diff --git "a/log/log-train-2023-02-05-17-58-35-2" "b/log/log-train-2023-02-05-17-58-35-2" new file mode 100644--- /dev/null +++ "b/log/log-train-2023-02-05-17-58-35-2" @@ -0,0 +1,25111 @@ +2023-02-05 17:58:35,364 INFO [train.py:973] (2/4) Training started +2023-02-05 17:58:35,365 INFO [train.py:983] (2/4) Device: cuda:2 +2023-02-05 17:58:35,412 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': 'r7n07', 'IP address': '10.1.7.7'}, 'world_size': 4, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 1, '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} +2023-02-05 17:58:35,412 INFO [train.py:994] (2/4) About to create model +2023-02-05 17:58:36,048 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. +2023-02-05 17:58:36,060 INFO [train.py:998] (2/4) Number of model parameters: 20697573 +2023-02-05 17:58:51,148 INFO [train.py:1013] (2/4) Using DDP +2023-02-05 17:58:51,427 INFO [asr_datamodule.py:420] (2/4) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts +2023-02-05 17:58:52,644 INFO [asr_datamodule.py:224] (2/4) Enable MUSAN +2023-02-05 17:58:52,645 INFO [asr_datamodule.py:225] (2/4) About to get Musan cuts +2023-02-05 17:58:54,523 INFO [asr_datamodule.py:249] (2/4) Enable SpecAugment +2023-02-05 17:58:54,523 INFO [asr_datamodule.py:250] (2/4) Time warp factor: 80 +2023-02-05 17:58:54,523 INFO [asr_datamodule.py:260] (2/4) Num frame mask: 10 +2023-02-05 17:58:54,523 INFO [asr_datamodule.py:273] (2/4) About to create train dataset +2023-02-05 17:58:54,523 INFO [asr_datamodule.py:300] (2/4) Using DynamicBucketingSampler. +2023-02-05 17:58:54,545 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 17:58:57,031 INFO [asr_datamodule.py:316] (2/4) About to create train dataloader +2023-02-05 17:58:57,031 INFO [asr_datamodule.py:430] (2/4) About to get dev-clean cuts +2023-02-05 17:58:57,033 INFO [asr_datamodule.py:437] (2/4) About to get dev-other cuts +2023-02-05 17:58:57,033 INFO [asr_datamodule.py:347] (2/4) About to create dev dataset +2023-02-05 17:58:57,380 INFO [asr_datamodule.py:364] (2/4) About to create dev dataloader +2023-02-05 17:59:06,521 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 17:59:11,987 INFO [train.py:901] (2/4) Epoch 1, batch 0, loss[loss=7.164, simple_loss=6.476, pruned_loss=6.868, over 7655.00 frames. ], tot_loss[loss=7.164, simple_loss=6.476, pruned_loss=6.868, over 7655.00 frames. ], batch size: 19, lr: 2.50e-02, grad_scale: 2.0 +2023-02-05 17:59:11,988 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 17:59:24,181 INFO [train.py:935] (2/4) Epoch 1, validation: loss=6.888, simple_loss=6.229, pruned_loss=6.575, over 944034.00 frames. +2023-02-05 17:59:24,182 INFO [train.py:936] (2/4) Maximum memory allocated so far is 4557MB +2023-02-05 17:59:31,367 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=4.10 vs. limit=2.0 +2023-02-05 17:59:37,738 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 17:59:48,696 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=5.29 vs. limit=2.0 +2023-02-05 17:59:55,489 INFO [train.py:901] (2/4) Epoch 1, batch 50, loss[loss=1.453, simple_loss=1.287, pruned_loss=1.483, over 8366.00 frames. ], tot_loss[loss=2.149, simple_loss=1.943, pruned_loss=1.971, over 359040.72 frames. ], batch size: 24, lr: 2.75e-02, grad_scale: 0.25 +2023-02-05 17:59:56,162 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:00:11,289 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 18:00:13,743 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:00:28,699 INFO [train.py:901] (2/4) Epoch 1, batch 100, loss[loss=1.087, simple_loss=0.9315, pruned_loss=1.231, over 7800.00 frames. ], tot_loss[loss=1.64, simple_loss=1.46, pruned_loss=1.614, over 641797.83 frames. ], batch size: 19, lr: 3.00e-02, grad_scale: 0.0625 +2023-02-05 18:00:28,821 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:00:32,052 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 18:00:32,809 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 3.087e+01 6.689e+01 1.862e+02 6.030e+02 6.185e+04, threshold=3.723e+02, percent-clipped=0.0 +2023-02-05 18:01:00,488 INFO [train.py:901] (2/4) Epoch 1, batch 150, loss[loss=1.113, simple_loss=0.95, pruned_loss=1.183, over 8342.00 frames. ], tot_loss[loss=1.408, simple_loss=1.238, pruned_loss=1.438, over 861772.80 frames. ], batch size: 26, lr: 3.25e-02, grad_scale: 0.0625 +2023-02-05 18:01:14,169 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=44.96 vs. limit=5.0 +2023-02-05 18:01:34,365 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=6.87 vs. limit=2.0 +2023-02-05 18:01:34,586 INFO [train.py:901] (2/4) Epoch 1, batch 200, loss[loss=1.017, simple_loss=0.8697, pruned_loss=1.003, over 8367.00 frames. ], tot_loss[loss=1.266, simple_loss=1.103, pruned_loss=1.303, over 1028310.46 frames. ], batch size: 49, lr: 3.50e-02, grad_scale: 0.125 +2023-02-05 18:01:36,296 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=31.05 vs. limit=5.0 +2023-02-05 18:01:37,983 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 3.848e+01 5.119e+01 6.630e+01 8.708e+01 3.236e+02, threshold=1.326e+02, percent-clipped=1.0 +2023-02-05 18:01:39,357 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=4.37 vs. limit=2.0 +2023-02-05 18:01:40,247 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6301, 4.6300, 4.6300, 4.6301, 4.6301, 4.6301, 4.6301, 4.6301], + device='cuda:2'), covar=tensor([2.6490e-05, 2.1035e-05, 5.3192e-05, 4.7522e-05, 3.6533e-05, 4.2642e-05, + 2.0418e-05, 3.2286e-05], device='cuda:2'), in_proj_covar=tensor([0.0014, 0.0014, 0.0014, 0.0014, 0.0014, 0.0014, 0.0014, 0.0014], + device='cuda:2'), out_proj_covar=tensor([9.3120e-06, 9.4909e-06, 9.3863e-06, 9.1082e-06, 9.5719e-06, 9.2246e-06, + 9.5521e-06, 9.3924e-06], device='cuda:2') +2023-02-05 18:02:05,438 INFO [train.py:901] (2/4) Epoch 1, batch 250, loss[loss=0.9433, simple_loss=0.7969, pruned_loss=0.9224, over 8583.00 frames. ], tot_loss[loss=1.179, simple_loss=1.019, pruned_loss=1.205, over 1158044.96 frames. ], batch size: 31, lr: 3.75e-02, grad_scale: 0.125 +2023-02-05 18:02:14,822 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 18:02:14,929 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9630, 2.9630, 2.9630, 2.9630, 2.9630, 2.9630, 2.9630, 2.9630], + device='cuda:2'), covar=tensor([3.2164e-05, 3.3923e-05, 5.4673e-05, 4.8642e-05, 3.3072e-05, 3.6856e-05, + 5.7300e-05, 3.8083e-05], device='cuda:2'), in_proj_covar=tensor([0.0013, 0.0013, 0.0013, 0.0014, 0.0013, 0.0013, 0.0014, 0.0013], + device='cuda:2'), out_proj_covar=tensor([9.1469e-06, 9.0843e-06, 9.1827e-06, 8.9071e-06, 9.1140e-06, 8.9794e-06, + 8.9708e-06, 8.8985e-06], device='cuda:2') +2023-02-05 18:02:17,543 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=8.68 vs. limit=2.0 +2023-02-05 18:02:22,941 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 18:02:37,911 INFO [train.py:901] (2/4) Epoch 1, batch 300, loss[loss=1.009, simple_loss=0.8393, pruned_loss=0.987, over 8287.00 frames. ], tot_loss[loss=1.12, simple_loss=0.9605, pruned_loss=1.135, over 1259241.78 frames. ], batch size: 23, lr: 4.00e-02, grad_scale: 0.25 +2023-02-05 18:02:42,331 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=306.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:02:42,690 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 4.041e+01 5.570e+01 7.201e+01 9.677e+01 1.807e+02, threshold=1.440e+02, percent-clipped=6.0 +2023-02-05 18:02:47,402 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:02:48,870 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2023-02-05 18:03:10,257 INFO [train.py:901] (2/4) Epoch 1, batch 350, loss[loss=1.078, simple_loss=0.8901, pruned_loss=1.031, over 8322.00 frames. ], tot_loss[loss=1.084, simple_loss=0.9222, pruned_loss=1.084, over 1344327.74 frames. ], batch size: 25, lr: 4.25e-02, grad_scale: 0.25 +2023-02-05 18:03:42,314 INFO [train.py:901] (2/4) Epoch 1, batch 400, loss[loss=0.8752, simple_loss=0.7162, pruned_loss=0.8232, over 7981.00 frames. ], tot_loss[loss=1.053, simple_loss=0.8887, pruned_loss=1.04, over 1402798.25 frames. ], batch size: 21, lr: 4.50e-02, grad_scale: 0.5 +2023-02-05 18:03:44,610 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:03:45,466 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 4.847e+01 5.714e+01 6.661e+01 8.261e+01 1.252e+02, threshold=1.332e+02, percent-clipped=0.0 +2023-02-05 18:03:55,288 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:04:11,519 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=445.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:04:15,508 INFO [train.py:901] (2/4) Epoch 1, batch 450, loss[loss=0.9806, simple_loss=0.8007, pruned_loss=0.8916, over 8511.00 frames. ], tot_loss[loss=1.03, simple_loss=0.8627, pruned_loss=1.002, over 1449653.42 frames. ], batch size: 28, lr: 4.75e-02, grad_scale: 0.5 +2023-02-05 18:04:36,885 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=5.64 vs. limit=5.0 +2023-02-05 18:04:45,722 INFO [train.py:901] (2/4) Epoch 1, batch 500, loss[loss=0.9982, simple_loss=0.8112, pruned_loss=0.8869, over 8352.00 frames. ], tot_loss[loss=1.012, simple_loss=0.8426, pruned_loss=0.9674, over 1487974.45 frames. ], batch size: 24, lr: 4.99e-02, grad_scale: 1.0 +2023-02-05 18:04:47,282 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=7.51 vs. limit=2.0 +2023-02-05 18:04:49,478 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 5.283e+01 6.268e+01 7.626e+01 9.977e+01 2.238e+02, threshold=1.525e+02, percent-clipped=10.0 +2023-02-05 18:05:16,926 INFO [train.py:901] (2/4) Epoch 1, batch 550, loss[loss=0.8864, simple_loss=0.726, pruned_loss=0.7485, over 8099.00 frames. ], tot_loss[loss=0.9934, simple_loss=0.8237, pruned_loss=0.9289, over 1514485.22 frames. ], batch size: 23, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:05:22,229 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:05:30,975 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=8.94 vs. limit=5.0 +2023-02-05 18:05:34,657 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=580.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:05:39,265 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:05:47,855 INFO [train.py:901] (2/4) Epoch 1, batch 600, loss[loss=0.8791, simple_loss=0.7289, pruned_loss=0.7012, over 8093.00 frames. ], tot_loss[loss=0.9781, simple_loss=0.8099, pruned_loss=0.8897, over 1534892.54 frames. ], batch size: 21, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:05:51,146 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 5.986e+01 8.101e+01 1.064e+02 1.512e+02 3.340e+02, threshold=2.128e+02, percent-clipped=22.0 +2023-02-05 18:05:51,944 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:05:57,565 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 18:06:15,545 INFO [train.py:901] (2/4) Epoch 1, batch 650, loss[loss=0.7856, simple_loss=0.656, pruned_loss=0.6015, over 7651.00 frames. ], tot_loss[loss=0.958, simple_loss=0.7941, pruned_loss=0.8449, over 1554479.62 frames. ], batch size: 19, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:06:16,980 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=5.07 vs. limit=2.0 +2023-02-05 18:06:20,641 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=658.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:06:25,234 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=8.69 vs. limit=5.0 +2023-02-05 18:06:31,053 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:06:35,971 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=2.62 vs. limit=2.0 +2023-02-05 18:06:44,409 INFO [train.py:901] (2/4) Epoch 1, batch 700, loss[loss=0.8922, simple_loss=0.7397, pruned_loss=0.6791, over 8091.00 frames. ], tot_loss[loss=0.9333, simple_loss=0.7755, pruned_loss=0.7976, over 1566963.43 frames. ], batch size: 21, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:06:45,066 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:06:48,201 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.109e+02 3.132e+02 4.412e+02 1.990e+03, threshold=6.264e+02, percent-clipped=73.0 +2023-02-05 18:07:14,475 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=749.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:07:15,369 INFO [train.py:901] (2/4) Epoch 1, batch 750, loss[loss=0.6782, simple_loss=0.5795, pruned_loss=0.4738, over 7427.00 frames. ], tot_loss[loss=0.9072, simple_loss=0.7563, pruned_loss=0.7508, over 1581831.75 frames. ], batch size: 17, lr: 4.97e-02, grad_scale: 1.0 +2023-02-05 18:07:25,623 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 18:07:26,856 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:07:32,325 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 18:07:43,624 INFO [train.py:901] (2/4) Epoch 1, batch 800, loss[loss=0.7198, simple_loss=0.6115, pruned_loss=0.5011, over 7689.00 frames. ], tot_loss[loss=0.8789, simple_loss=0.7356, pruned_loss=0.705, over 1588781.46 frames. ], batch size: 18, lr: 4.97e-02, grad_scale: 2.0 +2023-02-05 18:07:46,599 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.447e+02 2.528e+02 3.354e+02 4.455e+02 1.086e+03, threshold=6.708e+02, percent-clipped=4.0 +2023-02-05 18:07:51,303 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:07:58,528 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2958, 1.3372, 1.0677, 1.2178, 1.1082, 1.0115, 1.1676, 1.2563], + device='cuda:2'), covar=tensor([0.6132, 0.7841, 0.9596, 0.7309, 0.9089, 0.9021, 0.7759, 0.9574], + device='cuda:2'), in_proj_covar=tensor([0.0058, 0.0067, 0.0073, 0.0067, 0.0073, 0.0072, 0.0066, 0.0077], + device='cuda:2'), out_proj_covar=tensor([4.1819e-05, 4.9480e-05, 4.9706e-05, 4.6487e-05, 4.9191e-05, 4.6920e-05, + 4.3869e-05, 5.3112e-05], device='cuda:2') +2023-02-05 18:08:05,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:08:11,180 INFO [train.py:901] (2/4) Epoch 1, batch 850, loss[loss=0.7833, simple_loss=0.675, pruned_loss=0.5215, over 8458.00 frames. ], tot_loss[loss=0.8563, simple_loss=0.7195, pruned_loss=0.6662, over 1594074.66 frames. ], batch size: 27, lr: 4.96e-02, grad_scale: 2.0 +2023-02-05 18:08:11,882 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7108, 1.0789, 2.0276, 1.5175, 1.2097, 1.1850, 1.1840, 1.7279], + device='cuda:2'), covar=tensor([1.0189, 3.1751, 0.8547, 1.0032, 1.8385, 2.0228, 2.1747, 0.8871], + device='cuda:2'), in_proj_covar=tensor([0.0044, 0.0060, 0.0043, 0.0044, 0.0054, 0.0067, 0.0067, 0.0040], + device='cuda:2'), out_proj_covar=tensor([2.5911e-05, 4.3421e-05, 2.4613e-05, 2.3377e-05, 3.1656e-05, 4.4629e-05, + 3.9660e-05, 2.2601e-05], device='cuda:2') +2023-02-05 18:08:22,426 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:08:22,899 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:08:42,859 INFO [train.py:901] (2/4) Epoch 1, batch 900, loss[loss=0.7653, simple_loss=0.6598, pruned_loss=0.5028, over 8528.00 frames. ], tot_loss[loss=0.8305, simple_loss=0.7009, pruned_loss=0.6277, over 1597467.70 frames. ], batch size: 28, lr: 4.96e-02, grad_scale: 2.0 +2023-02-05 18:08:46,407 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 3.070e+02 3.818e+02 4.702e+02 7.623e+02, threshold=7.636e+02, percent-clipped=5.0 +2023-02-05 18:08:55,931 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=924.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:08:58,989 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=930.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:09:10,090 INFO [train.py:901] (2/4) Epoch 1, batch 950, loss[loss=0.7806, simple_loss=0.6694, pruned_loss=0.512, over 8465.00 frames. ], tot_loss[loss=0.8084, simple_loss=0.685, pruned_loss=0.5945, over 1602868.80 frames. ], batch size: 27, lr: 4.96e-02, grad_scale: 2.0 +2023-02-05 18:09:10,749 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=952.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:09:26,428 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 18:09:36,685 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4710, 1.6700, 3.2811, 2.1683, 2.6066, 3.1310, 1.4641, 1.3278], + device='cuda:2'), covar=tensor([1.3165, 2.2754, 0.3017, 0.9729, 1.1421, 0.6313, 1.8099, 1.8365], + device='cuda:2'), in_proj_covar=tensor([0.0060, 0.0068, 0.0038, 0.0055, 0.0060, 0.0048, 0.0073, 0.0065], + device='cuda:2'), out_proj_covar=tensor([3.9192e-05, 4.7741e-05, 2.1334e-05, 3.4574e-05, 3.9925e-05, 2.8983e-05, + 4.7570e-05, 4.4730e-05], device='cuda:2') +2023-02-05 18:09:37,672 INFO [train.py:901] (2/4) Epoch 1, batch 1000, loss[loss=0.7542, simple_loss=0.6571, pruned_loss=0.4752, over 8355.00 frames. ], tot_loss[loss=0.7845, simple_loss=0.668, pruned_loss=0.5616, over 1606355.82 frames. ], batch size: 24, lr: 4.95e-02, grad_scale: 2.0 +2023-02-05 18:09:40,951 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 3.215e+02 4.159e+02 4.799e+02 1.770e+03, threshold=8.319e+02, percent-clipped=6.0 +2023-02-05 18:09:52,911 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:09:53,904 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 18:09:59,186 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:10:02,615 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:05,084 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 18:10:05,582 INFO [train.py:901] (2/4) Epoch 1, batch 1050, loss[loss=0.6874, simple_loss=0.5967, pruned_loss=0.432, over 8141.00 frames. ], tot_loss[loss=0.7644, simple_loss=0.6535, pruned_loss=0.5342, over 1602933.40 frames. ], batch size: 22, lr: 4.95e-02, grad_scale: 2.0 +2023-02-05 18:10:07,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:14,061 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:10:33,046 INFO [train.py:901] (2/4) Epoch 1, batch 1100, loss[loss=0.6044, simple_loss=0.5278, pruned_loss=0.3727, over 7536.00 frames. ], tot_loss[loss=0.7451, simple_loss=0.6398, pruned_loss=0.5085, over 1601890.91 frames. ], batch size: 18, lr: 4.94e-02, grad_scale: 2.0 +2023-02-05 18:10:36,093 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 3.463e+02 4.480e+02 5.452e+02 1.232e+03, threshold=8.959e+02, percent-clipped=3.0 +2023-02-05 18:10:43,739 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:56,880 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:10:59,912 INFO [train.py:901] (2/4) Epoch 1, batch 1150, loss[loss=0.703, simple_loss=0.6215, pruned_loss=0.4212, over 8499.00 frames. ], tot_loss[loss=0.7284, simple_loss=0.6281, pruned_loss=0.4863, over 1604852.76 frames. ], batch size: 29, lr: 4.94e-02, grad_scale: 2.0 +2023-02-05 18:11:01,584 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 18:11:11,048 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2023-02-05 18:11:11,758 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:11:27,723 INFO [train.py:901] (2/4) Epoch 1, batch 1200, loss[loss=0.6869, simple_loss=0.594, pruned_loss=0.4235, over 8497.00 frames. ], tot_loss[loss=0.7137, simple_loss=0.6181, pruned_loss=0.4667, over 1611427.83 frames. ], batch size: 26, lr: 4.93e-02, grad_scale: 4.0 +2023-02-05 18:11:30,964 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 3.424e+02 4.173e+02 5.178e+02 8.029e+02, threshold=8.346e+02, percent-clipped=0.0 +2023-02-05 18:11:32,133 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:11:45,904 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0807, 1.3445, 1.3680, 1.0871, 1.1982, 1.3888, 0.7115, 0.9102], + device='cuda:2'), covar=tensor([0.5114, 0.4500, 0.4305, 0.6412, 0.7333, 0.3541, 0.8490, 0.6987], + device='cuda:2'), in_proj_covar=tensor([0.0069, 0.0072, 0.0068, 0.0070, 0.0088, 0.0066, 0.0083, 0.0079], + device='cuda:2'), out_proj_covar=tensor([4.7009e-05, 4.7788e-05, 4.4192e-05, 4.6527e-05, 6.3388e-05, 3.9561e-05, + 5.5886e-05, 5.1384e-05], device='cuda:2') +2023-02-05 18:11:54,243 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9688, 1.3314, 2.5612, 1.3441, 1.8403, 1.8713, 1.1259, 1.5104], + device='cuda:2'), covar=tensor([1.3166, 1.7315, 0.3142, 0.9488, 1.1208, 0.6498, 1.2656, 1.3139], + device='cuda:2'), in_proj_covar=tensor([0.0069, 0.0074, 0.0043, 0.0061, 0.0073, 0.0055, 0.0078, 0.0077], + device='cuda:2'), out_proj_covar=tensor([4.6758e-05, 5.1502e-05, 2.4205e-05, 3.9120e-05, 5.0068e-05, 3.4794e-05, + 4.9914e-05, 5.3807e-05], device='cuda:2') +2023-02-05 18:11:56,782 INFO [train.py:901] (2/4) Epoch 1, batch 1250, loss[loss=0.6093, simple_loss=0.5432, pruned_loss=0.356, over 8250.00 frames. ], tot_loss[loss=0.7012, simple_loss=0.6093, pruned_loss=0.4505, over 1606753.31 frames. ], batch size: 22, lr: 4.92e-02, grad_scale: 4.0 +2023-02-05 18:12:21,163 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:24,272 INFO [train.py:901] (2/4) Epoch 1, batch 1300, loss[loss=0.6638, simple_loss=0.5914, pruned_loss=0.3862, over 7819.00 frames. ], tot_loss[loss=0.6901, simple_loss=0.602, pruned_loss=0.4355, over 1610130.10 frames. ], batch size: 20, lr: 4.92e-02, grad_scale: 4.0 +2023-02-05 18:12:24,479 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:27,427 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.398e+02 3.917e+02 4.747e+02 6.152e+02 9.080e+02, threshold=9.493e+02, percent-clipped=1.0 +2023-02-05 18:12:30,301 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0625, 0.8300, 0.9138, 1.1842, 0.6064, 0.8213, 0.7952, 1.0749], + device='cuda:2'), covar=tensor([1.0489, 1.3925, 1.0122, 0.5149, 1.2366, 1.3617, 1.1965, 1.1750], + device='cuda:2'), in_proj_covar=tensor([0.0126, 0.0123, 0.0112, 0.0097, 0.0134, 0.0137, 0.0127, 0.0133], + device='cuda:2'), out_proj_covar=tensor([8.6420e-05, 8.6967e-05, 8.0697e-05, 5.8041e-05, 9.5128e-05, 9.3522e-05, + 9.0672e-05, 9.1660e-05], device='cuda:2') +2023-02-05 18:12:34,731 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:36,272 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:36,784 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1324.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:37,937 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:51,930 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:53,374 INFO [train.py:901] (2/4) Epoch 1, batch 1350, loss[loss=0.6161, simple_loss=0.5506, pruned_loss=0.3552, over 8078.00 frames. ], tot_loss[loss=0.6774, simple_loss=0.5934, pruned_loss=0.4206, over 1607357.92 frames. ], batch size: 21, lr: 4.91e-02, grad_scale: 4.0 +2023-02-05 18:13:22,445 INFO [train.py:901] (2/4) Epoch 1, batch 1400, loss[loss=0.6686, simple_loss=0.5952, pruned_loss=0.3859, over 8251.00 frames. ], tot_loss[loss=0.6652, simple_loss=0.5854, pruned_loss=0.4065, over 1612873.53 frames. ], batch size: 24, lr: 4.91e-02, grad_scale: 4.0 +2023-02-05 18:13:25,824 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 3.466e+02 4.520e+02 5.912e+02 1.396e+03, threshold=9.040e+02, percent-clipped=6.0 +2023-02-05 18:13:50,939 INFO [train.py:901] (2/4) Epoch 1, batch 1450, loss[loss=0.6044, simple_loss=0.5534, pruned_loss=0.3343, over 8616.00 frames. ], tot_loss[loss=0.6548, simple_loss=0.5783, pruned_loss=0.3947, over 1611459.68 frames. ], batch size: 39, lr: 4.90e-02, grad_scale: 4.0 +2023-02-05 18:13:51,590 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1452.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:13:54,963 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 18:14:03,117 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1149, 1.9115, 1.3579, 3.4861, 2.3641, 2.5654, 2.0757, 3.3724], + device='cuda:2'), covar=tensor([0.5288, 0.9685, 1.8939, 0.1409, 0.7151, 0.5328, 0.8154, 0.1918], + device='cuda:2'), in_proj_covar=tensor([0.0135, 0.0157, 0.0225, 0.0099, 0.0151, 0.0144, 0.0177, 0.0131], + device='cuda:2'), out_proj_covar=tensor([8.6278e-05, 1.1000e-04, 1.4765e-04, 6.2294e-05, 1.0497e-04, 8.9922e-05, + 1.1418e-04, 7.6896e-05], device='cuda:2') +2023-02-05 18:14:21,297 INFO [train.py:901] (2/4) Epoch 1, batch 1500, loss[loss=0.4871, simple_loss=0.4435, pruned_loss=0.2709, over 7430.00 frames. ], tot_loss[loss=0.6437, simple_loss=0.5706, pruned_loss=0.3832, over 1609638.73 frames. ], batch size: 17, lr: 4.89e-02, grad_scale: 4.0 +2023-02-05 18:14:22,376 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 +2023-02-05 18:14:24,733 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.084e+02 4.059e+02 4.884e+02 5.820e+02 1.191e+03, threshold=9.769e+02, percent-clipped=4.0 +2023-02-05 18:14:29,250 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1515.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:14:50,474 INFO [train.py:901] (2/4) Epoch 1, batch 1550, loss[loss=0.6331, simple_loss=0.5639, pruned_loss=0.3607, over 8312.00 frames. ], tot_loss[loss=0.6371, simple_loss=0.5659, pruned_loss=0.3754, over 1610204.72 frames. ], batch size: 25, lr: 4.89e-02, grad_scale: 4.0 +2023-02-05 18:15:08,649 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1580.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:15:10,861 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1584.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:15:20,750 INFO [train.py:901] (2/4) Epoch 1, batch 1600, loss[loss=0.6757, simple_loss=0.598, pruned_loss=0.3865, over 8615.00 frames. ], tot_loss[loss=0.6322, simple_loss=0.5626, pruned_loss=0.3692, over 1612563.73 frames. ], batch size: 39, lr: 4.88e-02, grad_scale: 8.0 +2023-02-05 18:15:23,989 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1605.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:15:24,958 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.844e+02 4.893e+02 6.465e+02 8.597e+02 2.177e+03, threshold=1.293e+03, percent-clipped=12.0 +2023-02-05 18:15:37,781 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1629.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:15:38,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1630.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:15:50,682 INFO [train.py:901] (2/4) Epoch 1, batch 1650, loss[loss=0.6347, simple_loss=0.5735, pruned_loss=0.3535, over 8191.00 frames. ], tot_loss[loss=0.6269, simple_loss=0.5598, pruned_loss=0.3625, over 1616981.38 frames. ], batch size: 23, lr: 4.87e-02, grad_scale: 8.0 +2023-02-05 18:15:54,304 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-05 18:16:13,744 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8345, 2.3321, 3.1848, 2.4130, 2.5359, 3.4083, 3.4891, 3.2347], + device='cuda:2'), covar=tensor([0.2115, 0.3258, 0.0435, 0.2043, 0.1620, 0.0400, 0.0288, 0.0547], + device='cuda:2'), in_proj_covar=tensor([0.0125, 0.0137, 0.0075, 0.0120, 0.0104, 0.0067, 0.0063, 0.0083], + device='cuda:2'), out_proj_covar=tensor([8.9322e-05, 1.0254e-04, 4.2933e-05, 7.8463e-05, 7.1411e-05, 3.8906e-05, + 3.5956e-05, 4.9323e-05], device='cuda:2') +2023-02-05 18:16:21,965 INFO [train.py:901] (2/4) Epoch 1, batch 1700, loss[loss=0.5989, simple_loss=0.5377, pruned_loss=0.3349, over 8352.00 frames. ], tot_loss[loss=0.6192, simple_loss=0.5555, pruned_loss=0.3543, over 1617660.71 frames. ], batch size: 24, lr: 4.86e-02, grad_scale: 8.0 +2023-02-05 18:16:25,354 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.633e+02 4.287e+02 5.230e+02 6.455e+02 2.107e+03, threshold=1.046e+03, percent-clipped=2.0 +2023-02-05 18:16:47,459 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-05 18:16:51,240 INFO [train.py:901] (2/4) Epoch 1, batch 1750, loss[loss=0.4947, simple_loss=0.4636, pruned_loss=0.2636, over 6788.00 frames. ], tot_loss[loss=0.6138, simple_loss=0.5527, pruned_loss=0.3481, over 1618518.39 frames. ], batch size: 15, lr: 4.86e-02, grad_scale: 8.0 +2023-02-05 18:17:11,148 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4938, 1.6440, 4.2318, 2.9136, 4.1564, 3.9189, 3.8618, 3.7726], + device='cuda:2'), covar=tensor([0.0173, 0.2616, 0.0230, 0.0627, 0.0264, 0.0217, 0.0242, 0.0368], + device='cuda:2'), in_proj_covar=tensor([0.0048, 0.0134, 0.0064, 0.0078, 0.0065, 0.0060, 0.0071, 0.0081], + device='cuda:2'), out_proj_covar=tensor([2.8501e-05, 8.5486e-05, 3.7921e-05, 5.1943e-05, 3.6545e-05, 3.4185e-05, + 4.2278e-05, 4.8152e-05], device='cuda:2') +2023-02-05 18:17:18,071 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1796.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:17:21,115 INFO [train.py:901] (2/4) Epoch 1, batch 1800, loss[loss=0.5591, simple_loss=0.5284, pruned_loss=0.295, over 8673.00 frames. ], tot_loss[loss=0.6039, simple_loss=0.5462, pruned_loss=0.3396, over 1615340.67 frames. ], batch size: 34, lr: 4.85e-02, grad_scale: 8.0 +2023-02-05 18:17:24,721 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.688e+02 4.554e+02 5.596e+02 6.733e+02 1.418e+03, threshold=1.119e+03, percent-clipped=4.0 +2023-02-05 18:17:52,119 INFO [train.py:901] (2/4) Epoch 1, batch 1850, loss[loss=0.5874, simple_loss=0.5335, pruned_loss=0.3224, over 8365.00 frames. ], tot_loss[loss=0.5973, simple_loss=0.5418, pruned_loss=0.3335, over 1616639.27 frames. ], batch size: 24, lr: 4.84e-02, grad_scale: 8.0 +2023-02-05 18:17:55,088 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1856.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:18:06,738 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1875.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:18:13,295 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1886.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:18:14,343 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1888.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:18:21,896 INFO [train.py:901] (2/4) Epoch 1, batch 1900, loss[loss=0.5729, simple_loss=0.5403, pruned_loss=0.3028, over 8454.00 frames. ], tot_loss[loss=0.5896, simple_loss=0.5373, pruned_loss=0.3267, over 1616792.05 frames. ], batch size: 25, lr: 4.83e-02, grad_scale: 8.0 +2023-02-05 18:18:25,476 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.326e+02 4.483e+02 5.242e+02 7.443e+02 2.270e+03, threshold=1.048e+03, percent-clipped=7.0 +2023-02-05 18:18:27,930 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1911.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:18:27,945 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1911.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:18:37,733 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1928.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:18:45,005 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 18:18:52,609 INFO [train.py:901] (2/4) Epoch 1, batch 1950, loss[loss=0.5976, simple_loss=0.5504, pruned_loss=0.3227, over 8328.00 frames. ], tot_loss[loss=0.5892, simple_loss=0.5371, pruned_loss=0.3252, over 1615701.81 frames. ], batch size: 25, lr: 4.83e-02, grad_scale: 8.0 +2023-02-05 18:18:55,539 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 18:19:05,749 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1973.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:19:11,336 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 18:19:12,085 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8047, 1.8299, 1.0994, 1.5557, 1.7272, 1.0125, 1.6067, 2.2253], + device='cuda:2'), covar=tensor([0.3789, 0.4028, 0.5655, 0.4000, 0.3799, 0.6380, 0.3388, 0.2639], + device='cuda:2'), in_proj_covar=tensor([0.0133, 0.0124, 0.0112, 0.0117, 0.0143, 0.0129, 0.0113, 0.0118], + device='cuda:2'), out_proj_covar=tensor([1.0037e-04, 9.0365e-05, 8.5071e-05, 8.8396e-05, 1.0665e-04, 9.4161e-05, + 8.7248e-05, 8.9818e-05], device='cuda:2') +2023-02-05 18:19:23,723 INFO [train.py:901] (2/4) Epoch 1, batch 2000, loss[loss=0.5661, simple_loss=0.5401, pruned_loss=0.2961, over 8619.00 frames. ], tot_loss[loss=0.5828, simple_loss=0.5334, pruned_loss=0.3197, over 1613653.06 frames. ], batch size: 31, lr: 4.82e-02, grad_scale: 8.0 +2023-02-05 18:19:27,549 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.734e+02 4.600e+02 5.655e+02 7.771e+02 1.691e+03, threshold=1.131e+03, percent-clipped=5.0 +2023-02-05 18:19:50,359 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2043.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:19:56,663 INFO [train.py:901] (2/4) Epoch 1, batch 2050, loss[loss=0.5672, simple_loss=0.5329, pruned_loss=0.3007, over 8286.00 frames. ], tot_loss[loss=0.5751, simple_loss=0.5295, pruned_loss=0.3132, over 1613646.71 frames. ], batch size: 23, lr: 4.81e-02, grad_scale: 8.0 +2023-02-05 18:20:17,890 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5776, 1.5264, 2.5126, 1.4633, 2.0870, 2.8933, 2.9682, 2.4555], + device='cuda:2'), covar=tensor([0.2911, 0.3557, 0.0492, 0.3073, 0.1418, 0.0328, 0.0307, 0.0454], + device='cuda:2'), in_proj_covar=tensor([0.0168, 0.0180, 0.0096, 0.0165, 0.0144, 0.0084, 0.0079, 0.0096], + device='cuda:2'), out_proj_covar=tensor([1.1817e-04, 1.2844e-04, 5.8833e-05, 1.1005e-04, 1.0260e-04, 5.3408e-05, + 4.6832e-05, 6.1049e-05], device='cuda:2') +2023-02-05 18:20:21,051 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2088.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:20:29,072 INFO [train.py:901] (2/4) Epoch 1, batch 2100, loss[loss=0.5522, simple_loss=0.5284, pruned_loss=0.288, over 8343.00 frames. ], tot_loss[loss=0.5707, simple_loss=0.5277, pruned_loss=0.3091, over 1611808.85 frames. ], batch size: 26, lr: 4.80e-02, grad_scale: 16.0 +2023-02-05 18:20:32,716 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.532e+02 4.654e+02 5.875e+02 8.240e+02 2.515e+03, threshold=1.175e+03, percent-clipped=11.0 +2023-02-05 18:21:01,648 INFO [train.py:901] (2/4) Epoch 1, batch 2150, loss[loss=0.4429, simple_loss=0.4299, pruned_loss=0.2279, over 7935.00 frames. ], tot_loss[loss=0.5608, simple_loss=0.5218, pruned_loss=0.3016, over 1611613.52 frames. ], batch size: 20, lr: 4.79e-02, grad_scale: 16.0 +2023-02-05 18:21:11,759 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2167.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:21:29,905 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2192.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:21:35,018 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2200.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:21:35,574 INFO [train.py:901] (2/4) Epoch 1, batch 2200, loss[loss=0.485, simple_loss=0.4651, pruned_loss=0.2525, over 7803.00 frames. ], tot_loss[loss=0.5565, simple_loss=0.5199, pruned_loss=0.2979, over 1615110.02 frames. ], batch size: 19, lr: 4.78e-02, grad_scale: 16.0 +2023-02-05 18:21:37,175 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-05 18:21:39,335 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 3.885e+02 5.100e+02 6.280e+02 1.293e+03, threshold=1.020e+03, percent-clipped=3.0 +2023-02-05 18:21:46,995 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2219.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:21:55,782 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2232.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:22:07,870 INFO [train.py:901] (2/4) Epoch 1, batch 2250, loss[loss=0.5949, simple_loss=0.5616, pruned_loss=0.3141, over 8334.00 frames. ], tot_loss[loss=0.5519, simple_loss=0.5178, pruned_loss=0.294, over 1615207.91 frames. ], batch size: 25, lr: 4.77e-02, grad_scale: 16.0 +2023-02-05 18:22:41,040 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2299.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:22:42,091 INFO [train.py:901] (2/4) Epoch 1, batch 2300, loss[loss=0.4525, simple_loss=0.4323, pruned_loss=0.2364, over 7554.00 frames. ], tot_loss[loss=0.5448, simple_loss=0.5133, pruned_loss=0.289, over 1611272.47 frames. ], batch size: 18, lr: 4.77e-02, grad_scale: 16.0 +2023-02-05 18:22:45,952 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.442e+02 5.272e+02 6.513e+02 7.975e+02 1.884e+03, threshold=1.303e+03, percent-clipped=9.0 +2023-02-05 18:22:51,196 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2315.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:22:56,966 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2324.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:03,177 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2334.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:23:09,695 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2344.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:12,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2347.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:23:14,708 INFO [train.py:901] (2/4) Epoch 1, batch 2350, loss[loss=0.4633, simple_loss=0.4707, pruned_loss=0.2279, over 8330.00 frames. ], tot_loss[loss=0.5392, simple_loss=0.5102, pruned_loss=0.2847, over 1606326.26 frames. ], batch size: 25, lr: 4.76e-02, grad_scale: 16.0 +2023-02-05 18:23:19,250 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2358.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:21,088 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5141, 1.3793, 1.4629, 1.6497, 1.2357, 1.1624, 0.9843, 1.6211], + device='cuda:2'), covar=tensor([0.2647, 0.2618, 0.2100, 0.0985, 0.2899, 0.3164, 0.3620, 0.2316], + device='cuda:2'), in_proj_covar=tensor([0.0183, 0.0177, 0.0157, 0.0128, 0.0222, 0.0198, 0.0219, 0.0184], + device='cuda:2'), out_proj_covar=tensor([1.3583e-04, 1.3226e-04, 1.2538e-04, 8.9005e-05, 1.6174e-04, 1.4613e-04, + 1.6180e-04, 1.3970e-04], device='cuda:2') +2023-02-05 18:23:26,041 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2369.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:46,437 INFO [train.py:901] (2/4) Epoch 1, batch 2400, loss[loss=0.5104, simple_loss=0.4952, pruned_loss=0.2628, over 8526.00 frames. ], tot_loss[loss=0.5354, simple_loss=0.5085, pruned_loss=0.2816, over 1609826.72 frames. ], batch size: 28, lr: 4.75e-02, grad_scale: 16.0 +2023-02-05 18:23:50,348 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.591e+02 4.467e+02 5.905e+02 7.151e+02 1.301e+03, threshold=1.181e+03, percent-clipped=0.0 +2023-02-05 18:24:16,548 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2023-02-05 18:24:20,801 INFO [train.py:901] (2/4) Epoch 1, batch 2450, loss[loss=0.5304, simple_loss=0.5139, pruned_loss=0.2735, over 8132.00 frames. ], tot_loss[loss=0.5332, simple_loss=0.5079, pruned_loss=0.2797, over 1614520.76 frames. ], batch size: 22, lr: 4.74e-02, grad_scale: 16.0 +2023-02-05 18:24:52,767 INFO [train.py:901] (2/4) Epoch 1, batch 2500, loss[loss=0.6053, simple_loss=0.5438, pruned_loss=0.3334, over 6977.00 frames. ], tot_loss[loss=0.5277, simple_loss=0.5042, pruned_loss=0.2759, over 1611009.03 frames. ], batch size: 71, lr: 4.73e-02, grad_scale: 16.0 +2023-02-05 18:24:54,530 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2023-02-05 18:24:56,548 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 5.238e+02 6.448e+02 8.237e+02 1.660e+03, threshold=1.290e+03, percent-clipped=6.0 +2023-02-05 18:24:58,276 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 18:25:25,592 INFO [train.py:901] (2/4) Epoch 1, batch 2550, loss[loss=0.4743, simple_loss=0.4834, pruned_loss=0.2326, over 7650.00 frames. ], tot_loss[loss=0.5252, simple_loss=0.5025, pruned_loss=0.2741, over 1614790.43 frames. ], batch size: 19, lr: 4.72e-02, grad_scale: 16.0 +2023-02-05 18:25:38,472 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2571.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:25:51,085 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2590.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:25:52,533 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2023-02-05 18:25:54,860 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2596.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:25:57,891 INFO [train.py:901] (2/4) Epoch 1, batch 2600, loss[loss=0.4654, simple_loss=0.4657, pruned_loss=0.2326, over 8237.00 frames. ], tot_loss[loss=0.5192, simple_loss=0.4989, pruned_loss=0.2699, over 1614168.11 frames. ], batch size: 22, lr: 4.71e-02, grad_scale: 16.0 +2023-02-05 18:25:59,372 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2603.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:26:01,609 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.388e+02 4.352e+02 5.534e+02 7.344e+02 1.370e+03, threshold=1.107e+03, percent-clipped=3.0 +2023-02-05 18:26:06,886 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2615.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:26:15,228 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2628.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:26:24,305 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3763, 2.1351, 4.0811, 4.1128, 2.7067, 1.2097, 1.9856, 2.8596], + device='cuda:2'), covar=tensor([0.3256, 0.2769, 0.0205, 0.0323, 0.1856, 0.2902, 0.2261, 0.1817], + device='cuda:2'), in_proj_covar=tensor([0.0167, 0.0129, 0.0061, 0.0079, 0.0145, 0.0135, 0.0133, 0.0147], + device='cuda:2'), out_proj_covar=tensor([1.0886e-04, 8.2293e-05, 3.5962e-05, 4.5528e-05, 8.9661e-05, 8.3390e-05, + 8.2727e-05, 8.8429e-05], device='cuda:2') +2023-02-05 18:26:31,161 INFO [train.py:901] (2/4) Epoch 1, batch 2650, loss[loss=0.4574, simple_loss=0.4768, pruned_loss=0.219, over 8199.00 frames. ], tot_loss[loss=0.515, simple_loss=0.4969, pruned_loss=0.2667, over 1613026.59 frames. ], batch size: 23, lr: 4.70e-02, grad_scale: 16.0 +2023-02-05 18:26:54,440 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 18:27:03,471 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-05 18:27:03,816 INFO [train.py:901] (2/4) Epoch 1, batch 2700, loss[loss=0.4226, simple_loss=0.4388, pruned_loss=0.2032, over 8079.00 frames. ], tot_loss[loss=0.513, simple_loss=0.496, pruned_loss=0.2651, over 1614713.32 frames. ], batch size: 21, lr: 4.69e-02, grad_scale: 16.0 +2023-02-05 18:27:04,568 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2702.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:27:05,224 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2703.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:27:08,306 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.214e+02 4.351e+02 5.311e+02 6.408e+02 1.471e+03, threshold=1.062e+03, percent-clipped=4.0 +2023-02-05 18:27:15,001 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6314, 1.7586, 1.5639, 1.7922, 1.6312, 1.7748, 1.3882, 1.8914], + device='cuda:2'), covar=tensor([0.1427, 0.1580, 0.2307, 0.0858, 0.2224, 0.1604, 0.2686, 0.1355], + device='cuda:2'), in_proj_covar=tensor([0.0107, 0.0103, 0.0139, 0.0084, 0.0122, 0.0104, 0.0147, 0.0109], + device='cuda:2'), out_proj_covar=tensor([7.5862e-05, 7.1819e-05, 9.3531e-05, 5.8992e-05, 8.7493e-05, 7.1684e-05, + 1.0255e-04, 7.2005e-05], device='cuda:2') +2023-02-05 18:27:37,286 INFO [train.py:901] (2/4) Epoch 1, batch 2750, loss[loss=0.5237, simple_loss=0.5009, pruned_loss=0.2733, over 7011.00 frames. ], tot_loss[loss=0.509, simple_loss=0.4941, pruned_loss=0.262, over 1614165.88 frames. ], batch size: 71, lr: 4.68e-02, grad_scale: 16.0 +2023-02-05 18:28:11,565 INFO [train.py:901] (2/4) Epoch 1, batch 2800, loss[loss=0.4534, simple_loss=0.4503, pruned_loss=0.2282, over 7649.00 frames. ], tot_loss[loss=0.5044, simple_loss=0.4912, pruned_loss=0.2589, over 1613126.36 frames. ], batch size: 19, lr: 4.67e-02, grad_scale: 16.0 +2023-02-05 18:28:15,265 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.340e+02 3.916e+02 4.898e+02 6.530e+02 2.276e+03, threshold=9.797e+02, percent-clipped=2.0 +2023-02-05 18:28:21,891 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2817.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:28:38,539 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9241, 1.4245, 5.2981, 2.6250, 5.4342, 4.8987, 4.9499, 4.9941], + device='cuda:2'), covar=tensor([0.0156, 0.4279, 0.0189, 0.1101, 0.0238, 0.0200, 0.0365, 0.0296], + device='cuda:2'), in_proj_covar=tensor([0.0079, 0.0232, 0.0100, 0.0125, 0.0110, 0.0111, 0.0118, 0.0127], + device='cuda:2'), out_proj_covar=tensor([5.0212e-05, 1.4106e-04, 6.6818e-05, 8.4086e-05, 6.4305e-05, 6.4503e-05, + 7.3779e-05, 7.7152e-05], device='cuda:2') +2023-02-05 18:28:38,548 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2842.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:28:44,179 INFO [train.py:901] (2/4) Epoch 1, batch 2850, loss[loss=0.4516, simple_loss=0.4466, pruned_loss=0.2283, over 7818.00 frames. ], tot_loss[loss=0.5012, simple_loss=0.4891, pruned_loss=0.2567, over 1610956.00 frames. ], batch size: 20, lr: 4.66e-02, grad_scale: 16.0 +2023-02-05 18:29:18,784 INFO [train.py:901] (2/4) Epoch 1, batch 2900, loss[loss=0.3887, simple_loss=0.4083, pruned_loss=0.1845, over 7426.00 frames. ], tot_loss[loss=0.5003, simple_loss=0.4886, pruned_loss=0.256, over 1608796.63 frames. ], batch size: 17, lr: 4.65e-02, grad_scale: 16.0 +2023-02-05 18:29:22,667 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.417e+02 4.413e+02 5.664e+02 7.338e+02 1.737e+03, threshold=1.133e+03, percent-clipped=8.0 +2023-02-05 18:29:48,929 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 18:29:52,148 INFO [train.py:901] (2/4) Epoch 1, batch 2950, loss[loss=0.402, simple_loss=0.4211, pruned_loss=0.1915, over 7544.00 frames. ], tot_loss[loss=0.4975, simple_loss=0.4871, pruned_loss=0.254, over 1609439.06 frames. ], batch size: 18, lr: 4.64e-02, grad_scale: 16.0 +2023-02-05 18:29:54,925 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2955.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:30:25,896 INFO [train.py:901] (2/4) Epoch 1, batch 3000, loss[loss=0.4192, simple_loss=0.4393, pruned_loss=0.1995, over 8578.00 frames. ], tot_loss[loss=0.4961, simple_loss=0.4864, pruned_loss=0.2529, over 1613129.39 frames. ], batch size: 34, lr: 4.63e-02, grad_scale: 16.0 +2023-02-05 18:30:25,897 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 18:30:40,788 INFO [train.py:935] (2/4) Epoch 1, validation: loss=0.4518, simple_loss=0.5106, pruned_loss=0.1966, over 944034.00 frames. +2023-02-05 18:30:40,788 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6115MB +2023-02-05 18:30:44,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.692e+02 4.264e+02 5.642e+02 7.781e+02 1.743e+03, threshold=1.128e+03, percent-clipped=6.0 +2023-02-05 18:31:07,324 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3037.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:31:13,909 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3047.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:31:16,509 INFO [train.py:901] (2/4) Epoch 1, batch 3050, loss[loss=0.5479, simple_loss=0.5357, pruned_loss=0.28, over 8350.00 frames. ], tot_loss[loss=0.4946, simple_loss=0.486, pruned_loss=0.2516, over 1611625.04 frames. ], batch size: 26, lr: 4.62e-02, grad_scale: 16.0 +2023-02-05 18:31:30,841 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3073.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:31:47,505 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3098.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:31:49,292 INFO [train.py:901] (2/4) Epoch 1, batch 3100, loss[loss=0.5595, simple_loss=0.5355, pruned_loss=0.2918, over 8555.00 frames. ], tot_loss[loss=0.4982, simple_loss=0.4879, pruned_loss=0.2542, over 1613310.70 frames. ], batch size: 31, lr: 4.61e-02, grad_scale: 16.0 +2023-02-05 18:31:53,106 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.570e+02 4.257e+02 6.045e+02 8.311e+02 2.838e+03, threshold=1.209e+03, percent-clipped=13.0 +2023-02-05 18:32:24,455 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 +2023-02-05 18:32:24,769 INFO [train.py:901] (2/4) Epoch 1, batch 3150, loss[loss=0.5046, simple_loss=0.5037, pruned_loss=0.2527, over 8293.00 frames. ], tot_loss[loss=0.4984, simple_loss=0.4887, pruned_loss=0.2541, over 1615510.93 frames. ], batch size: 23, lr: 4.60e-02, grad_scale: 16.0 +2023-02-05 18:32:32,227 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3162.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:32:47,642 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3186.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:32:57,065 INFO [train.py:901] (2/4) Epoch 1, batch 3200, loss[loss=0.4646, simple_loss=0.4696, pruned_loss=0.2298, over 8256.00 frames. ], tot_loss[loss=0.4941, simple_loss=0.4859, pruned_loss=0.2512, over 1614507.72 frames. ], batch size: 24, lr: 4.59e-02, grad_scale: 16.0 +2023-02-05 18:33:00,911 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 4.232e+02 5.266e+02 6.948e+02 2.778e+03, threshold=1.053e+03, percent-clipped=2.0 +2023-02-05 18:33:32,107 INFO [train.py:901] (2/4) Epoch 1, batch 3250, loss[loss=0.5349, simple_loss=0.5155, pruned_loss=0.2772, over 8030.00 frames. ], tot_loss[loss=0.4945, simple_loss=0.4859, pruned_loss=0.2515, over 1614118.96 frames. ], batch size: 22, lr: 4.58e-02, grad_scale: 16.0 +2023-02-05 18:33:47,473 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1886, 1.0719, 3.0741, 1.3571, 2.7689, 2.5644, 2.6246, 2.5718], + device='cuda:2'), covar=tensor([0.0221, 0.3097, 0.0322, 0.1188, 0.0388, 0.0394, 0.0466, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0080, 0.0242, 0.0103, 0.0135, 0.0119, 0.0121, 0.0119, 0.0134], + device='cuda:2'), out_proj_covar=tensor([5.0069e-05, 1.4327e-04, 6.7418e-05, 9.0741e-05, 7.2105e-05, 7.2046e-05, + 7.5567e-05, 8.4699e-05], device='cuda:2') +2023-02-05 18:33:48,111 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9692, 4.4180, 3.5573, 1.3440, 3.3146, 3.7767, 3.7886, 3.2156], + device='cuda:2'), covar=tensor([0.1041, 0.0401, 0.0713, 0.3969, 0.0542, 0.0459, 0.0971, 0.0607], + device='cuda:2'), in_proj_covar=tensor([0.0158, 0.0121, 0.0143, 0.0200, 0.0110, 0.0092, 0.0144, 0.0103], + device='cuda:2'), out_proj_covar=tensor([1.1819e-04, 9.8272e-05, 9.5450e-05, 1.3694e-04, 7.4376e-05, 6.6610e-05, + 1.1293e-04, 7.1706e-05], device='cuda:2') +2023-02-05 18:33:57,953 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8476, 1.2865, 5.0142, 2.5378, 5.1985, 4.4977, 4.7893, 4.8133], + device='cuda:2'), covar=tensor([0.0107, 0.3652, 0.0187, 0.1026, 0.0181, 0.0194, 0.0301, 0.0245], + device='cuda:2'), in_proj_covar=tensor([0.0079, 0.0244, 0.0104, 0.0135, 0.0120, 0.0121, 0.0120, 0.0133], + device='cuda:2'), out_proj_covar=tensor([4.9538e-05, 1.4432e-04, 6.8518e-05, 9.1354e-05, 7.2619e-05, 7.2649e-05, + 7.5828e-05, 8.4223e-05], device='cuda:2') +2023-02-05 18:34:04,442 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3299.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:34:05,630 INFO [train.py:901] (2/4) Epoch 1, batch 3300, loss[loss=0.5965, simple_loss=0.5429, pruned_loss=0.325, over 7244.00 frames. ], tot_loss[loss=0.4943, simple_loss=0.4866, pruned_loss=0.251, over 1614843.89 frames. ], batch size: 71, lr: 4.57e-02, grad_scale: 16.0 +2023-02-05 18:34:05,853 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3301.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:34:06,488 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7524, 2.5058, 1.3749, 1.5842, 2.2580, 1.9431, 1.4862, 2.1616], + device='cuda:2'), covar=tensor([0.1887, 0.1329, 0.2801, 0.1327, 0.2240, 0.1473, 0.4175, 0.1536], + device='cuda:2'), in_proj_covar=tensor([0.0157, 0.0129, 0.0191, 0.0112, 0.0175, 0.0135, 0.0212, 0.0148], + device='cuda:2'), out_proj_covar=tensor([1.1224e-04, 9.2370e-05, 1.3281e-04, 8.3220e-05, 1.2822e-04, 9.8638e-05, + 1.4975e-04, 1.0509e-04], device='cuda:2') +2023-02-05 18:34:08,973 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3306.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:34:09,426 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 4.334e+02 5.638e+02 7.160e+02 2.697e+03, threshold=1.128e+03, percent-clipped=10.0 +2023-02-05 18:34:28,315 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 +2023-02-05 18:34:39,416 INFO [train.py:901] (2/4) Epoch 1, batch 3350, loss[loss=0.4783, simple_loss=0.4668, pruned_loss=0.2449, over 8040.00 frames. ], tot_loss[loss=0.4866, simple_loss=0.4811, pruned_loss=0.2461, over 1612398.56 frames. ], batch size: 20, lr: 4.56e-02, grad_scale: 16.0 +2023-02-05 18:35:01,941 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3381.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:35:14,988 INFO [train.py:901] (2/4) Epoch 1, batch 3400, loss[loss=0.5244, simple_loss=0.5159, pruned_loss=0.2665, over 8587.00 frames. ], tot_loss[loss=0.4845, simple_loss=0.48, pruned_loss=0.2445, over 1617446.77 frames. ], batch size: 31, lr: 4.55e-02, grad_scale: 16.0 +2023-02-05 18:35:19,028 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.486e+02 3.960e+02 5.068e+02 6.311e+02 1.481e+03, threshold=1.014e+03, percent-clipped=3.0 +2023-02-05 18:35:23,816 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3414.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:35:26,554 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3418.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:35:43,703 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3443.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:35:48,737 INFO [train.py:901] (2/4) Epoch 1, batch 3450, loss[loss=0.4272, simple_loss=0.4322, pruned_loss=0.2111, over 7808.00 frames. ], tot_loss[loss=0.4834, simple_loss=0.4792, pruned_loss=0.2438, over 1615659.69 frames. ], batch size: 20, lr: 4.54e-02, grad_scale: 16.0 +2023-02-05 18:36:21,031 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3496.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:36:24,211 INFO [train.py:901] (2/4) Epoch 1, batch 3500, loss[loss=0.3858, simple_loss=0.4045, pruned_loss=0.1835, over 7255.00 frames. ], tot_loss[loss=0.4827, simple_loss=0.4788, pruned_loss=0.2433, over 1615504.84 frames. ], batch size: 16, lr: 4.53e-02, grad_scale: 16.0 +2023-02-05 18:36:28,198 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.305e+02 4.405e+02 5.773e+02 7.537e+02 2.537e+03, threshold=1.155e+03, percent-clipped=7.0 +2023-02-05 18:36:36,229 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 18:36:57,804 INFO [train.py:901] (2/4) Epoch 1, batch 3550, loss[loss=0.4734, simple_loss=0.4701, pruned_loss=0.2384, over 8249.00 frames. ], tot_loss[loss=0.4809, simple_loss=0.4775, pruned_loss=0.2421, over 1614737.37 frames. ], batch size: 24, lr: 4.51e-02, grad_scale: 16.0 +2023-02-05 18:37:02,106 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3557.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:37:07,150 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3564.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:37:19,181 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3582.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:37:22,793 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1284, 4.3868, 3.8645, 1.4882, 3.6094, 3.8880, 3.8940, 3.4229], + device='cuda:2'), covar=tensor([0.0711, 0.0307, 0.0542, 0.3263, 0.0400, 0.0415, 0.0712, 0.0484], + device='cuda:2'), in_proj_covar=tensor([0.0182, 0.0138, 0.0162, 0.0215, 0.0121, 0.0104, 0.0166, 0.0113], + device='cuda:2'), out_proj_covar=tensor([1.3900e-04, 1.0937e-04, 1.0719e-04, 1.4812e-04, 8.2449e-05, 7.5362e-05, + 1.2759e-04, 7.9803e-05], device='cuda:2') +2023-02-05 18:37:33,298 INFO [train.py:901] (2/4) Epoch 1, batch 3600, loss[loss=0.4789, simple_loss=0.4775, pruned_loss=0.2401, over 8259.00 frames. ], tot_loss[loss=0.4843, simple_loss=0.4791, pruned_loss=0.2447, over 1610019.17 frames. ], batch size: 24, lr: 4.50e-02, grad_scale: 16.0 +2023-02-05 18:37:33,625 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2023-02-05 18:37:37,961 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.853e+02 4.660e+02 6.337e+02 8.772e+02 4.832e+03, threshold=1.267e+03, percent-clipped=11.0 +2023-02-05 18:38:06,596 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3650.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:38:06,677 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4546, 1.6179, 1.0510, 1.5138, 1.5539, 1.3908, 1.3216, 2.1162], + device='cuda:2'), covar=tensor([0.1505, 0.1233, 0.2712, 0.0715, 0.1898, 0.1447, 0.2624, 0.0832], + device='cuda:2'), in_proj_covar=tensor([0.0195, 0.0151, 0.0229, 0.0134, 0.0203, 0.0169, 0.0239, 0.0180], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 18:38:07,049 INFO [train.py:901] (2/4) Epoch 1, batch 3650, loss[loss=0.5549, simple_loss=0.5268, pruned_loss=0.2915, over 7972.00 frames. ], tot_loss[loss=0.4794, simple_loss=0.4759, pruned_loss=0.2415, over 1608574.46 frames. ], batch size: 21, lr: 4.49e-02, grad_scale: 16.0 +2023-02-05 18:38:19,613 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.7531, 1.0942, 1.2028, 1.0812, 0.8705, 1.1900, 0.2056, 0.8236], + device='cuda:2'), covar=tensor([0.0616, 0.0627, 0.0498, 0.0407, 0.0710, 0.0348, 0.1493, 0.0825], + device='cuda:2'), in_proj_covar=tensor([0.0129, 0.0125, 0.0106, 0.0112, 0.0125, 0.0095, 0.0162, 0.0127], + device='cuda:2'), out_proj_covar=tensor([9.2613e-05, 9.6194e-05, 7.6196e-05, 8.1479e-05, 9.3805e-05, 6.5106e-05, + 1.2417e-04, 1.0064e-04], device='cuda:2') +2023-02-05 18:38:19,628 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3670.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:38:36,829 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3694.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:38:37,586 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3695.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:38:40,432 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 18:38:41,136 INFO [train.py:901] (2/4) Epoch 1, batch 3700, loss[loss=0.4501, simple_loss=0.4558, pruned_loss=0.2222, over 7936.00 frames. ], tot_loss[loss=0.4834, simple_loss=0.4794, pruned_loss=0.2437, over 1616929.86 frames. ], batch size: 20, lr: 4.48e-02, grad_scale: 16.0 +2023-02-05 18:38:45,140 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.178e+02 4.586e+02 6.278e+02 1.050e+03 3.437e+03, threshold=1.256e+03, percent-clipped=14.0 +2023-02-05 18:39:17,454 INFO [train.py:901] (2/4) Epoch 1, batch 3750, loss[loss=0.5377, simple_loss=0.5322, pruned_loss=0.2716, over 8359.00 frames. ], tot_loss[loss=0.4828, simple_loss=0.4798, pruned_loss=0.2429, over 1621562.39 frames. ], batch size: 48, lr: 4.47e-02, grad_scale: 16.0 +2023-02-05 18:39:18,333 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3752.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:39:27,123 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3765.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:39:32,484 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2023-02-05 18:39:35,219 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3777.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:39:38,644 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9431, 2.0628, 1.8825, 3.0307, 1.9609, 1.7249, 2.0953, 2.0482], + device='cuda:2'), covar=tensor([0.1929, 0.1850, 0.1659, 0.0276, 0.2463, 0.1946, 0.2526, 0.1955], + device='cuda:2'), in_proj_covar=tensor([0.0248, 0.0256, 0.0242, 0.0156, 0.0314, 0.0290, 0.0339, 0.0248], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-05 18:39:51,681 INFO [train.py:901] (2/4) Epoch 1, batch 3800, loss[loss=0.4504, simple_loss=0.4454, pruned_loss=0.2277, over 8249.00 frames. ], tot_loss[loss=0.4788, simple_loss=0.4767, pruned_loss=0.2404, over 1618437.81 frames. ], batch size: 24, lr: 4.46e-02, grad_scale: 16.0 +2023-02-05 18:39:55,874 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.457e+02 5.389e+02 6.979e+02 9.091e+02 1.609e+03, threshold=1.396e+03, percent-clipped=5.0 +2023-02-05 18:40:24,140 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2844, 1.9301, 2.0272, 1.6997, 1.2884, 2.0726, 0.4958, 1.3136], + device='cuda:2'), covar=tensor([0.0846, 0.0547, 0.0359, 0.0552, 0.0812, 0.0387, 0.1791, 0.0926], + device='cuda:2'), in_proj_covar=tensor([0.0137, 0.0128, 0.0112, 0.0121, 0.0131, 0.0099, 0.0170, 0.0137], + device='cuda:2'), out_proj_covar=tensor([9.9139e-05, 9.9916e-05, 8.0945e-05, 8.7123e-05, 9.9756e-05, 6.8807e-05, + 1.3293e-04, 1.1100e-04], device='cuda:2') +2023-02-05 18:40:27,870 INFO [train.py:901] (2/4) Epoch 1, batch 3850, loss[loss=0.5425, simple_loss=0.5343, pruned_loss=0.2754, over 8562.00 frames. ], tot_loss[loss=0.4788, simple_loss=0.4767, pruned_loss=0.2404, over 1616838.12 frames. ], batch size: 39, lr: 4.45e-02, grad_scale: 16.0 +2023-02-05 18:40:40,587 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2023-02-05 18:40:46,554 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 18:40:56,729 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 18:41:00,999 INFO [train.py:901] (2/4) Epoch 1, batch 3900, loss[loss=0.4638, simple_loss=0.4652, pruned_loss=0.2312, over 8291.00 frames. ], tot_loss[loss=0.4779, simple_loss=0.4764, pruned_loss=0.2397, over 1621623.71 frames. ], batch size: 23, lr: 4.44e-02, grad_scale: 16.0 +2023-02-05 18:41:04,025 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-05 18:41:04,999 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 3.102e+02 5.552e+02 7.100e+02 9.321e+02 1.906e+03, threshold=1.420e+03, percent-clipped=2.0 +2023-02-05 18:41:05,743 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3908.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:41:29,951 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3944.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:41:35,339 INFO [train.py:901] (2/4) Epoch 1, batch 3950, loss[loss=0.5847, simple_loss=0.5569, pruned_loss=0.3062, over 7049.00 frames. ], tot_loss[loss=0.4754, simple_loss=0.4754, pruned_loss=0.2377, over 1623444.83 frames. ], batch size: 71, lr: 4.43e-02, grad_scale: 16.0 +2023-02-05 18:41:46,993 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7714, 2.2444, 2.3694, 2.4412, 1.9568, 1.0162, 2.1528, 2.1402], + device='cuda:2'), covar=tensor([0.1743, 0.0857, 0.0664, 0.0479, 0.1052, 0.1642, 0.0409, 0.0696], + device='cuda:2'), in_proj_covar=tensor([0.0130, 0.0088, 0.0066, 0.0077, 0.0094, 0.0106, 0.0107, 0.0104], + device='cuda:2'), out_proj_covar=tensor([7.8475e-05, 5.0052e-05, 3.6312e-05, 4.4707e-05, 5.5055e-05, 5.5944e-05, + 5.6340e-05, 5.6760e-05], device='cuda:2') +2023-02-05 18:42:10,919 INFO [train.py:901] (2/4) Epoch 1, batch 4000, loss[loss=0.3896, simple_loss=0.4098, pruned_loss=0.1847, over 7433.00 frames. ], tot_loss[loss=0.4748, simple_loss=0.4752, pruned_loss=0.2372, over 1624922.46 frames. ], batch size: 17, lr: 4.42e-02, grad_scale: 8.0 +2023-02-05 18:42:15,518 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.262e+02 4.572e+02 5.687e+02 7.371e+02 1.820e+03, threshold=1.137e+03, percent-clipped=4.0 +2023-02-05 18:42:15,728 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3684, 1.7238, 1.3901, 1.5564, 1.7652, 1.4237, 1.5775, 1.7915], + device='cuda:2'), covar=tensor([0.2668, 0.2743, 0.3060, 0.2734, 0.1785, 0.2991, 0.2204, 0.2055], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0257, 0.0240, 0.0248, 0.0256, 0.0238, 0.0250, 0.0243], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-05 18:42:24,563 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4021.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:42:25,855 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4023.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:42:36,382 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4038.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:42:42,709 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4046.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:42:46,085 INFO [train.py:901] (2/4) Epoch 1, batch 4050, loss[loss=0.4471, simple_loss=0.4619, pruned_loss=0.2161, over 8507.00 frames. ], tot_loss[loss=0.4758, simple_loss=0.4759, pruned_loss=0.2378, over 1621497.76 frames. ], batch size: 26, lr: 4.41e-02, grad_scale: 8.0 +2023-02-05 18:43:13,072 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-05 18:43:22,338 INFO [train.py:901] (2/4) Epoch 1, batch 4100, loss[loss=0.5816, simple_loss=0.556, pruned_loss=0.3035, over 8576.00 frames. ], tot_loss[loss=0.4746, simple_loss=0.4748, pruned_loss=0.2371, over 1620928.94 frames. ], batch size: 31, lr: 4.40e-02, grad_scale: 8.0 +2023-02-05 18:43:26,883 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.479e+02 4.889e+02 6.474e+02 8.616e+02 2.054e+03, threshold=1.295e+03, percent-clipped=5.0 +2023-02-05 18:43:46,473 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6873, 3.1273, 1.7038, 2.4843, 2.5502, 2.6278, 1.8212, 2.5440], + device='cuda:2'), covar=tensor([0.1561, 0.0758, 0.2325, 0.0839, 0.1700, 0.1557, 0.2943, 0.1257], + device='cuda:2'), in_proj_covar=tensor([0.0222, 0.0164, 0.0264, 0.0161, 0.0237, 0.0196, 0.0267, 0.0208], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 18:43:56,549 INFO [train.py:901] (2/4) Epoch 1, batch 4150, loss[loss=0.42, simple_loss=0.4486, pruned_loss=0.1957, over 8469.00 frames. ], tot_loss[loss=0.469, simple_loss=0.4714, pruned_loss=0.2333, over 1621056.17 frames. ], batch size: 25, lr: 4.39e-02, grad_scale: 8.0 +2023-02-05 18:43:58,155 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4153.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:44:02,263 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4159.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:44:33,532 INFO [train.py:901] (2/4) Epoch 1, batch 4200, loss[loss=0.5268, simple_loss=0.5115, pruned_loss=0.2711, over 7130.00 frames. ], tot_loss[loss=0.4671, simple_loss=0.4702, pruned_loss=0.232, over 1617783.58 frames. ], batch size: 73, lr: 4.38e-02, grad_scale: 8.0 +2023-02-05 18:44:38,312 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 4.057e+02 5.109e+02 6.409e+02 1.525e+03, threshold=1.022e+03, percent-clipped=2.0 +2023-02-05 18:44:44,325 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 18:44:50,725 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.70 vs. limit=5.0 +2023-02-05 18:45:05,002 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 18:45:07,075 INFO [train.py:901] (2/4) Epoch 1, batch 4250, loss[loss=0.4038, simple_loss=0.431, pruned_loss=0.1883, over 8467.00 frames. ], tot_loss[loss=0.4664, simple_loss=0.4693, pruned_loss=0.2318, over 1619991.30 frames. ], batch size: 25, lr: 4.36e-02, grad_scale: 8.0 +2023-02-05 18:45:14,147 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6325, 2.0766, 2.1410, 2.3865, 1.5522, 1.2192, 1.9985, 2.0212], + device='cuda:2'), covar=tensor([0.2286, 0.0981, 0.0631, 0.0528, 0.1029, 0.1546, 0.0938, 0.1047], + device='cuda:2'), in_proj_covar=tensor([0.0145, 0.0094, 0.0070, 0.0076, 0.0098, 0.0112, 0.0119, 0.0110], + device='cuda:2'), out_proj_covar=tensor([8.5845e-05, 5.3153e-05, 3.8652e-05, 4.3017e-05, 5.5241e-05, 6.1308e-05, + 6.6512e-05, 5.9167e-05], device='cuda:2') +2023-02-05 18:45:25,182 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5850, 2.1941, 1.6647, 1.8560, 1.9380, 1.9465, 2.5789, 2.6205], + device='cuda:2'), covar=tensor([0.2135, 0.2960, 0.2894, 0.2558, 0.2472, 0.2981, 0.2058, 0.1808], + device='cuda:2'), in_proj_covar=tensor([0.0257, 0.0262, 0.0243, 0.0253, 0.0265, 0.0242, 0.0258, 0.0244], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 18:45:26,606 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4279.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:45:33,325 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:45:42,872 INFO [train.py:901] (2/4) Epoch 1, batch 4300, loss[loss=0.5123, simple_loss=0.5103, pruned_loss=0.2571, over 8353.00 frames. ], tot_loss[loss=0.4652, simple_loss=0.4685, pruned_loss=0.2309, over 1619079.61 frames. ], batch size: 26, lr: 4.35e-02, grad_scale: 8.0 +2023-02-05 18:45:45,725 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4304.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:45:47,020 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4306.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:45:48,889 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.647e+02 4.666e+02 6.207e+02 8.078e+02 1.600e+03, threshold=1.241e+03, percent-clipped=6.0 +2023-02-05 18:46:03,750 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-05 18:46:08,342 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1891, 1.6788, 3.3003, 1.0019, 2.0098, 1.7250, 1.1207, 1.9478], + device='cuda:2'), covar=tensor([0.1404, 0.1484, 0.0236, 0.1544, 0.1244, 0.1952, 0.1481, 0.1189], + device='cuda:2'), in_proj_covar=tensor([0.0195, 0.0207, 0.0149, 0.0206, 0.0243, 0.0268, 0.0206, 0.0236], + device='cuda:2'), out_proj_covar=tensor([1.3432e-04, 1.4800e-04, 9.7854e-05, 1.3998e-04, 1.5791e-04, 1.9178e-04, + 1.3279e-04, 1.5358e-04], device='cuda:2') +2023-02-05 18:46:18,301 INFO [train.py:901] (2/4) Epoch 1, batch 4350, loss[loss=0.4664, simple_loss=0.465, pruned_loss=0.2339, over 7797.00 frames. ], tot_loss[loss=0.4629, simple_loss=0.4667, pruned_loss=0.2296, over 1612849.84 frames. ], batch size: 20, lr: 4.34e-02, grad_scale: 8.0 +2023-02-05 18:46:37,371 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 18:46:52,944 INFO [train.py:901] (2/4) Epoch 1, batch 4400, loss[loss=0.3859, simple_loss=0.41, pruned_loss=0.1809, over 7525.00 frames. ], tot_loss[loss=0.4592, simple_loss=0.4639, pruned_loss=0.2272, over 1607430.15 frames. ], batch size: 18, lr: 4.33e-02, grad_scale: 8.0 +2023-02-05 18:46:54,533 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4403.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:46:57,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 4.338e+02 5.789e+02 7.262e+02 1.136e+03, threshold=1.158e+03, percent-clipped=0.0 +2023-02-05 18:46:58,934 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4409.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:47:18,600 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4434.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:47:21,208 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 18:47:25,469 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1282, 1.4098, 1.5980, 1.6412, 1.4894, 1.1588, 1.3984, 1.5585], + device='cuda:2'), covar=tensor([0.2598, 0.1050, 0.0686, 0.0750, 0.0957, 0.1518, 0.1185, 0.0952], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0107, 0.0078, 0.0088, 0.0112, 0.0125, 0.0131, 0.0127], + device='cuda:2'), out_proj_covar=tensor([1.0117e-04, 6.1551e-05, 4.2938e-05, 5.0339e-05, 6.2559e-05, 6.8881e-05, + 7.3920e-05, 6.8342e-05], device='cuda:2') +2023-02-05 18:47:29,979 INFO [train.py:901] (2/4) Epoch 1, batch 4450, loss[loss=0.3938, simple_loss=0.4048, pruned_loss=0.1914, over 7224.00 frames. ], tot_loss[loss=0.4584, simple_loss=0.4635, pruned_loss=0.2267, over 1609859.49 frames. ], batch size: 16, lr: 4.32e-02, grad_scale: 8.0 +2023-02-05 18:48:04,125 INFO [train.py:901] (2/4) Epoch 1, batch 4500, loss[loss=0.4435, simple_loss=0.4654, pruned_loss=0.2108, over 8246.00 frames. ], tot_loss[loss=0.4567, simple_loss=0.4621, pruned_loss=0.2256, over 1609661.73 frames. ], batch size: 24, lr: 4.31e-02, grad_scale: 8.0 +2023-02-05 18:48:04,970 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5872, 1.3709, 2.8554, 1.5395, 2.1410, 3.2303, 3.0966, 2.8541], + device='cuda:2'), covar=tensor([0.2125, 0.2420, 0.0384, 0.2470, 0.1095, 0.0253, 0.0328, 0.0441], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0248, 0.0135, 0.0238, 0.0181, 0.0103, 0.0104, 0.0149], + device='cuda:2'), out_proj_covar=tensor([1.8068e-04, 1.9319e-04, 1.2062e-04, 1.7884e-04, 1.6182e-04, 8.4974e-05, + 9.4653e-05, 1.2548e-04], device='cuda:2') +2023-02-05 18:48:05,597 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4503.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:48:09,057 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.251e+02 4.383e+02 5.863e+02 8.313e+02 2.632e+03, threshold=1.173e+03, percent-clipped=9.0 +2023-02-05 18:48:15,323 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 18:48:41,811 INFO [train.py:901] (2/4) Epoch 1, batch 4550, loss[loss=0.4245, simple_loss=0.444, pruned_loss=0.2025, over 8790.00 frames. ], tot_loss[loss=0.4516, simple_loss=0.4592, pruned_loss=0.222, over 1613261.10 frames. ], batch size: 32, lr: 4.30e-02, grad_scale: 8.0 +2023-02-05 18:48:47,550 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0412, 2.0814, 1.5364, 1.4062, 1.8910, 1.7400, 2.0376, 2.3106], + device='cuda:2'), covar=tensor([0.2060, 0.2534, 0.2591, 0.2696, 0.1893, 0.2412, 0.1968, 0.1603], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0281, 0.0261, 0.0267, 0.0279, 0.0254, 0.0263, 0.0259], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 18:48:58,681 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0022, 1.3988, 3.9020, 2.1643, 3.5868, 3.2659, 3.3540, 3.3552], + device='cuda:2'), covar=tensor([0.0146, 0.3026, 0.0200, 0.1056, 0.0303, 0.0280, 0.0280, 0.0313], + device='cuda:2'), in_proj_covar=tensor([0.0103, 0.0292, 0.0134, 0.0168, 0.0149, 0.0149, 0.0144, 0.0163], + device='cuda:2'), out_proj_covar=tensor([6.5236e-05, 1.6856e-04, 8.5137e-05, 1.1351e-04, 8.8889e-05, 9.3388e-05, + 9.1333e-05, 1.0776e-04], device='cuda:2') +2023-02-05 18:49:16,696 INFO [train.py:901] (2/4) Epoch 1, batch 4600, loss[loss=0.5059, simple_loss=0.4858, pruned_loss=0.263, over 7981.00 frames. ], tot_loss[loss=0.4536, simple_loss=0.4602, pruned_loss=0.2235, over 1612171.89 frames. ], batch size: 21, lr: 4.29e-02, grad_scale: 8.0 +2023-02-05 18:49:21,483 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.209e+02 3.983e+02 5.037e+02 6.922e+02 1.236e+03, threshold=1.007e+03, percent-clipped=2.0 +2023-02-05 18:49:28,473 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4618.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:49:51,612 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4650.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:49:52,226 INFO [train.py:901] (2/4) Epoch 1, batch 4650, loss[loss=0.4739, simple_loss=0.481, pruned_loss=0.2334, over 8440.00 frames. ], tot_loss[loss=0.4523, simple_loss=0.4591, pruned_loss=0.2228, over 1609879.06 frames. ], batch size: 27, lr: 4.28e-02, grad_scale: 8.0 +2023-02-05 18:49:59,118 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4659.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:50:16,195 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4684.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:50:27,574 INFO [train.py:901] (2/4) Epoch 1, batch 4700, loss[loss=0.4776, simple_loss=0.496, pruned_loss=0.2296, over 8450.00 frames. ], tot_loss[loss=0.4482, simple_loss=0.4568, pruned_loss=0.2199, over 1613615.04 frames. ], batch size: 27, lr: 4.27e-02, grad_scale: 8.0 +2023-02-05 18:50:32,365 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.254e+02 4.576e+02 5.443e+02 6.674e+02 1.320e+03, threshold=1.089e+03, percent-clipped=4.0 +2023-02-05 18:51:01,881 INFO [train.py:901] (2/4) Epoch 1, batch 4750, loss[loss=0.3784, simple_loss=0.3913, pruned_loss=0.1827, over 7682.00 frames. ], tot_loss[loss=0.4452, simple_loss=0.4549, pruned_loss=0.2177, over 1619036.15 frames. ], batch size: 18, lr: 4.26e-02, grad_scale: 8.0 +2023-02-05 18:51:12,268 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4765.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:51:21,687 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 18:51:23,819 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 18:51:37,814 INFO [train.py:901] (2/4) Epoch 1, batch 4800, loss[loss=0.4089, simple_loss=0.4212, pruned_loss=0.1983, over 7259.00 frames. ], tot_loss[loss=0.4457, simple_loss=0.4551, pruned_loss=0.2181, over 1620927.19 frames. ], batch size: 16, lr: 4.25e-02, grad_scale: 8.0 +2023-02-05 18:51:42,622 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.690e+02 4.367e+02 5.327e+02 7.244e+02 1.939e+03, threshold=1.065e+03, percent-clipped=6.0 +2023-02-05 18:52:11,409 INFO [train.py:901] (2/4) Epoch 1, batch 4850, loss[loss=0.4086, simple_loss=0.4257, pruned_loss=0.1958, over 7965.00 frames. ], tot_loss[loss=0.4465, simple_loss=0.4554, pruned_loss=0.2188, over 1619949.48 frames. ], batch size: 21, lr: 4.24e-02, grad_scale: 8.0 +2023-02-05 18:52:13,503 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 18:52:18,628 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2023-02-05 18:52:27,489 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4874.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:52:47,400 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4899.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:52:48,546 INFO [train.py:901] (2/4) Epoch 1, batch 4900, loss[loss=0.453, simple_loss=0.4688, pruned_loss=0.2186, over 8751.00 frames. ], tot_loss[loss=0.4437, simple_loss=0.4537, pruned_loss=0.2168, over 1622333.13 frames. ], batch size: 39, lr: 4.23e-02, grad_scale: 8.0 +2023-02-05 18:52:53,380 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.332e+02 4.394e+02 5.447e+02 6.722e+02 1.310e+03, threshold=1.089e+03, percent-clipped=5.0 +2023-02-05 18:53:22,702 INFO [train.py:901] (2/4) Epoch 1, batch 4950, loss[loss=0.5518, simple_loss=0.5064, pruned_loss=0.2986, over 7232.00 frames. ], tot_loss[loss=0.4416, simple_loss=0.4519, pruned_loss=0.2156, over 1618576.08 frames. ], batch size: 16, lr: 4.21e-02, grad_scale: 8.0 +2023-02-05 18:53:59,108 INFO [train.py:901] (2/4) Epoch 1, batch 5000, loss[loss=0.4609, simple_loss=0.4695, pruned_loss=0.2262, over 8507.00 frames. ], tot_loss[loss=0.4419, simple_loss=0.4521, pruned_loss=0.2158, over 1617287.07 frames. ], batch size: 26, lr: 4.20e-02, grad_scale: 8.0 +2023-02-05 18:54:04,640 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.658e+02 4.358e+02 5.438e+02 7.182e+02 1.797e+03, threshold=1.088e+03, percent-clipped=3.0 +2023-02-05 18:54:13,645 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5021.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:54:30,644 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5046.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:54:33,890 INFO [train.py:901] (2/4) Epoch 1, batch 5050, loss[loss=0.4102, simple_loss=0.4373, pruned_loss=0.1916, over 8506.00 frames. ], tot_loss[loss=0.4387, simple_loss=0.4497, pruned_loss=0.2139, over 1612316.44 frames. ], batch size: 26, lr: 4.19e-02, grad_scale: 8.0 +2023-02-05 18:54:50,690 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 18:55:08,921 INFO [train.py:901] (2/4) Epoch 1, batch 5100, loss[loss=0.4272, simple_loss=0.4203, pruned_loss=0.2171, over 7791.00 frames. ], tot_loss[loss=0.4396, simple_loss=0.4504, pruned_loss=0.2144, over 1609015.81 frames. ], batch size: 19, lr: 4.18e-02, grad_scale: 8.0 +2023-02-05 18:55:13,360 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2023-02-05 18:55:13,597 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.507e+02 4.431e+02 5.257e+02 6.582e+02 1.311e+03, threshold=1.051e+03, percent-clipped=2.0 +2023-02-05 18:55:45,841 INFO [train.py:901] (2/4) Epoch 1, batch 5150, loss[loss=0.4781, simple_loss=0.4955, pruned_loss=0.2304, over 8444.00 frames. ], tot_loss[loss=0.4422, simple_loss=0.4524, pruned_loss=0.216, over 1609892.01 frames. ], batch size: 27, lr: 4.17e-02, grad_scale: 8.0 +2023-02-05 18:56:19,014 INFO [train.py:901] (2/4) Epoch 1, batch 5200, loss[loss=0.4503, simple_loss=0.4585, pruned_loss=0.221, over 8036.00 frames. ], tot_loss[loss=0.4417, simple_loss=0.4517, pruned_loss=0.2159, over 1606498.06 frames. ], batch size: 22, lr: 4.16e-02, grad_scale: 8.0 +2023-02-05 18:56:23,576 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 3.937e+02 5.264e+02 6.479e+02 1.558e+03, threshold=1.053e+03, percent-clipped=7.0 +2023-02-05 18:56:51,633 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 18:56:55,106 INFO [train.py:901] (2/4) Epoch 1, batch 5250, loss[loss=0.448, simple_loss=0.4418, pruned_loss=0.227, over 6008.00 frames. ], tot_loss[loss=0.4389, simple_loss=0.45, pruned_loss=0.2139, over 1605941.76 frames. ], batch size: 13, lr: 4.15e-02, grad_scale: 8.0 +2023-02-05 18:57:28,852 INFO [train.py:901] (2/4) Epoch 1, batch 5300, loss[loss=0.4444, simple_loss=0.463, pruned_loss=0.2129, over 8486.00 frames. ], tot_loss[loss=0.4413, simple_loss=0.4517, pruned_loss=0.2155, over 1605390.83 frames. ], batch size: 29, lr: 4.14e-02, grad_scale: 8.0 +2023-02-05 18:57:33,640 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.076e+02 4.278e+02 4.955e+02 6.641e+02 1.586e+03, threshold=9.909e+02, percent-clipped=4.0 +2023-02-05 18:57:34,876 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2023-02-05 18:58:04,007 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=5.13 vs. limit=5.0 +2023-02-05 18:58:04,342 INFO [train.py:901] (2/4) Epoch 1, batch 5350, loss[loss=0.4835, simple_loss=0.4571, pruned_loss=0.255, over 7693.00 frames. ], tot_loss[loss=0.4428, simple_loss=0.4528, pruned_loss=0.2165, over 1606023.76 frames. ], batch size: 18, lr: 4.13e-02, grad_scale: 8.0 +2023-02-05 18:58:13,884 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7144, 1.8607, 1.8519, 2.7527, 1.6128, 1.2182, 1.7633, 2.0003], + device='cuda:2'), covar=tensor([0.1380, 0.1511, 0.1216, 0.0227, 0.1907, 0.2243, 0.1868, 0.1201], + device='cuda:2'), in_proj_covar=tensor([0.0278, 0.0298, 0.0281, 0.0167, 0.0334, 0.0321, 0.0373, 0.0278], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 18:58:39,818 INFO [train.py:901] (2/4) Epoch 1, batch 5400, loss[loss=0.4673, simple_loss=0.4834, pruned_loss=0.2256, over 8351.00 frames. ], tot_loss[loss=0.4401, simple_loss=0.4506, pruned_loss=0.2148, over 1611284.98 frames. ], batch size: 24, lr: 4.12e-02, grad_scale: 8.0 +2023-02-05 18:58:44,298 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.977e+02 4.515e+02 5.788e+02 7.308e+02 1.362e+03, threshold=1.158e+03, percent-clipped=5.0 +2023-02-05 18:59:13,401 INFO [train.py:901] (2/4) Epoch 1, batch 5450, loss[loss=0.4047, simple_loss=0.4344, pruned_loss=0.1875, over 8248.00 frames. ], tot_loss[loss=0.4396, simple_loss=0.4502, pruned_loss=0.2145, over 1612048.72 frames. ], batch size: 22, lr: 4.11e-02, grad_scale: 8.0 +2023-02-05 18:59:41,792 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 18:59:49,949 INFO [train.py:901] (2/4) Epoch 1, batch 5500, loss[loss=0.4907, simple_loss=0.4968, pruned_loss=0.2423, over 8504.00 frames. ], tot_loss[loss=0.4371, simple_loss=0.4487, pruned_loss=0.2127, over 1614030.06 frames. ], batch size: 28, lr: 4.10e-02, grad_scale: 8.0 +2023-02-05 18:59:54,517 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.397e+02 4.451e+02 5.295e+02 6.340e+02 1.239e+03, threshold=1.059e+03, percent-clipped=2.0 +2023-02-05 19:00:23,628 INFO [train.py:901] (2/4) Epoch 1, batch 5550, loss[loss=0.4758, simple_loss=0.4721, pruned_loss=0.2398, over 8123.00 frames. ], tot_loss[loss=0.4376, simple_loss=0.4493, pruned_loss=0.213, over 1617561.77 frames. ], batch size: 22, lr: 4.09e-02, grad_scale: 8.0 +2023-02-05 19:01:00,923 INFO [train.py:901] (2/4) Epoch 1, batch 5600, loss[loss=0.4322, simple_loss=0.4642, pruned_loss=0.2001, over 8455.00 frames. ], tot_loss[loss=0.4337, simple_loss=0.4466, pruned_loss=0.2104, over 1611139.69 frames. ], batch size: 27, lr: 4.08e-02, grad_scale: 8.0 +2023-02-05 19:01:05,771 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 3.916e+02 5.301e+02 6.582e+02 1.340e+03, threshold=1.060e+03, percent-clipped=3.0 +2023-02-05 19:01:34,541 INFO [train.py:901] (2/4) Epoch 1, batch 5650, loss[loss=0.3311, simple_loss=0.3582, pruned_loss=0.152, over 7685.00 frames. ], tot_loss[loss=0.433, simple_loss=0.446, pruned_loss=0.21, over 1608537.34 frames. ], batch size: 18, lr: 4.07e-02, grad_scale: 8.0 +2023-02-05 19:01:35,630 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 19:01:45,693 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 19:01:45,826 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5668.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:02:09,329 INFO [train.py:901] (2/4) Epoch 1, batch 5700, loss[loss=0.4667, simple_loss=0.4804, pruned_loss=0.2265, over 8489.00 frames. ], tot_loss[loss=0.4358, simple_loss=0.4479, pruned_loss=0.2119, over 1611451.45 frames. ], batch size: 28, lr: 4.06e-02, grad_scale: 8.0 +2023-02-05 19:02:15,264 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 3.140e+02 4.740e+02 5.744e+02 8.008e+02 1.790e+03, threshold=1.149e+03, percent-clipped=10.0 +2023-02-05 19:02:44,478 INFO [train.py:901] (2/4) Epoch 1, batch 5750, loss[loss=0.3805, simple_loss=0.4061, pruned_loss=0.1774, over 7780.00 frames. ], tot_loss[loss=0.4316, simple_loss=0.4449, pruned_loss=0.2091, over 1608981.46 frames. ], batch size: 19, lr: 4.05e-02, grad_scale: 8.0 +2023-02-05 19:02:51,395 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 19:02:51,839 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 19:02:54,503 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-05 19:02:59,859 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5773.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:03:19,619 INFO [train.py:901] (2/4) Epoch 1, batch 5800, loss[loss=0.4244, simple_loss=0.4386, pruned_loss=0.2051, over 8086.00 frames. ], tot_loss[loss=0.4293, simple_loss=0.4432, pruned_loss=0.2077, over 1605198.22 frames. ], batch size: 21, lr: 4.04e-02, grad_scale: 8.0 +2023-02-05 19:03:24,537 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.458e+02 3.671e+02 4.595e+02 5.667e+02 1.405e+03, threshold=9.190e+02, percent-clipped=2.0 +2023-02-05 19:03:55,603 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.29 vs. limit=5.0 +2023-02-05 19:03:57,239 INFO [train.py:901] (2/4) Epoch 1, batch 5850, loss[loss=0.4226, simple_loss=0.4086, pruned_loss=0.2183, over 7540.00 frames. ], tot_loss[loss=0.4303, simple_loss=0.4438, pruned_loss=0.2084, over 1606018.02 frames. ], batch size: 18, lr: 4.03e-02, grad_scale: 8.0 +2023-02-05 19:04:15,201 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5876.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:04:32,485 INFO [train.py:901] (2/4) Epoch 1, batch 5900, loss[loss=0.4258, simple_loss=0.4263, pruned_loss=0.2126, over 7652.00 frames. ], tot_loss[loss=0.4282, simple_loss=0.4424, pruned_loss=0.2069, over 1607849.64 frames. ], batch size: 19, lr: 4.02e-02, grad_scale: 8.0 +2023-02-05 19:04:37,231 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 3.095e+02 4.155e+02 5.559e+02 6.668e+02 2.372e+03, threshold=1.112e+03, percent-clipped=6.0 +2023-02-05 19:04:54,075 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 19:05:09,347 INFO [train.py:901] (2/4) Epoch 1, batch 5950, loss[loss=0.4502, simple_loss=0.4651, pruned_loss=0.2177, over 8255.00 frames. ], tot_loss[loss=0.4284, simple_loss=0.4424, pruned_loss=0.2072, over 1605632.77 frames. ], batch size: 24, lr: 4.01e-02, grad_scale: 8.0 +2023-02-05 19:05:44,547 INFO [train.py:901] (2/4) Epoch 1, batch 6000, loss[loss=0.3853, simple_loss=0.414, pruned_loss=0.1783, over 8484.00 frames. ], tot_loss[loss=0.4286, simple_loss=0.4426, pruned_loss=0.2073, over 1607308.74 frames. ], batch size: 28, lr: 4.00e-02, grad_scale: 16.0 +2023-02-05 19:05:44,548 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 19:06:02,004 INFO [train.py:935] (2/4) Epoch 1, validation: loss=0.3351, simple_loss=0.4011, pruned_loss=0.1346, over 944034.00 frames. +2023-02-05 19:06:02,005 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 19:06:02,815 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.7432, 1.6809, 1.5357, 0.2610, 1.1541, 1.1077, 0.2178, 1.3994], + device='cuda:2'), covar=tensor([0.0707, 0.0569, 0.0375, 0.1045, 0.0527, 0.0643, 0.0912, 0.0281], + device='cuda:2'), in_proj_covar=tensor([0.0149, 0.0112, 0.0093, 0.0152, 0.0113, 0.0156, 0.0152, 0.0120], + device='cuda:2'), out_proj_covar=tensor([1.0927e-04, 8.0912e-05, 7.1124e-05, 1.1967e-04, 9.3822e-05, 1.1756e-04, + 1.1941e-04, 8.7026e-05], device='cuda:2') +2023-02-05 19:06:06,794 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.323e+02 3.694e+02 4.999e+02 6.330e+02 1.596e+03, threshold=9.998e+02, percent-clipped=5.0 +2023-02-05 19:06:06,984 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6008.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:06:09,470 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6012.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:06:35,735 INFO [train.py:901] (2/4) Epoch 1, batch 6050, loss[loss=0.4491, simple_loss=0.4272, pruned_loss=0.2355, over 7242.00 frames. ], tot_loss[loss=0.4336, simple_loss=0.4454, pruned_loss=0.2109, over 1608102.89 frames. ], batch size: 16, lr: 3.99e-02, grad_scale: 8.0 +2023-02-05 19:06:42,863 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:07:12,055 INFO [train.py:901] (2/4) Epoch 1, batch 6100, loss[loss=0.5123, simple_loss=0.5102, pruned_loss=0.2572, over 8027.00 frames. ], tot_loss[loss=0.4324, simple_loss=0.4451, pruned_loss=0.2099, over 1608829.34 frames. ], batch size: 22, lr: 3.98e-02, grad_scale: 8.0 +2023-02-05 19:07:17,503 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.508e+02 4.942e+02 6.048e+02 7.564e+02 1.774e+03, threshold=1.210e+03, percent-clipped=15.0 +2023-02-05 19:07:18,495 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8636, 2.1217, 2.8566, 3.4118, 2.1148, 1.5777, 1.9631, 2.3400], + device='cuda:2'), covar=tensor([0.1535, 0.0792, 0.0272, 0.0189, 0.0572, 0.0707, 0.0620, 0.0685], + device='cuda:2'), in_proj_covar=tensor([0.0278, 0.0177, 0.0125, 0.0145, 0.0179, 0.0192, 0.0198, 0.0214], + device='cuda:2'), out_proj_covar=tensor([1.6539e-04, 1.0798e-04, 7.5347e-05, 8.4724e-05, 1.0224e-04, 1.1456e-04, + 1.1366e-04, 1.2117e-04], device='cuda:2') +2023-02-05 19:07:23,139 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6117.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:07:28,996 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 19:07:29,787 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6127.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:07:45,967 INFO [train.py:901] (2/4) Epoch 1, batch 6150, loss[loss=0.3821, simple_loss=0.4162, pruned_loss=0.174, over 8441.00 frames. ], tot_loss[loss=0.4293, simple_loss=0.4435, pruned_loss=0.2075, over 1611466.19 frames. ], batch size: 27, lr: 3.97e-02, grad_scale: 8.0 +2023-02-05 19:07:47,408 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6153.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:07:56,772 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.71 vs. limit=5.0 +2023-02-05 19:08:12,166 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6188.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:22,933 INFO [train.py:901] (2/4) Epoch 1, batch 6200, loss[loss=0.4104, simple_loss=0.4425, pruned_loss=0.1892, over 8132.00 frames. ], tot_loss[loss=0.4296, simple_loss=0.4442, pruned_loss=0.2075, over 1611241.28 frames. ], batch size: 22, lr: 3.96e-02, grad_scale: 8.0 +2023-02-05 19:08:26,614 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0615, 0.9692, 4.0949, 1.8361, 3.5400, 3.3095, 3.4687, 3.4928], + device='cuda:2'), covar=tensor([0.0272, 0.3863, 0.0252, 0.1428, 0.0580, 0.0391, 0.0279, 0.0339], + device='cuda:2'), in_proj_covar=tensor([0.0124, 0.0318, 0.0160, 0.0195, 0.0189, 0.0178, 0.0151, 0.0174], + device='cuda:2'), out_proj_covar=tensor([7.6942e-05, 1.7727e-04, 1.0038e-04, 1.2605e-04, 1.0853e-04, 1.0798e-04, + 9.2981e-05, 1.1110e-04], device='cuda:2') +2023-02-05 19:08:28,567 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.743e+02 4.155e+02 5.130e+02 7.106e+02 1.864e+03, threshold=1.026e+03, percent-clipped=2.0 +2023-02-05 19:08:36,385 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:37,192 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6221.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:42,650 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6229.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:44,640 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6232.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:48,070 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6020, 1.8373, 3.3817, 1.0140, 2.6766, 2.0241, 1.5414, 2.2691], + device='cuda:2'), covar=tensor([0.1241, 0.1632, 0.0286, 0.1731, 0.1137, 0.1751, 0.1237, 0.1444], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0267, 0.0241, 0.0294, 0.0342, 0.0344, 0.0288, 0.0329], + 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') +2023-02-05 19:08:57,420 INFO [train.py:901] (2/4) Epoch 1, batch 6250, loss[loss=0.398, simple_loss=0.4146, pruned_loss=0.1907, over 7808.00 frames. ], tot_loss[loss=0.4261, simple_loss=0.4419, pruned_loss=0.2051, over 1605911.79 frames. ], batch size: 20, lr: 3.95e-02, grad_scale: 8.0 +2023-02-05 19:09:07,172 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5498, 1.8248, 2.6505, 2.5476, 1.9837, 1.4477, 1.5077, 1.9303], + device='cuda:2'), covar=tensor([0.1710, 0.0713, 0.0234, 0.0226, 0.0376, 0.0701, 0.0748, 0.0652], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0191, 0.0131, 0.0153, 0.0195, 0.0205, 0.0214, 0.0232], + device='cuda:2'), out_proj_covar=tensor([1.8059e-04, 1.1765e-04, 7.9702e-05, 8.8577e-05, 1.1087e-04, 1.2308e-04, + 1.2391e-04, 1.3132e-04], device='cuda:2') +2023-02-05 19:09:20,019 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6284.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:09:32,702 INFO [train.py:901] (2/4) Epoch 1, batch 6300, loss[loss=0.4121, simple_loss=0.4255, pruned_loss=0.1993, over 7207.00 frames. ], tot_loss[loss=0.4269, simple_loss=0.4428, pruned_loss=0.2055, over 1612128.34 frames. ], batch size: 16, lr: 3.94e-02, grad_scale: 8.0 +2023-02-05 19:09:38,777 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.783e+02 4.352e+02 5.159e+02 6.362e+02 1.735e+03, threshold=1.032e+03, percent-clipped=4.0 +2023-02-05 19:09:56,812 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6335.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:10:07,345 INFO [train.py:901] (2/4) Epoch 1, batch 6350, loss[loss=0.3744, simple_loss=0.4073, pruned_loss=0.1707, over 8030.00 frames. ], tot_loss[loss=0.4275, simple_loss=0.4429, pruned_loss=0.2061, over 1610152.93 frames. ], batch size: 22, lr: 3.93e-02, grad_scale: 8.0 +2023-02-05 19:10:08,103 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6352.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:10:28,923 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6383.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:10:40,794 INFO [train.py:901] (2/4) Epoch 1, batch 6400, loss[loss=0.3879, simple_loss=0.4034, pruned_loss=0.1862, over 7544.00 frames. ], tot_loss[loss=0.4272, simple_loss=0.4435, pruned_loss=0.2055, over 1614013.05 frames. ], batch size: 18, lr: 3.92e-02, grad_scale: 8.0 +2023-02-05 19:10:43,627 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6405.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:10:45,697 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.3543, 5.7443, 4.6109, 2.0624, 4.7005, 4.8152, 5.0302, 4.0620], + device='cuda:2'), covar=tensor([0.0874, 0.0271, 0.0723, 0.3563, 0.0393, 0.0561, 0.0723, 0.0456], + device='cuda:2'), in_proj_covar=tensor([0.0249, 0.0180, 0.0211, 0.0271, 0.0165, 0.0129, 0.0196, 0.0121], + device='cuda:2'), out_proj_covar=tensor([1.8835e-04, 1.2957e-04, 1.3808e-04, 1.7722e-04, 1.0664e-04, 9.3323e-05, + 1.3682e-04, 8.8519e-05], device='cuda:2') +2023-02-05 19:10:45,803 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6408.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:10:46,252 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.785e+02 4.017e+02 4.991e+02 6.603e+02 1.156e+03, threshold=9.981e+02, percent-clipped=3.0 +2023-02-05 19:11:09,690 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5968, 1.9536, 3.3702, 1.2112, 2.4039, 2.0034, 1.7763, 2.2025], + device='cuda:2'), covar=tensor([0.0969, 0.1190, 0.0292, 0.1289, 0.0969, 0.1467, 0.0969, 0.1054], + device='cuda:2'), in_proj_covar=tensor([0.0279, 0.0275, 0.0254, 0.0301, 0.0358, 0.0345, 0.0293, 0.0338], + 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') +2023-02-05 19:11:16,786 INFO [train.py:901] (2/4) Epoch 1, batch 6450, loss[loss=0.5176, simple_loss=0.5002, pruned_loss=0.2675, over 8100.00 frames. ], tot_loss[loss=0.4258, simple_loss=0.4426, pruned_loss=0.2045, over 1617186.22 frames. ], batch size: 23, lr: 3.91e-02, grad_scale: 8.0 +2023-02-05 19:11:27,830 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:36,487 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6480.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:39,744 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6485.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:41,921 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6488.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:47,716 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6497.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:50,292 INFO [train.py:901] (2/4) Epoch 1, batch 6500, loss[loss=0.4221, simple_loss=0.4512, pruned_loss=0.1965, over 8451.00 frames. ], tot_loss[loss=0.4227, simple_loss=0.4404, pruned_loss=0.2025, over 1616330.64 frames. ], batch size: 25, lr: 3.90e-02, grad_scale: 8.0 +2023-02-05 19:11:55,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.401e+02 4.204e+02 5.270e+02 6.161e+02 1.286e+03, threshold=1.054e+03, percent-clipped=6.0 +2023-02-05 19:11:58,504 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6513.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:03,288 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6520.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:11,141 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6532.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:25,067 INFO [train.py:901] (2/4) Epoch 1, batch 6550, loss[loss=0.3823, simple_loss=0.3997, pruned_loss=0.1824, over 7787.00 frames. ], tot_loss[loss=0.4195, simple_loss=0.4383, pruned_loss=0.2003, over 1613471.32 frames. ], batch size: 19, lr: 3.89e-02, grad_scale: 8.0 +2023-02-05 19:12:35,938 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6565.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:37,935 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 19:12:41,472 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6573.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:53,795 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6591.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:57,646 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 19:13:00,152 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-05 19:13:00,390 INFO [train.py:901] (2/4) Epoch 1, batch 6600, loss[loss=0.4031, simple_loss=0.4149, pruned_loss=0.1957, over 7543.00 frames. ], tot_loss[loss=0.4197, simple_loss=0.438, pruned_loss=0.2007, over 1613668.03 frames. ], batch size: 18, lr: 3.89e-02, grad_scale: 8.0 +2023-02-05 19:13:05,684 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 4.035e+02 4.985e+02 6.404e+02 1.328e+03, threshold=9.970e+02, percent-clipped=3.0 +2023-02-05 19:13:07,907 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6612.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:10,618 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6616.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:18,642 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6628.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:23,004 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.19 vs. limit=5.0 +2023-02-05 19:13:23,446 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4069, 2.0365, 1.6310, 1.6690, 2.0865, 1.7326, 2.1585, 2.3411], + device='cuda:2'), covar=tensor([0.1390, 0.1944, 0.2208, 0.2133, 0.1253, 0.1932, 0.1429, 0.1098], + device='cuda:2'), in_proj_covar=tensor([0.0278, 0.0289, 0.0291, 0.0284, 0.0278, 0.0262, 0.0266, 0.0262], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], + device='cuda:2') +2023-02-05 19:13:31,558 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6647.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:31,594 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6647.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:34,155 INFO [train.py:901] (2/4) Epoch 1, batch 6650, loss[loss=0.343, simple_loss=0.3679, pruned_loss=0.1591, over 5585.00 frames. ], tot_loss[loss=0.4182, simple_loss=0.437, pruned_loss=0.1997, over 1606485.04 frames. ], batch size: 12, lr: 3.88e-02, grad_scale: 8.0 +2023-02-05 19:13:56,263 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6680.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:01,297 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6688.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:09,931 INFO [train.py:901] (2/4) Epoch 1, batch 6700, loss[loss=0.4228, simple_loss=0.4247, pruned_loss=0.2105, over 7714.00 frames. ], tot_loss[loss=0.4166, simple_loss=0.4357, pruned_loss=0.1988, over 1608214.52 frames. ], batch size: 18, lr: 3.87e-02, grad_scale: 8.0 +2023-02-05 19:14:15,398 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.351e+02 4.140e+02 4.960e+02 6.260e+02 1.494e+03, threshold=9.921e+02, percent-clipped=3.0 +2023-02-05 19:14:25,029 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6723.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:38,640 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6743.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:38,875 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 19:14:42,150 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6748.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:44,021 INFO [train.py:901] (2/4) Epoch 1, batch 6750, loss[loss=0.4093, simple_loss=0.4341, pruned_loss=0.1923, over 8503.00 frames. ], tot_loss[loss=0.4143, simple_loss=0.434, pruned_loss=0.1973, over 1607737.38 frames. ], batch size: 28, lr: 3.86e-02, grad_scale: 8.0 +2023-02-05 19:15:00,943 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6776.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:14,379 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 19:15:19,964 INFO [train.py:901] (2/4) Epoch 1, batch 6800, loss[loss=0.4865, simple_loss=0.4927, pruned_loss=0.2401, over 8348.00 frames. ], tot_loss[loss=0.416, simple_loss=0.4353, pruned_loss=0.1983, over 1605933.06 frames. ], batch size: 26, lr: 3.85e-02, grad_scale: 8.0 +2023-02-05 19:15:20,161 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6801.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:25,326 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 4.226e+02 5.434e+02 7.341e+02 1.725e+03, threshold=1.087e+03, percent-clipped=4.0 +2023-02-05 19:15:35,602 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6824.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:39,016 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6829.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:54,376 INFO [train.py:901] (2/4) Epoch 1, batch 6850, loss[loss=0.4261, simple_loss=0.4409, pruned_loss=0.2057, over 8031.00 frames. ], tot_loss[loss=0.4171, simple_loss=0.4366, pruned_loss=0.1988, over 1608196.97 frames. ], batch size: 22, lr: 3.84e-02, grad_scale: 8.0 +2023-02-05 19:15:57,382 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6251, 1.1717, 3.2746, 1.3562, 1.8762, 3.5950, 3.2016, 3.1866], + device='cuda:2'), covar=tensor([0.1802, 0.2330, 0.0336, 0.2720, 0.1143, 0.0278, 0.0360, 0.0529], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0258, 0.0156, 0.0251, 0.0188, 0.0121, 0.0124, 0.0175], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 19:16:04,830 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 19:16:06,395 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6868.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:23,446 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6893.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:25,434 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6896.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:16:29,305 INFO [train.py:901] (2/4) Epoch 1, batch 6900, loss[loss=0.3463, simple_loss=0.3855, pruned_loss=0.1535, over 8087.00 frames. ], tot_loss[loss=0.4155, simple_loss=0.435, pruned_loss=0.198, over 1604986.78 frames. ], batch size: 21, lr: 3.83e-02, grad_scale: 8.0 +2023-02-05 19:16:31,390 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6903.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:35,803 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.469e+02 3.796e+02 4.754e+02 6.076e+02 1.448e+03, threshold=9.507e+02, percent-clipped=2.0 +2023-02-05 19:16:40,306 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 +2023-02-05 19:16:48,743 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6927.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:49,399 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6928.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:54,887 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6936.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:56,886 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6939.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:00,412 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6944.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:00,457 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6944.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:05,062 INFO [train.py:901] (2/4) Epoch 1, batch 6950, loss[loss=0.441, simple_loss=0.4626, pruned_loss=0.2097, over 8530.00 frames. ], tot_loss[loss=0.4168, simple_loss=0.4365, pruned_loss=0.1986, over 1611676.81 frames. ], batch size: 39, lr: 3.82e-02, grad_scale: 8.0 +2023-02-05 19:17:11,204 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-05 19:17:11,443 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0019, 1.9555, 3.3666, 1.4720, 2.5018, 2.3363, 1.9695, 2.4172], + device='cuda:2'), covar=tensor([0.0721, 0.1022, 0.0186, 0.1114, 0.0783, 0.0891, 0.0686, 0.0802], + device='cuda:2'), in_proj_covar=tensor([0.0291, 0.0293, 0.0270, 0.0319, 0.0375, 0.0345, 0.0300, 0.0354], + 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') +2023-02-05 19:17:12,145 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6961.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:17,893 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6969.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:32,944 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6991.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:38,573 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6999.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:39,833 INFO [train.py:901] (2/4) Epoch 1, batch 7000, loss[loss=0.4244, simple_loss=0.4496, pruned_loss=0.1996, over 8479.00 frames. ], tot_loss[loss=0.4146, simple_loss=0.4344, pruned_loss=0.1974, over 1610219.84 frames. ], batch size: 29, lr: 3.81e-02, grad_scale: 8.0 +2023-02-05 19:17:45,243 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.380e+02 4.090e+02 4.918e+02 6.048e+02 1.151e+03, threshold=9.836e+02, percent-clipped=6.0 +2023-02-05 19:17:57,721 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7024.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:58,275 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1610, 3.2655, 2.7782, 1.1205, 2.7265, 2.7348, 2.8754, 2.3271], + device='cuda:2'), covar=tensor([0.0970, 0.0522, 0.0967, 0.3801, 0.0549, 0.0619, 0.1029, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0258, 0.0187, 0.0216, 0.0290, 0.0168, 0.0133, 0.0202, 0.0130], + device='cuda:2'), out_proj_covar=tensor([1.9296e-04, 1.3380e-04, 1.4189e-04, 1.8658e-04, 1.0993e-04, 9.3640e-05, + 1.3958e-04, 9.3973e-05], device='cuda:2') +2023-02-05 19:18:16,023 INFO [train.py:901] (2/4) Epoch 1, batch 7050, loss[loss=0.4486, simple_loss=0.4665, pruned_loss=0.2153, over 8459.00 frames. ], tot_loss[loss=0.4154, simple_loss=0.435, pruned_loss=0.198, over 1613419.82 frames. ], batch size: 27, lr: 3.80e-02, grad_scale: 8.0 +2023-02-05 19:18:26,091 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7066.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:18:50,228 INFO [train.py:901] (2/4) Epoch 1, batch 7100, loss[loss=0.3676, simple_loss=0.4006, pruned_loss=0.1672, over 7814.00 frames. ], tot_loss[loss=0.4159, simple_loss=0.4359, pruned_loss=0.1979, over 1614697.48 frames. ], batch size: 20, lr: 3.79e-02, grad_scale: 8.0 +2023-02-05 19:18:53,881 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7106.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:18:55,757 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.243e+02 3.791e+02 4.613e+02 6.150e+02 1.722e+03, threshold=9.225e+02, percent-clipped=5.0 +2023-02-05 19:19:25,907 INFO [train.py:901] (2/4) Epoch 1, batch 7150, loss[loss=0.5054, simple_loss=0.4905, pruned_loss=0.2601, over 7826.00 frames. ], tot_loss[loss=0.4138, simple_loss=0.4345, pruned_loss=0.1966, over 1613884.61 frames. ], batch size: 20, lr: 3.78e-02, grad_scale: 8.0 +2023-02-05 19:19:46,604 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7181.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:19:56,063 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7195.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:19:59,475 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7200.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:19:59,951 INFO [train.py:901] (2/4) Epoch 1, batch 7200, loss[loss=0.3397, simple_loss=0.3864, pruned_loss=0.1465, over 8132.00 frames. ], tot_loss[loss=0.4139, simple_loss=0.4342, pruned_loss=0.1968, over 1610319.80 frames. ], batch size: 22, lr: 3.78e-02, grad_scale: 8.0 +2023-02-05 19:20:05,328 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.325e+02 4.231e+02 5.262e+02 7.053e+02 1.685e+03, threshold=1.052e+03, percent-clipped=7.0 +2023-02-05 19:20:13,056 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:20:16,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7225.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:20:24,760 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7193, 1.1832, 5.2130, 2.4781, 3.9697, 4.6037, 5.0921, 5.0195], + device='cuda:2'), covar=tensor([0.0775, 0.4518, 0.0477, 0.1257, 0.1537, 0.0475, 0.0304, 0.0510], + device='cuda:2'), in_proj_covar=tensor([0.0155, 0.0340, 0.0179, 0.0214, 0.0225, 0.0200, 0.0170, 0.0203], + device='cuda:2'), out_proj_covar=tensor([9.5622e-05, 1.8661e-04, 1.0917e-04, 1.3572e-04, 1.2867e-04, 1.1939e-04, + 1.0304e-04, 1.2815e-04], device='cuda:2') +2023-02-05 19:20:25,922 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7240.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:20:33,003 INFO [train.py:901] (2/4) Epoch 1, batch 7250, loss[loss=0.4045, simple_loss=0.4287, pruned_loss=0.1902, over 7656.00 frames. ], tot_loss[loss=0.4149, simple_loss=0.4345, pruned_loss=0.1977, over 1612109.00 frames. ], batch size: 19, lr: 3.77e-02, grad_scale: 8.0 +2023-02-05 19:20:33,214 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4886, 1.9760, 1.1139, 1.9897, 1.7721, 1.4535, 1.5258, 2.1967], + device='cuda:2'), covar=tensor([0.1421, 0.0744, 0.1523, 0.0709, 0.1146, 0.1108, 0.1608, 0.0801], + device='cuda:2'), in_proj_covar=tensor([0.0334, 0.0234, 0.0342, 0.0274, 0.0319, 0.0276, 0.0330, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 19:20:49,155 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7271.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:21:03,983 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7293.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:21:09,003 INFO [train.py:901] (2/4) Epoch 1, batch 7300, loss[loss=0.3889, simple_loss=0.4094, pruned_loss=0.1842, over 7669.00 frames. ], tot_loss[loss=0.4151, simple_loss=0.4347, pruned_loss=0.1977, over 1611192.11 frames. ], batch size: 19, lr: 3.76e-02, grad_scale: 8.0 +2023-02-05 19:21:14,304 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.413e+02 4.263e+02 5.448e+02 6.514e+02 1.215e+03, threshold=1.090e+03, percent-clipped=2.0 +2023-02-05 19:21:42,628 INFO [train.py:901] (2/4) Epoch 1, batch 7350, loss[loss=0.3589, simple_loss=0.3777, pruned_loss=0.1701, over 7791.00 frames. ], tot_loss[loss=0.4152, simple_loss=0.4346, pruned_loss=0.1979, over 1611213.70 frames. ], batch size: 19, lr: 3.75e-02, grad_scale: 8.0 +2023-02-05 19:21:45,537 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7355.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:21:50,110 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7362.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:21:56,016 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-05 19:22:03,009 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1531, 1.6142, 1.0406, 1.8403, 1.3357, 1.0706, 1.0615, 1.9658], + device='cuda:2'), covar=tensor([0.1290, 0.0859, 0.2258, 0.0741, 0.1885, 0.1668, 0.1817, 0.0745], + device='cuda:2'), in_proj_covar=tensor([0.0342, 0.0246, 0.0367, 0.0286, 0.0333, 0.0289, 0.0349, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 19:22:08,387 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9069, 3.0949, 2.6061, 1.0909, 2.5008, 2.6621, 2.7648, 2.1475], + device='cuda:2'), covar=tensor([0.1341, 0.0645, 0.1128, 0.4240, 0.0728, 0.0671, 0.1182, 0.0791], + device='cuda:2'), in_proj_covar=tensor([0.0265, 0.0185, 0.0230, 0.0294, 0.0177, 0.0140, 0.0213, 0.0134], + device='cuda:2'), out_proj_covar=tensor([1.9670e-04, 1.3098e-04, 1.5045e-04, 1.8912e-04, 1.1403e-04, 1.0044e-04, + 1.4600e-04, 9.3769e-05], device='cuda:2') +2023-02-05 19:22:08,467 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7386.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:22:09,133 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7387.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:22:18,186 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-05 19:22:18,974 INFO [train.py:901] (2/4) Epoch 1, batch 7400, loss[loss=0.3591, simple_loss=0.3925, pruned_loss=0.1628, over 7649.00 frames. ], tot_loss[loss=0.4134, simple_loss=0.4333, pruned_loss=0.1967, over 1609237.52 frames. ], batch size: 19, lr: 3.74e-02, grad_scale: 8.0 +2023-02-05 19:22:24,408 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.824e+02 4.270e+02 5.603e+02 6.704e+02 2.452e+03, threshold=1.121e+03, percent-clipped=4.0 +2023-02-05 19:22:25,141 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7410.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:22:35,071 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7425.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:22:52,554 INFO [train.py:901] (2/4) Epoch 1, batch 7450, loss[loss=0.4557, simple_loss=0.4646, pruned_loss=0.2233, over 8243.00 frames. ], tot_loss[loss=0.4128, simple_loss=0.4332, pruned_loss=0.1962, over 1611261.04 frames. ], batch size: 22, lr: 3.73e-02, grad_scale: 8.0 +2023-02-05 19:22:53,051 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-05 19:22:56,078 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-05 19:23:27,513 INFO [train.py:901] (2/4) Epoch 1, batch 7500, loss[loss=0.4377, simple_loss=0.4695, pruned_loss=0.203, over 8458.00 frames. ], tot_loss[loss=0.4148, simple_loss=0.4347, pruned_loss=0.1975, over 1612780.12 frames. ], batch size: 29, lr: 3.72e-02, grad_scale: 8.0 +2023-02-05 19:23:34,187 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.398e+02 4.060e+02 5.044e+02 6.934e+02 1.457e+03, threshold=1.009e+03, percent-clipped=3.0 +2023-02-05 19:23:45,044 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7525.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:23:45,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7525.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:24:02,105 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2023-02-05 19:24:02,230 INFO [train.py:901] (2/4) Epoch 1, batch 7550, loss[loss=0.3303, simple_loss=0.3557, pruned_loss=0.1525, over 7695.00 frames. ], tot_loss[loss=0.4141, simple_loss=0.434, pruned_loss=0.1971, over 1613594.58 frames. ], batch size: 18, lr: 3.72e-02, grad_scale: 8.0 +2023-02-05 19:24:36,293 INFO [train.py:901] (2/4) Epoch 1, batch 7600, loss[loss=0.3435, simple_loss=0.3643, pruned_loss=0.1614, over 7923.00 frames. ], tot_loss[loss=0.4132, simple_loss=0.4328, pruned_loss=0.1968, over 1608464.32 frames. ], batch size: 20, lr: 3.71e-02, grad_scale: 8.0 +2023-02-05 19:24:41,738 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.765e+02 4.361e+02 5.460e+02 6.853e+02 1.164e+03, threshold=1.092e+03, percent-clipped=2.0 +2023-02-05 19:24:43,938 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7611.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:25:03,352 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7636.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:25:03,853 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7637.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:05,958 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7640.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:06,827 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=5.04 vs. limit=5.0 +2023-02-05 19:25:07,273 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7642.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:07,351 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7642.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:13,187 INFO [train.py:901] (2/4) Epoch 1, batch 7650, loss[loss=0.3755, simple_loss=0.4165, pruned_loss=0.1672, over 8327.00 frames. ], tot_loss[loss=0.4108, simple_loss=0.4315, pruned_loss=0.195, over 1607552.57 frames. ], batch size: 25, lr: 3.70e-02, grad_scale: 8.0 +2023-02-05 19:25:23,893 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7667.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:46,144 INFO [train.py:901] (2/4) Epoch 1, batch 7700, loss[loss=0.5156, simple_loss=0.5122, pruned_loss=0.2595, over 8348.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.4308, pruned_loss=0.1937, over 1613273.23 frames. ], batch size: 26, lr: 3.69e-02, grad_scale: 8.0 +2023-02-05 19:25:51,307 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.517e+02 4.083e+02 4.742e+02 6.161e+02 2.101e+03, threshold=9.483e+02, percent-clipped=6.0 +2023-02-05 19:26:07,592 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-05 19:26:14,379 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4737, 1.8799, 3.2077, 1.1496, 2.1277, 1.9325, 1.4136, 2.2233], + device='cuda:2'), covar=tensor([0.1086, 0.1294, 0.0274, 0.1493, 0.1059, 0.1427, 0.1042, 0.1056], + device='cuda:2'), in_proj_covar=tensor([0.0322, 0.0318, 0.0306, 0.0340, 0.0399, 0.0369, 0.0318, 0.0386], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 19:26:21,659 INFO [train.py:901] (2/4) Epoch 1, batch 7750, loss[loss=0.418, simple_loss=0.4553, pruned_loss=0.1904, over 8358.00 frames. ], tot_loss[loss=0.4088, simple_loss=0.4307, pruned_loss=0.1935, over 1612770.67 frames. ], batch size: 24, lr: 3.68e-02, grad_scale: 8.0 +2023-02-05 19:26:23,165 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7752.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:26:29,120 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7761.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:26:34,416 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7769.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:26:42,677 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7781.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:26:56,367 INFO [train.py:901] (2/4) Epoch 1, batch 7800, loss[loss=0.428, simple_loss=0.4527, pruned_loss=0.2016, over 8528.00 frames. ], tot_loss[loss=0.4072, simple_loss=0.4299, pruned_loss=0.1923, over 1615872.58 frames. ], batch size: 28, lr: 3.67e-02, grad_scale: 8.0 +2023-02-05 19:26:59,825 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7806.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:27:01,643 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.208e+02 3.720e+02 4.585e+02 5.523e+02 1.290e+03, threshold=9.170e+02, percent-clipped=3.0 +2023-02-05 19:27:07,211 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9586, 3.0596, 1.8203, 2.4712, 2.7740, 1.8730, 2.1206, 2.9091], + device='cuda:2'), covar=tensor([0.1606, 0.0764, 0.1175, 0.1065, 0.1061, 0.1301, 0.1586, 0.0806], + device='cuda:2'), in_proj_covar=tensor([0.0326, 0.0231, 0.0346, 0.0274, 0.0319, 0.0281, 0.0326, 0.0269], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], + device='cuda:2') +2023-02-05 19:27:09,238 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7820.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:27:11,967 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8347, 2.0160, 1.9511, 2.6324, 1.4457, 1.1985, 1.8922, 1.9493], + device='cuda:2'), covar=tensor([0.1221, 0.1361, 0.1255, 0.0397, 0.1829, 0.2050, 0.1672, 0.1155], + device='cuda:2'), in_proj_covar=tensor([0.0300, 0.0329, 0.0312, 0.0201, 0.0343, 0.0332, 0.0393, 0.0291], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 19:27:29,719 INFO [train.py:901] (2/4) Epoch 1, batch 7850, loss[loss=0.4551, simple_loss=0.4745, pruned_loss=0.2179, over 8626.00 frames. ], tot_loss[loss=0.4069, simple_loss=0.4295, pruned_loss=0.1922, over 1617868.75 frames. ], batch size: 34, lr: 3.66e-02, grad_scale: 8.0 +2023-02-05 19:27:52,023 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7884.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:27:59,879 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7896.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:28:03,018 INFO [train.py:901] (2/4) Epoch 1, batch 7900, loss[loss=0.4539, simple_loss=0.461, pruned_loss=0.2234, over 6642.00 frames. ], tot_loss[loss=0.406, simple_loss=0.4286, pruned_loss=0.1916, over 1612900.23 frames. ], batch size: 72, lr: 3.66e-02, grad_scale: 8.0 +2023-02-05 19:28:08,433 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 3.732e+02 4.923e+02 6.190e+02 1.863e+03, threshold=9.845e+02, percent-clipped=5.0 +2023-02-05 19:28:16,374 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7921.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:28:16,631 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-05 19:28:35,812 INFO [train.py:901] (2/4) Epoch 1, batch 7950, loss[loss=0.4412, simple_loss=0.4711, pruned_loss=0.2056, over 8339.00 frames. ], tot_loss[loss=0.4074, simple_loss=0.4299, pruned_loss=0.1924, over 1613344.87 frames. ], batch size: 25, lr: 3.65e-02, grad_scale: 8.0 +2023-02-05 19:28:45,404 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-05 19:28:59,119 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7986.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:29:10,062 INFO [train.py:901] (2/4) Epoch 1, batch 8000, loss[loss=0.5517, simple_loss=0.5169, pruned_loss=0.2933, over 6615.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4284, pruned_loss=0.1913, over 1610131.70 frames. ], batch size: 71, lr: 3.64e-02, grad_scale: 8.0 +2023-02-05 19:29:15,098 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8008.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:29:15,548 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.650e+02 3.959e+02 4.934e+02 6.403e+02 1.426e+03, threshold=9.868e+02, percent-clipped=4.0 +2023-02-05 19:29:29,569 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4159, 2.0298, 1.2013, 2.0030, 1.9489, 1.2772, 1.7432, 2.4443], + device='cuda:2'), covar=tensor([0.1454, 0.0649, 0.1508, 0.0907, 0.1067, 0.1434, 0.1310, 0.0793], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0227, 0.0356, 0.0288, 0.0334, 0.0304, 0.0342, 0.0287], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 19:29:31,430 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8033.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:29:43,012 INFO [train.py:901] (2/4) Epoch 1, batch 8050, loss[loss=0.4386, simple_loss=0.4441, pruned_loss=0.2165, over 6831.00 frames. ], tot_loss[loss=0.4064, simple_loss=0.4284, pruned_loss=0.1922, over 1596063.85 frames. ], batch size: 71, lr: 3.63e-02, grad_scale: 16.0 +2023-02-05 19:29:53,046 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0026, 1.8516, 3.1078, 1.1702, 1.8527, 2.3540, 0.5054, 1.5992], + device='cuda:2'), covar=tensor([0.0667, 0.0575, 0.0275, 0.0399, 0.0435, 0.0317, 0.1497, 0.0642], + device='cuda:2'), in_proj_covar=tensor([0.0142, 0.0114, 0.0095, 0.0133, 0.0115, 0.0083, 0.0163, 0.0132], + device='cuda:2'), out_proj_covar=tensor([1.1974e-04, 1.0761e-04, 8.3139e-05, 1.1257e-04, 1.0679e-04, 7.3238e-05, + 1.4342e-04, 1.1716e-04], device='cuda:2') +2023-02-05 19:30:16,951 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 19:30:20,852 INFO [train.py:901] (2/4) Epoch 2, batch 0, loss[loss=0.3829, simple_loss=0.4024, pruned_loss=0.1816, over 7713.00 frames. ], tot_loss[loss=0.3829, simple_loss=0.4024, pruned_loss=0.1816, over 7713.00 frames. ], batch size: 18, lr: 3.56e-02, grad_scale: 8.0 +2023-02-05 19:30:20,853 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 19:30:32,397 INFO [train.py:935] (2/4) Epoch 2, validation: loss=0.3107, simple_loss=0.3861, pruned_loss=0.1176, over 944034.00 frames. +2023-02-05 19:30:32,398 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 19:30:44,062 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8101.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:30:46,623 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 19:30:46,681 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8105.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:30:49,928 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.305e+02 3.846e+02 4.676e+02 6.027e+02 1.450e+03, threshold=9.352e+02, percent-clipped=5.0 +2023-02-05 19:30:56,768 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7788, 1.8838, 1.8499, 2.5534, 0.9706, 1.0852, 1.6427, 1.8515], + device='cuda:2'), covar=tensor([0.1222, 0.1525, 0.1394, 0.0376, 0.2416, 0.2550, 0.2023, 0.1308], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0323, 0.0300, 0.0196, 0.0331, 0.0343, 0.0380, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0004, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 19:31:06,753 INFO [train.py:901] (2/4) Epoch 2, batch 50, loss[loss=0.4761, simple_loss=0.4615, pruned_loss=0.2453, over 8240.00 frames. ], tot_loss[loss=0.4027, simple_loss=0.4274, pruned_loss=0.189, over 364121.76 frames. ], batch size: 22, lr: 3.55e-02, grad_scale: 8.0 +2023-02-05 19:31:11,116 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8140.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:31:20,787 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 19:31:28,341 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8164.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:31:29,177 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8165.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:31:41,604 INFO [train.py:901] (2/4) Epoch 2, batch 100, loss[loss=0.4268, simple_loss=0.4365, pruned_loss=0.2086, over 8035.00 frames. ], tot_loss[loss=0.406, simple_loss=0.4302, pruned_loss=0.1909, over 644360.20 frames. ], batch size: 22, lr: 3.54e-02, grad_scale: 8.0 +2023-02-05 19:31:44,283 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 19:31:59,434 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.305e+02 4.246e+02 4.943e+02 6.491e+02 9.375e+02, threshold=9.885e+02, percent-clipped=1.0 +2023-02-05 19:32:06,342 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:32:07,758 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5228, 1.9354, 3.0883, 0.8696, 2.2812, 1.7347, 1.4636, 2.0157], + device='cuda:2'), covar=tensor([0.1199, 0.1271, 0.0401, 0.1931, 0.1036, 0.1752, 0.1051, 0.1401], + device='cuda:2'), in_proj_covar=tensor([0.0324, 0.0316, 0.0312, 0.0356, 0.0403, 0.0372, 0.0324, 0.0388], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 19:32:15,464 INFO [train.py:901] (2/4) Epoch 2, batch 150, loss[loss=0.3343, simple_loss=0.3745, pruned_loss=0.147, over 7693.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4306, pruned_loss=0.1902, over 863622.01 frames. ], batch size: 18, lr: 3.53e-02, grad_scale: 8.0 +2023-02-05 19:32:47,787 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8279.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:32:50,393 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8283.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:32:50,923 INFO [train.py:901] (2/4) Epoch 2, batch 200, loss[loss=0.3713, simple_loss=0.4049, pruned_loss=0.1689, over 8468.00 frames. ], tot_loss[loss=0.4022, simple_loss=0.4276, pruned_loss=0.1884, over 1029688.39 frames. ], batch size: 25, lr: 3.52e-02, grad_scale: 8.0 +2023-02-05 19:32:53,835 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 +2023-02-05 19:33:08,598 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.581e+02 3.727e+02 4.975e+02 6.903e+02 1.681e+03, threshold=9.950e+02, percent-clipped=7.0 +2023-02-05 19:33:24,845 INFO [train.py:901] (2/4) Epoch 2, batch 250, loss[loss=0.4005, simple_loss=0.4331, pruned_loss=0.184, over 7980.00 frames. ], tot_loss[loss=0.4011, simple_loss=0.427, pruned_loss=0.1877, over 1161484.74 frames. ], batch size: 21, lr: 3.52e-02, grad_scale: 8.0 +2023-02-05 19:33:36,306 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 19:33:40,657 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8357.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:33:45,997 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 19:33:58,466 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8382.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:33:59,607 INFO [train.py:901] (2/4) Epoch 2, batch 300, loss[loss=0.3802, simple_loss=0.4375, pruned_loss=0.1615, over 8251.00 frames. ], tot_loss[loss=0.4027, simple_loss=0.4283, pruned_loss=0.1885, over 1264525.68 frames. ], batch size: 24, lr: 3.51e-02, grad_scale: 8.0 +2023-02-05 19:34:18,662 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 4.043e+02 4.737e+02 5.583e+02 9.957e+02, threshold=9.474e+02, percent-clipped=1.0 +2023-02-05 19:34:19,482 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0731, 2.7201, 4.9477, 1.0727, 2.8216, 2.4383, 2.0344, 2.3323], + device='cuda:2'), covar=tensor([0.1038, 0.1161, 0.0183, 0.1633, 0.1063, 0.1409, 0.0921, 0.1528], + device='cuda:2'), in_proj_covar=tensor([0.0331, 0.0308, 0.0310, 0.0350, 0.0402, 0.0370, 0.0324, 0.0390], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 19:34:35,500 INFO [train.py:901] (2/4) Epoch 2, batch 350, loss[loss=0.3169, simple_loss=0.3545, pruned_loss=0.1397, over 7706.00 frames. ], tot_loss[loss=0.4006, simple_loss=0.4269, pruned_loss=0.1871, over 1343388.97 frames. ], batch size: 18, lr: 3.50e-02, grad_scale: 8.0 +2023-02-05 19:34:36,952 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.5898, 1.0330, 3.7178, 1.4639, 3.0583, 3.1067, 3.1760, 3.2028], + device='cuda:2'), covar=tensor([0.0402, 0.3525, 0.0319, 0.1564, 0.1028, 0.0440, 0.0364, 0.0494], + device='cuda:2'), in_proj_covar=tensor([0.0166, 0.0349, 0.0193, 0.0220, 0.0252, 0.0212, 0.0177, 0.0222], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 19:34:42,215 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2999, 1.3207, 1.5424, 1.1102, 0.9581, 1.4457, 0.1561, 0.9121], + device='cuda:2'), covar=tensor([0.0575, 0.0530, 0.0266, 0.0372, 0.0507, 0.0276, 0.1465, 0.0647], + device='cuda:2'), in_proj_covar=tensor([0.0143, 0.0119, 0.0102, 0.0145, 0.0118, 0.0089, 0.0171, 0.0142], + device='cuda:2'), out_proj_covar=tensor([1.2182e-04, 1.1227e-04, 9.1913e-05, 1.2420e-04, 1.1222e-04, 8.1590e-05, + 1.4938e-04, 1.2798e-04], device='cuda:2') +2023-02-05 19:35:03,563 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8476.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:35:09,447 INFO [train.py:901] (2/4) Epoch 2, batch 400, loss[loss=0.377, simple_loss=0.4055, pruned_loss=0.1742, over 8140.00 frames. ], tot_loss[loss=0.4001, simple_loss=0.4263, pruned_loss=0.187, over 1405704.73 frames. ], batch size: 22, lr: 3.49e-02, grad_scale: 8.0 +2023-02-05 19:35:13,313 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-05 19:35:20,909 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8501.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:35:27,438 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.659e+02 4.339e+02 4.887e+02 6.099e+02 1.134e+03, threshold=9.773e+02, percent-clipped=6.0 +2023-02-05 19:35:43,481 INFO [train.py:901] (2/4) Epoch 2, batch 450, loss[loss=0.3456, simple_loss=0.3932, pruned_loss=0.149, over 8099.00 frames. ], tot_loss[loss=0.401, simple_loss=0.4268, pruned_loss=0.1877, over 1451341.83 frames. ], batch size: 23, lr: 3.49e-02, grad_scale: 8.0 +2023-02-05 19:35:44,344 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8535.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:36:01,873 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8560.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:36:13,318 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4780, 1.4363, 3.1697, 1.5868, 2.2987, 3.6588, 3.3380, 3.3107], + device='cuda:2'), covar=tensor([0.1639, 0.1886, 0.0351, 0.2057, 0.0714, 0.0202, 0.0243, 0.0395], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0260, 0.0170, 0.0253, 0.0184, 0.0138, 0.0131, 0.0192], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 19:36:18,002 INFO [train.py:901] (2/4) Epoch 2, batch 500, loss[loss=0.3554, simple_loss=0.3865, pruned_loss=0.1622, over 7635.00 frames. ], tot_loss[loss=0.3984, simple_loss=0.4249, pruned_loss=0.186, over 1488710.60 frames. ], batch size: 19, lr: 3.48e-02, grad_scale: 8.0 +2023-02-05 19:36:26,943 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.3502, 5.6069, 4.5388, 1.7783, 4.6647, 4.8734, 5.0307, 4.2619], + device='cuda:2'), covar=tensor([0.0652, 0.0313, 0.0713, 0.3531, 0.0285, 0.0290, 0.0959, 0.0348], + device='cuda:2'), in_proj_covar=tensor([0.0280, 0.0185, 0.0229, 0.0303, 0.0193, 0.0144, 0.0204, 0.0137], + device='cuda:2'), out_proj_covar=tensor([2.0096e-04, 1.2857e-04, 1.4659e-04, 1.9091e-04, 1.2387e-04, 1.0357e-04, + 1.3701e-04, 9.6944e-05], device='cuda:2') +2023-02-05 19:36:36,157 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.361e+02 3.910e+02 4.803e+02 5.619e+02 9.699e+02, threshold=9.605e+02, percent-clipped=0.0 +2023-02-05 19:36:47,558 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8627.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:36:52,675 INFO [train.py:901] (2/4) Epoch 2, batch 550, loss[loss=0.434, simple_loss=0.4582, pruned_loss=0.2049, over 8324.00 frames. ], tot_loss[loss=0.3988, simple_loss=0.4255, pruned_loss=0.1861, over 1517622.11 frames. ], batch size: 25, lr: 3.47e-02, grad_scale: 8.0 +2023-02-05 19:37:24,170 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8084, 2.2485, 1.8078, 2.8343, 1.3320, 1.0781, 1.7886, 2.0606], + device='cuda:2'), covar=tensor([0.1066, 0.1105, 0.1144, 0.0265, 0.1882, 0.2316, 0.1978, 0.1274], + device='cuda:2'), in_proj_covar=tensor([0.0306, 0.0329, 0.0317, 0.0209, 0.0343, 0.0352, 0.0395, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:2') +2023-02-05 19:37:26,524 INFO [train.py:901] (2/4) Epoch 2, batch 600, loss[loss=0.4575, simple_loss=0.4519, pruned_loss=0.2315, over 8074.00 frames. ], tot_loss[loss=0.3996, simple_loss=0.4256, pruned_loss=0.1868, over 1537610.90 frames. ], batch size: 21, lr: 3.46e-02, grad_scale: 8.0 +2023-02-05 19:37:43,320 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.752e+02 3.934e+02 5.073e+02 6.758e+02 1.500e+03, threshold=1.015e+03, percent-clipped=5.0 +2023-02-05 19:37:44,752 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 19:37:59,735 INFO [train.py:901] (2/4) Epoch 2, batch 650, loss[loss=0.4907, simple_loss=0.4715, pruned_loss=0.255, over 8533.00 frames. ], tot_loss[loss=0.3983, simple_loss=0.4247, pruned_loss=0.186, over 1555707.93 frames. ], batch size: 49, lr: 3.46e-02, grad_scale: 8.0 +2023-02-05 19:38:05,378 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8742.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:38:31,196 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8778.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:38:32,083 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2023-02-05 19:38:35,550 INFO [train.py:901] (2/4) Epoch 2, batch 700, loss[loss=0.4477, simple_loss=0.4539, pruned_loss=0.2207, over 8467.00 frames. ], tot_loss[loss=0.395, simple_loss=0.4221, pruned_loss=0.184, over 1568470.09 frames. ], batch size: 25, lr: 3.45e-02, grad_scale: 8.0 +2023-02-05 19:38:53,117 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.421e+02 3.759e+02 4.676e+02 6.060e+02 1.461e+03, threshold=9.352e+02, percent-clipped=1.0 +2023-02-05 19:39:09,177 INFO [train.py:901] (2/4) Epoch 2, batch 750, loss[loss=0.4535, simple_loss=0.4628, pruned_loss=0.2221, over 8454.00 frames. ], tot_loss[loss=0.3955, simple_loss=0.4222, pruned_loss=0.1844, over 1574651.53 frames. ], batch size: 27, lr: 3.44e-02, grad_scale: 8.0 +2023-02-05 19:39:26,413 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 19:39:35,571 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 19:39:42,526 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3476, 1.8328, 1.4828, 1.1817, 1.9969, 1.5978, 1.9084, 2.1006], + device='cuda:2'), covar=tensor([0.1217, 0.2050, 0.2370, 0.2283, 0.1148, 0.2083, 0.1453, 0.1137], + device='cuda:2'), in_proj_covar=tensor([0.0257, 0.0279, 0.0292, 0.0275, 0.0256, 0.0256, 0.0251, 0.0245], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:2') +2023-02-05 19:39:44,339 INFO [train.py:901] (2/4) Epoch 2, batch 800, loss[loss=0.3304, simple_loss=0.3648, pruned_loss=0.148, over 7437.00 frames. ], tot_loss[loss=0.3989, simple_loss=0.4243, pruned_loss=0.1867, over 1584741.25 frames. ], batch size: 17, lr: 3.43e-02, grad_scale: 8.0 +2023-02-05 19:40:02,285 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.406e+02 4.043e+02 5.225e+02 6.708e+02 1.302e+03, threshold=1.045e+03, percent-clipped=9.0 +2023-02-05 19:40:18,495 INFO [train.py:901] (2/4) Epoch 2, batch 850, loss[loss=0.3706, simple_loss=0.3947, pruned_loss=0.1732, over 7812.00 frames. ], tot_loss[loss=0.3978, simple_loss=0.4243, pruned_loss=0.1856, over 1596403.37 frames. ], batch size: 20, lr: 3.43e-02, grad_scale: 8.0 +2023-02-05 19:40:26,052 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8945.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:40:32,050 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4872, 1.9593, 1.9189, 0.5696, 1.9664, 1.3513, 0.5034, 1.3631], + device='cuda:2'), covar=tensor([0.0197, 0.0122, 0.0157, 0.0426, 0.0199, 0.0436, 0.0532, 0.0201], + device='cuda:2'), in_proj_covar=tensor([0.0161, 0.0123, 0.0105, 0.0164, 0.0119, 0.0203, 0.0171, 0.0143], + device='cuda:2'), out_proj_covar=tensor([1.1400e-04, 8.7058e-05, 8.0642e-05, 1.1866e-04, 9.2732e-05, 1.5596e-04, + 1.2641e-04, 1.0579e-04], device='cuda:2') +2023-02-05 19:40:42,766 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-05 19:40:52,653 INFO [train.py:901] (2/4) Epoch 2, batch 900, loss[loss=0.4446, simple_loss=0.4576, pruned_loss=0.2158, over 8139.00 frames. ], tot_loss[loss=0.3959, simple_loss=0.4234, pruned_loss=0.1842, over 1604380.43 frames. ], batch size: 22, lr: 3.42e-02, grad_scale: 8.0 +2023-02-05 19:41:03,939 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8998.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:08,132 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9004.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:12,012 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.317e+02 3.660e+02 4.402e+02 6.333e+02 1.420e+03, threshold=8.805e+02, percent-clipped=4.0 +2023-02-05 19:41:21,850 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9023.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:27,228 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9031.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:29,113 INFO [train.py:901] (2/4) Epoch 2, batch 950, loss[loss=0.4266, simple_loss=0.4535, pruned_loss=0.1998, over 8343.00 frames. ], tot_loss[loss=0.3949, simple_loss=0.4233, pruned_loss=0.1832, over 1610552.64 frames. ], batch size: 24, lr: 3.41e-02, grad_scale: 8.0 +2023-02-05 19:41:57,091 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 19:42:04,017 INFO [train.py:901] (2/4) Epoch 2, batch 1000, loss[loss=0.3726, simple_loss=0.3981, pruned_loss=0.1735, over 7643.00 frames. ], tot_loss[loss=0.395, simple_loss=0.4226, pruned_loss=0.1837, over 1604017.94 frames. ], batch size: 19, lr: 3.40e-02, grad_scale: 8.0 +2023-02-05 19:42:22,612 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.505e+02 3.676e+02 4.681e+02 5.718e+02 9.745e+02, threshold=9.362e+02, percent-clipped=2.0 +2023-02-05 19:42:30,647 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9122.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:42:31,277 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 19:42:39,163 INFO [train.py:901] (2/4) Epoch 2, batch 1050, loss[loss=0.3853, simple_loss=0.4131, pruned_loss=0.1788, over 7977.00 frames. ], tot_loss[loss=0.3959, simple_loss=0.4227, pruned_loss=0.1846, over 1604430.27 frames. ], batch size: 21, lr: 3.40e-02, grad_scale: 8.0 +2023-02-05 19:42:43,232 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 19:43:12,160 INFO [train.py:901] (2/4) Epoch 2, batch 1100, loss[loss=0.409, simple_loss=0.4468, pruned_loss=0.1857, over 8632.00 frames. ], tot_loss[loss=0.3945, simple_loss=0.4217, pruned_loss=0.1836, over 1610236.82 frames. ], batch size: 39, lr: 3.39e-02, grad_scale: 8.0 +2023-02-05 19:43:25,819 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.9653, 1.2069, 5.1880, 2.4207, 3.8908, 4.5145, 4.8101, 4.7724], + device='cuda:2'), covar=tensor([0.0816, 0.4456, 0.0544, 0.1393, 0.1979, 0.0445, 0.0466, 0.0674], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0353, 0.0204, 0.0240, 0.0274, 0.0219, 0.0201, 0.0236], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 19:43:30,062 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.440e+02 3.895e+02 4.986e+02 6.293e+02 1.172e+03, threshold=9.973e+02, percent-clipped=2.0 +2023-02-05 19:43:47,492 INFO [train.py:901] (2/4) Epoch 2, batch 1150, loss[loss=0.3532, simple_loss=0.3999, pruned_loss=0.1533, over 8453.00 frames. ], tot_loss[loss=0.395, simple_loss=0.4223, pruned_loss=0.1838, over 1610764.94 frames. ], batch size: 29, lr: 3.38e-02, grad_scale: 8.0 +2023-02-05 19:43:49,761 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9237.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:43:50,987 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 19:44:11,412 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2580, 4.5551, 3.9148, 1.7215, 3.7246, 3.6812, 4.1057, 3.2629], + device='cuda:2'), covar=tensor([0.0763, 0.0216, 0.0525, 0.3584, 0.0408, 0.0607, 0.0556, 0.0554], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0199, 0.0241, 0.0316, 0.0201, 0.0154, 0.0219, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 19:44:22,131 INFO [train.py:901] (2/4) Epoch 2, batch 1200, loss[loss=0.4481, simple_loss=0.4703, pruned_loss=0.213, over 8467.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.4216, pruned_loss=0.1825, over 1609362.32 frames. ], batch size: 25, lr: 3.38e-02, grad_scale: 8.0 +2023-02-05 19:44:23,917 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-05 19:44:25,538 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9289.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:44:41,020 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.810e+02 4.160e+02 4.885e+02 6.720e+02 4.965e+03, threshold=9.769e+02, percent-clipped=5.0 +2023-02-05 19:44:56,723 INFO [train.py:901] (2/4) Epoch 2, batch 1250, loss[loss=0.3875, simple_loss=0.4252, pruned_loss=0.1749, over 8096.00 frames. ], tot_loss[loss=0.3965, simple_loss=0.4235, pruned_loss=0.1847, over 1610866.07 frames. ], batch size: 23, lr: 3.37e-02, grad_scale: 4.0 +2023-02-05 19:45:07,486 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9348.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:45:25,818 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9375.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:45:31,820 INFO [train.py:901] (2/4) Epoch 2, batch 1300, loss[loss=0.3839, simple_loss=0.4253, pruned_loss=0.1713, over 8524.00 frames. ], tot_loss[loss=0.3956, simple_loss=0.423, pruned_loss=0.1841, over 1610277.76 frames. ], batch size: 31, lr: 3.36e-02, grad_scale: 4.0 +2023-02-05 19:45:45,735 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9404.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:45:50,303 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.161e+02 4.162e+02 5.656e+02 7.688e+02 2.529e+03, threshold=1.131e+03, percent-clipped=11.0 +2023-02-05 19:45:52,204 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 19:45:57,231 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0032, 1.8875, 1.3662, 1.2799, 1.9480, 1.4134, 1.3821, 2.0457], + device='cuda:2'), covar=tensor([0.1297, 0.1701, 0.2519, 0.2259, 0.1047, 0.1919, 0.1524, 0.0966], + device='cuda:2'), in_proj_covar=tensor([0.0256, 0.0272, 0.0295, 0.0271, 0.0253, 0.0250, 0.0245, 0.0234], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:2') +2023-02-05 19:46:05,050 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:06,285 INFO [train.py:901] (2/4) Epoch 2, batch 1350, loss[loss=0.4649, simple_loss=0.4786, pruned_loss=0.2256, over 8677.00 frames. ], tot_loss[loss=0.3953, simple_loss=0.4227, pruned_loss=0.184, over 1612450.98 frames. ], batch size: 34, lr: 3.36e-02, grad_scale: 4.0 +2023-02-05 19:46:09,322 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2023-02-05 19:46:27,277 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9463.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:27,301 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6587, 2.1999, 1.1792, 1.9080, 1.7596, 1.3007, 1.3433, 2.0329], + device='cuda:2'), covar=tensor([0.1487, 0.0589, 0.1483, 0.1036, 0.1196, 0.1461, 0.1732, 0.1018], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0241, 0.0352, 0.0295, 0.0341, 0.0307, 0.0359, 0.0301], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 19:46:41,365 INFO [train.py:901] (2/4) Epoch 2, batch 1400, loss[loss=0.4184, simple_loss=0.4626, pruned_loss=0.1871, over 8322.00 frames. ], tot_loss[loss=0.3953, simple_loss=0.4228, pruned_loss=0.1839, over 1613584.35 frames. ], batch size: 25, lr: 3.35e-02, grad_scale: 4.0 +2023-02-05 19:46:45,510 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9490.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:47,576 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9493.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:59,486 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.192e+02 3.889e+02 4.981e+02 6.326e+02 1.555e+03, threshold=9.962e+02, percent-clipped=1.0 +2023-02-05 19:47:04,255 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9518.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:47:06,512 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-05 19:47:15,069 INFO [train.py:901] (2/4) Epoch 2, batch 1450, loss[loss=0.3953, simple_loss=0.4335, pruned_loss=0.1786, over 8114.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4222, pruned_loss=0.1841, over 1613202.01 frames. ], batch size: 23, lr: 3.34e-02, grad_scale: 4.0 +2023-02-05 19:47:19,042 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 19:47:49,261 INFO [train.py:901] (2/4) Epoch 2, batch 1500, loss[loss=0.3635, simple_loss=0.4058, pruned_loss=0.1607, over 8240.00 frames. ], tot_loss[loss=0.3915, simple_loss=0.42, pruned_loss=0.1814, over 1615710.51 frames. ], batch size: 22, lr: 3.33e-02, grad_scale: 4.0 +2023-02-05 19:48:01,346 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9602.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:48:07,902 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.496e+02 4.006e+02 4.905e+02 6.157e+02 1.300e+03, threshold=9.811e+02, percent-clipped=3.0 +2023-02-05 19:48:23,387 INFO [train.py:901] (2/4) Epoch 2, batch 1550, loss[loss=0.3511, simple_loss=0.394, pruned_loss=0.1541, over 7808.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4192, pruned_loss=0.1803, over 1619092.97 frames. ], batch size: 20, lr: 3.33e-02, grad_scale: 4.0 +2023-02-05 19:48:41,696 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9660.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:48:45,214 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-05 19:48:50,894 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 +2023-02-05 19:48:52,341 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9676.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:48:57,567 INFO [train.py:901] (2/4) Epoch 2, batch 1600, loss[loss=0.3373, simple_loss=0.3871, pruned_loss=0.1438, over 8032.00 frames. ], tot_loss[loss=0.392, simple_loss=0.4208, pruned_loss=0.1816, over 1613209.07 frames. ], batch size: 22, lr: 3.32e-02, grad_scale: 8.0 +2023-02-05 19:48:58,392 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9685.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:17,085 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.659e+02 4.192e+02 5.177e+02 6.492e+02 1.266e+03, threshold=1.035e+03, percent-clipped=2.0 +2023-02-05 19:49:22,888 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:33,627 INFO [train.py:901] (2/4) Epoch 2, batch 1650, loss[loss=0.3685, simple_loss=0.403, pruned_loss=0.167, over 8075.00 frames. ], tot_loss[loss=0.3886, simple_loss=0.4184, pruned_loss=0.1794, over 1614202.70 frames. ], batch size: 21, lr: 3.31e-02, grad_scale: 8.0 +2023-02-05 19:49:40,254 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9744.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:41,550 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9746.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:47,790 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-05 19:49:51,527 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9761.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:58,956 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9771.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:50:02,237 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9776.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:50:03,227 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.17 vs. limit=2.0 +2023-02-05 19:50:07,343 INFO [train.py:901] (2/4) Epoch 2, batch 1700, loss[loss=0.4222, simple_loss=0.4528, pruned_loss=0.1958, over 8502.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4197, pruned_loss=0.1805, over 1616226.92 frames. ], batch size: 28, lr: 3.31e-02, grad_scale: 8.0 +2023-02-05 19:50:24,477 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2023, 2.1911, 1.9616, 0.5015, 1.9098, 1.4238, 0.3970, 2.0077], + device='cuda:2'), covar=tensor([0.0239, 0.0068, 0.0145, 0.0335, 0.0145, 0.0279, 0.0369, 0.0091], + device='cuda:2'), in_proj_covar=tensor([0.0168, 0.0123, 0.0110, 0.0173, 0.0121, 0.0207, 0.0175, 0.0143], + device='cuda:2'), out_proj_covar=tensor([1.1481e-04, 8.5389e-05, 8.3123e-05, 1.2257e-04, 9.1607e-05, 1.5285e-04, + 1.2569e-04, 1.0142e-04], device='cuda:2') +2023-02-05 19:50:26,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.640e+02 4.068e+02 5.098e+02 6.535e+02 1.207e+03, threshold=1.020e+03, percent-clipped=5.0 +2023-02-05 19:50:42,243 INFO [train.py:901] (2/4) Epoch 2, batch 1750, loss[loss=0.4101, simple_loss=0.4425, pruned_loss=0.1888, over 8322.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4189, pruned_loss=0.1798, over 1617709.55 frames. ], batch size: 25, lr: 3.30e-02, grad_scale: 8.0 +2023-02-05 19:50:44,397 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8239, 2.2589, 4.5746, 1.2491, 3.0334, 2.2529, 1.9116, 2.3507], + device='cuda:2'), covar=tensor([0.1044, 0.1372, 0.0241, 0.1672, 0.0947, 0.1522, 0.0897, 0.1566], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0337, 0.0347, 0.0382, 0.0430, 0.0403, 0.0343, 0.0423], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 19:51:16,319 INFO [train.py:901] (2/4) Epoch 2, batch 1800, loss[loss=0.4002, simple_loss=0.4161, pruned_loss=0.1922, over 8326.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.4194, pruned_loss=0.1804, over 1622942.53 frames. ], batch size: 25, lr: 3.29e-02, grad_scale: 8.0 +2023-02-05 19:51:21,270 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9891.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:51:34,083 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.365e+02 4.111e+02 5.198e+02 6.626e+02 1.120e+03, threshold=1.040e+03, percent-clipped=3.0 +2023-02-05 19:51:49,954 INFO [train.py:901] (2/4) Epoch 2, batch 1850, loss[loss=0.4157, simple_loss=0.4438, pruned_loss=0.1938, over 8641.00 frames. ], tot_loss[loss=0.3874, simple_loss=0.4174, pruned_loss=0.1787, over 1619014.06 frames. ], batch size: 49, lr: 3.29e-02, grad_scale: 8.0 +2023-02-05 19:51:58,649 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9946.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:52:01,473 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0558, 1.7301, 1.3944, 1.1372, 1.9990, 1.5886, 1.5625, 2.0100], + device='cuda:2'), covar=tensor([0.1162, 0.1781, 0.2504, 0.2153, 0.0870, 0.1890, 0.1285, 0.0840], + device='cuda:2'), in_proj_covar=tensor([0.0247, 0.0267, 0.0290, 0.0258, 0.0240, 0.0250, 0.0234, 0.0227], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:2') +2023-02-05 19:52:08,259 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-05 19:52:13,469 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3849, 1.5459, 2.2579, 0.9768, 1.8512, 1.5340, 1.3488, 1.5809], + device='cuda:2'), covar=tensor([0.1039, 0.1052, 0.0440, 0.1648, 0.0871, 0.1569, 0.0971, 0.1007], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0341, 0.0359, 0.0385, 0.0437, 0.0400, 0.0346, 0.0427], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 19:52:24,359 INFO [train.py:901] (2/4) Epoch 2, batch 1900, loss[loss=0.4002, simple_loss=0.4348, pruned_loss=0.1828, over 8449.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4158, pruned_loss=0.177, over 1617412.44 frames. ], batch size: 29, lr: 3.28e-02, grad_scale: 8.0 +2023-02-05 19:52:36,761 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3960, 1.4470, 1.2303, 1.4954, 1.3553, 1.1686, 1.1382, 1.7996], + device='cuda:2'), covar=tensor([0.0949, 0.0591, 0.1300, 0.0664, 0.0974, 0.1248, 0.1111, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0369, 0.0250, 0.0370, 0.0298, 0.0354, 0.0314, 0.0369, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 19:52:39,962 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4050, 0.9354, 4.5199, 2.0111, 3.8878, 3.6244, 4.0105, 4.0155], + device='cuda:2'), covar=tensor([0.0326, 0.3613, 0.0217, 0.1390, 0.0805, 0.0334, 0.0279, 0.0338], + device='cuda:2'), in_proj_covar=tensor([0.0179, 0.0361, 0.0209, 0.0252, 0.0285, 0.0235, 0.0202, 0.0240], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 19:52:43,800 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 3.513e+02 4.327e+02 5.785e+02 1.080e+03, threshold=8.653e+02, percent-clipped=1.0 +2023-02-05 19:52:49,917 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10020.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:52:54,607 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 19:52:57,444 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9163, 2.2817, 1.8806, 2.7669, 1.1635, 1.3314, 1.8089, 2.2439], + device='cuda:2'), covar=tensor([0.1169, 0.1506, 0.1672, 0.0385, 0.2505, 0.2446, 0.2198, 0.1402], + device='cuda:2'), in_proj_covar=tensor([0.0307, 0.0326, 0.0319, 0.0221, 0.0334, 0.0349, 0.0383, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:2') +2023-02-05 19:52:59,199 INFO [train.py:901] (2/4) Epoch 2, batch 1950, loss[loss=0.4023, simple_loss=0.433, pruned_loss=0.1858, over 8342.00 frames. ], tot_loss[loss=0.3885, simple_loss=0.4183, pruned_loss=0.1794, over 1617293.36 frames. ], batch size: 26, lr: 3.27e-02, grad_scale: 8.0 +2023-02-05 19:53:06,844 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 19:53:15,911 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10057.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:53:19,396 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:53:25,587 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 19:53:33,042 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10080.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:53:33,717 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4033, 2.0820, 3.3599, 3.1256, 2.8009, 2.0168, 1.5446, 1.9102], + device='cuda:2'), covar=tensor([0.0894, 0.0990, 0.0208, 0.0282, 0.0394, 0.0434, 0.0601, 0.0847], + device='cuda:2'), in_proj_covar=tensor([0.0444, 0.0353, 0.0256, 0.0293, 0.0373, 0.0328, 0.0349, 0.0387], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 19:53:35,461 INFO [train.py:901] (2/4) Epoch 2, batch 2000, loss[loss=0.3437, simple_loss=0.3757, pruned_loss=0.1558, over 7978.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4181, pruned_loss=0.1793, over 1617823.30 frames. ], batch size: 21, lr: 3.27e-02, grad_scale: 8.0 +2023-02-05 19:53:50,416 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10105.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:53:55,727 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.648e+02 4.167e+02 5.413e+02 6.926e+02 6.671e+03, threshold=1.083e+03, percent-clipped=14.0 +2023-02-05 19:54:10,561 INFO [train.py:901] (2/4) Epoch 2, batch 2050, loss[loss=0.3545, simple_loss=0.3952, pruned_loss=0.1569, over 8133.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4159, pruned_loss=0.1784, over 1616483.04 frames. ], batch size: 22, lr: 3.26e-02, grad_scale: 4.0 +2023-02-05 19:54:11,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10135.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:54:16,423 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-05 19:54:16,760 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.4249, 1.4231, 5.2435, 2.3122, 4.7862, 4.5565, 4.8102, 4.9938], + device='cuda:2'), covar=tensor([0.0293, 0.3427, 0.0189, 0.1422, 0.0737, 0.0279, 0.0260, 0.0276], + device='cuda:2'), in_proj_covar=tensor([0.0177, 0.0358, 0.0213, 0.0246, 0.0289, 0.0229, 0.0207, 0.0244], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 19:54:19,453 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10147.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:54:36,885 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10172.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:54:45,486 INFO [train.py:901] (2/4) Epoch 2, batch 2100, loss[loss=0.3794, simple_loss=0.415, pruned_loss=0.1719, over 8085.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.4148, pruned_loss=0.1769, over 1612942.51 frames. ], batch size: 21, lr: 3.25e-02, grad_scale: 4.0 +2023-02-05 19:55:06,153 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.637e+02 3.788e+02 4.646e+02 5.840e+02 1.328e+03, threshold=9.292e+02, percent-clipped=3.0 +2023-02-05 19:55:11,268 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:55:20,260 INFO [train.py:901] (2/4) Epoch 2, batch 2150, loss[loss=0.3923, simple_loss=0.4294, pruned_loss=0.1776, over 8250.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4132, pruned_loss=0.1753, over 1609033.03 frames. ], batch size: 24, lr: 3.25e-02, grad_scale: 4.0 +2023-02-05 19:55:50,895 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 19:55:53,988 INFO [train.py:901] (2/4) Epoch 2, batch 2200, loss[loss=0.3849, simple_loss=0.4141, pruned_loss=0.1779, over 8104.00 frames. ], tot_loss[loss=0.383, simple_loss=0.4139, pruned_loss=0.1761, over 1612611.80 frames. ], batch size: 23, lr: 3.24e-02, grad_scale: 4.0 +2023-02-05 19:56:13,115 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2023-02-05 19:56:14,593 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 3.803e+02 4.971e+02 6.310e+02 1.458e+03, threshold=9.942e+02, percent-clipped=6.0 +2023-02-05 19:56:18,160 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10317.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:56:29,272 INFO [train.py:901] (2/4) Epoch 2, batch 2250, loss[loss=0.4924, simple_loss=0.4834, pruned_loss=0.2507, over 7316.00 frames. ], tot_loss[loss=0.3858, simple_loss=0.4157, pruned_loss=0.1779, over 1613953.78 frames. ], batch size: 71, lr: 3.24e-02, grad_scale: 4.0 +2023-02-05 19:56:34,612 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10342.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:56:44,802 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 19:57:03,211 INFO [train.py:901] (2/4) Epoch 2, batch 2300, loss[loss=0.3445, simple_loss=0.3935, pruned_loss=0.1478, over 8609.00 frames. ], tot_loss[loss=0.3857, simple_loss=0.4155, pruned_loss=0.178, over 1614980.19 frames. ], batch size: 34, lr: 3.23e-02, grad_scale: 4.0 +2023-02-05 19:57:08,295 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10391.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:15,022 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10401.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:18,756 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.59 vs. limit=5.0 +2023-02-05 19:57:23,810 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.989e+02 5.161e+02 7.086e+02 1.471e+03, threshold=1.032e+03, percent-clipped=7.0 +2023-02-05 19:57:25,917 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10416.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:31,805 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10424.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:57:33,873 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10427.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:39,136 INFO [train.py:901] (2/4) Epoch 2, batch 2350, loss[loss=0.3765, simple_loss=0.4054, pruned_loss=0.1739, over 8358.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4149, pruned_loss=0.1775, over 1616469.40 frames. ], batch size: 24, lr: 3.22e-02, grad_scale: 4.0 +2023-02-05 19:57:39,514 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2023-02-05 19:58:05,152 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10472.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:07,808 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10476.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:12,888 INFO [train.py:901] (2/4) Epoch 2, batch 2400, loss[loss=0.3463, simple_loss=0.3974, pruned_loss=0.1475, over 8459.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4138, pruned_loss=0.1766, over 1614948.63 frames. ], batch size: 27, lr: 3.22e-02, grad_scale: 8.0 +2023-02-05 19:58:21,309 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7973, 1.0927, 3.2226, 1.0601, 2.0409, 3.8237, 3.3594, 3.3385], + device='cuda:2'), covar=tensor([0.1402, 0.1951, 0.0336, 0.2270, 0.0845, 0.0181, 0.0285, 0.0405], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0263, 0.0180, 0.0254, 0.0195, 0.0149, 0.0137, 0.0210], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 19:58:24,669 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10501.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:32,502 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.364e+02 3.956e+02 5.047e+02 6.263e+02 1.564e+03, threshold=1.009e+03, percent-clipped=2.0 +2023-02-05 19:58:34,740 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10516.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:47,327 INFO [train.py:901] (2/4) Epoch 2, batch 2450, loss[loss=0.3974, simple_loss=0.4337, pruned_loss=0.1806, over 8553.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4151, pruned_loss=0.1769, over 1618996.30 frames. ], batch size: 49, lr: 3.21e-02, grad_scale: 8.0 +2023-02-05 19:58:50,860 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10539.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:59:22,254 INFO [train.py:901] (2/4) Epoch 2, batch 2500, loss[loss=0.3689, simple_loss=0.3947, pruned_loss=0.1715, over 7811.00 frames. ], tot_loss[loss=0.382, simple_loss=0.4132, pruned_loss=0.1755, over 1618846.60 frames. ], batch size: 20, lr: 3.20e-02, grad_scale: 8.0 +2023-02-05 19:59:36,326 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1463, 1.1126, 4.0945, 1.6459, 3.4926, 3.5457, 3.6758, 3.6650], + device='cuda:2'), covar=tensor([0.0234, 0.2926, 0.0235, 0.1207, 0.0878, 0.0304, 0.0259, 0.0355], + device='cuda:2'), in_proj_covar=tensor([0.0177, 0.0355, 0.0216, 0.0241, 0.0286, 0.0228, 0.0215, 0.0244], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 19:59:42,154 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.520e+02 3.522e+02 4.438e+02 6.473e+02 1.354e+03, threshold=8.876e+02, percent-clipped=4.0 +2023-02-05 19:59:54,190 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4677, 2.0938, 3.2439, 3.0263, 2.9317, 2.0215, 1.6402, 2.4449], + device='cuda:2'), covar=tensor([0.0755, 0.0887, 0.0147, 0.0201, 0.0272, 0.0425, 0.0612, 0.0565], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0372, 0.0256, 0.0299, 0.0383, 0.0336, 0.0359, 0.0406], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 19:59:55,943 INFO [train.py:901] (2/4) Epoch 2, batch 2550, loss[loss=0.3289, simple_loss=0.3814, pruned_loss=0.1382, over 7930.00 frames. ], tot_loss[loss=0.3822, simple_loss=0.413, pruned_loss=0.1757, over 1617305.37 frames. ], batch size: 20, lr: 3.20e-02, grad_scale: 8.0 +2023-02-05 20:00:31,344 INFO [train.py:901] (2/4) Epoch 2, batch 2600, loss[loss=0.3952, simple_loss=0.43, pruned_loss=0.1802, over 8551.00 frames. ], tot_loss[loss=0.3815, simple_loss=0.4123, pruned_loss=0.1753, over 1613263.59 frames. ], batch size: 31, lr: 3.19e-02, grad_scale: 8.0 +2023-02-05 20:00:34,223 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5017, 2.0122, 1.5037, 2.0730, 1.9919, 1.3061, 1.5443, 2.1713], + device='cuda:2'), covar=tensor([0.1232, 0.0746, 0.1231, 0.0793, 0.1003, 0.1394, 0.1245, 0.0705], + device='cuda:2'), in_proj_covar=tensor([0.0362, 0.0237, 0.0356, 0.0304, 0.0350, 0.0314, 0.0352, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 20:00:44,745 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2756, 2.3787, 2.1084, 2.6333, 1.8501, 1.8621, 1.9755, 2.3972], + device='cuda:2'), covar=tensor([0.0844, 0.1066, 0.1231, 0.0419, 0.1574, 0.1426, 0.1487, 0.0844], + device='cuda:2'), in_proj_covar=tensor([0.0310, 0.0331, 0.0322, 0.0217, 0.0331, 0.0340, 0.0383, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:00:50,490 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 4.188e+02 4.914e+02 6.333e+02 1.432e+03, threshold=9.828e+02, percent-clipped=6.0 +2023-02-05 20:01:05,101 INFO [train.py:901] (2/4) Epoch 2, batch 2650, loss[loss=0.3803, simple_loss=0.3987, pruned_loss=0.1809, over 7978.00 frames. ], tot_loss[loss=0.3805, simple_loss=0.412, pruned_loss=0.1744, over 1610838.54 frames. ], batch size: 21, lr: 3.19e-02, grad_scale: 8.0 +2023-02-05 20:01:24,204 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10762.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:30,879 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10771.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:31,718 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10772.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:40,305 INFO [train.py:901] (2/4) Epoch 2, batch 2700, loss[loss=0.4327, simple_loss=0.4658, pruned_loss=0.1998, over 8757.00 frames. ], tot_loss[loss=0.3789, simple_loss=0.4107, pruned_loss=0.1735, over 1608343.37 frames. ], batch size: 30, lr: 3.18e-02, grad_scale: 8.0 +2023-02-05 20:01:46,658 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10792.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:48,791 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10795.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:01:50,095 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10797.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:02:01,027 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.290e+02 4.005e+02 5.458e+02 7.000e+02 2.619e+03, threshold=1.092e+03, percent-clipped=7.0 +2023-02-05 20:02:03,279 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10816.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:02:06,059 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10820.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:02:15,171 INFO [train.py:901] (2/4) Epoch 2, batch 2750, loss[loss=0.4329, simple_loss=0.4565, pruned_loss=0.2047, over 8320.00 frames. ], tot_loss[loss=0.3773, simple_loss=0.4095, pruned_loss=0.1726, over 1605287.11 frames. ], batch size: 26, lr: 3.17e-02, grad_scale: 8.0 +2023-02-05 20:02:49,756 INFO [train.py:901] (2/4) Epoch 2, batch 2800, loss[loss=0.4238, simple_loss=0.4348, pruned_loss=0.2064, over 7539.00 frames. ], tot_loss[loss=0.3789, simple_loss=0.4109, pruned_loss=0.1734, over 1607923.91 frames. ], batch size: 18, lr: 3.17e-02, grad_scale: 8.0 +2023-02-05 20:02:51,260 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10886.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:03:03,319 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10903.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:03:10,621 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.367e+02 3.535e+02 4.531e+02 6.001e+02 1.335e+03, threshold=9.062e+02, percent-clipped=2.0 +2023-02-05 20:03:23,137 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10931.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:03:25,018 INFO [train.py:901] (2/4) Epoch 2, batch 2850, loss[loss=0.4037, simple_loss=0.4539, pruned_loss=0.1767, over 8647.00 frames. ], tot_loss[loss=0.3776, simple_loss=0.4105, pruned_loss=0.1723, over 1608844.50 frames. ], batch size: 34, lr: 3.16e-02, grad_scale: 8.0 +2023-02-05 20:03:40,880 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-05 20:03:59,105 INFO [train.py:901] (2/4) Epoch 2, batch 2900, loss[loss=0.3313, simple_loss=0.3962, pruned_loss=0.1332, over 7807.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4111, pruned_loss=0.1727, over 1602820.78 frames. ], batch size: 20, lr: 3.16e-02, grad_scale: 8.0 +2023-02-05 20:04:18,979 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0117, 2.3707, 1.8584, 2.8176, 1.1710, 1.3303, 1.5907, 2.0444], + device='cuda:2'), covar=tensor([0.1015, 0.0947, 0.1434, 0.0344, 0.2004, 0.2148, 0.2232, 0.1148], + device='cuda:2'), in_proj_covar=tensor([0.0307, 0.0320, 0.0316, 0.0208, 0.0325, 0.0329, 0.0378, 0.0302], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:04:19,454 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.551e+02 4.216e+02 5.196e+02 6.845e+02 2.226e+03, threshold=1.039e+03, percent-clipped=10.0 +2023-02-05 20:04:34,448 INFO [train.py:901] (2/4) Epoch 2, batch 2950, loss[loss=0.5775, simple_loss=0.5428, pruned_loss=0.3061, over 8629.00 frames. ], tot_loss[loss=0.3793, simple_loss=0.4108, pruned_loss=0.1739, over 1601661.66 frames. ], batch size: 39, lr: 3.15e-02, grad_scale: 8.0 +2023-02-05 20:04:39,268 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 20:04:41,043 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 20:05:08,646 INFO [train.py:901] (2/4) Epoch 2, batch 3000, loss[loss=0.333, simple_loss=0.3747, pruned_loss=0.1456, over 7649.00 frames. ], tot_loss[loss=0.3793, simple_loss=0.4114, pruned_loss=0.1737, over 1605900.41 frames. ], batch size: 19, lr: 3.14e-02, grad_scale: 8.0 +2023-02-05 20:05:08,646 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 20:05:18,635 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4863, 1.6152, 1.5266, 2.3043, 1.1066, 1.0762, 1.5614, 1.5726], + device='cuda:2'), covar=tensor([0.1496, 0.1671, 0.1785, 0.0535, 0.2258, 0.2648, 0.2011, 0.1435], + device='cuda:2'), in_proj_covar=tensor([0.0307, 0.0324, 0.0313, 0.0213, 0.0320, 0.0331, 0.0372, 0.0302], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:05:24,855 INFO [train.py:935] (2/4) Epoch 2, validation: loss=0.2878, simple_loss=0.369, pruned_loss=0.1033, over 944034.00 frames. +2023-02-05 20:05:24,856 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 20:05:40,501 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11106.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:05:45,160 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.542e+02 3.795e+02 4.955e+02 6.193e+02 1.384e+03, threshold=9.910e+02, percent-clipped=4.0 +2023-02-05 20:06:00,078 INFO [train.py:901] (2/4) Epoch 2, batch 3050, loss[loss=0.391, simple_loss=0.4317, pruned_loss=0.1752, over 8478.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4122, pruned_loss=0.174, over 1609913.67 frames. ], batch size: 25, lr: 3.14e-02, grad_scale: 8.0 +2023-02-05 20:06:01,582 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11136.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:05,919 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11142.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:24,327 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11167.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:35,406 INFO [train.py:901] (2/4) Epoch 2, batch 3100, loss[loss=0.3254, simple_loss=0.3742, pruned_loss=0.1384, over 8082.00 frames. ], tot_loss[loss=0.378, simple_loss=0.4105, pruned_loss=0.1728, over 1609017.22 frames. ], batch size: 21, lr: 3.13e-02, grad_scale: 8.0 +2023-02-05 20:06:37,614 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11187.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:40,867 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11192.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:55,397 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11212.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:55,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.637e+02 3.930e+02 4.987e+02 6.652e+02 1.229e+03, threshold=9.974e+02, percent-clipped=5.0 +2023-02-05 20:06:58,267 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4639, 2.4919, 2.6664, 0.4062, 2.5494, 1.8421, 1.0658, 1.6419], + device='cuda:2'), covar=tensor([0.0205, 0.0074, 0.0108, 0.0317, 0.0131, 0.0280, 0.0317, 0.0164], + device='cuda:2'), in_proj_covar=tensor([0.0176, 0.0130, 0.0122, 0.0182, 0.0133, 0.0227, 0.0189, 0.0157], + device='cuda:2'), out_proj_covar=tensor([1.1370e-04, 8.5132e-05, 8.6157e-05, 1.2077e-04, 9.3833e-05, 1.5867e-04, + 1.2916e-04, 1.0555e-04], device='cuda:2') +2023-02-05 20:07:01,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11221.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:07:10,344 INFO [train.py:901] (2/4) Epoch 2, batch 3150, loss[loss=0.3711, simple_loss=0.4138, pruned_loss=0.1642, over 8346.00 frames. ], tot_loss[loss=0.3778, simple_loss=0.411, pruned_loss=0.1723, over 1614598.65 frames. ], batch size: 26, lr: 3.13e-02, grad_scale: 8.0 +2023-02-05 20:07:20,132 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11247.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:07:22,966 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11251.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:07:35,649 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.32 vs. limit=2.0 +2023-02-05 20:07:40,490 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.18 vs. limit=5.0 +2023-02-05 20:07:46,079 INFO [train.py:901] (2/4) Epoch 2, batch 3200, loss[loss=0.3089, simple_loss=0.3588, pruned_loss=0.1295, over 7980.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4089, pruned_loss=0.1707, over 1608843.02 frames. ], batch size: 21, lr: 3.12e-02, grad_scale: 8.0 +2023-02-05 20:07:48,949 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9401, 1.3284, 2.0778, 0.7977, 2.1909, 2.1855, 2.1515, 1.9046], + device='cuda:2'), covar=tensor([0.1737, 0.1716, 0.0642, 0.2751, 0.0648, 0.0612, 0.0609, 0.0924], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0258, 0.0171, 0.0242, 0.0190, 0.0148, 0.0137, 0.0208], + 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') +2023-02-05 20:07:56,013 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2023-02-05 20:08:06,205 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 3.889e+02 4.508e+02 6.050e+02 1.565e+03, threshold=9.016e+02, percent-clipped=4.0 +2023-02-05 20:08:07,294 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.67 vs. limit=5.0 +2023-02-05 20:08:21,234 INFO [train.py:901] (2/4) Epoch 2, batch 3250, loss[loss=0.374, simple_loss=0.4161, pruned_loss=0.1659, over 8111.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4061, pruned_loss=0.1683, over 1608499.03 frames. ], batch size: 23, lr: 3.11e-02, grad_scale: 8.0 +2023-02-05 20:08:39,992 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11362.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:08:55,029 INFO [train.py:901] (2/4) Epoch 2, batch 3300, loss[loss=0.3557, simple_loss=0.4035, pruned_loss=0.154, over 7906.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4066, pruned_loss=0.1686, over 1610900.33 frames. ], batch size: 20, lr: 3.11e-02, grad_scale: 8.0 +2023-02-05 20:09:16,011 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.686e+02 3.650e+02 4.417e+02 5.589e+02 1.513e+03, threshold=8.834e+02, percent-clipped=8.0 +2023-02-05 20:09:30,176 INFO [train.py:901] (2/4) Epoch 2, batch 3350, loss[loss=0.3582, simple_loss=0.3887, pruned_loss=0.1638, over 7809.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4068, pruned_loss=0.1691, over 1609208.03 frames. ], batch size: 20, lr: 3.10e-02, grad_scale: 8.0 +2023-02-05 20:10:00,609 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11477.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:04,735 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2846, 1.8814, 1.4173, 1.1923, 1.9363, 1.5762, 2.0331, 1.7453], + device='cuda:2'), covar=tensor([0.1086, 0.1639, 0.2358, 0.2128, 0.1075, 0.1994, 0.1165, 0.0979], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0262, 0.0289, 0.0259, 0.0233, 0.0251, 0.0230, 0.0222], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:10:05,251 INFO [train.py:901] (2/4) Epoch 2, batch 3400, loss[loss=0.3528, simple_loss=0.4039, pruned_loss=0.1509, over 8362.00 frames. ], tot_loss[loss=0.376, simple_loss=0.41, pruned_loss=0.171, over 1616508.90 frames. ], batch size: 24, lr: 3.10e-02, grad_scale: 8.0 +2023-02-05 20:10:17,706 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11502.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:21,204 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11507.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:25,774 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 3.730e+02 4.591e+02 5.662e+02 1.223e+03, threshold=9.181e+02, percent-clipped=5.0 +2023-02-05 20:10:39,501 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:40,654 INFO [train.py:901] (2/4) Epoch 2, batch 3450, loss[loss=0.3188, simple_loss=0.3582, pruned_loss=0.1397, over 7523.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.41, pruned_loss=0.1715, over 1619050.79 frames. ], batch size: 18, lr: 3.09e-02, grad_scale: 8.0 +2023-02-05 20:10:42,062 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11536.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:51,626 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11550.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:11:15,640 INFO [train.py:901] (2/4) Epoch 2, batch 3500, loss[loss=0.3388, simple_loss=0.398, pruned_loss=0.1398, over 8364.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4093, pruned_loss=0.1704, over 1620444.98 frames. ], batch size: 24, lr: 3.09e-02, grad_scale: 8.0 +2023-02-05 20:11:35,936 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.906e+02 4.071e+02 4.877e+02 6.297e+02 1.257e+03, threshold=9.753e+02, percent-clipped=3.0 +2023-02-05 20:11:39,547 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11618.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:11:40,719 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 20:11:50,751 INFO [train.py:901] (2/4) Epoch 2, batch 3550, loss[loss=0.3787, simple_loss=0.4288, pruned_loss=0.1643, over 8459.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4091, pruned_loss=0.1706, over 1619186.62 frames. ], batch size: 29, lr: 3.08e-02, grad_scale: 8.0 +2023-02-05 20:11:57,576 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11643.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:12:02,958 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11651.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:12:04,329 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11653.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:12:12,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1735, 1.4320, 1.2645, 1.9175, 1.1307, 0.9616, 1.2007, 1.4676], + device='cuda:2'), covar=tensor([0.1632, 0.1476, 0.1848, 0.0602, 0.1634, 0.2600, 0.1650, 0.1144], + device='cuda:2'), in_proj_covar=tensor([0.0315, 0.0331, 0.0332, 0.0215, 0.0323, 0.0338, 0.0371, 0.0298], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:12:25,619 INFO [train.py:901] (2/4) Epoch 2, batch 3600, loss[loss=0.3814, simple_loss=0.4057, pruned_loss=0.1785, over 7791.00 frames. ], tot_loss[loss=0.3763, simple_loss=0.4097, pruned_loss=0.1714, over 1619225.70 frames. ], batch size: 19, lr: 3.08e-02, grad_scale: 8.0 +2023-02-05 20:12:45,381 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.676e+02 3.688e+02 4.691e+02 6.662e+02 1.491e+03, threshold=9.383e+02, percent-clipped=3.0 +2023-02-05 20:12:48,294 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11717.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:12:59,433 INFO [train.py:901] (2/4) Epoch 2, batch 3650, loss[loss=0.3069, simple_loss=0.3478, pruned_loss=0.133, over 7652.00 frames. ], tot_loss[loss=0.3755, simple_loss=0.409, pruned_loss=0.171, over 1618393.87 frames. ], batch size: 19, lr: 3.07e-02, grad_scale: 8.0 +2023-02-05 20:13:25,959 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4989, 1.1253, 2.8990, 1.3218, 2.2572, 3.1656, 2.8646, 2.8247], + device='cuda:2'), covar=tensor([0.1473, 0.1687, 0.0422, 0.1938, 0.0667, 0.0269, 0.0396, 0.0531], + device='cuda:2'), in_proj_covar=tensor([0.0240, 0.0263, 0.0175, 0.0250, 0.0201, 0.0149, 0.0145, 0.0213], + 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') +2023-02-05 20:13:33,832 INFO [train.py:901] (2/4) Epoch 2, batch 3700, loss[loss=0.3646, simple_loss=0.3952, pruned_loss=0.167, over 7981.00 frames. ], tot_loss[loss=0.3759, simple_loss=0.4091, pruned_loss=0.1714, over 1619202.49 frames. ], batch size: 21, lr: 3.06e-02, grad_scale: 8.0 +2023-02-05 20:13:44,421 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 20:13:45,468 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0506, 1.8704, 2.9611, 2.5767, 2.4929, 1.8318, 1.4806, 1.5654], + device='cuda:2'), covar=tensor([0.0722, 0.0632, 0.0127, 0.0215, 0.0248, 0.0321, 0.0455, 0.0532], + device='cuda:2'), in_proj_covar=tensor([0.0470, 0.0384, 0.0280, 0.0314, 0.0404, 0.0351, 0.0373, 0.0406], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:13:53,762 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.910e+02 4.224e+02 5.211e+02 6.213e+02 2.304e+03, threshold=1.042e+03, percent-clipped=10.0 +2023-02-05 20:13:58,270 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 +2023-02-05 20:14:08,519 INFO [train.py:901] (2/4) Epoch 2, batch 3750, loss[loss=0.3467, simple_loss=0.3927, pruned_loss=0.1504, over 8662.00 frames. ], tot_loss[loss=0.3758, simple_loss=0.4087, pruned_loss=0.1715, over 1614824.06 frames. ], batch size: 34, lr: 3.06e-02, grad_scale: 8.0 +2023-02-05 20:14:08,633 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11834.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:14:28,578 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11864.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:14:33,029 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5845, 1.2356, 4.5360, 1.9071, 3.9661, 3.6762, 3.9794, 3.9014], + device='cuda:2'), covar=tensor([0.0207, 0.2725, 0.0174, 0.1285, 0.0608, 0.0319, 0.0255, 0.0295], + device='cuda:2'), in_proj_covar=tensor([0.0180, 0.0360, 0.0218, 0.0258, 0.0302, 0.0236, 0.0218, 0.0248], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:14:43,033 INFO [train.py:901] (2/4) Epoch 2, batch 3800, loss[loss=0.4234, simple_loss=0.4086, pruned_loss=0.2191, over 7540.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4086, pruned_loss=0.1712, over 1614979.95 frames. ], batch size: 18, lr: 3.05e-02, grad_scale: 8.0 +2023-02-05 20:14:46,610 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 20:14:49,684 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11894.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:14:54,224 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 +2023-02-05 20:14:58,776 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:02,634 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.750e+02 4.056e+02 4.773e+02 6.198e+02 1.391e+03, threshold=9.546e+02, percent-clipped=3.0 +2023-02-05 20:15:16,340 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11932.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:17,493 INFO [train.py:901] (2/4) Epoch 2, batch 3850, loss[loss=0.3915, simple_loss=0.4214, pruned_loss=0.1808, over 8642.00 frames. ], tot_loss[loss=0.3745, simple_loss=0.4079, pruned_loss=0.1705, over 1615567.49 frames. ], batch size: 39, lr: 3.05e-02, grad_scale: 8.0 +2023-02-05 20:15:20,311 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11938.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:25,285 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1705, 1.6375, 2.8100, 0.9321, 1.9869, 1.3148, 1.3119, 1.4040], + device='cuda:2'), covar=tensor([0.1372, 0.1375, 0.0457, 0.2143, 0.1064, 0.1913, 0.1208, 0.1440], + device='cuda:2'), in_proj_covar=tensor([0.0364, 0.0353, 0.0388, 0.0412, 0.0453, 0.0414, 0.0369, 0.0468], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 20:15:39,750 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11966.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:42,734 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-05 20:15:47,061 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 20:15:51,652 INFO [train.py:901] (2/4) Epoch 2, batch 3900, loss[loss=0.3758, simple_loss=0.4166, pruned_loss=0.1675, over 8516.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4084, pruned_loss=0.1706, over 1617547.84 frames. ], batch size: 26, lr: 3.04e-02, grad_scale: 8.0 +2023-02-05 20:16:01,108 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11997.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:06,095 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12002.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:10,862 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:13,165 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.113e+02 3.926e+02 4.686e+02 5.678e+02 1.222e+03, threshold=9.373e+02, percent-clipped=4.0 +2023-02-05 20:16:19,135 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 20:16:28,132 INFO [train.py:901] (2/4) Epoch 2, batch 3950, loss[loss=0.3349, simple_loss=0.3654, pruned_loss=0.1521, over 7716.00 frames. ], tot_loss[loss=0.3755, simple_loss=0.4088, pruned_loss=0.1711, over 1613608.06 frames. ], batch size: 18, lr: 3.04e-02, grad_scale: 8.0 +2023-02-05 20:16:36,270 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2209, 2.1283, 3.1773, 2.8075, 2.6261, 1.8086, 1.5058, 1.8413], + device='cuda:2'), covar=tensor([0.0706, 0.0640, 0.0129, 0.0213, 0.0283, 0.0353, 0.0477, 0.0536], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0393, 0.0294, 0.0329, 0.0415, 0.0363, 0.0373, 0.0413], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:16:46,990 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:56,012 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6081, 2.4064, 3.5326, 3.2889, 2.8834, 2.1795, 1.7429, 2.2158], + device='cuda:2'), covar=tensor([0.0587, 0.0634, 0.0124, 0.0180, 0.0256, 0.0316, 0.0414, 0.0522], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0400, 0.0303, 0.0331, 0.0422, 0.0369, 0.0378, 0.0420], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:17:02,479 INFO [train.py:901] (2/4) Epoch 2, batch 4000, loss[loss=0.3235, simple_loss=0.3561, pruned_loss=0.1455, over 7531.00 frames. ], tot_loss[loss=0.3758, simple_loss=0.409, pruned_loss=0.1713, over 1608440.43 frames. ], batch size: 18, lr: 3.03e-02, grad_scale: 8.0 +2023-02-05 20:17:09,212 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12094.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:17:22,650 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12112.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:17:23,108 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.955e+02 4.453e+02 5.904e+02 7.845e+02 2.502e+03, threshold=1.181e+03, percent-clipped=13.0 +2023-02-05 20:17:32,963 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3113, 1.9821, 1.4587, 2.0121, 1.7167, 1.3745, 1.5867, 2.0157], + device='cuda:2'), covar=tensor([0.1100, 0.0547, 0.0977, 0.0595, 0.0798, 0.0995, 0.0900, 0.0558], + device='cuda:2'), in_proj_covar=tensor([0.0368, 0.0249, 0.0352, 0.0313, 0.0364, 0.0317, 0.0360, 0.0327], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 20:17:36,859 INFO [train.py:901] (2/4) Epoch 2, batch 4050, loss[loss=0.3027, simple_loss=0.3331, pruned_loss=0.1362, over 7462.00 frames. ], tot_loss[loss=0.3763, simple_loss=0.4097, pruned_loss=0.1715, over 1609382.84 frames. ], batch size: 17, lr: 3.03e-02, grad_scale: 16.0 +2023-02-05 20:18:06,022 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12176.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:18:07,289 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12178.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:18:11,164 INFO [train.py:901] (2/4) Epoch 2, batch 4100, loss[loss=0.367, simple_loss=0.3927, pruned_loss=0.1707, over 7408.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4096, pruned_loss=0.1718, over 1609405.68 frames. ], batch size: 17, lr: 3.02e-02, grad_scale: 16.0 +2023-02-05 20:18:27,598 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12208.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:18:30,904 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.458e+02 3.728e+02 4.672e+02 5.863e+02 2.072e+03, threshold=9.344e+02, percent-clipped=1.0 +2023-02-05 20:18:47,039 INFO [train.py:901] (2/4) Epoch 2, batch 4150, loss[loss=0.3659, simple_loss=0.4003, pruned_loss=0.1658, over 8131.00 frames. ], tot_loss[loss=0.3742, simple_loss=0.4079, pruned_loss=0.1702, over 1602930.85 frames. ], batch size: 22, lr: 3.02e-02, grad_scale: 16.0 +2023-02-05 20:19:08,924 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12265.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:20,419 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12282.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:21,722 INFO [train.py:901] (2/4) Epoch 2, batch 4200, loss[loss=0.3425, simple_loss=0.4033, pruned_loss=0.1408, over 8446.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4054, pruned_loss=0.1687, over 1601638.42 frames. ], batch size: 27, lr: 3.01e-02, grad_scale: 16.0 +2023-02-05 20:19:25,928 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12290.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:28,635 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12293.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:38,116 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1885, 1.7723, 1.8522, 1.3376, 1.0553, 1.9109, 0.3588, 0.8622], + device='cuda:2'), covar=tensor([0.1149, 0.0553, 0.0292, 0.0596, 0.0915, 0.0327, 0.1633, 0.0925], + device='cuda:2'), in_proj_covar=tensor([0.0111, 0.0094, 0.0084, 0.0140, 0.0121, 0.0079, 0.0148, 0.0117], + device='cuda:2'), out_proj_covar=tensor([1.1063e-04, 9.9779e-05, 8.3575e-05, 1.3434e-04, 1.2477e-04, 7.9035e-05, + 1.4442e-04, 1.2076e-04], device='cuda:2') +2023-02-05 20:19:40,058 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12310.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:42,058 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.324e+02 3.573e+02 4.694e+02 5.833e+02 1.413e+03, threshold=9.388e+02, percent-clipped=6.0 +2023-02-05 20:19:43,526 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 20:19:49,246 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12323.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:57,044 INFO [train.py:901] (2/4) Epoch 2, batch 4250, loss[loss=0.38, simple_loss=0.4171, pruned_loss=0.1715, over 8249.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.4067, pruned_loss=0.1698, over 1605582.62 frames. ], batch size: 24, lr: 3.01e-02, grad_scale: 16.0 +2023-02-05 20:20:02,188 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3773, 1.5599, 2.3625, 1.0749, 1.9115, 1.6524, 1.4299, 1.6264], + device='cuda:2'), covar=tensor([0.1065, 0.1167, 0.0454, 0.1696, 0.0792, 0.1430, 0.0951, 0.1038], + device='cuda:2'), in_proj_covar=tensor([0.0371, 0.0362, 0.0390, 0.0418, 0.0475, 0.0422, 0.0371, 0.0458], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 20:20:06,019 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:06,619 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 20:20:17,106 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 +2023-02-05 20:20:20,975 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12368.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:32,224 INFO [train.py:901] (2/4) Epoch 2, batch 4300, loss[loss=0.4125, simple_loss=0.4439, pruned_loss=0.1906, over 8487.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4062, pruned_loss=0.1682, over 1606222.54 frames. ], batch size: 26, lr: 3.00e-02, grad_scale: 16.0 +2023-02-05 20:20:38,551 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12393.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:41,192 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12397.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:53,214 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.244e+02 3.864e+02 4.648e+02 5.983e+02 1.525e+03, threshold=9.296e+02, percent-clipped=6.0 +2023-02-05 20:21:00,843 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12425.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:05,819 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:06,902 INFO [train.py:901] (2/4) Epoch 2, batch 4350, loss[loss=0.438, simple_loss=0.4458, pruned_loss=0.2151, over 8473.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4052, pruned_loss=0.1667, over 1608544.18 frames. ], batch size: 29, lr: 2.99e-02, grad_scale: 8.0 +2023-02-05 20:21:09,737 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12438.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:22,839 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3375, 1.7413, 2.0613, 1.5478, 0.9986, 1.9354, 0.5229, 1.1127], + device='cuda:2'), covar=tensor([0.1196, 0.0700, 0.0219, 0.0635, 0.1457, 0.0410, 0.2084, 0.0944], + device='cuda:2'), in_proj_covar=tensor([0.0117, 0.0096, 0.0086, 0.0144, 0.0126, 0.0081, 0.0155, 0.0119], + device='cuda:2'), out_proj_covar=tensor([1.1647e-04, 1.0284e-04, 8.7467e-05, 1.3871e-04, 1.2978e-04, 8.1588e-05, + 1.5059e-04, 1.2420e-04], device='cuda:2') +2023-02-05 20:21:23,526 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12457.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:26,039 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12461.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:28,052 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12464.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:38,126 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 20:21:42,135 INFO [train.py:901] (2/4) Epoch 2, batch 4400, loss[loss=0.39, simple_loss=0.4162, pruned_loss=0.1819, over 7919.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4055, pruned_loss=0.1672, over 1611190.46 frames. ], batch size: 20, lr: 2.99e-02, grad_scale: 8.0 +2023-02-05 20:22:02,398 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.494e+02 4.041e+02 4.964e+02 6.742e+02 1.213e+03, threshold=9.928e+02, percent-clipped=4.0 +2023-02-05 20:22:16,720 INFO [train.py:901] (2/4) Epoch 2, batch 4450, loss[loss=0.3848, simple_loss=0.4274, pruned_loss=0.1711, over 8346.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4051, pruned_loss=0.167, over 1610794.29 frames. ], batch size: 26, lr: 2.98e-02, grad_scale: 8.0 +2023-02-05 20:22:17,400 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 20:22:27,253 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12549.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:29,911 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12553.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:45,079 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12574.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:49,161 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12579.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:52,394 INFO [train.py:901] (2/4) Epoch 2, batch 4500, loss[loss=0.3454, simple_loss=0.3987, pruned_loss=0.1461, over 8323.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.4057, pruned_loss=0.1673, over 1611843.87 frames. ], batch size: 25, lr: 2.98e-02, grad_scale: 8.0 +2023-02-05 20:23:06,030 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:23:12,558 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 20:23:13,220 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.312e+02 4.309e+02 5.092e+02 6.256e+02 1.421e+03, threshold=1.018e+03, percent-clipped=5.0 +2023-02-05 20:23:27,083 INFO [train.py:901] (2/4) Epoch 2, batch 4550, loss[loss=0.3405, simple_loss=0.3821, pruned_loss=0.1495, over 7793.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4046, pruned_loss=0.1662, over 1611221.94 frames. ], batch size: 19, lr: 2.97e-02, grad_scale: 8.0 +2023-02-05 20:23:29,978 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.4125, 2.5420, 4.2569, 4.2177, 3.1349, 2.6133, 1.8219, 2.5414], + device='cuda:2'), covar=tensor([0.0515, 0.0746, 0.0108, 0.0201, 0.0299, 0.0281, 0.0448, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0494, 0.0404, 0.0294, 0.0339, 0.0420, 0.0365, 0.0387, 0.0420], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:23:40,670 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12653.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:23:57,695 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12678.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:23:59,781 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12681.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:01,608 INFO [train.py:901] (2/4) Epoch 2, batch 4600, loss[loss=0.3701, simple_loss=0.4116, pruned_loss=0.1643, over 8693.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4049, pruned_loss=0.1667, over 1607136.44 frames. ], batch size: 39, lr: 2.97e-02, grad_scale: 8.0 +2023-02-05 20:24:17,921 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12706.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:23,151 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.401e+02 3.817e+02 4.647e+02 5.826e+02 1.354e+03, threshold=9.293e+02, percent-clipped=3.0 +2023-02-05 20:24:25,442 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12717.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:37,082 INFO [train.py:901] (2/4) Epoch 2, batch 4650, loss[loss=0.3833, simple_loss=0.4131, pruned_loss=0.1767, over 8300.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4049, pruned_loss=0.1671, over 1607458.67 frames. ], batch size: 23, lr: 2.96e-02, grad_scale: 8.0 +2023-02-05 20:24:42,619 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12742.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:09,736 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12781.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:11,634 INFO [train.py:901] (2/4) Epoch 2, batch 4700, loss[loss=0.3656, simple_loss=0.4172, pruned_loss=0.157, over 8337.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4051, pruned_loss=0.167, over 1608307.73 frames. ], batch size: 25, lr: 2.96e-02, grad_scale: 8.0 +2023-02-05 20:25:15,192 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9866, 3.1171, 2.7074, 1.3487, 2.7276, 2.7043, 2.7722, 2.4656], + device='cuda:2'), covar=tensor([0.1215, 0.0731, 0.1102, 0.4144, 0.0731, 0.0833, 0.1327, 0.0767], + device='cuda:2'), in_proj_covar=tensor([0.0313, 0.0209, 0.0255, 0.0335, 0.0230, 0.0178, 0.0237, 0.0162], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:25:28,728 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12808.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:29,549 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12809.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:32,791 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.540e+02 4.122e+02 5.358e+02 6.927e+02 1.344e+03, threshold=1.072e+03, percent-clipped=8.0 +2023-02-05 20:25:47,170 INFO [train.py:901] (2/4) Epoch 2, batch 4750, loss[loss=0.3417, simple_loss=0.3691, pruned_loss=0.1571, over 7439.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4048, pruned_loss=0.167, over 1609829.39 frames. ], batch size: 17, lr: 2.95e-02, grad_scale: 8.0 +2023-02-05 20:25:47,386 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12834.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:59,155 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=5.61 vs. limit=5.0 +2023-02-05 20:26:17,700 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.40 vs. limit=5.0 +2023-02-05 20:26:18,679 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 20:26:20,732 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 20:26:22,726 INFO [train.py:901] (2/4) Epoch 2, batch 4800, loss[loss=0.3324, simple_loss=0.365, pruned_loss=0.1499, over 7238.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4036, pruned_loss=0.1657, over 1610289.31 frames. ], batch size: 16, lr: 2.95e-02, grad_scale: 8.0 +2023-02-05 20:26:25,715 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4006, 2.0427, 3.3548, 3.1039, 2.7846, 2.0541, 1.3019, 1.8328], + device='cuda:2'), covar=tensor([0.0639, 0.0741, 0.0142, 0.0257, 0.0298, 0.0346, 0.0553, 0.0594], + device='cuda:2'), in_proj_covar=tensor([0.0503, 0.0413, 0.0307, 0.0354, 0.0431, 0.0377, 0.0401, 0.0435], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:26:42,251 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6582, 3.4488, 2.9520, 4.2237, 1.7471, 2.0442, 2.1117, 3.1753], + device='cuda:2'), covar=tensor([0.1117, 0.1190, 0.1206, 0.0216, 0.2273, 0.2041, 0.2719, 0.1072], + device='cuda:2'), in_proj_covar=tensor([0.0323, 0.0340, 0.0332, 0.0227, 0.0327, 0.0338, 0.0392, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:26:43,405 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 3.678e+02 4.471e+02 5.888e+02 1.234e+03, threshold=8.941e+02, percent-clipped=3.0 +2023-02-05 20:26:49,695 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12923.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:26:57,685 INFO [train.py:901] (2/4) Epoch 2, batch 4850, loss[loss=0.5133, simple_loss=0.4893, pruned_loss=0.2687, over 6701.00 frames. ], tot_loss[loss=0.3663, simple_loss=0.4025, pruned_loss=0.165, over 1607262.35 frames. ], batch size: 71, lr: 2.94e-02, grad_scale: 8.0 +2023-02-05 20:27:12,733 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 20:27:32,108 INFO [train.py:901] (2/4) Epoch 2, batch 4900, loss[loss=0.3677, simple_loss=0.4125, pruned_loss=0.1615, over 8395.00 frames. ], tot_loss[loss=0.3706, simple_loss=0.4057, pruned_loss=0.1677, over 1610356.76 frames. ], batch size: 49, lr: 2.94e-02, grad_scale: 8.0 +2023-02-05 20:27:37,246 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-05 20:27:53,286 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 4.170e+02 5.532e+02 7.452e+02 1.588e+03, threshold=1.106e+03, percent-clipped=9.0 +2023-02-05 20:28:06,717 INFO [train.py:901] (2/4) Epoch 2, batch 4950, loss[loss=0.3662, simple_loss=0.3929, pruned_loss=0.1698, over 8244.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4058, pruned_loss=0.1673, over 1611471.10 frames. ], batch size: 22, lr: 2.93e-02, grad_scale: 8.0 +2023-02-05 20:28:41,844 INFO [train.py:901] (2/4) Epoch 2, batch 5000, loss[loss=0.3338, simple_loss=0.3716, pruned_loss=0.148, over 7822.00 frames. ], tot_loss[loss=0.3715, simple_loss=0.4068, pruned_loss=0.1681, over 1606215.03 frames. ], batch size: 20, lr: 2.93e-02, grad_scale: 8.0 +2023-02-05 20:29:02,460 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 4.113e+02 5.050e+02 6.511e+02 1.788e+03, threshold=1.010e+03, percent-clipped=5.0 +2023-02-05 20:29:09,704 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13125.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:29:15,837 INFO [train.py:901] (2/4) Epoch 2, batch 5050, loss[loss=0.3131, simple_loss=0.3719, pruned_loss=0.1272, over 8034.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.405, pruned_loss=0.1663, over 1604683.69 frames. ], batch size: 22, lr: 2.92e-02, grad_scale: 4.0 +2023-02-05 20:29:35,420 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-05 20:29:47,478 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13179.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:29:47,969 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 20:29:50,616 INFO [train.py:901] (2/4) Epoch 2, batch 5100, loss[loss=0.3986, simple_loss=0.4108, pruned_loss=0.1932, over 7233.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4052, pruned_loss=0.1662, over 1604624.65 frames. ], batch size: 16, lr: 2.92e-02, grad_scale: 4.0 +2023-02-05 20:29:51,469 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2922, 3.5411, 2.7355, 4.0211, 1.9111, 1.6192, 1.8596, 3.1673], + device='cuda:2'), covar=tensor([0.1262, 0.0966, 0.1542, 0.0281, 0.1887, 0.2275, 0.2237, 0.0793], + device='cuda:2'), in_proj_covar=tensor([0.0318, 0.0341, 0.0328, 0.0231, 0.0325, 0.0339, 0.0381, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:29:54,083 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0079, 1.2946, 5.6877, 2.2733, 5.2828, 5.0805, 5.4033, 5.4055], + device='cuda:2'), covar=tensor([0.0158, 0.3121, 0.0184, 0.1351, 0.0592, 0.0220, 0.0187, 0.0211], + device='cuda:2'), in_proj_covar=tensor([0.0188, 0.0382, 0.0242, 0.0276, 0.0327, 0.0260, 0.0244, 0.0265], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:30:04,628 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13204.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:30:11,535 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.243e+02 3.930e+02 4.883e+02 5.892e+02 1.355e+03, threshold=9.766e+02, percent-clipped=3.0 +2023-02-05 20:30:15,661 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4962, 2.2060, 1.5644, 2.0295, 1.7568, 1.1639, 1.6979, 2.2729], + device='cuda:2'), covar=tensor([0.1259, 0.0405, 0.1034, 0.0772, 0.0856, 0.1443, 0.1150, 0.0717], + device='cuda:2'), in_proj_covar=tensor([0.0382, 0.0250, 0.0362, 0.0322, 0.0367, 0.0342, 0.0383, 0.0340], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 20:30:24,602 INFO [train.py:901] (2/4) Epoch 2, batch 5150, loss[loss=0.3501, simple_loss=0.3755, pruned_loss=0.1623, over 7269.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4046, pruned_loss=0.1663, over 1605335.23 frames. ], batch size: 16, lr: 2.91e-02, grad_scale: 4.0 +2023-02-05 20:30:28,700 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13240.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:30:59,010 INFO [train.py:901] (2/4) Epoch 2, batch 5200, loss[loss=0.4419, simple_loss=0.4609, pruned_loss=0.2114, over 8355.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4038, pruned_loss=0.1654, over 1611148.00 frames. ], batch size: 24, lr: 2.91e-02, grad_scale: 8.0 +2023-02-05 20:31:06,110 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2023-02-05 20:31:16,381 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5513, 2.0314, 2.3469, 0.3638, 2.1783, 1.3874, 0.4871, 1.8175], + device='cuda:2'), covar=tensor([0.0117, 0.0062, 0.0070, 0.0185, 0.0101, 0.0221, 0.0219, 0.0075], + device='cuda:2'), in_proj_covar=tensor([0.0192, 0.0139, 0.0128, 0.0185, 0.0137, 0.0240, 0.0192, 0.0167], + device='cuda:2'), out_proj_covar=tensor([1.1591e-04, 8.4414e-05, 8.0699e-05, 1.0965e-04, 8.7943e-05, 1.5584e-04, + 1.1968e-04, 1.0186e-04], device='cuda:2') +2023-02-05 20:31:20,912 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 4.339e+02 5.206e+02 6.705e+02 1.063e+03, threshold=1.041e+03, percent-clipped=3.0 +2023-02-05 20:31:33,617 INFO [train.py:901] (2/4) Epoch 2, batch 5250, loss[loss=0.408, simple_loss=0.4399, pruned_loss=0.1881, over 8465.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4061, pruned_loss=0.1669, over 1614762.47 frames. ], batch size: 25, lr: 2.91e-02, grad_scale: 8.0 +2023-02-05 20:31:42,986 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 20:31:44,699 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 20:31:54,415 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4565, 1.9620, 1.9056, 0.5381, 1.8947, 1.3942, 0.3208, 1.9933], + device='cuda:2'), covar=tensor([0.0095, 0.0052, 0.0083, 0.0143, 0.0095, 0.0197, 0.0196, 0.0045], + device='cuda:2'), in_proj_covar=tensor([0.0191, 0.0138, 0.0127, 0.0186, 0.0138, 0.0238, 0.0190, 0.0167], + device='cuda:2'), out_proj_covar=tensor([1.1478e-04, 8.3747e-05, 8.0332e-05, 1.1020e-04, 8.8782e-05, 1.5401e-04, + 1.1824e-04, 1.0183e-04], device='cuda:2') +2023-02-05 20:32:07,576 INFO [train.py:901] (2/4) Epoch 2, batch 5300, loss[loss=0.3176, simple_loss=0.374, pruned_loss=0.1306, over 8294.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4051, pruned_loss=0.1659, over 1614034.39 frames. ], batch size: 23, lr: 2.90e-02, grad_scale: 8.0 +2023-02-05 20:32:19,194 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-05 20:32:29,092 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 3.821e+02 4.884e+02 6.417e+02 1.823e+03, threshold=9.767e+02, percent-clipped=6.0 +2023-02-05 20:32:42,522 INFO [train.py:901] (2/4) Epoch 2, batch 5350, loss[loss=0.3164, simple_loss=0.3527, pruned_loss=0.14, over 7798.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4049, pruned_loss=0.1669, over 1612949.53 frames. ], batch size: 19, lr: 2.90e-02, grad_scale: 8.0 +2023-02-05 20:32:54,622 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0756, 1.1578, 4.1611, 1.7526, 3.5822, 3.4848, 3.6101, 3.6309], + device='cuda:2'), covar=tensor([0.0303, 0.2871, 0.0287, 0.1372, 0.0967, 0.0405, 0.0292, 0.0332], + device='cuda:2'), in_proj_covar=tensor([0.0198, 0.0379, 0.0240, 0.0273, 0.0327, 0.0262, 0.0241, 0.0267], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:33:16,588 INFO [train.py:901] (2/4) Epoch 2, batch 5400, loss[loss=0.3479, simple_loss=0.4031, pruned_loss=0.1463, over 8194.00 frames. ], tot_loss[loss=0.3708, simple_loss=0.4063, pruned_loss=0.1677, over 1618560.28 frames. ], batch size: 23, lr: 2.89e-02, grad_scale: 8.0 +2023-02-05 20:33:24,899 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13496.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:33:38,013 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.355e+02 3.820e+02 4.559e+02 5.766e+02 1.205e+03, threshold=9.119e+02, percent-clipped=6.0 +2023-02-05 20:33:43,050 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13521.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:33:51,329 INFO [train.py:901] (2/4) Epoch 2, batch 5450, loss[loss=0.3755, simple_loss=0.4161, pruned_loss=0.1674, over 7963.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.407, pruned_loss=0.1684, over 1617725.90 frames. ], batch size: 21, lr: 2.89e-02, grad_scale: 8.0 +2023-02-05 20:34:25,974 INFO [train.py:901] (2/4) Epoch 2, batch 5500, loss[loss=0.3436, simple_loss=0.3959, pruned_loss=0.1457, over 8526.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4057, pruned_loss=0.1674, over 1615247.34 frames. ], batch size: 28, lr: 2.88e-02, grad_scale: 8.0 +2023-02-05 20:34:28,060 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 20:34:46,540 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.267e+02 3.726e+02 4.817e+02 6.308e+02 1.682e+03, threshold=9.635e+02, percent-clipped=6.0 +2023-02-05 20:34:53,639 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-05 20:34:59,986 INFO [train.py:901] (2/4) Epoch 2, batch 5550, loss[loss=0.3806, simple_loss=0.4267, pruned_loss=0.1673, over 8473.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.405, pruned_loss=0.1658, over 1617141.63 frames. ], batch size: 27, lr: 2.88e-02, grad_scale: 8.0 +2023-02-05 20:35:14,237 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 20:35:35,316 INFO [train.py:901] (2/4) Epoch 2, batch 5600, loss[loss=0.3918, simple_loss=0.4237, pruned_loss=0.18, over 8449.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4049, pruned_loss=0.1657, over 1610160.95 frames. ], batch size: 25, lr: 2.87e-02, grad_scale: 8.0 +2023-02-05 20:35:55,770 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 3.833e+02 4.619e+02 6.071e+02 1.383e+03, threshold=9.238e+02, percent-clipped=5.0 +2023-02-05 20:36:04,604 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6436, 1.5648, 3.4338, 1.0613, 2.3689, 4.0136, 3.5100, 3.4709], + device='cuda:2'), covar=tensor([0.1083, 0.1237, 0.0281, 0.1747, 0.0561, 0.0149, 0.0190, 0.0406], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0258, 0.0178, 0.0252, 0.0190, 0.0153, 0.0142, 0.0220], + 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') +2023-02-05 20:36:08,577 INFO [train.py:901] (2/4) Epoch 2, batch 5650, loss[loss=0.3148, simple_loss=0.35, pruned_loss=0.1398, over 7813.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4005, pruned_loss=0.1629, over 1605757.66 frames. ], batch size: 20, lr: 2.87e-02, grad_scale: 8.0 +2023-02-05 20:36:23,373 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13755.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:36:29,332 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0318, 2.2052, 4.7388, 1.2503, 2.9516, 2.2835, 1.9668, 2.2585], + device='cuda:2'), covar=tensor([0.0871, 0.1049, 0.0290, 0.1508, 0.0852, 0.1236, 0.0795, 0.1409], + device='cuda:2'), in_proj_covar=tensor([0.0386, 0.0369, 0.0403, 0.0430, 0.0490, 0.0423, 0.0386, 0.0484], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 20:36:34,197 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 20:36:43,547 INFO [train.py:901] (2/4) Epoch 2, batch 5700, loss[loss=0.2865, simple_loss=0.3398, pruned_loss=0.1165, over 7708.00 frames. ], tot_loss[loss=0.361, simple_loss=0.399, pruned_loss=0.1615, over 1606956.37 frames. ], batch size: 18, lr: 2.86e-02, grad_scale: 8.0 +2023-02-05 20:37:00,588 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-05 20:37:05,590 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.337e+02 4.261e+02 5.123e+02 6.631e+02 2.352e+03, threshold=1.025e+03, percent-clipped=5.0 +2023-02-05 20:37:18,906 INFO [train.py:901] (2/4) Epoch 2, batch 5750, loss[loss=0.3795, simple_loss=0.411, pruned_loss=0.1739, over 8561.00 frames. ], tot_loss[loss=0.3616, simple_loss=0.3996, pruned_loss=0.1618, over 1609634.38 frames. ], batch size: 34, lr: 2.86e-02, grad_scale: 8.0 +2023-02-05 20:37:38,953 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 20:37:54,460 INFO [train.py:901] (2/4) Epoch 2, batch 5800, loss[loss=0.4696, simple_loss=0.468, pruned_loss=0.2356, over 6914.00 frames. ], tot_loss[loss=0.3618, simple_loss=0.3998, pruned_loss=0.1619, over 1602541.40 frames. ], batch size: 71, lr: 2.85e-02, grad_scale: 8.0 +2023-02-05 20:38:15,740 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.601e+02 3.784e+02 4.729e+02 6.225e+02 2.390e+03, threshold=9.458e+02, percent-clipped=5.0 +2023-02-05 20:38:22,765 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-05 20:38:29,067 INFO [train.py:901] (2/4) Epoch 2, batch 5850, loss[loss=0.4047, simple_loss=0.4352, pruned_loss=0.1871, over 8327.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4025, pruned_loss=0.1636, over 1609007.68 frames. ], batch size: 25, lr: 2.85e-02, grad_scale: 8.0 +2023-02-05 20:38:48,631 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-05 20:39:03,909 INFO [train.py:901] (2/4) Epoch 2, batch 5900, loss[loss=0.3848, simple_loss=0.4326, pruned_loss=0.1685, over 8781.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4013, pruned_loss=0.163, over 1613218.17 frames. ], batch size: 40, lr: 2.84e-02, grad_scale: 8.0 +2023-02-05 20:39:27,076 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.452e+02 3.946e+02 4.724e+02 6.297e+02 1.551e+03, threshold=9.448e+02, percent-clipped=7.0 +2023-02-05 20:39:40,157 INFO [train.py:901] (2/4) Epoch 2, batch 5950, loss[loss=0.3489, simple_loss=0.4091, pruned_loss=0.1444, over 8512.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4032, pruned_loss=0.1646, over 1615087.89 frames. ], batch size: 28, lr: 2.84e-02, grad_scale: 8.0 +2023-02-05 20:40:14,637 INFO [train.py:901] (2/4) Epoch 2, batch 6000, loss[loss=0.4416, simple_loss=0.451, pruned_loss=0.2161, over 7349.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.405, pruned_loss=0.166, over 1619615.86 frames. ], batch size: 71, lr: 2.84e-02, grad_scale: 8.0 +2023-02-05 20:40:14,637 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 20:40:27,828 INFO [train.py:935] (2/4) Epoch 2, validation: loss=0.2758, simple_loss=0.3606, pruned_loss=0.0955, over 944034.00 frames. +2023-02-05 20:40:27,829 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 20:40:32,705 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14090.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:40:38,744 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14099.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:40:41,500 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5452, 4.6649, 4.1054, 1.6097, 4.0782, 3.6711, 4.2033, 3.1128], + device='cuda:2'), covar=tensor([0.0529, 0.0364, 0.0682, 0.4001, 0.0367, 0.0596, 0.0904, 0.0634], + device='cuda:2'), in_proj_covar=tensor([0.0315, 0.0213, 0.0258, 0.0343, 0.0228, 0.0185, 0.0237, 0.0162], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:40:49,504 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 3.733e+02 4.780e+02 6.772e+02 2.203e+03, threshold=9.561e+02, percent-clipped=10.0 +2023-02-05 20:40:53,729 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14121.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:41:02,697 INFO [train.py:901] (2/4) Epoch 2, batch 6050, loss[loss=0.3924, simple_loss=0.4137, pruned_loss=0.1856, over 7971.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4023, pruned_loss=0.164, over 1618807.59 frames. ], batch size: 21, lr: 2.83e-02, grad_scale: 8.0 +2023-02-05 20:41:05,432 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14138.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:41:34,583 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0604, 1.1425, 4.1026, 1.6773, 2.2205, 4.9383, 4.3459, 4.3783], + device='cuda:2'), covar=tensor([0.1164, 0.1804, 0.0285, 0.1803, 0.0700, 0.0167, 0.0222, 0.0401], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0255, 0.0180, 0.0253, 0.0191, 0.0159, 0.0146, 0.0225], + 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') +2023-02-05 20:41:37,174 INFO [train.py:901] (2/4) Epoch 2, batch 6100, loss[loss=0.3916, simple_loss=0.4312, pruned_loss=0.176, over 8344.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4035, pruned_loss=0.1648, over 1617873.47 frames. ], batch size: 26, lr: 2.83e-02, grad_scale: 8.0 +2023-02-05 20:41:58,436 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14214.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:41:58,921 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 3.920e+02 4.920e+02 6.492e+02 2.677e+03, threshold=9.840e+02, percent-clipped=6.0 +2023-02-05 20:42:08,076 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 20:42:09,016 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6913, 1.5715, 2.4592, 2.0380, 2.0469, 1.5152, 1.2181, 1.0290], + device='cuda:2'), covar=tensor([0.0716, 0.0597, 0.0165, 0.0221, 0.0272, 0.0348, 0.0408, 0.0540], + device='cuda:2'), in_proj_covar=tensor([0.0507, 0.0427, 0.0325, 0.0365, 0.0463, 0.0397, 0.0413, 0.0440], + 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') +2023-02-05 20:42:11,468 INFO [train.py:901] (2/4) Epoch 2, batch 6150, loss[loss=0.3925, simple_loss=0.4287, pruned_loss=0.1781, over 8411.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.4032, pruned_loss=0.1645, over 1620689.06 frames. ], batch size: 49, lr: 2.82e-02, grad_scale: 8.0 +2023-02-05 20:42:25,026 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4998, 2.1532, 1.3654, 2.0597, 1.7538, 1.1435, 1.3735, 1.8326], + device='cuda:2'), covar=tensor([0.1132, 0.0382, 0.1179, 0.0540, 0.0753, 0.1390, 0.1236, 0.0617], + device='cuda:2'), in_proj_covar=tensor([0.0372, 0.0249, 0.0360, 0.0321, 0.0355, 0.0331, 0.0364, 0.0333], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 20:42:29,273 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2023-02-05 20:42:46,421 INFO [train.py:901] (2/4) Epoch 2, batch 6200, loss[loss=0.3432, simple_loss=0.38, pruned_loss=0.1533, over 7432.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4024, pruned_loss=0.1633, over 1620152.80 frames. ], batch size: 17, lr: 2.82e-02, grad_scale: 8.0 +2023-02-05 20:43:08,135 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.406e+02 3.453e+02 4.846e+02 6.394e+02 2.249e+03, threshold=9.691e+02, percent-clipped=6.0 +2023-02-05 20:43:21,529 INFO [train.py:901] (2/4) Epoch 2, batch 6250, loss[loss=0.3989, simple_loss=0.4121, pruned_loss=0.1929, over 7535.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4006, pruned_loss=0.1622, over 1615153.77 frames. ], batch size: 18, lr: 2.81e-02, grad_scale: 8.0 +2023-02-05 20:43:22,060 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 20:43:42,213 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3762, 1.8279, 1.3862, 1.4885, 2.0446, 1.7828, 2.1057, 1.9002], + device='cuda:2'), covar=tensor([0.0954, 0.1452, 0.2292, 0.1623, 0.0771, 0.1622, 0.0959, 0.0788], + device='cuda:2'), in_proj_covar=tensor([0.0218, 0.0247, 0.0276, 0.0242, 0.0215, 0.0240, 0.0209, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:43:55,847 INFO [train.py:901] (2/4) Epoch 2, batch 6300, loss[loss=0.4118, simple_loss=0.4447, pruned_loss=0.1894, over 7960.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4018, pruned_loss=0.163, over 1617923.41 frames. ], batch size: 21, lr: 2.81e-02, grad_scale: 8.0 +2023-02-05 20:44:17,495 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.643e+02 3.823e+02 4.655e+02 5.877e+02 1.568e+03, threshold=9.309e+02, percent-clipped=4.0 +2023-02-05 20:44:28,391 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14431.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:44:30,244 INFO [train.py:901] (2/4) Epoch 2, batch 6350, loss[loss=0.4201, simple_loss=0.4399, pruned_loss=0.2002, over 7521.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4005, pruned_loss=0.1615, over 1615765.86 frames. ], batch size: 71, lr: 2.81e-02, grad_scale: 8.0 +2023-02-05 20:44:30,318 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14434.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:44:51,463 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14465.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:44:54,887 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14470.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:03,108 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14482.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:45:04,273 INFO [train.py:901] (2/4) Epoch 2, batch 6400, loss[loss=0.2981, simple_loss=0.3595, pruned_loss=0.1183, over 8084.00 frames. ], tot_loss[loss=0.3628, simple_loss=0.4013, pruned_loss=0.1622, over 1619838.25 frames. ], batch size: 21, lr: 2.80e-02, grad_scale: 8.0 +2023-02-05 20:45:12,405 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14495.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:17,905 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3139, 1.9177, 1.9850, 0.6587, 1.8987, 1.3043, 0.3311, 1.8597], + device='cuda:2'), covar=tensor([0.0090, 0.0048, 0.0063, 0.0129, 0.0072, 0.0190, 0.0176, 0.0044], + device='cuda:2'), in_proj_covar=tensor([0.0191, 0.0137, 0.0125, 0.0190, 0.0133, 0.0242, 0.0196, 0.0176], + device='cuda:2'), out_proj_covar=tensor([1.1004e-04, 7.8344e-05, 7.5238e-05, 1.0894e-04, 7.9981e-05, 1.4923e-04, + 1.1457e-04, 1.0304e-04], device='cuda:2') +2023-02-05 20:45:19,142 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14505.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:25,558 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 3.974e+02 5.065e+02 7.362e+02 1.328e+03, threshold=1.013e+03, percent-clipped=8.0 +2023-02-05 20:45:38,723 INFO [train.py:901] (2/4) Epoch 2, batch 6450, loss[loss=0.3722, simple_loss=0.4245, pruned_loss=0.1599, over 8284.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4019, pruned_loss=0.1622, over 1624469.95 frames. ], batch size: 23, lr: 2.80e-02, grad_scale: 8.0 +2023-02-05 20:45:48,962 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14549.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:46:10,579 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14580.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:46:13,166 INFO [train.py:901] (2/4) Epoch 2, batch 6500, loss[loss=0.2447, simple_loss=0.3116, pruned_loss=0.08892, over 6830.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4005, pruned_loss=0.1612, over 1618307.30 frames. ], batch size: 15, lr: 2.79e-02, grad_scale: 8.0 +2023-02-05 20:46:22,640 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14597.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:46:35,355 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.440e+02 3.999e+02 5.009e+02 6.288e+02 1.522e+03, threshold=1.002e+03, percent-clipped=8.0 +2023-02-05 20:46:48,433 INFO [train.py:901] (2/4) Epoch 2, batch 6550, loss[loss=0.3392, simple_loss=0.39, pruned_loss=0.1442, over 8250.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4026, pruned_loss=0.1627, over 1618362.82 frames. ], batch size: 24, lr: 2.79e-02, grad_scale: 8.0 +2023-02-05 20:47:16,630 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 20:47:23,548 INFO [train.py:901] (2/4) Epoch 2, batch 6600, loss[loss=0.3245, simple_loss=0.3635, pruned_loss=0.1427, over 7924.00 frames. ], tot_loss[loss=0.3628, simple_loss=0.4018, pruned_loss=0.1619, over 1616648.72 frames. ], batch size: 20, lr: 2.78e-02, grad_scale: 8.0 +2023-02-05 20:47:36,561 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 20:47:45,898 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.289e+02 3.681e+02 4.457e+02 5.556e+02 1.208e+03, threshold=8.913e+02, percent-clipped=4.0 +2023-02-05 20:47:58,943 INFO [train.py:901] (2/4) Epoch 2, batch 6650, loss[loss=0.3714, simple_loss=0.4152, pruned_loss=0.1638, over 8491.00 frames. ], tot_loss[loss=0.3632, simple_loss=0.4017, pruned_loss=0.1624, over 1613267.83 frames. ], batch size: 28, lr: 2.78e-02, grad_scale: 8.0 +2023-02-05 20:48:16,399 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14758.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:48:26,143 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4803, 1.8749, 1.9088, 1.6929, 1.1256, 1.8873, 0.4125, 1.0939], + device='cuda:2'), covar=tensor([0.1630, 0.0847, 0.0618, 0.0842, 0.2033, 0.0581, 0.2858, 0.1292], + device='cuda:2'), in_proj_covar=tensor([0.0113, 0.0101, 0.0086, 0.0140, 0.0139, 0.0083, 0.0161, 0.0120], + device='cuda:2'), out_proj_covar=tensor([1.2294e-04, 1.1777e-04, 9.4944e-05, 1.4791e-04, 1.5243e-04, 9.3962e-05, + 1.7045e-04, 1.3307e-04], device='cuda:2') +2023-02-05 20:48:28,648 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14775.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:48:34,790 INFO [train.py:901] (2/4) Epoch 2, batch 6700, loss[loss=0.3338, simple_loss=0.3784, pruned_loss=0.1447, over 7078.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.3986, pruned_loss=0.16, over 1609420.48 frames. ], batch size: 71, lr: 2.78e-02, grad_scale: 8.0 +2023-02-05 20:48:39,096 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2934, 1.1833, 1.3301, 1.1076, 1.4372, 1.2680, 1.3325, 1.2426], + device='cuda:2'), covar=tensor([0.1067, 0.1799, 0.2277, 0.1826, 0.0781, 0.1789, 0.0970, 0.0823], + device='cuda:2'), in_proj_covar=tensor([0.0213, 0.0240, 0.0276, 0.0239, 0.0208, 0.0239, 0.0203, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:48:45,397 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3288, 1.7597, 2.6963, 0.9440, 2.0081, 1.4674, 1.3519, 1.8871], + device='cuda:2'), covar=tensor([0.1286, 0.1274, 0.0472, 0.2273, 0.1115, 0.1829, 0.1138, 0.1342], + device='cuda:2'), in_proj_covar=tensor([0.0406, 0.0382, 0.0436, 0.0451, 0.0508, 0.0446, 0.0401, 0.0503], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 20:48:48,208 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6260, 2.3213, 1.5120, 2.0172, 1.8992, 1.4044, 1.7489, 1.9610], + device='cuda:2'), covar=tensor([0.1207, 0.0372, 0.1124, 0.0703, 0.0740, 0.1282, 0.0985, 0.0733], + device='cuda:2'), in_proj_covar=tensor([0.0381, 0.0246, 0.0369, 0.0322, 0.0360, 0.0341, 0.0363, 0.0331], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 20:48:50,186 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14805.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:48:53,130 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.62 vs. limit=5.0 +2023-02-05 20:48:56,685 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 3.873e+02 4.634e+02 6.203e+02 1.536e+03, threshold=9.268e+02, percent-clipped=6.0 +2023-02-05 20:49:04,584 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 +2023-02-05 20:49:07,134 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14830.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:09,756 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1463, 4.2939, 3.6404, 1.6273, 3.6043, 3.6953, 3.8502, 3.1891], + device='cuda:2'), covar=tensor([0.0926, 0.0461, 0.1014, 0.4623, 0.0668, 0.0562, 0.1151, 0.0655], + device='cuda:2'), in_proj_covar=tensor([0.0330, 0.0225, 0.0262, 0.0357, 0.0239, 0.0190, 0.0253, 0.0172], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 20:49:10,351 INFO [train.py:901] (2/4) Epoch 2, batch 6750, loss[loss=0.3053, simple_loss=0.3556, pruned_loss=0.1275, over 8077.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4009, pruned_loss=0.1622, over 1609973.38 frames. ], batch size: 21, lr: 2.77e-02, grad_scale: 8.0 +2023-02-05 20:49:11,909 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14836.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:17,972 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2172, 3.1757, 2.7857, 1.3824, 2.8218, 2.6397, 2.9405, 2.4559], + device='cuda:2'), covar=tensor([0.0854, 0.0552, 0.0913, 0.3630, 0.0640, 0.0829, 0.0971, 0.0687], + device='cuda:2'), in_proj_covar=tensor([0.0333, 0.0225, 0.0264, 0.0358, 0.0241, 0.0192, 0.0253, 0.0174], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 20:49:21,261 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14849.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:24,080 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14853.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:49:29,649 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14861.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:41,490 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14878.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:49:45,988 INFO [train.py:901] (2/4) Epoch 2, batch 6800, loss[loss=0.3056, simple_loss=0.3764, pruned_loss=0.1174, over 8245.00 frames. ], tot_loss[loss=0.3627, simple_loss=0.4013, pruned_loss=0.162, over 1609774.00 frames. ], batch size: 24, lr: 2.77e-02, grad_scale: 8.0 +2023-02-05 20:49:50,348 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14890.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:54,304 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 20:50:00,276 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-05 20:50:07,690 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 3.663e+02 4.715e+02 6.092e+02 1.805e+03, threshold=9.431e+02, percent-clipped=7.0 +2023-02-05 20:50:10,657 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4695, 1.2732, 1.3323, 1.1901, 1.4909, 1.3674, 1.0501, 1.3616], + device='cuda:2'), covar=tensor([0.1035, 0.1624, 0.2317, 0.1798, 0.0709, 0.1700, 0.1018, 0.0765], + device='cuda:2'), in_proj_covar=tensor([0.0213, 0.0240, 0.0276, 0.0240, 0.0208, 0.0241, 0.0202, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:50:21,310 INFO [train.py:901] (2/4) Epoch 2, batch 6850, loss[loss=0.3144, simple_loss=0.3524, pruned_loss=0.1382, over 7545.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4021, pruned_loss=0.1624, over 1612680.21 frames. ], batch size: 18, lr: 2.76e-02, grad_scale: 8.0 +2023-02-05 20:50:42,734 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14964.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:50:45,334 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 20:50:50,558 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2684, 1.4711, 2.3336, 0.9388, 1.7681, 1.4394, 1.3168, 1.4431], + device='cuda:2'), covar=tensor([0.1137, 0.1158, 0.0416, 0.1809, 0.0888, 0.1794, 0.0966, 0.1213], + device='cuda:2'), in_proj_covar=tensor([0.0415, 0.0389, 0.0450, 0.0464, 0.0513, 0.0467, 0.0410, 0.0519], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 20:50:57,102 INFO [train.py:901] (2/4) Epoch 2, batch 6900, loss[loss=0.3375, simple_loss=0.4031, pruned_loss=0.136, over 8466.00 frames. ], tot_loss[loss=0.3635, simple_loss=0.4019, pruned_loss=0.1626, over 1611890.24 frames. ], batch size: 25, lr: 2.76e-02, grad_scale: 8.0 +2023-02-05 20:50:58,971 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-05 20:51:19,285 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 4.191e+02 5.097e+02 7.005e+02 1.700e+03, threshold=1.019e+03, percent-clipped=5.0 +2023-02-05 20:51:32,576 INFO [train.py:901] (2/4) Epoch 2, batch 6950, loss[loss=0.391, simple_loss=0.4388, pruned_loss=0.1716, over 8109.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4011, pruned_loss=0.1621, over 1612367.21 frames. ], batch size: 23, lr: 2.75e-02, grad_scale: 8.0 +2023-02-05 20:51:56,461 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-05 20:52:08,404 INFO [train.py:901] (2/4) Epoch 2, batch 7000, loss[loss=0.4045, simple_loss=0.4505, pruned_loss=0.1793, over 8104.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.3986, pruned_loss=0.1595, over 1613172.38 frames. ], batch size: 23, lr: 2.75e-02, grad_scale: 8.0 +2023-02-05 20:52:14,182 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.00 vs. limit=5.0 +2023-02-05 20:52:21,506 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15102.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:52:30,559 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.233e+02 3.928e+02 4.810e+02 5.818e+02 1.410e+03, threshold=9.621e+02, percent-clipped=1.0 +2023-02-05 20:52:35,507 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0427, 2.3559, 2.0095, 2.7475, 1.8290, 1.6078, 2.0682, 2.2976], + device='cuda:2'), covar=tensor([0.0924, 0.0940, 0.1235, 0.0450, 0.1380, 0.1829, 0.1390, 0.0889], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0332, 0.0320, 0.0231, 0.0313, 0.0335, 0.0354, 0.0303], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:2') +2023-02-05 20:52:44,327 INFO [train.py:901] (2/4) Epoch 2, batch 7050, loss[loss=0.3111, simple_loss=0.3633, pruned_loss=0.1295, over 7657.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.3983, pruned_loss=0.1594, over 1611064.42 frames. ], batch size: 19, lr: 2.75e-02, grad_scale: 16.0 +2023-02-05 20:52:52,901 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15146.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:52:58,260 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.3138, 0.9523, 3.5823, 1.4454, 2.6510, 3.0817, 3.3378, 3.4023], + device='cuda:2'), covar=tensor([0.0611, 0.4047, 0.0571, 0.2086, 0.2019, 0.0690, 0.0570, 0.0675], + device='cuda:2'), in_proj_covar=tensor([0.0211, 0.0394, 0.0250, 0.0291, 0.0355, 0.0282, 0.0266, 0.0286], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:53:06,394 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9307, 2.0882, 4.0602, 3.7695, 2.8600, 2.2742, 1.5896, 1.8398], + device='cuda:2'), covar=tensor([0.0740, 0.1063, 0.0195, 0.0258, 0.0517, 0.0387, 0.0528, 0.0936], + device='cuda:2'), in_proj_covar=tensor([0.0532, 0.0444, 0.0343, 0.0386, 0.0496, 0.0418, 0.0437, 0.0450], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:53:10,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15171.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:18,766 INFO [train.py:901] (2/4) Epoch 2, batch 7100, loss[loss=0.4032, simple_loss=0.4454, pruned_loss=0.1805, over 8523.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.3982, pruned_loss=0.1592, over 1614464.50 frames. ], batch size: 28, lr: 2.74e-02, grad_scale: 16.0 +2023-02-05 20:53:23,894 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3747, 1.9406, 3.2649, 2.6981, 2.4355, 1.9523, 1.4975, 1.6328], + device='cuda:2'), covar=tensor([0.0691, 0.0828, 0.0148, 0.0310, 0.0407, 0.0391, 0.0527, 0.0773], + device='cuda:2'), in_proj_covar=tensor([0.0534, 0.0451, 0.0347, 0.0389, 0.0501, 0.0422, 0.0442, 0.0454], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 20:53:26,482 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9771, 4.0176, 3.6329, 1.5845, 3.6006, 3.5005, 3.6232, 2.9936], + device='cuda:2'), covar=tensor([0.0844, 0.0580, 0.0908, 0.4282, 0.0580, 0.0640, 0.1335, 0.0728], + device='cuda:2'), in_proj_covar=tensor([0.0321, 0.0228, 0.0258, 0.0344, 0.0237, 0.0190, 0.0247, 0.0170], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:53:39,782 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15213.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:40,997 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.137e+02 3.718e+02 4.413e+02 5.855e+02 1.165e+03, threshold=8.826e+02, percent-clipped=3.0 +2023-02-05 20:53:42,524 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15217.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:44,502 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:53,628 INFO [train.py:901] (2/4) Epoch 2, batch 7150, loss[loss=0.3657, simple_loss=0.4138, pruned_loss=0.1588, over 8520.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.3986, pruned_loss=0.1593, over 1616944.27 frames. ], batch size: 28, lr: 2.74e-02, grad_scale: 16.0 +2023-02-05 20:54:02,059 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15245.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:54:17,101 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4778, 2.0052, 1.9825, 0.5823, 2.0922, 1.4059, 0.4227, 1.8970], + device='cuda:2'), covar=tensor([0.0142, 0.0057, 0.0068, 0.0170, 0.0067, 0.0241, 0.0224, 0.0070], + device='cuda:2'), in_proj_covar=tensor([0.0201, 0.0135, 0.0124, 0.0186, 0.0139, 0.0251, 0.0201, 0.0175], + device='cuda:2'), out_proj_covar=tensor([1.1355e-04, 7.5475e-05, 7.1644e-05, 1.0169e-04, 8.2241e-05, 1.5283e-04, + 1.1501e-04, 9.9122e-05], device='cuda:2') +2023-02-05 20:54:29,183 INFO [train.py:901] (2/4) Epoch 2, batch 7200, loss[loss=0.4204, simple_loss=0.4175, pruned_loss=0.2117, over 5973.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.3976, pruned_loss=0.1584, over 1613144.58 frames. ], batch size: 13, lr: 2.73e-02, grad_scale: 16.0 +2023-02-05 20:54:51,169 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.528e+02 3.704e+02 4.905e+02 6.625e+02 1.855e+03, threshold=9.809e+02, percent-clipped=12.0 +2023-02-05 20:55:04,885 INFO [train.py:901] (2/4) Epoch 2, batch 7250, loss[loss=0.438, simple_loss=0.4658, pruned_loss=0.2051, over 8045.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.3981, pruned_loss=0.1588, over 1617907.13 frames. ], batch size: 22, lr: 2.73e-02, grad_scale: 8.0 +2023-02-05 20:55:25,881 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 +2023-02-05 20:55:39,918 INFO [train.py:901] (2/4) Epoch 2, batch 7300, loss[loss=0.3076, simple_loss=0.3721, pruned_loss=0.1215, over 8521.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.3985, pruned_loss=0.1594, over 1618141.98 frames. ], batch size: 26, lr: 2.73e-02, grad_scale: 8.0 +2023-02-05 20:56:02,317 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.194e+02 3.434e+02 4.292e+02 5.923e+02 1.449e+03, threshold=8.584e+02, percent-clipped=5.0 +2023-02-05 20:56:05,230 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5872, 2.5762, 1.6894, 2.3583, 2.1659, 1.3619, 1.4369, 2.0321], + device='cuda:2'), covar=tensor([0.1403, 0.0427, 0.0873, 0.0593, 0.0740, 0.1164, 0.1234, 0.0781], + device='cuda:2'), in_proj_covar=tensor([0.0386, 0.0251, 0.0355, 0.0324, 0.0362, 0.0330, 0.0364, 0.0330], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 20:56:14,878 INFO [train.py:901] (2/4) Epoch 2, batch 7350, loss[loss=0.3013, simple_loss=0.3545, pruned_loss=0.1241, over 7797.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.3987, pruned_loss=0.1598, over 1619186.88 frames. ], batch size: 19, lr: 2.72e-02, grad_scale: 8.0 +2023-02-05 20:56:42,773 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15473.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:56:43,905 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-05 20:56:49,847 INFO [train.py:901] (2/4) Epoch 2, batch 7400, loss[loss=0.3454, simple_loss=0.392, pruned_loss=0.1494, over 8467.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.3976, pruned_loss=0.1588, over 1618761.01 frames. ], batch size: 25, lr: 2.72e-02, grad_scale: 8.0 +2023-02-05 20:56:59,515 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15498.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:57:01,981 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-05 20:57:04,160 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6316, 2.0016, 2.1248, 0.6277, 2.0909, 1.3316, 0.4878, 1.9042], + device='cuda:2'), covar=tensor([0.0105, 0.0038, 0.0050, 0.0135, 0.0066, 0.0195, 0.0176, 0.0051], + device='cuda:2'), in_proj_covar=tensor([0.0200, 0.0134, 0.0124, 0.0185, 0.0130, 0.0249, 0.0201, 0.0176], + device='cuda:2'), out_proj_covar=tensor([1.1159e-04, 7.3433e-05, 7.0393e-05, 1.0050e-04, 7.5858e-05, 1.5033e-04, + 1.1429e-04, 9.8739e-05], device='cuda:2') +2023-02-05 20:57:11,815 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 3.940e+02 4.956e+02 6.362e+02 1.377e+03, threshold=9.912e+02, percent-clipped=7.0 +2023-02-05 20:57:24,678 INFO [train.py:901] (2/4) Epoch 2, batch 7450, loss[loss=0.3401, simple_loss=0.3902, pruned_loss=0.145, over 7982.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.3977, pruned_loss=0.1596, over 1615262.58 frames. ], batch size: 21, lr: 2.71e-02, grad_scale: 8.0 +2023-02-05 20:57:30,904 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9388, 1.8822, 1.7639, 2.3629, 1.1677, 1.2468, 1.3955, 1.9274], + device='cuda:2'), covar=tensor([0.1194, 0.1433, 0.1566, 0.0650, 0.1887, 0.2657, 0.2144, 0.1122], + device='cuda:2'), in_proj_covar=tensor([0.0311, 0.0324, 0.0328, 0.0224, 0.0308, 0.0329, 0.0359, 0.0304], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0003, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 20:57:40,476 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15557.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:57:41,773 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-05 20:57:49,494 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-05 20:57:50,501 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3220, 1.4820, 1.2770, 1.6850, 1.2218, 1.0602, 1.0837, 1.4678], + device='cuda:2'), covar=tensor([0.0830, 0.0515, 0.0997, 0.0607, 0.0780, 0.1100, 0.0941, 0.0650], + device='cuda:2'), in_proj_covar=tensor([0.0377, 0.0246, 0.0348, 0.0321, 0.0345, 0.0323, 0.0354, 0.0324], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], + device='cuda:2') +2023-02-05 20:57:51,016 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8271, 2.8669, 2.5040, 1.3783, 2.4951, 2.3954, 2.5931, 2.1675], + device='cuda:2'), covar=tensor([0.1143, 0.0757, 0.1104, 0.4070, 0.0767, 0.0981, 0.1370, 0.0916], + device='cuda:2'), in_proj_covar=tensor([0.0330, 0.0221, 0.0259, 0.0348, 0.0233, 0.0198, 0.0247, 0.0175], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:57:59,034 INFO [train.py:901] (2/4) Epoch 2, batch 7500, loss[loss=0.321, simple_loss=0.3754, pruned_loss=0.1333, over 8599.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.3976, pruned_loss=0.1591, over 1620575.91 frames. ], batch size: 34, lr: 2.71e-02, grad_scale: 8.0 +2023-02-05 20:58:21,355 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.209e+02 3.662e+02 4.519e+02 5.678e+02 1.466e+03, threshold=9.038e+02, percent-clipped=6.0 +2023-02-05 20:58:26,262 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9314, 1.0930, 3.0660, 0.9753, 2.4305, 2.5604, 2.6426, 2.6723], + device='cuda:2'), covar=tensor([0.0528, 0.3237, 0.0380, 0.2063, 0.1362, 0.0589, 0.0574, 0.0644], + device='cuda:2'), in_proj_covar=tensor([0.0213, 0.0397, 0.0252, 0.0294, 0.0353, 0.0278, 0.0270, 0.0282], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 20:58:34,053 INFO [train.py:901] (2/4) Epoch 2, batch 7550, loss[loss=0.3504, simple_loss=0.4037, pruned_loss=0.1486, over 8338.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.3964, pruned_loss=0.1579, over 1622062.82 frames. ], batch size: 26, lr: 2.71e-02, grad_scale: 8.0 +2023-02-05 20:59:00,923 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15672.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:59:08,584 INFO [train.py:901] (2/4) Epoch 2, batch 7600, loss[loss=0.3756, simple_loss=0.4156, pruned_loss=0.1679, over 8634.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.3964, pruned_loss=0.1579, over 1621737.69 frames. ], batch size: 34, lr: 2.70e-02, grad_scale: 8.0 +2023-02-05 20:59:31,058 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 3.634e+02 4.473e+02 6.191e+02 1.516e+03, threshold=8.946e+02, percent-clipped=5.0 +2023-02-05 20:59:43,083 INFO [train.py:901] (2/4) Epoch 2, batch 7650, loss[loss=0.3159, simple_loss=0.3557, pruned_loss=0.138, over 7704.00 frames. ], tot_loss[loss=0.356, simple_loss=0.3961, pruned_loss=0.158, over 1619511.56 frames. ], batch size: 18, lr: 2.70e-02, grad_scale: 8.0 +2023-02-05 21:00:19,422 INFO [train.py:901] (2/4) Epoch 2, batch 7700, loss[loss=0.3438, simple_loss=0.3779, pruned_loss=0.1549, over 6444.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.3932, pruned_loss=0.1555, over 1615082.24 frames. ], batch size: 14, lr: 2.69e-02, grad_scale: 8.0 +2023-02-05 21:00:41,051 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 3.880e+02 4.902e+02 6.175e+02 1.322e+03, threshold=9.805e+02, percent-clipped=4.0 +2023-02-05 21:00:42,376 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 21:00:51,197 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-05 21:00:53,924 INFO [train.py:901] (2/4) Epoch 2, batch 7750, loss[loss=0.3417, simple_loss=0.3903, pruned_loss=0.1465, over 8102.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.393, pruned_loss=0.1549, over 1616612.37 frames. ], batch size: 23, lr: 2.69e-02, grad_scale: 8.0 +2023-02-05 21:01:28,170 INFO [train.py:901] (2/4) Epoch 2, batch 7800, loss[loss=0.3443, simple_loss=0.3744, pruned_loss=0.1571, over 7419.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.3912, pruned_loss=0.154, over 1611137.29 frames. ], batch size: 17, lr: 2.69e-02, grad_scale: 8.0 +2023-02-05 21:01:40,343 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15901.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:01:50,917 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.227e+02 3.569e+02 4.742e+02 5.990e+02 9.896e+02, threshold=9.484e+02, percent-clipped=1.0 +2023-02-05 21:01:59,057 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15928.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:02:02,905 INFO [train.py:901] (2/4) Epoch 2, batch 7850, loss[loss=0.3475, simple_loss=0.3992, pruned_loss=0.1479, over 8336.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.3904, pruned_loss=0.1532, over 1608015.87 frames. ], batch size: 25, lr: 2.68e-02, grad_scale: 8.0 +2023-02-05 21:02:15,679 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15953.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:02:36,218 INFO [train.py:901] (2/4) Epoch 2, batch 7900, loss[loss=0.3557, simple_loss=0.3965, pruned_loss=0.1575, over 8464.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3917, pruned_loss=0.1535, over 1614522.06 frames. ], batch size: 27, lr: 2.68e-02, grad_scale: 8.0 +2023-02-05 21:02:39,746 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7835, 1.9813, 2.3096, 0.7233, 2.1991, 1.3201, 0.5473, 1.5270], + device='cuda:2'), covar=tensor([0.0119, 0.0063, 0.0066, 0.0161, 0.0083, 0.0256, 0.0227, 0.0104], + device='cuda:2'), in_proj_covar=tensor([0.0205, 0.0141, 0.0125, 0.0190, 0.0138, 0.0253, 0.0203, 0.0179], + device='cuda:2'), out_proj_covar=tensor([1.1241e-04, 7.5918e-05, 6.9377e-05, 1.0293e-04, 7.9174e-05, 1.5025e-04, + 1.1324e-04, 9.9462e-05], device='cuda:2') +2023-02-05 21:02:58,236 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.267e+02 3.808e+02 4.602e+02 5.936e+02 1.299e+03, threshold=9.205e+02, percent-clipped=9.0 +2023-02-05 21:03:00,387 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16019.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:03:10,210 INFO [train.py:901] (2/4) Epoch 2, batch 7950, loss[loss=0.3495, simple_loss=0.3947, pruned_loss=0.1522, over 8650.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.3932, pruned_loss=0.1542, over 1616064.70 frames. ], batch size: 27, lr: 2.68e-02, grad_scale: 8.0 +2023-02-05 21:03:43,323 INFO [train.py:901] (2/4) Epoch 2, batch 8000, loss[loss=0.3644, simple_loss=0.4034, pruned_loss=0.1627, over 8194.00 frames. ], tot_loss[loss=0.354, simple_loss=0.3952, pruned_loss=0.1564, over 1614360.32 frames. ], batch size: 23, lr: 2.67e-02, grad_scale: 8.0 +2023-02-05 21:03:56,071 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16103.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:04:04,534 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.336e+02 4.123e+02 4.991e+02 6.647e+02 1.461e+03, threshold=9.983e+02, percent-clipped=10.0 +2023-02-05 21:04:16,503 INFO [train.py:901] (2/4) Epoch 2, batch 8050, loss[loss=0.3042, simple_loss=0.3542, pruned_loss=0.1271, over 7529.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.3918, pruned_loss=0.1548, over 1598533.46 frames. ], batch size: 18, lr: 2.67e-02, grad_scale: 8.0 +2023-02-05 21:04:27,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9195, 1.0108, 1.1046, 0.8440, 0.6964, 1.2339, 0.2374, 0.6125], + device='cuda:2'), covar=tensor([0.1506, 0.1665, 0.0813, 0.1481, 0.2160, 0.0579, 0.3270, 0.1928], + device='cuda:2'), in_proj_covar=tensor([0.0109, 0.0096, 0.0082, 0.0146, 0.0145, 0.0080, 0.0150, 0.0110], + device='cuda:2'), out_proj_covar=tensor([1.2640e-04, 1.1616e-04, 9.4297e-05, 1.5998e-04, 1.6178e-04, 9.5338e-05, + 1.6892e-04, 1.3118e-04], device='cuda:2') +2023-02-05 21:04:51,313 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 21:04:55,113 INFO [train.py:901] (2/4) Epoch 3, batch 0, loss[loss=0.3868, simple_loss=0.4165, pruned_loss=0.1785, over 8497.00 frames. ], tot_loss[loss=0.3868, simple_loss=0.4165, pruned_loss=0.1785, over 8497.00 frames. ], batch size: 49, lr: 2.53e-02, grad_scale: 8.0 +2023-02-05 21:04:55,114 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 21:05:06,962 INFO [train.py:935] (2/4) Epoch 3, validation: loss=0.2731, simple_loss=0.3579, pruned_loss=0.09417, over 944034.00 frames. +2023-02-05 21:05:06,963 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 21:05:07,103 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16167.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:05:14,140 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2574, 2.6282, 4.1171, 3.8018, 3.2511, 2.7121, 1.7577, 2.0637], + device='cuda:2'), covar=tensor([0.0700, 0.0863, 0.0144, 0.0277, 0.0368, 0.0370, 0.0563, 0.0782], + device='cuda:2'), in_proj_covar=tensor([0.0540, 0.0461, 0.0352, 0.0404, 0.0496, 0.0430, 0.0455, 0.0464], + 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') +2023-02-05 21:05:23,570 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 21:05:42,767 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.402e+02 4.065e+02 5.070e+02 6.931e+02 1.670e+03, threshold=1.014e+03, percent-clipped=5.0 +2023-02-05 21:05:42,788 INFO [train.py:901] (2/4) Epoch 3, batch 50, loss[loss=0.3457, simple_loss=0.3911, pruned_loss=0.1501, over 8472.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.3937, pruned_loss=0.1574, over 365506.55 frames. ], batch size: 25, lr: 2.53e-02, grad_scale: 4.0 +2023-02-05 21:05:58,800 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 21:06:02,997 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16245.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:06:18,204 INFO [train.py:901] (2/4) Epoch 3, batch 100, loss[loss=0.4973, simple_loss=0.4797, pruned_loss=0.2575, over 6681.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.3957, pruned_loss=0.1589, over 637404.75 frames. ], batch size: 72, lr: 2.53e-02, grad_scale: 4.0 +2023-02-05 21:06:18,929 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 21:06:20,108 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.67 vs. limit=5.0 +2023-02-05 21:06:53,416 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.291e+02 3.520e+02 4.471e+02 5.811e+02 1.196e+03, threshold=8.942e+02, percent-clipped=3.0 +2023-02-05 21:06:53,436 INFO [train.py:901] (2/4) Epoch 3, batch 150, loss[loss=0.3028, simple_loss=0.3628, pruned_loss=0.1214, over 8496.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3917, pruned_loss=0.1535, over 851880.94 frames. ], batch size: 28, lr: 2.52e-02, grad_scale: 4.0 +2023-02-05 21:07:23,168 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16360.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:07:24,938 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16363.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:07:27,430 INFO [train.py:901] (2/4) Epoch 3, batch 200, loss[loss=0.324, simple_loss=0.3674, pruned_loss=0.1402, over 7914.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3924, pruned_loss=0.1532, over 1020458.04 frames. ], batch size: 20, lr: 2.52e-02, grad_scale: 4.0 +2023-02-05 21:07:27,654 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4997, 2.0219, 2.2897, 0.6817, 2.3039, 1.5856, 0.5649, 1.8375], + device='cuda:2'), covar=tensor([0.0142, 0.0061, 0.0096, 0.0188, 0.0096, 0.0239, 0.0238, 0.0079], + device='cuda:2'), in_proj_covar=tensor([0.0206, 0.0145, 0.0125, 0.0192, 0.0134, 0.0254, 0.0210, 0.0178], + device='cuda:2'), out_proj_covar=tensor([1.1189e-04, 7.7989e-05, 6.8373e-05, 1.0217e-04, 7.4906e-05, 1.4812e-04, + 1.1632e-04, 9.7388e-05], device='cuda:2') +2023-02-05 21:07:42,151 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16389.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:08:01,474 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 3.609e+02 4.419e+02 5.456e+02 1.161e+03, threshold=8.837e+02, percent-clipped=3.0 +2023-02-05 21:08:01,494 INFO [train.py:901] (2/4) Epoch 3, batch 250, loss[loss=0.4054, simple_loss=0.4373, pruned_loss=0.1868, over 8030.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.3931, pruned_loss=0.1531, over 1155405.59 frames. ], batch size: 22, lr: 2.51e-02, grad_scale: 4.0 +2023-02-05 21:08:13,913 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 21:08:15,455 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0813, 3.0842, 2.0539, 2.3589, 2.5597, 2.0014, 2.1615, 2.7397], + device='cuda:2'), covar=tensor([0.0915, 0.0394, 0.0724, 0.0707, 0.0549, 0.0862, 0.0972, 0.0709], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0246, 0.0340, 0.0305, 0.0350, 0.0313, 0.0352, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:08:22,557 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 21:08:22,628 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16447.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:08:27,798 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6736, 2.2422, 4.6700, 2.7782, 4.2579, 4.1717, 4.3521, 4.3910], + device='cuda:2'), covar=tensor([0.0243, 0.2031, 0.0232, 0.1205, 0.0640, 0.0292, 0.0269, 0.0258], + device='cuda:2'), in_proj_covar=tensor([0.0214, 0.0403, 0.0262, 0.0303, 0.0370, 0.0290, 0.0276, 0.0302], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 21:08:35,575 INFO [train.py:901] (2/4) Epoch 3, batch 300, loss[loss=0.3435, simple_loss=0.3932, pruned_loss=0.1469, over 7979.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.3935, pruned_loss=0.1536, over 1258969.13 frames. ], batch size: 21, lr: 2.51e-02, grad_scale: 4.0 +2023-02-05 21:08:43,558 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16478.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:09:05,163 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16511.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:09:09,088 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 3.752e+02 4.774e+02 5.919e+02 1.248e+03, threshold=9.549e+02, percent-clipped=6.0 +2023-02-05 21:09:09,108 INFO [train.py:901] (2/4) Epoch 3, batch 350, loss[loss=0.3085, simple_loss=0.3743, pruned_loss=0.1214, over 8248.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.3933, pruned_loss=0.1541, over 1342244.78 frames. ], batch size: 24, lr: 2.51e-02, grad_scale: 4.0 +2023-02-05 21:09:14,648 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1799, 1.1493, 4.2943, 1.7999, 3.5746, 3.5648, 3.8125, 3.7704], + device='cuda:2'), covar=tensor([0.0320, 0.3270, 0.0244, 0.1649, 0.0983, 0.0427, 0.0374, 0.0411], + device='cuda:2'), in_proj_covar=tensor([0.0215, 0.0405, 0.0271, 0.0309, 0.0375, 0.0292, 0.0282, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 21:09:22,196 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-05 21:09:34,321 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.4147, 2.1242, 1.8411, 1.4664, 2.6487, 1.9191, 2.4978, 2.2090], + device='cuda:2'), covar=tensor([0.0854, 0.1491, 0.2092, 0.1774, 0.0764, 0.1721, 0.0875, 0.0788], + device='cuda:2'), in_proj_covar=tensor([0.0207, 0.0233, 0.0267, 0.0230, 0.0203, 0.0229, 0.0197, 0.0196], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:09:40,931 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16562.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:09:44,001 INFO [train.py:901] (2/4) Epoch 3, batch 400, loss[loss=0.3233, simple_loss=0.3857, pruned_loss=0.1304, over 8359.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.3934, pruned_loss=0.1545, over 1400851.77 frames. ], batch size: 24, lr: 2.50e-02, grad_scale: 8.0 +2023-02-05 21:10:18,119 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16616.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:10:18,545 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.210e+02 3.588e+02 4.493e+02 6.059e+02 1.047e+03, threshold=8.987e+02, percent-clipped=2.0 +2023-02-05 21:10:18,565 INFO [train.py:901] (2/4) Epoch 3, batch 450, loss[loss=0.3311, simple_loss=0.3882, pruned_loss=0.137, over 8768.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.3931, pruned_loss=0.154, over 1446955.29 frames. ], batch size: 30, lr: 2.50e-02, grad_scale: 8.0 +2023-02-05 21:10:24,823 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16626.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:10:35,584 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16641.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:10:53,050 INFO [train.py:901] (2/4) Epoch 3, batch 500, loss[loss=0.3492, simple_loss=0.3943, pruned_loss=0.1521, over 8191.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.3932, pruned_loss=0.1536, over 1488938.22 frames. ], batch size: 23, lr: 2.50e-02, grad_scale: 8.0 +2023-02-05 21:11:27,933 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 3.547e+02 4.664e+02 6.145e+02 2.246e+03, threshold=9.327e+02, percent-clipped=7.0 +2023-02-05 21:11:27,953 INFO [train.py:901] (2/4) Epoch 3, batch 550, loss[loss=0.2917, simple_loss=0.3599, pruned_loss=0.1118, over 8081.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3904, pruned_loss=0.1513, over 1511594.36 frames. ], batch size: 21, lr: 2.49e-02, grad_scale: 8.0 +2023-02-05 21:11:38,639 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16733.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:11:39,412 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16734.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:11:56,556 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16759.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:12:01,682 INFO [train.py:901] (2/4) Epoch 3, batch 600, loss[loss=0.3535, simple_loss=0.397, pruned_loss=0.155, over 8251.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3911, pruned_loss=0.152, over 1538533.85 frames. ], batch size: 24, lr: 2.49e-02, grad_scale: 8.0 +2023-02-05 21:12:16,376 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 21:12:20,620 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3300, 1.5410, 2.8793, 0.9352, 2.0308, 1.6511, 1.4178, 1.6648], + device='cuda:2'), covar=tensor([0.1308, 0.1711, 0.0419, 0.2581, 0.1125, 0.1877, 0.1106, 0.1671], + device='cuda:2'), in_proj_covar=tensor([0.0427, 0.0401, 0.0460, 0.0488, 0.0535, 0.0468, 0.0429, 0.0534], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 21:12:21,238 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3969, 2.1589, 1.5793, 1.5151, 1.7147, 1.7607, 2.0609, 1.8499], + device='cuda:2'), covar=tensor([0.0830, 0.1202, 0.1918, 0.1567, 0.0863, 0.1482, 0.0938, 0.0634], + device='cuda:2'), in_proj_covar=tensor([0.0208, 0.0237, 0.0270, 0.0236, 0.0207, 0.0234, 0.0201, 0.0199], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:12:22,377 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7138, 1.7649, 4.2316, 2.0182, 2.5059, 5.1036, 4.4843, 4.5760], + device='cuda:2'), covar=tensor([0.0926, 0.1443, 0.0228, 0.1750, 0.0705, 0.0136, 0.0240, 0.0385], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0267, 0.0200, 0.0268, 0.0200, 0.0167, 0.0164, 0.0241], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 21:12:36,656 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.715e+02 4.834e+02 5.984e+02 1.404e+03, threshold=9.668e+02, percent-clipped=7.0 +2023-02-05 21:12:36,676 INFO [train.py:901] (2/4) Epoch 3, batch 650, loss[loss=0.3438, simple_loss=0.3912, pruned_loss=0.1482, over 8343.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3899, pruned_loss=0.1521, over 1550364.40 frames. ], batch size: 26, lr: 2.49e-02, grad_scale: 8.0 +2023-02-05 21:12:37,560 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16818.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:12:54,030 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16843.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:12:57,234 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16848.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:10,252 INFO [train.py:901] (2/4) Epoch 3, batch 700, loss[loss=0.3465, simple_loss=0.3983, pruned_loss=0.1474, over 8251.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.3915, pruned_loss=0.1538, over 1564797.02 frames. ], batch size: 24, lr: 2.48e-02, grad_scale: 8.0 +2023-02-05 21:13:20,420 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16882.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:23,045 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16886.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:13:38,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16907.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:44,828 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.086e+02 3.932e+02 4.613e+02 6.231e+02 2.383e+03, threshold=9.225e+02, percent-clipped=5.0 +2023-02-05 21:13:44,850 INFO [train.py:901] (2/4) Epoch 3, batch 750, loss[loss=0.4173, simple_loss=0.4392, pruned_loss=0.1977, over 8667.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.3924, pruned_loss=0.1539, over 1579189.35 frames. ], batch size: 49, lr: 2.48e-02, grad_scale: 8.0 +2023-02-05 21:13:49,809 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16924.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:57,352 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-05 21:13:59,079 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 21:14:07,691 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 21:14:19,211 INFO [train.py:901] (2/4) Epoch 3, batch 800, loss[loss=0.3278, simple_loss=0.3768, pruned_loss=0.1394, over 8134.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.3917, pruned_loss=0.1519, over 1590978.83 frames. ], batch size: 22, lr: 2.48e-02, grad_scale: 8.0 +2023-02-05 21:14:53,628 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.211e+02 3.452e+02 4.368e+02 5.287e+02 1.393e+03, threshold=8.735e+02, percent-clipped=4.0 +2023-02-05 21:14:53,649 INFO [train.py:901] (2/4) Epoch 3, batch 850, loss[loss=0.3083, simple_loss=0.3602, pruned_loss=0.1282, over 7671.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3907, pruned_loss=0.1515, over 1599862.25 frames. ], batch size: 19, lr: 2.47e-02, grad_scale: 8.0 +2023-02-05 21:15:01,681 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2797, 2.7884, 2.2983, 3.2744, 1.9316, 2.0091, 2.1444, 2.7969], + device='cuda:2'), covar=tensor([0.0901, 0.0877, 0.1006, 0.0360, 0.1476, 0.1605, 0.1409, 0.0731], + device='cuda:2'), in_proj_covar=tensor([0.0316, 0.0330, 0.0314, 0.0229, 0.0301, 0.0330, 0.0354, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:15:19,772 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5327, 2.4621, 1.5560, 2.0267, 2.0377, 1.0831, 1.5298, 1.9586], + device='cuda:2'), covar=tensor([0.1296, 0.0420, 0.0980, 0.0582, 0.0703, 0.1245, 0.1003, 0.0773], + device='cuda:2'), in_proj_covar=tensor([0.0364, 0.0249, 0.0346, 0.0305, 0.0350, 0.0315, 0.0355, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:15:28,355 INFO [train.py:901] (2/4) Epoch 3, batch 900, loss[loss=0.3613, simple_loss=0.403, pruned_loss=0.1598, over 8500.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.3911, pruned_loss=0.1518, over 1604536.98 frames. ], batch size: 26, lr: 2.47e-02, grad_scale: 8.0 +2023-02-05 21:15:53,797 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17104.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:16:02,281 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.375e+02 3.695e+02 4.540e+02 5.760e+02 9.795e+02, threshold=9.080e+02, percent-clipped=3.0 +2023-02-05 21:16:02,301 INFO [train.py:901] (2/4) Epoch 3, batch 950, loss[loss=0.2867, simple_loss=0.347, pruned_loss=0.1133, over 7930.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3906, pruned_loss=0.1513, over 1606181.19 frames. ], batch size: 20, lr: 2.47e-02, grad_scale: 8.0 +2023-02-05 21:16:10,473 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17129.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:16:25,706 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 21:16:36,857 INFO [train.py:901] (2/4) Epoch 3, batch 1000, loss[loss=0.3981, simple_loss=0.426, pruned_loss=0.1851, over 8458.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.3925, pruned_loss=0.1531, over 1605285.89 frames. ], batch size: 50, lr: 2.46e-02, grad_scale: 8.0 +2023-02-05 21:16:57,949 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 21:17:03,570 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17207.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:17:07,058 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5670, 2.5036, 1.5752, 1.9535, 2.1263, 1.2231, 1.5105, 2.1599], + device='cuda:2'), covar=tensor([0.1549, 0.0515, 0.1106, 0.0724, 0.0700, 0.1383, 0.1339, 0.0999], + device='cuda:2'), in_proj_covar=tensor([0.0368, 0.0250, 0.0349, 0.0311, 0.0346, 0.0320, 0.0361, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:17:08,327 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5284, 1.2859, 1.3429, 1.1889, 1.1734, 1.3472, 1.1675, 1.0993], + device='cuda:2'), covar=tensor([0.0842, 0.1605, 0.2191, 0.1661, 0.0821, 0.1731, 0.0998, 0.0758], + device='cuda:2'), in_proj_covar=tensor([0.0200, 0.0231, 0.0265, 0.0230, 0.0201, 0.0231, 0.0198, 0.0194], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:17:10,140 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 4.093e+02 4.952e+02 6.088e+02 1.030e+03, threshold=9.904e+02, percent-clipped=7.0 +2023-02-05 21:17:10,160 INFO [train.py:901] (2/4) Epoch 3, batch 1050, loss[loss=0.3539, simple_loss=0.3889, pruned_loss=0.1595, over 8078.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.3932, pruned_loss=0.1535, over 1608544.28 frames. ], batch size: 21, lr: 2.46e-02, grad_scale: 8.0 +2023-02-05 21:17:10,171 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 21:17:11,624 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2568, 4.3499, 3.7648, 1.9395, 3.7620, 3.8816, 3.9415, 3.2561], + device='cuda:2'), covar=tensor([0.0843, 0.0494, 0.0824, 0.4312, 0.0591, 0.0748, 0.0997, 0.0716], + device='cuda:2'), in_proj_covar=tensor([0.0342, 0.0228, 0.0267, 0.0358, 0.0251, 0.0199, 0.0259, 0.0188], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:17:19,697 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17230.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:17:45,190 INFO [train.py:901] (2/4) Epoch 3, batch 1100, loss[loss=0.3523, simple_loss=0.4029, pruned_loss=0.1508, over 8132.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.3921, pruned_loss=0.1523, over 1612415.92 frames. ], batch size: 22, lr: 2.46e-02, grad_scale: 8.0 +2023-02-05 21:17:45,923 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17268.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:18:01,487 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-05 21:18:19,093 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.272e+02 3.840e+02 4.434e+02 5.714e+02 1.415e+03, threshold=8.869e+02, percent-clipped=3.0 +2023-02-05 21:18:19,113 INFO [train.py:901] (2/4) Epoch 3, batch 1150, loss[loss=0.295, simple_loss=0.3392, pruned_loss=0.1254, over 7256.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.3917, pruned_loss=0.1516, over 1615890.31 frames. ], batch size: 16, lr: 2.45e-02, grad_scale: 8.0 +2023-02-05 21:18:22,484 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 21:18:38,635 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17345.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:18:52,840 INFO [train.py:901] (2/4) Epoch 3, batch 1200, loss[loss=0.4143, simple_loss=0.443, pruned_loss=0.1928, over 8192.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.3927, pruned_loss=0.1526, over 1618697.57 frames. ], batch size: 23, lr: 2.45e-02, grad_scale: 8.0 +2023-02-05 21:19:01,229 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2023-02-05 21:19:02,176 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17380.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:19:04,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17383.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:19:17,666 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17401.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:19:21,582 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1621, 4.2432, 3.6845, 1.4596, 3.6571, 3.5095, 3.8795, 3.1211], + device='cuda:2'), covar=tensor([0.0757, 0.0551, 0.0925, 0.4198, 0.0544, 0.0586, 0.1201, 0.0593], + device='cuda:2'), in_proj_covar=tensor([0.0338, 0.0227, 0.0269, 0.0352, 0.0245, 0.0196, 0.0252, 0.0184], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:19:28,364 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 3.772e+02 4.989e+02 5.905e+02 9.785e+02, threshold=9.978e+02, percent-clipped=4.0 +2023-02-05 21:19:28,384 INFO [train.py:901] (2/4) Epoch 3, batch 1250, loss[loss=0.3271, simple_loss=0.3704, pruned_loss=0.1419, over 7922.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3908, pruned_loss=0.1512, over 1618157.85 frames. ], batch size: 20, lr: 2.45e-02, grad_scale: 8.0 +2023-02-05 21:20:02,610 INFO [train.py:901] (2/4) Epoch 3, batch 1300, loss[loss=0.3617, simple_loss=0.4072, pruned_loss=0.1581, over 8503.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.3933, pruned_loss=0.1527, over 1620908.40 frames. ], batch size: 28, lr: 2.44e-02, grad_scale: 8.0 +2023-02-05 21:20:19,165 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2023-02-05 21:20:37,547 INFO [train.py:901] (2/4) Epoch 3, batch 1350, loss[loss=0.3215, simple_loss=0.3857, pruned_loss=0.1287, over 8312.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.3926, pruned_loss=0.1532, over 1614728.05 frames. ], batch size: 25, lr: 2.44e-02, grad_scale: 4.0 +2023-02-05 21:20:38,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 4.258e+02 5.812e+02 8.345e+02 8.746e+03, threshold=1.162e+03, percent-clipped=16.0 +2023-02-05 21:21:00,236 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17551.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:21:11,045 INFO [train.py:901] (2/4) Epoch 3, batch 1400, loss[loss=0.2801, simple_loss=0.3302, pruned_loss=0.115, over 7538.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.3906, pruned_loss=0.152, over 1612862.66 frames. ], batch size: 18, lr: 2.44e-02, grad_scale: 4.0 +2023-02-05 21:21:34,757 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17601.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:21:46,905 INFO [train.py:901] (2/4) Epoch 3, batch 1450, loss[loss=0.2959, simple_loss=0.3541, pruned_loss=0.1188, over 8355.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3898, pruned_loss=0.1514, over 1612722.39 frames. ], batch size: 24, lr: 2.43e-02, grad_scale: 4.0 +2023-02-05 21:21:47,590 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.138e+02 3.309e+02 4.161e+02 5.035e+02 1.114e+03, threshold=8.322e+02, percent-clipped=0.0 +2023-02-05 21:21:48,907 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 21:21:53,236 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17626.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:22:02,526 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17639.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:22:19,041 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17664.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:22:20,427 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17666.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:22:20,937 INFO [train.py:901] (2/4) Epoch 3, batch 1500, loss[loss=0.3558, simple_loss=0.4122, pruned_loss=0.1497, over 8646.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3898, pruned_loss=0.1516, over 1612444.05 frames. ], batch size: 34, lr: 2.43e-02, grad_scale: 4.0 +2023-02-05 21:22:56,184 INFO [train.py:901] (2/4) Epoch 3, batch 1550, loss[loss=0.3554, simple_loss=0.3989, pruned_loss=0.156, over 8078.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3885, pruned_loss=0.1505, over 1609115.67 frames. ], batch size: 21, lr: 2.43e-02, grad_scale: 4.0 +2023-02-05 21:22:56,831 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.415e+02 3.678e+02 4.620e+02 5.892e+02 1.697e+03, threshold=9.239e+02, percent-clipped=9.0 +2023-02-05 21:22:57,023 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:23:01,086 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17724.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:23:16,060 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17745.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:23:17,957 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17748.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:23:30,932 INFO [train.py:901] (2/4) Epoch 3, batch 1600, loss[loss=0.3308, simple_loss=0.3755, pruned_loss=0.143, over 7675.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3882, pruned_loss=0.1501, over 1607795.46 frames. ], batch size: 19, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:24:05,145 INFO [train.py:901] (2/4) Epoch 3, batch 1650, loss[loss=0.3152, simple_loss=0.3864, pruned_loss=0.122, over 8195.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3878, pruned_loss=0.1489, over 1614862.01 frames. ], batch size: 23, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:24:05,806 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 3.309e+02 4.132e+02 5.477e+02 8.650e+02, threshold=8.264e+02, percent-clipped=0.0 +2023-02-05 21:24:20,681 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17839.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:24:21,857 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5664, 5.4627, 4.7746, 1.5514, 4.9307, 4.9875, 5.0699, 4.4192], + device='cuda:2'), covar=tensor([0.0683, 0.0489, 0.1015, 0.5330, 0.0417, 0.0601, 0.1069, 0.0597], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0232, 0.0278, 0.0364, 0.0257, 0.0210, 0.0269, 0.0192], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:24:35,253 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17860.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:24:39,697 INFO [train.py:901] (2/4) Epoch 3, batch 1700, loss[loss=0.3891, simple_loss=0.417, pruned_loss=0.1806, over 8581.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3881, pruned_loss=0.149, over 1616384.14 frames. ], batch size: 34, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:24:42,706 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-05 21:25:09,505 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 +2023-02-05 21:25:13,896 INFO [train.py:901] (2/4) Epoch 3, batch 1750, loss[loss=0.3689, simple_loss=0.4122, pruned_loss=0.1628, over 8289.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3872, pruned_loss=0.1485, over 1616948.04 frames. ], batch size: 23, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:25:14,598 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 3.998e+02 5.161e+02 6.686e+02 1.470e+03, threshold=1.032e+03, percent-clipped=12.0 +2023-02-05 21:25:17,622 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17922.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:25:35,809 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17947.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:25:48,896 INFO [train.py:901] (2/4) Epoch 3, batch 1800, loss[loss=0.3301, simple_loss=0.3889, pruned_loss=0.1357, over 8341.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3877, pruned_loss=0.1493, over 1616373.22 frames. ], batch size: 26, lr: 2.41e-02, grad_scale: 8.0 +2023-02-05 21:25:57,025 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17978.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:26:25,065 INFO [train.py:901] (2/4) Epoch 3, batch 1850, loss[loss=0.2961, simple_loss=0.3482, pruned_loss=0.1219, over 7647.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.3892, pruned_loss=0.1506, over 1616689.37 frames. ], batch size: 19, lr: 2.41e-02, grad_scale: 8.0 +2023-02-05 21:26:25,636 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.340e+02 3.564e+02 4.327e+02 5.819e+02 2.228e+03, threshold=8.654e+02, percent-clipped=8.0 +2023-02-05 21:26:39,173 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9617, 4.0756, 3.5775, 1.5986, 3.4744, 3.2775, 3.7085, 3.0696], + device='cuda:2'), covar=tensor([0.0884, 0.0563, 0.0964, 0.4797, 0.0694, 0.0847, 0.0994, 0.0770], + device='cuda:2'), in_proj_covar=tensor([0.0340, 0.0232, 0.0281, 0.0365, 0.0246, 0.0204, 0.0258, 0.0188], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:26:55,942 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18062.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:26:59,906 INFO [train.py:901] (2/4) Epoch 3, batch 1900, loss[loss=0.3443, simple_loss=0.4003, pruned_loss=0.1441, over 8509.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3885, pruned_loss=0.1501, over 1615758.01 frames. ], batch size: 28, lr: 2.41e-02, grad_scale: 8.0 +2023-02-05 21:27:17,421 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18092.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:27:19,613 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18095.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:27:24,169 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 21:27:32,751 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 21:27:34,376 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18116.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:27:34,800 INFO [train.py:901] (2/4) Epoch 3, batch 1950, loss[loss=0.3197, simple_loss=0.3756, pruned_loss=0.1319, over 8039.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3875, pruned_loss=0.1491, over 1615296.34 frames. ], batch size: 22, lr: 2.40e-02, grad_scale: 8.0 +2023-02-05 21:27:35,484 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.131e+02 3.385e+02 4.094e+02 5.586e+02 1.173e+03, threshold=8.188e+02, percent-clipped=3.0 +2023-02-05 21:27:36,197 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 21:27:37,027 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18120.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:27:39,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1692, 4.3007, 3.6948, 1.6964, 3.7001, 3.4570, 4.0261, 3.0502], + device='cuda:2'), covar=tensor([0.1040, 0.0585, 0.1071, 0.5133, 0.0652, 0.0747, 0.1222, 0.0798], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0240, 0.0287, 0.0378, 0.0252, 0.0215, 0.0270, 0.0196], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:27:51,205 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18141.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:27:55,040 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 21:28:02,893 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-05 21:28:07,359 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1411, 1.5761, 1.3102, 1.6648, 1.3894, 1.1024, 1.1266, 1.4628], + device='cuda:2'), covar=tensor([0.0957, 0.0512, 0.0970, 0.0550, 0.0779, 0.1138, 0.0951, 0.0702], + device='cuda:2'), in_proj_covar=tensor([0.0365, 0.0251, 0.0343, 0.0308, 0.0347, 0.0308, 0.0356, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:28:09,116 INFO [train.py:901] (2/4) Epoch 3, batch 2000, loss[loss=0.4709, simple_loss=0.482, pruned_loss=0.2299, over 8396.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3879, pruned_loss=0.1499, over 1617748.92 frames. ], batch size: 48, lr: 2.40e-02, grad_scale: 8.0 +2023-02-05 21:28:17,041 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18177.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:27,336 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18192.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:28,012 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18193.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:38,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18207.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:44,799 INFO [train.py:901] (2/4) Epoch 3, batch 2050, loss[loss=0.2619, simple_loss=0.3097, pruned_loss=0.107, over 7682.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3876, pruned_loss=0.1497, over 1619471.65 frames. ], batch size: 18, lr: 2.40e-02, grad_scale: 8.0 +2023-02-05 21:28:46,138 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.474e+02 3.817e+02 4.995e+02 6.129e+02 1.664e+03, threshold=9.991e+02, percent-clipped=7.0 +2023-02-05 21:29:19,889 INFO [train.py:901] (2/4) Epoch 3, batch 2100, loss[loss=0.3824, simple_loss=0.4135, pruned_loss=0.1756, over 8332.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3886, pruned_loss=0.1504, over 1622639.56 frames. ], batch size: 26, lr: 2.39e-02, grad_scale: 8.0 +2023-02-05 21:29:55,197 INFO [train.py:901] (2/4) Epoch 3, batch 2150, loss[loss=0.3342, simple_loss=0.3963, pruned_loss=0.136, over 8339.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3892, pruned_loss=0.1505, over 1622368.64 frames. ], batch size: 26, lr: 2.39e-02, grad_scale: 8.0 +2023-02-05 21:29:55,882 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.297e+02 3.744e+02 4.718e+02 5.936e+02 1.452e+03, threshold=9.436e+02, percent-clipped=4.0 +2023-02-05 21:29:59,202 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18322.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:30:13,276 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2952, 1.5688, 2.3630, 0.9873, 1.7345, 1.5349, 1.3361, 1.6377], + device='cuda:2'), covar=tensor([0.1094, 0.1205, 0.0434, 0.2096, 0.0885, 0.1663, 0.1037, 0.1119], + device='cuda:2'), in_proj_covar=tensor([0.0436, 0.0398, 0.0470, 0.0487, 0.0535, 0.0478, 0.0426, 0.0540], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 21:30:31,069 INFO [train.py:901] (2/4) Epoch 3, batch 2200, loss[loss=0.4312, simple_loss=0.444, pruned_loss=0.2093, over 8137.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3901, pruned_loss=0.1509, over 1620904.12 frames. ], batch size: 22, lr: 2.39e-02, grad_scale: 8.0 +2023-02-05 21:30:38,558 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-05 21:31:06,829 INFO [train.py:901] (2/4) Epoch 3, batch 2250, loss[loss=0.2884, simple_loss=0.3328, pruned_loss=0.122, over 7529.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3903, pruned_loss=0.151, over 1621700.74 frames. ], batch size: 18, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:31:07,500 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.379e+02 3.424e+02 4.222e+02 5.561e+02 1.530e+03, threshold=8.445e+02, percent-clipped=2.0 +2023-02-05 21:31:07,729 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8164, 2.2974, 1.5895, 2.0915, 1.7372, 1.2803, 1.5982, 2.0171], + device='cuda:2'), covar=tensor([0.1126, 0.0539, 0.0976, 0.0632, 0.0845, 0.1203, 0.1100, 0.0746], + device='cuda:2'), in_proj_covar=tensor([0.0368, 0.0254, 0.0340, 0.0313, 0.0348, 0.0308, 0.0358, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:31:18,177 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18433.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:21,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18437.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:36,360 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18458.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:39,797 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18463.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:42,291 INFO [train.py:901] (2/4) Epoch 3, batch 2300, loss[loss=0.39, simple_loss=0.4269, pruned_loss=0.1766, over 8191.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3886, pruned_loss=0.1493, over 1620234.84 frames. ], batch size: 23, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:31:56,792 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18488.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:11,144 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18508.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:17,111 INFO [train.py:901] (2/4) Epoch 3, batch 2350, loss[loss=0.2818, simple_loss=0.3371, pruned_loss=0.1133, over 8093.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3872, pruned_loss=0.1488, over 1621784.06 frames. ], batch size: 21, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:32:17,761 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.449e+02 3.759e+02 4.661e+02 5.652e+02 9.227e+02, threshold=9.323e+02, percent-clipped=1.0 +2023-02-05 21:32:23,349 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18526.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:32:30,480 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18536.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:31,168 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18537.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:37,151 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3790, 1.6686, 2.1834, 1.3286, 1.0320, 1.9216, 0.2971, 1.0209], + device='cuda:2'), covar=tensor([0.2970, 0.2171, 0.0748, 0.2212, 0.5121, 0.1089, 0.6199, 0.2139], + device='cuda:2'), in_proj_covar=tensor([0.0110, 0.0102, 0.0081, 0.0153, 0.0160, 0.0078, 0.0152, 0.0111], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:32:51,314 INFO [train.py:901] (2/4) Epoch 3, batch 2400, loss[loss=0.4109, simple_loss=0.4305, pruned_loss=0.1957, over 7973.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3875, pruned_loss=0.1495, over 1623929.42 frames. ], batch size: 21, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:32:59,586 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-05 21:33:03,868 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2023, 1.4080, 2.2916, 0.9981, 2.1271, 2.3691, 2.3801, 1.9458], + device='cuda:2'), covar=tensor([0.1219, 0.1137, 0.0469, 0.2047, 0.0538, 0.0465, 0.0419, 0.0863], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0261, 0.0198, 0.0259, 0.0200, 0.0168, 0.0167, 0.0246], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 21:33:20,500 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2023-02-05 21:33:24,822 INFO [train.py:901] (2/4) Epoch 3, batch 2450, loss[loss=0.3436, simple_loss=0.3618, pruned_loss=0.1627, over 7536.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3882, pruned_loss=0.1501, over 1622622.07 frames. ], batch size: 18, lr: 2.37e-02, grad_scale: 8.0 +2023-02-05 21:33:25,539 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 3.618e+02 4.763e+02 6.456e+02 1.024e+03, threshold=9.527e+02, percent-clipped=2.0 +2023-02-05 21:33:49,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18651.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:33:49,832 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18652.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:33:59,418 INFO [train.py:901] (2/4) Epoch 3, batch 2500, loss[loss=0.3269, simple_loss=0.3782, pruned_loss=0.1378, over 8582.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3865, pruned_loss=0.1484, over 1617446.87 frames. ], batch size: 39, lr: 2.37e-02, grad_scale: 8.0 +2023-02-05 21:34:17,682 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18692.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:34:18,343 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18693.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:34:28,091 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8083, 2.6522, 1.8314, 2.1732, 2.0540, 1.4471, 1.7517, 2.2284], + device='cuda:2'), covar=tensor([0.1233, 0.0403, 0.0859, 0.0649, 0.0721, 0.1141, 0.1018, 0.0826], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0243, 0.0337, 0.0308, 0.0331, 0.0309, 0.0349, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:34:33,853 INFO [train.py:901] (2/4) Epoch 3, batch 2550, loss[loss=0.3455, simple_loss=0.3911, pruned_loss=0.15, over 8193.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3876, pruned_loss=0.1493, over 1617584.65 frames. ], batch size: 23, lr: 2.37e-02, grad_scale: 8.0 +2023-02-05 21:34:34,506 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 3.889e+02 4.529e+02 5.619e+02 1.309e+03, threshold=9.058e+02, percent-clipped=5.0 +2023-02-05 21:34:34,733 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18718.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:35:08,025 INFO [train.py:901] (2/4) Epoch 3, batch 2600, loss[loss=0.3459, simple_loss=0.3904, pruned_loss=0.1507, over 8436.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3854, pruned_loss=0.1478, over 1610692.61 frames. ], batch size: 27, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:35:28,075 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.94 vs. limit=5.0 +2023-02-05 21:35:35,642 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1185, 1.2242, 4.0960, 1.5347, 2.1931, 4.8024, 4.4458, 4.3569], + device='cuda:2'), covar=tensor([0.1381, 0.1818, 0.0277, 0.2101, 0.0826, 0.0317, 0.0309, 0.0529], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0264, 0.0205, 0.0264, 0.0206, 0.0173, 0.0174, 0.0250], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 21:35:44,450 INFO [train.py:901] (2/4) Epoch 3, batch 2650, loss[loss=0.3089, simple_loss=0.3579, pruned_loss=0.1299, over 7809.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3861, pruned_loss=0.1482, over 1614860.55 frames. ], batch size: 19, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:35:45,136 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.426e+02 4.272e+02 5.708e+02 1.020e+03, threshold=8.544e+02, percent-clipped=5.0 +2023-02-05 21:36:08,384 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18852.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:36:12,057 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.6743, 2.9007, 4.3118, 3.8412, 3.5216, 2.8222, 2.1101, 2.4835], + device='cuda:2'), covar=tensor([0.0617, 0.0980, 0.0209, 0.0318, 0.0396, 0.0392, 0.0526, 0.0872], + device='cuda:2'), in_proj_covar=tensor([0.0571, 0.0498, 0.0400, 0.0449, 0.0546, 0.0459, 0.0486, 0.0483], + 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') +2023-02-05 21:36:19,356 INFO [train.py:901] (2/4) Epoch 3, batch 2700, loss[loss=0.3828, simple_loss=0.4047, pruned_loss=0.1805, over 6829.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.386, pruned_loss=0.1479, over 1610362.16 frames. ], batch size: 71, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:36:21,520 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18870.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:36:29,668 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8110, 2.1208, 1.7477, 2.6671, 1.2392, 1.2884, 2.0409, 2.2403], + device='cuda:2'), covar=tensor([0.1113, 0.1500, 0.1741, 0.0624, 0.2068, 0.2690, 0.1889, 0.1147], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0312, 0.0310, 0.0230, 0.0290, 0.0316, 0.0336, 0.0303], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:36:29,702 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4480, 1.9573, 1.8447, 0.5780, 1.7926, 1.3787, 0.4328, 1.9586], + device='cuda:2'), covar=tensor([0.0130, 0.0071, 0.0073, 0.0150, 0.0119, 0.0224, 0.0205, 0.0054], + device='cuda:2'), in_proj_covar=tensor([0.0217, 0.0153, 0.0131, 0.0199, 0.0150, 0.0269, 0.0211, 0.0179], + device='cuda:2'), out_proj_covar=tensor([1.0840e-04, 7.6736e-05, 6.4520e-05, 9.6488e-05, 7.7425e-05, 1.4440e-04, + 1.0793e-04, 9.0180e-05], device='cuda:2') +2023-02-05 21:36:40,055 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-05 21:36:42,538 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9905, 1.6293, 3.3515, 1.0254, 1.9711, 3.5933, 3.4078, 3.1188], + device='cuda:2'), covar=tensor([0.1075, 0.1352, 0.0287, 0.2203, 0.0794, 0.0267, 0.0312, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0258, 0.0200, 0.0258, 0.0205, 0.0170, 0.0173, 0.0246], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 21:36:47,163 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18907.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:36:47,843 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18908.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:36:53,561 INFO [train.py:901] (2/4) Epoch 3, batch 2750, loss[loss=0.3392, simple_loss=0.3638, pruned_loss=0.1573, over 7422.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3863, pruned_loss=0.1482, over 1611024.92 frames. ], batch size: 17, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:36:54,225 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 3.360e+02 4.052e+02 5.079e+02 9.265e+02, threshold=8.105e+02, percent-clipped=2.0 +2023-02-05 21:37:05,008 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18932.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:05,660 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18933.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:15,749 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18948.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:28,077 INFO [train.py:901] (2/4) Epoch 3, batch 2800, loss[loss=0.339, simple_loss=0.39, pruned_loss=0.1439, over 8595.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3863, pruned_loss=0.1479, over 1615446.26 frames. ], batch size: 39, lr: 2.35e-02, grad_scale: 8.0 +2023-02-05 21:37:28,265 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18967.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:31,702 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1972, 3.5989, 1.8218, 2.0820, 2.6103, 1.4368, 1.7274, 2.8791], + device='cuda:2'), covar=tensor([0.1251, 0.0314, 0.1131, 0.0954, 0.0894, 0.1369, 0.1387, 0.0997], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0241, 0.0336, 0.0303, 0.0340, 0.0317, 0.0351, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:37:41,140 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18985.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:37:49,287 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7201, 2.0361, 3.4092, 2.7864, 2.6446, 1.9214, 1.3819, 1.4011], + device='cuda:2'), covar=tensor([0.0738, 0.1059, 0.0189, 0.0383, 0.0429, 0.0492, 0.0653, 0.0968], + device='cuda:2'), in_proj_covar=tensor([0.0569, 0.0499, 0.0397, 0.0447, 0.0550, 0.0462, 0.0485, 0.0490], + 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') +2023-02-05 21:38:03,342 INFO [train.py:901] (2/4) Epoch 3, batch 2850, loss[loss=0.3488, simple_loss=0.3826, pruned_loss=0.1574, over 7979.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.385, pruned_loss=0.147, over 1607822.56 frames. ], batch size: 21, lr: 2.35e-02, grad_scale: 8.0 +2023-02-05 21:38:03,917 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 3.511e+02 4.402e+02 5.555e+02 1.104e+03, threshold=8.804e+02, percent-clipped=5.0 +2023-02-05 21:38:04,210 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6593, 2.1484, 3.3011, 2.9373, 2.8634, 2.2026, 1.5611, 2.1291], + device='cuda:2'), covar=tensor([0.0553, 0.0888, 0.0153, 0.0282, 0.0334, 0.0371, 0.0522, 0.0668], + device='cuda:2'), in_proj_covar=tensor([0.0575, 0.0502, 0.0399, 0.0451, 0.0555, 0.0463, 0.0485, 0.0493], + 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') +2023-02-05 21:38:15,997 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19036.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:38:37,619 INFO [train.py:901] (2/4) Epoch 3, batch 2900, loss[loss=0.3171, simple_loss=0.3662, pruned_loss=0.134, over 8355.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3866, pruned_loss=0.148, over 1608159.67 frames. ], batch size: 24, lr: 2.35e-02, grad_scale: 8.0 +2023-02-05 21:39:00,010 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5466, 1.9226, 2.0280, 1.6073, 1.1570, 2.1454, 0.3729, 1.2952], + device='cuda:2'), covar=tensor([0.2756, 0.2212, 0.1541, 0.2293, 0.4740, 0.0654, 0.5949, 0.2214], + device='cuda:2'), in_proj_covar=tensor([0.0111, 0.0100, 0.0080, 0.0147, 0.0163, 0.0077, 0.0149, 0.0109], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:39:02,476 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 21:39:04,034 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0741, 3.5006, 2.0819, 2.3250, 2.7767, 1.9749, 1.7832, 2.6272], + device='cuda:2'), covar=tensor([0.1219, 0.0406, 0.0819, 0.0841, 0.0562, 0.0952, 0.1254, 0.0940], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0243, 0.0334, 0.0307, 0.0340, 0.0315, 0.0346, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:39:11,756 INFO [train.py:901] (2/4) Epoch 3, batch 2950, loss[loss=0.3708, simple_loss=0.4117, pruned_loss=0.1649, over 8185.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3859, pruned_loss=0.1475, over 1608472.81 frames. ], batch size: 23, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:39:12,415 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.179e+02 3.613e+02 4.498e+02 5.900e+02 1.326e+03, threshold=8.996e+02, percent-clipped=8.0 +2023-02-05 21:39:32,495 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19147.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:39:35,236 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19151.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:39:46,226 INFO [train.py:901] (2/4) Epoch 3, batch 3000, loss[loss=0.3521, simple_loss=0.3832, pruned_loss=0.1606, over 7798.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3861, pruned_loss=0.1475, over 1614511.96 frames. ], batch size: 19, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:39:46,226 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 21:39:58,667 INFO [train.py:935] (2/4) Epoch 3, validation: loss=0.2584, simple_loss=0.3473, pruned_loss=0.08481, over 944034.00 frames. +2023-02-05 21:39:58,667 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 21:40:29,494 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2023-02-05 21:40:33,769 INFO [train.py:901] (2/4) Epoch 3, batch 3050, loss[loss=0.2731, simple_loss=0.3448, pruned_loss=0.1007, over 7805.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3855, pruned_loss=0.1469, over 1617244.04 frames. ], batch size: 20, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:40:34,454 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 3.526e+02 4.458e+02 6.217e+02 1.354e+03, threshold=8.917e+02, percent-clipped=3.0 +2023-02-05 21:40:37,621 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-05 21:40:38,079 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19223.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:40:50,812 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19241.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:40:55,430 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19248.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:41:07,578 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19266.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:41:07,999 INFO [train.py:901] (2/4) Epoch 3, batch 3100, loss[loss=0.3384, simple_loss=0.3766, pruned_loss=0.1501, over 8037.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.384, pruned_loss=0.1459, over 1613006.69 frames. ], batch size: 20, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:41:25,129 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-05 21:41:26,059 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19292.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:41:43,833 INFO [train.py:901] (2/4) Epoch 3, batch 3150, loss[loss=0.2924, simple_loss=0.3324, pruned_loss=0.1262, over 8088.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3848, pruned_loss=0.1464, over 1612711.21 frames. ], batch size: 21, lr: 2.33e-02, grad_scale: 8.0 +2023-02-05 21:41:44,474 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.556e+02 3.507e+02 4.387e+02 6.193e+02 1.521e+03, threshold=8.773e+02, percent-clipped=4.0 +2023-02-05 21:42:00,580 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8886, 1.3499, 3.3864, 1.2093, 2.3913, 3.7491, 3.5149, 3.2702], + device='cuda:2'), covar=tensor([0.1063, 0.1539, 0.0324, 0.1892, 0.0716, 0.0208, 0.0300, 0.0521], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0264, 0.0204, 0.0268, 0.0206, 0.0172, 0.0175, 0.0250], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 21:42:11,539 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.68 vs. limit=5.0 +2023-02-05 21:42:17,836 INFO [train.py:901] (2/4) Epoch 3, batch 3200, loss[loss=0.2973, simple_loss=0.342, pruned_loss=0.1263, over 7525.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3843, pruned_loss=0.1464, over 1607013.47 frames. ], batch size: 18, lr: 2.33e-02, grad_scale: 8.0 +2023-02-05 21:42:45,881 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19407.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:42:45,922 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19407.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:42:49,237 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19412.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:42:53,531 INFO [train.py:901] (2/4) Epoch 3, batch 3250, loss[loss=0.3204, simple_loss=0.3902, pruned_loss=0.1253, over 8104.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3848, pruned_loss=0.1465, over 1609515.72 frames. ], batch size: 23, lr: 2.33e-02, grad_scale: 8.0 +2023-02-05 21:42:54,129 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 3.440e+02 4.583e+02 5.736e+02 1.373e+03, threshold=9.167e+02, percent-clipped=8.0 +2023-02-05 21:43:03,730 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:43:26,821 INFO [train.py:901] (2/4) Epoch 3, batch 3300, loss[loss=0.3318, simple_loss=0.3877, pruned_loss=0.138, over 8022.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3839, pruned_loss=0.1461, over 1610494.57 frames. ], batch size: 22, lr: 2.32e-02, grad_scale: 8.0 +2023-02-05 21:43:27,954 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-05 21:43:43,378 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19491.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:43:49,040 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.19 vs. limit=5.0 +2023-02-05 21:44:01,025 INFO [train.py:901] (2/4) Epoch 3, batch 3350, loss[loss=0.3586, simple_loss=0.4161, pruned_loss=0.1505, over 8261.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3838, pruned_loss=0.1462, over 1611208.03 frames. ], batch size: 24, lr: 2.32e-02, grad_scale: 16.0 +2023-02-05 21:44:01,692 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 3.690e+02 4.650e+02 5.581e+02 1.223e+03, threshold=9.300e+02, percent-clipped=5.0 +2023-02-05 21:44:35,822 INFO [train.py:901] (2/4) Epoch 3, batch 3400, loss[loss=0.4033, simple_loss=0.4156, pruned_loss=0.1955, over 6832.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3845, pruned_loss=0.1463, over 1610675.78 frames. ], batch size: 71, lr: 2.32e-02, grad_scale: 16.0 +2023-02-05 21:45:02,613 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19606.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:45:09,816 INFO [train.py:901] (2/4) Epoch 3, batch 3450, loss[loss=0.2712, simple_loss=0.3261, pruned_loss=0.1081, over 8245.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3835, pruned_loss=0.146, over 1609872.12 frames. ], batch size: 22, lr: 2.32e-02, grad_scale: 16.0 +2023-02-05 21:45:10,436 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.543e+02 3.801e+02 4.733e+02 6.108e+02 1.526e+03, threshold=9.466e+02, percent-clipped=4.0 +2023-02-05 21:45:23,376 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19636.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:45:30,105 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7523, 1.5719, 2.5657, 1.3934, 2.0985, 2.7991, 2.6509, 2.5477], + device='cuda:2'), covar=tensor([0.0959, 0.1231, 0.0693, 0.1557, 0.0822, 0.0293, 0.0349, 0.0497], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0262, 0.0200, 0.0260, 0.0205, 0.0171, 0.0174, 0.0245], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 21:45:42,898 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19663.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:45:45,374 INFO [train.py:901] (2/4) Epoch 3, batch 3500, loss[loss=0.3256, simple_loss=0.3753, pruned_loss=0.138, over 8133.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3848, pruned_loss=0.1475, over 1608377.60 frames. ], batch size: 22, lr: 2.31e-02, grad_scale: 16.0 +2023-02-05 21:45:58,035 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 21:45:59,527 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19688.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:46:00,151 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0629, 1.1469, 3.1688, 0.9666, 2.7229, 2.6771, 2.7883, 2.7351], + device='cuda:2'), covar=tensor([0.0449, 0.2778, 0.0393, 0.1840, 0.1146, 0.0580, 0.0457, 0.0577], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0412, 0.0285, 0.0315, 0.0385, 0.0309, 0.0301, 0.0325], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 21:46:00,220 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1699, 1.5868, 1.3092, 1.7406, 1.3344, 1.0608, 1.2972, 1.5346], + device='cuda:2'), covar=tensor([0.0947, 0.0504, 0.1006, 0.0504, 0.0714, 0.1142, 0.0792, 0.0665], + device='cuda:2'), in_proj_covar=tensor([0.0363, 0.0246, 0.0335, 0.0311, 0.0339, 0.0317, 0.0351, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 21:46:07,364 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3383, 1.6330, 1.2678, 1.9261, 0.8890, 1.0991, 1.3655, 1.6122], + device='cuda:2'), covar=tensor([0.1388, 0.1153, 0.1663, 0.0755, 0.1760, 0.2491, 0.1505, 0.1180], + device='cuda:2'), in_proj_covar=tensor([0.0299, 0.0304, 0.0309, 0.0233, 0.0291, 0.0313, 0.0330, 0.0297], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:46:15,476 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.8922, 1.1231, 0.8433, 1.3087, 0.6747, 0.6883, 0.9340, 1.1539], + device='cuda:2'), covar=tensor([0.0995, 0.0943, 0.1313, 0.0598, 0.1309, 0.1751, 0.1055, 0.0803], + device='cuda:2'), in_proj_covar=tensor([0.0300, 0.0304, 0.0310, 0.0233, 0.0293, 0.0313, 0.0331, 0.0298], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:46:19,299 INFO [train.py:901] (2/4) Epoch 3, batch 3550, loss[loss=0.3137, simple_loss=0.3742, pruned_loss=0.1266, over 8245.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.387, pruned_loss=0.149, over 1609757.30 frames. ], batch size: 24, lr: 2.31e-02, grad_scale: 16.0 +2023-02-05 21:46:19,960 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 3.514e+02 4.193e+02 5.166e+02 1.109e+03, threshold=8.387e+02, percent-clipped=2.0 +2023-02-05 21:46:37,151 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6920, 2.2593, 4.4685, 1.0902, 2.8096, 2.1683, 1.6314, 2.3394], + device='cuda:2'), covar=tensor([0.1277, 0.1362, 0.0469, 0.2465, 0.1182, 0.1806, 0.1167, 0.1982], + device='cuda:2'), in_proj_covar=tensor([0.0431, 0.0401, 0.0476, 0.0488, 0.0540, 0.0477, 0.0423, 0.0535], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 21:46:40,592 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5656, 1.8653, 2.2015, 1.5760, 0.9741, 1.9172, 0.3250, 1.3160], + device='cuda:2'), covar=tensor([0.2569, 0.1547, 0.0857, 0.1867, 0.4911, 0.0668, 0.4823, 0.1773], + device='cuda:2'), in_proj_covar=tensor([0.0103, 0.0090, 0.0078, 0.0145, 0.0159, 0.0075, 0.0144, 0.0104], + device='cuda:2'), out_proj_covar=tensor([1.2990e-04, 1.1744e-04, 1.0077e-04, 1.7279e-04, 1.8760e-04, 9.8662e-05, + 1.7485e-04, 1.3789e-04], device='cuda:2') +2023-02-05 21:46:46,364 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19756.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:46:54,372 INFO [train.py:901] (2/4) Epoch 3, batch 3600, loss[loss=0.3619, simple_loss=0.4033, pruned_loss=0.1603, over 8531.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3849, pruned_loss=0.1472, over 1607624.84 frames. ], batch size: 49, lr: 2.31e-02, grad_scale: 16.0 +2023-02-05 21:47:28,224 INFO [train.py:901] (2/4) Epoch 3, batch 3650, loss[loss=0.316, simple_loss=0.3794, pruned_loss=0.1263, over 8034.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3848, pruned_loss=0.1468, over 1608813.39 frames. ], batch size: 22, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:47:28,895 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 3.610e+02 4.497e+02 5.952e+02 1.837e+03, threshold=8.994e+02, percent-clipped=7.0 +2023-02-05 21:47:58,279 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3544, 1.6158, 1.9192, 1.3240, 0.7912, 1.8140, 0.2444, 1.2384], + device='cuda:2'), covar=tensor([0.2707, 0.1262, 0.0855, 0.1956, 0.3951, 0.0670, 0.5212, 0.1821], + device='cuda:2'), in_proj_covar=tensor([0.0107, 0.0093, 0.0078, 0.0150, 0.0166, 0.0077, 0.0151, 0.0106], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:47:58,726 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 21:47:59,619 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19862.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:48:02,764 INFO [train.py:901] (2/4) Epoch 3, batch 3700, loss[loss=0.395, simple_loss=0.433, pruned_loss=0.1786, over 8361.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3839, pruned_loss=0.1462, over 1608072.16 frames. ], batch size: 26, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:48:05,648 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19871.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:48:14,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.5023, 3.5649, 3.0770, 1.6496, 3.0438, 2.9820, 3.2694, 2.5585], + device='cuda:2'), covar=tensor([0.0929, 0.0703, 0.1008, 0.4018, 0.0670, 0.0769, 0.1137, 0.0868], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0241, 0.0283, 0.0367, 0.0257, 0.0213, 0.0263, 0.0199], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 21:48:17,451 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19887.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:48:37,847 INFO [train.py:901] (2/4) Epoch 3, batch 3750, loss[loss=0.3237, simple_loss=0.3461, pruned_loss=0.1507, over 7643.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3831, pruned_loss=0.1457, over 1606832.03 frames. ], batch size: 19, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:48:38,369 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 3.342e+02 4.116e+02 5.480e+02 1.463e+03, threshold=8.233e+02, percent-clipped=1.0 +2023-02-05 21:48:45,548 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.57 vs. limit=5.0 +2023-02-05 21:48:52,341 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-05 21:49:12,027 INFO [train.py:901] (2/4) Epoch 3, batch 3800, loss[loss=0.3784, simple_loss=0.4258, pruned_loss=0.1655, over 8517.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3842, pruned_loss=0.1464, over 1607345.85 frames. ], batch size: 28, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:49:20,776 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19980.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:49:29,965 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2023-02-05 21:49:48,451 INFO [train.py:901] (2/4) Epoch 3, batch 3850, loss[loss=0.3475, simple_loss=0.3941, pruned_loss=0.1504, over 8130.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3852, pruned_loss=0.1464, over 1614089.66 frames. ], batch size: 22, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:49:49,086 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 3.536e+02 4.444e+02 5.257e+02 1.055e+03, threshold=8.889e+02, percent-clipped=4.0 +2023-02-05 21:49:57,544 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 21:50:01,547 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 21:50:01,930 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 21:50:02,728 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20038.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:50:19,034 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-05 21:50:22,596 INFO [train.py:901] (2/4) Epoch 3, batch 3900, loss[loss=0.2579, simple_loss=0.3199, pruned_loss=0.09796, over 7819.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3864, pruned_loss=0.147, over 1620042.64 frames. ], batch size: 20, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:50:41,536 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20095.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:50:50,429 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20107.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:50:51,159 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3450, 1.6520, 2.0659, 1.7607, 0.8249, 1.8786, 0.2949, 1.4239], + device='cuda:2'), covar=tensor([0.3024, 0.1566, 0.1114, 0.1647, 0.4366, 0.0740, 0.5983, 0.2115], + device='cuda:2'), in_proj_covar=tensor([0.0108, 0.0093, 0.0081, 0.0142, 0.0164, 0.0075, 0.0150, 0.0108], + device='cuda:2'), out_proj_covar=tensor([1.3734e-04, 1.2106e-04, 1.0481e-04, 1.7228e-04, 1.9372e-04, 9.8992e-05, + 1.8253e-04, 1.4211e-04], device='cuda:2') +2023-02-05 21:50:56,844 INFO [train.py:901] (2/4) Epoch 3, batch 3950, loss[loss=0.3549, simple_loss=0.4058, pruned_loss=0.152, over 8502.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3854, pruned_loss=0.1461, over 1619394.44 frames. ], batch size: 28, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:50:57,405 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 3.492e+02 4.461e+02 6.032e+02 1.371e+03, threshold=8.922e+02, percent-clipped=4.0 +2023-02-05 21:51:05,052 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20127.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:51:06,976 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20130.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:51:21,460 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20152.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:51:31,209 INFO [train.py:901] (2/4) Epoch 3, batch 4000, loss[loss=0.3048, simple_loss=0.3461, pruned_loss=0.1317, over 7570.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3838, pruned_loss=0.1447, over 1618387.39 frames. ], batch size: 18, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:52:05,183 INFO [train.py:901] (2/4) Epoch 3, batch 4050, loss[loss=0.2823, simple_loss=0.3432, pruned_loss=0.1107, over 7802.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3828, pruned_loss=0.1435, over 1621755.18 frames. ], batch size: 20, lr: 2.28e-02, grad_scale: 16.0 +2023-02-05 21:52:05,854 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.226e+02 3.505e+02 4.242e+02 5.307e+02 1.364e+03, threshold=8.485e+02, percent-clipped=4.0 +2023-02-05 21:52:09,350 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20222.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:52:40,358 INFO [train.py:901] (2/4) Epoch 3, batch 4100, loss[loss=0.2805, simple_loss=0.3272, pruned_loss=0.1169, over 7263.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3826, pruned_loss=0.1439, over 1617262.76 frames. ], batch size: 16, lr: 2.28e-02, grad_scale: 8.0 +2023-02-05 21:52:41,887 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0236, 1.2181, 3.1803, 0.9257, 2.5602, 2.6644, 2.7837, 2.7506], + device='cuda:2'), covar=tensor([0.0495, 0.2911, 0.0438, 0.2040, 0.1298, 0.0560, 0.0541, 0.0713], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0420, 0.0299, 0.0327, 0.0397, 0.0308, 0.0306, 0.0340], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 21:53:14,414 INFO [train.py:901] (2/4) Epoch 3, batch 4150, loss[loss=0.4247, simple_loss=0.4506, pruned_loss=0.1995, over 8562.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3818, pruned_loss=0.143, over 1617210.17 frames. ], batch size: 49, lr: 2.28e-02, grad_scale: 8.0 +2023-02-05 21:53:15,797 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 3.849e+02 4.660e+02 5.932e+02 1.097e+03, threshold=9.320e+02, percent-clipped=6.0 +2023-02-05 21:53:38,373 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20351.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:53:49,990 INFO [train.py:901] (2/4) Epoch 3, batch 4200, loss[loss=0.3059, simple_loss=0.3678, pruned_loss=0.122, over 8285.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3821, pruned_loss=0.143, over 1619214.69 frames. ], batch size: 23, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:53:55,432 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 21:53:56,307 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:00,118 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20382.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:16,269 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20406.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:16,803 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 21:54:21,047 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2023-02-05 21:54:24,063 INFO [train.py:901] (2/4) Epoch 3, batch 4250, loss[loss=0.3482, simple_loss=0.3973, pruned_loss=0.1495, over 8296.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3831, pruned_loss=0.1446, over 1614946.27 frames. ], batch size: 23, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:54:25,370 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.360e+02 3.627e+02 5.036e+02 6.332e+02 1.636e+03, threshold=1.007e+03, percent-clipped=4.0 +2023-02-05 21:54:37,009 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20436.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:47,726 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20451.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:49,942 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6510, 1.9345, 3.4879, 1.1631, 1.9959, 1.8440, 1.7194, 1.8500], + device='cuda:2'), covar=tensor([0.1255, 0.1416, 0.0454, 0.2315, 0.1248, 0.1866, 0.1033, 0.1819], + device='cuda:2'), in_proj_covar=tensor([0.0440, 0.0404, 0.0479, 0.0496, 0.0539, 0.0482, 0.0426, 0.0541], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 21:54:59,263 INFO [train.py:901] (2/4) Epoch 3, batch 4300, loss[loss=0.3592, simple_loss=0.4177, pruned_loss=0.1503, over 8788.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3834, pruned_loss=0.1444, over 1615971.31 frames. ], batch size: 30, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:55:04,853 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20474.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:55:20,176 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20497.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:55:33,531 INFO [train.py:901] (2/4) Epoch 3, batch 4350, loss[loss=0.3109, simple_loss=0.3713, pruned_loss=0.1253, over 8255.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3825, pruned_loss=0.1439, over 1611032.93 frames. ], batch size: 24, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:55:34,898 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 3.452e+02 4.356e+02 5.638e+02 1.577e+03, threshold=8.711e+02, percent-clipped=2.0 +2023-02-05 21:55:46,525 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 21:55:50,083 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 +2023-02-05 21:56:05,194 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-05 21:56:06,880 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20566.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:06,996 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20566.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:07,507 INFO [train.py:901] (2/4) Epoch 3, batch 4400, loss[loss=0.3036, simple_loss=0.3534, pruned_loss=0.1269, over 8084.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3821, pruned_loss=0.1442, over 1609299.96 frames. ], batch size: 21, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:56:24,130 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20589.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:27,495 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 21:56:36,783 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20606.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:36,819 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1403, 2.0490, 1.4766, 1.2252, 1.6840, 1.5994, 2.2657, 2.1028], + device='cuda:2'), covar=tensor([0.0799, 0.1397, 0.2406, 0.1801, 0.0926, 0.1864, 0.0956, 0.0764], + device='cuda:2'), in_proj_covar=tensor([0.0186, 0.0219, 0.0256, 0.0218, 0.0186, 0.0220, 0.0181, 0.0183], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 21:56:44,139 INFO [train.py:901] (2/4) Epoch 3, batch 4450, loss[loss=0.3012, simple_loss=0.3523, pruned_loss=0.1251, over 8092.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3793, pruned_loss=0.1418, over 1609318.72 frames. ], batch size: 21, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:56:45,438 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 3.404e+02 4.420e+02 6.069e+02 1.310e+03, threshold=8.839e+02, percent-clipped=8.0 +2023-02-05 21:57:06,677 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4110, 1.9798, 3.4903, 0.9611, 2.1373, 1.6743, 1.4803, 1.9100], + device='cuda:2'), covar=tensor([0.1375, 0.1499, 0.0603, 0.2591, 0.1438, 0.2037, 0.1209, 0.2067], + device='cuda:2'), in_proj_covar=tensor([0.0436, 0.0408, 0.0478, 0.0494, 0.0545, 0.0480, 0.0424, 0.0543], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 21:57:18,492 INFO [train.py:901] (2/4) Epoch 3, batch 4500, loss[loss=0.3138, simple_loss=0.3677, pruned_loss=0.1299, over 7974.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3791, pruned_loss=0.1414, over 1610526.01 frames. ], batch size: 21, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:57:20,854 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 21:57:27,456 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20681.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:57:46,864 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2023-02-05 21:57:53,170 INFO [train.py:901] (2/4) Epoch 3, batch 4550, loss[loss=0.3484, simple_loss=0.3912, pruned_loss=0.1528, over 8353.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3787, pruned_loss=0.1408, over 1610735.71 frames. ], batch size: 49, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:57:54,484 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.207e+02 3.483e+02 4.570e+02 6.300e+02 1.347e+03, threshold=9.139e+02, percent-clipped=2.0 +2023-02-05 21:58:14,828 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20750.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:58:17,080 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20753.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:58:26,243 INFO [train.py:901] (2/4) Epoch 3, batch 4600, loss[loss=0.3695, simple_loss=0.4175, pruned_loss=0.1608, over 8567.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3796, pruned_loss=0.1414, over 1614550.48 frames. ], batch size: 31, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 21:58:34,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20778.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:58:35,797 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20780.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:00,497 INFO [train.py:901] (2/4) Epoch 3, batch 4650, loss[loss=0.3351, simple_loss=0.3754, pruned_loss=0.1474, over 8043.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3809, pruned_loss=0.1421, over 1617293.29 frames. ], batch size: 22, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 21:59:02,525 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.001e+02 3.299e+02 4.239e+02 5.426e+02 9.400e+02, threshold=8.478e+02, percent-clipped=1.0 +2023-02-05 21:59:04,810 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20822.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:21,398 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20845.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:22,738 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20847.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:34,642 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20865.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:35,793 INFO [train.py:901] (2/4) Epoch 3, batch 4700, loss[loss=0.2789, simple_loss=0.338, pruned_loss=0.1099, over 7509.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3811, pruned_loss=0.1425, over 1619807.93 frames. ], batch size: 18, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 21:59:37,911 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20870.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:39,888 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3838, 1.7393, 2.9394, 1.0633, 1.9513, 1.7706, 1.3909, 1.7748], + device='cuda:2'), covar=tensor([0.1315, 0.1450, 0.0531, 0.2520, 0.1185, 0.1989, 0.1288, 0.1648], + device='cuda:2'), in_proj_covar=tensor([0.0439, 0.0413, 0.0475, 0.0501, 0.0549, 0.0481, 0.0427, 0.0547], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 21:59:54,902 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20895.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:08,494 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1074, 3.8930, 2.0988, 1.9853, 2.8398, 1.9533, 2.5019, 2.8016], + device='cuda:2'), covar=tensor([0.1241, 0.0331, 0.0730, 0.0913, 0.0570, 0.0944, 0.1017, 0.0887], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0244, 0.0328, 0.0314, 0.0339, 0.0324, 0.0350, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 22:00:08,978 INFO [train.py:901] (2/4) Epoch 3, batch 4750, loss[loss=0.2954, simple_loss=0.3692, pruned_loss=0.1108, over 8435.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3821, pruned_loss=0.1433, over 1618413.50 frames. ], batch size: 29, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 22:00:10,297 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.464e+02 3.634e+02 4.432e+02 5.821e+02 1.296e+03, threshold=8.863e+02, percent-clipped=5.0 +2023-02-05 22:00:23,262 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20937.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:24,393 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 22:00:26,467 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 22:00:32,658 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20950.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:41,459 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20962.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:44,560 INFO [train.py:901] (2/4) Epoch 3, batch 4800, loss[loss=0.3313, simple_loss=0.373, pruned_loss=0.1448, over 8089.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3816, pruned_loss=0.1426, over 1616135.75 frames. ], batch size: 21, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:01:15,904 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.34 vs. limit=5.0 +2023-02-05 22:01:18,149 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 22:01:18,803 INFO [train.py:901] (2/4) Epoch 3, batch 4850, loss[loss=0.461, simple_loss=0.4673, pruned_loss=0.2274, over 6752.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3814, pruned_loss=0.1427, over 1614291.05 frames. ], batch size: 72, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:01:20,192 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.459e+02 3.687e+02 4.412e+02 5.668e+02 1.155e+03, threshold=8.825e+02, percent-clipped=6.0 +2023-02-05 22:01:21,142 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3936, 1.5986, 1.6232, 0.7439, 1.5858, 1.2492, 0.2971, 1.6231], + device='cuda:2'), covar=tensor([0.0102, 0.0079, 0.0086, 0.0128, 0.0090, 0.0258, 0.0213, 0.0050], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0165, 0.0138, 0.0214, 0.0159, 0.0282, 0.0224, 0.0194], + device='cuda:2'), out_proj_covar=tensor([1.1142e-04, 7.8030e-05, 6.3686e-05, 9.7799e-05, 7.6079e-05, 1.4329e-04, + 1.0844e-04, 9.0591e-05], device='cuda:2') +2023-02-05 22:01:42,126 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4653, 2.1205, 2.3074, 1.6664, 1.1286, 2.1866, 0.3238, 1.5115], + device='cuda:2'), covar=tensor([0.2738, 0.1176, 0.0610, 0.1947, 0.3779, 0.0663, 0.4856, 0.1669], + device='cuda:2'), in_proj_covar=tensor([0.0107, 0.0101, 0.0080, 0.0147, 0.0164, 0.0081, 0.0146, 0.0109], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:01:51,979 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21065.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:01:53,048 INFO [train.py:901] (2/4) Epoch 3, batch 4900, loss[loss=0.4778, simple_loss=0.4809, pruned_loss=0.2373, over 8484.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3826, pruned_loss=0.1438, over 1615127.43 frames. ], batch size: 28, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:01:55,915 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21070.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:27,712 INFO [train.py:901] (2/4) Epoch 3, batch 4950, loss[loss=0.3551, simple_loss=0.4017, pruned_loss=0.1542, over 8539.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3816, pruned_loss=0.1429, over 1612729.58 frames. ], batch size: 28, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:02:29,094 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 3.569e+02 4.502e+02 6.229e+02 1.133e+03, threshold=9.004e+02, percent-clipped=2.0 +2023-02-05 22:02:30,685 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21121.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:41,327 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21136.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:48,064 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21146.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:51,482 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21151.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:03:01,832 INFO [train.py:901] (2/4) Epoch 3, batch 5000, loss[loss=0.2955, simple_loss=0.3465, pruned_loss=0.1222, over 7641.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3782, pruned_loss=0.1411, over 1608975.69 frames. ], batch size: 19, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:03:03,404 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-05 22:03:08,657 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21176.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:03:37,220 INFO [train.py:901] (2/4) Epoch 3, batch 5050, loss[loss=0.3646, simple_loss=0.4142, pruned_loss=0.1574, over 8437.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3791, pruned_loss=0.1423, over 1607764.11 frames. ], batch size: 27, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:03:38,541 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 3.325e+02 4.224e+02 5.254e+02 1.187e+03, threshold=8.447e+02, percent-clipped=3.0 +2023-02-05 22:03:57,071 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 22:04:11,245 INFO [train.py:901] (2/4) Epoch 3, batch 5100, loss[loss=0.3111, simple_loss=0.3633, pruned_loss=0.1295, over 8520.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3806, pruned_loss=0.1431, over 1608505.82 frames. ], batch size: 28, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:04:16,072 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7332, 2.2070, 1.6596, 2.6919, 1.2699, 1.3770, 1.6646, 2.0983], + device='cuda:2'), covar=tensor([0.1297, 0.1293, 0.2044, 0.0503, 0.2124, 0.2627, 0.1972, 0.1415], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0308, 0.0303, 0.0229, 0.0293, 0.0316, 0.0329, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 22:04:46,361 INFO [train.py:901] (2/4) Epoch 3, batch 5150, loss[loss=0.4027, simple_loss=0.4357, pruned_loss=0.1848, over 8369.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3787, pruned_loss=0.1416, over 1606350.73 frames. ], batch size: 24, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:04:47,675 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.142e+02 3.453e+02 4.061e+02 5.332e+02 1.278e+03, threshold=8.122e+02, percent-clipped=4.0 +2023-02-05 22:04:50,023 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21321.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:04:56,024 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21330.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:06,474 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:06,535 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:19,994 INFO [train.py:901] (2/4) Epoch 3, batch 5200, loss[loss=0.3906, simple_loss=0.4047, pruned_loss=0.1882, over 7929.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3792, pruned_loss=0.1416, over 1612731.99 frames. ], batch size: 20, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:05:21,044 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-05 22:05:52,848 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21414.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:54,862 INFO [train.py:901] (2/4) Epoch 3, batch 5250, loss[loss=0.3636, simple_loss=0.4013, pruned_loss=0.163, over 7795.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3804, pruned_loss=0.1425, over 1615463.36 frames. ], batch size: 20, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:05:54,877 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 22:05:56,261 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 3.353e+02 4.281e+02 5.765e+02 2.364e+03, threshold=8.563e+02, percent-clipped=11.0 +2023-02-05 22:06:30,401 INFO [train.py:901] (2/4) Epoch 3, batch 5300, loss[loss=0.3289, simple_loss=0.3889, pruned_loss=0.1344, over 8188.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3801, pruned_loss=0.1422, over 1613636.26 frames. ], batch size: 23, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:06:39,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21480.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:07:04,804 INFO [train.py:901] (2/4) Epoch 3, batch 5350, loss[loss=0.3187, simple_loss=0.3606, pruned_loss=0.1384, over 7970.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3798, pruned_loss=0.1417, over 1609394.43 frames. ], batch size: 21, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:07:06,084 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.137e+02 3.338e+02 4.128e+02 5.460e+02 1.129e+03, threshold=8.255e+02, percent-clipped=3.0 +2023-02-05 22:07:13,626 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21529.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:07:38,553 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-05 22:07:40,164 INFO [train.py:901] (2/4) Epoch 3, batch 5400, loss[loss=0.3791, simple_loss=0.4218, pruned_loss=0.1682, over 8462.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3791, pruned_loss=0.1415, over 1604972.00 frames. ], batch size: 27, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:07:59,341 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:12,095 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:14,745 INFO [train.py:901] (2/4) Epoch 3, batch 5450, loss[loss=0.3465, simple_loss=0.3946, pruned_loss=0.1492, over 8509.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3777, pruned_loss=0.1404, over 1605349.87 frames. ], batch size: 26, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:08:16,069 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.864e+02 3.746e+02 4.366e+02 5.874e+02 2.172e+03, threshold=8.732e+02, percent-clipped=6.0 +2023-02-05 22:08:24,279 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21631.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:27,054 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9280, 1.3509, 3.4463, 1.4628, 2.2749, 3.8013, 3.4113, 3.3295], + device='cuda:2'), covar=tensor([0.1024, 0.1541, 0.0326, 0.1811, 0.0683, 0.0236, 0.0374, 0.0493], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0262, 0.0200, 0.0255, 0.0207, 0.0181, 0.0180, 0.0252], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:08:41,814 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 22:08:49,812 INFO [train.py:901] (2/4) Epoch 3, batch 5500, loss[loss=0.3317, simple_loss=0.3829, pruned_loss=0.1402, over 8192.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3796, pruned_loss=0.1414, over 1605301.79 frames. ], batch size: 23, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:08:55,223 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:56,633 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.4468, 1.8858, 5.4293, 2.2466, 4.7463, 4.6477, 5.0146, 4.9252], + device='cuda:2'), covar=tensor([0.0313, 0.2871, 0.0232, 0.1488, 0.0855, 0.0372, 0.0282, 0.0338], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0427, 0.0308, 0.0339, 0.0399, 0.0322, 0.0309, 0.0349], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 22:09:05,908 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21690.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:09:09,887 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0577, 3.1287, 2.7353, 1.7332, 2.7083, 2.6298, 2.9022, 2.2736], + device='cuda:2'), covar=tensor([0.1008, 0.0729, 0.1009, 0.3466, 0.0757, 0.0888, 0.1369, 0.0858], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0240, 0.0288, 0.0370, 0.0266, 0.0215, 0.0265, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:09:23,546 INFO [train.py:901] (2/4) Epoch 3, batch 5550, loss[loss=0.3657, simple_loss=0.4112, pruned_loss=0.1601, over 8604.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.381, pruned_loss=0.1418, over 1611475.70 frames. ], batch size: 31, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:09:23,780 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5998, 1.8739, 2.2486, 1.7490, 1.0346, 2.1141, 0.3488, 1.2970], + device='cuda:2'), covar=tensor([0.3193, 0.2389, 0.1248, 0.2110, 0.5373, 0.0916, 0.6385, 0.2223], + device='cuda:2'), in_proj_covar=tensor([0.0117, 0.0107, 0.0083, 0.0150, 0.0177, 0.0084, 0.0150, 0.0115], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:09:24,912 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 3.296e+02 4.063e+02 5.206e+02 8.291e+02, threshold=8.125e+02, percent-clipped=0.0 +2023-02-05 22:09:58,432 INFO [train.py:901] (2/4) Epoch 3, batch 5600, loss[loss=0.3642, simple_loss=0.4061, pruned_loss=0.1611, over 8716.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3815, pruned_loss=0.141, over 1615609.79 frames. ], batch size: 49, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:10:08,495 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21781.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:11,828 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21785.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:14,461 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21789.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:25,389 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21805.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:28,851 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21810.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:33,430 INFO [train.py:901] (2/4) Epoch 3, batch 5650, loss[loss=0.3039, simple_loss=0.3598, pruned_loss=0.124, over 8111.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3815, pruned_loss=0.1415, over 1617473.77 frames. ], batch size: 23, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:10:34,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 3.614e+02 4.526e+02 5.980e+02 8.654e+02, threshold=9.051e+02, percent-clipped=4.0 +2023-02-05 22:10:45,265 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 22:10:56,879 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21851.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:11:02,289 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-05 22:11:07,115 INFO [train.py:901] (2/4) Epoch 3, batch 5700, loss[loss=0.3493, simple_loss=0.3789, pruned_loss=0.1599, over 8075.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3821, pruned_loss=0.1428, over 1615940.83 frames. ], batch size: 21, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:11:07,935 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21868.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:11:13,370 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21876.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:11:39,355 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-05 22:11:42,902 INFO [train.py:901] (2/4) Epoch 3, batch 5750, loss[loss=0.2374, simple_loss=0.3028, pruned_loss=0.086, over 7787.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3803, pruned_loss=0.142, over 1609793.14 frames. ], batch size: 19, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:11:44,221 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.339e+02 3.657e+02 4.422e+02 5.345e+02 1.248e+03, threshold=8.845e+02, percent-clipped=3.0 +2023-02-05 22:11:49,705 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 22:12:10,247 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21957.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:12:11,339 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-05 22:12:16,965 INFO [train.py:901] (2/4) Epoch 3, batch 5800, loss[loss=0.3548, simple_loss=0.3918, pruned_loss=0.1589, over 8655.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3802, pruned_loss=0.1419, over 1609468.72 frames. ], batch size: 39, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:12:19,906 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1505, 1.3601, 2.3107, 1.0752, 2.2328, 2.4601, 2.2803, 2.0714], + device='cuda:2'), covar=tensor([0.1115, 0.1035, 0.0396, 0.1652, 0.0441, 0.0338, 0.0449, 0.0689], + device='cuda:2'), in_proj_covar=tensor([0.0228, 0.0254, 0.0190, 0.0247, 0.0199, 0.0174, 0.0170, 0.0246], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:12:22,435 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21975.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:12:25,859 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0111, 1.8529, 2.8476, 2.3532, 2.3038, 1.8671, 1.3838, 1.0088], + device='cuda:2'), covar=tensor([0.1092, 0.1001, 0.0239, 0.0456, 0.0450, 0.0534, 0.0687, 0.1040], + device='cuda:2'), in_proj_covar=tensor([0.0622, 0.0534, 0.0453, 0.0489, 0.0603, 0.0499, 0.0509, 0.0517], + 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') +2023-02-05 22:12:30,996 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21988.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:12:52,162 INFO [train.py:901] (2/4) Epoch 3, batch 5850, loss[loss=0.2978, simple_loss=0.3628, pruned_loss=0.1164, over 8501.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3798, pruned_loss=0.1412, over 1613734.28 frames. ], batch size: 28, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:12:53,405 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 3.662e+02 4.461e+02 5.594e+02 1.608e+03, threshold=8.923e+02, percent-clipped=8.0 +2023-02-05 22:13:11,817 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22045.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:22,187 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:25,941 INFO [train.py:901] (2/4) Epoch 3, batch 5900, loss[loss=0.4242, simple_loss=0.4411, pruned_loss=0.2036, over 7417.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3786, pruned_loss=0.1406, over 1608007.15 frames. ], batch size: 72, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:13:28,808 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22070.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:30,183 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22072.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:39,565 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22086.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:42,233 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22090.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:57,695 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3333, 1.5881, 1.4892, 1.2226, 1.5759, 1.4306, 1.6816, 1.6560], + device='cuda:2'), covar=tensor([0.0646, 0.1345, 0.1979, 0.1616, 0.0764, 0.1741, 0.0870, 0.0677], + device='cuda:2'), in_proj_covar=tensor([0.0181, 0.0216, 0.0256, 0.0216, 0.0184, 0.0218, 0.0179, 0.0180], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:14:00,338 INFO [train.py:901] (2/4) Epoch 3, batch 5950, loss[loss=0.3197, simple_loss=0.3712, pruned_loss=0.1341, over 8500.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3777, pruned_loss=0.14, over 1607219.34 frames. ], batch size: 26, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:14:02,405 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.045e+02 3.353e+02 4.485e+02 5.691e+02 1.558e+03, threshold=8.970e+02, percent-clipped=6.0 +2023-02-05 22:14:07,180 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22125.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:14:23,954 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22148.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:14:36,950 INFO [train.py:901] (2/4) Epoch 3, batch 6000, loss[loss=0.3636, simple_loss=0.409, pruned_loss=0.159, over 8505.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3776, pruned_loss=0.1395, over 1611060.81 frames. ], batch size: 39, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:14:36,951 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 22:14:49,936 INFO [train.py:935] (2/4) Epoch 3, validation: loss=0.2472, simple_loss=0.3383, pruned_loss=0.07805, over 944034.00 frames. +2023-02-05 22:14:49,937 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 22:15:08,336 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22194.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:12,103 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-05 22:15:21,642 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22212.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:25,116 INFO [train.py:901] (2/4) Epoch 3, batch 6050, loss[loss=0.2928, simple_loss=0.3504, pruned_loss=0.1176, over 7241.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3786, pruned_loss=0.1407, over 1616588.32 frames. ], batch size: 16, lr: 2.18e-02, grad_scale: 8.0 +2023-02-05 22:15:26,473 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 3.417e+02 4.364e+02 5.364e+02 3.571e+03, threshold=8.727e+02, percent-clipped=6.0 +2023-02-05 22:15:33,728 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2023-02-05 22:15:36,171 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22233.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:40,935 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22240.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:51,050 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22255.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:59,773 INFO [train.py:901] (2/4) Epoch 3, batch 6100, loss[loss=0.3227, simple_loss=0.3831, pruned_loss=0.1311, over 8513.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3788, pruned_loss=0.1403, over 1616655.71 frames. ], batch size: 26, lr: 2.18e-02, grad_scale: 16.0 +2023-02-05 22:16:18,445 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 22:16:35,144 INFO [train.py:901] (2/4) Epoch 3, batch 6150, loss[loss=0.3623, simple_loss=0.3982, pruned_loss=0.1632, over 8492.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3788, pruned_loss=0.1409, over 1616512.68 frames. ], batch size: 26, lr: 2.18e-02, grad_scale: 16.0 +2023-02-05 22:16:36,466 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.469e+02 3.615e+02 4.380e+02 5.688e+02 1.525e+03, threshold=8.759e+02, percent-clipped=2.0 +2023-02-05 22:16:41,827 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22327.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:42,487 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22328.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:45,079 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22332.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:54,538 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:59,259 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22353.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:17:04,952 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-05 22:17:08,710 INFO [train.py:901] (2/4) Epoch 3, batch 6200, loss[loss=0.3329, simple_loss=0.3797, pruned_loss=0.143, over 8070.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3782, pruned_loss=0.1402, over 1612090.02 frames. ], batch size: 21, lr: 2.18e-02, grad_scale: 16.0 +2023-02-05 22:17:09,125 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-05 22:17:11,706 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22371.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:17:19,684 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9125, 0.9831, 1.0714, 0.8872, 0.6535, 1.1082, 0.0551, 0.7855], + device='cuda:2'), covar=tensor([0.1999, 0.1657, 0.1063, 0.1539, 0.3641, 0.0633, 0.4170, 0.1910], + device='cuda:2'), in_proj_covar=tensor([0.0115, 0.0108, 0.0079, 0.0150, 0.0169, 0.0075, 0.0144, 0.0115], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:17:27,294 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-05 22:17:34,529 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22403.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:17:40,520 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1411, 1.2964, 1.2314, 0.1980, 1.1813, 0.9631, 0.0788, 1.1711], + device='cuda:2'), covar=tensor([0.0088, 0.0068, 0.0069, 0.0152, 0.0079, 0.0237, 0.0192, 0.0077], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0159, 0.0139, 0.0213, 0.0157, 0.0283, 0.0226, 0.0188], + device='cuda:2'), out_proj_covar=tensor([1.0651e-04, 7.1615e-05, 6.2392e-05, 9.5690e-05, 7.3334e-05, 1.3844e-04, + 1.0564e-04, 8.4073e-05], device='cuda:2') +2023-02-05 22:17:44,417 INFO [train.py:901] (2/4) Epoch 3, batch 6250, loss[loss=0.2899, simple_loss=0.3395, pruned_loss=0.1202, over 7281.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3787, pruned_loss=0.1409, over 1610927.22 frames. ], batch size: 16, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:17:45,753 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.208e+02 3.506e+02 4.308e+02 5.585e+02 1.214e+03, threshold=8.617e+02, percent-clipped=6.0 +2023-02-05 22:18:05,793 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22447.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:18:19,126 INFO [train.py:901] (2/4) Epoch 3, batch 6300, loss[loss=0.3837, simple_loss=0.4251, pruned_loss=0.1711, over 8259.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3788, pruned_loss=0.1408, over 1612973.57 frames. ], batch size: 24, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:18:20,234 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.39 vs. limit=5.0 +2023-02-05 22:18:25,230 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1957, 1.7925, 2.9274, 2.3707, 2.3231, 1.7709, 1.3844, 1.0706], + device='cuda:2'), covar=tensor([0.1045, 0.1072, 0.0243, 0.0484, 0.0470, 0.0594, 0.0755, 0.1124], + device='cuda:2'), in_proj_covar=tensor([0.0618, 0.0528, 0.0453, 0.0488, 0.0603, 0.0495, 0.0511, 0.0518], + 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') +2023-02-05 22:18:36,439 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22492.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:18:39,274 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7906, 2.0545, 1.5915, 1.4430, 1.8945, 1.8073, 2.3120, 2.3473], + device='cuda:2'), covar=tensor([0.0544, 0.1362, 0.1987, 0.1692, 0.0748, 0.1646, 0.0859, 0.0600], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0215, 0.0252, 0.0214, 0.0179, 0.0217, 0.0178, 0.0179], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:18:39,316 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22496.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:18:54,107 INFO [train.py:901] (2/4) Epoch 3, batch 6350, loss[loss=0.274, simple_loss=0.3286, pruned_loss=0.1097, over 7813.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3776, pruned_loss=0.14, over 1612513.71 frames. ], batch size: 20, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:18:55,438 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 3.537e+02 4.368e+02 5.315e+02 1.494e+03, threshold=8.736e+02, percent-clipped=5.0 +2023-02-05 22:18:57,019 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22521.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:08,840 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22538.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:28,747 INFO [train.py:901] (2/4) Epoch 3, batch 6400, loss[loss=0.3433, simple_loss=0.4063, pruned_loss=0.1401, over 8024.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3763, pruned_loss=0.1385, over 1612947.79 frames. ], batch size: 22, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:19:35,384 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22577.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:39,442 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22583.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:40,140 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7924, 2.6095, 4.6987, 1.1717, 2.6343, 2.4809, 1.8263, 2.3728], + device='cuda:2'), covar=tensor([0.1238, 0.1434, 0.0501, 0.2584, 0.1307, 0.1721, 0.1160, 0.2053], + device='cuda:2'), in_proj_covar=tensor([0.0454, 0.0419, 0.0501, 0.0512, 0.0556, 0.0492, 0.0445, 0.0562], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:19:50,049 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22599.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:55,833 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22607.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:56,537 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22608.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:20:03,200 INFO [train.py:901] (2/4) Epoch 3, batch 6450, loss[loss=0.3801, simple_loss=0.4263, pruned_loss=0.167, over 8039.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3768, pruned_loss=0.1382, over 1618696.52 frames. ], batch size: 22, lr: 2.16e-02, grad_scale: 16.0 +2023-02-05 22:20:04,480 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 3.557e+02 4.436e+02 5.729e+02 1.082e+03, threshold=8.871e+02, percent-clipped=7.0 +2023-02-05 22:20:21,543 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 +2023-02-05 22:20:28,490 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22653.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:20:31,150 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22657.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:20:37,604 INFO [train.py:901] (2/4) Epoch 3, batch 6500, loss[loss=0.2758, simple_loss=0.3191, pruned_loss=0.1162, over 5524.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3779, pruned_loss=0.1388, over 1619915.37 frames. ], batch size: 12, lr: 2.16e-02, grad_scale: 16.0 +2023-02-05 22:20:37,825 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7094, 3.1728, 3.2262, 1.9698, 1.6360, 3.0202, 0.7633, 2.5907], + device='cuda:2'), covar=tensor([0.2591, 0.1891, 0.0754, 0.2732, 0.4903, 0.0793, 0.5960, 0.1371], + device='cuda:2'), in_proj_covar=tensor([0.0116, 0.0110, 0.0082, 0.0153, 0.0175, 0.0077, 0.0145, 0.0111], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:20:55,236 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22692.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:01,238 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4388, 2.1345, 2.0963, 0.8258, 2.0157, 1.4550, 0.5280, 1.6224], + device='cuda:2'), covar=tensor([0.0158, 0.0084, 0.0099, 0.0194, 0.0116, 0.0292, 0.0234, 0.0089], + device='cuda:2'), in_proj_covar=tensor([0.0246, 0.0168, 0.0143, 0.0221, 0.0164, 0.0297, 0.0234, 0.0197], + device='cuda:2'), out_proj_covar=tensor([1.1229e-04, 7.5322e-05, 6.4148e-05, 9.8566e-05, 7.6339e-05, 1.4518e-04, + 1.0787e-04, 8.6885e-05], device='cuda:2') +2023-02-05 22:21:02,666 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22703.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:03,420 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.06 vs. limit=5.0 +2023-02-05 22:21:09,917 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22714.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:11,757 INFO [train.py:901] (2/4) Epoch 3, batch 6550, loss[loss=0.3895, simple_loss=0.4286, pruned_loss=0.1752, over 8496.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3777, pruned_loss=0.1389, over 1616235.53 frames. ], batch size: 26, lr: 2.16e-02, grad_scale: 16.0 +2023-02-05 22:21:13,164 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.155e+02 3.258e+02 3.883e+02 5.357e+02 1.264e+03, threshold=7.766e+02, percent-clipped=3.0 +2023-02-05 22:21:19,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22728.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:28,616 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 22:21:32,835 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22747.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:47,206 INFO [train.py:901] (2/4) Epoch 3, batch 6600, loss[loss=0.3811, simple_loss=0.4335, pruned_loss=0.1644, over 8475.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.377, pruned_loss=0.1382, over 1618275.15 frames. ], batch size: 25, lr: 2.16e-02, grad_scale: 8.0 +2023-02-05 22:21:47,903 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 22:22:19,031 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22812.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:22,347 INFO [train.py:901] (2/4) Epoch 3, batch 6650, loss[loss=0.2858, simple_loss=0.3384, pruned_loss=0.1166, over 7924.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3774, pruned_loss=0.138, over 1619311.07 frames. ], batch size: 20, lr: 2.16e-02, grad_scale: 8.0 +2023-02-05 22:22:24,341 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.373e+02 3.456e+02 4.169e+02 5.335e+02 9.931e+02, threshold=8.339e+02, percent-clipped=8.0 +2023-02-05 22:22:34,130 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4527, 2.4714, 1.7365, 2.2196, 1.9535, 1.2751, 1.8645, 1.9867], + device='cuda:2'), covar=tensor([0.1049, 0.0401, 0.0896, 0.0571, 0.0674, 0.1206, 0.0906, 0.0630], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0230, 0.0323, 0.0313, 0.0327, 0.0313, 0.0337, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 22:22:40,066 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22843.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:41,434 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3942, 1.2686, 3.0124, 1.0501, 2.1952, 3.2957, 3.1550, 2.8219], + device='cuda:2'), covar=tensor([0.1280, 0.1492, 0.0371, 0.2083, 0.0656, 0.0238, 0.0383, 0.0605], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0262, 0.0199, 0.0260, 0.0206, 0.0178, 0.0177, 0.0249], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:22:53,739 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22862.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:54,496 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22863.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:57,072 INFO [train.py:901] (2/4) Epoch 3, batch 6700, loss[loss=0.2968, simple_loss=0.3652, pruned_loss=0.1142, over 8037.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3751, pruned_loss=0.1368, over 1615906.40 frames. ], batch size: 22, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:23:04,028 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5608, 2.0624, 3.5429, 1.0630, 2.2440, 1.8028, 1.5223, 1.9593], + device='cuda:2'), covar=tensor([0.1203, 0.1524, 0.0526, 0.2573, 0.1288, 0.1974, 0.1248, 0.1964], + device='cuda:2'), in_proj_covar=tensor([0.0445, 0.0413, 0.0490, 0.0507, 0.0546, 0.0478, 0.0432, 0.0552], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:23:12,645 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22888.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:23:17,363 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7513, 2.1260, 1.6461, 2.6205, 1.0687, 1.4376, 1.8178, 2.2723], + device='cuda:2'), covar=tensor([0.1175, 0.1308, 0.1817, 0.0596, 0.1986, 0.2288, 0.1529, 0.1043], + device='cuda:2'), in_proj_covar=tensor([0.0299, 0.0304, 0.0300, 0.0227, 0.0272, 0.0310, 0.0313, 0.0293], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:2') +2023-02-05 22:23:26,901 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22909.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:23:32,721 INFO [train.py:901] (2/4) Epoch 3, batch 6750, loss[loss=0.3721, simple_loss=0.4144, pruned_loss=0.165, over 8533.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3758, pruned_loss=0.1374, over 1618634.50 frames. ], batch size: 49, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:23:34,753 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 3.597e+02 4.402e+02 5.483e+02 1.400e+03, threshold=8.804e+02, percent-clipped=7.0 +2023-02-05 22:23:44,514 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22934.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:23:53,624 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22948.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:05,896 INFO [train.py:901] (2/4) Epoch 3, batch 6800, loss[loss=0.3252, simple_loss=0.3737, pruned_loss=0.1383, over 7245.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3762, pruned_loss=0.1381, over 1613547.49 frames. ], batch size: 16, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:24:05,905 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 22:24:08,774 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22970.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:10,828 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22973.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:26,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22995.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:30,242 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23001.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:35,032 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4357, 2.1928, 1.4637, 1.8057, 1.9365, 1.1709, 1.5077, 1.9137], + device='cuda:2'), covar=tensor([0.0963, 0.0341, 0.0883, 0.0573, 0.0525, 0.1045, 0.0837, 0.0598], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0235, 0.0329, 0.0314, 0.0331, 0.0312, 0.0340, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 22:24:41,423 INFO [train.py:901] (2/4) Epoch 3, batch 6850, loss[loss=0.3383, simple_loss=0.4036, pruned_loss=0.1365, over 8457.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.376, pruned_loss=0.1376, over 1615448.66 frames. ], batch size: 25, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:24:43,431 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 3.425e+02 4.505e+02 5.413e+02 1.323e+03, threshold=9.011e+02, percent-clipped=6.0 +2023-02-05 22:24:54,848 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 22:25:15,252 INFO [train.py:901] (2/4) Epoch 3, batch 6900, loss[loss=0.32, simple_loss=0.3811, pruned_loss=0.1295, over 8199.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.378, pruned_loss=0.1394, over 1615182.86 frames. ], batch size: 23, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:25:50,201 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23116.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:25:50,684 INFO [train.py:901] (2/4) Epoch 3, batch 6950, loss[loss=0.2701, simple_loss=0.333, pruned_loss=0.1035, over 7920.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3782, pruned_loss=0.1395, over 1613917.13 frames. ], batch size: 20, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:25:51,596 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23118.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:25:52,722 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 3.525e+02 4.440e+02 6.025e+02 1.140e+03, threshold=8.880e+02, percent-clipped=3.0 +2023-02-05 22:25:57,666 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23126.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:02,144 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-05 22:26:09,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23143.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:14,266 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 22:26:18,629 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23156.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:26,188 INFO [train.py:901] (2/4) Epoch 3, batch 7000, loss[loss=0.2882, simple_loss=0.3544, pruned_loss=0.111, over 7964.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3765, pruned_loss=0.1381, over 1613900.79 frames. ], batch size: 21, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:26:33,858 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5121, 1.7964, 2.0553, 1.7497, 1.0059, 1.8909, 0.4232, 1.3197], + device='cuda:2'), covar=tensor([0.3294, 0.1750, 0.1005, 0.1363, 0.5172, 0.0810, 0.5118, 0.1826], + device='cuda:2'), in_proj_covar=tensor([0.0120, 0.0116, 0.0088, 0.0158, 0.0185, 0.0081, 0.0151, 0.0116], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:26:39,926 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23187.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:01,199 INFO [train.py:901] (2/4) Epoch 3, batch 7050, loss[loss=0.3497, simple_loss=0.391, pruned_loss=0.1542, over 8257.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3759, pruned_loss=0.1382, over 1610028.17 frames. ], batch size: 48, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:27:03,862 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 3.682e+02 4.488e+02 5.424e+02 1.788e+03, threshold=8.977e+02, percent-clipped=6.0 +2023-02-05 22:27:14,102 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23235.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:18,794 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3336, 1.4440, 4.4868, 1.8831, 3.7737, 3.6977, 3.9600, 3.9024], + device='cuda:2'), covar=tensor([0.0388, 0.3175, 0.0291, 0.1818, 0.1021, 0.0506, 0.0381, 0.0465], + device='cuda:2'), in_proj_covar=tensor([0.0263, 0.0429, 0.0321, 0.0340, 0.0405, 0.0334, 0.0316, 0.0356], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 22:27:22,741 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23247.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:36,329 INFO [train.py:901] (2/4) Epoch 3, batch 7100, loss[loss=0.3406, simple_loss=0.3901, pruned_loss=0.1455, over 8298.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3754, pruned_loss=0.138, over 1609888.31 frames. ], batch size: 23, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:27:39,094 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23271.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:59,149 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23302.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:28:08,849 INFO [train.py:901] (2/4) Epoch 3, batch 7150, loss[loss=0.3742, simple_loss=0.4144, pruned_loss=0.167, over 7971.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3757, pruned_loss=0.1385, over 1607928.34 frames. ], batch size: 21, lr: 2.13e-02, grad_scale: 8.0 +2023-02-05 22:28:10,871 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 3.845e+02 4.572e+02 5.960e+02 1.048e+03, threshold=9.143e+02, percent-clipped=2.0 +2023-02-05 22:28:43,301 INFO [train.py:901] (2/4) Epoch 3, batch 7200, loss[loss=0.3152, simple_loss=0.3556, pruned_loss=0.1373, over 7420.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3741, pruned_loss=0.1371, over 1608585.78 frames. ], batch size: 17, lr: 2.13e-02, grad_scale: 8.0 +2023-02-05 22:28:47,566 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23372.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:29:04,753 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23397.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:29:17,775 INFO [train.py:901] (2/4) Epoch 3, batch 7250, loss[loss=0.2975, simple_loss=0.3373, pruned_loss=0.1288, over 7250.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3743, pruned_loss=0.1376, over 1606714.98 frames. ], batch size: 16, lr: 2.13e-02, grad_scale: 4.0 +2023-02-05 22:29:20,311 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.518e+02 3.505e+02 4.323e+02 5.847e+02 9.851e+02, threshold=8.646e+02, percent-clipped=2.0 +2023-02-05 22:29:52,887 INFO [train.py:901] (2/4) Epoch 3, batch 7300, loss[loss=0.3481, simple_loss=0.3941, pruned_loss=0.1511, over 8252.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3749, pruned_loss=0.1378, over 1610457.10 frames. ], batch size: 49, lr: 2.13e-02, grad_scale: 4.0 +2023-02-05 22:29:55,076 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23470.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:02,050 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7357, 2.7862, 1.6830, 2.1452, 2.3297, 1.4994, 2.0740, 2.2517], + device='cuda:2'), covar=tensor([0.1281, 0.0432, 0.0967, 0.0676, 0.0579, 0.1139, 0.0829, 0.0735], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0236, 0.0324, 0.0318, 0.0339, 0.0319, 0.0341, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 22:30:08,364 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4040, 1.3897, 4.5357, 1.7156, 3.7150, 3.6726, 3.9828, 3.9384], + device='cuda:2'), covar=tensor([0.0388, 0.3341, 0.0254, 0.1973, 0.1158, 0.0493, 0.0389, 0.0565], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0439, 0.0321, 0.0349, 0.0417, 0.0346, 0.0320, 0.0368], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 22:30:14,541 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1491, 3.1375, 2.7675, 1.4301, 2.7535, 2.6483, 2.8770, 2.5307], + device='cuda:2'), covar=tensor([0.1273, 0.0800, 0.1161, 0.4543, 0.0904, 0.1212, 0.1516, 0.0994], + device='cuda:2'), in_proj_covar=tensor([0.0363, 0.0249, 0.0276, 0.0365, 0.0269, 0.0223, 0.0267, 0.0208], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:30:20,013 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4745, 1.6478, 1.4607, 1.2217, 1.4995, 1.3590, 1.8361, 1.7031], + device='cuda:2'), covar=tensor([0.0638, 0.1208, 0.1985, 0.1638, 0.0714, 0.1639, 0.0835, 0.0616], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0208, 0.0243, 0.0210, 0.0171, 0.0211, 0.0174, 0.0174], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:30:21,343 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23506.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:28,673 INFO [train.py:901] (2/4) Epoch 3, batch 7350, loss[loss=0.3204, simple_loss=0.3717, pruned_loss=0.1345, over 8237.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3748, pruned_loss=0.1376, over 1611603.18 frames. ], batch size: 22, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:30:31,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.204e+02 3.295e+02 4.174e+02 5.897e+02 1.266e+03, threshold=8.348e+02, percent-clipped=6.0 +2023-02-05 22:30:35,723 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23527.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:45,675 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-05 22:30:52,325 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23552.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:56,354 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23558.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:03,073 INFO [train.py:901] (2/4) Epoch 3, batch 7400, loss[loss=0.3172, simple_loss=0.3624, pruned_loss=0.136, over 5952.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3764, pruned_loss=0.1389, over 1608166.45 frames. ], batch size: 13, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:31:03,948 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9639, 1.1986, 1.7757, 0.8506, 1.3700, 1.1639, 1.0446, 1.1818], + device='cuda:2'), covar=tensor([0.0942, 0.1120, 0.0399, 0.1802, 0.0765, 0.1393, 0.0943, 0.1167], + device='cuda:2'), in_proj_covar=tensor([0.0444, 0.0413, 0.0490, 0.0502, 0.0550, 0.0488, 0.0435, 0.0556], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:31:05,768 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-05 22:31:11,871 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23579.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:14,783 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23583.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:16,181 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:20,318 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23591.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:38,767 INFO [train.py:901] (2/4) Epoch 3, batch 7450, loss[loss=0.3401, simple_loss=0.4043, pruned_loss=0.1379, over 8479.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3765, pruned_loss=0.1397, over 1605341.28 frames. ], batch size: 27, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:31:41,482 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.151e+02 3.560e+02 4.542e+02 5.434e+02 8.209e+02, threshold=9.083e+02, percent-clipped=0.0 +2023-02-05 22:31:44,206 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-05 22:32:11,863 INFO [train.py:901] (2/4) Epoch 3, batch 7500, loss[loss=0.3469, simple_loss=0.3996, pruned_loss=0.1471, over 8571.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3774, pruned_loss=0.1399, over 1611750.49 frames. ], batch size: 31, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:32:23,287 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4554, 1.8849, 1.8400, 0.8191, 1.7982, 1.3744, 0.4665, 1.7838], + device='cuda:2'), covar=tensor([0.0122, 0.0078, 0.0087, 0.0143, 0.0110, 0.0230, 0.0194, 0.0057], + device='cuda:2'), in_proj_covar=tensor([0.0245, 0.0175, 0.0144, 0.0223, 0.0170, 0.0294, 0.0238, 0.0196], + device='cuda:2'), out_proj_covar=tensor([1.0878e-04, 7.6709e-05, 6.2373e-05, 9.6665e-05, 7.7073e-05, 1.3988e-04, + 1.0696e-04, 8.5106e-05], device='cuda:2') +2023-02-05 22:32:28,861 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-05 22:32:31,249 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23694.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:32:39,264 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23706.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:32:47,010 INFO [train.py:901] (2/4) Epoch 3, batch 7550, loss[loss=0.2915, simple_loss=0.3545, pruned_loss=0.1143, over 8243.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.378, pruned_loss=0.1396, over 1615639.04 frames. ], batch size: 22, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:32:49,788 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.055e+02 3.573e+02 4.120e+02 5.568e+02 9.909e+02, threshold=8.240e+02, percent-clipped=1.0 +2023-02-05 22:33:21,016 INFO [train.py:901] (2/4) Epoch 3, batch 7600, loss[loss=0.3514, simple_loss=0.4081, pruned_loss=0.1474, over 8320.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3777, pruned_loss=0.1395, over 1611783.16 frames. ], batch size: 25, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:33:21,256 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4435, 1.8664, 2.1130, 1.0626, 1.9960, 1.3739, 0.7501, 1.7674], + device='cuda:2'), covar=tensor([0.0154, 0.0084, 0.0086, 0.0171, 0.0125, 0.0289, 0.0227, 0.0079], + device='cuda:2'), in_proj_covar=tensor([0.0246, 0.0176, 0.0145, 0.0226, 0.0173, 0.0300, 0.0240, 0.0202], + device='cuda:2'), out_proj_covar=tensor([1.0904e-04, 7.7271e-05, 6.2765e-05, 9.7671e-05, 7.8611e-05, 1.4334e-04, + 1.0812e-04, 8.7947e-05], device='cuda:2') +2023-02-05 22:33:25,982 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2023-02-05 22:33:55,880 INFO [train.py:901] (2/4) Epoch 3, batch 7650, loss[loss=0.2751, simple_loss=0.3454, pruned_loss=0.1024, over 8462.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3763, pruned_loss=0.1388, over 1614282.62 frames. ], batch size: 25, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:33:58,394 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.349e+02 3.333e+02 4.379e+02 5.791e+02 1.321e+03, threshold=8.759e+02, percent-clipped=7.0 +2023-02-05 22:34:12,402 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-05 22:34:13,553 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23841.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:34:19,388 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23850.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:34:29,028 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 22:34:30,200 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23866.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:34:30,645 INFO [train.py:901] (2/4) Epoch 3, batch 7700, loss[loss=0.3377, simple_loss=0.3959, pruned_loss=0.1398, over 8512.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3757, pruned_loss=0.1385, over 1616919.03 frames. ], batch size: 26, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:34:44,965 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7431, 1.9660, 3.0255, 1.2792, 2.5285, 1.9708, 1.7835, 2.2123], + device='cuda:2'), covar=tensor([0.0970, 0.1261, 0.0434, 0.1952, 0.0756, 0.1397, 0.0912, 0.1320], + device='cuda:2'), in_proj_covar=tensor([0.0449, 0.0412, 0.0498, 0.0506, 0.0549, 0.0489, 0.0431, 0.0562], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:34:50,582 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-05 22:34:50,859 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-05 22:35:04,773 INFO [train.py:901] (2/4) Epoch 3, batch 7750, loss[loss=0.3372, simple_loss=0.3913, pruned_loss=0.1416, over 8509.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3747, pruned_loss=0.1375, over 1618350.54 frames. ], batch size: 26, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:35:08,100 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 3.458e+02 4.167e+02 5.729e+02 1.393e+03, threshold=8.335e+02, percent-clipped=8.0 +2023-02-05 22:35:27,666 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23950.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:37,131 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23962.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:39,048 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23965.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:40,188 INFO [train.py:901] (2/4) Epoch 3, batch 7800, loss[loss=0.3238, simple_loss=0.3586, pruned_loss=0.1445, over 8088.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3738, pruned_loss=0.1369, over 1615855.38 frames. ], batch size: 21, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:35:45,559 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23975.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:53,490 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23987.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:36:14,028 INFO [train.py:901] (2/4) Epoch 3, batch 7850, loss[loss=0.239, simple_loss=0.3065, pruned_loss=0.08577, over 7793.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3746, pruned_loss=0.1371, over 1614619.40 frames. ], batch size: 19, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:36:16,553 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.230e+02 3.608e+02 4.565e+02 5.801e+02 1.089e+03, threshold=9.129e+02, percent-clipped=5.0 +2023-02-05 22:36:18,831 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3799, 1.9866, 3.3372, 2.6294, 2.6462, 2.0062, 1.3868, 1.2561], + device='cuda:2'), covar=tensor([0.1198, 0.1487, 0.0271, 0.0628, 0.0615, 0.0627, 0.0830, 0.1414], + device='cuda:2'), in_proj_covar=tensor([0.0637, 0.0556, 0.0463, 0.0519, 0.0630, 0.0512, 0.0529, 0.0536], + 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') +2023-02-05 22:36:39,257 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24055.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:36:47,366 INFO [train.py:901] (2/4) Epoch 3, batch 7900, loss[loss=0.4175, simple_loss=0.4433, pruned_loss=0.1958, over 6804.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3744, pruned_loss=0.1368, over 1617189.51 frames. ], batch size: 71, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:37:20,406 INFO [train.py:901] (2/4) Epoch 3, batch 7950, loss[loss=0.3536, simple_loss=0.3945, pruned_loss=0.1563, over 8362.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3749, pruned_loss=0.1369, over 1616148.48 frames. ], batch size: 24, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:37:23,166 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 3.295e+02 4.369e+02 5.897e+02 2.335e+03, threshold=8.738e+02, percent-clipped=5.0 +2023-02-05 22:37:45,312 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-05 22:37:54,048 INFO [train.py:901] (2/4) Epoch 3, batch 8000, loss[loss=0.3301, simple_loss=0.3667, pruned_loss=0.1467, over 7452.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3745, pruned_loss=0.137, over 1614930.49 frames. ], batch size: 17, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:37:54,334 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0922, 1.1614, 4.2038, 1.6394, 3.6567, 3.5261, 3.7254, 3.6168], + device='cuda:2'), covar=tensor([0.0313, 0.3215, 0.0313, 0.1968, 0.0944, 0.0518, 0.0395, 0.0500], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0439, 0.0334, 0.0356, 0.0425, 0.0347, 0.0333, 0.0372], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 22:37:59,711 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.6729, 3.7470, 3.3094, 1.5643, 3.3547, 3.2449, 3.5037, 2.8711], + device='cuda:2'), covar=tensor([0.1031, 0.0654, 0.1019, 0.4208, 0.0693, 0.0767, 0.1172, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0248, 0.0290, 0.0366, 0.0273, 0.0234, 0.0263, 0.0214], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:38:02,053 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-05 22:38:25,113 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.27 vs. limit=5.0 +2023-02-05 22:38:27,963 INFO [train.py:901] (2/4) Epoch 3, batch 8050, loss[loss=0.4157, simple_loss=0.4356, pruned_loss=0.1979, over 6955.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3727, pruned_loss=0.1375, over 1597158.95 frames. ], batch size: 71, lr: 2.09e-02, grad_scale: 8.0 +2023-02-05 22:38:30,754 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 3.328e+02 4.149e+02 5.404e+02 3.135e+03, threshold=8.298e+02, percent-clipped=6.0 +2023-02-05 22:38:30,985 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24221.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:38:44,820 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1263, 2.7771, 3.9348, 3.2146, 3.0575, 2.2642, 1.4843, 1.9689], + device='cuda:2'), covar=tensor([0.0871, 0.1158, 0.0237, 0.0505, 0.0590, 0.0581, 0.0813, 0.1136], + device='cuda:2'), in_proj_covar=tensor([0.0630, 0.0549, 0.0468, 0.0516, 0.0622, 0.0510, 0.0523, 0.0528], + 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') +2023-02-05 22:38:48,118 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24246.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:39:03,944 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 22:39:07,722 INFO [train.py:901] (2/4) Epoch 4, batch 0, loss[loss=0.3011, simple_loss=0.3579, pruned_loss=0.1222, over 8026.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3579, pruned_loss=0.1222, over 8026.00 frames. ], batch size: 22, lr: 1.96e-02, grad_scale: 8.0 +2023-02-05 22:39:07,722 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 22:39:18,719 INFO [train.py:935] (2/4) Epoch 4, validation: loss=0.2476, simple_loss=0.3384, pruned_loss=0.07836, over 944034.00 frames. +2023-02-05 22:39:18,719 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6564MB +2023-02-05 22:39:32,880 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.6978, 3.6689, 3.2487, 1.7623, 3.1514, 3.0984, 3.4193, 2.9311], + device='cuda:2'), covar=tensor([0.1171, 0.0666, 0.1092, 0.5078, 0.0967, 0.1239, 0.1271, 0.1101], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0247, 0.0288, 0.0371, 0.0276, 0.0233, 0.0268, 0.0216], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:39:34,134 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 22:39:52,985 INFO [train.py:901] (2/4) Epoch 4, batch 50, loss[loss=0.2869, simple_loss=0.3583, pruned_loss=0.1077, over 8491.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.375, pruned_loss=0.1356, over 365837.07 frames. ], batch size: 26, lr: 1.96e-02, grad_scale: 8.0 +2023-02-05 22:40:03,957 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4029, 1.7756, 3.2740, 2.6655, 2.5747, 1.8999, 1.3475, 1.2670], + device='cuda:2'), covar=tensor([0.1165, 0.1665, 0.0267, 0.0559, 0.0610, 0.0682, 0.0871, 0.1426], + device='cuda:2'), in_proj_covar=tensor([0.0631, 0.0555, 0.0470, 0.0516, 0.0634, 0.0513, 0.0532, 0.0535], + 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') +2023-02-05 22:40:07,592 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.017e+02 3.527e+02 4.250e+02 5.116e+02 9.987e+02, threshold=8.500e+02, percent-clipped=2.0 +2023-02-05 22:40:08,999 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 22:40:27,954 INFO [train.py:901] (2/4) Epoch 4, batch 100, loss[loss=0.379, simple_loss=0.4239, pruned_loss=0.1671, over 8360.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3738, pruned_loss=0.1375, over 637599.97 frames. ], batch size: 24, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:40:31,335 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 22:40:48,362 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9156, 1.3866, 2.9918, 1.5036, 2.5771, 2.5658, 2.7345, 2.7140], + device='cuda:2'), covar=tensor([0.0346, 0.2389, 0.0476, 0.1827, 0.0921, 0.0518, 0.0366, 0.0433], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0433, 0.0324, 0.0351, 0.0413, 0.0343, 0.0325, 0.0365], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 22:41:01,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24399.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:41:02,095 INFO [train.py:901] (2/4) Epoch 4, batch 150, loss[loss=0.2991, simple_loss=0.3557, pruned_loss=0.1213, over 8141.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.372, pruned_loss=0.1358, over 854350.89 frames. ], batch size: 22, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:41:17,163 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 3.490e+02 4.203e+02 5.614e+02 1.653e+03, threshold=8.406e+02, percent-clipped=4.0 +2023-02-05 22:41:37,211 INFO [train.py:901] (2/4) Epoch 4, batch 200, loss[loss=0.364, simple_loss=0.4092, pruned_loss=0.1594, over 8019.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3713, pruned_loss=0.1346, over 1020946.82 frames. ], batch size: 22, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:42:02,656 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-05 22:42:03,947 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0130, 2.4003, 4.8065, 1.3269, 2.8339, 2.3523, 1.8669, 2.6899], + device='cuda:2'), covar=tensor([0.1118, 0.1592, 0.0454, 0.2452, 0.1175, 0.1857, 0.1147, 0.1888], + device='cuda:2'), in_proj_covar=tensor([0.0468, 0.0424, 0.0511, 0.0519, 0.0567, 0.0498, 0.0451, 0.0577], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:42:11,040 INFO [train.py:901] (2/4) Epoch 4, batch 250, loss[loss=0.3639, simple_loss=0.4059, pruned_loss=0.1609, over 8622.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3697, pruned_loss=0.1329, over 1153564.52 frames. ], batch size: 49, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:42:12,491 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4537, 1.5421, 4.2051, 2.0050, 2.4697, 4.7966, 4.6641, 4.0719], + device='cuda:2'), covar=tensor([0.1106, 0.1539, 0.0282, 0.1700, 0.0778, 0.0344, 0.0254, 0.0636], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0266, 0.0213, 0.0266, 0.0207, 0.0185, 0.0191, 0.0269], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:42:16,536 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5109, 2.0366, 1.3666, 2.6755, 1.3118, 1.2232, 1.7503, 2.1948], + device='cuda:2'), covar=tensor([0.2021, 0.1707, 0.3222, 0.0535, 0.2113, 0.3184, 0.1848, 0.1252], + device='cuda:2'), in_proj_covar=tensor([0.0294, 0.0296, 0.0305, 0.0224, 0.0285, 0.0306, 0.0320, 0.0291], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:42:20,371 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24514.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:42:23,557 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 22:42:24,837 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.156e+02 3.531e+02 4.434e+02 5.277e+02 1.190e+03, threshold=8.868e+02, percent-clipped=4.0 +2023-02-05 22:42:31,623 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 22:42:46,001 INFO [train.py:901] (2/4) Epoch 4, batch 300, loss[loss=0.3167, simple_loss=0.381, pruned_loss=0.1262, over 8283.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3714, pruned_loss=0.134, over 1256769.72 frames. ], batch size: 23, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:42:57,003 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24565.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:43:12,327 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24587.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:43:16,435 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4696, 2.0942, 2.1202, 0.7628, 2.0289, 1.3109, 0.5689, 1.7895], + device='cuda:2'), covar=tensor([0.0161, 0.0061, 0.0060, 0.0161, 0.0102, 0.0259, 0.0229, 0.0068], + device='cuda:2'), in_proj_covar=tensor([0.0256, 0.0175, 0.0143, 0.0225, 0.0170, 0.0302, 0.0239, 0.0206], + device='cuda:2'), out_proj_covar=tensor([1.1207e-04, 7.5289e-05, 6.1270e-05, 9.5059e-05, 7.5759e-05, 1.4152e-04, + 1.0535e-04, 8.8042e-05], device='cuda:2') +2023-02-05 22:43:21,552 INFO [train.py:901] (2/4) Epoch 4, batch 350, loss[loss=0.2605, simple_loss=0.3114, pruned_loss=0.1048, over 7915.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3698, pruned_loss=0.1323, over 1336906.19 frames. ], batch size: 20, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:43:31,764 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5210, 2.0137, 2.1217, 0.6280, 1.9808, 1.4557, 0.6019, 1.7378], + device='cuda:2'), covar=tensor([0.0161, 0.0060, 0.0057, 0.0163, 0.0111, 0.0237, 0.0212, 0.0079], + device='cuda:2'), in_proj_covar=tensor([0.0259, 0.0178, 0.0146, 0.0229, 0.0172, 0.0306, 0.0242, 0.0210], + device='cuda:2'), out_proj_covar=tensor([1.1364e-04, 7.6424e-05, 6.2785e-05, 9.6814e-05, 7.6290e-05, 1.4319e-04, + 1.0660e-04, 8.9545e-05], device='cuda:2') +2023-02-05 22:43:34,090 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-05 22:43:35,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 3.300e+02 4.421e+02 5.071e+02 1.044e+03, threshold=8.841e+02, percent-clipped=4.0 +2023-02-05 22:43:52,422 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 22:43:56,477 INFO [train.py:901] (2/4) Epoch 4, batch 400, loss[loss=0.3465, simple_loss=0.3954, pruned_loss=0.1488, over 8553.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3708, pruned_loss=0.1328, over 1396852.00 frames. ], batch size: 39, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:44:04,951 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-05 22:44:30,026 INFO [train.py:901] (2/4) Epoch 4, batch 450, loss[loss=0.3246, simple_loss=0.3724, pruned_loss=0.1384, over 7930.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3726, pruned_loss=0.1336, over 1447405.54 frames. ], batch size: 20, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:44:44,808 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 3.414e+02 4.548e+02 5.600e+02 1.007e+03, threshold=9.096e+02, percent-clipped=5.0 +2023-02-05 22:45:04,969 INFO [train.py:901] (2/4) Epoch 4, batch 500, loss[loss=0.301, simple_loss=0.3566, pruned_loss=0.1227, over 8242.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3737, pruned_loss=0.1341, over 1486425.30 frames. ], batch size: 22, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:45:19,918 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24770.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:45:28,219 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24783.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:45:36,834 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24795.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:45:40,053 INFO [train.py:901] (2/4) Epoch 4, batch 550, loss[loss=0.3306, simple_loss=0.382, pruned_loss=0.1397, over 8129.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3743, pruned_loss=0.1338, over 1520617.20 frames. ], batch size: 22, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:45:50,772 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7140, 2.0395, 2.9213, 1.2130, 2.7887, 1.5980, 1.4149, 1.8563], + device='cuda:2'), covar=tensor([0.0231, 0.0102, 0.0086, 0.0186, 0.0110, 0.0275, 0.0240, 0.0125], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0174, 0.0142, 0.0222, 0.0167, 0.0300, 0.0235, 0.0207], + device='cuda:2'), out_proj_covar=tensor([1.1079e-04, 7.4644e-05, 6.0301e-05, 9.3264e-05, 7.3447e-05, 1.4024e-04, + 1.0275e-04, 8.7701e-05], device='cuda:2') +2023-02-05 22:45:53,858 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 3.369e+02 4.426e+02 5.591e+02 8.767e+02, threshold=8.852e+02, percent-clipped=0.0 +2023-02-05 22:46:08,701 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5704, 1.8408, 1.6031, 2.3752, 1.2623, 1.2209, 1.5462, 2.0627], + device='cuda:2'), covar=tensor([0.1257, 0.1411, 0.1787, 0.0609, 0.1848, 0.2683, 0.1917, 0.1015], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0299, 0.0309, 0.0227, 0.0278, 0.0308, 0.0323, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:46:13,959 INFO [train.py:901] (2/4) Epoch 4, batch 600, loss[loss=0.3192, simple_loss=0.3717, pruned_loss=0.1333, over 8447.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.374, pruned_loss=0.1331, over 1547852.35 frames. ], batch size: 49, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:46:16,993 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-05 22:46:24,941 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24866.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:46:28,959 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 22:46:43,929 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9019, 2.3406, 1.8630, 2.7374, 1.6016, 1.3826, 1.7767, 2.3080], + device='cuda:2'), covar=tensor([0.1253, 0.1239, 0.1598, 0.0543, 0.1741, 0.2597, 0.1874, 0.1148], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0300, 0.0306, 0.0226, 0.0275, 0.0310, 0.0317, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:46:49,153 INFO [train.py:901] (2/4) Epoch 4, batch 650, loss[loss=0.3044, simple_loss=0.369, pruned_loss=0.1199, over 8475.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3717, pruned_loss=0.132, over 1560888.10 frames. ], batch size: 25, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:46:55,190 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:47:03,759 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 3.310e+02 4.230e+02 5.108e+02 1.167e+03, threshold=8.459e+02, percent-clipped=4.0 +2023-02-05 22:47:10,602 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24931.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:47:24,033 INFO [train.py:901] (2/4) Epoch 4, batch 700, loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 8774.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3727, pruned_loss=0.1324, over 1576635.04 frames. ], batch size: 40, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:47:59,149 INFO [train.py:901] (2/4) Epoch 4, batch 750, loss[loss=0.3419, simple_loss=0.395, pruned_loss=0.1444, over 8658.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3714, pruned_loss=0.1313, over 1589679.47 frames. ], batch size: 34, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:48:08,254 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25013.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:48:13,342 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 3.175e+02 4.108e+02 5.247e+02 1.235e+03, threshold=8.217e+02, percent-clipped=4.0 +2023-02-05 22:48:14,035 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 22:48:15,517 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25024.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:48:22,616 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 22:48:30,669 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25046.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:48:33,225 INFO [train.py:901] (2/4) Epoch 4, batch 800, loss[loss=0.3102, simple_loss=0.3685, pruned_loss=0.1259, over 8462.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3712, pruned_loss=0.1318, over 1594933.26 frames. ], batch size: 27, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:48:50,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1825, 4.1975, 3.6835, 1.8535, 3.6939, 3.6124, 3.9127, 3.3450], + device='cuda:2'), covar=tensor([0.0654, 0.0469, 0.0931, 0.3969, 0.0699, 0.0820, 0.1066, 0.0798], + device='cuda:2'), in_proj_covar=tensor([0.0365, 0.0243, 0.0289, 0.0372, 0.0280, 0.0228, 0.0263, 0.0213], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:49:01,544 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-05 22:49:06,962 INFO [train.py:901] (2/4) Epoch 4, batch 850, loss[loss=0.2899, simple_loss=0.3424, pruned_loss=0.1187, over 7448.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3702, pruned_loss=0.1319, over 1597413.04 frames. ], batch size: 17, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:49:11,901 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4012, 2.1070, 2.3053, 1.0462, 2.2790, 1.5868, 0.7790, 1.8753], + device='cuda:2'), covar=tensor([0.0174, 0.0062, 0.0057, 0.0145, 0.0099, 0.0215, 0.0213, 0.0082], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0176, 0.0143, 0.0220, 0.0168, 0.0297, 0.0240, 0.0208], + device='cuda:2'), out_proj_covar=tensor([1.0997e-04, 7.4906e-05, 6.0257e-05, 9.1788e-05, 7.3460e-05, 1.3828e-04, + 1.0417e-04, 8.7632e-05], device='cuda:2') +2023-02-05 22:49:22,453 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 3.301e+02 4.277e+02 5.478e+02 1.022e+03, threshold=8.554e+02, percent-clipped=4.0 +2023-02-05 22:49:26,600 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25127.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:49:28,074 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0664, 1.5213, 1.4254, 1.1412, 1.3559, 1.3016, 1.4335, 1.4933], + device='cuda:2'), covar=tensor([0.0753, 0.1473, 0.2231, 0.1765, 0.0796, 0.1966, 0.0944, 0.0711], + device='cuda:2'), in_proj_covar=tensor([0.0169, 0.0204, 0.0247, 0.0206, 0.0167, 0.0212, 0.0171, 0.0175], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:49:42,455 INFO [train.py:901] (2/4) Epoch 4, batch 900, loss[loss=0.3238, simple_loss=0.3793, pruned_loss=0.1342, over 8645.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3702, pruned_loss=0.1319, over 1601826.08 frames. ], batch size: 49, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:49:42,805 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-05 22:50:16,545 INFO [train.py:901] (2/4) Epoch 4, batch 950, loss[loss=0.3221, simple_loss=0.3669, pruned_loss=0.1386, over 8084.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3675, pruned_loss=0.1302, over 1600090.65 frames. ], batch size: 21, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:50:23,515 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25210.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:50:28,346 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6427, 1.2599, 5.5178, 2.2324, 4.9657, 4.6779, 5.3139, 5.1450], + device='cuda:2'), covar=tensor([0.0236, 0.3847, 0.0245, 0.1890, 0.0813, 0.0474, 0.0242, 0.0339], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0439, 0.0338, 0.0353, 0.0413, 0.0353, 0.0322, 0.0370], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 22:50:30,879 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.215e+02 3.501e+02 4.488e+02 5.717e+02 1.063e+03, threshold=8.976e+02, percent-clipped=5.0 +2023-02-05 22:50:40,883 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 22:50:46,576 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25242.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:50:49,882 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4610, 1.3992, 4.6132, 1.8823, 3.9620, 3.7771, 4.1587, 4.0733], + device='cuda:2'), covar=tensor([0.0378, 0.3784, 0.0308, 0.2168, 0.1044, 0.0586, 0.0392, 0.0482], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0442, 0.0341, 0.0355, 0.0413, 0.0355, 0.0326, 0.0372], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 22:50:51,550 INFO [train.py:901] (2/4) Epoch 4, batch 1000, loss[loss=0.3901, simple_loss=0.4213, pruned_loss=0.1794, over 8319.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3692, pruned_loss=0.1313, over 1607376.30 frames. ], batch size: 25, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:51:12,236 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25280.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:13,325 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 22:51:25,837 INFO [train.py:901] (2/4) Epoch 4, batch 1050, loss[loss=0.3166, simple_loss=0.3779, pruned_loss=0.1276, over 8310.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3695, pruned_loss=0.1318, over 1605759.63 frames. ], batch size: 25, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:51:26,415 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 22:51:27,148 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25302.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:28,968 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25305.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:31,081 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 +2023-02-05 22:51:39,506 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 3.519e+02 4.399e+02 5.664e+02 1.146e+03, threshold=8.797e+02, percent-clipped=2.0 +2023-02-05 22:51:41,049 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2445, 1.5371, 1.4543, 1.1264, 1.5807, 1.3031, 1.6224, 1.5339], + device='cuda:2'), covar=tensor([0.0688, 0.1210, 0.2017, 0.1565, 0.0704, 0.1736, 0.0945, 0.0667], + device='cuda:2'), in_proj_covar=tensor([0.0166, 0.0202, 0.0243, 0.0202, 0.0164, 0.0209, 0.0171, 0.0174], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:51:42,398 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25325.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:43,771 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25327.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:53,723 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7881, 2.1036, 1.8467, 2.7739, 1.3718, 1.2673, 1.9082, 2.2976], + device='cuda:2'), covar=tensor([0.1104, 0.1238, 0.1460, 0.0362, 0.1523, 0.2184, 0.1339, 0.1012], + device='cuda:2'), in_proj_covar=tensor([0.0288, 0.0299, 0.0307, 0.0221, 0.0277, 0.0309, 0.0315, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 22:51:58,863 INFO [train.py:901] (2/4) Epoch 4, batch 1100, loss[loss=0.2619, simple_loss=0.3177, pruned_loss=0.103, over 7526.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3714, pruned_loss=0.1328, over 1612306.99 frames. ], batch size: 18, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:52:05,139 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25357.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:52:09,895 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6566, 1.7205, 2.0226, 1.6430, 1.2801, 2.0393, 0.4642, 1.1095], + device='cuda:2'), covar=tensor([0.3578, 0.2670, 0.2064, 0.3845, 0.6284, 0.1603, 0.7435, 0.2982], + device='cuda:2'), in_proj_covar=tensor([0.0114, 0.0110, 0.0082, 0.0156, 0.0188, 0.0080, 0.0153, 0.0115], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:52:27,703 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5190, 4.6084, 3.9865, 1.8057, 3.9643, 3.9865, 4.2010, 3.5742], + device='cuda:2'), covar=tensor([0.0874, 0.0456, 0.0885, 0.4714, 0.0752, 0.0787, 0.1408, 0.0797], + device='cuda:2'), in_proj_covar=tensor([0.0369, 0.0249, 0.0292, 0.0382, 0.0292, 0.0232, 0.0272, 0.0222], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 22:52:34,557 INFO [train.py:901] (2/4) Epoch 4, batch 1150, loss[loss=0.3212, simple_loss=0.3736, pruned_loss=0.1344, over 8462.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.371, pruned_loss=0.1327, over 1608618.05 frames. ], batch size: 49, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:52:37,395 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 22:52:49,211 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.195e+02 3.278e+02 3.972e+02 4.649e+02 8.065e+02, threshold=7.944e+02, percent-clipped=0.0 +2023-02-05 22:53:06,104 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25446.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 22:53:08,496 INFO [train.py:901] (2/4) Epoch 4, batch 1200, loss[loss=0.2584, simple_loss=0.3292, pruned_loss=0.09382, over 7799.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3703, pruned_loss=0.1325, over 1608664.76 frames. ], batch size: 19, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:53:23,226 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25472.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:53:36,619 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5596, 1.7998, 1.9332, 1.5071, 1.0104, 2.0597, 0.3926, 0.9168], + device='cuda:2'), covar=tensor([0.4158, 0.2623, 0.1488, 0.2860, 0.7601, 0.1078, 0.6953, 0.2991], + device='cuda:2'), in_proj_covar=tensor([0.0119, 0.0113, 0.0086, 0.0163, 0.0194, 0.0083, 0.0158, 0.0119], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 22:53:42,019 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25498.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:53:43,163 INFO [train.py:901] (2/4) Epoch 4, batch 1250, loss[loss=0.2776, simple_loss=0.327, pruned_loss=0.1141, over 7924.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3699, pruned_loss=0.1327, over 1611679.63 frames. ], batch size: 20, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:53:57,794 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 3.538e+02 4.328e+02 6.105e+02 1.271e+03, threshold=8.657e+02, percent-clipped=4.0 +2023-02-05 22:53:59,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25523.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:54:06,838 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-05 22:54:18,017 INFO [train.py:901] (2/4) Epoch 4, batch 1300, loss[loss=0.3877, simple_loss=0.4299, pruned_loss=0.1728, over 8540.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3698, pruned_loss=0.1324, over 1611551.30 frames. ], batch size: 31, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:54:39,386 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25581.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:54:53,160 INFO [train.py:901] (2/4) Epoch 4, batch 1350, loss[loss=0.3494, simple_loss=0.3972, pruned_loss=0.1508, over 8592.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.371, pruned_loss=0.1324, over 1613095.85 frames. ], batch size: 31, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:54:57,488 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25606.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:55:08,861 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.283e+02 4.098e+02 5.393e+02 1.175e+03, threshold=8.196e+02, percent-clipped=3.0 +2023-02-05 22:55:28,873 INFO [train.py:901] (2/4) Epoch 4, batch 1400, loss[loss=0.3071, simple_loss=0.3619, pruned_loss=0.1261, over 8493.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3704, pruned_loss=0.1321, over 1614641.34 frames. ], batch size: 49, lr: 1.91e-02, grad_scale: 8.0 +2023-02-05 22:55:29,807 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2964, 1.8262, 1.8097, 0.5523, 1.8603, 1.2238, 0.4792, 1.5941], + device='cuda:2'), covar=tensor([0.0152, 0.0072, 0.0070, 0.0168, 0.0079, 0.0286, 0.0228, 0.0071], + device='cuda:2'), in_proj_covar=tensor([0.0256, 0.0184, 0.0146, 0.0226, 0.0169, 0.0301, 0.0252, 0.0211], + device='cuda:2'), out_proj_covar=tensor([1.0935e-04, 7.7768e-05, 6.0936e-05, 9.3158e-05, 7.2259e-05, 1.3835e-04, + 1.0928e-04, 8.8565e-05], device='cuda:2') +2023-02-05 22:56:03,160 INFO [train.py:901] (2/4) Epoch 4, batch 1450, loss[loss=0.2987, simple_loss=0.3486, pruned_loss=0.1244, over 7788.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3687, pruned_loss=0.131, over 1613631.09 frames. ], batch size: 19, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:56:05,847 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 22:56:14,685 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-05 22:56:18,904 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 3.243e+02 3.964e+02 4.847e+02 1.034e+03, threshold=7.929e+02, percent-clipped=2.0 +2023-02-05 22:56:23,195 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25728.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:56:38,595 INFO [train.py:901] (2/4) Epoch 4, batch 1500, loss[loss=0.2873, simple_loss=0.35, pruned_loss=0.1123, over 8077.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3686, pruned_loss=0.131, over 1615469.80 frames. ], batch size: 21, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:56:40,765 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25753.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:57:05,989 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25790.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 22:57:12,574 INFO [train.py:901] (2/4) Epoch 4, batch 1550, loss[loss=0.2707, simple_loss=0.3416, pruned_loss=0.09994, over 8498.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3677, pruned_loss=0.13, over 1612763.09 frames. ], batch size: 28, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:57:20,794 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 +2023-02-05 22:57:27,006 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 3.100e+02 3.836e+02 5.066e+02 1.009e+03, threshold=7.672e+02, percent-clipped=5.0 +2023-02-05 22:57:46,740 INFO [train.py:901] (2/4) Epoch 4, batch 1600, loss[loss=0.3198, simple_loss=0.3676, pruned_loss=0.136, over 7821.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3688, pruned_loss=0.1309, over 1614571.71 frames. ], batch size: 20, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:58:04,753 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25876.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 22:58:21,019 INFO [train.py:901] (2/4) Epoch 4, batch 1650, loss[loss=0.3951, simple_loss=0.4226, pruned_loss=0.1838, over 7149.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3679, pruned_loss=0.1309, over 1606248.51 frames. ], batch size: 71, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:58:24,588 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25905.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 22:58:35,953 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.227e+02 3.823e+02 4.768e+02 5.766e+02 1.707e+03, threshold=9.535e+02, percent-clipped=9.0 +2023-02-05 22:58:56,117 INFO [train.py:901] (2/4) Epoch 4, batch 1700, loss[loss=0.3178, simple_loss=0.3689, pruned_loss=0.1333, over 8495.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3683, pruned_loss=0.1309, over 1600160.83 frames. ], batch size: 26, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:59:14,220 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.46 vs. limit=5.0 +2023-02-05 22:59:31,188 INFO [train.py:901] (2/4) Epoch 4, batch 1750, loss[loss=0.2703, simple_loss=0.3302, pruned_loss=0.1052, over 7824.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3683, pruned_loss=0.1308, over 1606282.16 frames. ], batch size: 20, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 22:59:35,850 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1015, 1.6169, 3.2868, 1.3897, 2.1769, 3.7145, 3.5391, 3.2163], + device='cuda:2'), covar=tensor([0.1089, 0.1458, 0.0391, 0.2071, 0.0863, 0.0266, 0.0326, 0.0599], + device='cuda:2'), in_proj_covar=tensor([0.0236, 0.0266, 0.0212, 0.0277, 0.0217, 0.0194, 0.0196, 0.0266], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 22:59:47,002 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 3.187e+02 3.816e+02 4.801e+02 8.317e+02, threshold=7.632e+02, percent-clipped=0.0 +2023-02-05 23:00:06,097 INFO [train.py:901] (2/4) Epoch 4, batch 1800, loss[loss=0.2887, simple_loss=0.3349, pruned_loss=0.1213, over 7704.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3672, pruned_loss=0.1305, over 1604689.85 frames. ], batch size: 18, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:00:41,282 INFO [train.py:901] (2/4) Epoch 4, batch 1850, loss[loss=0.3179, simple_loss=0.3796, pruned_loss=0.1281, over 8503.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3664, pruned_loss=0.1294, over 1607075.80 frames. ], batch size: 26, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:00:55,439 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26120.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:00:56,608 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.260e+02 3.379e+02 4.261e+02 5.084e+02 1.608e+03, threshold=8.521e+02, percent-clipped=6.0 +2023-02-05 23:01:15,412 INFO [train.py:901] (2/4) Epoch 4, batch 1900, loss[loss=0.3001, simple_loss=0.3399, pruned_loss=0.1301, over 7787.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3656, pruned_loss=0.129, over 1610720.67 frames. ], batch size: 19, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:01:22,879 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26161.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:01:29,033 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 +2023-02-05 23:01:40,600 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26186.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:01:40,662 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26186.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:01:41,083 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 23:01:49,628 INFO [train.py:901] (2/4) Epoch 4, batch 1950, loss[loss=0.3514, simple_loss=0.4022, pruned_loss=0.1503, over 8545.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3662, pruned_loss=0.1294, over 1613397.18 frames. ], batch size: 49, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:01:52,431 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 23:02:04,048 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26220.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:02:05,152 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.303e+02 3.684e+02 4.572e+02 6.046e+02 1.247e+03, threshold=9.144e+02, percent-clipped=2.0 +2023-02-05 23:02:10,406 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 23:02:22,285 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-05 23:02:24,334 INFO [train.py:901] (2/4) Epoch 4, batch 2000, loss[loss=0.3698, simple_loss=0.4134, pruned_loss=0.1631, over 8030.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3665, pruned_loss=0.1295, over 1615926.87 frames. ], batch size: 22, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:02:36,631 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26268.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:02:59,488 INFO [train.py:901] (2/4) Epoch 4, batch 2050, loss[loss=0.2981, simple_loss=0.3627, pruned_loss=0.1167, over 8103.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3675, pruned_loss=0.1304, over 1618516.98 frames. ], batch size: 23, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:03:14,377 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.223e+02 3.433e+02 4.198e+02 5.260e+02 1.263e+03, threshold=8.396e+02, percent-clipped=5.0 +2023-02-05 23:03:14,623 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6531, 1.7920, 2.0379, 1.5471, 0.8512, 1.9652, 0.3664, 1.2768], + device='cuda:2'), covar=tensor([0.3874, 0.2025, 0.0875, 0.2384, 0.7274, 0.0978, 0.6407, 0.2540], + device='cuda:2'), in_proj_covar=tensor([0.0121, 0.0110, 0.0082, 0.0159, 0.0191, 0.0085, 0.0150, 0.0119], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 23:03:24,280 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26335.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:03:34,743 INFO [train.py:901] (2/4) Epoch 4, batch 2100, loss[loss=0.2548, simple_loss=0.3253, pruned_loss=0.0922, over 8089.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3682, pruned_loss=0.1309, over 1617834.26 frames. ], batch size: 21, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:04:08,106 INFO [train.py:901] (2/4) Epoch 4, batch 2150, loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.1209, over 8025.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3683, pruned_loss=0.1307, over 1618645.20 frames. ], batch size: 22, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:04:24,393 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.431e+02 3.407e+02 4.210e+02 5.616e+02 1.521e+03, threshold=8.419e+02, percent-clipped=4.0 +2023-02-05 23:04:31,146 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:04:43,605 INFO [train.py:901] (2/4) Epoch 4, batch 2200, loss[loss=0.2715, simple_loss=0.3407, pruned_loss=0.1012, over 8517.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3679, pruned_loss=0.1299, over 1621762.02 frames. ], batch size: 28, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:04:53,163 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26464.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:05:18,123 INFO [train.py:901] (2/4) Epoch 4, batch 2250, loss[loss=0.3501, simple_loss=0.3973, pruned_loss=0.1514, over 8508.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3678, pruned_loss=0.1305, over 1615511.80 frames. ], batch size: 28, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:05:25,219 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3849, 1.9794, 2.1826, 0.8263, 2.1937, 1.5225, 0.6542, 1.7672], + device='cuda:2'), covar=tensor([0.0188, 0.0077, 0.0081, 0.0162, 0.0095, 0.0249, 0.0221, 0.0081], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0193, 0.0156, 0.0231, 0.0183, 0.0311, 0.0254, 0.0218], + device='cuda:2'), out_proj_covar=tensor([1.1417e-04, 8.0643e-05, 6.3799e-05, 9.4011e-05, 7.7814e-05, 1.4057e-04, + 1.0803e-04, 8.9445e-05], device='cuda:2') +2023-02-05 23:05:33,090 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 3.188e+02 3.857e+02 4.748e+02 9.287e+02, threshold=7.714e+02, percent-clipped=1.0 +2023-02-05 23:05:38,936 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26530.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:05:52,824 INFO [train.py:901] (2/4) Epoch 4, batch 2300, loss[loss=0.2805, simple_loss=0.3324, pruned_loss=0.1143, over 6789.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3679, pruned_loss=0.1306, over 1613745.33 frames. ], batch size: 15, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:06:12,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26579.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:06:21,882 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26591.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:06:27,546 INFO [train.py:901] (2/4) Epoch 4, batch 2350, loss[loss=0.4086, simple_loss=0.4307, pruned_loss=0.1933, over 7210.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3674, pruned_loss=0.1302, over 1610550.18 frames. ], batch size: 72, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:06:35,960 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26612.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:06:38,760 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26616.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:06:42,410 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 3.505e+02 4.841e+02 5.770e+02 1.247e+03, threshold=9.683e+02, percent-clipped=6.0 +2023-02-05 23:06:58,652 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26645.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:07:00,513 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2517, 1.3127, 2.2810, 1.1360, 2.1547, 2.4682, 2.4547, 2.0758], + device='cuda:2'), covar=tensor([0.0912, 0.1072, 0.0451, 0.1683, 0.0479, 0.0333, 0.0371, 0.0659], + device='cuda:2'), in_proj_covar=tensor([0.0222, 0.0257, 0.0205, 0.0262, 0.0208, 0.0186, 0.0194, 0.0259], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:07:01,512 INFO [train.py:901] (2/4) Epoch 4, batch 2400, loss[loss=0.3169, simple_loss=0.3697, pruned_loss=0.1321, over 7929.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3701, pruned_loss=0.1319, over 1615571.09 frames. ], batch size: 20, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:07:26,619 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1534, 1.1903, 4.3761, 1.6370, 3.6203, 3.5542, 3.8171, 3.7129], + device='cuda:2'), covar=tensor([0.0378, 0.3695, 0.0317, 0.2459, 0.0937, 0.0563, 0.0467, 0.0503], + device='cuda:2'), in_proj_covar=tensor([0.0274, 0.0450, 0.0351, 0.0368, 0.0433, 0.0366, 0.0347, 0.0384], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:07:37,131 INFO [train.py:901] (2/4) Epoch 4, batch 2450, loss[loss=0.3072, simple_loss=0.3739, pruned_loss=0.1202, over 8335.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3699, pruned_loss=0.1317, over 1615814.21 frames. ], batch size: 25, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:07:48,166 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-05 23:07:51,856 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 3.211e+02 4.300e+02 5.616e+02 1.854e+03, threshold=8.599e+02, percent-clipped=7.0 +2023-02-05 23:07:55,325 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26727.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:08:10,583 INFO [train.py:901] (2/4) Epoch 4, batch 2500, loss[loss=0.3255, simple_loss=0.3792, pruned_loss=0.1359, over 8030.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3688, pruned_loss=0.1308, over 1612083.40 frames. ], batch size: 22, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:08:24,858 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1543, 1.9079, 2.9441, 2.4621, 2.4096, 1.8856, 1.3670, 1.0491], + device='cuda:2'), covar=tensor([0.1455, 0.1491, 0.0343, 0.0671, 0.0700, 0.0804, 0.0936, 0.1586], + device='cuda:2'), in_proj_covar=tensor([0.0663, 0.0590, 0.0489, 0.0565, 0.0675, 0.0546, 0.0546, 0.0556], + 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') +2023-02-05 23:08:29,250 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26776.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:08:45,150 INFO [train.py:901] (2/4) Epoch 4, batch 2550, loss[loss=0.2528, simple_loss=0.3093, pruned_loss=0.09817, over 7434.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3683, pruned_loss=0.1308, over 1613364.23 frames. ], batch size: 17, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:09:01,300 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.235e+02 3.301e+02 4.146e+02 5.074e+02 1.055e+03, threshold=8.293e+02, percent-clipped=2.0 +2023-02-05 23:09:10,582 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26835.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:09:16,260 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-05 23:09:20,514 INFO [train.py:901] (2/4) Epoch 4, batch 2600, loss[loss=0.251, simple_loss=0.322, pruned_loss=0.09, over 7800.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3677, pruned_loss=0.1298, over 1614236.57 frames. ], batch size: 20, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:09:27,765 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26860.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:09:50,369 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26891.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:09:56,423 INFO [train.py:901] (2/4) Epoch 4, batch 2650, loss[loss=0.2871, simple_loss=0.3573, pruned_loss=0.1084, over 7640.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3658, pruned_loss=0.1284, over 1609728.97 frames. ], batch size: 19, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:09:57,280 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26901.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:10:12,341 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 3.245e+02 3.916e+02 5.024e+02 1.006e+03, threshold=7.831e+02, percent-clipped=3.0 +2023-02-05 23:10:12,506 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26922.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:10:15,318 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26926.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:10:32,094 INFO [train.py:901] (2/4) Epoch 4, batch 2700, loss[loss=0.3496, simple_loss=0.3851, pruned_loss=0.1571, over 8354.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3667, pruned_loss=0.1296, over 1608789.96 frames. ], batch size: 24, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:10:44,345 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26968.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:10:45,657 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5975, 1.9706, 5.7493, 2.1141, 5.0463, 4.6955, 5.2789, 5.1508], + device='cuda:2'), covar=tensor([0.0404, 0.3152, 0.0183, 0.2151, 0.0813, 0.0433, 0.0296, 0.0342], + device='cuda:2'), in_proj_covar=tensor([0.0280, 0.0458, 0.0356, 0.0372, 0.0445, 0.0368, 0.0352, 0.0394], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:10:47,100 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8779, 2.2649, 3.9214, 3.0702, 3.1346, 2.0422, 1.4919, 1.6842], + device='cuda:2'), covar=tensor([0.1202, 0.1661, 0.0321, 0.0704, 0.0727, 0.0785, 0.0824, 0.1625], + device='cuda:2'), in_proj_covar=tensor([0.0668, 0.0594, 0.0494, 0.0564, 0.0679, 0.0546, 0.0547, 0.0561], + 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') +2023-02-05 23:10:54,314 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26983.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:11:05,897 INFO [train.py:901] (2/4) Epoch 4, batch 2750, loss[loss=0.3148, simple_loss=0.3552, pruned_loss=0.1372, over 7800.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3665, pruned_loss=0.1295, over 1605739.83 frames. ], batch size: 19, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:11:12,363 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27008.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:11:21,382 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 3.589e+02 4.354e+02 5.460e+02 1.197e+03, threshold=8.707e+02, percent-clipped=9.0 +2023-02-05 23:11:40,845 INFO [train.py:901] (2/4) Epoch 4, batch 2800, loss[loss=0.3061, simple_loss=0.3709, pruned_loss=0.1207, over 8258.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3672, pruned_loss=0.1299, over 1609820.39 frames. ], batch size: 24, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:11:52,397 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3625, 1.5401, 1.5879, 1.2883, 0.8860, 1.6307, 0.0636, 1.0711], + device='cuda:2'), covar=tensor([0.4288, 0.2391, 0.1564, 0.2043, 0.6177, 0.1197, 0.6025, 0.2548], + device='cuda:2'), in_proj_covar=tensor([0.0125, 0.0114, 0.0083, 0.0162, 0.0200, 0.0084, 0.0155, 0.0121], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 23:12:04,130 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-05 23:12:14,851 INFO [train.py:901] (2/4) Epoch 4, batch 2850, loss[loss=0.3494, simple_loss=0.3932, pruned_loss=0.1528, over 8506.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3671, pruned_loss=0.1304, over 1607061.07 frames. ], batch size: 26, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:12:30,250 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 3.374e+02 4.464e+02 5.831e+02 1.992e+03, threshold=8.927e+02, percent-clipped=6.0 +2023-02-05 23:12:30,479 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2869, 1.6706, 1.5154, 0.4988, 1.5397, 1.2692, 0.2454, 1.5608], + device='cuda:2'), covar=tensor([0.0178, 0.0090, 0.0080, 0.0158, 0.0115, 0.0283, 0.0251, 0.0077], + device='cuda:2'), in_proj_covar=tensor([0.0265, 0.0191, 0.0158, 0.0225, 0.0183, 0.0309, 0.0255, 0.0214], + device='cuda:2'), out_proj_covar=tensor([1.1074e-04, 7.8909e-05, 6.3577e-05, 9.0078e-05, 7.7392e-05, 1.3781e-04, + 1.0779e-04, 8.6775e-05], device='cuda:2') +2023-02-05 23:12:47,569 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27147.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:12:49,277 INFO [train.py:901] (2/4) Epoch 4, batch 2900, loss[loss=0.3301, simple_loss=0.3763, pruned_loss=0.1419, over 7429.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3676, pruned_loss=0.1312, over 1603831.89 frames. ], batch size: 17, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:13:00,624 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27166.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:13:05,282 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27172.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:13:11,805 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 23:13:24,074 INFO [train.py:901] (2/4) Epoch 4, batch 2950, loss[loss=0.2905, simple_loss=0.3586, pruned_loss=0.1112, over 8082.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3672, pruned_loss=0.131, over 1605416.53 frames. ], batch size: 21, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:13:38,737 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 3.092e+02 3.649e+02 5.055e+02 1.216e+03, threshold=7.299e+02, percent-clipped=3.0 +2023-02-05 23:13:58,832 INFO [train.py:901] (2/4) Epoch 4, batch 3000, loss[loss=0.4599, simple_loss=0.4529, pruned_loss=0.2335, over 7477.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3662, pruned_loss=0.1299, over 1604771.34 frames. ], batch size: 73, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:13:58,832 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 23:14:11,273 INFO [train.py:935] (2/4) Epoch 4, validation: loss=0.2374, simple_loss=0.3304, pruned_loss=0.07225, over 944034.00 frames. +2023-02-05 23:14:11,274 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6599MB +2023-02-05 23:14:23,016 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27266.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:14:45,721 INFO [train.py:901] (2/4) Epoch 4, batch 3050, loss[loss=0.3566, simple_loss=0.4055, pruned_loss=0.1539, over 8581.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3655, pruned_loss=0.1289, over 1608633.68 frames. ], batch size: 31, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:14:54,658 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27312.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:14:55,443 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0828, 1.2443, 3.1642, 0.9101, 2.7101, 2.6911, 2.9253, 2.8413], + device='cuda:2'), covar=tensor([0.0509, 0.2840, 0.0575, 0.2336, 0.1352, 0.0768, 0.0576, 0.0694], + device='cuda:2'), in_proj_covar=tensor([0.0282, 0.0458, 0.0357, 0.0371, 0.0447, 0.0368, 0.0364, 0.0402], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:15:01,945 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 3.415e+02 4.317e+02 5.768e+02 1.933e+03, threshold=8.634e+02, percent-clipped=10.0 +2023-02-05 23:15:20,630 INFO [train.py:901] (2/4) Epoch 4, batch 3100, loss[loss=0.3306, simple_loss=0.3775, pruned_loss=0.1418, over 7695.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3656, pruned_loss=0.1289, over 1611171.28 frames. ], batch size: 76, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:15:41,815 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27381.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:15:54,817 INFO [train.py:901] (2/4) Epoch 4, batch 3150, loss[loss=0.29, simple_loss=0.3412, pruned_loss=0.1194, over 7177.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3654, pruned_loss=0.1287, over 1607839.92 frames. ], batch size: 16, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:16:09,489 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 3.237e+02 4.041e+02 5.193e+02 1.210e+03, threshold=8.082e+02, percent-clipped=3.0 +2023-02-05 23:16:13,659 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27427.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:16:16,356 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8755, 2.1231, 3.4101, 1.3178, 2.5876, 2.1650, 1.9705, 2.2932], + device='cuda:2'), covar=tensor([0.0876, 0.1229, 0.0309, 0.2104, 0.0835, 0.1352, 0.0852, 0.1254], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0417, 0.0499, 0.0512, 0.0560, 0.0494, 0.0430, 0.0560], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:16:29,612 INFO [train.py:901] (2/4) Epoch 4, batch 3200, loss[loss=0.2361, simple_loss=0.3011, pruned_loss=0.08549, over 7326.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3658, pruned_loss=0.1291, over 1608985.94 frames. ], batch size: 16, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:16:50,835 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8420, 2.7484, 2.4928, 1.5484, 2.4825, 2.4693, 2.5743, 2.2075], + device='cuda:2'), covar=tensor([0.1245, 0.1026, 0.1225, 0.4062, 0.1074, 0.1091, 0.1518, 0.1046], + device='cuda:2'), in_proj_covar=tensor([0.0366, 0.0260, 0.0301, 0.0383, 0.0290, 0.0238, 0.0282, 0.0221], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 23:17:03,119 INFO [train.py:901] (2/4) Epoch 4, batch 3250, loss[loss=0.3038, simple_loss=0.3543, pruned_loss=0.1266, over 7430.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3667, pruned_loss=0.1291, over 1615289.42 frames. ], batch size: 17, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:17:07,514 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-05 23:17:10,580 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27510.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:17:18,422 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 3.449e+02 4.059e+02 4.930e+02 7.939e+02, threshold=8.117e+02, percent-clipped=0.0 +2023-02-05 23:17:37,477 INFO [train.py:901] (2/4) Epoch 4, batch 3300, loss[loss=0.2972, simple_loss=0.3606, pruned_loss=0.1169, over 8294.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3647, pruned_loss=0.1279, over 1609533.93 frames. ], batch size: 23, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:18:12,287 INFO [train.py:901] (2/4) Epoch 4, batch 3350, loss[loss=0.2735, simple_loss=0.3252, pruned_loss=0.1109, over 7810.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3644, pruned_loss=0.1274, over 1612580.35 frames. ], batch size: 20, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:18:12,670 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2144, 1.9550, 2.8706, 2.3484, 2.3728, 1.9950, 1.4369, 1.0170], + device='cuda:2'), covar=tensor([0.1307, 0.1365, 0.0324, 0.0672, 0.0607, 0.0660, 0.0835, 0.1396], + device='cuda:2'), in_proj_covar=tensor([0.0670, 0.0605, 0.0510, 0.0570, 0.0678, 0.0556, 0.0557, 0.0564], + 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') +2023-02-05 23:18:28,393 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 3.326e+02 4.176e+02 5.439e+02 1.733e+03, threshold=8.353e+02, percent-clipped=9.0 +2023-02-05 23:18:30,492 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27625.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:18:38,368 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27637.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:18:46,470 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4252, 1.8724, 1.8743, 0.8781, 1.8809, 1.3511, 0.3836, 1.6839], + device='cuda:2'), covar=tensor([0.0186, 0.0101, 0.0091, 0.0151, 0.0098, 0.0305, 0.0266, 0.0073], + device='cuda:2'), in_proj_covar=tensor([0.0267, 0.0196, 0.0157, 0.0227, 0.0182, 0.0313, 0.0255, 0.0217], + device='cuda:2'), out_proj_covar=tensor([1.1082e-04, 8.0562e-05, 6.2599e-05, 9.0173e-05, 7.5884e-05, 1.3866e-04, + 1.0689e-04, 8.7278e-05], device='cuda:2') +2023-02-05 23:18:46,885 INFO [train.py:901] (2/4) Epoch 4, batch 3400, loss[loss=0.3182, simple_loss=0.378, pruned_loss=0.1292, over 7935.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.365, pruned_loss=0.1281, over 1611997.90 frames. ], batch size: 20, lr: 1.84e-02, grad_scale: 16.0 +2023-02-05 23:18:55,799 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27662.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:19:10,318 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27683.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:19:10,336 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4564, 1.8425, 3.2107, 1.0596, 2.2843, 1.8246, 1.5606, 1.8441], + device='cuda:2'), covar=tensor([0.1388, 0.1639, 0.0604, 0.2774, 0.1285, 0.2048, 0.1378, 0.1879], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0428, 0.0512, 0.0518, 0.0567, 0.0495, 0.0443, 0.0568], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:19:21,470 INFO [train.py:901] (2/4) Epoch 4, batch 3450, loss[loss=0.2952, simple_loss=0.3551, pruned_loss=0.1177, over 8285.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3652, pruned_loss=0.1282, over 1613750.03 frames. ], batch size: 23, lr: 1.84e-02, grad_scale: 16.0 +2023-02-05 23:19:26,946 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27708.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:19:36,059 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.499e+02 3.357e+02 4.072e+02 5.275e+02 9.264e+02, threshold=8.144e+02, percent-clipped=1.0 +2023-02-05 23:19:40,426 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-05 23:19:43,549 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27732.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:19:46,192 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.3829, 5.3356, 4.8680, 2.1195, 4.7839, 4.9752, 4.9231, 4.3181], + device='cuda:2'), covar=tensor([0.0637, 0.0389, 0.0631, 0.4155, 0.0610, 0.0596, 0.0955, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0364, 0.0257, 0.0299, 0.0383, 0.0283, 0.0233, 0.0275, 0.0216], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 23:19:55,909 INFO [train.py:901] (2/4) Epoch 4, batch 3500, loss[loss=0.2779, simple_loss=0.3324, pruned_loss=0.1117, over 7782.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3649, pruned_loss=0.1276, over 1613360.29 frames. ], batch size: 19, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:20:10,690 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 23:20:31,114 INFO [train.py:901] (2/4) Epoch 4, batch 3550, loss[loss=0.2927, simple_loss=0.3431, pruned_loss=0.1212, over 7639.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3641, pruned_loss=0.1271, over 1614607.14 frames. ], batch size: 19, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:20:46,085 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 3.262e+02 3.955e+02 5.254e+02 1.114e+03, threshold=7.909e+02, percent-clipped=8.0 +2023-02-05 23:21:05,473 INFO [train.py:901] (2/4) Epoch 4, batch 3600, loss[loss=0.3037, simple_loss=0.3714, pruned_loss=0.1179, over 8342.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3646, pruned_loss=0.1275, over 1609612.21 frames. ], batch size: 26, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:21:27,351 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27881.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:21:32,203 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 23:21:33,264 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5920, 5.5407, 4.5843, 2.3203, 4.7807, 5.1870, 5.0568, 4.6766], + device='cuda:2'), covar=tensor([0.0755, 0.0608, 0.1002, 0.4309, 0.0668, 0.0735, 0.1133, 0.0790], + device='cuda:2'), in_proj_covar=tensor([0.0364, 0.0262, 0.0294, 0.0384, 0.0287, 0.0240, 0.0281, 0.0217], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 23:21:39,940 INFO [train.py:901] (2/4) Epoch 4, batch 3650, loss[loss=0.3041, simple_loss=0.363, pruned_loss=0.1226, over 8460.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3642, pruned_loss=0.1271, over 1611308.29 frames. ], batch size: 25, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:21:44,920 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27906.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:21:56,097 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.998e+02 3.334e+02 3.945e+02 4.811e+02 1.062e+03, threshold=7.891e+02, percent-clipped=4.0 +2023-02-05 23:22:13,494 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 23:22:14,786 INFO [train.py:901] (2/4) Epoch 4, batch 3700, loss[loss=0.3544, simple_loss=0.3999, pruned_loss=0.1545, over 8634.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3652, pruned_loss=0.1283, over 1607250.22 frames. ], batch size: 34, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:22:49,604 INFO [train.py:901] (2/4) Epoch 4, batch 3750, loss[loss=0.3563, simple_loss=0.4097, pruned_loss=0.1514, over 8483.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3664, pruned_loss=0.1289, over 1610502.25 frames. ], batch size: 29, lr: 1.83e-02, grad_scale: 8.0 +2023-02-05 23:23:05,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.245e+02 3.553e+02 4.442e+02 6.055e+02 1.985e+03, threshold=8.883e+02, percent-clipped=11.0 +2023-02-05 23:23:24,893 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-05 23:23:25,245 INFO [train.py:901] (2/4) Epoch 4, batch 3800, loss[loss=0.2972, simple_loss=0.3682, pruned_loss=0.113, over 8293.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3654, pruned_loss=0.1283, over 1608394.58 frames. ], batch size: 23, lr: 1.83e-02, grad_scale: 8.0 +2023-02-05 23:23:42,796 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28076.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:24:00,305 INFO [train.py:901] (2/4) Epoch 4, batch 3850, loss[loss=0.2872, simple_loss=0.3435, pruned_loss=0.1155, over 8096.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3641, pruned_loss=0.1275, over 1606321.76 frames. ], batch size: 21, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:24:15,224 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 3.238e+02 4.124e+02 5.182e+02 9.210e+02, threshold=8.247e+02, percent-clipped=1.0 +2023-02-05 23:24:17,311 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 23:24:34,689 INFO [train.py:901] (2/4) Epoch 4, batch 3900, loss[loss=0.264, simple_loss=0.3171, pruned_loss=0.1054, over 7697.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3648, pruned_loss=0.1277, over 1606153.61 frames. ], batch size: 18, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:25:02,727 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28191.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:25:08,591 INFO [train.py:901] (2/4) Epoch 4, batch 3950, loss[loss=0.3299, simple_loss=0.3788, pruned_loss=0.1405, over 8512.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3641, pruned_loss=0.1278, over 1601837.18 frames. ], batch size: 26, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:25:17,220 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 23:25:24,837 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 3.357e+02 4.080e+02 5.453e+02 1.389e+03, threshold=8.161e+02, percent-clipped=8.0 +2023-02-05 23:25:41,193 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28247.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:25:43,104 INFO [train.py:901] (2/4) Epoch 4, batch 4000, loss[loss=0.2743, simple_loss=0.3223, pruned_loss=0.1132, over 7532.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3644, pruned_loss=0.1281, over 1603018.43 frames. ], batch size: 18, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:25:59,931 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28273.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:26:17,616 INFO [train.py:901] (2/4) Epoch 4, batch 4050, loss[loss=0.3013, simple_loss=0.3574, pruned_loss=0.1226, over 8126.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3642, pruned_loss=0.1275, over 1609295.22 frames. ], batch size: 22, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:26:21,158 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28305.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:26:34,456 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 3.482e+02 4.201e+02 5.400e+02 1.078e+03, threshold=8.403e+02, percent-clipped=4.0 +2023-02-05 23:26:52,366 INFO [train.py:901] (2/4) Epoch 4, batch 4100, loss[loss=0.2923, simple_loss=0.3681, pruned_loss=0.1083, over 8524.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3645, pruned_loss=0.1271, over 1612662.18 frames. ], batch size: 28, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:27:27,345 INFO [train.py:901] (2/4) Epoch 4, batch 4150, loss[loss=0.4092, simple_loss=0.4221, pruned_loss=0.1982, over 7098.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3644, pruned_loss=0.1271, over 1611063.33 frames. ], batch size: 72, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:27:43,618 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.222e+02 3.372e+02 4.170e+02 5.520e+02 1.384e+03, threshold=8.341e+02, percent-clipped=6.0 +2023-02-05 23:28:00,677 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28447.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:28:02,477 INFO [train.py:901] (2/4) Epoch 4, batch 4200, loss[loss=0.2592, simple_loss=0.3278, pruned_loss=0.09533, over 8018.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3647, pruned_loss=0.127, over 1610069.15 frames. ], batch size: 22, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:28:07,669 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 23:28:12,245 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-05 23:28:17,491 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28472.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:28:29,056 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 23:28:36,443 INFO [train.py:901] (2/4) Epoch 4, batch 4250, loss[loss=0.3343, simple_loss=0.3881, pruned_loss=0.1402, over 8464.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3653, pruned_loss=0.1276, over 1611254.64 frames. ], batch size: 27, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:28:39,210 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28504.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:28:43,325 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28510.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:28:51,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 3.170e+02 4.105e+02 5.662e+02 1.430e+03, threshold=8.210e+02, percent-clipped=9.0 +2023-02-05 23:29:10,385 INFO [train.py:901] (2/4) Epoch 4, batch 4300, loss[loss=0.3003, simple_loss=0.3591, pruned_loss=0.1207, over 8738.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3648, pruned_loss=0.1272, over 1610353.14 frames. ], batch size: 39, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:29:38,464 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28591.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:29:45,232 INFO [train.py:901] (2/4) Epoch 4, batch 4350, loss[loss=0.3266, simple_loss=0.3563, pruned_loss=0.1484, over 7794.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.364, pruned_loss=0.1267, over 1610409.11 frames. ], batch size: 19, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:29:57,580 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28617.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:29:58,761 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 23:30:01,427 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 3.285e+02 3.917e+02 4.771e+02 1.131e+03, threshold=7.833e+02, percent-clipped=1.0 +2023-02-05 23:30:19,077 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28649.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:30:19,682 INFO [train.py:901] (2/4) Epoch 4, batch 4400, loss[loss=0.2615, simple_loss=0.3233, pruned_loss=0.09981, over 7785.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3619, pruned_loss=0.125, over 1612134.87 frames. ], batch size: 19, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:30:41,088 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 23:30:54,262 INFO [train.py:901] (2/4) Epoch 4, batch 4450, loss[loss=0.2608, simple_loss=0.3229, pruned_loss=0.09938, over 7544.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3623, pruned_loss=0.1252, over 1611008.58 frames. ], batch size: 18, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:30:58,485 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28706.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:31:10,734 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.181e+02 3.229e+02 4.056e+02 4.786e+02 8.259e+02, threshold=8.113e+02, percent-clipped=1.0 +2023-02-05 23:31:17,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28732.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:31:30,205 INFO [train.py:901] (2/4) Epoch 4, batch 4500, loss[loss=0.3, simple_loss=0.3671, pruned_loss=0.1164, over 8335.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3626, pruned_loss=0.125, over 1613799.80 frames. ], batch size: 25, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:31:36,214 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 23:31:39,879 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28764.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:32:05,450 INFO [train.py:901] (2/4) Epoch 4, batch 4550, loss[loss=0.2766, simple_loss=0.3477, pruned_loss=0.1028, over 7802.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3624, pruned_loss=0.125, over 1614188.38 frames. ], batch size: 19, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:32:21,335 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 3.133e+02 4.046e+02 5.517e+02 1.256e+03, threshold=8.093e+02, percent-clipped=3.0 +2023-02-05 23:32:39,633 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:32:40,898 INFO [train.py:901] (2/4) Epoch 4, batch 4600, loss[loss=0.2928, simple_loss=0.3634, pruned_loss=0.1111, over 8491.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3608, pruned_loss=0.1244, over 1608491.16 frames. ], batch size: 29, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:32:43,601 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28854.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:33:00,447 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28879.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:33:10,535 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-05 23:33:14,777 INFO [train.py:901] (2/4) Epoch 4, batch 4650, loss[loss=0.3085, simple_loss=0.3625, pruned_loss=0.1273, over 8589.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3607, pruned_loss=0.1246, over 1609407.66 frames. ], batch size: 34, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:33:30,675 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 3.425e+02 4.570e+02 5.631e+02 1.457e+03, threshold=9.141e+02, percent-clipped=7.0 +2023-02-05 23:33:49,337 INFO [train.py:901] (2/4) Epoch 4, batch 4700, loss[loss=0.3279, simple_loss=0.3761, pruned_loss=0.1399, over 7050.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3625, pruned_loss=0.1255, over 1615442.82 frames. ], batch size: 72, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:33:58,491 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28962.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:33:59,159 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28963.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:03,899 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28969.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:15,889 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28987.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:16,552 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28988.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:34:24,389 INFO [train.py:901] (2/4) Epoch 4, batch 4750, loss[loss=0.3222, simple_loss=0.378, pruned_loss=0.1332, over 8617.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3614, pruned_loss=0.1245, over 1609667.12 frames. ], batch size: 50, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:34:32,591 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4230, 1.8282, 1.5192, 1.4376, 1.4075, 1.5014, 1.6826, 1.7739], + device='cuda:2'), covar=tensor([0.0658, 0.1237, 0.1847, 0.1453, 0.0729, 0.1604, 0.0892, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0153, 0.0196, 0.0232, 0.0196, 0.0152, 0.0201, 0.0161, 0.0164], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:34:33,284 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29013.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:34:38,670 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29020.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:38,753 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29020.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:40,432 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.078e+02 3.145e+02 3.754e+02 5.040e+02 8.107e+02, threshold=7.508e+02, percent-clipped=0.0 +2023-02-05 23:34:40,461 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 23:34:42,486 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 23:34:56,216 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29045.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:58,838 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3640, 1.5128, 1.6113, 1.3022, 0.7753, 1.6621, 0.1155, 1.0166], + device='cuda:2'), covar=tensor([0.3140, 0.2466, 0.1066, 0.2384, 0.6242, 0.0911, 0.5484, 0.2401], + device='cuda:2'), in_proj_covar=tensor([0.0126, 0.0110, 0.0079, 0.0157, 0.0198, 0.0082, 0.0140, 0.0118], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-05 23:34:59,298 INFO [train.py:901] (2/4) Epoch 4, batch 4800, loss[loss=0.3535, simple_loss=0.3929, pruned_loss=0.157, over 8578.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3611, pruned_loss=0.1247, over 1609564.05 frames. ], batch size: 34, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:35:33,998 INFO [train.py:901] (2/4) Epoch 4, batch 4850, loss[loss=0.3329, simple_loss=0.3602, pruned_loss=0.1528, over 7968.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3605, pruned_loss=0.1244, over 1608414.51 frames. ], batch size: 21, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:35:34,009 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 23:35:41,372 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-05 23:35:49,588 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.099e+02 3.374e+02 4.405e+02 6.016e+02 1.134e+03, threshold=8.810e+02, percent-clipped=7.0 +2023-02-05 23:36:08,337 INFO [train.py:901] (2/4) Epoch 4, batch 4900, loss[loss=0.3283, simple_loss=0.3719, pruned_loss=0.1424, over 8240.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3605, pruned_loss=0.1238, over 1611897.86 frames. ], batch size: 22, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:36:08,612 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7718, 2.3507, 3.9158, 2.9962, 2.9341, 2.2471, 1.6795, 1.7431], + device='cuda:2'), covar=tensor([0.1522, 0.1942, 0.0335, 0.0969, 0.1016, 0.0812, 0.0954, 0.1984], + device='cuda:2'), in_proj_covar=tensor([0.0685, 0.0614, 0.0523, 0.0580, 0.0695, 0.0572, 0.0562, 0.0573], + 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') +2023-02-05 23:36:41,947 INFO [train.py:901] (2/4) Epoch 4, batch 4950, loss[loss=0.2808, simple_loss=0.3577, pruned_loss=0.102, over 8245.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3613, pruned_loss=0.1244, over 1611663.61 frames. ], batch size: 24, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:36:56,272 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29219.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:36:58,772 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 3.208e+02 3.912e+02 5.596e+02 9.849e+02, threshold=7.824e+02, percent-clipped=2.0 +2023-02-05 23:36:58,865 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29223.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:00,343 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29225.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:12,962 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29244.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:16,652 INFO [train.py:901] (2/4) Epoch 4, batch 5000, loss[loss=0.3407, simple_loss=0.3836, pruned_loss=0.1489, over 8296.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3625, pruned_loss=0.1246, over 1616944.16 frames. ], batch size: 23, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:37:16,810 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9529, 1.5624, 4.0982, 1.7571, 3.6123, 3.4789, 3.6599, 3.6073], + device='cuda:2'), covar=tensor([0.0371, 0.3037, 0.0338, 0.2096, 0.0889, 0.0552, 0.0410, 0.0455], + device='cuda:2'), in_proj_covar=tensor([0.0288, 0.0460, 0.0356, 0.0373, 0.0439, 0.0370, 0.0357, 0.0398], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:37:16,879 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29250.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:18,900 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2285, 1.4524, 4.3951, 1.5365, 3.7624, 3.6403, 3.9007, 3.8445], + device='cuda:2'), covar=tensor([0.0383, 0.3385, 0.0335, 0.2430, 0.1203, 0.0620, 0.0412, 0.0486], + device='cuda:2'), in_proj_covar=tensor([0.0288, 0.0460, 0.0356, 0.0373, 0.0439, 0.0370, 0.0357, 0.0398], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:37:19,516 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29254.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:35,452 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5870, 5.6475, 5.0248, 2.1991, 4.9053, 5.3100, 5.1684, 4.7035], + device='cuda:2'), covar=tensor([0.0651, 0.0416, 0.0816, 0.4169, 0.0572, 0.0416, 0.1000, 0.0572], + device='cuda:2'), in_proj_covar=tensor([0.0369, 0.0268, 0.0293, 0.0381, 0.0291, 0.0238, 0.0288, 0.0223], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 23:37:51,640 INFO [train.py:901] (2/4) Epoch 4, batch 5050, loss[loss=0.2958, simple_loss=0.3598, pruned_loss=0.1159, over 7919.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3625, pruned_loss=0.124, over 1614817.70 frames. ], batch size: 20, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:38:07,700 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 3.436e+02 4.072e+02 5.001e+02 1.022e+03, threshold=8.144e+02, percent-clipped=3.0 +2023-02-05 23:38:14,946 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 23:38:18,459 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29338.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:38:26,649 INFO [train.py:901] (2/4) Epoch 4, batch 5100, loss[loss=0.3068, simple_loss=0.3417, pruned_loss=0.136, over 7720.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3617, pruned_loss=0.1244, over 1613470.64 frames. ], batch size: 18, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:38:36,225 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29364.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:38:44,636 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8328, 3.8379, 3.3675, 1.6473, 3.3452, 3.0285, 3.4847, 2.9574], + device='cuda:2'), covar=tensor([0.0897, 0.0695, 0.1008, 0.4907, 0.0905, 0.0942, 0.1330, 0.0887], + device='cuda:2'), in_proj_covar=tensor([0.0372, 0.0270, 0.0296, 0.0390, 0.0298, 0.0243, 0.0293, 0.0223], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-05 23:38:57,977 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0384, 2.4049, 3.9247, 3.2335, 3.0781, 2.4129, 1.6163, 1.8254], + device='cuda:2'), covar=tensor([0.1280, 0.1818, 0.0404, 0.0738, 0.0860, 0.0720, 0.0893, 0.1743], + device='cuda:2'), in_proj_covar=tensor([0.0695, 0.0616, 0.0528, 0.0574, 0.0698, 0.0572, 0.0564, 0.0573], + 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') +2023-02-05 23:39:00,484 INFO [train.py:901] (2/4) Epoch 4, batch 5150, loss[loss=0.2792, simple_loss=0.3513, pruned_loss=0.1035, over 8464.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3614, pruned_loss=0.124, over 1615706.03 frames. ], batch size: 27, lr: 1.78e-02, grad_scale: 8.0 +2023-02-05 23:39:16,241 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.119e+02 3.145e+02 3.888e+02 4.871e+02 1.199e+03, threshold=7.777e+02, percent-clipped=1.0 +2023-02-05 23:39:35,373 INFO [train.py:901] (2/4) Epoch 4, batch 5200, loss[loss=0.312, simple_loss=0.3771, pruned_loss=0.1234, over 8478.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3637, pruned_loss=0.1263, over 1613269.63 frames. ], batch size: 29, lr: 1.78e-02, grad_scale: 8.0 +2023-02-05 23:39:54,955 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29479.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:40:09,487 INFO [train.py:901] (2/4) Epoch 4, batch 5250, loss[loss=0.3392, simple_loss=0.3926, pruned_loss=0.1428, over 7802.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3633, pruned_loss=0.126, over 1614464.33 frames. ], batch size: 20, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:40:12,203 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 23:40:25,985 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.346e+02 3.507e+02 4.371e+02 5.555e+02 1.318e+03, threshold=8.742e+02, percent-clipped=11.0 +2023-02-05 23:40:39,165 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2023-02-05 23:40:43,419 INFO [train.py:901] (2/4) Epoch 4, batch 5300, loss[loss=0.3275, simple_loss=0.3774, pruned_loss=0.1388, over 8631.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3631, pruned_loss=0.1258, over 1610913.14 frames. ], batch size: 34, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:40:51,103 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2480, 1.6415, 3.3769, 0.9675, 2.5304, 1.5513, 1.2993, 2.0186], + device='cuda:2'), covar=tensor([0.1852, 0.1946, 0.0515, 0.3508, 0.1125, 0.2521, 0.1851, 0.1971], + device='cuda:2'), in_proj_covar=tensor([0.0455, 0.0429, 0.0505, 0.0517, 0.0560, 0.0500, 0.0437, 0.0576], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:40:57,670 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29569.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:02,754 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-05 23:41:14,744 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29594.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:17,286 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29598.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:18,612 INFO [train.py:901] (2/4) Epoch 4, batch 5350, loss[loss=0.3094, simple_loss=0.364, pruned_loss=0.1274, over 7935.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3644, pruned_loss=0.126, over 1618452.86 frames. ], batch size: 20, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:41:32,363 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29619.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:35,509 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 3.127e+02 4.006e+02 4.952e+02 2.682e+03, threshold=8.012e+02, percent-clipped=7.0 +2023-02-05 23:41:50,011 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-05 23:41:53,621 INFO [train.py:901] (2/4) Epoch 4, batch 5400, loss[loss=0.3108, simple_loss=0.3638, pruned_loss=0.1289, over 8595.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.366, pruned_loss=0.1275, over 1616693.92 frames. ], batch size: 49, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:42:14,527 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29680.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:42:28,608 INFO [train.py:901] (2/4) Epoch 4, batch 5450, loss[loss=0.2339, simple_loss=0.3036, pruned_loss=0.08204, over 7549.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3638, pruned_loss=0.1256, over 1615367.36 frames. ], batch size: 18, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:42:37,304 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29713.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:42:44,948 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.180e+02 3.089e+02 4.007e+02 5.016e+02 9.074e+02, threshold=8.014e+02, percent-clipped=4.0 +2023-02-05 23:42:52,701 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29735.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:42:57,993 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 23:43:02,828 INFO [train.py:901] (2/4) Epoch 4, batch 5500, loss[loss=0.2953, simple_loss=0.3646, pruned_loss=0.113, over 8184.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3632, pruned_loss=0.1254, over 1610364.91 frames. ], batch size: 23, lr: 1.77e-02, grad_scale: 4.0 +2023-02-05 23:43:10,333 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29760.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:43:12,460 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-05 23:43:20,938 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6340, 1.4395, 5.6025, 2.2588, 4.8577, 4.7332, 5.1886, 5.0539], + device='cuda:2'), covar=tensor([0.0346, 0.3786, 0.0221, 0.2107, 0.0931, 0.0445, 0.0342, 0.0418], + device='cuda:2'), in_proj_covar=tensor([0.0284, 0.0452, 0.0348, 0.0374, 0.0434, 0.0371, 0.0355, 0.0391], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-05 23:43:24,177 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-05 23:43:38,373 INFO [train.py:901] (2/4) Epoch 4, batch 5550, loss[loss=0.2514, simple_loss=0.3282, pruned_loss=0.08728, over 7817.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3634, pruned_loss=0.1256, over 1614423.66 frames. ], batch size: 19, lr: 1.77e-02, grad_scale: 4.0 +2023-02-05 23:43:38,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4995, 1.3314, 1.4073, 1.1904, 1.0276, 1.3306, 1.2410, 1.3341], + device='cuda:2'), covar=tensor([0.0653, 0.1186, 0.2005, 0.1437, 0.0668, 0.1547, 0.0784, 0.0551], + device='cuda:2'), in_proj_covar=tensor([0.0152, 0.0196, 0.0232, 0.0196, 0.0152, 0.0198, 0.0160, 0.0163], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:43:51,783 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29820.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:43:53,127 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6566, 1.9932, 2.3179, 0.9692, 2.2584, 1.5170, 0.7439, 1.8161], + device='cuda:2'), covar=tensor([0.0182, 0.0090, 0.0068, 0.0166, 0.0113, 0.0285, 0.0248, 0.0092], + device='cuda:2'), in_proj_covar=tensor([0.0281, 0.0198, 0.0157, 0.0238, 0.0191, 0.0326, 0.0261, 0.0225], + device='cuda:2'), out_proj_covar=tensor([1.1202e-04, 7.8135e-05, 5.9470e-05, 9.1674e-05, 7.6732e-05, 1.3897e-04, + 1.0497e-04, 8.8034e-05], device='cuda:2') +2023-02-05 23:43:54,237 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 3.211e+02 3.931e+02 4.808e+02 9.688e+02, threshold=7.861e+02, percent-clipped=2.0 +2023-02-05 23:44:12,158 INFO [train.py:901] (2/4) Epoch 4, batch 5600, loss[loss=0.2971, simple_loss=0.3564, pruned_loss=0.1189, over 8294.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.363, pruned_loss=0.1253, over 1614532.78 frames. ], batch size: 23, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:44:31,748 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8157, 2.4705, 4.6485, 1.2985, 3.1876, 2.0824, 1.9245, 2.5633], + device='cuda:2'), covar=tensor([0.1313, 0.1619, 0.0584, 0.2729, 0.1172, 0.2188, 0.1224, 0.2083], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0422, 0.0496, 0.0510, 0.0551, 0.0493, 0.0433, 0.0554], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:44:39,041 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6438, 1.7820, 1.5031, 1.1490, 1.5017, 1.5566, 1.7317, 1.6765], + device='cuda:2'), covar=tensor([0.0605, 0.1250, 0.1934, 0.1657, 0.0756, 0.1675, 0.0936, 0.0666], + device='cuda:2'), in_proj_covar=tensor([0.0152, 0.0195, 0.0230, 0.0195, 0.0152, 0.0198, 0.0160, 0.0165], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:44:46,060 INFO [train.py:901] (2/4) Epoch 4, batch 5650, loss[loss=0.3656, simple_loss=0.4072, pruned_loss=0.1621, over 8179.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3635, pruned_loss=0.126, over 1617947.88 frames. ], batch size: 48, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:44:55,377 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29913.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:45:03,295 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 3.236e+02 4.025e+02 5.119e+02 8.732e+02, threshold=8.050e+02, percent-clipped=2.0 +2023-02-05 23:45:03,336 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 23:45:10,182 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4099, 1.1998, 4.5044, 1.7076, 3.8605, 3.7195, 4.0652, 3.9920], + device='cuda:2'), covar=tensor([0.0433, 0.3706, 0.0319, 0.2299, 0.1038, 0.0523, 0.0416, 0.0457], + device='cuda:2'), in_proj_covar=tensor([0.0294, 0.0464, 0.0365, 0.0383, 0.0451, 0.0378, 0.0366, 0.0401], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:45:10,970 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1118, 1.9891, 3.3268, 0.8307, 2.1464, 1.3406, 1.4736, 1.9360], + device='cuda:2'), covar=tensor([0.2306, 0.1874, 0.0791, 0.4316, 0.1765, 0.3040, 0.1869, 0.2530], + device='cuda:2'), in_proj_covar=tensor([0.0452, 0.0427, 0.0499, 0.0516, 0.0559, 0.0492, 0.0435, 0.0561], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:45:18,368 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2537, 1.5725, 1.2803, 1.8954, 0.8059, 1.1213, 1.1908, 1.5228], + device='cuda:2'), covar=tensor([0.1310, 0.1020, 0.1485, 0.0699, 0.1627, 0.2105, 0.1349, 0.1065], + device='cuda:2'), in_proj_covar=tensor([0.0280, 0.0287, 0.0298, 0.0222, 0.0264, 0.0293, 0.0306, 0.0275], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:45:20,771 INFO [train.py:901] (2/4) Epoch 4, batch 5700, loss[loss=0.2726, simple_loss=0.3379, pruned_loss=0.1036, over 8019.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3626, pruned_loss=0.1256, over 1615327.13 frames. ], batch size: 22, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:45:34,467 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29969.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:45:52,113 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29994.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:45:55,950 INFO [train.py:901] (2/4) Epoch 4, batch 5750, loss[loss=0.2759, simple_loss=0.3456, pruned_loss=0.1031, over 8511.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3612, pruned_loss=0.1241, over 1614984.26 frames. ], batch size: 26, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:46:07,130 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 23:46:13,252 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.352e+02 3.278e+02 4.024e+02 4.787e+02 1.009e+03, threshold=8.047e+02, percent-clipped=4.0 +2023-02-05 23:46:13,360 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30024.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:46:17,493 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30028.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:46:25,007 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3605, 1.6697, 2.8401, 1.0707, 1.9916, 1.6175, 1.4281, 1.7189], + device='cuda:2'), covar=tensor([0.1579, 0.1642, 0.0530, 0.3093, 0.1270, 0.2212, 0.1381, 0.1763], + device='cuda:2'), in_proj_covar=tensor([0.0451, 0.0423, 0.0502, 0.0512, 0.0557, 0.0492, 0.0436, 0.0561], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:46:32,285 INFO [train.py:901] (2/4) Epoch 4, batch 5800, loss[loss=0.2686, simple_loss=0.3249, pruned_loss=0.1062, over 7428.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3589, pruned_loss=0.1228, over 1609990.58 frames. ], batch size: 17, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:47:02,997 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 23:47:06,526 INFO [train.py:901] (2/4) Epoch 4, batch 5850, loss[loss=0.2782, simple_loss=0.3369, pruned_loss=0.1097, over 8093.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3594, pruned_loss=0.1234, over 1610899.29 frames. ], batch size: 21, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:47:13,369 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7653, 1.6922, 5.7152, 2.0663, 5.0148, 4.8142, 5.3330, 5.2688], + device='cuda:2'), covar=tensor([0.0341, 0.3216, 0.0230, 0.2171, 0.0820, 0.0437, 0.0318, 0.0350], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0457, 0.0364, 0.0374, 0.0450, 0.0379, 0.0361, 0.0404], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-05 23:47:23,087 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 3.427e+02 4.657e+02 5.932e+02 9.223e+02, threshold=9.314e+02, percent-clipped=4.0 +2023-02-05 23:47:33,285 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30139.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:47:33,521 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-05 23:47:35,875 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30143.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:47:41,675 INFO [train.py:901] (2/4) Epoch 4, batch 5900, loss[loss=0.2877, simple_loss=0.3576, pruned_loss=0.1089, over 8573.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3601, pruned_loss=0.1238, over 1616103.71 frames. ], batch size: 31, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:47:51,360 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30164.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:48:15,960 INFO [train.py:901] (2/4) Epoch 4, batch 5950, loss[loss=0.3506, simple_loss=0.4005, pruned_loss=0.1503, over 8652.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3607, pruned_loss=0.124, over 1618165.45 frames. ], batch size: 34, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:48:32,430 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 3.143e+02 3.968e+02 4.977e+02 1.070e+03, threshold=7.937e+02, percent-clipped=1.0 +2023-02-05 23:48:35,342 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30227.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:48:50,175 INFO [train.py:901] (2/4) Epoch 4, batch 6000, loss[loss=0.2519, simple_loss=0.3053, pruned_loss=0.09923, over 7778.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3614, pruned_loss=0.1242, over 1618615.54 frames. ], batch size: 19, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:48:50,175 INFO [train.py:926] (2/4) Computing validation loss +2023-02-05 23:48:58,490 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4411, 1.7435, 2.7208, 1.1491, 2.0493, 1.7198, 1.5251, 1.8164], + device='cuda:2'), covar=tensor([0.1295, 0.1718, 0.0507, 0.2844, 0.1164, 0.2027, 0.1375, 0.1797], + device='cuda:2'), in_proj_covar=tensor([0.0460, 0.0435, 0.0512, 0.0523, 0.0563, 0.0499, 0.0447, 0.0574], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-05 23:49:02,857 INFO [train.py:935] (2/4) Epoch 4, validation: loss=0.2338, simple_loss=0.3275, pruned_loss=0.07005, over 944034.00 frames. +2023-02-05 23:49:02,858 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6599MB +2023-02-05 23:49:10,944 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.12 vs. limit=5.0 +2023-02-05 23:49:22,555 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30279.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:49:25,897 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30284.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:49:27,951 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6633, 2.0908, 1.6715, 2.1895, 1.4802, 1.4488, 1.7094, 2.0378], + device='cuda:2'), covar=tensor([0.0962, 0.0806, 0.1289, 0.0548, 0.1330, 0.1580, 0.1044, 0.0787], + device='cuda:2'), in_proj_covar=tensor([0.0274, 0.0280, 0.0295, 0.0223, 0.0262, 0.0288, 0.0299, 0.0274], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:49:37,708 INFO [train.py:901] (2/4) Epoch 4, batch 6050, loss[loss=0.274, simple_loss=0.3314, pruned_loss=0.1083, over 7553.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3622, pruned_loss=0.1252, over 1616590.61 frames. ], batch size: 18, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:49:44,056 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30309.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:49:53,935 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 3.338e+02 3.992e+02 4.649e+02 1.183e+03, threshold=7.984e+02, percent-clipped=3.0 +2023-02-05 23:50:12,471 INFO [train.py:901] (2/4) Epoch 4, batch 6100, loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1239, over 8140.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.362, pruned_loss=0.1245, over 1616361.33 frames. ], batch size: 22, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:50:14,756 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4075, 2.7541, 1.6351, 1.9166, 2.0006, 1.4289, 1.6751, 1.9637], + device='cuda:2'), covar=tensor([0.1319, 0.0302, 0.0873, 0.0628, 0.0656, 0.1172, 0.1052, 0.0738], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0234, 0.0306, 0.0302, 0.0330, 0.0311, 0.0334, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 23:50:24,482 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4930, 2.7253, 1.5717, 2.0689, 2.0148, 1.2929, 1.8000, 2.0560], + device='cuda:2'), covar=tensor([0.1301, 0.0408, 0.0965, 0.0575, 0.0639, 0.1269, 0.0995, 0.0849], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0235, 0.0307, 0.0304, 0.0331, 0.0313, 0.0336, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 23:50:32,436 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30378.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:50:39,277 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 23:50:44,131 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30395.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:50:47,296 INFO [train.py:901] (2/4) Epoch 4, batch 6150, loss[loss=0.3077, simple_loss=0.3571, pruned_loss=0.1292, over 7805.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3619, pruned_loss=0.1244, over 1615562.29 frames. ], batch size: 19, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:51:02,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30420.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:51:05,068 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 3.511e+02 4.267e+02 5.249e+02 1.089e+03, threshold=8.535e+02, percent-clipped=6.0 +2023-02-05 23:51:23,134 INFO [train.py:901] (2/4) Epoch 4, batch 6200, loss[loss=0.3335, simple_loss=0.3859, pruned_loss=0.1406, over 8247.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.363, pruned_loss=0.1251, over 1615184.87 frames. ], batch size: 22, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:51:47,746 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7484, 2.4657, 3.1109, 1.0401, 3.2020, 1.8928, 1.2973, 1.6534], + device='cuda:2'), covar=tensor([0.0312, 0.0093, 0.0085, 0.0268, 0.0112, 0.0284, 0.0360, 0.0179], + device='cuda:2'), in_proj_covar=tensor([0.0285, 0.0201, 0.0166, 0.0248, 0.0198, 0.0332, 0.0266, 0.0230], + device='cuda:2'), out_proj_covar=tensor([1.1253e-04, 7.8713e-05, 6.2006e-05, 9.4146e-05, 7.8766e-05, 1.3988e-04, + 1.0582e-04, 8.8970e-05], device='cuda:2') +2023-02-05 23:51:48,270 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30487.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:51:57,460 INFO [train.py:901] (2/4) Epoch 4, batch 6250, loss[loss=0.2824, simple_loss=0.3491, pruned_loss=0.1078, over 8495.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3626, pruned_loss=0.125, over 1612983.33 frames. ], batch size: 39, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:52:14,459 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 3.291e+02 3.933e+02 5.014e+02 1.132e+03, threshold=7.866e+02, percent-clipped=4.0 +2023-02-05 23:52:17,389 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2500, 2.6540, 1.5234, 1.8995, 1.7426, 1.1709, 1.6273, 1.7406], + device='cuda:2'), covar=tensor([0.1550, 0.0359, 0.1103, 0.0736, 0.0914, 0.1573, 0.1277, 0.0966], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0242, 0.0311, 0.0311, 0.0335, 0.0317, 0.0340, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-05 23:52:22,911 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30535.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:52:32,703 INFO [train.py:901] (2/4) Epoch 4, batch 6300, loss[loss=0.281, simple_loss=0.333, pruned_loss=0.1145, over 7437.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3615, pruned_loss=0.1245, over 1611772.09 frames. ], batch size: 17, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:52:39,499 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30560.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:52:47,315 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30571.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:53:06,768 INFO [train.py:901] (2/4) Epoch 4, batch 6350, loss[loss=0.2664, simple_loss=0.3423, pruned_loss=0.09524, over 8470.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3614, pruned_loss=0.1239, over 1610274.00 frames. ], batch size: 31, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:53:08,354 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30602.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:53:19,901 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8960, 2.0806, 1.6739, 2.6068, 1.2227, 1.3652, 1.7005, 2.2164], + device='cuda:2'), covar=tensor([0.0961, 0.1243, 0.1587, 0.0565, 0.1706, 0.2142, 0.1630, 0.0903], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0280, 0.0289, 0.0221, 0.0260, 0.0285, 0.0298, 0.0269], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:53:23,786 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 3.165e+02 3.849e+02 5.077e+02 1.430e+03, threshold=7.697e+02, percent-clipped=4.0 +2023-02-05 23:53:42,465 INFO [train.py:901] (2/4) Epoch 4, batch 6400, loss[loss=0.2488, simple_loss=0.3234, pruned_loss=0.08714, over 8027.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3615, pruned_loss=0.124, over 1614310.37 frames. ], batch size: 22, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:53:58,098 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30673.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:54:07,693 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30686.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:54:16,879 INFO [train.py:901] (2/4) Epoch 4, batch 6450, loss[loss=0.2828, simple_loss=0.3604, pruned_loss=0.1026, over 8362.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3609, pruned_loss=0.1231, over 1615365.57 frames. ], batch size: 24, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:54:31,540 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30722.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:54:32,830 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 3.022e+02 3.987e+02 5.645e+02 1.412e+03, threshold=7.975e+02, percent-clipped=10.0 +2023-02-05 23:54:50,958 INFO [train.py:901] (2/4) Epoch 4, batch 6500, loss[loss=0.2831, simple_loss=0.3649, pruned_loss=0.1007, over 8245.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3605, pruned_loss=0.1226, over 1616285.54 frames. ], batch size: 24, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:55:01,899 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8993, 2.2142, 3.0702, 1.2162, 3.0201, 1.9484, 1.4081, 1.9042], + device='cuda:2'), covar=tensor([0.0262, 0.0111, 0.0082, 0.0206, 0.0123, 0.0277, 0.0326, 0.0133], + device='cuda:2'), in_proj_covar=tensor([0.0281, 0.0200, 0.0162, 0.0239, 0.0195, 0.0332, 0.0261, 0.0224], + device='cuda:2'), out_proj_covar=tensor([1.1033e-04, 7.7747e-05, 6.0131e-05, 9.0085e-05, 7.7270e-05, 1.3957e-04, + 1.0364e-04, 8.6134e-05], device='cuda:2') +2023-02-05 23:55:26,186 INFO [train.py:901] (2/4) Epoch 4, batch 6550, loss[loss=0.278, simple_loss=0.3504, pruned_loss=0.1028, over 8470.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3584, pruned_loss=0.1214, over 1614947.30 frames. ], batch size: 25, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:55:42,638 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.464e+02 3.539e+02 4.251e+02 5.114e+02 1.135e+03, threshold=8.501e+02, percent-clipped=1.0 +2023-02-05 23:55:50,034 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 23:55:51,484 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30837.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:55:54,909 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2023-02-05 23:56:00,637 INFO [train.py:901] (2/4) Epoch 4, batch 6600, loss[loss=0.3099, simple_loss=0.3759, pruned_loss=0.122, over 8494.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3597, pruned_loss=0.1224, over 1616200.90 frames. ], batch size: 28, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:56:06,245 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30858.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:08,701 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 23:56:24,233 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30883.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:35,345 INFO [train.py:901] (2/4) Epoch 4, batch 6650, loss[loss=0.2984, simple_loss=0.3463, pruned_loss=0.1253, over 7784.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3579, pruned_loss=0.1214, over 1613930.04 frames. ], batch size: 19, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:56:50,088 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30921.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:51,871 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 3.362e+02 4.352e+02 5.461e+02 1.446e+03, threshold=8.703e+02, percent-clipped=3.0 +2023-02-05 23:57:04,064 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30942.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:04,651 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30943.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:09,163 INFO [train.py:901] (2/4) Epoch 4, batch 6700, loss[loss=0.2856, simple_loss=0.3476, pruned_loss=0.1118, over 8477.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3584, pruned_loss=0.1222, over 1614571.80 frames. ], batch size: 25, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:57:10,676 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30952.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:13,688 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 23:57:21,644 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30967.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:45,059 INFO [train.py:901] (2/4) Epoch 4, batch 6750, loss[loss=0.3135, simple_loss=0.3695, pruned_loss=0.1288, over 8329.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3595, pruned_loss=0.1232, over 1614934.39 frames. ], batch size: 25, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:57:56,392 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31017.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:58:00,825 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.225e+02 3.317e+02 4.136e+02 5.252e+02 1.678e+03, threshold=8.272e+02, percent-clipped=4.0 +2023-02-05 23:58:04,873 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31029.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:58:18,933 INFO [train.py:901] (2/4) Epoch 4, batch 6800, loss[loss=0.2559, simple_loss=0.3384, pruned_loss=0.08664, over 8247.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3598, pruned_loss=0.1243, over 1613581.81 frames. ], batch size: 24, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:58:19,597 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 23:58:19,773 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1553, 1.6840, 1.4609, 1.1262, 1.3996, 1.4242, 1.5187, 1.7485], + device='cuda:2'), covar=tensor([0.0616, 0.1219, 0.1944, 0.1615, 0.0652, 0.1627, 0.0849, 0.0562], + device='cuda:2'), in_proj_covar=tensor([0.0151, 0.0193, 0.0232, 0.0195, 0.0148, 0.0198, 0.0160, 0.0159], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:58:48,706 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31093.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:58:53,201 INFO [train.py:901] (2/4) Epoch 4, batch 6850, loss[loss=0.2514, simple_loss=0.312, pruned_loss=0.09539, over 7811.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3593, pruned_loss=0.123, over 1614993.38 frames. ], batch size: 19, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:59:06,897 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31118.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:59:10,006 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 23:59:10,567 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 3.178e+02 3.797e+02 5.313e+02 1.260e+03, threshold=7.594e+02, percent-clipped=4.0 +2023-02-05 23:59:13,446 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6946, 3.0508, 2.4501, 4.0332, 1.7540, 1.7601, 2.1858, 3.2101], + device='cuda:2'), covar=tensor([0.0945, 0.1397, 0.1377, 0.0311, 0.1945, 0.2367, 0.2063, 0.1294], + device='cuda:2'), in_proj_covar=tensor([0.0275, 0.0284, 0.0299, 0.0223, 0.0272, 0.0295, 0.0308, 0.0276], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-05 23:59:16,201 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31132.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:59:28,756 INFO [train.py:901] (2/4) Epoch 4, batch 6900, loss[loss=0.2991, simple_loss=0.3675, pruned_loss=0.1153, over 8625.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3601, pruned_loss=0.1237, over 1610824.46 frames. ], batch size: 34, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:00:03,327 INFO [train.py:901] (2/4) Epoch 4, batch 6950, loss[loss=0.2515, simple_loss=0.3111, pruned_loss=0.09592, over 7801.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3606, pruned_loss=0.1235, over 1609915.57 frames. ], batch size: 19, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:00:18,141 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 00:00:20,075 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 3.425e+02 4.122e+02 5.302e+02 9.579e+02, threshold=8.244e+02, percent-clipped=6.0 +2023-02-06 00:00:23,107 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 +2023-02-06 00:00:38,134 INFO [train.py:901] (2/4) Epoch 4, batch 7000, loss[loss=0.2872, simple_loss=0.359, pruned_loss=0.1077, over 8292.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3613, pruned_loss=0.1242, over 1613335.07 frames. ], batch size: 23, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:00:48,782 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31265.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:00:53,584 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0394, 2.2856, 3.8466, 3.0713, 3.1938, 2.2299, 1.6267, 1.5466], + device='cuda:2'), covar=tensor([0.1384, 0.1987, 0.0394, 0.0906, 0.0891, 0.0816, 0.0953, 0.2035], + device='cuda:2'), in_proj_covar=tensor([0.0709, 0.0625, 0.0529, 0.0609, 0.0727, 0.0595, 0.0574, 0.0592], + 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') +2023-02-06 00:00:54,246 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2265, 1.7285, 1.7168, 0.5871, 1.7479, 1.1930, 0.2781, 1.5480], + device='cuda:2'), covar=tensor([0.0186, 0.0076, 0.0080, 0.0162, 0.0104, 0.0320, 0.0276, 0.0080], + device='cuda:2'), in_proj_covar=tensor([0.0280, 0.0200, 0.0165, 0.0244, 0.0194, 0.0326, 0.0267, 0.0231], + device='cuda:2'), out_proj_covar=tensor([1.0915e-04, 7.6842e-05, 6.1419e-05, 9.1168e-05, 7.5797e-05, 1.3587e-04, + 1.0511e-04, 8.8277e-05], device='cuda:2') +2023-02-06 00:00:56,885 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2296, 2.2509, 1.6349, 1.9591, 1.8569, 1.3120, 1.6796, 1.9855], + device='cuda:2'), covar=tensor([0.1040, 0.0323, 0.0817, 0.0408, 0.0463, 0.1037, 0.0708, 0.0536], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0241, 0.0309, 0.0305, 0.0323, 0.0317, 0.0338, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:01:03,281 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:01:09,175 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31296.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:01:11,662 INFO [train.py:901] (2/4) Epoch 4, batch 7050, loss[loss=0.3136, simple_loss=0.379, pruned_loss=0.1241, over 8334.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3627, pruned_loss=0.1247, over 1615898.82 frames. ], batch size: 25, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:01:17,359 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7458, 2.2960, 1.8217, 2.9638, 1.2075, 1.4131, 1.8196, 2.3669], + device='cuda:2'), covar=tensor([0.1285, 0.1211, 0.1619, 0.0522, 0.1987, 0.2347, 0.1714, 0.1200], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0280, 0.0294, 0.0217, 0.0264, 0.0294, 0.0296, 0.0272], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 00:01:28,375 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 3.221e+02 3.873e+02 5.326e+02 1.178e+03, threshold=7.746e+02, percent-clipped=8.0 +2023-02-06 00:01:47,447 INFO [train.py:901] (2/4) Epoch 4, batch 7100, loss[loss=0.3715, simple_loss=0.4045, pruned_loss=0.1693, over 8504.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3632, pruned_loss=0.1252, over 1615592.65 frames. ], batch size: 26, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:01:57,592 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4522, 1.9826, 3.1433, 2.5376, 2.6549, 2.1344, 1.4671, 1.1232], + device='cuda:2'), covar=tensor([0.1314, 0.1571, 0.0323, 0.0786, 0.0724, 0.0809, 0.0949, 0.1660], + device='cuda:2'), in_proj_covar=tensor([0.0704, 0.0629, 0.0528, 0.0606, 0.0718, 0.0593, 0.0574, 0.0591], + 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') +2023-02-06 00:02:02,610 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31373.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:02:08,043 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31380.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:13,373 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:21,059 INFO [train.py:901] (2/4) Epoch 4, batch 7150, loss[loss=0.3061, simple_loss=0.3524, pruned_loss=0.1299, over 7232.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3612, pruned_loss=0.1233, over 1616077.70 frames. ], batch size: 16, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:02:22,722 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:28,579 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31411.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:28,656 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7612, 1.4935, 2.2813, 1.8460, 2.0260, 1.6269, 1.2380, 0.7068], + device='cuda:2'), covar=tensor([0.1625, 0.1561, 0.0419, 0.0817, 0.0659, 0.0892, 0.0951, 0.1511], + device='cuda:2'), in_proj_covar=tensor([0.0706, 0.0630, 0.0533, 0.0606, 0.0719, 0.0595, 0.0574, 0.0590], + 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') +2023-02-06 00:02:29,932 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:37,019 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 3.187e+02 3.955e+02 5.000e+02 8.847e+02, threshold=7.910e+02, percent-clipped=2.0 +2023-02-06 00:02:39,207 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:56,011 INFO [train.py:901] (2/4) Epoch 4, batch 7200, loss[loss=0.3491, simple_loss=0.3907, pruned_loss=0.1538, over 6740.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3608, pruned_loss=0.123, over 1614624.57 frames. ], batch size: 71, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:03:22,242 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31488.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:03:28,007 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 00:03:30,966 INFO [train.py:901] (2/4) Epoch 4, batch 7250, loss[loss=0.2539, simple_loss=0.3095, pruned_loss=0.09909, over 7695.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3595, pruned_loss=0.1225, over 1613500.94 frames. ], batch size: 18, lr: 1.73e-02, grad_scale: 16.0 +2023-02-06 00:03:32,538 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1496, 2.0937, 3.4400, 2.1969, 2.8058, 3.7921, 3.6122, 3.5810], + device='cuda:2'), covar=tensor([0.0956, 0.1089, 0.0667, 0.1418, 0.0820, 0.0241, 0.0331, 0.0402], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0263, 0.0216, 0.0263, 0.0220, 0.0197, 0.0208, 0.0268], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:03:35,400 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5840, 1.9679, 3.3556, 1.1012, 2.4454, 1.8278, 1.5759, 2.0138], + device='cuda:2'), covar=tensor([0.1429, 0.1774, 0.0636, 0.3015, 0.1223, 0.2284, 0.1308, 0.2063], + device='cuda:2'), in_proj_covar=tensor([0.0454, 0.0429, 0.0506, 0.0510, 0.0553, 0.0503, 0.0436, 0.0571], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:03:37,360 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31509.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:03:47,334 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.140e+02 3.150e+02 3.858e+02 4.938e+02 9.845e+02, threshold=7.715e+02, percent-clipped=4.0 +2023-02-06 00:03:57,286 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.73 vs. limit=5.0 +2023-02-06 00:04:05,046 INFO [train.py:901] (2/4) Epoch 4, batch 7300, loss[loss=0.3061, simple_loss=0.3668, pruned_loss=0.1227, over 8620.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3583, pruned_loss=0.1218, over 1615511.84 frames. ], batch size: 34, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:04:40,415 INFO [train.py:901] (2/4) Epoch 4, batch 7350, loss[loss=0.3139, simple_loss=0.373, pruned_loss=0.1274, over 8451.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3585, pruned_loss=0.1221, over 1614310.51 frames. ], batch size: 27, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:04:57,282 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.182e+02 2.774e+02 3.613e+02 4.483e+02 1.102e+03, threshold=7.227e+02, percent-clipped=2.0 +2023-02-06 00:04:59,997 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 00:05:05,415 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:14,568 INFO [train.py:901] (2/4) Epoch 4, batch 7400, loss[loss=0.2738, simple_loss=0.3339, pruned_loss=0.1068, over 8205.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3592, pruned_loss=0.1229, over 1611822.19 frames. ], batch size: 23, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:05:19,259 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 00:05:20,117 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:21,987 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:26,686 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:37,008 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:43,497 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31692.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:48,578 INFO [train.py:901] (2/4) Epoch 4, batch 7450, loss[loss=0.2738, simple_loss=0.345, pruned_loss=0.1013, over 8297.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3607, pruned_loss=0.1235, over 1615009.00 frames. ], batch size: 23, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:05:58,000 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 00:06:04,224 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:06:05,435 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 3.216e+02 3.933e+02 5.503e+02 1.387e+03, threshold=7.866e+02, percent-clipped=9.0 +2023-02-06 00:06:19,851 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31744.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:06:23,709 INFO [train.py:901] (2/4) Epoch 4, batch 7500, loss[loss=0.335, simple_loss=0.3847, pruned_loss=0.1427, over 8463.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.361, pruned_loss=0.123, over 1620453.63 frames. ], batch size: 25, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:06:36,764 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31769.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:06:37,959 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:06:40,259 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 00:06:57,583 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1761, 1.2116, 3.2005, 0.9674, 2.7357, 2.6696, 2.9595, 2.8627], + device='cuda:2'), covar=tensor([0.0510, 0.3276, 0.0546, 0.2496, 0.1321, 0.0897, 0.0586, 0.0698], + device='cuda:2'), in_proj_covar=tensor([0.0304, 0.0467, 0.0378, 0.0390, 0.0450, 0.0387, 0.0374, 0.0417], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 00:06:58,136 INFO [train.py:901] (2/4) Epoch 4, batch 7550, loss[loss=0.3784, simple_loss=0.4139, pruned_loss=0.1714, over 8502.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3611, pruned_loss=0.1239, over 1620856.69 frames. ], batch size: 26, lr: 1.72e-02, grad_scale: 8.0 +2023-02-06 00:07:02,985 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:07:16,184 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 2.842e+02 3.963e+02 5.244e+02 1.193e+03, threshold=7.926e+02, percent-clipped=8.0 +2023-02-06 00:07:27,220 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:07:34,061 INFO [train.py:901] (2/4) Epoch 4, batch 7600, loss[loss=0.379, simple_loss=0.4118, pruned_loss=0.1731, over 8577.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.361, pruned_loss=0.1237, over 1622129.15 frames. ], batch size: 49, lr: 1.72e-02, grad_scale: 8.0 +2023-02-06 00:07:36,151 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31853.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:07:58,613 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:05,425 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31896.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:07,983 INFO [train.py:901] (2/4) Epoch 4, batch 7650, loss[loss=0.2802, simple_loss=0.3543, pruned_loss=0.1031, over 8334.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3601, pruned_loss=0.1228, over 1622875.85 frames. ], batch size: 26, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:08:26,156 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 3.181e+02 3.860e+02 4.828e+02 9.649e+02, threshold=7.720e+02, percent-clipped=2.0 +2023-02-06 00:08:31,516 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31933.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:43,637 INFO [train.py:901] (2/4) Epoch 4, batch 7700, loss[loss=0.2669, simple_loss=0.3154, pruned_loss=0.1092, over 7513.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3589, pruned_loss=0.1216, over 1622949.69 frames. ], batch size: 18, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:08:55,986 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31968.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:56,022 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31968.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:09:06,798 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 00:09:17,470 INFO [train.py:901] (2/4) Epoch 4, batch 7750, loss[loss=0.3055, simple_loss=0.3723, pruned_loss=0.1194, over 8196.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3598, pruned_loss=0.1221, over 1623633.47 frames. ], batch size: 23, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:09:35,964 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.165e+02 3.163e+02 3.927e+02 5.355e+02 1.239e+03, threshold=7.853e+02, percent-clipped=4.0 +2023-02-06 00:09:53,603 INFO [train.py:901] (2/4) Epoch 4, batch 7800, loss[loss=0.3393, simple_loss=0.389, pruned_loss=0.1448, over 8474.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3604, pruned_loss=0.1223, over 1623642.65 frames. ], batch size: 25, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:10:05,168 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:10:27,395 INFO [train.py:901] (2/4) Epoch 4, batch 7850, loss[loss=0.3969, simple_loss=0.4365, pruned_loss=0.1787, over 8625.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3595, pruned_loss=0.1215, over 1622289.28 frames. ], batch size: 34, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:10:43,942 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.460e+02 3.521e+02 4.480e+02 6.179e+02 1.308e+03, threshold=8.960e+02, percent-clipped=13.0 +2023-02-06 00:10:55,615 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:00,787 INFO [train.py:901] (2/4) Epoch 4, batch 7900, loss[loss=0.3347, simple_loss=0.3877, pruned_loss=0.1409, over 8622.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3594, pruned_loss=0.1215, over 1623963.72 frames. ], batch size: 50, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:11:00,854 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:12,544 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:21,562 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32181.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:24,163 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32185.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:34,316 INFO [train.py:901] (2/4) Epoch 4, batch 7950, loss[loss=0.3433, simple_loss=0.3957, pruned_loss=0.1454, over 8331.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3607, pruned_loss=0.1224, over 1624542.83 frames. ], batch size: 26, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:11:42,539 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 00:11:51,236 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32224.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:11:51,629 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.248e+02 3.536e+02 4.226e+02 5.315e+02 1.259e+03, threshold=8.452e+02, percent-clipped=4.0 +2023-02-06 00:12:01,777 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32240.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:08,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32249.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:12:08,749 INFO [train.py:901] (2/4) Epoch 4, batch 8000, loss[loss=0.257, simple_loss=0.3179, pruned_loss=0.09802, over 7436.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.359, pruned_loss=0.1212, over 1622283.73 frames. ], batch size: 17, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:12:13,076 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 00:12:19,085 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32265.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:22,646 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 00:12:26,800 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32277.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:37,101 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3975, 2.0740, 1.4158, 1.7757, 1.6702, 1.2352, 1.4721, 1.8076], + device='cuda:2'), covar=tensor([0.0936, 0.0310, 0.0842, 0.0480, 0.0629, 0.1139, 0.0816, 0.0665], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0241, 0.0316, 0.0309, 0.0324, 0.0316, 0.0341, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:12:42,459 INFO [train.py:901] (2/4) Epoch 4, batch 8050, loss[loss=0.2434, simple_loss=0.314, pruned_loss=0.08645, over 7227.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3575, pruned_loss=0.1222, over 1599171.16 frames. ], batch size: 16, lr: 1.70e-02, grad_scale: 8.0 +2023-02-06 00:12:42,657 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32300.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:50,705 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:58,916 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.697e+02 3.496e+02 4.220e+02 5.135e+02 1.064e+03, threshold=8.441e+02, percent-clipped=2.0 +2023-02-06 00:13:15,884 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 00:13:19,660 INFO [train.py:901] (2/4) Epoch 5, batch 0, loss[loss=0.3705, simple_loss=0.4112, pruned_loss=0.1649, over 8715.00 frames. ], tot_loss[loss=0.3705, simple_loss=0.4112, pruned_loss=0.1649, over 8715.00 frames. ], batch size: 34, lr: 1.59e-02, grad_scale: 8.0 +2023-02-06 00:13:19,660 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 00:13:31,617 INFO [train.py:935] (2/4) Epoch 5, validation: loss=0.2309, simple_loss=0.3254, pruned_loss=0.06822, over 944034.00 frames. +2023-02-06 00:13:31,618 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6599MB +2023-02-06 00:13:46,441 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 00:13:46,611 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:06,998 INFO [train.py:901] (2/4) Epoch 5, batch 50, loss[loss=0.3227, simple_loss=0.3775, pruned_loss=0.1339, over 8564.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3653, pruned_loss=0.1242, over 369329.72 frames. ], batch size: 31, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:14:14,083 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:22,921 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 00:14:36,523 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 3.148e+02 3.721e+02 4.839e+02 1.477e+03, threshold=7.442e+02, percent-clipped=1.0 +2023-02-06 00:14:38,052 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:41,775 INFO [train.py:901] (2/4) Epoch 5, batch 100, loss[loss=0.3893, simple_loss=0.4212, pruned_loss=0.1787, over 8715.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3595, pruned_loss=0.1215, over 648626.42 frames. ], batch size: 34, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:14:44,566 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32437.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:45,035 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 00:14:50,603 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9268, 3.8867, 3.5286, 1.7914, 3.4906, 3.6469, 3.6970, 3.2017], + device='cuda:2'), covar=tensor([0.1310, 0.0836, 0.1084, 0.5199, 0.0880, 0.0906, 0.1454, 0.0875], + device='cuda:2'), in_proj_covar=tensor([0.0380, 0.0282, 0.0301, 0.0402, 0.0306, 0.0263, 0.0294, 0.0233], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 00:15:02,119 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:15:15,812 INFO [train.py:901] (2/4) Epoch 5, batch 150, loss[loss=0.2457, simple_loss=0.3277, pruned_loss=0.08187, over 8111.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3577, pruned_loss=0.1198, over 862437.31 frames. ], batch size: 23, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:15:43,038 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:15:45,458 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 3.007e+02 3.818e+02 4.644e+02 8.323e+02, threshold=7.636e+02, percent-clipped=1.0 +2023-02-06 00:15:50,791 INFO [train.py:901] (2/4) Epoch 5, batch 200, loss[loss=0.3033, simple_loss=0.3582, pruned_loss=0.1242, over 8254.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3558, pruned_loss=0.12, over 1029899.50 frames. ], batch size: 22, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:15:59,818 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:06,415 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:09,713 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0536, 1.1600, 1.0799, 0.2270, 1.1398, 0.9115, 0.1400, 1.0812], + device='cuda:2'), covar=tensor([0.0138, 0.0100, 0.0079, 0.0177, 0.0097, 0.0295, 0.0232, 0.0094], + device='cuda:2'), in_proj_covar=tensor([0.0291, 0.0204, 0.0164, 0.0251, 0.0200, 0.0328, 0.0268, 0.0232], + device='cuda:2'), out_proj_covar=tensor([1.1034e-04, 7.7110e-05, 6.0176e-05, 9.2228e-05, 7.6379e-05, 1.3345e-04, + 1.0301e-04, 8.6821e-05], device='cuda:2') +2023-02-06 00:16:23,703 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:24,858 INFO [train.py:901] (2/4) Epoch 5, batch 250, loss[loss=0.2894, simple_loss=0.3498, pruned_loss=0.1145, over 8356.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3556, pruned_loss=0.1204, over 1156375.59 frames. ], batch size: 24, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:16:36,163 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 00:16:45,160 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:46,291 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 00:16:54,483 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 3.241e+02 4.131e+02 4.869e+02 1.219e+03, threshold=8.263e+02, percent-clipped=9.0 +2023-02-06 00:17:00,673 INFO [train.py:901] (2/4) Epoch 5, batch 300, loss[loss=0.3234, simple_loss=0.381, pruned_loss=0.133, over 8376.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3578, pruned_loss=0.1219, over 1257844.03 frames. ], batch size: 49, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:17:03,006 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:10,885 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:24,181 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7561, 2.3459, 4.5918, 1.1956, 2.9962, 2.4300, 1.7410, 2.4904], + device='cuda:2'), covar=tensor([0.1442, 0.1802, 0.0609, 0.3273, 0.1227, 0.2060, 0.1399, 0.2286], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0435, 0.0517, 0.0525, 0.0565, 0.0497, 0.0445, 0.0577], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:17:27,573 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32673.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:34,177 INFO [train.py:901] (2/4) Epoch 5, batch 350, loss[loss=0.2364, simple_loss=0.3028, pruned_loss=0.085, over 7800.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3583, pruned_loss=0.1218, over 1342656.99 frames. ], batch size: 19, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:17:34,412 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:41,060 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.61 vs. limit=5.0 +2023-02-06 00:17:51,720 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:18:04,032 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 3.189e+02 4.031e+02 4.810e+02 8.158e+02, threshold=8.062e+02, percent-clipped=0.0 +2023-02-06 00:18:09,327 INFO [train.py:901] (2/4) Epoch 5, batch 400, loss[loss=0.3371, simple_loss=0.3878, pruned_loss=0.1432, over 8688.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3584, pruned_loss=0.1208, over 1404086.24 frames. ], batch size: 34, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:18:30,397 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6502, 2.7547, 1.8207, 2.2364, 2.3004, 1.5161, 2.1210, 2.1765], + device='cuda:2'), covar=tensor([0.1105, 0.0432, 0.0824, 0.0510, 0.0541, 0.1040, 0.0739, 0.0790], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0240, 0.0308, 0.0304, 0.0316, 0.0312, 0.0336, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:18:43,757 INFO [train.py:901] (2/4) Epoch 5, batch 450, loss[loss=0.2754, simple_loss=0.349, pruned_loss=0.1008, over 8459.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3573, pruned_loss=0.1194, over 1452948.47 frames. ], batch size: 29, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:19:05,621 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.12 vs. limit=5.0 +2023-02-06 00:19:12,442 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.263e+02 3.122e+02 4.068e+02 4.898e+02 9.897e+02, threshold=8.137e+02, percent-clipped=5.0 +2023-02-06 00:19:17,686 INFO [train.py:901] (2/4) Epoch 5, batch 500, loss[loss=0.3144, simple_loss=0.3636, pruned_loss=0.1326, over 7978.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3561, pruned_loss=0.1187, over 1489196.20 frames. ], batch size: 21, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:19:38,471 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3592, 2.7049, 1.7494, 1.9907, 2.1399, 1.2862, 1.8817, 2.2776], + device='cuda:2'), covar=tensor([0.1504, 0.0332, 0.0926, 0.0685, 0.0677, 0.1226, 0.1008, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0239, 0.0310, 0.0303, 0.0320, 0.0315, 0.0337, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:19:52,905 INFO [train.py:901] (2/4) Epoch 5, batch 550, loss[loss=0.2999, simple_loss=0.3628, pruned_loss=0.1185, over 8312.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3556, pruned_loss=0.1188, over 1516623.21 frames. ], batch size: 23, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:20:21,235 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.184e+02 3.133e+02 3.697e+02 5.126e+02 1.321e+03, threshold=7.393e+02, percent-clipped=4.0 +2023-02-06 00:20:21,442 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9634, 2.2049, 1.7163, 2.8238, 1.3089, 1.2475, 1.6535, 2.2997], + device='cuda:2'), covar=tensor([0.1004, 0.1280, 0.1728, 0.0465, 0.1888, 0.2717, 0.2046, 0.1359], + device='cuda:2'), in_proj_covar=tensor([0.0274, 0.0272, 0.0294, 0.0218, 0.0255, 0.0284, 0.0290, 0.0269], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 00:20:26,722 INFO [train.py:901] (2/4) Epoch 5, batch 600, loss[loss=0.2401, simple_loss=0.2986, pruned_loss=0.09077, over 7431.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3534, pruned_loss=0.1175, over 1537015.84 frames. ], batch size: 17, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:20:36,137 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.6645, 1.2546, 3.7852, 1.3437, 3.2555, 3.1395, 3.3761, 3.3434], + device='cuda:2'), covar=tensor([0.0418, 0.3507, 0.0469, 0.2528, 0.1159, 0.0758, 0.0503, 0.0556], + device='cuda:2'), in_proj_covar=tensor([0.0310, 0.0478, 0.0381, 0.0393, 0.0460, 0.0381, 0.0382, 0.0421], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 00:20:50,124 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 00:21:02,154 INFO [train.py:901] (2/4) Epoch 5, batch 650, loss[loss=0.2854, simple_loss=0.3323, pruned_loss=0.1193, over 7544.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3549, pruned_loss=0.119, over 1551534.56 frames. ], batch size: 18, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:21:02,983 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32984.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:21:21,553 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1643, 3.1060, 2.8528, 1.4263, 2.8250, 2.8422, 2.9051, 2.6086], + device='cuda:2'), covar=tensor([0.1265, 0.0883, 0.1278, 0.4944, 0.0972, 0.0974, 0.1542, 0.0925], + device='cuda:2'), in_proj_covar=tensor([0.0381, 0.0280, 0.0315, 0.0409, 0.0303, 0.0266, 0.0296, 0.0238], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 00:21:30,782 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 3.090e+02 3.854e+02 5.024e+02 8.355e+02, threshold=7.708e+02, percent-clipped=4.0 +2023-02-06 00:21:36,138 INFO [train.py:901] (2/4) Epoch 5, batch 700, loss[loss=0.3237, simple_loss=0.3623, pruned_loss=0.1426, over 7926.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.355, pruned_loss=0.1191, over 1564778.23 frames. ], batch size: 20, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:21:44,894 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4213, 1.6393, 1.4643, 1.3272, 1.5274, 1.3358, 1.8160, 1.5699], + device='cuda:2'), covar=tensor([0.0577, 0.1244, 0.1889, 0.1444, 0.0624, 0.1584, 0.0744, 0.0590], + device='cuda:2'), in_proj_covar=tensor([0.0143, 0.0187, 0.0228, 0.0187, 0.0142, 0.0196, 0.0152, 0.0157], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 00:22:11,096 INFO [train.py:901] (2/4) Epoch 5, batch 750, loss[loss=0.3219, simple_loss=0.3802, pruned_loss=0.1319, over 8587.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3553, pruned_loss=0.1196, over 1572958.27 frames. ], batch size: 31, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:22:14,422 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:22:36,868 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 00:22:40,967 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 3.306e+02 4.079e+02 5.042e+02 1.499e+03, threshold=8.159e+02, percent-clipped=7.0 +2023-02-06 00:22:45,500 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 00:22:46,149 INFO [train.py:901] (2/4) Epoch 5, batch 800, loss[loss=0.2903, simple_loss=0.3592, pruned_loss=0.1106, over 8322.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3542, pruned_loss=0.1188, over 1579681.82 frames. ], batch size: 25, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:23:13,721 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33173.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:23:19,948 INFO [train.py:901] (2/4) Epoch 5, batch 850, loss[loss=0.3127, simple_loss=0.3784, pruned_loss=0.1235, over 8514.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3568, pruned_loss=0.1203, over 1593164.32 frames. ], batch size: 26, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:23:49,970 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.888e+02 3.855e+02 5.468e+02 1.103e+03, threshold=7.709e+02, percent-clipped=2.0 +2023-02-06 00:23:56,030 INFO [train.py:901] (2/4) Epoch 5, batch 900, loss[loss=0.2914, simple_loss=0.3429, pruned_loss=0.1199, over 7199.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3555, pruned_loss=0.1197, over 1593739.72 frames. ], batch size: 16, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:24:06,311 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3180, 2.7814, 1.6508, 1.9617, 2.0366, 1.3141, 1.8485, 2.0099], + device='cuda:2'), covar=tensor([0.1258, 0.0235, 0.0883, 0.0587, 0.0630, 0.1165, 0.0906, 0.0773], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0240, 0.0312, 0.0305, 0.0322, 0.0313, 0.0338, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:24:29,752 INFO [train.py:901] (2/4) Epoch 5, batch 950, loss[loss=0.2873, simple_loss=0.3618, pruned_loss=0.1064, over 8202.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3551, pruned_loss=0.1187, over 1597258.93 frames. ], batch size: 23, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:25:00,483 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8385, 1.3877, 3.2460, 1.1803, 2.2004, 3.4133, 3.4665, 3.0307], + device='cuda:2'), covar=tensor([0.1016, 0.1501, 0.0351, 0.2065, 0.0749, 0.0287, 0.0336, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0262, 0.0219, 0.0260, 0.0221, 0.0195, 0.0213, 0.0268], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:25:01,024 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 3.004e+02 3.759e+02 4.642e+02 8.675e+02, threshold=7.519e+02, percent-clipped=2.0 +2023-02-06 00:25:03,068 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:25:05,105 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 00:25:06,446 INFO [train.py:901] (2/4) Epoch 5, batch 1000, loss[loss=0.2946, simple_loss=0.3542, pruned_loss=0.1176, over 8502.00 frames. ], tot_loss[loss=0.296, simple_loss=0.355, pruned_loss=0.1185, over 1602352.53 frames. ], batch size: 26, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:25:40,439 INFO [train.py:901] (2/4) Epoch 5, batch 1050, loss[loss=0.2375, simple_loss=0.295, pruned_loss=0.08996, over 7704.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3548, pruned_loss=0.1188, over 1602712.05 frames. ], batch size: 18, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:25:40,443 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 00:25:47,549 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2573, 1.2436, 3.6318, 1.4196, 2.6256, 2.9230, 3.2067, 3.2275], + device='cuda:2'), covar=tensor([0.0994, 0.4872, 0.0981, 0.3253, 0.2579, 0.1425, 0.0950, 0.1133], + device='cuda:2'), in_proj_covar=tensor([0.0314, 0.0476, 0.0383, 0.0392, 0.0466, 0.0383, 0.0380, 0.0423], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 00:25:52,110 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 00:26:08,784 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 3.252e+02 3.786e+02 4.850e+02 9.380e+02, threshold=7.572e+02, percent-clipped=3.0 +2023-02-06 00:26:13,583 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33431.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:26:14,851 INFO [train.py:901] (2/4) Epoch 5, batch 1100, loss[loss=0.2643, simple_loss=0.336, pruned_loss=0.09623, over 7965.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3544, pruned_loss=0.1187, over 1606192.93 frames. ], batch size: 21, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:26:23,004 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:26:50,044 INFO [train.py:901] (2/4) Epoch 5, batch 1150, loss[loss=0.2798, simple_loss=0.3308, pruned_loss=0.1144, over 7975.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3541, pruned_loss=0.1181, over 1609713.32 frames. ], batch size: 21, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:26:54,937 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:26:59,859 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 00:27:02,054 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 00:27:13,081 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:27:18,265 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.110e+02 3.101e+02 4.052e+02 5.357e+02 1.331e+03, threshold=8.105e+02, percent-clipped=11.0 +2023-02-06 00:27:23,596 INFO [train.py:901] (2/4) Epoch 5, batch 1200, loss[loss=0.2942, simple_loss=0.3668, pruned_loss=0.1108, over 8243.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3538, pruned_loss=0.117, over 1615373.38 frames. ], batch size: 24, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:27:32,563 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:27:34,633 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0960, 1.2695, 1.1956, 0.2991, 1.2620, 0.9907, 0.2180, 1.1756], + device='cuda:2'), covar=tensor([0.0124, 0.0096, 0.0087, 0.0157, 0.0086, 0.0322, 0.0231, 0.0088], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0209, 0.0174, 0.0254, 0.0209, 0.0351, 0.0282, 0.0243], + device='cuda:2'), out_proj_covar=tensor([1.1046e-04, 7.7795e-05, 6.2877e-05, 9.1910e-05, 7.8908e-05, 1.4121e-04, + 1.0697e-04, 8.9627e-05], device='cuda:2') +2023-02-06 00:27:38,026 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5878, 1.9538, 2.3298, 1.0084, 2.4471, 1.5221, 0.8096, 1.8437], + device='cuda:2'), covar=tensor([0.0194, 0.0094, 0.0063, 0.0181, 0.0070, 0.0279, 0.0290, 0.0087], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0209, 0.0174, 0.0254, 0.0208, 0.0349, 0.0281, 0.0243], + device='cuda:2'), out_proj_covar=tensor([1.1037e-04, 7.7643e-05, 6.2731e-05, 9.1967e-05, 7.8688e-05, 1.4054e-04, + 1.0671e-04, 8.9448e-05], device='cuda:2') +2023-02-06 00:28:00,122 INFO [train.py:901] (2/4) Epoch 5, batch 1250, loss[loss=0.3365, simple_loss=0.3961, pruned_loss=0.1384, over 8650.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3534, pruned_loss=0.1177, over 1610638.06 frames. ], batch size: 34, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:28:09,238 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5226, 1.6439, 1.4975, 1.1980, 1.5047, 1.3952, 1.7372, 1.8544], + device='cuda:2'), covar=tensor([0.0597, 0.1293, 0.1849, 0.1482, 0.0657, 0.1754, 0.0830, 0.0589], + device='cuda:2'), in_proj_covar=tensor([0.0145, 0.0186, 0.0226, 0.0187, 0.0141, 0.0197, 0.0154, 0.0158], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 00:28:29,019 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 3.057e+02 3.737e+02 5.343e+02 1.068e+03, threshold=7.474e+02, percent-clipped=1.0 +2023-02-06 00:28:34,046 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:28:34,551 INFO [train.py:901] (2/4) Epoch 5, batch 1300, loss[loss=0.2793, simple_loss=0.3489, pruned_loss=0.1048, over 8490.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.353, pruned_loss=0.1172, over 1612616.69 frames. ], batch size: 28, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:28:48,194 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2215, 2.2912, 1.6105, 1.9156, 1.7695, 1.2919, 1.4698, 1.8449], + device='cuda:2'), covar=tensor([0.0971, 0.0330, 0.0880, 0.0460, 0.0521, 0.1102, 0.0863, 0.0639], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0239, 0.0310, 0.0301, 0.0318, 0.0309, 0.0336, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:29:10,153 INFO [train.py:901] (2/4) Epoch 5, batch 1350, loss[loss=0.2626, simple_loss=0.3358, pruned_loss=0.09469, over 8661.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3523, pruned_loss=0.1172, over 1611698.74 frames. ], batch size: 34, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:29:18,826 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:29:21,575 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33699.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:29:38,304 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33724.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:29:39,349 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 3.141e+02 3.942e+02 4.566e+02 9.800e+02, threshold=7.885e+02, percent-clipped=1.0 +2023-02-06 00:29:44,190 INFO [train.py:901] (2/4) Epoch 5, batch 1400, loss[loss=0.317, simple_loss=0.3733, pruned_loss=0.1304, over 8202.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3534, pruned_loss=0.1175, over 1615035.18 frames. ], batch size: 23, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:30:02,698 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 00:30:18,171 INFO [train.py:901] (2/4) Epoch 5, batch 1450, loss[loss=0.3202, simple_loss=0.3729, pruned_loss=0.1337, over 8478.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3539, pruned_loss=0.118, over 1615953.03 frames. ], batch size: 25, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:30:32,272 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 00:30:32,497 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:30:35,063 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0495, 1.6839, 3.8761, 1.9131, 2.3866, 4.3445, 4.0264, 3.7642], + device='cuda:2'), covar=tensor([0.1115, 0.1507, 0.0433, 0.1678, 0.0983, 0.0215, 0.0430, 0.0527], + device='cuda:2'), in_proj_covar=tensor([0.0239, 0.0273, 0.0223, 0.0262, 0.0228, 0.0201, 0.0226, 0.0280], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 00:30:49,117 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.073e+02 3.068e+02 3.705e+02 5.190e+02 1.303e+03, threshold=7.410e+02, percent-clipped=4.0 +2023-02-06 00:30:50,032 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:30:54,029 INFO [train.py:901] (2/4) Epoch 5, batch 1500, loss[loss=0.2586, simple_loss=0.3168, pruned_loss=0.1002, over 7691.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3538, pruned_loss=0.1177, over 1617803.73 frames. ], batch size: 18, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:30:54,780 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33834.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:31:12,656 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 +2023-02-06 00:31:27,611 INFO [train.py:901] (2/4) Epoch 5, batch 1550, loss[loss=0.2715, simple_loss=0.3201, pruned_loss=0.1114, over 6821.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3527, pruned_loss=0.1174, over 1612219.89 frames. ], batch size: 15, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:31:29,110 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9734, 2.0916, 2.3990, 1.3074, 2.2877, 1.7700, 1.5397, 1.8883], + device='cuda:2'), covar=tensor([0.0211, 0.0099, 0.0062, 0.0180, 0.0096, 0.0199, 0.0248, 0.0106], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0210, 0.0177, 0.0259, 0.0209, 0.0348, 0.0279, 0.0244], + device='cuda:2'), out_proj_covar=tensor([1.1086e-04, 7.7402e-05, 6.3712e-05, 9.3764e-05, 7.8667e-05, 1.3926e-04, + 1.0551e-04, 8.9919e-05], device='cuda:2') +2023-02-06 00:31:31,091 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33888.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:31:31,251 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2023-02-06 00:31:48,413 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33913.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:31:58,126 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 3.419e+02 4.027e+02 4.998e+02 8.696e+02, threshold=8.054e+02, percent-clipped=2.0 +2023-02-06 00:32:03,146 INFO [train.py:901] (2/4) Epoch 5, batch 1600, loss[loss=0.2739, simple_loss=0.3469, pruned_loss=0.1004, over 8301.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3522, pruned_loss=0.117, over 1615462.42 frames. ], batch size: 23, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:32:14,803 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33949.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:32:20,124 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:32:37,058 INFO [train.py:901] (2/4) Epoch 5, batch 1650, loss[loss=0.3019, simple_loss=0.3707, pruned_loss=0.1166, over 8026.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3517, pruned_loss=0.1158, over 1618884.92 frames. ], batch size: 22, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:33:00,643 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 +2023-02-06 00:33:07,304 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.926e+02 3.722e+02 4.611e+02 9.053e+02, threshold=7.444e+02, percent-clipped=4.0 +2023-02-06 00:33:11,843 INFO [train.py:901] (2/4) Epoch 5, batch 1700, loss[loss=0.2279, simple_loss=0.2893, pruned_loss=0.08325, over 7714.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.354, pruned_loss=0.1169, over 1621944.15 frames. ], batch size: 18, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:33:17,289 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34039.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:33:33,625 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:33:36,450 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8899, 2.2676, 4.5686, 1.2540, 2.8977, 2.4600, 1.7582, 2.6021], + device='cuda:2'), covar=tensor([0.1326, 0.1748, 0.0537, 0.3079, 0.1250, 0.2107, 0.1367, 0.1981], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0451, 0.0527, 0.0538, 0.0578, 0.0518, 0.0450, 0.0586], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:33:45,392 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3810, 2.0435, 3.1168, 2.6654, 2.6633, 2.0462, 1.4891, 1.3183], + device='cuda:2'), covar=tensor([0.1442, 0.1670, 0.0373, 0.0750, 0.0754, 0.0845, 0.0916, 0.1744], + device='cuda:2'), in_proj_covar=tensor([0.0713, 0.0657, 0.0564, 0.0640, 0.0750, 0.0608, 0.0591, 0.0616], + 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') +2023-02-06 00:33:47,846 INFO [train.py:901] (2/4) Epoch 5, batch 1750, loss[loss=0.2144, simple_loss=0.2853, pruned_loss=0.07179, over 7700.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3523, pruned_loss=0.1155, over 1623699.01 frames. ], batch size: 18, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:34:16,524 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.177e+02 3.118e+02 3.687e+02 4.787e+02 9.448e+02, threshold=7.373e+02, percent-clipped=7.0 +2023-02-06 00:34:21,853 INFO [train.py:901] (2/4) Epoch 5, batch 1800, loss[loss=0.2578, simple_loss=0.3251, pruned_loss=0.09521, over 7649.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3526, pruned_loss=0.1156, over 1621342.28 frames. ], batch size: 19, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:34:36,711 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:34:43,249 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:34:57,300 INFO [train.py:901] (2/4) Epoch 5, batch 1850, loss[loss=0.2348, simple_loss=0.3176, pruned_loss=0.07597, over 8230.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3519, pruned_loss=0.1156, over 1618947.18 frames. ], batch size: 22, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:35:12,359 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34205.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:35:26,214 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 3.489e+02 4.150e+02 5.670e+02 1.027e+03, threshold=8.299e+02, percent-clipped=7.0 +2023-02-06 00:35:29,069 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34230.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:35:30,903 INFO [train.py:901] (2/4) Epoch 5, batch 1900, loss[loss=0.2628, simple_loss=0.3313, pruned_loss=0.09713, over 8253.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3523, pruned_loss=0.1157, over 1618599.70 frames. ], batch size: 24, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:36:05,945 INFO [train.py:901] (2/4) Epoch 5, batch 1950, loss[loss=0.3396, simple_loss=0.3927, pruned_loss=0.1432, over 8579.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3519, pruned_loss=0.1157, over 1617453.68 frames. ], batch size: 39, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:36:09,863 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 00:36:18,719 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:36:23,388 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 00:36:35,388 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.967e+02 3.945e+02 4.927e+02 1.257e+03, threshold=7.890e+02, percent-clipped=2.0 +2023-02-06 00:36:39,083 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4530, 2.2947, 2.5481, 2.1317, 1.5305, 2.6190, 0.8697, 1.9499], + device='cuda:2'), covar=tensor([0.3309, 0.2791, 0.1117, 0.3496, 0.7540, 0.1056, 0.7161, 0.2759], + device='cuda:2'), in_proj_covar=tensor([0.0130, 0.0123, 0.0084, 0.0167, 0.0217, 0.0085, 0.0149, 0.0125], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 00:36:40,168 INFO [train.py:901] (2/4) Epoch 5, batch 2000, loss[loss=0.3461, simple_loss=0.3906, pruned_loss=0.1507, over 8027.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3522, pruned_loss=0.1161, over 1616881.56 frames. ], batch size: 22, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:36:42,286 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 00:37:14,366 INFO [train.py:901] (2/4) Epoch 5, batch 2050, loss[loss=0.3248, simple_loss=0.3801, pruned_loss=0.1347, over 8662.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3524, pruned_loss=0.1165, over 1615616.00 frames. ], batch size: 34, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:37:17,170 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5295, 1.8106, 1.9898, 1.7850, 1.0903, 2.0836, 0.3620, 1.2614], + device='cuda:2'), covar=tensor([0.4257, 0.2737, 0.1425, 0.2651, 0.6797, 0.1099, 0.6874, 0.2970], + device='cuda:2'), in_proj_covar=tensor([0.0130, 0.0122, 0.0083, 0.0167, 0.0213, 0.0084, 0.0147, 0.0123], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 00:37:31,121 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:33,924 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:38,677 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34416.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:45,186 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 3.245e+02 4.066e+02 4.898e+02 1.293e+03, threshold=8.132e+02, percent-clipped=4.0 +2023-02-06 00:37:49,847 INFO [train.py:901] (2/4) Epoch 5, batch 2100, loss[loss=0.2145, simple_loss=0.2854, pruned_loss=0.07183, over 7664.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3526, pruned_loss=0.1168, over 1615585.35 frames. ], batch size: 19, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:37:51,414 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:52,722 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7166, 2.0145, 1.5013, 2.3573, 0.9737, 1.4167, 1.5064, 1.9338], + device='cuda:2'), covar=tensor([0.0963, 0.1151, 0.1698, 0.0643, 0.1933, 0.2239, 0.1587, 0.1101], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0268, 0.0287, 0.0223, 0.0260, 0.0288, 0.0290, 0.0264], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 00:38:23,173 INFO [train.py:901] (2/4) Epoch 5, batch 2150, loss[loss=0.3058, simple_loss=0.3783, pruned_loss=0.1166, over 8239.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3535, pruned_loss=0.117, over 1617895.16 frames. ], batch size: 22, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:38:30,768 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6441, 1.9198, 3.3721, 1.1944, 2.3518, 1.7788, 1.6085, 2.1157], + device='cuda:2'), covar=tensor([0.1312, 0.1701, 0.0622, 0.2755, 0.1246, 0.2251, 0.1292, 0.1889], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0448, 0.0532, 0.0530, 0.0577, 0.0520, 0.0445, 0.0581], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:38:39,926 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:38:50,620 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:38:51,960 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1846, 1.4443, 4.3763, 1.7706, 3.7815, 3.6025, 3.9078, 3.7908], + device='cuda:2'), covar=tensor([0.0490, 0.3793, 0.0398, 0.2466, 0.1142, 0.0659, 0.0521, 0.0639], + device='cuda:2'), in_proj_covar=tensor([0.0321, 0.0489, 0.0399, 0.0408, 0.0484, 0.0393, 0.0398, 0.0440], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 00:38:53,806 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.929e+02 3.753e+02 4.663e+02 1.529e+03, threshold=7.506e+02, percent-clipped=2.0 +2023-02-06 00:38:53,933 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8629, 5.9102, 5.1298, 2.4727, 5.1898, 5.4757, 5.4198, 5.0124], + device='cuda:2'), covar=tensor([0.0518, 0.0380, 0.0711, 0.4141, 0.0550, 0.0539, 0.1014, 0.0573], + device='cuda:2'), in_proj_covar=tensor([0.0391, 0.0289, 0.0310, 0.0401, 0.0309, 0.0270, 0.0300, 0.0244], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 00:38:59,149 INFO [train.py:901] (2/4) Epoch 5, batch 2200, loss[loss=0.3045, simple_loss=0.3665, pruned_loss=0.1212, over 8243.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3531, pruned_loss=0.1167, over 1619436.48 frames. ], batch size: 24, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:39:09,611 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.84 vs. limit=5.0 +2023-02-06 00:39:20,089 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3397, 1.2305, 4.4342, 1.6561, 3.8619, 3.6957, 3.8894, 3.8772], + device='cuda:2'), covar=tensor([0.0356, 0.3797, 0.0351, 0.2441, 0.0998, 0.0594, 0.0520, 0.0500], + device='cuda:2'), in_proj_covar=tensor([0.0322, 0.0489, 0.0400, 0.0411, 0.0483, 0.0395, 0.0403, 0.0440], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 00:39:32,427 INFO [train.py:901] (2/4) Epoch 5, batch 2250, loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 8464.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3531, pruned_loss=0.1174, over 1618850.29 frames. ], batch size: 25, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:39:39,440 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.02 vs. limit=5.0 +2023-02-06 00:39:52,223 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:39:59,511 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:02,635 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 3.688e+02 4.883e+02 6.349e+02 4.437e+03, threshold=9.766e+02, percent-clipped=16.0 +2023-02-06 00:40:07,909 INFO [train.py:901] (2/4) Epoch 5, batch 2300, loss[loss=0.3049, simple_loss=0.3655, pruned_loss=0.1221, over 8193.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3519, pruned_loss=0.1167, over 1615001.92 frames. ], batch size: 23, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:40:12,727 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34640.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:16,071 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:34,981 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:42,248 INFO [train.py:901] (2/4) Epoch 5, batch 2350, loss[loss=0.2958, simple_loss=0.3665, pruned_loss=0.1125, over 8505.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3534, pruned_loss=0.1172, over 1618207.50 frames. ], batch size: 28, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:40:51,695 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34697.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:41:11,430 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 3.189e+02 4.018e+02 4.942e+02 1.178e+03, threshold=8.036e+02, percent-clipped=1.0 +2023-02-06 00:41:12,934 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34728.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:41:16,102 INFO [train.py:901] (2/4) Epoch 5, batch 2400, loss[loss=0.3197, simple_loss=0.374, pruned_loss=0.1327, over 8510.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3534, pruned_loss=0.1177, over 1616868.42 frames. ], batch size: 26, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:41:47,415 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34777.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:41:51,215 INFO [train.py:901] (2/4) Epoch 5, batch 2450, loss[loss=0.2828, simple_loss=0.3478, pruned_loss=0.1089, over 8734.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3542, pruned_loss=0.1186, over 1613644.38 frames. ], batch size: 34, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:42:04,208 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:42:14,155 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1720, 1.8298, 2.8919, 2.2981, 2.3161, 1.9143, 1.4528, 1.0054], + device='cuda:2'), covar=tensor([0.1498, 0.1492, 0.0345, 0.0738, 0.0723, 0.0790, 0.0928, 0.1551], + device='cuda:2'), in_proj_covar=tensor([0.0726, 0.0661, 0.0564, 0.0639, 0.0744, 0.0609, 0.0593, 0.0617], + 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') +2023-02-06 00:42:19,462 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.991e+02 3.791e+02 4.954e+02 1.109e+03, threshold=7.583e+02, percent-clipped=3.0 +2023-02-06 00:42:24,154 INFO [train.py:901] (2/4) Epoch 5, batch 2500, loss[loss=0.2735, simple_loss=0.3426, pruned_loss=0.1022, over 8197.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3535, pruned_loss=0.1182, over 1617482.10 frames. ], batch size: 23, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:42:55,975 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34878.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:42:59,761 INFO [train.py:901] (2/4) Epoch 5, batch 2550, loss[loss=0.2764, simple_loss=0.3355, pruned_loss=0.1086, over 8235.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3517, pruned_loss=0.1166, over 1616454.45 frames. ], batch size: 22, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:43:13,486 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34903.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:43:29,138 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 2.947e+02 3.618e+02 4.736e+02 1.253e+03, threshold=7.237e+02, percent-clipped=4.0 +2023-02-06 00:43:33,913 INFO [train.py:901] (2/4) Epoch 5, batch 2600, loss[loss=0.3373, simple_loss=0.3903, pruned_loss=0.1421, over 8369.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3545, pruned_loss=0.1184, over 1619250.16 frames. ], batch size: 24, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:43:49,010 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34955.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:44:09,526 INFO [train.py:901] (2/4) Epoch 5, batch 2650, loss[loss=0.2813, simple_loss=0.3383, pruned_loss=0.1121, over 7820.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3549, pruned_loss=0.1188, over 1620806.48 frames. ], batch size: 20, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:44:10,293 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34984.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:44:13,019 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34988.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:44:39,307 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.922e+02 3.827e+02 4.980e+02 8.274e+02, threshold=7.654e+02, percent-clipped=5.0 +2023-02-06 00:44:41,488 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3077, 1.4882, 1.5317, 0.8240, 1.5746, 1.1234, 0.6827, 1.4233], + device='cuda:2'), covar=tensor([0.0162, 0.0087, 0.0059, 0.0155, 0.0091, 0.0254, 0.0212, 0.0065], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0216, 0.0177, 0.0259, 0.0203, 0.0347, 0.0270, 0.0246], + device='cuda:2'), out_proj_covar=tensor([1.1211e-04, 7.8453e-05, 6.2863e-05, 9.3077e-05, 7.5833e-05, 1.3713e-04, + 9.9953e-05, 8.9828e-05], device='cuda:2') +2023-02-06 00:44:43,766 INFO [train.py:901] (2/4) Epoch 5, batch 2700, loss[loss=0.3468, simple_loss=0.3856, pruned_loss=0.1539, over 8498.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3537, pruned_loss=0.1175, over 1623190.24 frames. ], batch size: 26, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:44:44,126 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.52 vs. limit=5.0 +2023-02-06 00:45:09,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:10,728 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35072.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:45:18,323 INFO [train.py:901] (2/4) Epoch 5, batch 2750, loss[loss=0.2983, simple_loss=0.3616, pruned_loss=0.1175, over 8181.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3538, pruned_loss=0.1177, over 1619691.90 frames. ], batch size: 23, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:45:29,196 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4036, 1.8614, 3.3191, 2.5844, 2.6773, 1.9161, 1.4033, 1.3165], + device='cuda:2'), covar=tensor([0.1703, 0.2051, 0.0436, 0.1010, 0.0973, 0.0926, 0.0985, 0.2009], + device='cuda:2'), in_proj_covar=tensor([0.0738, 0.0666, 0.0564, 0.0649, 0.0747, 0.0619, 0.0594, 0.0621], + 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') +2023-02-06 00:45:29,757 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35099.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:32,516 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:37,951 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35110.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:39,343 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7726, 2.9122, 1.8413, 2.0867, 2.4404, 1.5716, 2.2082, 2.2969], + device='cuda:2'), covar=tensor([0.1123, 0.0209, 0.0833, 0.0589, 0.0462, 0.1028, 0.0775, 0.0769], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0240, 0.0313, 0.0308, 0.0321, 0.0304, 0.0336, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:45:45,352 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8099, 2.3000, 1.7019, 2.7335, 1.6347, 1.4968, 1.8634, 2.2436], + device='cuda:2'), covar=tensor([0.1094, 0.0979, 0.1365, 0.0507, 0.1308, 0.1928, 0.1436, 0.0921], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0272, 0.0294, 0.0227, 0.0255, 0.0289, 0.0288, 0.0263], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 00:45:48,687 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.004e+02 3.039e+02 3.659e+02 5.251e+02 1.248e+03, threshold=7.317e+02, percent-clipped=8.0 +2023-02-06 00:45:53,667 INFO [train.py:901] (2/4) Epoch 5, batch 2800, loss[loss=0.2491, simple_loss=0.3193, pruned_loss=0.08946, over 7914.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3533, pruned_loss=0.1172, over 1619498.94 frames. ], batch size: 20, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:46:16,473 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-06 00:46:27,512 INFO [train.py:901] (2/4) Epoch 5, batch 2850, loss[loss=0.2782, simple_loss=0.3374, pruned_loss=0.1096, over 8022.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3531, pruned_loss=0.1179, over 1613735.84 frames. ], batch size: 22, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:46:30,531 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35187.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:46:34,577 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1752, 1.8337, 1.6046, 1.4350, 1.6327, 1.6733, 2.1908, 1.7184], + device='cuda:2'), covar=tensor([0.0590, 0.1250, 0.1811, 0.1502, 0.0670, 0.1625, 0.0822, 0.0656], + device='cuda:2'), in_proj_covar=tensor([0.0143, 0.0184, 0.0225, 0.0187, 0.0139, 0.0194, 0.0150, 0.0158], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 00:46:46,215 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4964, 1.6055, 2.3726, 1.0883, 1.7572, 1.7419, 1.4547, 1.4214], + device='cuda:2'), covar=tensor([0.1252, 0.1507, 0.0592, 0.2693, 0.1025, 0.2046, 0.1301, 0.1492], + device='cuda:2'), in_proj_covar=tensor([0.0462, 0.0437, 0.0519, 0.0526, 0.0576, 0.0508, 0.0447, 0.0581], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:46:58,369 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 2.990e+02 3.598e+02 4.675e+02 1.498e+03, threshold=7.197e+02, percent-clipped=4.0 +2023-02-06 00:47:03,665 INFO [train.py:901] (2/4) Epoch 5, batch 2900, loss[loss=0.3125, simple_loss=0.3711, pruned_loss=0.1269, over 8659.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3535, pruned_loss=0.1186, over 1614036.76 frames. ], batch size: 34, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:47:10,554 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2081, 1.7422, 2.6996, 2.2969, 2.3088, 1.9542, 1.4519, 1.3932], + device='cuda:2'), covar=tensor([0.1148, 0.1620, 0.0353, 0.0673, 0.0628, 0.0665, 0.0772, 0.1337], + device='cuda:2'), in_proj_covar=tensor([0.0729, 0.0665, 0.0559, 0.0645, 0.0748, 0.0612, 0.0593, 0.0611], + 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') +2023-02-06 00:47:36,539 INFO [train.py:901] (2/4) Epoch 5, batch 2950, loss[loss=0.3443, simple_loss=0.3931, pruned_loss=0.1477, over 8407.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3531, pruned_loss=0.1179, over 1609054.61 frames. ], batch size: 49, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:47:41,853 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 00:47:54,166 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5429, 4.5887, 4.0044, 1.8127, 4.0764, 3.9962, 4.2315, 3.6249], + device='cuda:2'), covar=tensor([0.0714, 0.0553, 0.0916, 0.4690, 0.0666, 0.0826, 0.1191, 0.0715], + device='cuda:2'), in_proj_covar=tensor([0.0383, 0.0279, 0.0309, 0.0393, 0.0302, 0.0267, 0.0292, 0.0236], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 00:48:06,832 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 3.185e+02 3.825e+02 4.988e+02 1.295e+03, threshold=7.649e+02, percent-clipped=4.0 +2023-02-06 00:48:07,076 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:12,109 INFO [train.py:901] (2/4) Epoch 5, batch 3000, loss[loss=0.3211, simple_loss=0.3784, pruned_loss=0.1319, over 8664.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3527, pruned_loss=0.1174, over 1607296.12 frames. ], batch size: 39, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:48:12,109 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 00:48:25,509 INFO [train.py:935] (2/4) Epoch 5, validation: loss=0.2228, simple_loss=0.319, pruned_loss=0.0633, over 944034.00 frames. +2023-02-06 00:48:25,509 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 00:48:34,488 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5193, 1.5792, 1.4836, 1.3641, 0.7791, 1.6212, 0.0843, 1.0144], + device='cuda:2'), covar=tensor([0.3225, 0.2284, 0.1114, 0.2253, 0.6226, 0.0974, 0.5533, 0.2258], + device='cuda:2'), in_proj_covar=tensor([0.0123, 0.0124, 0.0078, 0.0163, 0.0202, 0.0080, 0.0142, 0.0119], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 00:48:39,261 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35351.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:41,986 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:44,693 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:47,340 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35363.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:48:59,229 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35380.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:01,098 INFO [train.py:901] (2/4) Epoch 5, batch 3050, loss[loss=0.2839, simple_loss=0.3515, pruned_loss=0.1081, over 8645.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3535, pruned_loss=0.1175, over 1615156.85 frames. ], batch size: 34, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:49:01,954 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:07,210 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:21,413 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1442, 1.6249, 1.2411, 1.6533, 1.4496, 1.0833, 1.1097, 1.4335], + device='cuda:2'), covar=tensor([0.0766, 0.0332, 0.0779, 0.0384, 0.0525, 0.0892, 0.0689, 0.0502], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0240, 0.0313, 0.0308, 0.0322, 0.0307, 0.0344, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:49:29,669 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.238e+02 3.010e+02 3.735e+02 4.816e+02 9.592e+02, threshold=7.471e+02, percent-clipped=3.0 +2023-02-06 00:49:34,199 INFO [train.py:901] (2/4) Epoch 5, batch 3100, loss[loss=0.3456, simple_loss=0.4, pruned_loss=0.1456, over 8697.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3529, pruned_loss=0.1173, over 1612464.48 frames. ], batch size: 34, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:49:36,283 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1551, 1.3344, 2.2968, 1.0400, 2.3448, 2.4341, 2.4256, 2.0881], + device='cuda:2'), covar=tensor([0.1091, 0.1171, 0.0499, 0.2007, 0.0487, 0.0367, 0.0535, 0.0786], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0265, 0.0225, 0.0264, 0.0221, 0.0198, 0.0229, 0.0276], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 00:49:41,039 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35443.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:49:48,740 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35454.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:59,648 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35468.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:50:09,986 INFO [train.py:901] (2/4) Epoch 5, batch 3150, loss[loss=0.3054, simple_loss=0.3691, pruned_loss=0.1208, over 8193.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3532, pruned_loss=0.117, over 1610995.41 frames. ], batch size: 23, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:50:39,622 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.108e+02 3.249e+02 4.087e+02 5.030e+02 9.472e+02, threshold=8.174e+02, percent-clipped=3.0 +2023-02-06 00:50:43,440 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 00:50:44,419 INFO [train.py:901] (2/4) Epoch 5, batch 3200, loss[loss=0.3147, simple_loss=0.3619, pruned_loss=0.1338, over 8031.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3541, pruned_loss=0.1179, over 1610285.94 frames. ], batch size: 22, lr: 1.51e-02, grad_scale: 8.0 +2023-02-06 00:50:54,843 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:51:09,461 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35569.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:51:20,160 INFO [train.py:901] (2/4) Epoch 5, batch 3250, loss[loss=0.2882, simple_loss=0.3639, pruned_loss=0.1062, over 8256.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3537, pruned_loss=0.1173, over 1616424.11 frames. ], batch size: 24, lr: 1.51e-02, grad_scale: 8.0 +2023-02-06 00:51:35,383 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5381, 1.5059, 1.6992, 1.5575, 1.2085, 1.6790, 0.8322, 1.3322], + device='cuda:2'), covar=tensor([0.2778, 0.1556, 0.0887, 0.1455, 0.3830, 0.0735, 0.3652, 0.1735], + device='cuda:2'), in_proj_covar=tensor([0.0126, 0.0125, 0.0080, 0.0165, 0.0204, 0.0081, 0.0140, 0.0120], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 00:51:43,807 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2992, 1.9945, 3.0845, 2.5049, 2.6102, 2.0060, 1.4036, 1.4387], + device='cuda:2'), covar=tensor([0.1895, 0.2057, 0.0484, 0.0954, 0.0860, 0.1011, 0.1147, 0.1972], + device='cuda:2'), in_proj_covar=tensor([0.0738, 0.0667, 0.0571, 0.0652, 0.0760, 0.0623, 0.0605, 0.0617], + 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') +2023-02-06 00:51:50,492 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 3.378e+02 4.149e+02 5.121e+02 1.146e+03, threshold=8.298e+02, percent-clipped=3.0 +2023-02-06 00:51:55,280 INFO [train.py:901] (2/4) Epoch 5, batch 3300, loss[loss=0.2926, simple_loss=0.3561, pruned_loss=0.1145, over 8318.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3547, pruned_loss=0.1177, over 1613704.43 frames. ], batch size: 25, lr: 1.51e-02, grad_scale: 8.0 +2023-02-06 00:52:04,544 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0899, 1.6540, 1.1371, 1.4769, 1.2087, 0.9542, 1.2289, 1.4106], + device='cuda:2'), covar=tensor([0.1058, 0.0426, 0.1315, 0.0638, 0.0854, 0.1528, 0.0971, 0.0779], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0236, 0.0312, 0.0302, 0.0318, 0.0307, 0.0344, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 00:52:12,160 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1541, 1.1992, 1.0666, 1.0112, 0.7948, 1.1944, 0.0121, 0.8895], + device='cuda:2'), covar=tensor([0.3825, 0.2337, 0.1360, 0.2314, 0.5847, 0.1034, 0.5168, 0.2555], + device='cuda:2'), in_proj_covar=tensor([0.0125, 0.0124, 0.0080, 0.0164, 0.0204, 0.0081, 0.0140, 0.0121], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 00:52:30,151 INFO [train.py:901] (2/4) Epoch 5, batch 3350, loss[loss=0.319, simple_loss=0.3636, pruned_loss=0.1372, over 8232.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3557, pruned_loss=0.1181, over 1616864.88 frames. ], batch size: 22, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:52:47,835 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35707.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:53:01,469 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 3.555e+02 4.125e+02 4.946e+02 1.065e+03, threshold=8.250e+02, percent-clipped=5.0 +2023-02-06 00:53:06,230 INFO [train.py:901] (2/4) Epoch 5, batch 3400, loss[loss=0.2854, simple_loss=0.3476, pruned_loss=0.1116, over 8511.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3555, pruned_loss=0.1182, over 1619401.18 frames. ], batch size: 26, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:53:08,409 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:53:09,277 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4735, 1.9298, 3.0909, 1.1243, 2.2264, 1.7011, 1.5706, 1.8636], + device='cuda:2'), covar=tensor([0.1438, 0.1539, 0.0567, 0.2931, 0.1194, 0.2246, 0.1357, 0.1990], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0443, 0.0525, 0.0531, 0.0573, 0.0513, 0.0446, 0.0590], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:53:28,169 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6462, 1.9002, 1.9039, 1.3574, 0.9078, 2.0132, 0.2754, 1.2517], + device='cuda:2'), covar=tensor([0.3136, 0.1839, 0.1132, 0.2946, 0.7359, 0.0961, 0.5482, 0.2059], + device='cuda:2'), in_proj_covar=tensor([0.0123, 0.0125, 0.0080, 0.0164, 0.0206, 0.0083, 0.0141, 0.0119], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 00:53:28,797 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1901, 1.4084, 1.4022, 1.2474, 1.3729, 1.3059, 1.6840, 1.6744], + device='cuda:2'), covar=tensor([0.0659, 0.1311, 0.1847, 0.1566, 0.0658, 0.1691, 0.0795, 0.0596], + device='cuda:2'), in_proj_covar=tensor([0.0142, 0.0185, 0.0223, 0.0186, 0.0138, 0.0196, 0.0148, 0.0159], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 00:53:36,086 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-06 00:53:39,718 INFO [train.py:901] (2/4) Epoch 5, batch 3450, loss[loss=0.2562, simple_loss=0.3304, pruned_loss=0.09096, over 8466.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3554, pruned_loss=0.1181, over 1618389.19 frames. ], batch size: 25, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:53:43,893 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:53:56,605 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1376, 1.2509, 3.3692, 1.2187, 2.3084, 3.8931, 3.7331, 3.3224], + device='cuda:2'), covar=tensor([0.0930, 0.1658, 0.0356, 0.2054, 0.0835, 0.0191, 0.0312, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0227, 0.0263, 0.0218, 0.0259, 0.0217, 0.0195, 0.0225, 0.0268], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 00:54:05,362 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8251, 2.5160, 4.5419, 1.2864, 2.6679, 2.1919, 2.0155, 2.3103], + device='cuda:2'), covar=tensor([0.1543, 0.1735, 0.0608, 0.3387, 0.1725, 0.2473, 0.1356, 0.2758], + device='cuda:2'), in_proj_covar=tensor([0.0468, 0.0443, 0.0525, 0.0537, 0.0579, 0.0516, 0.0447, 0.0594], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 00:54:07,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35822.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:54:09,940 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:10,372 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.123e+02 3.051e+02 3.738e+02 4.571e+02 6.690e+02, threshold=7.475e+02, percent-clipped=0.0 +2023-02-06 00:54:12,584 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35829.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:15,091 INFO [train.py:901] (2/4) Epoch 5, batch 3500, loss[loss=0.297, simple_loss=0.3562, pruned_loss=0.119, over 8106.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3531, pruned_loss=0.1163, over 1614461.58 frames. ], batch size: 23, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:54:27,023 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35850.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:27,675 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:40,687 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 00:54:48,819 INFO [train.py:901] (2/4) Epoch 5, batch 3550, loss[loss=0.2729, simple_loss=0.3401, pruned_loss=0.1028, over 8338.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3528, pruned_loss=0.1166, over 1610845.35 frames. ], batch size: 25, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:54:54,910 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35892.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:55:19,532 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.232e+02 3.318e+02 3.882e+02 4.908e+02 1.221e+03, threshold=7.763e+02, percent-clipped=6.0 +2023-02-06 00:55:19,958 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2023-02-06 00:55:23,997 INFO [train.py:901] (2/4) Epoch 5, batch 3600, loss[loss=0.3259, simple_loss=0.3653, pruned_loss=0.1433, over 7921.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3533, pruned_loss=0.117, over 1611294.50 frames. ], batch size: 20, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:55:31,761 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 00:55:52,653 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 00:55:57,697 INFO [train.py:901] (2/4) Epoch 5, batch 3650, loss[loss=0.277, simple_loss=0.3586, pruned_loss=0.09768, over 8487.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3537, pruned_loss=0.1172, over 1617321.84 frames. ], batch size: 28, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:56:00,559 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8246, 1.4462, 3.1974, 1.3826, 2.2263, 3.5715, 3.3498, 2.9954], + device='cuda:2'), covar=tensor([0.1097, 0.1320, 0.0402, 0.1789, 0.0680, 0.0210, 0.0351, 0.0591], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0264, 0.0221, 0.0259, 0.0218, 0.0197, 0.0226, 0.0269], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 00:56:15,074 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:56:27,680 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 3.345e+02 4.197e+02 5.280e+02 9.599e+02, threshold=8.394e+02, percent-clipped=10.0 +2023-02-06 00:56:32,338 INFO [train.py:901] (2/4) Epoch 5, batch 3700, loss[loss=0.3066, simple_loss=0.3626, pruned_loss=0.1253, over 8448.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3535, pruned_loss=0.1171, over 1613864.70 frames. ], batch size: 27, lr: 1.50e-02, grad_scale: 16.0 +2023-02-06 00:56:40,794 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 00:57:04,793 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36078.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:57:07,929 INFO [train.py:901] (2/4) Epoch 5, batch 3750, loss[loss=0.2784, simple_loss=0.3401, pruned_loss=0.1083, over 8493.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3532, pruned_loss=0.117, over 1614360.89 frames. ], batch size: 49, lr: 1.50e-02, grad_scale: 16.0 +2023-02-06 00:57:21,391 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36103.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:57:24,067 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7259, 1.2483, 3.8815, 1.3838, 3.3239, 3.2258, 3.4194, 3.3799], + device='cuda:2'), covar=tensor([0.0468, 0.3311, 0.0415, 0.2523, 0.1065, 0.0665, 0.0478, 0.0576], + device='cuda:2'), in_proj_covar=tensor([0.0337, 0.0494, 0.0410, 0.0421, 0.0489, 0.0404, 0.0397, 0.0450], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 00:57:24,154 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:57:36,067 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7189, 2.0748, 2.1979, 0.9880, 2.2433, 1.5648, 0.6259, 1.7956], + device='cuda:2'), covar=tensor([0.0168, 0.0107, 0.0091, 0.0201, 0.0118, 0.0294, 0.0288, 0.0097], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0221, 0.0176, 0.0264, 0.0203, 0.0349, 0.0274, 0.0246], + device='cuda:2'), out_proj_covar=tensor([1.0850e-04, 7.9083e-05, 6.2023e-05, 9.3622e-05, 7.4231e-05, 1.3604e-04, + 1.0045e-04, 8.8113e-05], device='cuda:2') +2023-02-06 00:57:37,191 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 3.033e+02 3.704e+02 4.599e+02 1.470e+03, threshold=7.408e+02, percent-clipped=9.0 +2023-02-06 00:57:40,783 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:57:41,247 INFO [train.py:901] (2/4) Epoch 5, batch 3800, loss[loss=0.266, simple_loss=0.3326, pruned_loss=0.09973, over 8349.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.352, pruned_loss=0.1157, over 1617855.83 frames. ], batch size: 24, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:57:41,318 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:58:09,342 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36173.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:58:15,719 INFO [train.py:901] (2/4) Epoch 5, batch 3850, loss[loss=0.4042, simple_loss=0.4168, pruned_loss=0.1958, over 6554.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3521, pruned_loss=0.1161, over 1618745.12 frames. ], batch size: 71, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:58:27,233 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36199.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:58:29,596 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.86 vs. limit=5.0 +2023-02-06 00:58:41,067 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 00:58:44,299 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.15 vs. limit=5.0 +2023-02-06 00:58:45,741 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 3.284e+02 4.097e+02 5.243e+02 1.380e+03, threshold=8.194e+02, percent-clipped=10.0 +2023-02-06 00:58:49,706 INFO [train.py:901] (2/4) Epoch 5, batch 3900, loss[loss=0.2502, simple_loss=0.3025, pruned_loss=0.09897, over 7230.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.352, pruned_loss=0.1156, over 1619226.21 frames. ], batch size: 16, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:58:59,605 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36248.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:09,645 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:24,818 INFO [train.py:901] (2/4) Epoch 5, batch 3950, loss[loss=0.2771, simple_loss=0.3435, pruned_loss=0.1053, over 7818.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3511, pruned_loss=0.1151, over 1616310.63 frames. ], batch size: 20, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:59:28,399 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:28,422 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:54,567 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 2.959e+02 3.540e+02 4.519e+02 1.633e+03, threshold=7.079e+02, percent-clipped=6.0 +2023-02-06 00:59:58,368 INFO [train.py:901] (2/4) Epoch 5, batch 4000, loss[loss=0.2765, simple_loss=0.3278, pruned_loss=0.1126, over 7830.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3512, pruned_loss=0.1152, over 1618803.85 frames. ], batch size: 20, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:00:22,607 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1130, 4.0985, 3.6605, 2.1052, 3.6243, 3.7200, 3.7566, 3.3492], + device='cuda:2'), covar=tensor([0.0858, 0.0608, 0.1089, 0.4109, 0.0798, 0.0735, 0.1315, 0.0848], + device='cuda:2'), in_proj_covar=tensor([0.0387, 0.0280, 0.0314, 0.0399, 0.0312, 0.0270, 0.0298, 0.0241], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 01:00:31,740 INFO [train.py:901] (2/4) Epoch 5, batch 4050, loss[loss=0.2791, simple_loss=0.3299, pruned_loss=0.1141, over 7245.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3513, pruned_loss=0.1151, over 1615915.69 frames. ], batch size: 16, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:01:03,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 3.036e+02 3.577e+02 4.439e+02 7.437e+02, threshold=7.154e+02, percent-clipped=1.0 +2023-02-06 01:01:07,957 INFO [train.py:901] (2/4) Epoch 5, batch 4100, loss[loss=0.2759, simple_loss=0.3535, pruned_loss=0.09918, over 8441.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3512, pruned_loss=0.1151, over 1617041.63 frames. ], batch size: 27, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:01:41,918 INFO [train.py:901] (2/4) Epoch 5, batch 4150, loss[loss=0.2551, simple_loss=0.3197, pruned_loss=0.09527, over 8030.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3498, pruned_loss=0.1141, over 1615466.52 frames. ], batch size: 22, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:01:56,692 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:09,789 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7941, 2.1183, 3.7351, 3.0538, 3.1518, 2.2549, 1.6850, 2.0739], + device='cuda:2'), covar=tensor([0.1515, 0.2191, 0.0447, 0.0891, 0.0796, 0.0923, 0.0947, 0.1816], + device='cuda:2'), in_proj_covar=tensor([0.0745, 0.0668, 0.0567, 0.0654, 0.0754, 0.0621, 0.0601, 0.0614], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:02:13,614 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.751e+02 3.740e+02 4.679e+02 9.033e+02, threshold=7.480e+02, percent-clipped=3.0 +2023-02-06 01:02:15,168 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36529.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:17,736 INFO [train.py:901] (2/4) Epoch 5, batch 4200, loss[loss=0.288, simple_loss=0.3635, pruned_loss=0.1062, over 8308.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3489, pruned_loss=0.113, over 1613820.41 frames. ], batch size: 25, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:02:24,837 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:25,675 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36544.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:43,320 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 01:02:43,514 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36569.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:52,964 INFO [train.py:901] (2/4) Epoch 5, batch 4250, loss[loss=0.2958, simple_loss=0.3438, pruned_loss=0.1239, over 5566.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3501, pruned_loss=0.1142, over 1607388.09 frames. ], batch size: 12, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:02:53,363 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 01:02:58,784 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0803, 1.2928, 1.2061, 0.2034, 1.1588, 0.9146, 0.1063, 1.1216], + device='cuda:2'), covar=tensor([0.0134, 0.0108, 0.0088, 0.0199, 0.0127, 0.0334, 0.0255, 0.0100], + device='cuda:2'), in_proj_covar=tensor([0.0304, 0.0226, 0.0177, 0.0263, 0.0209, 0.0352, 0.0277, 0.0251], + device='cuda:2'), out_proj_covar=tensor([1.1029e-04, 8.0523e-05, 6.1631e-05, 9.2824e-05, 7.6212e-05, 1.3625e-04, + 1.0114e-04, 8.9913e-05], device='cuda:2') +2023-02-06 01:03:05,699 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 01:03:19,528 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.6778, 1.3992, 3.8286, 1.3777, 3.2439, 3.2056, 3.3438, 3.2788], + device='cuda:2'), covar=tensor([0.0506, 0.3419, 0.0469, 0.2686, 0.1306, 0.0740, 0.0582, 0.0662], + device='cuda:2'), in_proj_covar=tensor([0.0323, 0.0481, 0.0405, 0.0415, 0.0476, 0.0398, 0.0388, 0.0442], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 01:03:20,912 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:24,276 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36626.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:24,817 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 3.120e+02 3.802e+02 4.654e+02 9.583e+02, threshold=7.605e+02, percent-clipped=3.0 +2023-02-06 01:03:27,106 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:28,971 INFO [train.py:901] (2/4) Epoch 5, batch 4300, loss[loss=0.2516, simple_loss=0.3053, pruned_loss=0.09894, over 7286.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3518, pruned_loss=0.1149, over 1610615.26 frames. ], batch size: 16, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:03:42,648 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-06 01:03:47,180 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:52,222 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 01:04:05,000 INFO [train.py:901] (2/4) Epoch 5, batch 4350, loss[loss=0.3142, simple_loss=0.3513, pruned_loss=0.1386, over 7536.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3518, pruned_loss=0.1148, over 1607169.75 frames. ], batch size: 18, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:04:28,889 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36718.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:04:34,836 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 3.265e+02 4.032e+02 4.973e+02 1.053e+03, threshold=8.064e+02, percent-clipped=5.0 +2023-02-06 01:04:36,256 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 01:04:38,954 INFO [train.py:901] (2/4) Epoch 5, batch 4400, loss[loss=0.2504, simple_loss=0.3082, pruned_loss=0.09636, over 7282.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3514, pruned_loss=0.1149, over 1605546.59 frames. ], batch size: 16, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:04:56,225 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 01:05:15,304 INFO [train.py:901] (2/4) Epoch 5, batch 4450, loss[loss=0.2689, simple_loss=0.3325, pruned_loss=0.1027, over 8328.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3514, pruned_loss=0.1149, over 1606220.02 frames. ], batch size: 25, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:05:18,074 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 01:05:43,195 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36823.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:05:45,716 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 3.029e+02 3.648e+02 4.687e+02 9.435e+02, threshold=7.296e+02, percent-clipped=4.0 +2023-02-06 01:05:49,708 INFO [train.py:901] (2/4) Epoch 5, batch 4500, loss[loss=0.2918, simple_loss=0.3436, pruned_loss=0.1201, over 7694.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3503, pruned_loss=0.1141, over 1605875.23 frames. ], batch size: 18, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:06:14,972 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 01:06:25,763 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.67 vs. limit=5.0 +2023-02-06 01:06:26,110 INFO [train.py:901] (2/4) Epoch 5, batch 4550, loss[loss=0.2822, simple_loss=0.3607, pruned_loss=0.1018, over 8455.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3499, pruned_loss=0.1136, over 1608133.05 frames. ], batch size: 25, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:06:47,951 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36914.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:06:56,558 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 3.174e+02 3.779e+02 4.790e+02 8.988e+02, threshold=7.559e+02, percent-clipped=4.0 +2023-02-06 01:06:58,109 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:00,569 INFO [train.py:901] (2/4) Epoch 5, batch 4600, loss[loss=0.347, simple_loss=0.3829, pruned_loss=0.1556, over 8509.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3503, pruned_loss=0.1134, over 1613203.47 frames. ], batch size: 28, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:07:04,717 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:23,548 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:25,629 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36970.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:29,034 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:30,332 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36976.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:35,633 INFO [train.py:901] (2/4) Epoch 5, batch 4650, loss[loss=0.3363, simple_loss=0.3841, pruned_loss=0.1443, over 8475.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3504, pruned_loss=0.1144, over 1605553.08 frames. ], batch size: 49, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:08:02,861 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:08:06,067 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 3.207e+02 3.974e+02 5.163e+02 9.904e+02, threshold=7.949e+02, percent-clipped=4.0 +2023-02-06 01:08:10,735 INFO [train.py:901] (2/4) Epoch 5, batch 4700, loss[loss=0.1808, simple_loss=0.2634, pruned_loss=0.04907, over 7418.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3501, pruned_loss=0.1144, over 1603125.91 frames. ], batch size: 17, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:08:29,881 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37062.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:08:43,597 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:08:44,118 INFO [train.py:901] (2/4) Epoch 5, batch 4750, loss[loss=0.2726, simple_loss=0.3478, pruned_loss=0.09866, over 8484.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3509, pruned_loss=0.1151, over 1606045.08 frames. ], batch size: 26, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:08:45,691 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37085.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:08:49,139 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37089.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:09:08,550 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.4091, 5.4158, 4.8348, 2.1624, 4.7617, 5.0409, 5.0727, 4.5045], + device='cuda:2'), covar=tensor([0.0594, 0.0345, 0.0725, 0.4399, 0.0663, 0.0500, 0.0841, 0.0505], + device='cuda:2'), in_proj_covar=tensor([0.0399, 0.0290, 0.0324, 0.0406, 0.0323, 0.0270, 0.0305, 0.0247], + 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') +2023-02-06 01:09:15,970 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 3.010e+02 3.846e+02 4.879e+02 1.523e+03, threshold=7.692e+02, percent-clipped=5.0 +2023-02-06 01:09:17,400 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 01:09:20,175 INFO [train.py:901] (2/4) Epoch 5, batch 4800, loss[loss=0.3066, simple_loss=0.3637, pruned_loss=0.1247, over 8570.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3511, pruned_loss=0.1152, over 1611440.11 frames. ], batch size: 31, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:09:20,180 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 01:09:44,353 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37167.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:09:51,279 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37177.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:09:51,932 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3775, 1.4345, 2.2243, 1.0578, 2.1841, 2.4081, 2.3979, 2.0381], + device='cuda:2'), covar=tensor([0.1053, 0.1085, 0.0552, 0.1996, 0.0549, 0.0383, 0.0591, 0.0833], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0270, 0.0225, 0.0262, 0.0225, 0.0198, 0.0231, 0.0273], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 01:09:55,226 INFO [train.py:901] (2/4) Epoch 5, batch 4850, loss[loss=0.3262, simple_loss=0.3769, pruned_loss=0.1377, over 8414.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3502, pruned_loss=0.1144, over 1613742.72 frames. ], batch size: 48, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:10:10,652 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 01:10:12,995 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2023-02-06 01:10:27,460 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.956e+02 3.581e+02 4.871e+02 1.087e+03, threshold=7.163e+02, percent-clipped=6.0 +2023-02-06 01:10:31,481 INFO [train.py:901] (2/4) Epoch 5, batch 4900, loss[loss=0.3863, simple_loss=0.4125, pruned_loss=0.1801, over 6752.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3503, pruned_loss=0.1147, over 1610056.49 frames. ], batch size: 72, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:10:59,368 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37273.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:04,934 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9463, 3.7934, 2.4936, 2.1489, 2.6222, 2.2620, 2.2209, 2.7478], + device='cuda:2'), covar=tensor([0.1298, 0.0291, 0.0760, 0.0708, 0.0663, 0.0922, 0.1020, 0.0938], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0243, 0.0305, 0.0298, 0.0319, 0.0304, 0.0339, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 01:11:05,621 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37282.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:11:06,073 INFO [train.py:901] (2/4) Epoch 5, batch 4950, loss[loss=0.2874, simple_loss=0.3428, pruned_loss=0.116, over 8331.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3496, pruned_loss=0.1141, over 1606031.97 frames. ], batch size: 25, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:11:08,195 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:31,043 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37320.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:35,592 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 3.052e+02 3.616e+02 4.696e+02 1.143e+03, threshold=7.231e+02, percent-clipped=5.0 +2023-02-06 01:11:40,305 INFO [train.py:901] (2/4) Epoch 5, batch 5000, loss[loss=0.2345, simple_loss=0.3045, pruned_loss=0.08227, over 7805.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3497, pruned_loss=0.1136, over 1609943.77 frames. ], batch size: 19, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:11:43,901 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:45,945 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37341.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:49,268 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37345.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:01,327 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:03,196 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:03,313 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:05,952 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37370.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:14,485 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37382.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:15,020 INFO [train.py:901] (2/4) Epoch 5, batch 5050, loss[loss=0.2765, simple_loss=0.3349, pruned_loss=0.1091, over 7801.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3505, pruned_loss=0.1144, over 1614325.20 frames. ], batch size: 19, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:12:18,411 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:44,491 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 3.496e+02 4.122e+02 5.072e+02 9.522e+02, threshold=8.245e+02, percent-clipped=6.0 +2023-02-06 01:12:48,503 INFO [train.py:901] (2/4) Epoch 5, batch 5100, loss[loss=0.3242, simple_loss=0.3767, pruned_loss=0.1358, over 8425.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3514, pruned_loss=0.1149, over 1613704.31 frames. ], batch size: 49, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:12:48,515 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 01:12:48,718 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37433.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:12:49,925 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:13:06,546 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37458.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:13:11,708 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7869, 1.4222, 3.3763, 1.3746, 2.5047, 3.7443, 3.5916, 3.2364], + device='cuda:2'), covar=tensor([0.1110, 0.1412, 0.0342, 0.1862, 0.0612, 0.0211, 0.0345, 0.0577], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0274, 0.0227, 0.0264, 0.0226, 0.0199, 0.0232, 0.0279], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 01:13:22,246 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:13:23,387 INFO [train.py:901] (2/4) Epoch 5, batch 5150, loss[loss=0.3044, simple_loss=0.3578, pruned_loss=0.1255, over 7816.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3494, pruned_loss=0.1141, over 1613149.00 frames. ], batch size: 20, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:13:53,504 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.119e+02 3.138e+02 3.879e+02 5.454e+02 1.167e+03, threshold=7.757e+02, percent-clipped=4.0 +2023-02-06 01:13:57,568 INFO [train.py:901] (2/4) Epoch 5, batch 5200, loss[loss=0.3599, simple_loss=0.3957, pruned_loss=0.162, over 6717.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3494, pruned_loss=0.114, over 1608978.33 frames. ], batch size: 71, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:14:01,148 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37538.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:14:19,393 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37563.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:14:33,076 INFO [train.py:901] (2/4) Epoch 5, batch 5250, loss[loss=0.2624, simple_loss=0.3396, pruned_loss=0.09259, over 8133.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.35, pruned_loss=0.1141, over 1603641.99 frames. ], batch size: 22, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:14:40,789 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8626, 2.1190, 2.2596, 1.7328, 1.1775, 2.2159, 0.3775, 1.1931], + device='cuda:2'), covar=tensor([0.3867, 0.1888, 0.0965, 0.3172, 0.6800, 0.1048, 0.6850, 0.2978], + device='cuda:2'), in_proj_covar=tensor([0.0129, 0.0123, 0.0079, 0.0172, 0.0209, 0.0083, 0.0149, 0.0127], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:14:45,218 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 01:15:03,494 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 3.205e+02 3.781e+02 5.298e+02 9.083e+02, threshold=7.562e+02, percent-clipped=4.0 +2023-02-06 01:15:05,550 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:15:07,535 INFO [train.py:901] (2/4) Epoch 5, batch 5300, loss[loss=0.2922, simple_loss=0.3538, pruned_loss=0.1153, over 6823.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3491, pruned_loss=0.1128, over 1607822.16 frames. ], batch size: 15, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:15:15,193 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:15:32,585 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:15:42,453 INFO [train.py:901] (2/4) Epoch 5, batch 5350, loss[loss=0.3462, simple_loss=0.3906, pruned_loss=0.1508, over 8732.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3499, pruned_loss=0.1136, over 1612684.51 frames. ], batch size: 49, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:15:47,816 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:05,030 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37716.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:11,798 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:12,347 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 3.472e+02 4.206e+02 5.536e+02 1.524e+03, threshold=8.412e+02, percent-clipped=7.0 +2023-02-06 01:16:17,027 INFO [train.py:901] (2/4) Epoch 5, batch 5400, loss[loss=0.3083, simple_loss=0.3609, pruned_loss=0.1278, over 8499.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3517, pruned_loss=0.1155, over 1612699.56 frames. ], batch size: 26, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:16:19,920 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:25,230 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:36,758 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37762.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:50,838 INFO [train.py:901] (2/4) Epoch 5, batch 5450, loss[loss=0.2707, simple_loss=0.3153, pruned_loss=0.113, over 7696.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3486, pruned_loss=0.1127, over 1607838.91 frames. ], batch size: 18, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:17:22,464 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.936e+02 3.882e+02 5.021e+02 1.156e+03, threshold=7.764e+02, percent-clipped=3.0 +2023-02-06 01:17:26,498 INFO [train.py:901] (2/4) Epoch 5, batch 5500, loss[loss=0.3139, simple_loss=0.3798, pruned_loss=0.124, over 8476.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3496, pruned_loss=0.1126, over 1611808.78 frames. ], batch size: 25, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:17:30,497 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 01:17:32,018 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:18:00,732 INFO [train.py:901] (2/4) Epoch 5, batch 5550, loss[loss=0.3861, simple_loss=0.4318, pruned_loss=0.1702, over 8464.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3499, pruned_loss=0.1134, over 1609741.76 frames. ], batch size: 27, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:18:32,131 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 3.124e+02 3.941e+02 5.093e+02 9.977e+02, threshold=7.882e+02, percent-clipped=4.0 +2023-02-06 01:18:36,201 INFO [train.py:901] (2/4) Epoch 5, batch 5600, loss[loss=0.2917, simple_loss=0.3433, pruned_loss=0.12, over 7823.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3504, pruned_loss=0.114, over 1613593.54 frames. ], batch size: 20, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:18:39,364 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 01:19:10,749 INFO [train.py:901] (2/4) Epoch 5, batch 5650, loss[loss=0.288, simple_loss=0.3481, pruned_loss=0.1139, over 8238.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3488, pruned_loss=0.113, over 1611346.32 frames. ], batch size: 22, lr: 1.47e-02, grad_scale: 4.0 +2023-02-06 01:19:20,217 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37997.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:19:23,903 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:19:34,403 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 01:19:41,409 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:19:42,550 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 3.005e+02 3.717e+02 4.758e+02 1.120e+03, threshold=7.434e+02, percent-clipped=3.0 +2023-02-06 01:19:45,907 INFO [train.py:901] (2/4) Epoch 5, batch 5700, loss[loss=0.2466, simple_loss=0.3148, pruned_loss=0.08919, over 7204.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.349, pruned_loss=0.1138, over 1607092.40 frames. ], batch size: 16, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:20:20,736 INFO [train.py:901] (2/4) Epoch 5, batch 5750, loss[loss=0.2842, simple_loss=0.3395, pruned_loss=0.1144, over 8243.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3502, pruned_loss=0.1144, over 1607233.47 frames. ], batch size: 22, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:20:20,968 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4300, 1.8084, 1.9554, 1.3111, 0.9500, 1.9720, 0.2127, 1.0927], + device='cuda:2'), covar=tensor([0.4668, 0.1813, 0.0929, 0.3247, 0.6763, 0.0770, 0.6319, 0.2573], + device='cuda:2'), in_proj_covar=tensor([0.0130, 0.0120, 0.0079, 0.0171, 0.0210, 0.0079, 0.0145, 0.0124], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:20:30,238 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:20:37,194 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 01:20:41,417 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0248, 1.0497, 3.2734, 0.8666, 2.7250, 2.7266, 2.8508, 2.7704], + device='cuda:2'), covar=tensor([0.0652, 0.3440, 0.0553, 0.2690, 0.1427, 0.0785, 0.0645, 0.0798], + device='cuda:2'), in_proj_covar=tensor([0.0340, 0.0498, 0.0417, 0.0433, 0.0491, 0.0414, 0.0406, 0.0453], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 01:20:45,514 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5570, 1.7964, 3.4252, 1.1259, 2.2221, 1.8345, 1.5029, 1.9347], + device='cuda:2'), covar=tensor([0.1526, 0.1874, 0.0611, 0.3356, 0.1441, 0.2408, 0.1546, 0.2245], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0445, 0.0523, 0.0538, 0.0576, 0.0517, 0.0446, 0.0589], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 01:20:46,797 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:20:50,581 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 3.053e+02 3.876e+02 4.925e+02 1.023e+03, threshold=7.752e+02, percent-clipped=4.0 +2023-02-06 01:20:54,528 INFO [train.py:901] (2/4) Epoch 5, batch 5800, loss[loss=0.2536, simple_loss=0.3314, pruned_loss=0.08794, over 8293.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3492, pruned_loss=0.1132, over 1610191.51 frames. ], batch size: 23, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:21:29,404 INFO [train.py:901] (2/4) Epoch 5, batch 5850, loss[loss=0.3158, simple_loss=0.3781, pruned_loss=0.1268, over 8293.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3494, pruned_loss=0.1134, over 1612480.88 frames. ], batch size: 49, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:22:00,285 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 01:22:01,294 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 3.016e+02 3.759e+02 4.889e+02 1.185e+03, threshold=7.518e+02, percent-clipped=2.0 +2023-02-06 01:22:02,502 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 01:22:04,818 INFO [train.py:901] (2/4) Epoch 5, batch 5900, loss[loss=0.2444, simple_loss=0.328, pruned_loss=0.08045, over 8439.00 frames. ], tot_loss[loss=0.287, simple_loss=0.349, pruned_loss=0.1125, over 1609737.10 frames. ], batch size: 27, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:22:40,895 INFO [train.py:901] (2/4) Epoch 5, batch 5950, loss[loss=0.2627, simple_loss=0.3185, pruned_loss=0.1034, over 7254.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.349, pruned_loss=0.1124, over 1612827.24 frames. ], batch size: 16, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:23:04,141 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 +2023-02-06 01:23:12,021 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.179e+02 3.191e+02 3.790e+02 5.332e+02 1.075e+03, threshold=7.580e+02, percent-clipped=7.0 +2023-02-06 01:23:14,240 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0110, 2.5288, 1.9172, 2.9868, 1.4930, 1.6865, 1.7357, 2.3544], + device='cuda:2'), covar=tensor([0.0947, 0.0844, 0.1236, 0.0490, 0.1561, 0.1713, 0.1679, 0.1153], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0256, 0.0278, 0.0219, 0.0246, 0.0278, 0.0290, 0.0262], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 01:23:15,478 INFO [train.py:901] (2/4) Epoch 5, batch 6000, loss[loss=0.2818, simple_loss=0.3459, pruned_loss=0.1088, over 7969.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3494, pruned_loss=0.1123, over 1618832.42 frames. ], batch size: 21, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:23:15,478 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 01:23:28,274 INFO [train.py:935] (2/4) Epoch 5, validation: loss=0.2196, simple_loss=0.3162, pruned_loss=0.06146, over 944034.00 frames. +2023-02-06 01:23:28,275 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 01:23:33,797 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38341.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:23:58,477 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38378.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:24:01,709 INFO [train.py:901] (2/4) Epoch 5, batch 6050, loss[loss=0.2921, simple_loss=0.3317, pruned_loss=0.1263, over 6822.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.348, pruned_loss=0.1117, over 1617246.16 frames. ], batch size: 15, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:24:33,770 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 3.108e+02 3.868e+02 4.827e+02 8.119e+02, threshold=7.737e+02, percent-clipped=1.0 +2023-02-06 01:24:37,062 INFO [train.py:901] (2/4) Epoch 5, batch 6100, loss[loss=0.2657, simple_loss=0.329, pruned_loss=0.1012, over 8231.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3486, pruned_loss=0.1121, over 1619274.59 frames. ], batch size: 22, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:24:40,778 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2023-02-06 01:24:53,393 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38456.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:24:54,054 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4377, 2.4369, 1.4411, 3.1499, 1.4932, 1.2381, 2.0167, 2.2393], + device='cuda:2'), covar=tensor([0.2386, 0.1796, 0.3004, 0.0439, 0.2279, 0.3321, 0.1961, 0.1536], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0254, 0.0280, 0.0219, 0.0249, 0.0281, 0.0291, 0.0268], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 01:24:57,956 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7503, 1.5041, 3.2083, 1.2603, 2.1563, 3.5729, 3.4278, 3.0728], + device='cuda:2'), covar=tensor([0.1152, 0.1460, 0.0371, 0.2063, 0.0778, 0.0290, 0.0394, 0.0628], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0269, 0.0224, 0.0263, 0.0225, 0.0201, 0.0232, 0.0276], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 01:25:09,662 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 01:25:10,932 INFO [train.py:901] (2/4) Epoch 5, batch 6150, loss[loss=0.2663, simple_loss=0.3401, pruned_loss=0.09622, over 8476.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3481, pruned_loss=0.112, over 1620259.62 frames. ], batch size: 25, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:25:32,112 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.29 vs. limit=5.0 +2023-02-06 01:25:42,282 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 3.089e+02 4.008e+02 5.119e+02 1.011e+03, threshold=8.016e+02, percent-clipped=7.0 +2023-02-06 01:25:45,617 INFO [train.py:901] (2/4) Epoch 5, batch 6200, loss[loss=0.3141, simple_loss=0.3696, pruned_loss=0.1294, over 8800.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3465, pruned_loss=0.1111, over 1617467.25 frames. ], batch size: 40, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:26:20,117 INFO [train.py:901] (2/4) Epoch 5, batch 6250, loss[loss=0.2957, simple_loss=0.3539, pruned_loss=0.1188, over 8589.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3458, pruned_loss=0.1111, over 1617450.57 frames. ], batch size: 31, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:26:51,177 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.140e+02 3.239e+02 3.994e+02 4.997e+02 1.061e+03, threshold=7.988e+02, percent-clipped=3.0 +2023-02-06 01:26:54,599 INFO [train.py:901] (2/4) Epoch 5, batch 6300, loss[loss=0.279, simple_loss=0.3357, pruned_loss=0.1112, over 8088.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3467, pruned_loss=0.1116, over 1616296.64 frames. ], batch size: 21, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:27:29,613 INFO [train.py:901] (2/4) Epoch 5, batch 6350, loss[loss=0.2889, simple_loss=0.3375, pruned_loss=0.1201, over 8074.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3453, pruned_loss=0.111, over 1611710.91 frames. ], batch size: 21, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:27:49,930 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38712.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:27:56,422 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:28:00,326 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.862e+02 3.826e+02 4.732e+02 1.596e+03, threshold=7.652e+02, percent-clipped=5.0 +2023-02-06 01:28:03,611 INFO [train.py:901] (2/4) Epoch 5, batch 6400, loss[loss=0.3929, simple_loss=0.4177, pruned_loss=0.1841, over 6874.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.345, pruned_loss=0.1106, over 1613758.15 frames. ], batch size: 71, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:28:06,533 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:28:22,356 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4437, 4.3965, 3.9500, 1.6525, 3.8878, 3.9693, 4.2072, 3.6925], + device='cuda:2'), covar=tensor([0.0870, 0.0552, 0.0947, 0.5152, 0.0722, 0.0933, 0.0964, 0.0761], + device='cuda:2'), in_proj_covar=tensor([0.0399, 0.0294, 0.0322, 0.0406, 0.0310, 0.0272, 0.0303, 0.0253], + 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') +2023-02-06 01:28:38,892 INFO [train.py:901] (2/4) Epoch 5, batch 6450, loss[loss=0.3068, simple_loss=0.3665, pruned_loss=0.1235, over 8425.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3471, pruned_loss=0.1116, over 1618536.23 frames. ], batch size: 27, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:29:09,816 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.470e+02 3.536e+02 4.141e+02 5.010e+02 9.096e+02, threshold=8.281e+02, percent-clipped=4.0 +2023-02-06 01:29:11,283 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9232, 3.4507, 2.7055, 4.3660, 1.9228, 2.1786, 2.6657, 3.7336], + device='cuda:2'), covar=tensor([0.0845, 0.0983, 0.1190, 0.0231, 0.1591, 0.1794, 0.1443, 0.0697], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0252, 0.0282, 0.0220, 0.0243, 0.0277, 0.0282, 0.0254], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 01:29:13,134 INFO [train.py:901] (2/4) Epoch 5, batch 6500, loss[loss=0.2643, simple_loss=0.3373, pruned_loss=0.09564, over 8450.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3473, pruned_loss=0.1123, over 1614844.00 frames. ], batch size: 25, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:29:16,037 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38837.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:29:18,076 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4501, 2.0606, 3.3164, 2.7674, 2.6455, 2.1361, 1.5185, 1.3394], + device='cuda:2'), covar=tensor([0.1948, 0.2172, 0.0452, 0.1067, 0.1124, 0.1081, 0.1161, 0.2221], + device='cuda:2'), in_proj_covar=tensor([0.0745, 0.0674, 0.0577, 0.0660, 0.0766, 0.0635, 0.0604, 0.0625], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:29:18,660 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:29:20,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1115, 1.1151, 1.0517, 1.0715, 0.8195, 1.1907, 0.0785, 0.8410], + device='cuda:2'), covar=tensor([0.3097, 0.2269, 0.1060, 0.1857, 0.5375, 0.0848, 0.4470, 0.2478], + device='cuda:2'), in_proj_covar=tensor([0.0128, 0.0124, 0.0080, 0.0166, 0.0209, 0.0078, 0.0140, 0.0125], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:29:48,036 INFO [train.py:901] (2/4) Epoch 5, batch 6550, loss[loss=0.2904, simple_loss=0.3691, pruned_loss=0.1058, over 8334.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3476, pruned_loss=0.1126, over 1612181.77 frames. ], batch size: 25, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:29:49,535 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0713, 1.2247, 4.2352, 1.6192, 3.6956, 3.5549, 3.8199, 3.7151], + device='cuda:2'), covar=tensor([0.0453, 0.3706, 0.0421, 0.2435, 0.1032, 0.0593, 0.0431, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0332, 0.0481, 0.0416, 0.0422, 0.0481, 0.0401, 0.0399, 0.0442], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-06 01:30:06,593 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9826, 1.4298, 1.3352, 1.0755, 1.1372, 1.3230, 1.5178, 1.4728], + device='cuda:2'), covar=tensor([0.0598, 0.1252, 0.1952, 0.1579, 0.0627, 0.1531, 0.0744, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0135, 0.0178, 0.0220, 0.0183, 0.0130, 0.0187, 0.0142, 0.0153], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 01:30:19,275 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 3.239e+02 3.737e+02 4.952e+02 1.438e+03, threshold=7.474e+02, percent-clipped=4.0 +2023-02-06 01:30:22,012 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 01:30:22,676 INFO [train.py:901] (2/4) Epoch 5, batch 6600, loss[loss=0.2855, simple_loss=0.3496, pruned_loss=0.1107, over 8327.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3484, pruned_loss=0.1138, over 1613202.23 frames. ], batch size: 26, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:30:39,818 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 01:30:51,529 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5834, 2.4245, 1.5581, 2.1251, 1.9352, 1.2992, 1.7736, 1.8729], + device='cuda:2'), covar=tensor([0.0902, 0.0267, 0.0924, 0.0443, 0.0624, 0.1219, 0.0730, 0.0695], + device='cuda:2'), in_proj_covar=tensor([0.0341, 0.0235, 0.0305, 0.0299, 0.0315, 0.0309, 0.0335, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 01:30:58,095 INFO [train.py:901] (2/4) Epoch 5, batch 6650, loss[loss=0.284, simple_loss=0.3539, pruned_loss=0.1071, over 8031.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3478, pruned_loss=0.1124, over 1616385.15 frames. ], batch size: 22, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:31:29,853 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 3.068e+02 3.660e+02 4.252e+02 1.265e+03, threshold=7.321e+02, percent-clipped=3.0 +2023-02-06 01:31:33,318 INFO [train.py:901] (2/4) Epoch 5, batch 6700, loss[loss=0.2514, simple_loss=0.3242, pruned_loss=0.08925, over 8140.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3473, pruned_loss=0.1117, over 1617823.59 frames. ], batch size: 22, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:31:49,152 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0110, 2.4435, 3.9997, 3.1168, 3.3226, 2.3161, 1.7759, 1.9957], + device='cuda:2'), covar=tensor([0.1669, 0.2039, 0.0479, 0.1101, 0.0927, 0.1012, 0.1038, 0.2179], + device='cuda:2'), in_proj_covar=tensor([0.0753, 0.0683, 0.0582, 0.0675, 0.0786, 0.0648, 0.0615, 0.0638], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:32:08,096 INFO [train.py:901] (2/4) Epoch 5, batch 6750, loss[loss=0.3047, simple_loss=0.3601, pruned_loss=0.1247, over 8462.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3468, pruned_loss=0.1116, over 1615341.43 frames. ], batch size: 27, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:32:15,841 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:32:26,943 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4290, 1.9214, 3.6853, 1.0743, 2.5999, 1.7326, 1.4573, 2.0980], + device='cuda:2'), covar=tensor([0.1719, 0.1827, 0.0546, 0.3411, 0.1294, 0.2417, 0.1642, 0.2285], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0448, 0.0516, 0.0533, 0.0570, 0.0509, 0.0442, 0.0587], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 01:32:33,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39118.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:32:40,320 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.802e+02 3.437e+02 4.300e+02 8.945e+02, threshold=6.874e+02, percent-clipped=2.0 +2023-02-06 01:32:43,709 INFO [train.py:901] (2/4) Epoch 5, batch 6800, loss[loss=0.2925, simple_loss=0.3411, pruned_loss=0.1219, over 7445.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.347, pruned_loss=0.112, over 1615449.87 frames. ], batch size: 17, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:32:49,774 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5331, 1.7730, 2.1406, 1.0631, 2.2269, 1.3290, 0.6737, 1.5874], + device='cuda:2'), covar=tensor([0.0214, 0.0111, 0.0079, 0.0185, 0.0148, 0.0360, 0.0293, 0.0117], + device='cuda:2'), in_proj_covar=tensor([0.0317, 0.0226, 0.0194, 0.0273, 0.0220, 0.0363, 0.0286, 0.0268], + device='cuda:2'), out_proj_covar=tensor([1.1177e-04, 7.7858e-05, 6.6506e-05, 9.3822e-05, 7.7362e-05, 1.3607e-04, + 1.0147e-04, 9.3531e-05], device='cuda:2') +2023-02-06 01:32:52,269 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3078, 1.3710, 1.4540, 1.1433, 1.3539, 1.3285, 1.7211, 1.6401], + device='cuda:2'), covar=tensor([0.0543, 0.1300, 0.1803, 0.1442, 0.0588, 0.1600, 0.0699, 0.0533], + device='cuda:2'), in_proj_covar=tensor([0.0135, 0.0177, 0.0220, 0.0182, 0.0129, 0.0186, 0.0142, 0.0152], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 01:32:54,769 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 01:33:17,603 INFO [train.py:901] (2/4) Epoch 5, batch 6850, loss[loss=0.3004, simple_loss=0.3568, pruned_loss=0.122, over 7933.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3495, pruned_loss=0.1136, over 1619619.59 frames. ], batch size: 20, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:33:19,089 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:33:31,225 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:33:43,505 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 01:33:48,728 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 3.284e+02 3.960e+02 5.468e+02 1.321e+03, threshold=7.919e+02, percent-clipped=11.0 +2023-02-06 01:33:52,162 INFO [train.py:901] (2/4) Epoch 5, batch 6900, loss[loss=0.2903, simple_loss=0.3681, pruned_loss=0.1063, over 8495.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3503, pruned_loss=0.1143, over 1619730.83 frames. ], batch size: 26, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:33:59,429 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1955, 4.2106, 3.7824, 1.5850, 3.7202, 3.4965, 3.8313, 3.2372], + device='cuda:2'), covar=tensor([0.0747, 0.0533, 0.0912, 0.4701, 0.0757, 0.0994, 0.1136, 0.0922], + device='cuda:2'), in_proj_covar=tensor([0.0392, 0.0290, 0.0317, 0.0399, 0.0305, 0.0277, 0.0302, 0.0247], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 01:34:12,819 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0306, 1.1820, 4.2111, 1.5779, 3.6762, 3.4824, 3.7612, 3.6988], + device='cuda:2'), covar=tensor([0.0492, 0.3948, 0.0494, 0.2653, 0.1258, 0.0789, 0.0488, 0.0555], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0499, 0.0426, 0.0428, 0.0498, 0.0409, 0.0404, 0.0457], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 01:34:26,651 INFO [train.py:901] (2/4) Epoch 5, batch 6950, loss[loss=0.3508, simple_loss=0.4021, pruned_loss=0.1498, over 8358.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3509, pruned_loss=0.1141, over 1619854.26 frames. ], batch size: 24, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:34:38,150 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39300.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:34:39,409 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:34:49,520 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 01:34:51,219 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 01:34:58,303 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 3.231e+02 3.801e+02 5.196e+02 1.038e+03, threshold=7.603e+02, percent-clipped=4.0 +2023-02-06 01:35:01,662 INFO [train.py:901] (2/4) Epoch 5, batch 7000, loss[loss=0.2992, simple_loss=0.3715, pruned_loss=0.1135, over 8471.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3494, pruned_loss=0.1132, over 1616863.94 frames. ], batch size: 25, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:35:35,821 INFO [train.py:901] (2/4) Epoch 5, batch 7050, loss[loss=0.3119, simple_loss=0.3786, pruned_loss=0.1226, over 8248.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3485, pruned_loss=0.1122, over 1615662.30 frames. ], batch size: 24, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:35:58,318 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8433, 1.4641, 5.8037, 2.1309, 5.2647, 4.9002, 5.4377, 5.3783], + device='cuda:2'), covar=tensor([0.0242, 0.3612, 0.0184, 0.2081, 0.0774, 0.0519, 0.0305, 0.0357], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0501, 0.0436, 0.0432, 0.0507, 0.0420, 0.0415, 0.0467], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 01:36:06,897 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 2.867e+02 3.538e+02 4.706e+02 1.662e+03, threshold=7.075e+02, percent-clipped=2.0 +2023-02-06 01:36:10,285 INFO [train.py:901] (2/4) Epoch 5, batch 7100, loss[loss=0.2752, simple_loss=0.3251, pruned_loss=0.1127, over 7708.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3484, pruned_loss=0.1121, over 1616098.76 frames. ], batch size: 18, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:36:34,777 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6640, 2.0659, 3.6868, 1.0990, 2.7351, 1.9537, 1.6620, 2.1974], + device='cuda:2'), covar=tensor([0.1376, 0.1592, 0.0513, 0.3173, 0.1129, 0.2137, 0.1348, 0.1982], + device='cuda:2'), in_proj_covar=tensor([0.0465, 0.0447, 0.0515, 0.0526, 0.0565, 0.0505, 0.0442, 0.0584], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 01:36:44,799 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:36:45,999 INFO [train.py:901] (2/4) Epoch 5, batch 7150, loss[loss=0.2986, simple_loss=0.3623, pruned_loss=0.1174, over 8604.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3499, pruned_loss=0.113, over 1619399.75 frames. ], batch size: 39, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:37:17,185 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.912e+02 3.907e+02 4.774e+02 1.202e+03, threshold=7.813e+02, percent-clipped=7.0 +2023-02-06 01:37:20,760 INFO [train.py:901] (2/4) Epoch 5, batch 7200, loss[loss=0.3073, simple_loss=0.3666, pruned_loss=0.1239, over 8625.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3509, pruned_loss=0.1139, over 1618532.65 frames. ], batch size: 39, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:37:30,594 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:37:37,496 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:37:55,143 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:37:56,306 INFO [train.py:901] (2/4) Epoch 5, batch 7250, loss[loss=0.2382, simple_loss=0.3038, pruned_loss=0.08625, over 7679.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3487, pruned_loss=0.1127, over 1613170.49 frames. ], batch size: 18, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:38:13,473 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1570, 3.2654, 2.4570, 4.2173, 2.0503, 2.1025, 2.3496, 3.2345], + device='cuda:2'), covar=tensor([0.0747, 0.0980, 0.1314, 0.0279, 0.1580, 0.1790, 0.1864, 0.1013], + device='cuda:2'), in_proj_covar=tensor([0.0276, 0.0254, 0.0290, 0.0231, 0.0252, 0.0283, 0.0289, 0.0256], + 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') +2023-02-06 01:38:27,204 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 2.862e+02 3.679e+02 5.056e+02 1.142e+03, threshold=7.358e+02, percent-clipped=8.0 +2023-02-06 01:38:30,501 INFO [train.py:901] (2/4) Epoch 5, batch 7300, loss[loss=0.3206, simple_loss=0.3796, pruned_loss=0.1308, over 8653.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3488, pruned_loss=0.1128, over 1612915.86 frames. ], batch size: 34, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:38:39,559 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:38:47,110 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39657.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:38:50,516 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39662.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:39:05,558 INFO [train.py:901] (2/4) Epoch 5, batch 7350, loss[loss=0.2534, simple_loss=0.3137, pruned_loss=0.09653, over 7711.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3498, pruned_loss=0.1139, over 1612740.69 frames. ], batch size: 18, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:39:13,007 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1713, 1.6052, 1.6106, 1.1736, 1.1806, 1.4239, 1.6253, 1.5929], + device='cuda:2'), covar=tensor([0.0609, 0.1251, 0.1769, 0.1565, 0.0689, 0.1657, 0.0833, 0.0597], + device='cuda:2'), in_proj_covar=tensor([0.0134, 0.0177, 0.0218, 0.0182, 0.0131, 0.0188, 0.0143, 0.0151], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 01:39:15,966 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-06 01:39:16,374 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6767, 5.6475, 5.0564, 2.2453, 5.0353, 5.4727, 5.1733, 5.0050], + device='cuda:2'), covar=tensor([0.0550, 0.0451, 0.0824, 0.4451, 0.0626, 0.0504, 0.1030, 0.0523], + device='cuda:2'), in_proj_covar=tensor([0.0394, 0.0284, 0.0312, 0.0402, 0.0303, 0.0271, 0.0294, 0.0244], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 01:39:17,818 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0023, 2.3732, 1.8675, 2.8849, 1.5516, 1.4439, 1.6793, 2.3607], + device='cuda:2'), covar=tensor([0.1106, 0.1041, 0.1561, 0.0604, 0.1611, 0.2117, 0.1854, 0.1033], + device='cuda:2'), in_proj_covar=tensor([0.0276, 0.0254, 0.0288, 0.0228, 0.0249, 0.0282, 0.0289, 0.0255], + 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') +2023-02-06 01:39:27,328 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 01:39:33,464 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 01:39:36,162 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 2.846e+02 3.982e+02 4.999e+02 1.878e+03, threshold=7.964e+02, percent-clipped=11.0 +2023-02-06 01:39:39,632 INFO [train.py:901] (2/4) Epoch 5, batch 7400, loss[loss=0.2453, simple_loss=0.3142, pruned_loss=0.08817, over 7922.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3502, pruned_loss=0.1139, over 1614042.31 frames. ], batch size: 20, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:39:52,999 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 01:39:55,398 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 01:39:59,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:40:13,745 INFO [train.py:901] (2/4) Epoch 5, batch 7450, loss[loss=0.2509, simple_loss=0.3269, pruned_loss=0.08748, over 8245.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3497, pruned_loss=0.1131, over 1612384.03 frames. ], batch size: 24, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:40:15,958 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:40:28,289 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8433, 5.8414, 5.1236, 2.2412, 5.1013, 5.4809, 5.4629, 5.3616], + device='cuda:2'), covar=tensor([0.0553, 0.0490, 0.0888, 0.4730, 0.0681, 0.0468, 0.0937, 0.0585], + device='cuda:2'), in_proj_covar=tensor([0.0396, 0.0285, 0.0315, 0.0405, 0.0305, 0.0274, 0.0296, 0.0247], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:2') +2023-02-06 01:40:32,193 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 01:40:43,636 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:40:45,624 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 3.116e+02 3.779e+02 4.440e+02 1.107e+03, threshold=7.558e+02, percent-clipped=3.0 +2023-02-06 01:40:49,054 INFO [train.py:901] (2/4) Epoch 5, batch 7500, loss[loss=0.2735, simple_loss=0.3478, pruned_loss=0.09957, over 8620.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3497, pruned_loss=0.113, over 1614100.12 frames. ], batch size: 31, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:40:49,198 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:41:22,984 INFO [train.py:901] (2/4) Epoch 5, batch 7550, loss[loss=0.266, simple_loss=0.3197, pruned_loss=0.1061, over 7544.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3485, pruned_loss=0.113, over 1610727.57 frames. ], batch size: 18, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:41:48,010 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:41:54,547 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 3.144e+02 3.978e+02 5.379e+02 1.554e+03, threshold=7.957e+02, percent-clipped=6.0 +2023-02-06 01:41:57,848 INFO [train.py:901] (2/4) Epoch 5, batch 7600, loss[loss=0.2636, simple_loss=0.3259, pruned_loss=0.1006, over 7652.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.347, pruned_loss=0.1117, over 1607798.05 frames. ], batch size: 19, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:42:02,463 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:42:04,534 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39943.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:42:32,756 INFO [train.py:901] (2/4) Epoch 5, batch 7650, loss[loss=0.2848, simple_loss=0.3411, pruned_loss=0.1142, over 8359.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3467, pruned_loss=0.1122, over 1606024.46 frames. ], batch size: 24, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:42:33,867 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 01:42:45,886 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40001.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:42:49,489 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2023-02-06 01:42:56,997 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40017.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:43:05,593 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.281e+02 3.028e+02 3.689e+02 4.703e+02 1.290e+03, threshold=7.379e+02, percent-clipped=1.0 +2023-02-06 01:43:08,870 INFO [train.py:901] (2/4) Epoch 5, batch 7700, loss[loss=0.3034, simple_loss=0.3653, pruned_loss=0.1207, over 8509.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3463, pruned_loss=0.1117, over 1610215.06 frames. ], batch size: 26, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:43:15,432 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:43:30,101 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 01:43:37,803 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 01:43:44,273 INFO [train.py:901] (2/4) Epoch 5, batch 7750, loss[loss=0.2737, simple_loss=0.3495, pruned_loss=0.09897, over 8331.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3463, pruned_loss=0.1116, over 1611169.88 frames. ], batch size: 26, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:43:44,300 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 01:43:48,564 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3508, 2.2095, 1.5427, 1.9220, 1.7717, 1.2813, 1.8092, 1.6687], + device='cuda:2'), covar=tensor([0.0959, 0.0290, 0.0829, 0.0405, 0.0562, 0.1069, 0.0624, 0.0619], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0248, 0.0316, 0.0305, 0.0320, 0.0319, 0.0339, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 01:44:07,435 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40116.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:44:08,908 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 01:44:15,426 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.050e+02 3.016e+02 3.638e+02 4.428e+02 8.911e+02, threshold=7.276e+02, percent-clipped=8.0 +2023-02-06 01:44:16,231 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:44:18,943 INFO [train.py:901] (2/4) Epoch 5, batch 7800, loss[loss=0.2864, simple_loss=0.3485, pruned_loss=0.1122, over 8561.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3491, pruned_loss=0.1133, over 1615931.26 frames. ], batch size: 31, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:44:50,153 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40177.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:44:51,119 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-06 01:44:54,262 INFO [train.py:901] (2/4) Epoch 5, batch 7850, loss[loss=0.3014, simple_loss=0.3645, pruned_loss=0.1192, over 8659.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3478, pruned_loss=0.1122, over 1613487.89 frames. ], batch size: 39, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:45:03,202 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40196.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:14,679 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:20,118 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:24,747 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.315e+02 3.285e+02 3.978e+02 4.753e+02 1.108e+03, threshold=7.955e+02, percent-clipped=4.0 +2023-02-06 01:45:28,277 INFO [train.py:901] (2/4) Epoch 5, batch 7900, loss[loss=0.2562, simple_loss=0.3108, pruned_loss=0.1008, over 7799.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.348, pruned_loss=0.1127, over 1610365.16 frames. ], batch size: 19, lr: 1.42e-02, grad_scale: 16.0 +2023-02-06 01:45:30,495 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40236.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:35,934 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40244.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:41,990 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8245, 1.5188, 5.7363, 2.0561, 5.1582, 4.8605, 5.3342, 5.2270], + device='cuda:2'), covar=tensor([0.0317, 0.3550, 0.0257, 0.2338, 0.0907, 0.0593, 0.0371, 0.0377], + device='cuda:2'), in_proj_covar=tensor([0.0337, 0.0491, 0.0434, 0.0428, 0.0496, 0.0414, 0.0399, 0.0457], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 01:46:02,026 INFO [train.py:901] (2/4) Epoch 5, batch 7950, loss[loss=0.2501, simple_loss=0.3149, pruned_loss=0.09264, over 7708.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3478, pruned_loss=0.1122, over 1610583.08 frames. ], batch size: 18, lr: 1.42e-02, grad_scale: 16.0 +2023-02-06 01:46:07,558 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40290.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:46:09,028 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40292.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:46:33,072 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.942e+02 3.003e+02 3.931e+02 4.743e+02 9.937e+02, threshold=7.862e+02, percent-clipped=4.0 +2023-02-06 01:46:36,390 INFO [train.py:901] (2/4) Epoch 5, batch 8000, loss[loss=0.2782, simple_loss=0.3386, pruned_loss=0.1089, over 8031.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3482, pruned_loss=0.1125, over 1612445.96 frames. ], batch size: 22, lr: 1.42e-02, grad_scale: 16.0 +2023-02-06 01:46:46,581 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40348.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:47:03,145 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40372.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:47:10,475 INFO [train.py:901] (2/4) Epoch 5, batch 8050, loss[loss=0.2338, simple_loss=0.2938, pruned_loss=0.08691, over 7569.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3439, pruned_loss=0.1104, over 1604590.18 frames. ], batch size: 18, lr: 1.42e-02, grad_scale: 8.0 +2023-02-06 01:47:20,234 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40397.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:47:23,377 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 +2023-02-06 01:47:44,060 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 01:47:48,097 INFO [train.py:901] (2/4) Epoch 6, batch 0, loss[loss=0.3028, simple_loss=0.3503, pruned_loss=0.1276, over 7928.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3503, pruned_loss=0.1276, over 7928.00 frames. ], batch size: 20, lr: 1.33e-02, grad_scale: 8.0 +2023-02-06 01:47:48,097 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 01:47:59,054 INFO [train.py:935] (2/4) Epoch 6, validation: loss=0.2203, simple_loss=0.3165, pruned_loss=0.06206, over 944034.00 frames. +2023-02-06 01:47:59,054 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 01:48:07,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 3.052e+02 3.992e+02 5.098e+02 1.227e+03, threshold=7.983e+02, percent-clipped=7.0 +2023-02-06 01:48:13,426 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 01:48:26,724 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 01:48:34,118 INFO [train.py:901] (2/4) Epoch 6, batch 50, loss[loss=0.2769, simple_loss=0.3549, pruned_loss=0.09941, over 8323.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3485, pruned_loss=0.1135, over 361333.23 frames. ], batch size: 25, lr: 1.33e-02, grad_scale: 8.0 +2023-02-06 01:48:38,478 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 01:48:48,496 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 01:48:57,360 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40500.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:04,768 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:08,799 INFO [train.py:901] (2/4) Epoch 6, batch 100, loss[loss=0.283, simple_loss=0.3504, pruned_loss=0.1078, over 8349.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3452, pruned_loss=0.1093, over 641303.44 frames. ], batch size: 24, lr: 1.33e-02, grad_scale: 8.0 +2023-02-06 01:49:13,091 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 01:49:15,330 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40525.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:17,934 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.877e+02 3.627e+02 4.294e+02 7.601e+02, threshold=7.253e+02, percent-clipped=0.0 +2023-02-06 01:49:31,500 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:37,659 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:42,465 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3279, 2.0225, 3.2719, 2.5182, 2.6063, 2.0438, 1.5086, 1.4230], + device='cuda:2'), covar=tensor([0.1909, 0.2069, 0.0465, 0.1116, 0.1014, 0.1083, 0.1100, 0.2065], + device='cuda:2'), in_proj_covar=tensor([0.0770, 0.0707, 0.0600, 0.0693, 0.0798, 0.0662, 0.0627, 0.0654], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 01:49:44,247 INFO [train.py:901] (2/4) Epoch 6, batch 150, loss[loss=0.2582, simple_loss=0.3283, pruned_loss=0.09403, over 8781.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3436, pruned_loss=0.108, over 857818.61 frames. ], batch size: 30, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:49:49,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:54,328 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:50:19,246 INFO [train.py:901] (2/4) Epoch 6, batch 200, loss[loss=0.2484, simple_loss=0.3332, pruned_loss=0.08178, over 8521.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3436, pruned_loss=0.1077, over 1031245.66 frames. ], batch size: 28, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:50:28,759 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 3.079e+02 3.898e+02 5.213e+02 9.157e+02, threshold=7.795e+02, percent-clipped=3.0 +2023-02-06 01:50:32,268 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:50:54,056 INFO [train.py:901] (2/4) Epoch 6, batch 250, loss[loss=0.2876, simple_loss=0.3566, pruned_loss=0.1093, over 8364.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3445, pruned_loss=0.1081, over 1164507.81 frames. ], batch size: 24, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:50:56,273 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:50:58,361 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:51:04,095 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 01:51:12,263 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 01:51:13,018 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40692.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:51:15,103 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:51:29,243 INFO [train.py:901] (2/4) Epoch 6, batch 300, loss[loss=0.2427, simple_loss=0.3155, pruned_loss=0.08493, over 7783.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3452, pruned_loss=0.1088, over 1265635.33 frames. ], batch size: 19, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:51:38,581 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 3.025e+02 3.729e+02 4.724e+02 9.863e+02, threshold=7.458e+02, percent-clipped=3.0 +2023-02-06 01:51:52,710 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40749.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:52:03,708 INFO [train.py:901] (2/4) Epoch 6, batch 350, loss[loss=0.2848, simple_loss=0.338, pruned_loss=0.1158, over 7971.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3453, pruned_loss=0.1091, over 1344824.38 frames. ], batch size: 21, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:52:14,009 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 01:52:17,242 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5454, 1.8820, 1.9379, 0.7131, 2.0108, 1.3430, 0.3866, 1.6569], + device='cuda:2'), covar=tensor([0.0181, 0.0116, 0.0080, 0.0208, 0.0115, 0.0345, 0.0304, 0.0096], + device='cuda:2'), in_proj_covar=tensor([0.0324, 0.0234, 0.0195, 0.0282, 0.0228, 0.0374, 0.0300, 0.0274], + device='cuda:2'), out_proj_covar=tensor([1.1224e-04, 7.8652e-05, 6.5579e-05, 9.5827e-05, 7.8811e-05, 1.3856e-04, + 1.0426e-04, 9.3391e-05], device='cuda:2') +2023-02-06 01:52:32,434 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:52:38,237 INFO [train.py:901] (2/4) Epoch 6, batch 400, loss[loss=0.2675, simple_loss=0.3413, pruned_loss=0.0968, over 8466.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3456, pruned_loss=0.1095, over 1405806.51 frames. ], batch size: 25, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:52:46,905 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.939e+02 3.080e+02 3.801e+02 5.022e+02 1.220e+03, threshold=7.601e+02, percent-clipped=4.0 +2023-02-06 01:53:04,908 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40854.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:12,737 INFO [train.py:901] (2/4) Epoch 6, batch 450, loss[loss=0.2781, simple_loss=0.3484, pruned_loss=0.1039, over 8450.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3466, pruned_loss=0.1098, over 1454759.41 frames. ], batch size: 27, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:53:24,455 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40883.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:47,833 INFO [train.py:901] (2/4) Epoch 6, batch 500, loss[loss=0.2805, simple_loss=0.3503, pruned_loss=0.1054, over 8287.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3452, pruned_loss=0.1084, over 1490614.16 frames. ], batch size: 23, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:53:49,256 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:56,749 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:57,236 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 3.058e+02 3.738e+02 5.288e+02 8.550e+02, threshold=7.476e+02, percent-clipped=3.0 +2023-02-06 01:54:12,298 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:13,627 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40953.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:22,949 INFO [train.py:901] (2/4) Epoch 6, batch 550, loss[loss=0.2414, simple_loss=0.3054, pruned_loss=0.08873, over 8089.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3445, pruned_loss=0.1079, over 1520697.60 frames. ], batch size: 21, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:54:25,215 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:30,604 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:49,828 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:55,069 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41013.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:56,494 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:57,039 INFO [train.py:901] (2/4) Epoch 6, batch 600, loss[loss=0.3477, simple_loss=0.3921, pruned_loss=0.1516, over 7127.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3444, pruned_loss=0.1083, over 1535303.85 frames. ], batch size: 71, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:55:06,090 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.839e+02 3.515e+02 4.292e+02 8.268e+02, threshold=7.031e+02, percent-clipped=4.0 +2023-02-06 01:55:06,996 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41030.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:55:09,479 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 01:55:29,303 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41063.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:55:29,382 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41063.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:55:31,088 INFO [train.py:901] (2/4) Epoch 6, batch 650, loss[loss=0.2576, simple_loss=0.333, pruned_loss=0.09109, over 8334.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3471, pruned_loss=0.1104, over 1556442.31 frames. ], batch size: 25, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:55:43,236 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.27 vs. limit=5.0 +2023-02-06 01:55:46,217 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41088.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:56:05,822 INFO [train.py:901] (2/4) Epoch 6, batch 700, loss[loss=0.283, simple_loss=0.3512, pruned_loss=0.1074, over 8637.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.347, pruned_loss=0.1103, over 1571317.60 frames. ], batch size: 34, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:56:08,626 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:56:14,231 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:56:14,730 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.998e+02 3.776e+02 4.654e+02 1.221e+03, threshold=7.553e+02, percent-clipped=4.0 +2023-02-06 01:56:40,069 INFO [train.py:901] (2/4) Epoch 6, batch 750, loss[loss=0.2452, simple_loss=0.309, pruned_loss=0.09066, over 7793.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3462, pruned_loss=0.1102, over 1582511.42 frames. ], batch size: 19, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:56:52,788 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 01:57:00,440 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3155, 2.5639, 1.6238, 2.1055, 2.1475, 1.3909, 1.8639, 1.8827], + device='cuda:2'), covar=tensor([0.1127, 0.0265, 0.0856, 0.0537, 0.0554, 0.1138, 0.0773, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0339, 0.0237, 0.0306, 0.0298, 0.0311, 0.0315, 0.0335, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 01:57:00,945 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 01:57:15,154 INFO [train.py:901] (2/4) Epoch 6, batch 800, loss[loss=0.2549, simple_loss=0.321, pruned_loss=0.09433, over 7817.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3454, pruned_loss=0.1096, over 1589966.09 frames. ], batch size: 20, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:57:21,519 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41225.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:22,744 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41227.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:23,995 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.937e+02 3.578e+02 4.897e+02 8.076e+02, threshold=7.157e+02, percent-clipped=3.0 +2023-02-06 01:57:38,349 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:41,025 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6231, 2.5912, 4.6942, 1.2157, 3.4364, 2.1935, 1.7041, 2.8651], + device='cuda:2'), covar=tensor([0.1538, 0.1568, 0.0597, 0.3158, 0.1151, 0.2280, 0.1473, 0.2220], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0454, 0.0526, 0.0539, 0.0590, 0.0517, 0.0449, 0.0592], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 01:57:46,852 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:49,531 INFO [train.py:901] (2/4) Epoch 6, batch 850, loss[loss=0.2512, simple_loss=0.3255, pruned_loss=0.08841, over 7967.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3468, pruned_loss=0.1101, over 1598469.81 frames. ], batch size: 21, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:58:23,893 INFO [train.py:901] (2/4) Epoch 6, batch 900, loss[loss=0.3166, simple_loss=0.3786, pruned_loss=0.1273, over 8231.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3469, pruned_loss=0.1103, over 1607052.51 frames. ], batch size: 24, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:58:33,480 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.871e+02 3.405e+02 4.321e+02 1.147e+03, threshold=6.810e+02, percent-clipped=2.0 +2023-02-06 01:58:42,527 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:58:53,863 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:58:58,546 INFO [train.py:901] (2/4) Epoch 6, batch 950, loss[loss=0.3472, simple_loss=0.4, pruned_loss=0.1472, over 8624.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3447, pruned_loss=0.1086, over 1610569.18 frames. ], batch size: 39, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:59:06,367 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41377.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:11,061 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:21,425 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 01:59:26,997 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41407.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:28,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:32,969 INFO [train.py:901] (2/4) Epoch 6, batch 1000, loss[loss=0.3065, simple_loss=0.3659, pruned_loss=0.1236, over 8841.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3456, pruned_loss=0.1087, over 1615141.83 frames. ], batch size: 40, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:59:41,357 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 3.293e+02 3.921e+02 5.074e+02 1.211e+03, threshold=7.843e+02, percent-clipped=6.0 +2023-02-06 01:59:55,288 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 02:00:06,252 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:00:07,520 INFO [train.py:901] (2/4) Epoch 6, batch 1050, loss[loss=0.2685, simple_loss=0.3345, pruned_loss=0.1012, over 7710.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3458, pruned_loss=0.109, over 1612900.61 frames. ], batch size: 18, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:00:08,220 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 02:00:13,144 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:00:42,275 INFO [train.py:901] (2/4) Epoch 6, batch 1100, loss[loss=0.326, simple_loss=0.379, pruned_loss=0.1365, over 8441.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3458, pruned_loss=0.1091, over 1613472.10 frames. ], batch size: 49, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:00:46,701 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41522.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:00:51,107 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.937e+02 3.488e+02 4.376e+02 9.981e+02, threshold=6.976e+02, percent-clipped=3.0 +2023-02-06 02:01:13,000 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1797, 2.1446, 1.5123, 1.8958, 1.6880, 1.3066, 1.5088, 1.6569], + device='cuda:2'), covar=tensor([0.0940, 0.0267, 0.0897, 0.0466, 0.0584, 0.1124, 0.0797, 0.0698], + device='cuda:2'), in_proj_covar=tensor([0.0338, 0.0234, 0.0310, 0.0296, 0.0313, 0.0309, 0.0334, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 02:01:16,051 INFO [train.py:901] (2/4) Epoch 6, batch 1150, loss[loss=0.2894, simple_loss=0.3659, pruned_loss=0.1065, over 8522.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3446, pruned_loss=0.1083, over 1611927.27 frames. ], batch size: 39, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:01:18,818 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 02:01:25,451 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41579.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:01:38,066 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:01:50,379 INFO [train.py:901] (2/4) Epoch 6, batch 1200, loss[loss=0.2934, simple_loss=0.3523, pruned_loss=0.1173, over 8648.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3437, pruned_loss=0.1077, over 1614282.59 frames. ], batch size: 34, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:01:55,789 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:00,249 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 3.060e+02 3.864e+02 4.910e+02 1.275e+03, threshold=7.729e+02, percent-clipped=9.0 +2023-02-06 02:02:02,461 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5965, 1.5164, 3.0240, 1.1311, 2.0396, 3.3231, 3.2140, 2.8356], + device='cuda:2'), covar=tensor([0.1059, 0.1288, 0.0405, 0.1987, 0.0772, 0.0275, 0.0498, 0.0615], + device='cuda:2'), in_proj_covar=tensor([0.0244, 0.0272, 0.0233, 0.0267, 0.0234, 0.0211, 0.0248, 0.0285], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 02:02:03,213 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41633.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:19,548 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:24,290 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7299, 2.3038, 3.7425, 2.8894, 2.8104, 2.2923, 1.6609, 1.6071], + device='cuda:2'), covar=tensor([0.2009, 0.2546, 0.0500, 0.1331, 0.1338, 0.1204, 0.1219, 0.2659], + device='cuda:2'), in_proj_covar=tensor([0.0772, 0.0716, 0.0606, 0.0709, 0.0799, 0.0661, 0.0630, 0.0658], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 02:02:24,697 INFO [train.py:901] (2/4) Epoch 6, batch 1250, loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 8484.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3456, pruned_loss=0.109, over 1617422.97 frames. ], batch size: 28, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:02:29,690 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:59,806 INFO [train.py:901] (2/4) Epoch 6, batch 1300, loss[loss=0.2926, simple_loss=0.3575, pruned_loss=0.1139, over 7976.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3445, pruned_loss=0.1081, over 1614433.68 frames. ], batch size: 21, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:03:08,606 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 3.137e+02 4.028e+02 4.813e+02 9.668e+02, threshold=8.056e+02, percent-clipped=5.0 +2023-02-06 02:03:09,499 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:03:27,427 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:03:34,646 INFO [train.py:901] (2/4) Epoch 6, batch 1350, loss[loss=0.2651, simple_loss=0.3346, pruned_loss=0.09781, over 8290.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3437, pruned_loss=0.1077, over 1609444.43 frames. ], batch size: 23, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:03:42,962 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:00,418 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41803.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:09,625 INFO [train.py:901] (2/4) Epoch 6, batch 1400, loss[loss=0.2133, simple_loss=0.282, pruned_loss=0.07232, over 7252.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3426, pruned_loss=0.1072, over 1609412.87 frames. ], batch size: 16, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:04:18,118 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.090e+02 3.079e+02 3.704e+02 4.589e+02 8.838e+02, threshold=7.407e+02, percent-clipped=2.0 +2023-02-06 02:04:22,375 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41835.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:39,963 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41860.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:44,592 INFO [train.py:901] (2/4) Epoch 6, batch 1450, loss[loss=0.2734, simple_loss=0.3369, pruned_loss=0.1049, over 7804.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.342, pruned_loss=0.1073, over 1608355.64 frames. ], batch size: 20, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:04:47,868 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 02:05:18,590 INFO [train.py:901] (2/4) Epoch 6, batch 1500, loss[loss=0.2392, simple_loss=0.315, pruned_loss=0.08166, over 7806.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3428, pruned_loss=0.1075, over 1610731.02 frames. ], batch size: 20, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:05:24,646 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41924.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:05:27,835 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.922e+02 3.542e+02 4.432e+02 1.007e+03, threshold=7.084e+02, percent-clipped=2.0 +2023-02-06 02:05:53,224 INFO [train.py:901] (2/4) Epoch 6, batch 1550, loss[loss=0.2279, simple_loss=0.2884, pruned_loss=0.08373, over 7447.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3418, pruned_loss=0.1062, over 1616188.09 frames. ], batch size: 17, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:05:53,365 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41966.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:06:28,450 INFO [train.py:901] (2/4) Epoch 6, batch 1600, loss[loss=0.3487, simple_loss=0.3894, pruned_loss=0.154, over 6999.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.342, pruned_loss=0.1065, over 1613010.41 frames. ], batch size: 72, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:06:28,516 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:06:37,879 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.135e+02 3.132e+02 3.836e+02 5.392e+02 3.005e+03, threshold=7.672e+02, percent-clipped=11.0 +2023-02-06 02:07:03,789 INFO [train.py:901] (2/4) Epoch 6, batch 1650, loss[loss=0.2663, simple_loss=0.3317, pruned_loss=0.1004, over 8354.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3409, pruned_loss=0.1055, over 1612589.46 frames. ], batch size: 24, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:07:39,088 INFO [train.py:901] (2/4) Epoch 6, batch 1700, loss[loss=0.262, simple_loss=0.3417, pruned_loss=0.09116, over 8513.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3408, pruned_loss=0.1059, over 1611995.42 frames. ], batch size: 26, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:07:47,893 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.826e+02 3.670e+02 4.452e+02 1.049e+03, threshold=7.339e+02, percent-clipped=2.0 +2023-02-06 02:07:49,313 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42131.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:07:52,036 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1006, 1.1038, 3.2521, 0.9398, 2.7808, 2.7568, 2.9302, 2.8530], + device='cuda:2'), covar=tensor([0.0629, 0.3634, 0.0747, 0.2865, 0.1502, 0.0885, 0.0655, 0.0772], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0504, 0.0442, 0.0435, 0.0500, 0.0414, 0.0412, 0.0463], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 02:08:05,229 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8494, 2.9244, 3.0766, 2.3707, 1.4474, 3.4812, 0.6916, 1.6625], + device='cuda:2'), covar=tensor([0.3325, 0.1599, 0.0675, 0.2471, 0.6924, 0.0424, 0.5907, 0.2882], + device='cuda:2'), in_proj_covar=tensor([0.0131, 0.0132, 0.0082, 0.0179, 0.0216, 0.0083, 0.0145, 0.0135], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 02:08:14,469 INFO [train.py:901] (2/4) Epoch 6, batch 1750, loss[loss=0.2538, simple_loss=0.3238, pruned_loss=0.09186, over 8146.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3413, pruned_loss=0.1067, over 1610977.99 frames. ], batch size: 22, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:08:49,384 INFO [train.py:901] (2/4) Epoch 6, batch 1800, loss[loss=0.2825, simple_loss=0.343, pruned_loss=0.111, over 7971.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3424, pruned_loss=0.1075, over 1610853.75 frames. ], batch size: 21, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:08:59,176 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 3.032e+02 3.540e+02 4.353e+02 2.015e+03, threshold=7.080e+02, percent-clipped=5.0 +2023-02-06 02:09:24,921 INFO [train.py:901] (2/4) Epoch 6, batch 1850, loss[loss=0.2712, simple_loss=0.3477, pruned_loss=0.09729, over 8489.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3428, pruned_loss=0.1073, over 1613818.36 frames. ], batch size: 28, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:09:26,410 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:09:55,347 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42310.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:09:59,287 INFO [train.py:901] (2/4) Epoch 6, batch 1900, loss[loss=0.3067, simple_loss=0.362, pruned_loss=0.1257, over 8318.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3434, pruned_loss=0.1075, over 1614873.05 frames. ], batch size: 25, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:10:08,770 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.992e+02 2.715e+02 3.297e+02 4.142e+02 7.213e+02, threshold=6.594e+02, percent-clipped=2.0 +2023-02-06 02:10:23,937 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 02:10:34,058 INFO [train.py:901] (2/4) Epoch 6, batch 1950, loss[loss=0.3136, simple_loss=0.369, pruned_loss=0.1291, over 8198.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3433, pruned_loss=0.1073, over 1617865.64 frames. ], batch size: 23, lr: 1.30e-02, grad_scale: 16.0 +2023-02-06 02:10:36,639 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 02:10:46,220 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:10:48,987 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42387.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:10:56,181 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 02:10:56,318 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42397.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:11:06,406 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42412.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:11:08,979 INFO [train.py:901] (2/4) Epoch 6, batch 2000, loss[loss=0.2613, simple_loss=0.3189, pruned_loss=0.1018, over 7793.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3423, pruned_loss=0.1069, over 1615346.06 frames. ], batch size: 20, lr: 1.30e-02, grad_scale: 16.0 +2023-02-06 02:11:15,181 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42425.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:11:18,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.766e+02 3.581e+02 4.303e+02 8.011e+02, threshold=7.162e+02, percent-clipped=3.0 +2023-02-06 02:11:37,942 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3747, 1.5924, 1.5599, 1.2755, 1.3132, 1.3933, 1.9985, 1.6622], + device='cuda:2'), covar=tensor([0.0524, 0.1212, 0.1717, 0.1405, 0.0599, 0.1577, 0.0631, 0.0539], + device='cuda:2'), in_proj_covar=tensor([0.0128, 0.0173, 0.0211, 0.0178, 0.0125, 0.0183, 0.0138, 0.0149], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 02:11:43,879 INFO [train.py:901] (2/4) Epoch 6, batch 2050, loss[loss=0.2963, simple_loss=0.3597, pruned_loss=0.1165, over 8591.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3421, pruned_loss=0.1065, over 1619122.29 frames. ], batch size: 49, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:12:17,698 INFO [train.py:901] (2/4) Epoch 6, batch 2100, loss[loss=0.2718, simple_loss=0.3219, pruned_loss=0.1108, over 7795.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3438, pruned_loss=0.1081, over 1618184.03 frames. ], batch size: 19, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:12:23,879 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:12:27,686 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.916e+02 3.481e+02 4.572e+02 1.310e+03, threshold=6.962e+02, percent-clipped=2.0 +2023-02-06 02:12:52,386 INFO [train.py:901] (2/4) Epoch 6, batch 2150, loss[loss=0.2818, simple_loss=0.3598, pruned_loss=0.1019, over 8623.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3444, pruned_loss=0.1082, over 1614941.94 frames. ], batch size: 31, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:12:54,238 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.16 vs. limit=5.0 +2023-02-06 02:13:16,791 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 02:13:27,027 INFO [train.py:901] (2/4) Epoch 6, batch 2200, loss[loss=0.2906, simple_loss=0.3536, pruned_loss=0.1138, over 8023.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.344, pruned_loss=0.108, over 1615866.33 frames. ], batch size: 22, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:13:36,143 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 3.020e+02 3.729e+02 5.072e+02 1.122e+03, threshold=7.459e+02, percent-clipped=5.0 +2023-02-06 02:13:43,112 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:13:59,702 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:14:00,826 INFO [train.py:901] (2/4) Epoch 6, batch 2250, loss[loss=0.3157, simple_loss=0.3635, pruned_loss=0.134, over 6745.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3423, pruned_loss=0.1071, over 1611028.25 frames. ], batch size: 71, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:14:01,631 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:14:11,906 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42681.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:14:29,254 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42706.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:14:35,833 INFO [train.py:901] (2/4) Epoch 6, batch 2300, loss[loss=0.2883, simple_loss=0.3422, pruned_loss=0.1172, over 7653.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3431, pruned_loss=0.1076, over 1617058.23 frames. ], batch size: 19, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:14:45,246 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.977e+02 3.532e+02 4.435e+02 7.362e+02, threshold=7.063e+02, percent-clipped=0.0 +2023-02-06 02:14:53,236 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42741.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:15:11,251 INFO [train.py:901] (2/4) Epoch 6, batch 2350, loss[loss=0.315, simple_loss=0.3727, pruned_loss=0.1287, over 8529.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3442, pruned_loss=0.1083, over 1616182.67 frames. ], batch size: 28, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:15:46,904 INFO [train.py:901] (2/4) Epoch 6, batch 2400, loss[loss=0.308, simple_loss=0.3743, pruned_loss=0.1209, over 8248.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3429, pruned_loss=0.1075, over 1615535.86 frames. ], batch size: 24, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:15:56,306 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 3.099e+02 3.712e+02 4.452e+02 1.076e+03, threshold=7.425e+02, percent-clipped=4.0 +2023-02-06 02:16:14,324 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:20,856 INFO [train.py:901] (2/4) Epoch 6, batch 2450, loss[loss=0.3236, simple_loss=0.3849, pruned_loss=0.1311, over 8504.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.343, pruned_loss=0.1076, over 1615442.02 frames. ], batch size: 26, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:16:22,282 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:27,381 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 02:16:29,160 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:42,219 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:54,610 INFO [train.py:901] (2/4) Epoch 6, batch 2500, loss[loss=0.3004, simple_loss=0.3696, pruned_loss=0.1156, over 8323.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3428, pruned_loss=0.1069, over 1619263.75 frames. ], batch size: 25, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:17:05,200 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 3.094e+02 4.004e+02 4.995e+02 1.056e+03, threshold=8.009e+02, percent-clipped=4.0 +2023-02-06 02:17:15,514 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3211, 1.4792, 1.3971, 1.9471, 0.9709, 1.1554, 1.2643, 1.4138], + device='cuda:2'), covar=tensor([0.1130, 0.1027, 0.1333, 0.0593, 0.1230, 0.1925, 0.0997, 0.0938], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0253, 0.0285, 0.0228, 0.0247, 0.0283, 0.0284, 0.0258], + 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') +2023-02-06 02:17:29,431 INFO [train.py:901] (2/4) Epoch 6, batch 2550, loss[loss=0.2934, simple_loss=0.3392, pruned_loss=0.1238, over 7793.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3422, pruned_loss=0.107, over 1615355.03 frames. ], batch size: 19, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:17:41,648 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:18:01,224 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43011.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:18:04,439 INFO [train.py:901] (2/4) Epoch 6, batch 2600, loss[loss=0.2618, simple_loss=0.3278, pruned_loss=0.09789, over 8361.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3412, pruned_loss=0.1064, over 1614567.82 frames. ], batch size: 24, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:18:13,987 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.052e+02 3.779e+02 5.019e+02 1.784e+03, threshold=7.558e+02, percent-clipped=4.0 +2023-02-06 02:18:39,586 INFO [train.py:901] (2/4) Epoch 6, batch 2650, loss[loss=0.2769, simple_loss=0.3422, pruned_loss=0.1058, over 8111.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3397, pruned_loss=0.1056, over 1607706.90 frames. ], batch size: 23, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:18:43,297 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4779, 1.7888, 3.1095, 1.1035, 2.2811, 1.8044, 1.3701, 1.8937], + device='cuda:2'), covar=tensor([0.1548, 0.2102, 0.0644, 0.3475, 0.1370, 0.2480, 0.1654, 0.2175], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0455, 0.0521, 0.0538, 0.0584, 0.0520, 0.0445, 0.0589], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:19:05,116 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1709, 1.0974, 1.1278, 1.1036, 0.8377, 1.2492, 0.0284, 0.7853], + device='cuda:2'), covar=tensor([0.3080, 0.2555, 0.1079, 0.2130, 0.5575, 0.0822, 0.5121, 0.2514], + device='cuda:2'), in_proj_covar=tensor([0.0134, 0.0133, 0.0083, 0.0182, 0.0219, 0.0084, 0.0144, 0.0136], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 02:19:11,157 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43112.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:19:14,347 INFO [train.py:901] (2/4) Epoch 6, batch 2700, loss[loss=0.2528, simple_loss=0.3319, pruned_loss=0.08679, over 8359.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3388, pruned_loss=0.1049, over 1608675.26 frames. ], batch size: 24, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:19:15,204 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7834, 2.1118, 3.5744, 1.2618, 2.6431, 1.9795, 1.7035, 2.1834], + device='cuda:2'), covar=tensor([0.1377, 0.1934, 0.0549, 0.3336, 0.1320, 0.2359, 0.1404, 0.2188], + device='cuda:2'), in_proj_covar=tensor([0.0472, 0.0462, 0.0529, 0.0542, 0.0594, 0.0528, 0.0449, 0.0601], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:19:20,977 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43126.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:19:23,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.935e+02 3.532e+02 4.548e+02 1.003e+03, threshold=7.064e+02, percent-clipped=2.0 +2023-02-06 02:19:28,348 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:19:45,823 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 +2023-02-06 02:19:47,959 INFO [train.py:901] (2/4) Epoch 6, batch 2750, loss[loss=0.2879, simple_loss=0.3567, pruned_loss=0.1095, over 8287.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3392, pruned_loss=0.105, over 1611019.40 frames. ], batch size: 49, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:19:57,318 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4570, 1.9542, 2.1732, 0.9905, 2.3370, 1.3861, 0.6678, 1.8327], + device='cuda:2'), covar=tensor([0.0248, 0.0115, 0.0084, 0.0228, 0.0116, 0.0385, 0.0330, 0.0114], + device='cuda:2'), in_proj_covar=tensor([0.0321, 0.0235, 0.0206, 0.0288, 0.0227, 0.0381, 0.0298, 0.0273], + device='cuda:2'), out_proj_covar=tensor([1.0839e-04, 7.7382e-05, 6.8384e-05, 9.5718e-05, 7.6279e-05, 1.3795e-04, + 1.0173e-04, 9.1606e-05], device='cuda:2') +2023-02-06 02:20:02,630 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6565, 1.8257, 3.2679, 1.3187, 2.7864, 1.9649, 1.7410, 2.4896], + device='cuda:2'), covar=tensor([0.1712, 0.2103, 0.0570, 0.3352, 0.0995, 0.2234, 0.1758, 0.1600], + device='cuda:2'), in_proj_covar=tensor([0.0469, 0.0460, 0.0526, 0.0535, 0.0589, 0.0523, 0.0444, 0.0590], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:20:22,670 INFO [train.py:901] (2/4) Epoch 6, batch 2800, loss[loss=0.239, simple_loss=0.3051, pruned_loss=0.08645, over 7817.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3401, pruned_loss=0.1051, over 1614380.60 frames. ], batch size: 20, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:20:26,191 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:32,061 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.696e+02 3.315e+02 4.271e+02 8.534e+02, threshold=6.630e+02, percent-clipped=4.0 +2023-02-06 02:20:39,131 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:40,382 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43241.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:55,847 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43264.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:56,926 INFO [train.py:901] (2/4) Epoch 6, batch 2850, loss[loss=0.2729, simple_loss=0.3557, pruned_loss=0.0951, over 8320.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3419, pruned_loss=0.1063, over 1617440.42 frames. ], batch size: 25, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:21:11,142 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0534, 2.2364, 1.6649, 2.7790, 1.4320, 1.4231, 1.9094, 2.4604], + device='cuda:2'), covar=tensor([0.0986, 0.1221, 0.1480, 0.0541, 0.1474, 0.2053, 0.1316, 0.0847], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0254, 0.0282, 0.0229, 0.0247, 0.0284, 0.0286, 0.0254], + 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') +2023-02-06 02:21:31,678 INFO [train.py:901] (2/4) Epoch 6, batch 2900, loss[loss=0.2355, simple_loss=0.2939, pruned_loss=0.08858, over 7423.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3416, pruned_loss=0.1061, over 1614002.74 frames. ], batch size: 17, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:21:41,568 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.996e+02 3.885e+02 4.976e+02 9.964e+02, threshold=7.771e+02, percent-clipped=9.0 +2023-02-06 02:21:46,021 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:00,470 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43356.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:01,679 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 02:22:07,125 INFO [train.py:901] (2/4) Epoch 6, batch 2950, loss[loss=0.2471, simple_loss=0.3013, pruned_loss=0.09646, over 7229.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3414, pruned_loss=0.1062, over 1612057.58 frames. ], batch size: 16, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:22:17,900 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43382.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:36,001 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43407.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:41,761 INFO [train.py:901] (2/4) Epoch 6, batch 3000, loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.112, over 8512.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3403, pruned_loss=0.1055, over 1611425.30 frames. ], batch size: 26, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:22:41,761 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 02:22:53,878 INFO [train.py:935] (2/4) Epoch 6, validation: loss=0.2158, simple_loss=0.3124, pruned_loss=0.05962, over 944034.00 frames. +2023-02-06 02:22:53,879 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 02:23:03,880 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.288e+02 4.080e+02 5.339e+02 1.082e+03, threshold=8.161e+02, percent-clipped=5.0 +2023-02-06 02:23:28,757 INFO [train.py:901] (2/4) Epoch 6, batch 3050, loss[loss=0.3169, simple_loss=0.364, pruned_loss=0.1349, over 7979.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3422, pruned_loss=0.1071, over 1610863.96 frames. ], batch size: 21, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:24:03,331 INFO [train.py:901] (2/4) Epoch 6, batch 3100, loss[loss=0.2567, simple_loss=0.3268, pruned_loss=0.09328, over 7972.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3404, pruned_loss=0.1058, over 1612719.96 frames. ], batch size: 21, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:24:12,759 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.871e+02 3.509e+02 4.582e+02 1.148e+03, threshold=7.017e+02, percent-clipped=4.0 +2023-02-06 02:24:14,871 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:24:38,261 INFO [train.py:901] (2/4) Epoch 6, batch 3150, loss[loss=0.2275, simple_loss=0.2932, pruned_loss=0.08091, over 7726.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3411, pruned_loss=0.107, over 1606225.89 frames. ], batch size: 18, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:24:57,058 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:10,976 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43612.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:13,475 INFO [train.py:901] (2/4) Epoch 6, batch 3200, loss[loss=0.2486, simple_loss=0.3172, pruned_loss=0.08996, over 8024.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3417, pruned_loss=0.1073, over 1607159.12 frames. ], batch size: 22, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:25:14,372 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43617.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:23,589 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.828e+02 3.409e+02 4.222e+02 1.719e+03, threshold=6.818e+02, percent-clipped=4.0 +2023-02-06 02:25:28,581 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43637.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:29,270 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0064, 4.0926, 2.5033, 2.7308, 2.8803, 2.4438, 2.9435, 3.2798], + device='cuda:2'), covar=tensor([0.1289, 0.0254, 0.0740, 0.0669, 0.0563, 0.0858, 0.0671, 0.0709], + device='cuda:2'), in_proj_covar=tensor([0.0342, 0.0234, 0.0309, 0.0303, 0.0316, 0.0310, 0.0338, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 02:25:49,153 INFO [train.py:901] (2/4) Epoch 6, batch 3250, loss[loss=0.267, simple_loss=0.3314, pruned_loss=0.1013, over 7973.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3413, pruned_loss=0.1074, over 1606928.68 frames. ], batch size: 21, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:25:49,412 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1398, 1.9519, 3.0667, 2.3354, 2.4810, 1.9000, 1.3838, 1.1807], + device='cuda:2'), covar=tensor([0.2126, 0.2093, 0.0493, 0.1143, 0.1007, 0.1162, 0.1194, 0.2231], + device='cuda:2'), in_proj_covar=tensor([0.0792, 0.0722, 0.0624, 0.0712, 0.0801, 0.0670, 0.0635, 0.0663], + 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') +2023-02-06 02:25:57,411 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2493, 1.3037, 3.4358, 1.5500, 2.2211, 3.8195, 3.6985, 3.2082], + device='cuda:2'), covar=tensor([0.0918, 0.1655, 0.0346, 0.1954, 0.0930, 0.0230, 0.0384, 0.0659], + device='cuda:2'), in_proj_covar=tensor([0.0241, 0.0272, 0.0225, 0.0265, 0.0239, 0.0213, 0.0250, 0.0284], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 02:26:16,640 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8439, 3.7589, 3.4487, 1.8966, 3.3734, 3.3224, 3.5500, 2.9526], + device='cuda:2'), covar=tensor([0.0940, 0.0625, 0.0937, 0.4316, 0.0841, 0.0825, 0.1288, 0.0998], + device='cuda:2'), in_proj_covar=tensor([0.0405, 0.0297, 0.0334, 0.0411, 0.0321, 0.0285, 0.0316, 0.0260], + 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') +2023-02-06 02:26:17,077 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 02:26:23,414 INFO [train.py:901] (2/4) Epoch 6, batch 3300, loss[loss=0.2245, simple_loss=0.2868, pruned_loss=0.08107, over 7707.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3415, pruned_loss=0.1069, over 1613666.47 frames. ], batch size: 18, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:26:24,932 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3533, 2.1008, 1.4947, 1.9995, 1.8125, 1.2552, 1.5293, 1.8784], + device='cuda:2'), covar=tensor([0.0970, 0.0364, 0.0920, 0.0449, 0.0595, 0.1225, 0.0828, 0.0670], + device='cuda:2'), in_proj_covar=tensor([0.0334, 0.0230, 0.0304, 0.0297, 0.0310, 0.0304, 0.0330, 0.0302], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 02:26:33,006 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.968e+02 3.670e+02 5.054e+02 9.057e+02, threshold=7.341e+02, percent-clipped=6.0 +2023-02-06 02:26:43,374 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8405, 2.0899, 1.7408, 2.7053, 1.1035, 1.3448, 1.6287, 2.2672], + device='cuda:2'), covar=tensor([0.1106, 0.1019, 0.1358, 0.0577, 0.1572, 0.2092, 0.1517, 0.1057], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0251, 0.0277, 0.0225, 0.0243, 0.0276, 0.0282, 0.0251], + 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') +2023-02-06 02:26:58,035 INFO [train.py:901] (2/4) Epoch 6, batch 3350, loss[loss=0.2862, simple_loss=0.3447, pruned_loss=0.1139, over 8474.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3405, pruned_loss=0.106, over 1615812.17 frames. ], batch size: 29, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:27:25,486 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43805.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:27:33,330 INFO [train.py:901] (2/4) Epoch 6, batch 3400, loss[loss=0.3114, simple_loss=0.3663, pruned_loss=0.1283, over 7809.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3418, pruned_loss=0.1064, over 1618187.60 frames. ], batch size: 20, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:27:39,291 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7808, 5.8062, 4.9779, 2.2716, 5.1185, 5.4492, 5.4541, 4.7972], + device='cuda:2'), covar=tensor([0.0555, 0.0349, 0.0827, 0.4489, 0.0617, 0.0497, 0.0855, 0.0592], + device='cuda:2'), in_proj_covar=tensor([0.0407, 0.0303, 0.0339, 0.0421, 0.0325, 0.0290, 0.0324, 0.0266], + 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') +2023-02-06 02:27:42,441 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.693e+02 3.397e+02 4.441e+02 9.371e+02, threshold=6.793e+02, percent-clipped=2.0 +2023-02-06 02:27:46,358 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 02:27:47,659 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.78 vs. limit=5.0 +2023-02-06 02:28:07,549 INFO [train.py:901] (2/4) Epoch 6, batch 3450, loss[loss=0.2669, simple_loss=0.3213, pruned_loss=0.1063, over 7711.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3405, pruned_loss=0.1052, over 1619614.12 frames. ], batch size: 18, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:28:09,137 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3702, 2.4456, 3.1100, 1.2068, 3.1740, 1.9445, 1.4708, 1.9491], + device='cuda:2'), covar=tensor([0.0282, 0.0158, 0.0150, 0.0284, 0.0192, 0.0414, 0.0387, 0.0204], + device='cuda:2'), in_proj_covar=tensor([0.0332, 0.0244, 0.0206, 0.0297, 0.0239, 0.0394, 0.0306, 0.0279], + device='cuda:2'), out_proj_covar=tensor([1.1211e-04, 8.0431e-05, 6.7840e-05, 9.8214e-05, 8.0310e-05, 1.4249e-04, + 1.0373e-04, 9.3226e-05], device='cuda:2') +2023-02-06 02:28:14,344 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:28:24,929 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 02:28:42,228 INFO [train.py:901] (2/4) Epoch 6, batch 3500, loss[loss=0.2334, simple_loss=0.3, pruned_loss=0.0834, over 7656.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3399, pruned_loss=0.1045, over 1617600.67 frames. ], batch size: 19, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:28:50,530 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:28:52,404 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 3.111e+02 3.775e+02 4.956e+02 7.195e+02, threshold=7.550e+02, percent-clipped=1.0 +2023-02-06 02:28:59,179 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 02:29:16,409 INFO [train.py:901] (2/4) Epoch 6, batch 3550, loss[loss=0.3241, simple_loss=0.3609, pruned_loss=0.1437, over 8030.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3412, pruned_loss=0.106, over 1615640.09 frames. ], batch size: 22, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:29:23,547 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.52 vs. limit=5.0 +2023-02-06 02:29:24,742 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43977.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:29:34,248 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:29:52,632 INFO [train.py:901] (2/4) Epoch 6, batch 3600, loss[loss=0.2892, simple_loss=0.3496, pruned_loss=0.1144, over 8464.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.34, pruned_loss=0.1048, over 1617543.98 frames. ], batch size: 27, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:30:02,263 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.983e+02 3.632e+02 4.470e+02 1.452e+03, threshold=7.265e+02, percent-clipped=1.0 +2023-02-06 02:30:27,000 INFO [train.py:901] (2/4) Epoch 6, batch 3650, loss[loss=0.2525, simple_loss=0.3021, pruned_loss=0.1014, over 7806.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3404, pruned_loss=0.1056, over 1615007.06 frames. ], batch size: 20, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:31:00,588 INFO [train.py:901] (2/4) Epoch 6, batch 3700, loss[loss=0.2657, simple_loss=0.3405, pruned_loss=0.09546, over 8352.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3402, pruned_loss=0.1056, over 1614492.65 frames. ], batch size: 26, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:31:01,284 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 02:31:11,218 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 3.019e+02 3.651e+02 4.413e+02 8.839e+02, threshold=7.303e+02, percent-clipped=3.0 +2023-02-06 02:31:23,983 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44149.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:31:24,338 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 02:31:35,820 INFO [train.py:901] (2/4) Epoch 6, batch 3750, loss[loss=0.2501, simple_loss=0.314, pruned_loss=0.09311, over 8518.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3396, pruned_loss=0.105, over 1614563.39 frames. ], batch size: 28, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:32:09,331 INFO [train.py:901] (2/4) Epoch 6, batch 3800, loss[loss=0.2822, simple_loss=0.3628, pruned_loss=0.1008, over 8466.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.342, pruned_loss=0.1069, over 1621075.09 frames. ], batch size: 29, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:32:19,582 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.008e+02 3.761e+02 4.930e+02 1.044e+03, threshold=7.521e+02, percent-clipped=7.0 +2023-02-06 02:32:32,434 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:32:44,256 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44264.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:32:45,454 INFO [train.py:901] (2/4) Epoch 6, batch 3850, loss[loss=0.2252, simple_loss=0.2888, pruned_loss=0.08076, over 7543.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3413, pruned_loss=0.1064, over 1619939.53 frames. ], batch size: 18, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:32:48,967 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:32:49,771 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44272.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:33:02,820 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 02:33:20,475 INFO [train.py:901] (2/4) Epoch 6, batch 3900, loss[loss=0.2708, simple_loss=0.3422, pruned_loss=0.0997, over 8194.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3422, pruned_loss=0.1066, over 1617576.60 frames. ], batch size: 23, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:33:23,954 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44321.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:33:30,569 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 2.909e+02 3.535e+02 4.398e+02 8.405e+02, threshold=7.069e+02, percent-clipped=2.0 +2023-02-06 02:33:56,224 INFO [train.py:901] (2/4) Epoch 6, batch 3950, loss[loss=0.2857, simple_loss=0.3547, pruned_loss=0.1084, over 8142.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3418, pruned_loss=0.1061, over 1612914.70 frames. ], batch size: 22, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:34:09,756 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:34:30,790 INFO [train.py:901] (2/4) Epoch 6, batch 4000, loss[loss=0.233, simple_loss=0.3004, pruned_loss=0.08284, over 7542.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3404, pruned_loss=0.1058, over 1607438.20 frames. ], batch size: 18, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:34:40,321 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.805e+02 3.702e+02 4.857e+02 8.487e+02, threshold=7.405e+02, percent-clipped=7.0 +2023-02-06 02:34:44,614 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44436.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:35:05,733 INFO [train.py:901] (2/4) Epoch 6, batch 4050, loss[loss=0.2999, simple_loss=0.3665, pruned_loss=0.1166, over 8464.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3409, pruned_loss=0.1057, over 1615200.15 frames. ], batch size: 29, lr: 1.27e-02, grad_scale: 16.0 +2023-02-06 02:35:08,219 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2023-02-06 02:35:21,628 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2632, 1.8140, 3.0982, 1.1011, 2.2292, 1.6303, 1.4901, 1.7779], + device='cuda:2'), covar=tensor([0.1610, 0.1705, 0.0557, 0.3129, 0.1388, 0.2535, 0.1493, 0.2123], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0452, 0.0524, 0.0536, 0.0579, 0.0520, 0.0439, 0.0580], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:35:41,181 INFO [train.py:901] (2/4) Epoch 6, batch 4100, loss[loss=0.3286, simple_loss=0.3888, pruned_loss=0.1342, over 8500.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3414, pruned_loss=0.1053, over 1620365.57 frames. ], batch size: 26, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:35:44,017 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:35:50,484 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 3.131e+02 3.987e+02 5.314e+02 1.327e+03, threshold=7.973e+02, percent-clipped=4.0 +2023-02-06 02:36:00,536 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44545.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:36:00,598 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44545.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:36:14,412 INFO [train.py:901] (2/4) Epoch 6, batch 4150, loss[loss=0.2041, simple_loss=0.2785, pruned_loss=0.06484, over 7795.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.34, pruned_loss=0.1048, over 1616306.47 frames. ], batch size: 19, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:36:15,791 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44568.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:36:49,813 INFO [train.py:901] (2/4) Epoch 6, batch 4200, loss[loss=0.1915, simple_loss=0.2633, pruned_loss=0.05988, over 7806.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3387, pruned_loss=0.1044, over 1610649.90 frames. ], batch size: 19, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:36:58,994 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.001e+02 2.806e+02 3.559e+02 4.787e+02 1.284e+03, threshold=7.119e+02, percent-clipped=4.0 +2023-02-06 02:37:05,643 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 02:37:07,980 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:37:23,783 INFO [train.py:901] (2/4) Epoch 6, batch 4250, loss[loss=0.3322, simple_loss=0.3882, pruned_loss=0.138, over 8487.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3382, pruned_loss=0.1044, over 1612300.87 frames. ], batch size: 28, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:37:24,685 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:37:29,278 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 02:37:41,582 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44692.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:37:58,489 INFO [train.py:901] (2/4) Epoch 6, batch 4300, loss[loss=0.2376, simple_loss=0.303, pruned_loss=0.08608, over 7662.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3395, pruned_loss=0.1048, over 1616771.88 frames. ], batch size: 19, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:38:00,027 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44717.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:38:08,662 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.771e+02 3.321e+02 4.102e+02 9.930e+02, threshold=6.641e+02, percent-clipped=2.0 +2023-02-06 02:38:17,337 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.3674, 1.5286, 5.4484, 2.0341, 4.8159, 4.5243, 5.0227, 4.9452], + device='cuda:2'), covar=tensor([0.0384, 0.4000, 0.0290, 0.2784, 0.0882, 0.0581, 0.0398, 0.0418], + device='cuda:2'), in_proj_covar=tensor([0.0363, 0.0504, 0.0456, 0.0442, 0.0512, 0.0421, 0.0422, 0.0478], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 02:38:33,190 INFO [train.py:901] (2/4) Epoch 6, batch 4350, loss[loss=0.2934, simple_loss=0.3542, pruned_loss=0.1163, over 8526.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.339, pruned_loss=0.1048, over 1611805.69 frames. ], batch size: 39, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:38:45,512 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4088, 1.8396, 2.8270, 1.0849, 1.8980, 1.6639, 1.5464, 1.7629], + device='cuda:2'), covar=tensor([0.1424, 0.1700, 0.0609, 0.3236, 0.1391, 0.2447, 0.1431, 0.1926], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0459, 0.0536, 0.0543, 0.0588, 0.0529, 0.0443, 0.0597], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:38:55,443 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6191, 1.8427, 1.8862, 1.6268, 0.9068, 2.0580, 0.2020, 1.2668], + device='cuda:2'), covar=tensor([0.2913, 0.2222, 0.0895, 0.2164, 0.6099, 0.0554, 0.5035, 0.2196], + device='cuda:2'), in_proj_covar=tensor([0.0137, 0.0137, 0.0086, 0.0183, 0.0229, 0.0083, 0.0147, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 02:39:00,035 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 02:39:06,569 INFO [train.py:901] (2/4) Epoch 6, batch 4400, loss[loss=0.2918, simple_loss=0.3584, pruned_loss=0.1126, over 8464.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3391, pruned_loss=0.1051, over 1610827.88 frames. ], batch size: 27, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:39:17,274 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.185e+02 3.434e+02 4.206e+02 5.183e+02 1.151e+03, threshold=8.413e+02, percent-clipped=11.0 +2023-02-06 02:39:40,227 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 02:39:42,261 INFO [train.py:901] (2/4) Epoch 6, batch 4450, loss[loss=0.2786, simple_loss=0.3477, pruned_loss=0.1047, over 8448.00 frames. ], tot_loss[loss=0.275, simple_loss=0.339, pruned_loss=0.1055, over 1607199.09 frames. ], batch size: 27, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:39:45,222 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 +2023-02-06 02:39:58,545 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44889.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:40:13,883 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44912.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:40:16,359 INFO [train.py:901] (2/4) Epoch 6, batch 4500, loss[loss=0.3013, simple_loss=0.3461, pruned_loss=0.1283, over 7516.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3388, pruned_loss=0.105, over 1608019.31 frames. ], batch size: 18, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:40:26,435 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 3.100e+02 3.740e+02 5.266e+02 1.703e+03, threshold=7.479e+02, percent-clipped=4.0 +2023-02-06 02:40:31,753 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 02:40:48,245 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9173, 1.4148, 4.4168, 1.6364, 3.1580, 3.4246, 3.8484, 3.8763], + device='cuda:2'), covar=tensor([0.1116, 0.5565, 0.0842, 0.3740, 0.2320, 0.1315, 0.1064, 0.0995], + device='cuda:2'), in_proj_covar=tensor([0.0362, 0.0507, 0.0461, 0.0448, 0.0516, 0.0424, 0.0426, 0.0483], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 02:40:52,040 INFO [train.py:901] (2/4) Epoch 6, batch 4550, loss[loss=0.3109, simple_loss=0.3782, pruned_loss=0.1218, over 8316.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3392, pruned_loss=0.1051, over 1611154.98 frames. ], batch size: 25, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:41:18,800 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45004.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:41:27,059 INFO [train.py:901] (2/4) Epoch 6, batch 4600, loss[loss=0.3039, simple_loss=0.3566, pruned_loss=0.1256, over 6755.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3394, pruned_loss=0.1048, over 1613278.93 frames. ], batch size: 73, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:41:34,832 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:41:36,658 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.977e+02 3.732e+02 4.434e+02 1.135e+03, threshold=7.465e+02, percent-clipped=1.0 +2023-02-06 02:41:50,227 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9134, 2.3380, 1.7418, 2.8675, 1.2251, 1.4785, 2.0004, 2.6801], + device='cuda:2'), covar=tensor([0.1117, 0.1158, 0.1435, 0.0498, 0.1736, 0.2160, 0.1459, 0.0788], + device='cuda:2'), in_proj_covar=tensor([0.0265, 0.0247, 0.0279, 0.0223, 0.0242, 0.0280, 0.0286, 0.0252], + 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') +2023-02-06 02:42:02,765 INFO [train.py:901] (2/4) Epoch 6, batch 4650, loss[loss=0.3126, simple_loss=0.3638, pruned_loss=0.1307, over 7817.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3378, pruned_loss=0.1036, over 1613283.73 frames. ], batch size: 20, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:42:08,404 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45074.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:42:13,626 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3540, 4.3277, 3.9710, 1.8352, 3.8617, 3.8237, 3.9727, 3.4945], + device='cuda:2'), covar=tensor([0.0849, 0.0613, 0.0961, 0.4373, 0.0795, 0.0746, 0.1329, 0.0751], + device='cuda:2'), in_proj_covar=tensor([0.0407, 0.0304, 0.0332, 0.0413, 0.0328, 0.0292, 0.0315, 0.0264], + 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') +2023-02-06 02:42:25,263 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45099.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:42:33,712 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2334, 3.1239, 2.9309, 1.4686, 2.8068, 2.8440, 2.9738, 2.6673], + device='cuda:2'), covar=tensor([0.1208, 0.0802, 0.1198, 0.4251, 0.1094, 0.0978, 0.1416, 0.1013], + device='cuda:2'), in_proj_covar=tensor([0.0407, 0.0303, 0.0334, 0.0412, 0.0329, 0.0293, 0.0314, 0.0263], + 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') +2023-02-06 02:42:36,969 INFO [train.py:901] (2/4) Epoch 6, batch 4700, loss[loss=0.3493, simple_loss=0.4055, pruned_loss=0.1466, over 8328.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3401, pruned_loss=0.1051, over 1614166.60 frames. ], batch size: 26, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:42:46,387 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 3.187e+02 3.833e+02 4.569e+02 1.251e+03, threshold=7.667e+02, percent-clipped=2.0 +2023-02-06 02:43:02,585 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8319, 1.6187, 3.1469, 1.3261, 2.0291, 3.5557, 3.4129, 2.9677], + device='cuda:2'), covar=tensor([0.1115, 0.1412, 0.0423, 0.2164, 0.1049, 0.0276, 0.0475, 0.0720], + device='cuda:2'), in_proj_covar=tensor([0.0242, 0.0270, 0.0228, 0.0269, 0.0238, 0.0215, 0.0252, 0.0284], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 02:43:11,101 INFO [train.py:901] (2/4) Epoch 6, batch 4750, loss[loss=0.2448, simple_loss=0.3145, pruned_loss=0.08757, over 7801.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3408, pruned_loss=0.1052, over 1615383.75 frames. ], batch size: 20, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:43:30,404 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 02:43:31,802 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 02:43:37,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6503, 4.1087, 2.3432, 2.5853, 2.9647, 1.9746, 2.2697, 3.0916], + device='cuda:2'), covar=tensor([0.1690, 0.0267, 0.0920, 0.0826, 0.0659, 0.1199, 0.1145, 0.0969], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0237, 0.0312, 0.0305, 0.0320, 0.0315, 0.0336, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 02:43:37,696 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7700, 2.4182, 4.6036, 1.3353, 3.0567, 2.1747, 1.7886, 2.4006], + device='cuda:2'), covar=tensor([0.1362, 0.1626, 0.0581, 0.2956, 0.1229, 0.2226, 0.1337, 0.2228], + device='cuda:2'), in_proj_covar=tensor([0.0463, 0.0460, 0.0527, 0.0539, 0.0582, 0.0525, 0.0437, 0.0589], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:43:41,057 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2932, 1.7005, 2.8602, 1.1816, 1.9288, 1.6391, 1.4432, 1.5556], + device='cuda:2'), covar=tensor([0.1774, 0.1897, 0.0724, 0.3422, 0.1526, 0.2727, 0.1693, 0.2196], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0462, 0.0530, 0.0541, 0.0584, 0.0526, 0.0439, 0.0591], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:43:46,230 INFO [train.py:901] (2/4) Epoch 6, batch 4800, loss[loss=0.279, simple_loss=0.3529, pruned_loss=0.1026, over 8341.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3401, pruned_loss=0.1052, over 1611328.85 frames. ], batch size: 24, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:43:55,769 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 3.021e+02 3.501e+02 4.623e+02 8.497e+02, threshold=7.001e+02, percent-clipped=1.0 +2023-02-06 02:44:16,232 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45260.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:16,880 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1587, 1.1176, 2.2459, 1.0623, 2.1059, 2.4623, 2.4569, 2.0593], + device='cuda:2'), covar=tensor([0.1020, 0.1177, 0.0475, 0.1846, 0.0580, 0.0379, 0.0536, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0264, 0.0224, 0.0265, 0.0232, 0.0211, 0.0247, 0.0279], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 02:44:20,163 INFO [train.py:901] (2/4) Epoch 6, batch 4850, loss[loss=0.3309, simple_loss=0.3738, pruned_loss=0.144, over 8519.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3394, pruned_loss=0.1053, over 1611293.21 frames. ], batch size: 49, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:44:20,866 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 02:44:32,455 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3553, 1.3607, 1.4404, 1.2934, 0.8963, 1.4896, 0.4168, 1.1631], + device='cuda:2'), covar=tensor([0.2913, 0.1941, 0.0940, 0.2108, 0.5391, 0.0793, 0.5632, 0.2475], + device='cuda:2'), in_proj_covar=tensor([0.0138, 0.0136, 0.0088, 0.0185, 0.0228, 0.0083, 0.0147, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 02:44:32,469 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:35,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45285.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:50,596 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45308.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:56,001 INFO [train.py:901] (2/4) Epoch 6, batch 4900, loss[loss=0.2539, simple_loss=0.3249, pruned_loss=0.0915, over 8347.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3405, pruned_loss=0.1055, over 1615543.48 frames. ], batch size: 26, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:45:04,918 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7880, 1.4398, 5.8710, 2.2279, 5.1530, 4.8854, 5.3424, 5.2331], + device='cuda:2'), covar=tensor([0.0427, 0.4133, 0.0282, 0.2629, 0.0801, 0.0582, 0.0373, 0.0439], + device='cuda:2'), in_proj_covar=tensor([0.0366, 0.0514, 0.0456, 0.0444, 0.0510, 0.0416, 0.0427, 0.0477], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 02:45:05,437 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.896e+02 3.521e+02 4.501e+02 9.960e+02, threshold=7.042e+02, percent-clipped=7.0 +2023-02-06 02:45:30,240 INFO [train.py:901] (2/4) Epoch 6, batch 4950, loss[loss=0.2795, simple_loss=0.3619, pruned_loss=0.09855, over 8108.00 frames. ], tot_loss[loss=0.274, simple_loss=0.339, pruned_loss=0.1045, over 1614464.88 frames. ], batch size: 23, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:45:30,383 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45366.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:46:05,710 INFO [train.py:901] (2/4) Epoch 6, batch 5000, loss[loss=0.2638, simple_loss=0.3398, pruned_loss=0.09388, over 8316.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3395, pruned_loss=0.1049, over 1616265.47 frames. ], batch size: 25, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:46:06,589 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6825, 1.7376, 3.2901, 1.4319, 2.1533, 3.7158, 3.6524, 3.1875], + device='cuda:2'), covar=tensor([0.1220, 0.1357, 0.0362, 0.1987, 0.0889, 0.0248, 0.0434, 0.0620], + device='cuda:2'), in_proj_covar=tensor([0.0236, 0.0261, 0.0222, 0.0264, 0.0230, 0.0210, 0.0246, 0.0275], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 02:46:07,202 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:46:15,104 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.151e+02 3.255e+02 4.005e+02 4.887e+02 1.315e+03, threshold=8.009e+02, percent-clipped=7.0 +2023-02-06 02:46:16,063 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4650, 2.3723, 4.4668, 1.0509, 3.1684, 2.0515, 1.7671, 2.4412], + device='cuda:2'), covar=tensor([0.1719, 0.1881, 0.0681, 0.3690, 0.1301, 0.2614, 0.1611, 0.2340], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0463, 0.0525, 0.0540, 0.0586, 0.0524, 0.0444, 0.0591], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:46:24,042 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:46:40,007 INFO [train.py:901] (2/4) Epoch 6, batch 5050, loss[loss=0.2853, simple_loss=0.3599, pruned_loss=0.1054, over 8479.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3398, pruned_loss=0.1049, over 1616971.76 frames. ], batch size: 29, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:46:44,903 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0889, 3.4820, 2.3261, 2.4944, 2.8010, 1.8599, 2.5417, 2.8147], + device='cuda:2'), covar=tensor([0.1093, 0.0300, 0.0726, 0.0639, 0.0467, 0.1030, 0.0690, 0.0703], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0232, 0.0308, 0.0303, 0.0315, 0.0315, 0.0333, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 02:46:58,883 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 02:47:14,063 INFO [train.py:901] (2/4) Epoch 6, batch 5100, loss[loss=0.2093, simple_loss=0.2721, pruned_loss=0.07329, over 6766.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3401, pruned_loss=0.1051, over 1615120.49 frames. ], batch size: 15, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:47:24,714 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.842e+02 3.419e+02 4.219e+02 7.828e+02, threshold=6.837e+02, percent-clipped=0.0 +2023-02-06 02:47:26,968 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:47:43,445 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45558.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:47:49,315 INFO [train.py:901] (2/4) Epoch 6, batch 5150, loss[loss=0.3166, simple_loss=0.3693, pruned_loss=0.1319, over 8258.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3405, pruned_loss=0.1056, over 1611102.67 frames. ], batch size: 24, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:48:23,749 INFO [train.py:901] (2/4) Epoch 6, batch 5200, loss[loss=0.2239, simple_loss=0.2925, pruned_loss=0.0777, over 5974.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3398, pruned_loss=0.1045, over 1612375.79 frames. ], batch size: 13, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:48:34,004 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 3.204e+02 4.015e+02 4.654e+02 8.708e+02, threshold=8.029e+02, percent-clipped=4.0 +2023-02-06 02:48:57,693 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 02:48:59,757 INFO [train.py:901] (2/4) Epoch 6, batch 5250, loss[loss=0.3028, simple_loss=0.3635, pruned_loss=0.1211, over 8485.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3417, pruned_loss=0.1057, over 1615746.36 frames. ], batch size: 26, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:49:30,224 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45710.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:49:34,102 INFO [train.py:901] (2/4) Epoch 6, batch 5300, loss[loss=0.2564, simple_loss=0.3083, pruned_loss=0.1023, over 7698.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3402, pruned_loss=0.1059, over 1610645.59 frames. ], batch size: 18, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:49:36,947 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45720.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:49:43,549 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.935e+02 3.437e+02 4.667e+02 1.283e+03, threshold=6.874e+02, percent-clipped=3.0 +2023-02-06 02:50:09,978 INFO [train.py:901] (2/4) Epoch 6, batch 5350, loss[loss=0.2472, simple_loss=0.3334, pruned_loss=0.08052, over 8350.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3389, pruned_loss=0.1044, over 1611373.93 frames. ], batch size: 24, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:50:25,442 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:27,340 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45792.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:36,781 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3582, 1.8687, 3.2300, 2.5127, 2.6830, 2.0049, 1.5349, 1.3186], + device='cuda:2'), covar=tensor([0.2477, 0.2709, 0.0604, 0.1506, 0.1322, 0.1401, 0.1198, 0.2824], + device='cuda:2'), in_proj_covar=tensor([0.0803, 0.0739, 0.0628, 0.0725, 0.0837, 0.0675, 0.0638, 0.0673], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 02:50:42,757 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45814.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:42,777 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45814.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:43,910 INFO [train.py:901] (2/4) Epoch 6, batch 5400, loss[loss=0.2887, simple_loss=0.3556, pruned_loss=0.111, over 7242.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3407, pruned_loss=0.1057, over 1610119.59 frames. ], batch size: 71, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:50:49,968 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45825.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:50:53,703 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.842e+02 3.609e+02 4.644e+02 1.367e+03, threshold=7.218e+02, percent-clipped=2.0 +2023-02-06 02:50:59,048 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:51:17,300 INFO [train.py:901] (2/4) Epoch 6, batch 5450, loss[loss=0.2399, simple_loss=0.3023, pruned_loss=0.08873, over 7931.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.34, pruned_loss=0.1058, over 1605795.20 frames. ], batch size: 20, lr: 1.25e-02, grad_scale: 8.0 +2023-02-06 02:51:26,215 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:51:40,421 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.43 vs. limit=5.0 +2023-02-06 02:51:47,554 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 02:51:52,372 INFO [train.py:901] (2/4) Epoch 6, batch 5500, loss[loss=0.2886, simple_loss=0.3655, pruned_loss=0.1059, over 8348.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3399, pruned_loss=0.1052, over 1612125.25 frames. ], batch size: 26, lr: 1.25e-02, grad_scale: 8.0 +2023-02-06 02:51:54,758 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 +2023-02-06 02:52:03,079 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.821e+02 3.418e+02 4.385e+02 9.516e+02, threshold=6.836e+02, percent-clipped=4.0 +2023-02-06 02:52:25,879 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3466, 1.4675, 2.2777, 1.1757, 1.5879, 1.6457, 1.4134, 1.3819], + device='cuda:2'), covar=tensor([0.1433, 0.1969, 0.0720, 0.2976, 0.1289, 0.2423, 0.1511, 0.1663], + device='cuda:2'), in_proj_covar=tensor([0.0465, 0.0460, 0.0526, 0.0537, 0.0583, 0.0528, 0.0442, 0.0582], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:52:27,017 INFO [train.py:901] (2/4) Epoch 6, batch 5550, loss[loss=0.2377, simple_loss=0.307, pruned_loss=0.08415, over 8137.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3393, pruned_loss=0.1049, over 1609710.45 frames. ], batch size: 22, lr: 1.25e-02, grad_scale: 8.0 +2023-02-06 02:53:03,507 INFO [train.py:901] (2/4) Epoch 6, batch 5600, loss[loss=0.2318, simple_loss=0.3124, pruned_loss=0.07563, over 8295.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3395, pruned_loss=0.1045, over 1612667.41 frames. ], batch size: 23, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:53:13,368 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 2.809e+02 3.495e+02 4.670e+02 1.291e+03, threshold=6.989e+02, percent-clipped=6.0 +2023-02-06 02:53:36,015 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46064.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:53:37,306 INFO [train.py:901] (2/4) Epoch 6, batch 5650, loss[loss=0.2684, simple_loss=0.3339, pruned_loss=0.1014, over 8027.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3402, pruned_loss=0.1055, over 1613639.95 frames. ], batch size: 22, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:53:47,425 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46081.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:53:51,178 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 02:54:04,762 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46106.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:54:12,530 INFO [train.py:901] (2/4) Epoch 6, batch 5700, loss[loss=0.2588, simple_loss=0.3259, pruned_loss=0.09587, over 5094.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3393, pruned_loss=0.1047, over 1611246.23 frames. ], batch size: 11, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:54:22,543 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.829e+02 3.489e+02 4.392e+02 1.030e+03, threshold=6.978e+02, percent-clipped=3.0 +2023-02-06 02:54:25,977 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:54:35,703 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 02:54:46,331 INFO [train.py:901] (2/4) Epoch 6, batch 5750, loss[loss=0.2254, simple_loss=0.2922, pruned_loss=0.0793, over 8135.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3383, pruned_loss=0.1044, over 1608273.12 frames. ], batch size: 22, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:54:53,639 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 02:54:55,251 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46179.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:54:56,134 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 02:55:00,722 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0169, 1.5900, 1.3732, 1.2344, 1.0668, 1.2708, 1.6017, 1.2674], + device='cuda:2'), covar=tensor([0.0610, 0.1219, 0.1903, 0.1530, 0.0655, 0.1655, 0.0780, 0.0655], + device='cuda:2'), in_proj_covar=tensor([0.0122, 0.0170, 0.0212, 0.0174, 0.0121, 0.0179, 0.0136, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 02:55:21,181 INFO [train.py:901] (2/4) Epoch 6, batch 5800, loss[loss=0.3184, simple_loss=0.387, pruned_loss=0.1248, over 8506.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3395, pruned_loss=0.1049, over 1605597.27 frames. ], batch size: 28, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:55:24,800 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:55:32,629 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.949e+02 3.358e+02 4.338e+02 9.471e+02, threshold=6.717e+02, percent-clipped=1.0 +2023-02-06 02:55:45,833 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46251.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:55:49,917 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2296, 1.4659, 2.2489, 1.1166, 1.5895, 1.5743, 1.4068, 1.4211], + device='cuda:2'), covar=tensor([0.1550, 0.1843, 0.0677, 0.2947, 0.1281, 0.2307, 0.1492, 0.1653], + device='cuda:2'), in_proj_covar=tensor([0.0462, 0.0460, 0.0525, 0.0535, 0.0580, 0.0517, 0.0436, 0.0581], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:55:55,786 INFO [train.py:901] (2/4) Epoch 6, batch 5850, loss[loss=0.2493, simple_loss=0.3089, pruned_loss=0.09485, over 7926.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3375, pruned_loss=0.1037, over 1606125.03 frames. ], batch size: 20, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:55:59,171 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0055, 1.5682, 4.1960, 1.4821, 3.4590, 3.3759, 3.6787, 3.6382], + device='cuda:2'), covar=tensor([0.0506, 0.3906, 0.0493, 0.3283, 0.1332, 0.0884, 0.0614, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0512, 0.0460, 0.0446, 0.0511, 0.0422, 0.0434, 0.0473], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 02:56:14,983 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46294.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:56:29,476 INFO [train.py:901] (2/4) Epoch 6, batch 5900, loss[loss=0.3348, simple_loss=0.3574, pruned_loss=0.1561, over 7658.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3364, pruned_loss=0.1028, over 1609122.38 frames. ], batch size: 19, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:56:39,482 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 3.022e+02 3.849e+02 5.141e+02 8.536e+02, threshold=7.697e+02, percent-clipped=7.0 +2023-02-06 02:56:43,703 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:57:04,663 INFO [train.py:901] (2/4) Epoch 6, batch 5950, loss[loss=0.3053, simple_loss=0.3611, pruned_loss=0.1247, over 8469.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3369, pruned_loss=0.1029, over 1610961.08 frames. ], batch size: 27, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:57:06,801 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4946, 1.7443, 1.5444, 1.3064, 1.1858, 1.3270, 1.8957, 1.7002], + device='cuda:2'), covar=tensor([0.0473, 0.1171, 0.1637, 0.1331, 0.0637, 0.1577, 0.0722, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0122, 0.0171, 0.0211, 0.0175, 0.0121, 0.0181, 0.0136, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 02:57:38,437 INFO [train.py:901] (2/4) Epoch 6, batch 6000, loss[loss=0.3002, simple_loss=0.3624, pruned_loss=0.1191, over 6826.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3366, pruned_loss=0.1027, over 1611417.33 frames. ], batch size: 15, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:57:38,437 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 02:57:50,764 INFO [train.py:935] (2/4) Epoch 6, validation: loss=0.2127, simple_loss=0.3094, pruned_loss=0.05799, over 944034.00 frames. +2023-02-06 02:57:50,765 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 02:58:01,258 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.683e+02 3.226e+02 4.100e+02 1.140e+03, threshold=6.453e+02, percent-clipped=1.0 +2023-02-06 02:58:04,340 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:58:21,767 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46460.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:58:25,653 INFO [train.py:901] (2/4) Epoch 6, batch 6050, loss[loss=0.2984, simple_loss=0.3391, pruned_loss=0.1289, over 7717.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3366, pruned_loss=0.1027, over 1611486.53 frames. ], batch size: 18, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:58:56,092 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:59:02,119 INFO [train.py:901] (2/4) Epoch 6, batch 6100, loss[loss=0.2476, simple_loss=0.3072, pruned_loss=0.09403, over 7939.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3366, pruned_loss=0.1027, over 1610855.19 frames. ], batch size: 20, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:59:12,643 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 3.046e+02 3.657e+02 4.398e+02 9.620e+02, threshold=7.315e+02, percent-clipped=4.0 +2023-02-06 02:59:13,539 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:59:24,604 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 02:59:29,768 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4198, 2.0046, 3.1396, 1.1930, 2.1559, 1.8956, 1.6190, 1.8380], + device='cuda:2'), covar=tensor([0.1518, 0.1649, 0.0588, 0.3219, 0.1320, 0.2251, 0.1362, 0.2016], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0474, 0.0540, 0.0560, 0.0597, 0.0533, 0.0446, 0.0597], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 02:59:37,763 INFO [train.py:901] (2/4) Epoch 6, batch 6150, loss[loss=0.2316, simple_loss=0.2828, pruned_loss=0.09017, over 6366.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3373, pruned_loss=0.1034, over 1607979.05 frames. ], batch size: 14, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:59:56,265 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:11,863 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46613.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:13,806 INFO [train.py:901] (2/4) Epoch 6, batch 6200, loss[loss=0.2663, simple_loss=0.3252, pruned_loss=0.1037, over 8031.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3378, pruned_loss=0.1042, over 1605955.76 frames. ], batch size: 22, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 03:00:14,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46617.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:16,773 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:24,152 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 3.051e+02 3.861e+02 4.926e+02 1.016e+03, threshold=7.722e+02, percent-clipped=3.0 +2023-02-06 03:00:28,972 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46638.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:48,695 INFO [train.py:901] (2/4) Epoch 6, batch 6250, loss[loss=0.2573, simple_loss=0.3402, pruned_loss=0.08725, over 8506.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.337, pruned_loss=0.1036, over 1606796.18 frames. ], batch size: 39, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 03:01:22,871 INFO [train.py:901] (2/4) Epoch 6, batch 6300, loss[loss=0.2699, simple_loss=0.3358, pruned_loss=0.102, over 7795.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3376, pruned_loss=0.1038, over 1608182.07 frames. ], batch size: 20, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 03:01:34,435 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.734e+02 3.399e+02 4.377e+02 1.449e+03, threshold=6.797e+02, percent-clipped=4.0 +2023-02-06 03:01:38,092 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1689, 1.1568, 1.2541, 1.1904, 0.8213, 1.2494, 0.1882, 0.9612], + device='cuda:2'), covar=tensor([0.3202, 0.2029, 0.1025, 0.1744, 0.5030, 0.0931, 0.4027, 0.1824], + device='cuda:2'), in_proj_covar=tensor([0.0135, 0.0132, 0.0082, 0.0182, 0.0223, 0.0082, 0.0144, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:01:49,382 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:01:57,776 INFO [train.py:901] (2/4) Epoch 6, batch 6350, loss[loss=0.3177, simple_loss=0.3829, pruned_loss=0.1262, over 8465.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3381, pruned_loss=0.1041, over 1610534.83 frames. ], batch size: 27, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:02:32,255 INFO [train.py:901] (2/4) Epoch 6, batch 6400, loss[loss=0.2736, simple_loss=0.3478, pruned_loss=0.09976, over 8458.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3378, pruned_loss=0.1041, over 1609082.77 frames. ], batch size: 27, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:02:35,378 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 03:02:41,187 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:02:43,104 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.744e+02 3.578e+02 4.396e+02 9.504e+02, threshold=7.157e+02, percent-clipped=5.0 +2023-02-06 03:03:07,334 INFO [train.py:901] (2/4) Epoch 6, batch 6450, loss[loss=0.2908, simple_loss=0.354, pruned_loss=0.1138, over 8451.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3375, pruned_loss=0.1034, over 1609641.86 frames. ], batch size: 27, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:03:40,297 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5898, 5.6445, 4.9054, 2.0435, 4.9556, 5.2170, 5.2684, 4.7987], + device='cuda:2'), covar=tensor([0.0667, 0.0394, 0.0804, 0.4954, 0.0592, 0.0539, 0.0842, 0.0548], + device='cuda:2'), in_proj_covar=tensor([0.0397, 0.0304, 0.0327, 0.0411, 0.0315, 0.0291, 0.0304, 0.0263], + 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') +2023-02-06 03:03:41,565 INFO [train.py:901] (2/4) Epoch 6, batch 6500, loss[loss=0.2912, simple_loss=0.3639, pruned_loss=0.1093, over 8466.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3375, pruned_loss=0.1032, over 1610088.72 frames. ], batch size: 25, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:03:51,600 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.994e+02 3.001e+02 3.759e+02 4.377e+02 1.086e+03, threshold=7.517e+02, percent-clipped=1.0 +2023-02-06 03:04:09,448 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:04:14,548 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46964.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:04:15,764 INFO [train.py:901] (2/4) Epoch 6, batch 6550, loss[loss=0.2843, simple_loss=0.3657, pruned_loss=0.1014, over 8450.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3372, pruned_loss=0.1029, over 1611121.84 frames. ], batch size: 29, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:04:37,526 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 03:04:45,809 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47009.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:04:50,976 INFO [train.py:901] (2/4) Epoch 6, batch 6600, loss[loss=0.2619, simple_loss=0.3379, pruned_loss=0.09299, over 8463.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3378, pruned_loss=0.104, over 1611406.45 frames. ], batch size: 25, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:04:56,456 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 03:05:01,160 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.072e+02 2.927e+02 3.687e+02 4.772e+02 1.123e+03, threshold=7.374e+02, percent-clipped=4.0 +2023-02-06 03:05:03,298 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47034.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:05:08,620 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3614, 2.7181, 1.8213, 2.0352, 2.2348, 1.5225, 1.8829, 2.0997], + device='cuda:2'), covar=tensor([0.1194, 0.0252, 0.0823, 0.0536, 0.0501, 0.1041, 0.0856, 0.0746], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0233, 0.0309, 0.0302, 0.0312, 0.0315, 0.0340, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 03:05:25,535 INFO [train.py:901] (2/4) Epoch 6, batch 6650, loss[loss=0.2445, simple_loss=0.3085, pruned_loss=0.09022, over 7530.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3377, pruned_loss=0.1037, over 1608676.33 frames. ], batch size: 18, lr: 1.23e-02, grad_scale: 4.0 +2023-02-06 03:05:30,596 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:05:35,435 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47079.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:05:39,577 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9277, 1.6001, 2.2564, 1.8872, 2.0059, 1.7235, 1.3826, 0.6603], + device='cuda:2'), covar=tensor([0.2398, 0.2482, 0.0650, 0.1272, 0.0992, 0.1251, 0.1178, 0.2400], + device='cuda:2'), in_proj_covar=tensor([0.0798, 0.0735, 0.0638, 0.0722, 0.0824, 0.0676, 0.0635, 0.0674], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:05:59,664 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7566, 2.0446, 1.6288, 2.5734, 1.3137, 1.3265, 1.6358, 2.0460], + device='cuda:2'), covar=tensor([0.1103, 0.1069, 0.1659, 0.0628, 0.1498, 0.2154, 0.1635, 0.1169], + device='cuda:2'), in_proj_covar=tensor([0.0267, 0.0247, 0.0283, 0.0227, 0.0246, 0.0274, 0.0282, 0.0247], + 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') +2023-02-06 03:06:00,868 INFO [train.py:901] (2/4) Epoch 6, batch 6700, loss[loss=0.2558, simple_loss=0.3177, pruned_loss=0.09701, over 8246.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3387, pruned_loss=0.1039, over 1612248.21 frames. ], batch size: 22, lr: 1.23e-02, grad_scale: 4.0 +2023-02-06 03:06:12,492 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.777e+02 3.640e+02 4.922e+02 1.093e+03, threshold=7.281e+02, percent-clipped=6.0 +2023-02-06 03:06:25,941 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 03:06:34,840 INFO [train.py:901] (2/4) Epoch 6, batch 6750, loss[loss=0.2843, simple_loss=0.3535, pruned_loss=0.1075, over 8501.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3385, pruned_loss=0.1034, over 1613272.38 frames. ], batch size: 28, lr: 1.23e-02, grad_scale: 4.0 +2023-02-06 03:06:38,950 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47172.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:07:10,709 INFO [train.py:901] (2/4) Epoch 6, batch 6800, loss[loss=0.2567, simple_loss=0.3214, pruned_loss=0.09601, over 7665.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3381, pruned_loss=0.1033, over 1612089.62 frames. ], batch size: 19, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:07:12,676 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 03:07:12,931 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7075, 2.2069, 3.6891, 2.6741, 2.9644, 2.1579, 1.7361, 1.7751], + device='cuda:2'), covar=tensor([0.1992, 0.2380, 0.0556, 0.1443, 0.1311, 0.1245, 0.1141, 0.2581], + device='cuda:2'), in_proj_covar=tensor([0.0795, 0.0731, 0.0637, 0.0723, 0.0824, 0.0673, 0.0635, 0.0674], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:07:21,205 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.942e+02 3.591e+02 4.804e+02 1.528e+03, threshold=7.182e+02, percent-clipped=7.0 +2023-02-06 03:07:34,546 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7860, 3.7310, 3.3625, 1.5981, 3.2332, 3.2899, 3.4984, 2.9329], + device='cuda:2'), covar=tensor([0.0863, 0.0632, 0.0943, 0.4848, 0.0922, 0.0916, 0.1271, 0.0997], + device='cuda:2'), in_proj_covar=tensor([0.0400, 0.0303, 0.0326, 0.0414, 0.0319, 0.0291, 0.0304, 0.0264], + 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') +2023-02-06 03:07:40,019 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:07:44,568 INFO [train.py:901] (2/4) Epoch 6, batch 6850, loss[loss=0.3, simple_loss=0.3569, pruned_loss=0.1216, over 7650.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.339, pruned_loss=0.1037, over 1616129.97 frames. ], batch size: 19, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:07:52,848 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5948, 1.7935, 2.0161, 1.6308, 1.1053, 1.9683, 0.4314, 1.2187], + device='cuda:2'), covar=tensor([0.3174, 0.2424, 0.0752, 0.2044, 0.5440, 0.0825, 0.6752, 0.2690], + device='cuda:2'), in_proj_covar=tensor([0.0137, 0.0134, 0.0082, 0.0182, 0.0224, 0.0083, 0.0145, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:07:58,970 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:00,908 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 03:08:03,824 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3042, 2.4440, 1.9384, 3.0560, 1.4289, 1.4860, 2.0959, 2.5918], + device='cuda:2'), covar=tensor([0.0764, 0.1040, 0.1302, 0.0410, 0.1468, 0.1756, 0.1345, 0.0986], + device='cuda:2'), in_proj_covar=tensor([0.0265, 0.0249, 0.0278, 0.0224, 0.0247, 0.0272, 0.0279, 0.0247], + 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') +2023-02-06 03:08:11,541 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3984, 1.7717, 2.8862, 1.1847, 1.9997, 1.8008, 1.5113, 1.6944], + device='cuda:2'), covar=tensor([0.1491, 0.1702, 0.0635, 0.3058, 0.1294, 0.2180, 0.1500, 0.1927], + device='cuda:2'), in_proj_covar=tensor([0.0463, 0.0466, 0.0535, 0.0549, 0.0590, 0.0523, 0.0449, 0.0594], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:08:16,339 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-06 03:08:19,247 INFO [train.py:901] (2/4) Epoch 6, batch 6900, loss[loss=0.2813, simple_loss=0.343, pruned_loss=0.1098, over 7983.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3394, pruned_loss=0.1039, over 1619563.48 frames. ], batch size: 21, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:08:22,399 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 03:08:28,205 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:30,616 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 2.873e+02 3.537e+02 4.379e+02 9.664e+02, threshold=7.075e+02, percent-clipped=2.0 +2023-02-06 03:08:32,861 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47335.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:44,864 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:45,931 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 03:08:49,792 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47360.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:54,240 INFO [train.py:901] (2/4) Epoch 6, batch 6950, loss[loss=0.2834, simple_loss=0.351, pruned_loss=0.1079, over 8258.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.341, pruned_loss=0.1046, over 1618341.86 frames. ], batch size: 48, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:09:09,778 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 03:09:15,147 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47397.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:09:28,479 INFO [train.py:901] (2/4) Epoch 6, batch 7000, loss[loss=0.276, simple_loss=0.3466, pruned_loss=0.1027, over 8542.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3388, pruned_loss=0.1031, over 1615145.20 frames. ], batch size: 49, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:09:34,305 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 03:09:39,919 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.784e+02 3.553e+02 4.437e+02 1.281e+03, threshold=7.106e+02, percent-clipped=4.0 +2023-02-06 03:10:03,567 INFO [train.py:901] (2/4) Epoch 6, batch 7050, loss[loss=0.205, simple_loss=0.2868, pruned_loss=0.0616, over 7931.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3383, pruned_loss=0.1032, over 1612535.63 frames. ], batch size: 20, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:10:25,054 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 03:10:37,634 INFO [train.py:901] (2/4) Epoch 6, batch 7100, loss[loss=0.2547, simple_loss=0.3204, pruned_loss=0.09453, over 7816.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3379, pruned_loss=0.1032, over 1608984.85 frames. ], batch size: 20, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:10:48,827 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.154e+02 3.207e+02 3.842e+02 5.073e+02 1.424e+03, threshold=7.684e+02, percent-clipped=2.0 +2023-02-06 03:10:56,276 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:11:12,602 INFO [train.py:901] (2/4) Epoch 6, batch 7150, loss[loss=0.2471, simple_loss=0.3153, pruned_loss=0.08944, over 8136.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3381, pruned_loss=0.103, over 1611308.16 frames. ], batch size: 22, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:11:14,091 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47568.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:11:34,465 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1083, 2.3011, 1.8137, 2.6763, 1.2988, 1.5090, 1.7030, 2.4205], + device='cuda:2'), covar=tensor([0.0830, 0.0979, 0.1280, 0.0477, 0.1546, 0.1855, 0.1461, 0.0819], + device='cuda:2'), in_proj_covar=tensor([0.0267, 0.0250, 0.0278, 0.0226, 0.0248, 0.0270, 0.0279, 0.0245], + 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') +2023-02-06 03:11:37,948 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:11:46,767 INFO [train.py:901] (2/4) Epoch 6, batch 7200, loss[loss=0.2403, simple_loss=0.3175, pruned_loss=0.08154, over 8249.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3391, pruned_loss=0.1038, over 1615815.75 frames. ], batch size: 24, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:11:57,335 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6593, 1.9559, 2.1566, 1.5367, 0.9092, 2.1106, 0.2951, 1.3581], + device='cuda:2'), covar=tensor([0.3487, 0.2000, 0.0627, 0.2271, 0.6350, 0.0571, 0.4603, 0.1940], + device='cuda:2'), in_proj_covar=tensor([0.0135, 0.0132, 0.0080, 0.0176, 0.0220, 0.0080, 0.0140, 0.0134], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:11:57,791 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.983e+02 3.737e+02 4.630e+02 8.445e+02, threshold=7.473e+02, percent-clipped=4.0 +2023-02-06 03:12:22,028 INFO [train.py:901] (2/4) Epoch 6, batch 7250, loss[loss=0.275, simple_loss=0.343, pruned_loss=0.1035, over 8578.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.34, pruned_loss=0.1042, over 1619203.58 frames. ], batch size: 34, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:12:56,481 INFO [train.py:901] (2/4) Epoch 6, batch 7300, loss[loss=0.275, simple_loss=0.3459, pruned_loss=0.102, over 8182.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3399, pruned_loss=0.1039, over 1618279.59 frames. ], batch size: 23, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:12:57,902 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:13:07,209 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.071e+02 3.696e+02 4.839e+02 1.031e+03, threshold=7.393e+02, percent-clipped=2.0 +2023-02-06 03:13:13,283 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47741.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:13:23,400 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:13:30,053 INFO [train.py:901] (2/4) Epoch 6, batch 7350, loss[loss=0.2896, simple_loss=0.3492, pruned_loss=0.115, over 8134.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3402, pruned_loss=0.1043, over 1617872.14 frames. ], batch size: 22, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:13:48,823 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 03:13:59,322 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:14:05,871 INFO [train.py:901] (2/4) Epoch 6, batch 7400, loss[loss=0.2657, simple_loss=0.3332, pruned_loss=0.09908, over 7191.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3401, pruned_loss=0.1048, over 1615431.56 frames. ], batch size: 72, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:14:08,012 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 03:14:13,519 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47827.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:14:17,406 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 3.124e+02 3.904e+02 4.877e+02 9.892e+02, threshold=7.808e+02, percent-clipped=5.0 +2023-02-06 03:14:33,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:14:35,676 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0074, 1.2468, 4.2514, 1.5331, 3.6463, 3.6007, 3.8745, 3.6952], + device='cuda:2'), covar=tensor([0.0455, 0.3882, 0.0384, 0.2847, 0.1100, 0.0695, 0.0489, 0.0551], + device='cuda:2'), in_proj_covar=tensor([0.0365, 0.0503, 0.0448, 0.0444, 0.0500, 0.0418, 0.0423, 0.0468], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-06 03:14:40,241 INFO [train.py:901] (2/4) Epoch 6, batch 7450, loss[loss=0.2576, simple_loss=0.3227, pruned_loss=0.09631, over 8288.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3392, pruned_loss=0.1037, over 1617233.18 frames. ], batch size: 23, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:14:46,243 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 03:15:15,316 INFO [train.py:901] (2/4) Epoch 6, batch 7500, loss[loss=0.2795, simple_loss=0.3494, pruned_loss=0.1048, over 8493.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3402, pruned_loss=0.1046, over 1618859.68 frames. ], batch size: 29, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:15:25,968 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.098e+02 3.102e+02 3.706e+02 4.699e+02 1.511e+03, threshold=7.412e+02, percent-clipped=9.0 +2023-02-06 03:15:48,132 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4475, 1.7171, 1.8582, 1.5584, 1.0538, 1.8792, 0.3011, 1.2204], + device='cuda:2'), covar=tensor([0.3677, 0.1691, 0.0736, 0.1846, 0.5031, 0.0607, 0.3883, 0.2045], + device='cuda:2'), in_proj_covar=tensor([0.0138, 0.0133, 0.0081, 0.0182, 0.0227, 0.0082, 0.0143, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:15:49,281 INFO [train.py:901] (2/4) Epoch 6, batch 7550, loss[loss=0.2642, simple_loss=0.3285, pruned_loss=0.09996, over 7663.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3381, pruned_loss=0.1038, over 1613895.29 frames. ], batch size: 19, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:15:54,925 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:16:11,855 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:16:24,742 INFO [train.py:901] (2/4) Epoch 6, batch 7600, loss[loss=0.3025, simple_loss=0.3637, pruned_loss=0.1206, over 7116.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3392, pruned_loss=0.1045, over 1613116.27 frames. ], batch size: 71, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:16:32,091 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.36 vs. limit=5.0 +2023-02-06 03:16:34,118 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.07 vs. limit=5.0 +2023-02-06 03:16:37,192 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.897e+02 3.536e+02 4.611e+02 2.294e+03, threshold=7.072e+02, percent-clipped=5.0 +2023-02-06 03:17:01,525 INFO [train.py:901] (2/4) Epoch 6, batch 7650, loss[loss=0.3069, simple_loss=0.3808, pruned_loss=0.1165, over 8495.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.34, pruned_loss=0.1046, over 1615310.93 frames. ], batch size: 26, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:17:17,556 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48090.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:21,566 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5678, 2.0045, 2.2013, 1.0341, 2.2458, 1.3659, 0.5907, 1.7661], + device='cuda:2'), covar=tensor([0.0275, 0.0151, 0.0134, 0.0236, 0.0152, 0.0414, 0.0379, 0.0133], + device='cuda:2'), in_proj_covar=tensor([0.0337, 0.0248, 0.0205, 0.0300, 0.0240, 0.0383, 0.0314, 0.0286], + device='cuda:2'), out_proj_covar=tensor([1.1077e-04, 7.9298e-05, 6.4561e-05, 9.5806e-05, 7.7792e-05, 1.3366e-04, + 1.0300e-04, 9.2505e-05], device='cuda:2') +2023-02-06 03:17:24,183 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48100.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:32,465 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48112.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:34,938 INFO [train.py:901] (2/4) Epoch 6, batch 7700, loss[loss=0.2762, simple_loss=0.3468, pruned_loss=0.1028, over 8575.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3392, pruned_loss=0.1044, over 1614094.04 frames. ], batch size: 31, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:17:46,047 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 2.821e+02 3.617e+02 4.667e+02 9.808e+02, threshold=7.234e+02, percent-clipped=3.0 +2023-02-06 03:17:50,865 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:54,512 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 03:17:57,319 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 03:17:59,319 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:18:10,116 INFO [train.py:901] (2/4) Epoch 6, batch 7750, loss[loss=0.203, simple_loss=0.283, pruned_loss=0.06146, over 8082.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3385, pruned_loss=0.104, over 1616316.84 frames. ], batch size: 21, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:18:13,410 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48171.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:18:43,313 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48215.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:18:43,802 INFO [train.py:901] (2/4) Epoch 6, batch 7800, loss[loss=0.2298, simple_loss=0.2987, pruned_loss=0.08046, over 7922.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3398, pruned_loss=0.1047, over 1617348.54 frames. ], batch size: 20, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:18:53,444 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48230.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:18:54,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.145e+02 3.053e+02 3.731e+02 4.789e+02 1.133e+03, threshold=7.462e+02, percent-clipped=3.0 +2023-02-06 03:19:04,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4787, 1.7952, 2.8291, 1.1482, 1.8963, 1.6771, 1.5217, 1.7031], + device='cuda:2'), covar=tensor([0.1408, 0.1733, 0.0589, 0.3184, 0.1380, 0.2401, 0.1428, 0.1896], + device='cuda:2'), in_proj_covar=tensor([0.0465, 0.0466, 0.0528, 0.0547, 0.0593, 0.0530, 0.0447, 0.0588], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:19:16,578 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48265.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:19:17,077 INFO [train.py:901] (2/4) Epoch 6, batch 7850, loss[loss=0.2736, simple_loss=0.3466, pruned_loss=0.1003, over 8357.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3396, pruned_loss=0.1044, over 1616119.00 frames. ], batch size: 24, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:19:30,594 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48286.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:19:51,008 INFO [train.py:901] (2/4) Epoch 6, batch 7900, loss[loss=0.2697, simple_loss=0.3401, pruned_loss=0.09961, over 8355.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3384, pruned_loss=0.1032, over 1617128.01 frames. ], batch size: 24, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:20:01,871 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.927e+02 3.494e+02 4.326e+02 7.205e+02, threshold=6.988e+02, percent-clipped=0.0 +2023-02-06 03:20:06,675 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3155, 1.4282, 1.2687, 1.9225, 0.9855, 1.1449, 1.2948, 1.4613], + device='cuda:2'), covar=tensor([0.1091, 0.1064, 0.1575, 0.0613, 0.1220, 0.1957, 0.1062, 0.0967], + device='cuda:2'), in_proj_covar=tensor([0.0263, 0.0241, 0.0275, 0.0223, 0.0239, 0.0267, 0.0273, 0.0243], + 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') +2023-02-06 03:20:09,322 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:20:25,105 INFO [train.py:901] (2/4) Epoch 6, batch 7950, loss[loss=0.3076, simple_loss=0.3753, pruned_loss=0.1199, over 8195.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3409, pruned_loss=0.1047, over 1619601.84 frames. ], batch size: 23, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:20:37,998 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7784, 1.4604, 3.2550, 1.3314, 2.2625, 3.5901, 3.4969, 3.0057], + device='cuda:2'), covar=tensor([0.1050, 0.1491, 0.0330, 0.2011, 0.0753, 0.0270, 0.0400, 0.0653], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0272, 0.0223, 0.0265, 0.0233, 0.0209, 0.0257, 0.0276], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 03:20:58,691 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48415.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:20:59,245 INFO [train.py:901] (2/4) Epoch 6, batch 8000, loss[loss=0.2442, simple_loss=0.3254, pruned_loss=0.08146, over 8356.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3389, pruned_loss=0.1033, over 1618825.89 frames. ], batch size: 24, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:21:10,362 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.174e+02 2.873e+02 3.488e+02 4.217e+02 8.104e+02, threshold=6.977e+02, percent-clipped=2.0 +2023-02-06 03:21:11,764 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:21:16,772 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48441.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:21:33,916 INFO [train.py:901] (2/4) Epoch 6, batch 8050, loss[loss=0.2066, simple_loss=0.2824, pruned_loss=0.06542, over 7545.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3364, pruned_loss=0.103, over 1587867.38 frames. ], batch size: 18, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:21:37,657 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48471.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:21:48,692 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5522, 1.9719, 2.0689, 1.0348, 2.2025, 1.3784, 0.4989, 1.7963], + device='cuda:2'), covar=tensor([0.0310, 0.0173, 0.0143, 0.0297, 0.0157, 0.0460, 0.0455, 0.0134], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0254, 0.0205, 0.0305, 0.0240, 0.0391, 0.0316, 0.0291], + device='cuda:2'), out_proj_covar=tensor([1.1342e-04, 8.1646e-05, 6.4531e-05, 9.7245e-05, 7.7309e-05, 1.3615e-04, + 1.0318e-04, 9.3782e-05], device='cuda:2') +2023-02-06 03:21:54,571 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48496.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:22:07,192 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 03:22:10,939 INFO [train.py:901] (2/4) Epoch 7, batch 0, loss[loss=0.2699, simple_loss=0.3259, pruned_loss=0.1069, over 7538.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3259, pruned_loss=0.1069, over 7538.00 frames. ], batch size: 18, lr: 1.14e-02, grad_scale: 8.0 +2023-02-06 03:22:10,940 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 03:22:22,761 INFO [train.py:935] (2/4) Epoch 7, validation: loss=0.2113, simple_loss=0.3091, pruned_loss=0.05678, over 944034.00 frames. +2023-02-06 03:22:22,763 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 03:22:28,413 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6705, 1.8911, 1.5661, 2.4143, 1.0188, 1.4724, 1.5769, 1.8919], + device='cuda:2'), covar=tensor([0.0942, 0.1020, 0.1361, 0.0500, 0.1362, 0.1635, 0.1183, 0.0950], + device='cuda:2'), in_proj_covar=tensor([0.0263, 0.0239, 0.0276, 0.0221, 0.0240, 0.0267, 0.0273, 0.0244], + 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') +2023-02-06 03:22:37,615 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:22:38,075 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 03:22:39,696 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9919, 2.3085, 2.9876, 1.0234, 2.9609, 1.5682, 1.3707, 1.6258], + device='cuda:2'), covar=tensor([0.0386, 0.0195, 0.0143, 0.0337, 0.0192, 0.0440, 0.0464, 0.0269], + device='cuda:2'), in_proj_covar=tensor([0.0340, 0.0251, 0.0203, 0.0302, 0.0238, 0.0385, 0.0310, 0.0287], + device='cuda:2'), out_proj_covar=tensor([1.1150e-04, 8.0551e-05, 6.3816e-05, 9.6298e-05, 7.6935e-05, 1.3405e-04, + 1.0118e-04, 9.2305e-05], device='cuda:2') +2023-02-06 03:22:41,674 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8283, 2.0357, 1.6763, 2.6967, 1.0545, 1.4340, 1.7468, 1.9018], + device='cuda:2'), covar=tensor([0.1128, 0.1282, 0.1654, 0.0502, 0.1824, 0.2070, 0.1455, 0.1378], + device='cuda:2'), in_proj_covar=tensor([0.0266, 0.0242, 0.0278, 0.0224, 0.0244, 0.0271, 0.0277, 0.0247], + 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') +2023-02-06 03:22:45,424 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.727e+02 3.570e+02 4.321e+02 1.428e+03, threshold=7.140e+02, percent-clipped=5.0 +2023-02-06 03:22:53,283 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48542.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:22:55,822 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:22:56,685 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 03:22:57,629 INFO [train.py:901] (2/4) Epoch 7, batch 50, loss[loss=0.2557, simple_loss=0.3361, pruned_loss=0.08763, over 8196.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3386, pruned_loss=0.1024, over 367158.58 frames. ], batch size: 23, lr: 1.14e-02, grad_scale: 8.0 +2023-02-06 03:22:57,785 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:23:09,791 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48567.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:23:12,934 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 03:23:14,279 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48574.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:23:31,384 INFO [train.py:901] (2/4) Epoch 7, batch 100, loss[loss=0.2682, simple_loss=0.342, pruned_loss=0.09716, over 8290.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3377, pruned_loss=0.1027, over 645090.83 frames. ], batch size: 23, lr: 1.14e-02, grad_scale: 8.0 +2023-02-06 03:23:35,003 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 03:23:44,209 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2690, 1.5189, 2.2709, 1.1321, 1.7282, 1.5847, 1.4109, 1.3311], + device='cuda:2'), covar=tensor([0.1557, 0.1768, 0.0649, 0.3044, 0.1199, 0.2441, 0.1540, 0.1672], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0473, 0.0526, 0.0547, 0.0594, 0.0534, 0.0451, 0.0592], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:23:54,579 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.936e+02 3.434e+02 4.642e+02 8.961e+02, threshold=6.868e+02, percent-clipped=3.0 +2023-02-06 03:24:06,741 INFO [train.py:901] (2/4) Epoch 7, batch 150, loss[loss=0.233, simple_loss=0.3054, pruned_loss=0.08029, over 8515.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3377, pruned_loss=0.1032, over 859885.97 frames. ], batch size: 29, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:24:31,972 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:24:34,081 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48689.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:24:40,627 INFO [train.py:901] (2/4) Epoch 7, batch 200, loss[loss=0.2557, simple_loss=0.3316, pruned_loss=0.08986, over 8495.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3383, pruned_loss=0.1031, over 1029964.41 frames. ], batch size: 26, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:25:03,491 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.629e+02 3.306e+02 4.274e+02 1.004e+03, threshold=6.612e+02, percent-clipped=3.0 +2023-02-06 03:25:15,496 INFO [train.py:901] (2/4) Epoch 7, batch 250, loss[loss=0.2512, simple_loss=0.3269, pruned_loss=0.08778, over 8520.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.338, pruned_loss=0.1028, over 1158855.92 frames. ], batch size: 28, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:25:22,707 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48759.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:26,752 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 03:25:35,607 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 03:25:41,110 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:43,879 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5680, 1.4913, 3.1506, 1.3696, 2.0994, 3.3648, 3.3406, 2.7971], + device='cuda:2'), covar=tensor([0.1080, 0.1288, 0.0311, 0.1822, 0.0710, 0.0240, 0.0358, 0.0614], + device='cuda:2'), in_proj_covar=tensor([0.0239, 0.0272, 0.0226, 0.0269, 0.0235, 0.0213, 0.0264, 0.0278], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 03:25:50,541 INFO [train.py:901] (2/4) Epoch 7, batch 300, loss[loss=0.299, simple_loss=0.3649, pruned_loss=0.1165, over 8466.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3375, pruned_loss=0.1022, over 1262210.18 frames. ], batch size: 25, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:25:52,229 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:54,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48805.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:55,625 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7173, 3.0189, 2.2351, 3.6835, 1.9303, 1.7974, 2.4127, 3.0644], + device='cuda:2'), covar=tensor([0.0695, 0.0888, 0.1256, 0.0338, 0.1330, 0.1701, 0.1288, 0.0840], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0236, 0.0271, 0.0220, 0.0236, 0.0261, 0.0268, 0.0236], + 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') +2023-02-06 03:26:12,388 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:13,525 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.973e+02 3.476e+02 4.340e+02 1.124e+03, threshold=6.953e+02, percent-clipped=5.0 +2023-02-06 03:26:18,393 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:18,510 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4362, 1.9986, 3.1613, 2.4925, 2.7290, 2.1345, 1.5622, 1.3775], + device='cuda:2'), covar=tensor([0.2348, 0.2744, 0.0682, 0.1516, 0.1274, 0.1400, 0.1318, 0.2861], + device='cuda:2'), in_proj_covar=tensor([0.0800, 0.0742, 0.0640, 0.0731, 0.0829, 0.0683, 0.0638, 0.0672], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:26:25,097 INFO [train.py:901] (2/4) Epoch 7, batch 350, loss[loss=0.2571, simple_loss=0.3228, pruned_loss=0.09565, over 8246.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3371, pruned_loss=0.1021, over 1342098.44 frames. ], batch size: 22, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:26:30,610 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:43,227 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:27:00,371 INFO [train.py:901] (2/4) Epoch 7, batch 400, loss[loss=0.2901, simple_loss=0.3702, pruned_loss=0.105, over 8491.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.339, pruned_loss=0.1033, over 1406099.76 frames. ], batch size: 29, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:27:01,277 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48900.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:27:06,124 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 03:27:22,466 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.734e+02 3.619e+02 4.506e+02 1.679e+03, threshold=7.237e+02, percent-clipped=8.0 +2023-02-06 03:27:32,151 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48945.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:27:34,591 INFO [train.py:901] (2/4) Epoch 7, batch 450, loss[loss=0.2426, simple_loss=0.31, pruned_loss=0.0876, over 8098.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3387, pruned_loss=0.1021, over 1455325.31 frames. ], batch size: 23, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:27:49,649 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48970.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:28:10,025 INFO [train.py:901] (2/4) Epoch 7, batch 500, loss[loss=0.3092, simple_loss=0.3647, pruned_loss=0.1269, over 8108.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.338, pruned_loss=0.1013, over 1490278.32 frames. ], batch size: 23, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:28:24,469 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5305, 2.1860, 4.6791, 1.2110, 2.8495, 2.1151, 1.4784, 2.7694], + device='cuda:2'), covar=tensor([0.1672, 0.2050, 0.0594, 0.3371, 0.1523, 0.2521, 0.1728, 0.2271], + device='cuda:2'), in_proj_covar=tensor([0.0480, 0.0477, 0.0530, 0.0552, 0.0596, 0.0537, 0.0458, 0.0597], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:28:30,187 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 03:28:32,336 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.572e+02 3.184e+02 4.227e+02 8.649e+02, threshold=6.369e+02, percent-clipped=1.0 +2023-02-06 03:28:32,518 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7297, 1.4262, 2.7334, 1.1462, 2.0420, 2.9409, 2.8911, 2.4727], + device='cuda:2'), covar=tensor([0.0928, 0.1356, 0.0467, 0.2137, 0.0757, 0.0319, 0.0558, 0.0698], + device='cuda:2'), in_proj_covar=tensor([0.0240, 0.0277, 0.0229, 0.0272, 0.0238, 0.0214, 0.0267, 0.0281], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 03:28:43,904 INFO [train.py:901] (2/4) Epoch 7, batch 550, loss[loss=0.238, simple_loss=0.3075, pruned_loss=0.08427, over 7798.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3387, pruned_loss=0.102, over 1517613.21 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:28:50,291 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:28:52,346 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4114, 2.4685, 1.6069, 1.9681, 2.0727, 1.2965, 1.6079, 1.9389], + device='cuda:2'), covar=tensor([0.1120, 0.0308, 0.0952, 0.0531, 0.0559, 0.1242, 0.0958, 0.0785], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0232, 0.0319, 0.0301, 0.0316, 0.0313, 0.0347, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 03:29:07,256 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:07,273 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1482, 2.5578, 3.2512, 1.0688, 3.0388, 1.9798, 1.5219, 1.8160], + device='cuda:2'), covar=tensor([0.0390, 0.0209, 0.0108, 0.0372, 0.0230, 0.0445, 0.0439, 0.0264], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0256, 0.0212, 0.0315, 0.0245, 0.0403, 0.0323, 0.0298], + device='cuda:2'), out_proj_covar=tensor([1.1616e-04, 8.1790e-05, 6.6471e-05, 1.0018e-04, 7.8685e-05, 1.3979e-04, + 1.0517e-04, 9.5776e-05], device='cuda:2') +2023-02-06 03:29:07,958 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1023, 1.6603, 3.1538, 0.9573, 1.9408, 1.5598, 1.1478, 1.9016], + device='cuda:2'), covar=tensor([0.2133, 0.2104, 0.0630, 0.3951, 0.1704, 0.3025, 0.2214, 0.2330], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0469, 0.0523, 0.0541, 0.0587, 0.0528, 0.0452, 0.0590], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:29:19,499 INFO [train.py:901] (2/4) Epoch 7, batch 600, loss[loss=0.3052, simple_loss=0.3777, pruned_loss=0.1163, over 8464.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3384, pruned_loss=0.1024, over 1538317.57 frames. ], batch size: 25, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:29:31,450 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 03:29:41,663 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49130.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:42,811 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.845e+02 3.510e+02 4.694e+02 1.227e+03, threshold=7.020e+02, percent-clipped=5.0 +2023-02-06 03:29:54,583 INFO [train.py:901] (2/4) Epoch 7, batch 650, loss[loss=0.2546, simple_loss=0.3291, pruned_loss=0.09008, over 8630.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3384, pruned_loss=0.1024, over 1555048.41 frames. ], batch size: 34, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:29:58,984 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49155.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:59,706 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49156.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:17,665 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49181.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:18,990 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:24,026 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4102, 1.9001, 3.0530, 1.1807, 2.2334, 1.8687, 1.5113, 1.8058], + device='cuda:2'), covar=tensor([0.1756, 0.2089, 0.0717, 0.3915, 0.1367, 0.2586, 0.1814, 0.2272], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0474, 0.0530, 0.0551, 0.0590, 0.0531, 0.0455, 0.0596], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:30:29,826 INFO [train.py:901] (2/4) Epoch 7, batch 700, loss[loss=0.2742, simple_loss=0.3314, pruned_loss=0.1085, over 7942.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3385, pruned_loss=0.1031, over 1565809.77 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:30:31,258 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49200.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:54,550 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 2.876e+02 3.436e+02 4.276e+02 6.994e+02, threshold=6.873e+02, percent-clipped=0.0 +2023-02-06 03:31:03,863 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 03:31:06,145 INFO [train.py:901] (2/4) Epoch 7, batch 750, loss[loss=0.3634, simple_loss=0.4091, pruned_loss=0.1588, over 8469.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3389, pruned_loss=0.1038, over 1578942.37 frames. ], batch size: 27, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:31:09,786 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:31:18,084 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 03:31:26,425 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 03:31:40,962 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49298.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:31:41,432 INFO [train.py:901] (2/4) Epoch 7, batch 800, loss[loss=0.2915, simple_loss=0.3491, pruned_loss=0.117, over 7925.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3379, pruned_loss=0.1031, over 1590788.32 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:31:53,365 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49315.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:32:05,543 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.832e+02 3.318e+02 4.162e+02 1.224e+03, threshold=6.636e+02, percent-clipped=6.0 +2023-02-06 03:32:08,087 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.59 vs. limit=5.0 +2023-02-06 03:32:17,931 INFO [train.py:901] (2/4) Epoch 7, batch 850, loss[loss=0.2584, simple_loss=0.3256, pruned_loss=0.09562, over 7939.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3378, pruned_loss=0.1023, over 1602510.73 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:32:48,868 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 03:32:52,542 INFO [train.py:901] (2/4) Epoch 7, batch 900, loss[loss=0.2233, simple_loss=0.2924, pruned_loss=0.07712, over 7934.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3367, pruned_loss=0.1018, over 1605093.18 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:32:58,995 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7131, 5.9017, 4.9066, 2.3477, 5.0886, 5.4384, 5.2793, 4.8704], + device='cuda:2'), covar=tensor([0.0588, 0.0439, 0.0949, 0.4620, 0.0595, 0.0728, 0.1275, 0.0664], + device='cuda:2'), in_proj_covar=tensor([0.0414, 0.0320, 0.0345, 0.0430, 0.0330, 0.0311, 0.0319, 0.0271], + 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') +2023-02-06 03:33:17,124 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.812e+02 3.278e+02 4.578e+02 1.649e+03, threshold=6.556e+02, percent-clipped=8.0 +2023-02-06 03:33:20,691 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0802, 1.4261, 1.5756, 1.2665, 1.0672, 1.3447, 1.6060, 1.4891], + device='cuda:2'), covar=tensor([0.0538, 0.1297, 0.1676, 0.1427, 0.0620, 0.1526, 0.0754, 0.0589], + device='cuda:2'), in_proj_covar=tensor([0.0118, 0.0166, 0.0207, 0.0169, 0.0118, 0.0174, 0.0129, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 03:33:21,994 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:33:28,071 INFO [train.py:901] (2/4) Epoch 7, batch 950, loss[loss=0.3037, simple_loss=0.3763, pruned_loss=0.1155, over 8447.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3353, pruned_loss=0.1003, over 1606128.93 frames. ], batch size: 27, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:33:28,881 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49450.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:33:50,507 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 03:33:53,371 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1015, 4.0885, 3.6247, 1.8508, 3.5666, 3.6211, 3.8277, 3.2980], + device='cuda:2'), covar=tensor([0.0803, 0.0569, 0.0877, 0.4369, 0.0780, 0.0892, 0.0994, 0.0887], + device='cuda:2'), in_proj_covar=tensor([0.0410, 0.0315, 0.0342, 0.0424, 0.0325, 0.0309, 0.0314, 0.0270], + 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') +2023-02-06 03:34:03,657 INFO [train.py:901] (2/4) Epoch 7, batch 1000, loss[loss=0.2842, simple_loss=0.3527, pruned_loss=0.1078, over 8334.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3358, pruned_loss=0.1004, over 1611869.46 frames. ], batch size: 25, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:34:24,235 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 03:34:27,699 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 3.091e+02 3.599e+02 4.515e+02 1.445e+03, threshold=7.198e+02, percent-clipped=7.0 +2023-02-06 03:34:35,908 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 03:34:38,738 INFO [train.py:901] (2/4) Epoch 7, batch 1050, loss[loss=0.2655, simple_loss=0.3472, pruned_loss=0.09197, over 8102.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3352, pruned_loss=0.1007, over 1609452.78 frames. ], batch size: 23, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:34:43,050 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49554.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:34:54,543 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49571.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:00,748 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49579.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:13,256 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49596.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:14,560 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:15,147 INFO [train.py:901] (2/4) Epoch 7, batch 1100, loss[loss=0.2252, simple_loss=0.3007, pruned_loss=0.07489, over 7934.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3354, pruned_loss=0.1008, over 1610758.75 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:35:38,351 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.766e+02 3.386e+02 4.310e+02 6.415e+02, threshold=6.771e+02, percent-clipped=0.0 +2023-02-06 03:35:46,088 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 03:35:49,462 INFO [train.py:901] (2/4) Epoch 7, batch 1150, loss[loss=0.2899, simple_loss=0.3469, pruned_loss=0.1165, over 8026.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3343, pruned_loss=0.1001, over 1609700.90 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:35:55,626 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1423, 1.0732, 1.1426, 1.1055, 0.7976, 1.2015, 0.0299, 0.8573], + device='cuda:2'), covar=tensor([0.2562, 0.1939, 0.0869, 0.1543, 0.4995, 0.0706, 0.3812, 0.1921], + device='cuda:2'), in_proj_covar=tensor([0.0140, 0.0130, 0.0083, 0.0182, 0.0224, 0.0084, 0.0141, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:36:13,940 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 03:36:23,369 INFO [train.py:901] (2/4) Epoch 7, batch 1200, loss[loss=0.2751, simple_loss=0.3409, pruned_loss=0.1047, over 8511.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3348, pruned_loss=0.1005, over 1611912.44 frames. ], batch size: 49, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:36:33,699 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:36:47,174 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.915e+02 3.820e+02 5.048e+02 1.193e+03, threshold=7.640e+02, percent-clipped=11.0 +2023-02-06 03:36:57,970 INFO [train.py:901] (2/4) Epoch 7, batch 1250, loss[loss=0.2665, simple_loss=0.337, pruned_loss=0.09802, over 8392.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3329, pruned_loss=0.09995, over 1608436.72 frames. ], batch size: 49, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:37:09,616 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4144, 1.9059, 3.1012, 1.1532, 2.2477, 1.8535, 1.5654, 1.8612], + device='cuda:2'), covar=tensor([0.1653, 0.1898, 0.0717, 0.3419, 0.1460, 0.2556, 0.1581, 0.2190], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0480, 0.0538, 0.0552, 0.0601, 0.0534, 0.0455, 0.0599], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:37:10,636 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 03:37:12,336 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3411, 2.2349, 3.5402, 1.1330, 2.5742, 1.8158, 1.7457, 1.9687], + device='cuda:2'), covar=tensor([0.1789, 0.2015, 0.0643, 0.3560, 0.1535, 0.2730, 0.1597, 0.2560], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0480, 0.0538, 0.0553, 0.0602, 0.0535, 0.0456, 0.0599], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:37:22,187 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:37:22,255 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:37:29,679 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49794.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:37:32,800 INFO [train.py:901] (2/4) Epoch 7, batch 1300, loss[loss=0.2974, simple_loss=0.3577, pruned_loss=0.1186, over 7936.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3329, pruned_loss=0.1, over 1604460.53 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:37:57,653 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.650e+02 3.390e+02 4.402e+02 9.600e+02, threshold=6.781e+02, percent-clipped=3.0 +2023-02-06 03:38:08,115 INFO [train.py:901] (2/4) Epoch 7, batch 1350, loss[loss=0.2532, simple_loss=0.331, pruned_loss=0.08769, over 8355.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3327, pruned_loss=0.0999, over 1608806.34 frames. ], batch size: 24, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:38:28,832 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1064, 1.3066, 4.2826, 1.5697, 3.7475, 3.5755, 3.8120, 3.6848], + device='cuda:2'), covar=tensor([0.0397, 0.3567, 0.0383, 0.2606, 0.0908, 0.0616, 0.0433, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0366, 0.0509, 0.0461, 0.0447, 0.0508, 0.0423, 0.0426, 0.0481], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-06 03:38:42,135 INFO [train.py:901] (2/4) Epoch 7, batch 1400, loss[loss=0.2708, simple_loss=0.351, pruned_loss=0.09535, over 8466.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.334, pruned_loss=0.1006, over 1609924.84 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:38:43,013 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:38:49,798 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49909.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:39:07,149 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.921e+02 3.790e+02 4.996e+02 8.997e+02, threshold=7.579e+02, percent-clipped=6.0 +2023-02-06 03:39:11,338 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 03:39:18,087 INFO [train.py:901] (2/4) Epoch 7, batch 1450, loss[loss=0.237, simple_loss=0.3053, pruned_loss=0.08439, over 7816.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3339, pruned_loss=0.1005, over 1613340.51 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:39:26,267 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2063, 1.3965, 2.3376, 1.1520, 2.0366, 2.5281, 2.5000, 2.1456], + device='cuda:2'), covar=tensor([0.0968, 0.1065, 0.0440, 0.1915, 0.0593, 0.0366, 0.0535, 0.0752], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0275, 0.0231, 0.0272, 0.0236, 0.0215, 0.0269, 0.0281], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 03:39:31,684 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:39:41,670 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 +2023-02-06 03:39:49,099 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49994.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:39:52,397 INFO [train.py:901] (2/4) Epoch 7, batch 1500, loss[loss=0.2615, simple_loss=0.3415, pruned_loss=0.09074, over 8331.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3338, pruned_loss=0.1005, over 1612168.10 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:40:07,867 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 03:40:16,574 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.509e+02 2.823e+02 3.555e+02 4.038e+02 9.229e+02, threshold=7.110e+02, percent-clipped=3.0 +2023-02-06 03:40:27,887 INFO [train.py:901] (2/4) Epoch 7, batch 1550, loss[loss=0.3726, simple_loss=0.4146, pruned_loss=0.1653, over 8524.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3343, pruned_loss=0.1009, over 1607817.41 frames. ], batch size: 26, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:40:57,942 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0873, 3.0445, 3.3287, 2.4728, 1.7996, 3.2207, 0.5397, 2.1194], + device='cuda:2'), covar=tensor([0.2600, 0.1930, 0.0723, 0.2570, 0.6234, 0.0508, 0.5331, 0.2319], + device='cuda:2'), in_proj_covar=tensor([0.0141, 0.0134, 0.0082, 0.0186, 0.0225, 0.0084, 0.0144, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:41:02,311 INFO [train.py:901] (2/4) Epoch 7, batch 1600, loss[loss=0.321, simple_loss=0.3768, pruned_loss=0.1326, over 8031.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3349, pruned_loss=0.1012, over 1607621.45 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:41:22,665 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:41:25,916 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.849e+02 3.464e+02 4.418e+02 7.019e+02, threshold=6.928e+02, percent-clipped=0.0 +2023-02-06 03:41:35,387 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:41:35,570 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 03:41:37,156 INFO [train.py:901] (2/4) Epoch 7, batch 1650, loss[loss=0.2625, simple_loss=0.3358, pruned_loss=0.09455, over 8607.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3336, pruned_loss=0.09971, over 1608186.95 frames. ], batch size: 39, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:41:40,108 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1054, 1.3343, 1.2422, 0.7107, 1.3220, 1.0140, 0.4186, 1.2286], + device='cuda:2'), covar=tensor([0.0167, 0.0102, 0.0085, 0.0155, 0.0100, 0.0275, 0.0265, 0.0089], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0258, 0.0215, 0.0311, 0.0249, 0.0403, 0.0315, 0.0292], + device='cuda:2'), out_proj_covar=tensor([1.1342e-04, 8.1425e-05, 6.7136e-05, 9.8328e-05, 7.9676e-05, 1.3926e-04, + 1.0242e-04, 9.3340e-05], device='cuda:2') +2023-02-06 03:41:41,429 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50155.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:41:48,658 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50165.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:41:59,247 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:42:05,844 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50190.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:42:11,948 INFO [train.py:901] (2/4) Epoch 7, batch 1700, loss[loss=0.2807, simple_loss=0.3428, pruned_loss=0.1093, over 8106.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3341, pruned_loss=0.1001, over 1611089.86 frames. ], batch size: 23, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:42:22,984 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50215.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:42:35,242 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.886e+02 3.481e+02 4.608e+02 1.233e+03, threshold=6.962e+02, percent-clipped=3.0 +2023-02-06 03:42:41,502 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4592, 2.0479, 3.3814, 1.1173, 2.5005, 1.6914, 1.6307, 2.1408], + device='cuda:2'), covar=tensor([0.1544, 0.1861, 0.0633, 0.3315, 0.1279, 0.2652, 0.1437, 0.2172], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0472, 0.0530, 0.0550, 0.0590, 0.0534, 0.0451, 0.0592], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:42:42,134 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:42:46,680 INFO [train.py:901] (2/4) Epoch 7, batch 1750, loss[loss=0.2956, simple_loss=0.3577, pruned_loss=0.1167, over 8282.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3353, pruned_loss=0.1006, over 1615540.72 frames. ], batch size: 23, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:43:14,405 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1905, 1.2827, 2.3361, 1.1834, 2.2042, 2.5401, 2.5420, 2.1026], + device='cuda:2'), covar=tensor([0.0910, 0.1059, 0.0417, 0.1803, 0.0478, 0.0331, 0.0474, 0.0778], + device='cuda:2'), in_proj_covar=tensor([0.0240, 0.0278, 0.0232, 0.0275, 0.0239, 0.0215, 0.0273, 0.0283], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 03:43:14,442 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3100, 1.5048, 1.4112, 1.9517, 0.6812, 1.1393, 1.4449, 1.5164], + device='cuda:2'), covar=tensor([0.1133, 0.0940, 0.1456, 0.0603, 0.1434, 0.1927, 0.0955, 0.0845], + device='cuda:2'), in_proj_covar=tensor([0.0257, 0.0239, 0.0281, 0.0227, 0.0240, 0.0270, 0.0278, 0.0242], + 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') +2023-02-06 03:43:21,550 INFO [train.py:901] (2/4) Epoch 7, batch 1800, loss[loss=0.2629, simple_loss=0.3428, pruned_loss=0.09151, over 8325.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3358, pruned_loss=0.1008, over 1618443.42 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:43:44,800 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.922e+02 3.562e+02 4.379e+02 1.030e+03, threshold=7.125e+02, percent-clipped=4.0 +2023-02-06 03:43:56,170 INFO [train.py:901] (2/4) Epoch 7, batch 1850, loss[loss=0.2761, simple_loss=0.3414, pruned_loss=0.1054, over 8188.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3352, pruned_loss=0.1006, over 1615873.29 frames. ], batch size: 23, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:44:30,593 INFO [train.py:901] (2/4) Epoch 7, batch 1900, loss[loss=0.3132, simple_loss=0.3664, pruned_loss=0.13, over 8585.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3338, pruned_loss=0.09963, over 1611039.27 frames. ], batch size: 49, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:44:43,991 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 03:44:53,923 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.786e+02 3.637e+02 4.614e+02 8.948e+02, threshold=7.273e+02, percent-clipped=3.0 +2023-02-06 03:44:56,013 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 03:45:04,598 INFO [train.py:901] (2/4) Epoch 7, batch 1950, loss[loss=0.3575, simple_loss=0.3893, pruned_loss=0.1629, over 8062.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3337, pruned_loss=0.09975, over 1613139.17 frames. ], batch size: 73, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:45:12,441 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.33 vs. limit=5.0 +2023-02-06 03:45:15,311 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 03:45:33,288 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:45:39,204 INFO [train.py:901] (2/4) Epoch 7, batch 2000, loss[loss=0.2563, simple_loss=0.3321, pruned_loss=0.09028, over 7977.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3336, pruned_loss=0.09994, over 1610481.38 frames. ], batch size: 21, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:45:39,435 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50499.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:45:56,773 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:46:03,226 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.800e+02 3.583e+02 4.591e+02 1.075e+03, threshold=7.166e+02, percent-clipped=7.0 +2023-02-06 03:46:13,949 INFO [train.py:901] (2/4) Epoch 7, batch 2050, loss[loss=0.296, simple_loss=0.3602, pruned_loss=0.1159, over 8562.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3349, pruned_loss=0.1007, over 1615724.45 frames. ], batch size: 39, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:46:20,539 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50559.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:46:23,779 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:46:28,399 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3952, 2.0725, 3.6244, 1.1685, 2.5339, 1.9033, 1.6873, 2.2076], + device='cuda:2'), covar=tensor([0.1590, 0.1754, 0.0530, 0.3410, 0.1264, 0.2437, 0.1442, 0.2122], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0465, 0.0526, 0.0544, 0.0589, 0.0527, 0.0447, 0.0586], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:46:37,581 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5280, 1.2888, 4.6937, 1.7323, 4.0261, 3.9432, 4.2096, 4.0771], + device='cuda:2'), covar=tensor([0.0369, 0.3937, 0.0303, 0.2702, 0.0967, 0.0624, 0.0423, 0.0495], + device='cuda:2'), in_proj_covar=tensor([0.0370, 0.0509, 0.0464, 0.0456, 0.0515, 0.0429, 0.0432, 0.0493], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 03:46:47,297 INFO [train.py:901] (2/4) Epoch 7, batch 2100, loss[loss=0.3125, simple_loss=0.3778, pruned_loss=0.1236, over 8600.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3359, pruned_loss=0.1009, over 1619523.37 frames. ], batch size: 34, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:46:52,370 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:47:11,024 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 3.066e+02 3.697e+02 4.610e+02 1.063e+03, threshold=7.394e+02, percent-clipped=3.0 +2023-02-06 03:47:22,367 INFO [train.py:901] (2/4) Epoch 7, batch 2150, loss[loss=0.263, simple_loss=0.3323, pruned_loss=0.09685, over 8526.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3335, pruned_loss=0.1001, over 1610387.25 frames. ], batch size: 26, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:47:40,258 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50674.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:47:47,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2819, 1.4251, 1.6646, 1.3445, 1.2254, 1.3706, 1.8203, 1.8962], + device='cuda:2'), covar=tensor([0.0523, 0.1278, 0.1850, 0.1466, 0.0630, 0.1564, 0.0724, 0.0556], + device='cuda:2'), in_proj_covar=tensor([0.0117, 0.0167, 0.0208, 0.0171, 0.0118, 0.0176, 0.0130, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 03:47:56,608 INFO [train.py:901] (2/4) Epoch 7, batch 2200, loss[loss=0.2522, simple_loss=0.331, pruned_loss=0.08667, over 8105.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3328, pruned_loss=0.09941, over 1606621.17 frames. ], batch size: 23, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:48:05,939 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 03:48:20,824 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.939e+02 3.492e+02 4.230e+02 8.261e+02, threshold=6.983e+02, percent-clipped=2.0 +2023-02-06 03:48:30,767 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1382, 1.1054, 1.1385, 1.1270, 0.8776, 1.2491, 0.1132, 0.9026], + device='cuda:2'), covar=tensor([0.3378, 0.2010, 0.0853, 0.1534, 0.4922, 0.0782, 0.3930, 0.1878], + device='cuda:2'), in_proj_covar=tensor([0.0141, 0.0135, 0.0082, 0.0185, 0.0227, 0.0085, 0.0144, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:48:31,237 INFO [train.py:901] (2/4) Epoch 7, batch 2250, loss[loss=0.2849, simple_loss=0.3244, pruned_loss=0.1227, over 7684.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3338, pruned_loss=0.0997, over 1607993.60 frames. ], batch size: 18, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:49:05,383 INFO [train.py:901] (2/4) Epoch 7, batch 2300, loss[loss=0.275, simple_loss=0.3484, pruned_loss=0.1008, over 8029.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3354, pruned_loss=0.1008, over 1607944.96 frames. ], batch size: 22, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:49:06,291 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50800.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:49:23,399 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50826.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:49:28,605 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 3.140e+02 4.073e+02 5.620e+02 1.608e+03, threshold=8.146e+02, percent-clipped=16.0 +2023-02-06 03:49:29,696 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 03:49:38,685 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4631, 1.8102, 3.1382, 1.1433, 2.2593, 1.8663, 1.4449, 1.9314], + device='cuda:2'), covar=tensor([0.1513, 0.1922, 0.0496, 0.3406, 0.1268, 0.2395, 0.1612, 0.1965], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0473, 0.0535, 0.0557, 0.0596, 0.0537, 0.0457, 0.0596], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 03:49:39,826 INFO [train.py:901] (2/4) Epoch 7, batch 2350, loss[loss=0.235, simple_loss=0.3125, pruned_loss=0.07868, over 8073.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3359, pruned_loss=0.101, over 1609575.57 frames. ], batch size: 21, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:49:44,715 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5691, 2.6793, 1.7792, 2.1730, 2.1536, 1.3710, 1.9584, 2.0397], + device='cuda:2'), covar=tensor([0.1262, 0.0244, 0.1008, 0.0519, 0.0607, 0.1300, 0.0912, 0.0860], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0229, 0.0313, 0.0299, 0.0307, 0.0316, 0.0339, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 03:49:47,971 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:49:55,891 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 03:50:05,009 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:50:05,743 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4075, 1.9372, 3.0511, 2.4139, 2.5016, 2.1001, 1.5566, 1.1316], + device='cuda:2'), covar=tensor([0.2501, 0.2722, 0.0634, 0.1544, 0.1383, 0.1430, 0.1392, 0.3076], + device='cuda:2'), in_proj_covar=tensor([0.0835, 0.0770, 0.0668, 0.0767, 0.0864, 0.0709, 0.0664, 0.0707], + 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') +2023-02-06 03:50:14,220 INFO [train.py:901] (2/4) Epoch 7, batch 2400, loss[loss=0.2925, simple_loss=0.3606, pruned_loss=0.1122, over 7802.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3359, pruned_loss=0.1009, over 1610774.35 frames. ], batch size: 19, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:50:20,198 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50908.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:50:35,186 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50930.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:50:36,977 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.900e+02 3.414e+02 4.196e+02 7.276e+02, threshold=6.828e+02, percent-clipped=0.0 +2023-02-06 03:50:47,532 INFO [train.py:901] (2/4) Epoch 7, batch 2450, loss[loss=0.2805, simple_loss=0.3488, pruned_loss=0.1061, over 8500.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3359, pruned_loss=0.1012, over 1613194.25 frames. ], batch size: 28, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:50:48,446 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0726, 1.2978, 1.1757, 0.4326, 1.2010, 0.9871, 0.1521, 1.1657], + device='cuda:2'), covar=tensor([0.0187, 0.0157, 0.0148, 0.0245, 0.0169, 0.0425, 0.0337, 0.0133], + device='cuda:2'), in_proj_covar=tensor([0.0343, 0.0251, 0.0213, 0.0305, 0.0249, 0.0394, 0.0310, 0.0286], + device='cuda:2'), out_proj_covar=tensor([1.1103e-04, 7.8587e-05, 6.6356e-05, 9.6048e-05, 7.9174e-05, 1.3551e-04, + 1.0040e-04, 9.0933e-05], device='cuda:2') +2023-02-06 03:50:52,348 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50955.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:50:59,072 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50964.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:51:22,791 INFO [train.py:901] (2/4) Epoch 7, batch 2500, loss[loss=0.2903, simple_loss=0.3542, pruned_loss=0.1132, over 8506.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.335, pruned_loss=0.1003, over 1615345.44 frames. ], batch size: 28, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:51:39,933 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51023.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:51:46,386 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.634e+02 3.421e+02 4.023e+02 8.503e+02, threshold=6.842e+02, percent-clipped=1.0 +2023-02-06 03:51:56,889 INFO [train.py:901] (2/4) Epoch 7, batch 2550, loss[loss=0.267, simple_loss=0.3433, pruned_loss=0.09532, over 8462.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3357, pruned_loss=0.1009, over 1614830.89 frames. ], batch size: 25, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:51:58,478 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.98 vs. limit=5.0 +2023-02-06 03:52:04,252 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3504, 1.5351, 1.2860, 1.8881, 0.6935, 1.0798, 1.2767, 1.4673], + device='cuda:2'), covar=tensor([0.1143, 0.0970, 0.1506, 0.0694, 0.1580, 0.2100, 0.1039, 0.1024], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0236, 0.0275, 0.0221, 0.0238, 0.0266, 0.0274, 0.0241], + 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') +2023-02-06 03:52:18,135 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:52:31,080 INFO [train.py:901] (2/4) Epoch 7, batch 2600, loss[loss=0.3208, simple_loss=0.3685, pruned_loss=0.1366, over 8611.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3365, pruned_loss=0.1013, over 1618537.43 frames. ], batch size: 34, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:52:37,432 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 03:52:45,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6478, 1.7563, 2.0462, 1.7878, 1.0425, 2.1407, 0.2893, 1.2133], + device='cuda:2'), covar=tensor([0.2829, 0.2167, 0.0737, 0.2108, 0.6345, 0.0724, 0.4665, 0.2604], + device='cuda:2'), in_proj_covar=tensor([0.0142, 0.0138, 0.0083, 0.0186, 0.0234, 0.0085, 0.0145, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 03:52:54,868 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.187e+02 3.140e+02 3.874e+02 4.757e+02 8.436e+02, threshold=7.747e+02, percent-clipped=5.0 +2023-02-06 03:53:02,030 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51144.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:53:05,172 INFO [train.py:901] (2/4) Epoch 7, batch 2650, loss[loss=0.359, simple_loss=0.3951, pruned_loss=0.1615, over 8346.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3365, pruned_loss=0.1018, over 1617595.27 frames. ], batch size: 26, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:53:19,257 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51170.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:53:22,246 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 03:53:23,605 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 03:53:32,645 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51190.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:53:39,229 INFO [train.py:901] (2/4) Epoch 7, batch 2700, loss[loss=0.2476, simple_loss=0.3174, pruned_loss=0.08886, over 7978.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3355, pruned_loss=0.101, over 1613901.32 frames. ], batch size: 21, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:54:02,457 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 3.046e+02 3.584e+02 4.560e+02 9.753e+02, threshold=7.169e+02, percent-clipped=4.0 +2023-02-06 03:54:12,325 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8433, 1.6582, 5.9156, 2.1272, 5.2945, 5.0580, 5.5259, 5.4265], + device='cuda:2'), covar=tensor([0.0322, 0.3706, 0.0215, 0.2541, 0.0860, 0.0620, 0.0321, 0.0344], + device='cuda:2'), in_proj_covar=tensor([0.0379, 0.0520, 0.0469, 0.0458, 0.0525, 0.0437, 0.0432, 0.0490], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 03:54:14,234 INFO [train.py:901] (2/4) Epoch 7, batch 2750, loss[loss=0.2417, simple_loss=0.3282, pruned_loss=0.07756, over 8475.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3347, pruned_loss=0.09978, over 1612327.99 frames. ], batch size: 25, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:54:19,286 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 03:54:21,062 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:54:34,382 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51279.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:54:38,285 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51285.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:54:47,539 INFO [train.py:901] (2/4) Epoch 7, batch 2800, loss[loss=0.2718, simple_loss=0.3414, pruned_loss=0.1011, over 8492.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3342, pruned_loss=0.09928, over 1616582.88 frames. ], batch size: 26, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:54:51,134 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51304.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:54:53,748 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51308.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:55:11,783 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.811e+02 3.563e+02 4.674e+02 6.809e+02, threshold=7.126e+02, percent-clipped=0.0 +2023-02-06 03:55:22,692 INFO [train.py:901] (2/4) Epoch 7, batch 2850, loss[loss=0.2361, simple_loss=0.2946, pruned_loss=0.08879, over 7813.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3326, pruned_loss=0.09811, over 1614945.78 frames. ], batch size: 20, lr: 1.10e-02, grad_scale: 16.0 +2023-02-06 03:55:57,256 INFO [train.py:901] (2/4) Epoch 7, batch 2900, loss[loss=0.2483, simple_loss=0.3359, pruned_loss=0.08034, over 8294.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3348, pruned_loss=0.09992, over 1615119.19 frames. ], batch size: 23, lr: 1.10e-02, grad_scale: 16.0 +2023-02-06 03:55:58,211 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51400.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:04,947 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:13,542 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51423.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:14,145 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:19,650 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 03:56:20,272 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.833e+02 3.577e+02 4.732e+02 1.075e+03, threshold=7.153e+02, percent-clipped=9.0 +2023-02-06 03:56:32,325 INFO [train.py:901] (2/4) Epoch 7, batch 2950, loss[loss=0.2645, simple_loss=0.3317, pruned_loss=0.09863, over 8322.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3354, pruned_loss=0.0998, over 1615834.08 frames. ], batch size: 25, lr: 1.10e-02, grad_scale: 16.0 +2023-02-06 03:57:06,416 INFO [train.py:901] (2/4) Epoch 7, batch 3000, loss[loss=0.3211, simple_loss=0.3831, pruned_loss=0.1296, over 8519.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3364, pruned_loss=0.1011, over 1619514.01 frames. ], batch size: 50, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:57:06,416 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 03:57:21,709 INFO [train.py:935] (2/4) Epoch 7, validation: loss=0.2071, simple_loss=0.305, pruned_loss=0.05459, over 944034.00 frames. +2023-02-06 03:57:21,710 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 03:57:31,201 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:32,575 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51515.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:45,141 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.842e+02 3.422e+02 4.197e+02 1.269e+03, threshold=6.844e+02, percent-clipped=2.0 +2023-02-06 03:57:45,245 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51534.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:57:48,654 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:49,362 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51540.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:50,077 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51541.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:57:53,750 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 03:57:55,342 INFO [train.py:901] (2/4) Epoch 7, batch 3050, loss[loss=0.2524, simple_loss=0.3047, pruned_loss=0.1001, over 7719.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3358, pruned_loss=0.1008, over 1610909.79 frames. ], batch size: 18, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:58:06,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51566.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:58:29,894 INFO [train.py:901] (2/4) Epoch 7, batch 3100, loss[loss=0.2319, simple_loss=0.3125, pruned_loss=0.07558, over 8243.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3336, pruned_loss=0.09951, over 1610848.63 frames. ], batch size: 24, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:58:54,843 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 3.035e+02 3.902e+02 5.145e+02 1.067e+03, threshold=7.804e+02, percent-clipped=7.0 +2023-02-06 03:59:05,327 INFO [train.py:901] (2/4) Epoch 7, batch 3150, loss[loss=0.2204, simple_loss=0.2829, pruned_loss=0.079, over 7706.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3338, pruned_loss=0.1002, over 1613425.98 frames. ], batch size: 18, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:59:05,503 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51649.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:59:26,403 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51679.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:59:33,828 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 03:59:40,120 INFO [train.py:901] (2/4) Epoch 7, batch 3200, loss[loss=0.2597, simple_loss=0.3153, pruned_loss=0.102, over 7225.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3323, pruned_loss=0.09888, over 1613434.14 frames. ], batch size: 16, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:59:43,599 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:05,272 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.946e+02 3.588e+02 4.680e+02 7.788e+02, threshold=7.176e+02, percent-clipped=0.0 +2023-02-06 04:00:12,050 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51744.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:16,043 INFO [train.py:901] (2/4) Epoch 7, batch 3250, loss[loss=0.2855, simple_loss=0.3566, pruned_loss=0.1072, over 8507.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3335, pruned_loss=0.09976, over 1617896.36 frames. ], batch size: 29, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:00:19,467 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:26,781 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51765.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:46,822 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:49,337 INFO [train.py:901] (2/4) Epoch 7, batch 3300, loss[loss=0.2596, simple_loss=0.3359, pruned_loss=0.09166, over 8473.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3347, pruned_loss=0.1005, over 1618162.13 frames. ], batch size: 25, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:01:03,664 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51820.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:14,384 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.988e+02 3.662e+02 4.246e+02 9.313e+02, threshold=7.324e+02, percent-clipped=2.0 +2023-02-06 04:01:24,625 INFO [train.py:901] (2/4) Epoch 7, batch 3350, loss[loss=0.2558, simple_loss=0.3157, pruned_loss=0.09794, over 7925.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.334, pruned_loss=0.09973, over 1621952.33 frames. ], batch size: 20, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:01:30,918 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:32,346 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51859.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:39,540 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51869.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:59,346 INFO [train.py:901] (2/4) Epoch 7, batch 3400, loss[loss=0.2246, simple_loss=0.3077, pruned_loss=0.07076, over 8199.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3329, pruned_loss=0.09955, over 1617810.83 frames. ], batch size: 23, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:02:03,623 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51905.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:02:10,931 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 04:02:16,116 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9588, 2.3734, 1.7901, 2.7239, 1.4432, 1.4501, 1.9741, 2.4152], + device='cuda:2'), covar=tensor([0.0850, 0.0823, 0.1200, 0.0452, 0.1328, 0.1731, 0.1170, 0.0762], + device='cuda:2'), in_proj_covar=tensor([0.0252, 0.0233, 0.0275, 0.0219, 0.0239, 0.0263, 0.0272, 0.0237], + 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') +2023-02-06 04:02:20,930 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51930.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:02:23,260 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.723e+02 3.470e+02 4.144e+02 7.359e+02, threshold=6.940e+02, percent-clipped=1.0 +2023-02-06 04:02:24,206 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5417, 2.0931, 3.1251, 2.3747, 2.5028, 2.1918, 1.6748, 1.3690], + device='cuda:2'), covar=tensor([0.2579, 0.2844, 0.0679, 0.1512, 0.1571, 0.1612, 0.1540, 0.2869], + device='cuda:2'), in_proj_covar=tensor([0.0822, 0.0758, 0.0656, 0.0752, 0.0856, 0.0704, 0.0653, 0.0691], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:02:34,620 INFO [train.py:901] (2/4) Epoch 7, batch 3450, loss[loss=0.2487, simple_loss=0.3071, pruned_loss=0.09511, over 7924.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.331, pruned_loss=0.09814, over 1613872.44 frames. ], batch size: 20, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:02:50,958 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51972.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:03:04,793 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1373, 1.2768, 2.2359, 1.1108, 2.0722, 2.4135, 2.4580, 2.0453], + device='cuda:2'), covar=tensor([0.1178, 0.1360, 0.0589, 0.2207, 0.0678, 0.0427, 0.0647, 0.0930], + device='cuda:2'), in_proj_covar=tensor([0.0240, 0.0274, 0.0230, 0.0271, 0.0243, 0.0218, 0.0275, 0.0281], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 04:03:09,378 INFO [train.py:901] (2/4) Epoch 7, batch 3500, loss[loss=0.2268, simple_loss=0.298, pruned_loss=0.07785, over 7705.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.331, pruned_loss=0.0983, over 1610024.86 frames. ], batch size: 18, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:03:22,433 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 04:03:33,488 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.822e+02 3.302e+02 4.435e+02 1.594e+03, threshold=6.604e+02, percent-clipped=5.0 +2023-02-06 04:03:43,724 INFO [train.py:901] (2/4) Epoch 7, batch 3550, loss[loss=0.2506, simple_loss=0.3162, pruned_loss=0.09255, over 8080.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3314, pruned_loss=0.09823, over 1606880.93 frames. ], batch size: 21, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:03:50,004 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:19,842 INFO [train.py:901] (2/4) Epoch 7, batch 3600, loss[loss=0.2928, simple_loss=0.3593, pruned_loss=0.1131, over 8498.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.332, pruned_loss=0.09853, over 1611480.95 frames. ], batch size: 28, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:04:26,700 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:30,891 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52115.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:37,712 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:43,467 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.818e+02 3.176e+02 4.094e+02 8.086e+02, threshold=6.353e+02, percent-clipped=5.0 +2023-02-06 04:04:47,734 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:53,394 INFO [train.py:901] (2/4) Epoch 7, batch 3650, loss[loss=0.2482, simple_loss=0.3331, pruned_loss=0.08164, over 8517.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3316, pruned_loss=0.09807, over 1609982.35 frames. ], batch size: 49, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:04:54,178 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:23,184 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 04:05:28,575 INFO [train.py:901] (2/4) Epoch 7, batch 3700, loss[loss=0.2626, simple_loss=0.3293, pruned_loss=0.09794, over 8366.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3312, pruned_loss=0.09758, over 1610218.75 frames. ], batch size: 26, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:05:36,879 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52211.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:41,685 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7605, 1.9912, 2.9033, 1.5090, 2.4684, 1.9727, 1.8218, 2.2716], + device='cuda:2'), covar=tensor([0.1160, 0.1553, 0.0487, 0.2595, 0.0897, 0.1906, 0.1288, 0.1557], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0469, 0.0530, 0.0550, 0.0586, 0.0523, 0.0455, 0.0597], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 04:05:47,067 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52224.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:49,849 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52228.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:53,719 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.602e+02 3.554e+02 4.404e+02 9.700e+02, threshold=7.108e+02, percent-clipped=5.0 +2023-02-06 04:06:04,132 INFO [train.py:901] (2/4) Epoch 7, batch 3750, loss[loss=0.2496, simple_loss=0.3136, pruned_loss=0.09284, over 7665.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3314, pruned_loss=0.0977, over 1609966.52 frames. ], batch size: 19, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:06:05,809 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3599, 1.5460, 2.2475, 1.1247, 1.5603, 1.5850, 1.4815, 1.5362], + device='cuda:2'), covar=tensor([0.1473, 0.1919, 0.0699, 0.3327, 0.1385, 0.2561, 0.1574, 0.1735], + device='cuda:2'), in_proj_covar=tensor([0.0473, 0.0471, 0.0532, 0.0550, 0.0585, 0.0525, 0.0456, 0.0596], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 04:06:07,176 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:06:12,881 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6633, 3.1500, 2.3484, 4.0644, 1.8325, 1.9669, 2.2280, 3.0014], + device='cuda:2'), covar=tensor([0.0827, 0.0962, 0.1175, 0.0281, 0.1519, 0.1635, 0.1488, 0.1020], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0238, 0.0280, 0.0223, 0.0241, 0.0266, 0.0274, 0.0241], + 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') +2023-02-06 04:06:24,074 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.66 vs. limit=5.0 +2023-02-06 04:06:38,743 INFO [train.py:901] (2/4) Epoch 7, batch 3800, loss[loss=0.2849, simple_loss=0.3497, pruned_loss=0.11, over 8525.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3319, pruned_loss=0.09782, over 1611733.85 frames. ], batch size: 49, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:07:01,885 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52330.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:07:04,421 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.417e+02 2.780e+02 3.361e+02 4.228e+02 6.516e+02, threshold=6.722e+02, percent-clipped=0.0 +2023-02-06 04:07:15,844 INFO [train.py:901] (2/4) Epoch 7, batch 3850, loss[loss=0.2205, simple_loss=0.2958, pruned_loss=0.07255, over 7642.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3303, pruned_loss=0.09701, over 1611448.19 frames. ], batch size: 19, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:07:25,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3617, 1.5846, 2.3342, 1.2131, 1.6268, 1.6612, 1.5024, 1.5303], + device='cuda:2'), covar=tensor([0.1601, 0.1812, 0.0608, 0.3272, 0.1320, 0.2585, 0.1615, 0.1707], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0474, 0.0533, 0.0549, 0.0591, 0.0528, 0.0458, 0.0595], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 04:07:30,474 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 04:07:35,590 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.70 vs. limit=5.0 +2023-02-06 04:07:49,730 INFO [train.py:901] (2/4) Epoch 7, batch 3900, loss[loss=0.3023, simple_loss=0.351, pruned_loss=0.1268, over 7805.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3321, pruned_loss=0.09876, over 1607928.26 frames. ], batch size: 20, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:07:51,994 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:08:15,068 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.697e+02 3.207e+02 4.225e+02 1.297e+03, threshold=6.414e+02, percent-clipped=5.0 +2023-02-06 04:08:25,215 INFO [train.py:901] (2/4) Epoch 7, batch 3950, loss[loss=0.2571, simple_loss=0.3327, pruned_loss=0.09075, over 8105.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3322, pruned_loss=0.09846, over 1609955.30 frames. ], batch size: 23, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:08:48,005 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52480.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:00,582 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 04:09:00,796 INFO [train.py:901] (2/4) Epoch 7, batch 4000, loss[loss=0.3386, simple_loss=0.3857, pruned_loss=0.1457, over 8567.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3317, pruned_loss=0.09848, over 1608284.67 frames. ], batch size: 49, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:09:04,524 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7990, 2.4446, 2.8107, 1.0866, 2.7439, 1.6399, 1.6003, 1.6730], + device='cuda:2'), covar=tensor([0.0437, 0.0166, 0.0102, 0.0371, 0.0179, 0.0430, 0.0378, 0.0247], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0257, 0.0215, 0.0317, 0.0258, 0.0403, 0.0311, 0.0292], + device='cuda:2'), out_proj_covar=tensor([1.1301e-04, 8.0308e-05, 6.6344e-05, 9.9245e-05, 8.1370e-05, 1.3735e-04, + 9.9871e-05, 9.1926e-05], device='cuda:2') +2023-02-06 04:09:05,165 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52505.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:13,175 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:23,937 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 04:09:24,189 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.935e+02 3.629e+02 4.693e+02 1.248e+03, threshold=7.258e+02, percent-clipped=9.0 +2023-02-06 04:09:28,447 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1213, 2.2860, 1.9311, 2.6950, 1.2943, 1.5397, 1.9859, 2.5017], + device='cuda:2'), covar=tensor([0.0867, 0.0951, 0.1319, 0.0615, 0.1518, 0.1695, 0.1191, 0.0734], + device='cuda:2'), in_proj_covar=tensor([0.0258, 0.0240, 0.0282, 0.0224, 0.0243, 0.0268, 0.0278, 0.0240], + 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') +2023-02-06 04:09:35,740 INFO [train.py:901] (2/4) Epoch 7, batch 4050, loss[loss=0.2517, simple_loss=0.323, pruned_loss=0.09017, over 8619.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3312, pruned_loss=0.09741, over 1608291.52 frames. ], batch size: 39, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:09:39,787 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52555.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:52,443 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52573.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:10:10,473 INFO [train.py:901] (2/4) Epoch 7, batch 4100, loss[loss=0.2464, simple_loss=0.3212, pruned_loss=0.08578, over 8197.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3312, pruned_loss=0.09765, over 1608739.63 frames. ], batch size: 23, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:10:33,844 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.709e+02 3.346e+02 4.687e+02 1.096e+03, threshold=6.691e+02, percent-clipped=5.0 +2023-02-06 04:10:44,015 INFO [train.py:901] (2/4) Epoch 7, batch 4150, loss[loss=0.2422, simple_loss=0.3188, pruned_loss=0.08275, over 7975.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3326, pruned_loss=0.09849, over 1614684.16 frames. ], batch size: 21, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:10:59,829 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52670.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:11:02,381 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:11:20,448 INFO [train.py:901] (2/4) Epoch 7, batch 4200, loss[loss=0.3323, simple_loss=0.3687, pruned_loss=0.148, over 8129.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3328, pruned_loss=0.09882, over 1612138.47 frames. ], batch size: 22, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:11:27,420 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5557, 2.8017, 2.4354, 3.7685, 1.8110, 1.7828, 2.3880, 2.9066], + device='cuda:2'), covar=tensor([0.0801, 0.1028, 0.1221, 0.0339, 0.1448, 0.1727, 0.1298, 0.0906], + device='cuda:2'), in_proj_covar=tensor([0.0255, 0.0239, 0.0281, 0.0224, 0.0241, 0.0269, 0.0276, 0.0239], + 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') +2023-02-06 04:11:30,519 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 04:11:43,884 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.675e+02 3.334e+02 4.108e+02 1.082e+03, threshold=6.669e+02, percent-clipped=4.0 +2023-02-06 04:11:49,448 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5220, 1.9347, 1.9840, 1.1035, 2.1956, 1.3053, 0.6421, 1.8148], + device='cuda:2'), covar=tensor([0.0242, 0.0124, 0.0097, 0.0214, 0.0135, 0.0359, 0.0353, 0.0107], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0259, 0.0218, 0.0319, 0.0257, 0.0405, 0.0314, 0.0296], + device='cuda:2'), out_proj_covar=tensor([1.1252e-04, 8.0884e-05, 6.7414e-05, 9.9557e-05, 8.0962e-05, 1.3825e-04, + 1.0056e-04, 9.2977e-05], device='cuda:2') +2023-02-06 04:11:49,487 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3007, 1.7285, 3.1218, 2.3115, 2.5604, 1.8659, 1.4147, 1.4722], + device='cuda:2'), covar=tensor([0.2660, 0.3272, 0.0630, 0.1605, 0.1373, 0.1788, 0.1570, 0.2731], + device='cuda:2'), in_proj_covar=tensor([0.0830, 0.0774, 0.0667, 0.0762, 0.0858, 0.0711, 0.0658, 0.0706], + 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') +2023-02-06 04:11:53,177 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 04:11:53,286 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7438, 5.7763, 4.9637, 2.1293, 5.0400, 5.3475, 5.3472, 5.0005], + device='cuda:2'), covar=tensor([0.0548, 0.0374, 0.0758, 0.4373, 0.0625, 0.0708, 0.0927, 0.0554], + device='cuda:2'), in_proj_covar=tensor([0.0422, 0.0323, 0.0357, 0.0440, 0.0342, 0.0319, 0.0326, 0.0280], + 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') +2023-02-06 04:11:53,853 INFO [train.py:901] (2/4) Epoch 7, batch 4250, loss[loss=0.256, simple_loss=0.3321, pruned_loss=0.08996, over 8261.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3328, pruned_loss=0.099, over 1611608.55 frames. ], batch size: 24, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:12:08,140 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52770.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:10,301 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:17,100 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0724, 1.7144, 1.3719, 1.7304, 1.4418, 1.2320, 1.3157, 1.4304], + device='cuda:2'), covar=tensor([0.0893, 0.0399, 0.0899, 0.0391, 0.0549, 0.1022, 0.0690, 0.0629], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0236, 0.0305, 0.0298, 0.0307, 0.0314, 0.0335, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 04:12:22,418 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:28,470 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:28,933 INFO [train.py:901] (2/4) Epoch 7, batch 4300, loss[loss=0.2787, simple_loss=0.3544, pruned_loss=0.1015, over 8466.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3329, pruned_loss=0.09887, over 1611681.64 frames. ], batch size: 27, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:12:53,660 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.879e+02 3.462e+02 4.347e+02 1.112e+03, threshold=6.924e+02, percent-clipped=5.0 +2023-02-06 04:13:03,872 INFO [train.py:901] (2/4) Epoch 7, batch 4350, loss[loss=0.2502, simple_loss=0.316, pruned_loss=0.09225, over 7802.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3306, pruned_loss=0.09752, over 1607173.28 frames. ], batch size: 20, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:13:14,670 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4921, 1.9206, 2.1296, 1.0150, 2.3931, 1.2725, 0.7280, 1.7940], + device='cuda:2'), covar=tensor([0.0360, 0.0171, 0.0124, 0.0310, 0.0153, 0.0489, 0.0471, 0.0159], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0263, 0.0220, 0.0320, 0.0260, 0.0408, 0.0317, 0.0298], + device='cuda:2'), out_proj_covar=tensor([1.1267e-04, 8.2178e-05, 6.8024e-05, 9.9893e-05, 8.2111e-05, 1.3864e-04, + 1.0155e-04, 9.3844e-05], device='cuda:2') +2023-02-06 04:13:24,425 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 04:13:38,493 INFO [train.py:901] (2/4) Epoch 7, batch 4400, loss[loss=0.2863, simple_loss=0.3417, pruned_loss=0.1155, over 6889.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3309, pruned_loss=0.09758, over 1606317.75 frames. ], batch size: 72, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:13:50,776 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52917.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:13:56,762 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:14:02,505 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.715e+02 3.689e+02 4.508e+02 8.331e+02, threshold=7.379e+02, percent-clipped=6.0 +2023-02-06 04:14:06,678 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 04:14:13,200 INFO [train.py:901] (2/4) Epoch 7, batch 4450, loss[loss=0.3228, simple_loss=0.3642, pruned_loss=0.1407, over 8088.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3309, pruned_loss=0.09737, over 1611198.56 frames. ], batch size: 21, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:14:14,728 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:14:38,155 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4446, 1.7775, 3.1528, 2.4416, 2.5792, 1.8873, 1.4754, 1.2788], + device='cuda:2'), covar=tensor([0.2669, 0.3258, 0.0690, 0.1666, 0.1466, 0.1744, 0.1656, 0.3104], + device='cuda:2'), in_proj_covar=tensor([0.0829, 0.0774, 0.0676, 0.0768, 0.0863, 0.0712, 0.0662, 0.0711], + 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') +2023-02-06 04:14:46,593 INFO [train.py:901] (2/4) Epoch 7, batch 4500, loss[loss=0.3048, simple_loss=0.3628, pruned_loss=0.1233, over 8505.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3335, pruned_loss=0.09963, over 1610728.26 frames. ], batch size: 28, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:14:59,361 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 04:15:10,357 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53032.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:15:11,488 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 2.890e+02 3.405e+02 4.030e+02 1.067e+03, threshold=6.809e+02, percent-clipped=4.0 +2023-02-06 04:15:18,996 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:15:22,052 INFO [train.py:901] (2/4) Epoch 7, batch 4550, loss[loss=0.3101, simple_loss=0.3689, pruned_loss=0.1257, over 8521.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3319, pruned_loss=0.09816, over 1611910.01 frames. ], batch size: 28, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:15:33,123 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8002, 2.0608, 1.8534, 1.4310, 2.1148, 1.6320, 1.2251, 1.7359], + device='cuda:2'), covar=tensor([0.0244, 0.0142, 0.0122, 0.0207, 0.0136, 0.0256, 0.0312, 0.0129], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0264, 0.0221, 0.0323, 0.0262, 0.0409, 0.0319, 0.0298], + device='cuda:2'), out_proj_covar=tensor([1.1323e-04, 8.2176e-05, 6.8216e-05, 1.0069e-04, 8.2625e-05, 1.3878e-04, + 1.0222e-04, 9.3637e-05], device='cuda:2') +2023-02-06 04:15:36,990 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:15:39,628 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53074.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:15:56,227 INFO [train.py:901] (2/4) Epoch 7, batch 4600, loss[loss=0.242, simple_loss=0.3245, pruned_loss=0.07975, over 8289.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3322, pruned_loss=0.09789, over 1610957.17 frames. ], batch size: 23, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:16:06,495 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53114.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:16:20,974 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.951e+02 3.579e+02 4.375e+02 1.013e+03, threshold=7.158e+02, percent-clipped=5.0 +2023-02-06 04:16:31,864 INFO [train.py:901] (2/4) Epoch 7, batch 4650, loss[loss=0.2701, simple_loss=0.3417, pruned_loss=0.09929, over 8477.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3306, pruned_loss=0.09673, over 1611336.74 frames. ], batch size: 25, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:17:06,494 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3954, 1.9054, 3.4237, 1.0596, 2.4158, 1.8658, 1.4802, 2.2071], + device='cuda:2'), covar=tensor([0.1722, 0.2048, 0.0703, 0.3717, 0.1543, 0.2696, 0.1754, 0.2318], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0480, 0.0531, 0.0557, 0.0603, 0.0540, 0.0457, 0.0597], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 04:17:06,934 INFO [train.py:901] (2/4) Epoch 7, batch 4700, loss[loss=0.2459, simple_loss=0.3342, pruned_loss=0.07883, over 8200.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3312, pruned_loss=0.09716, over 1610900.43 frames. ], batch size: 23, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:17:27,472 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:17:30,531 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.793e+02 3.469e+02 4.275e+02 9.300e+02, threshold=6.939e+02, percent-clipped=3.0 +2023-02-06 04:17:34,228 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3209, 1.7826, 2.9669, 2.2245, 2.4383, 2.0164, 1.5099, 1.0996], + device='cuda:2'), covar=tensor([0.2448, 0.2852, 0.0618, 0.1638, 0.1295, 0.1451, 0.1330, 0.2986], + device='cuda:2'), in_proj_covar=tensor([0.0819, 0.0764, 0.0659, 0.0757, 0.0844, 0.0700, 0.0651, 0.0696], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:17:41,203 INFO [train.py:901] (2/4) Epoch 7, batch 4750, loss[loss=0.2298, simple_loss=0.2877, pruned_loss=0.08598, over 7798.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3312, pruned_loss=0.09745, over 1609492.22 frames. ], batch size: 19, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:17:48,716 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:17:56,643 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 04:18:00,062 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 04:18:02,329 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7024, 1.9082, 1.6036, 2.3158, 0.9707, 1.3929, 1.6657, 1.8302], + device='cuda:2'), covar=tensor([0.0822, 0.0962, 0.1315, 0.0529, 0.1513, 0.1728, 0.1021, 0.0962], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0238, 0.0279, 0.0224, 0.0239, 0.0266, 0.0276, 0.0238], + 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') +2023-02-06 04:18:09,725 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53288.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:18:16,605 INFO [train.py:901] (2/4) Epoch 7, batch 4800, loss[loss=0.2588, simple_loss=0.3348, pruned_loss=0.09144, over 8320.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3309, pruned_loss=0.09658, over 1613413.45 frames. ], batch size: 25, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:18:26,448 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53313.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:18:32,416 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:18:39,897 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53333.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:18:41,111 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.630e+02 3.191e+02 3.984e+02 9.617e+02, threshold=6.381e+02, percent-clipped=3.0 +2023-02-06 04:18:49,306 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8389, 1.9004, 4.2625, 1.9082, 2.2677, 4.9230, 4.9294, 4.3085], + device='cuda:2'), covar=tensor([0.0815, 0.1382, 0.0300, 0.1887, 0.0904, 0.0222, 0.0292, 0.0548], + device='cuda:2'), in_proj_covar=tensor([0.0240, 0.0278, 0.0233, 0.0274, 0.0244, 0.0220, 0.0279, 0.0283], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 04:18:50,454 INFO [train.py:901] (2/4) Epoch 7, batch 4850, loss[loss=0.2748, simple_loss=0.342, pruned_loss=0.1038, over 8416.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.332, pruned_loss=0.09773, over 1614145.31 frames. ], batch size: 27, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:18:51,160 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 04:19:16,477 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:19:26,608 INFO [train.py:901] (2/4) Epoch 7, batch 4900, loss[loss=0.2406, simple_loss=0.3145, pruned_loss=0.08331, over 8092.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3311, pruned_loss=0.0969, over 1616194.33 frames. ], batch size: 21, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:19:35,805 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 04:19:40,217 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:19:51,455 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.811e+02 3.269e+02 4.328e+02 9.769e+02, threshold=6.539e+02, percent-clipped=6.0 +2023-02-06 04:20:00,859 INFO [train.py:901] (2/4) Epoch 7, batch 4950, loss[loss=0.2954, simple_loss=0.3495, pruned_loss=0.1206, over 8334.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3326, pruned_loss=0.0988, over 1608016.17 frames. ], batch size: 26, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:20:20,045 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6658, 2.1983, 4.7299, 2.5943, 4.2811, 4.0504, 4.4552, 4.3021], + device='cuda:2'), covar=tensor([0.0461, 0.3313, 0.0483, 0.2621, 0.0806, 0.0636, 0.0406, 0.0472], + device='cuda:2'), in_proj_covar=tensor([0.0392, 0.0537, 0.0487, 0.0469, 0.0531, 0.0447, 0.0440, 0.0504], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 04:20:27,197 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53485.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:20:36,815 INFO [train.py:901] (2/4) Epoch 7, batch 5000, loss[loss=0.2698, simple_loss=0.3484, pruned_loss=0.09562, over 8434.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.332, pruned_loss=0.09819, over 1609752.56 frames. ], batch size: 29, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:20:45,259 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:20:46,626 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4233, 1.4739, 1.6344, 1.2977, 1.1065, 1.4478, 1.7503, 1.5766], + device='cuda:2'), covar=tensor([0.0552, 0.1249, 0.1955, 0.1510, 0.0639, 0.1583, 0.0707, 0.0614], + device='cuda:2'), in_proj_covar=tensor([0.0117, 0.0166, 0.0206, 0.0169, 0.0116, 0.0174, 0.0127, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 04:21:01,977 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:21:03,199 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.612e+02 3.156e+02 4.000e+02 8.821e+02, threshold=6.312e+02, percent-clipped=7.0 +2023-02-06 04:21:12,888 INFO [train.py:901] (2/4) Epoch 7, batch 5050, loss[loss=0.2573, simple_loss=0.3296, pruned_loss=0.0925, over 8736.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3306, pruned_loss=0.09709, over 1609507.85 frames. ], batch size: 30, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:21:31,984 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 04:21:46,773 INFO [train.py:901] (2/4) Epoch 7, batch 5100, loss[loss=0.3076, simple_loss=0.358, pruned_loss=0.1286, over 8369.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3321, pruned_loss=0.09834, over 1614187.53 frames. ], batch size: 49, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:21:51,221 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:22:04,007 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.5564, 1.6956, 3.6814, 1.9160, 3.3343, 3.1668, 3.3696, 3.3076], + device='cuda:2'), covar=tensor([0.0483, 0.2800, 0.0594, 0.2554, 0.0912, 0.0684, 0.0467, 0.0511], + device='cuda:2'), in_proj_covar=tensor([0.0390, 0.0525, 0.0483, 0.0464, 0.0527, 0.0439, 0.0434, 0.0495], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 04:22:13,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 2.889e+02 3.391e+02 4.238e+02 9.606e+02, threshold=6.783e+02, percent-clipped=10.0 +2023-02-06 04:22:23,446 INFO [train.py:901] (2/4) Epoch 7, batch 5150, loss[loss=0.1993, simple_loss=0.2794, pruned_loss=0.05964, over 7821.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3306, pruned_loss=0.09694, over 1607985.19 frames. ], batch size: 20, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:22:34,956 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:22:40,557 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=6.40 vs. limit=5.0 +2023-02-06 04:22:42,267 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:22:56,582 INFO [train.py:901] (2/4) Epoch 7, batch 5200, loss[loss=0.2279, simple_loss=0.298, pruned_loss=0.07896, over 7809.00 frames. ], tot_loss[loss=0.262, simple_loss=0.33, pruned_loss=0.09698, over 1605143.61 frames. ], batch size: 19, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:23:10,747 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:23:17,975 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53729.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:23:22,034 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.983e+02 3.078e+02 4.028e+02 5.378e+02 1.177e+03, threshold=8.056e+02, percent-clipped=8.0 +2023-02-06 04:23:28,936 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 04:23:32,186 INFO [train.py:901] (2/4) Epoch 7, batch 5250, loss[loss=0.2883, simple_loss=0.3548, pruned_loss=0.1109, over 8493.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3299, pruned_loss=0.09684, over 1605416.79 frames. ], batch size: 26, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:23:54,701 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53781.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:00,279 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:02,337 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53792.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:06,924 INFO [train.py:901] (2/4) Epoch 7, batch 5300, loss[loss=0.2872, simple_loss=0.3568, pruned_loss=0.1089, over 8459.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3295, pruned_loss=0.0971, over 1603825.30 frames. ], batch size: 25, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:24:17,554 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53814.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:31,353 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.647e+02 3.169e+02 3.870e+02 1.211e+03, threshold=6.339e+02, percent-clipped=2.0 +2023-02-06 04:24:39,056 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53844.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:42,222 INFO [train.py:901] (2/4) Epoch 7, batch 5350, loss[loss=0.2475, simple_loss=0.3178, pruned_loss=0.08863, over 7915.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3297, pruned_loss=0.09776, over 1604814.00 frames. ], batch size: 20, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:25:17,624 INFO [train.py:901] (2/4) Epoch 7, batch 5400, loss[loss=0.2495, simple_loss=0.321, pruned_loss=0.08903, over 8112.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3299, pruned_loss=0.09735, over 1608601.47 frames. ], batch size: 23, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:25:41,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.728e+02 3.458e+02 4.119e+02 1.009e+03, threshold=6.915e+02, percent-clipped=3.0 +2023-02-06 04:25:51,250 INFO [train.py:901] (2/4) Epoch 7, batch 5450, loss[loss=0.239, simple_loss=0.3079, pruned_loss=0.08507, over 7820.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3299, pruned_loss=0.09743, over 1606952.73 frames. ], batch size: 20, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:25:55,229 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.10 vs. limit=5.0 +2023-02-06 04:26:05,668 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.57 vs. limit=5.0 +2023-02-06 04:26:09,514 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:26:18,658 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 04:26:21,819 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 04:26:26,957 INFO [train.py:901] (2/4) Epoch 7, batch 5500, loss[loss=0.2994, simple_loss=0.3485, pruned_loss=0.1251, over 7316.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.33, pruned_loss=0.09724, over 1609797.33 frames. ], batch size: 72, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:26:27,120 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:26:52,075 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.794e+02 3.496e+02 4.646e+02 1.157e+03, threshold=6.993e+02, percent-clipped=7.0 +2023-02-06 04:26:53,576 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54037.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:01,154 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:01,615 INFO [train.py:901] (2/4) Epoch 7, batch 5550, loss[loss=0.2506, simple_loss=0.3273, pruned_loss=0.08699, over 8188.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3305, pruned_loss=0.097, over 1615206.29 frames. ], batch size: 23, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:27:10,528 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:18,052 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54073.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:23,949 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7408, 5.7439, 5.1672, 2.0912, 5.1716, 5.4991, 5.4391, 5.2071], + device='cuda:2'), covar=tensor([0.0544, 0.0447, 0.0727, 0.4367, 0.0560, 0.0577, 0.0897, 0.0544], + device='cuda:2'), in_proj_covar=tensor([0.0413, 0.0322, 0.0348, 0.0430, 0.0337, 0.0314, 0.0324, 0.0273], + 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') +2023-02-06 04:27:37,055 INFO [train.py:901] (2/4) Epoch 7, batch 5600, loss[loss=0.299, simple_loss=0.3579, pruned_loss=0.1201, over 8264.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3319, pruned_loss=0.09816, over 1615726.27 frames. ], batch size: 49, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:27:37,987 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54100.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:55,819 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:28:02,448 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.795e+02 3.455e+02 4.516e+02 9.788e+02, threshold=6.911e+02, percent-clipped=3.0 +2023-02-06 04:28:12,193 INFO [train.py:901] (2/4) Epoch 7, batch 5650, loss[loss=0.213, simple_loss=0.2862, pruned_loss=0.06985, over 7424.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3313, pruned_loss=0.09701, over 1617513.78 frames. ], batch size: 17, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:28:25,363 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 04:28:47,264 INFO [train.py:901] (2/4) Epoch 7, batch 5700, loss[loss=0.2618, simple_loss=0.3387, pruned_loss=0.09249, over 8623.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3299, pruned_loss=0.09624, over 1615657.54 frames. ], batch size: 34, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:29:01,029 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0626, 1.7082, 1.3137, 1.6270, 1.4380, 1.1149, 1.0920, 1.3957], + device='cuda:2'), covar=tensor([0.1011, 0.0385, 0.1051, 0.0490, 0.0617, 0.1272, 0.0913, 0.0666], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0239, 0.0311, 0.0303, 0.0314, 0.0319, 0.0342, 0.0325], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 04:29:12,985 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.786e+02 3.155e+02 4.023e+02 8.991e+02, threshold=6.311e+02, percent-clipped=4.0 +2023-02-06 04:29:22,459 INFO [train.py:901] (2/4) Epoch 7, batch 5750, loss[loss=0.2641, simple_loss=0.3292, pruned_loss=0.09951, over 8235.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3299, pruned_loss=0.09618, over 1611123.38 frames. ], batch size: 22, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:29:31,456 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 04:29:53,945 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 04:29:56,273 INFO [train.py:901] (2/4) Epoch 7, batch 5800, loss[loss=0.2402, simple_loss=0.3255, pruned_loss=0.07743, over 8493.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3292, pruned_loss=0.09559, over 1610385.04 frames. ], batch size: 28, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:30:22,027 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.127e+02 2.882e+02 3.737e+02 4.385e+02 9.194e+02, threshold=7.474e+02, percent-clipped=5.0 +2023-02-06 04:30:32,272 INFO [train.py:901] (2/4) Epoch 7, batch 5850, loss[loss=0.2532, simple_loss=0.3317, pruned_loss=0.08736, over 8350.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3292, pruned_loss=0.09529, over 1612145.49 frames. ], batch size: 26, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:30:58,370 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.48 vs. limit=5.0 +2023-02-06 04:31:06,107 INFO [train.py:901] (2/4) Epoch 7, batch 5900, loss[loss=0.2038, simple_loss=0.2697, pruned_loss=0.06892, over 7437.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3298, pruned_loss=0.0961, over 1610612.29 frames. ], batch size: 17, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:31:30,649 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.634e+02 3.151e+02 3.851e+02 7.879e+02, threshold=6.301e+02, percent-clipped=2.0 +2023-02-06 04:31:40,692 INFO [train.py:901] (2/4) Epoch 7, batch 5950, loss[loss=0.2333, simple_loss=0.3037, pruned_loss=0.08148, over 7434.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3288, pruned_loss=0.09569, over 1607797.69 frames. ], batch size: 17, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:32:14,299 INFO [train.py:901] (2/4) Epoch 7, batch 6000, loss[loss=0.2551, simple_loss=0.341, pruned_loss=0.08467, over 8109.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3301, pruned_loss=0.0966, over 1604740.36 frames. ], batch size: 23, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:32:14,299 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 04:32:26,544 INFO [train.py:935] (2/4) Epoch 7, validation: loss=0.2048, simple_loss=0.3036, pruned_loss=0.05298, over 944034.00 frames. +2023-02-06 04:32:26,544 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 04:32:49,996 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 04:32:50,868 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.687e+02 3.524e+02 4.445e+02 8.914e+02, threshold=7.048e+02, percent-clipped=8.0 +2023-02-06 04:33:00,125 INFO [train.py:901] (2/4) Epoch 7, batch 6050, loss[loss=0.2981, simple_loss=0.3728, pruned_loss=0.1117, over 8438.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3309, pruned_loss=0.09707, over 1612015.57 frames. ], batch size: 29, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:33:00,357 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3469, 1.7241, 1.5705, 0.9533, 1.6591, 1.2773, 0.2768, 1.5493], + device='cuda:2'), covar=tensor([0.0219, 0.0152, 0.0145, 0.0213, 0.0177, 0.0468, 0.0400, 0.0121], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0264, 0.0225, 0.0327, 0.0264, 0.0417, 0.0324, 0.0303], + device='cuda:2'), out_proj_covar=tensor([1.1182e-04, 8.1893e-05, 6.8761e-05, 1.0064e-04, 8.2835e-05, 1.4052e-04, + 1.0283e-04, 9.4368e-05], device='cuda:2') +2023-02-06 04:33:36,263 INFO [train.py:901] (2/4) Epoch 7, batch 6100, loss[loss=0.2462, simple_loss=0.3245, pruned_loss=0.08392, over 8502.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3309, pruned_loss=0.09766, over 1611664.60 frames. ], batch size: 26, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:33:51,051 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2460, 1.7799, 2.7691, 2.1228, 2.2241, 1.9443, 1.5132, 0.9964], + device='cuda:2'), covar=tensor([0.2658, 0.2831, 0.0622, 0.1589, 0.1276, 0.1576, 0.1585, 0.2648], + device='cuda:2'), in_proj_covar=tensor([0.0834, 0.0777, 0.0661, 0.0768, 0.0857, 0.0713, 0.0664, 0.0704], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:34:00,446 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 04:34:01,817 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.824e+02 3.447e+02 4.351e+02 1.012e+03, threshold=6.894e+02, percent-clipped=2.0 +2023-02-06 04:34:11,158 INFO [train.py:901] (2/4) Epoch 7, batch 6150, loss[loss=0.325, simple_loss=0.3812, pruned_loss=0.1344, over 8483.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3308, pruned_loss=0.09777, over 1609961.34 frames. ], batch size: 49, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:34:27,174 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3303, 1.8611, 3.1139, 2.3717, 2.5682, 2.0936, 1.4955, 1.2249], + device='cuda:2'), covar=tensor([0.2856, 0.3110, 0.0652, 0.1614, 0.1348, 0.1435, 0.1408, 0.3018], + device='cuda:2'), in_proj_covar=tensor([0.0837, 0.0780, 0.0665, 0.0771, 0.0862, 0.0717, 0.0667, 0.0710], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:34:34,034 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54682.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:34:46,675 INFO [train.py:901] (2/4) Epoch 7, batch 6200, loss[loss=0.2878, simple_loss=0.3542, pruned_loss=0.1107, over 8428.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3308, pruned_loss=0.09693, over 1614681.96 frames. ], batch size: 29, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:35:12,146 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.928e+02 3.624e+02 4.953e+02 9.267e+02, threshold=7.248e+02, percent-clipped=4.0 +2023-02-06 04:35:12,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4315, 1.3837, 2.3114, 1.1256, 2.1704, 2.4626, 2.4939, 2.0769], + device='cuda:2'), covar=tensor([0.0859, 0.1091, 0.0465, 0.1791, 0.0598, 0.0353, 0.0519, 0.0692], + device='cuda:2'), in_proj_covar=tensor([0.0241, 0.0273, 0.0231, 0.0270, 0.0243, 0.0216, 0.0275, 0.0279], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 04:35:21,774 INFO [train.py:901] (2/4) Epoch 7, batch 6250, loss[loss=0.2528, simple_loss=0.3207, pruned_loss=0.09242, over 7655.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.332, pruned_loss=0.09837, over 1614188.49 frames. ], batch size: 19, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:35:42,946 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7919, 3.8513, 2.3507, 2.4506, 2.9841, 2.0475, 2.5611, 2.7283], + device='cuda:2'), covar=tensor([0.1579, 0.0199, 0.0890, 0.0776, 0.0559, 0.1115, 0.0922, 0.0949], + device='cuda:2'), in_proj_covar=tensor([0.0339, 0.0229, 0.0307, 0.0294, 0.0307, 0.0309, 0.0332, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 04:35:55,498 INFO [train.py:901] (2/4) Epoch 7, batch 6300, loss[loss=0.2624, simple_loss=0.3301, pruned_loss=0.09735, over 7659.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3306, pruned_loss=0.0977, over 1613044.54 frames. ], batch size: 19, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:36:22,281 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 3.001e+02 3.662e+02 4.451e+02 9.002e+02, threshold=7.325e+02, percent-clipped=3.0 +2023-02-06 04:36:32,311 INFO [train.py:901] (2/4) Epoch 7, batch 6350, loss[loss=0.302, simple_loss=0.3706, pruned_loss=0.1167, over 8489.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3317, pruned_loss=0.09817, over 1616820.27 frames. ], batch size: 39, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:36:45,393 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2886, 2.2072, 1.6922, 2.0863, 1.7330, 1.2929, 1.6073, 1.7459], + device='cuda:2'), covar=tensor([0.1141, 0.0315, 0.0910, 0.0399, 0.0625, 0.1251, 0.0794, 0.0768], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0236, 0.0312, 0.0298, 0.0314, 0.0317, 0.0337, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 04:36:53,638 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54880.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:37:05,996 INFO [train.py:901] (2/4) Epoch 7, batch 6400, loss[loss=0.2383, simple_loss=0.3101, pruned_loss=0.08323, over 8251.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3317, pruned_loss=0.0984, over 1615826.27 frames. ], batch size: 22, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:37:31,193 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.661e+02 3.281e+02 3.949e+02 1.010e+03, threshold=6.562e+02, percent-clipped=2.0 +2023-02-06 04:37:40,710 INFO [train.py:901] (2/4) Epoch 7, batch 6450, loss[loss=0.2167, simple_loss=0.2878, pruned_loss=0.07276, over 7803.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3305, pruned_loss=0.09785, over 1609404.63 frames. ], batch size: 19, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:38:15,584 INFO [train.py:901] (2/4) Epoch 7, batch 6500, loss[loss=0.2275, simple_loss=0.3073, pruned_loss=0.0739, over 7982.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3288, pruned_loss=0.09617, over 1611816.43 frames. ], batch size: 21, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:38:33,804 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55026.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:38:39,665 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.546e+02 3.271e+02 4.197e+02 5.859e+02, threshold=6.542e+02, percent-clipped=0.0 +2023-02-06 04:38:49,579 INFO [train.py:901] (2/4) Epoch 7, batch 6550, loss[loss=0.2922, simple_loss=0.3652, pruned_loss=0.1096, over 8312.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3303, pruned_loss=0.09717, over 1612481.16 frames. ], batch size: 25, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:38:58,667 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6537, 1.9970, 2.1595, 1.5663, 1.1501, 2.2057, 0.3106, 1.4105], + device='cuda:2'), covar=tensor([0.2744, 0.1810, 0.0590, 0.2386, 0.5349, 0.0692, 0.4123, 0.2104], + device='cuda:2'), in_proj_covar=tensor([0.0142, 0.0143, 0.0084, 0.0191, 0.0228, 0.0089, 0.0147, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:39:12,198 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 04:39:25,813 INFO [train.py:901] (2/4) Epoch 7, batch 6600, loss[loss=0.2356, simple_loss=0.3098, pruned_loss=0.08074, over 8235.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3303, pruned_loss=0.09669, over 1616276.54 frames. ], batch size: 22, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:39:27,508 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 04:39:32,344 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 04:39:49,421 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.699e+02 3.503e+02 4.413e+02 7.218e+02, threshold=7.007e+02, percent-clipped=4.0 +2023-02-06 04:39:53,570 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55141.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:39:58,776 INFO [train.py:901] (2/4) Epoch 7, batch 6650, loss[loss=0.2219, simple_loss=0.297, pruned_loss=0.0734, over 7810.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3307, pruned_loss=0.0973, over 1615427.57 frames. ], batch size: 19, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:40:10,778 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55166.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:40:32,844 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2678, 4.2221, 3.7800, 1.7954, 3.7119, 3.8502, 3.9255, 3.5049], + device='cuda:2'), covar=tensor([0.1020, 0.0655, 0.1171, 0.5146, 0.0915, 0.0907, 0.1392, 0.1089], + device='cuda:2'), in_proj_covar=tensor([0.0418, 0.0322, 0.0356, 0.0437, 0.0343, 0.0319, 0.0331, 0.0282], + 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') +2023-02-06 04:40:34,123 INFO [train.py:901] (2/4) Epoch 7, batch 6700, loss[loss=0.234, simple_loss=0.3052, pruned_loss=0.08135, over 7564.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3317, pruned_loss=0.09778, over 1615971.95 frames. ], batch size: 18, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:40:38,422 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5504, 2.0406, 3.2263, 2.6289, 2.6723, 2.1102, 1.7036, 1.3920], + device='cuda:2'), covar=tensor([0.2544, 0.2886, 0.0787, 0.1662, 0.1557, 0.1577, 0.1361, 0.3148], + device='cuda:2'), in_proj_covar=tensor([0.0831, 0.0777, 0.0675, 0.0773, 0.0866, 0.0714, 0.0668, 0.0705], + 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') +2023-02-06 04:40:51,595 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55224.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:40:55,049 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:40:55,085 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8406, 1.4931, 1.5575, 1.2382, 1.0147, 1.3628, 1.4904, 1.4110], + device='cuda:2'), covar=tensor([0.0510, 0.1211, 0.1742, 0.1376, 0.0593, 0.1501, 0.0671, 0.0591], + device='cuda:2'), in_proj_covar=tensor([0.0113, 0.0166, 0.0206, 0.0169, 0.0115, 0.0172, 0.0126, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 04:40:58,789 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 3.031e+02 3.759e+02 4.673e+02 1.170e+03, threshold=7.519e+02, percent-clipped=9.0 +2023-02-06 04:41:07,993 INFO [train.py:901] (2/4) Epoch 7, batch 6750, loss[loss=0.2607, simple_loss=0.3193, pruned_loss=0.1011, over 7797.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.331, pruned_loss=0.09736, over 1615845.90 frames. ], batch size: 20, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:41:12,180 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 04:41:42,537 INFO [train.py:901] (2/4) Epoch 7, batch 6800, loss[loss=0.2464, simple_loss=0.3226, pruned_loss=0.08508, over 8239.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3312, pruned_loss=0.09751, over 1612063.03 frames. ], batch size: 22, lr: 1.07e-02, grad_scale: 16.0 +2023-02-06 04:41:47,229 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 04:42:08,558 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.859e+02 3.364e+02 4.161e+02 9.626e+02, threshold=6.728e+02, percent-clipped=3.0 +2023-02-06 04:42:09,378 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2429, 4.1686, 3.8246, 1.9090, 3.6790, 3.6166, 3.8356, 3.3257], + device='cuda:2'), covar=tensor([0.0735, 0.0519, 0.0892, 0.4278, 0.0875, 0.0940, 0.1099, 0.0904], + device='cuda:2'), in_proj_covar=tensor([0.0414, 0.0323, 0.0355, 0.0434, 0.0339, 0.0320, 0.0328, 0.0281], + 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') +2023-02-06 04:42:11,504 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55339.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:42:18,203 INFO [train.py:901] (2/4) Epoch 7, batch 6850, loss[loss=0.3136, simple_loss=0.3802, pruned_loss=0.1235, over 8246.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3312, pruned_loss=0.09765, over 1608956.45 frames. ], batch size: 24, lr: 1.06e-02, grad_scale: 16.0 +2023-02-06 04:42:19,222 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=5.28 vs. limit=5.0 +2023-02-06 04:42:25,226 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.24 vs. limit=2.0 +2023-02-06 04:42:34,174 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 04:42:50,635 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55397.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:42:52,432 INFO [train.py:901] (2/4) Epoch 7, batch 6900, loss[loss=0.2203, simple_loss=0.2966, pruned_loss=0.07199, over 8248.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3316, pruned_loss=0.09792, over 1608605.29 frames. ], batch size: 22, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:43:09,634 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55422.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:43:19,277 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.767e+02 3.318e+02 4.413e+02 7.718e+02, threshold=6.635e+02, percent-clipped=1.0 +2023-02-06 04:43:28,911 INFO [train.py:901] (2/4) Epoch 7, batch 6950, loss[loss=0.2464, simple_loss=0.3281, pruned_loss=0.08235, over 8482.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3315, pruned_loss=0.09755, over 1610460.66 frames. ], batch size: 28, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:43:34,097 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 04:43:46,611 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 04:44:02,255 INFO [train.py:901] (2/4) Epoch 7, batch 7000, loss[loss=0.2586, simple_loss=0.3158, pruned_loss=0.1007, over 7528.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3312, pruned_loss=0.09675, over 1614961.83 frames. ], batch size: 18, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:44:09,454 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:44:28,246 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.942e+02 3.699e+02 4.542e+02 1.220e+03, threshold=7.399e+02, percent-clipped=11.0 +2023-02-06 04:44:37,052 INFO [train.py:901] (2/4) Epoch 7, batch 7050, loss[loss=0.258, simple_loss=0.3363, pruned_loss=0.08984, over 6670.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3308, pruned_loss=0.09669, over 1611530.14 frames. ], batch size: 73, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:44:54,229 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:09,300 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:11,225 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2327, 1.5896, 1.4539, 1.2675, 1.1273, 1.3728, 1.7183, 1.6213], + device='cuda:2'), covar=tensor([0.0531, 0.1186, 0.1758, 0.1441, 0.0554, 0.1476, 0.0709, 0.0566], + device='cuda:2'), in_proj_covar=tensor([0.0112, 0.0166, 0.0204, 0.0167, 0.0112, 0.0171, 0.0126, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 04:45:11,741 INFO [train.py:901] (2/4) Epoch 7, batch 7100, loss[loss=0.3103, simple_loss=0.3699, pruned_loss=0.1253, over 8581.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3299, pruned_loss=0.09631, over 1609922.41 frames. ], batch size: 31, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:45:26,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:29,035 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 +2023-02-06 04:45:29,489 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55625.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:36,689 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.720e+02 3.297e+02 4.008e+02 7.250e+02, threshold=6.594e+02, percent-clipped=0.0 +2023-02-06 04:45:45,950 INFO [train.py:901] (2/4) Epoch 7, batch 7150, loss[loss=0.2481, simple_loss=0.3198, pruned_loss=0.08822, over 8132.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3297, pruned_loss=0.09579, over 1612609.45 frames. ], batch size: 22, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:46:10,053 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9558, 1.4885, 3.4360, 1.5469, 2.3093, 3.9134, 3.8695, 3.3506], + device='cuda:2'), covar=tensor([0.1066, 0.1422, 0.0332, 0.1878, 0.0829, 0.0229, 0.0363, 0.0638], + device='cuda:2'), in_proj_covar=tensor([0.0244, 0.0273, 0.0231, 0.0270, 0.0242, 0.0214, 0.0278, 0.0280], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 04:46:14,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:46:21,652 INFO [train.py:901] (2/4) Epoch 7, batch 7200, loss[loss=0.2276, simple_loss=0.2911, pruned_loss=0.08205, over 5114.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3296, pruned_loss=0.09631, over 1609877.96 frames. ], batch size: 11, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:46:42,577 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6948, 1.8409, 2.2578, 1.7367, 1.1337, 2.3371, 0.4081, 1.2947], + device='cuda:2'), covar=tensor([0.3428, 0.1951, 0.0611, 0.2602, 0.5223, 0.0624, 0.4391, 0.2208], + device='cuda:2'), in_proj_covar=tensor([0.0142, 0.0144, 0.0084, 0.0191, 0.0230, 0.0089, 0.0145, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:46:47,138 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.857e+02 3.487e+02 4.455e+02 1.230e+03, threshold=6.974e+02, percent-clipped=5.0 +2023-02-06 04:46:49,878 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5888, 4.6304, 4.0372, 1.8942, 4.0627, 4.1665, 4.2462, 3.7098], + device='cuda:2'), covar=tensor([0.0874, 0.0681, 0.1163, 0.4967, 0.0861, 0.0744, 0.1445, 0.0872], + device='cuda:2'), in_proj_covar=tensor([0.0411, 0.0325, 0.0354, 0.0436, 0.0336, 0.0315, 0.0327, 0.0277], + 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') +2023-02-06 04:46:55,810 INFO [train.py:901] (2/4) Epoch 7, batch 7250, loss[loss=0.2435, simple_loss=0.3134, pruned_loss=0.08683, over 7942.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3312, pruned_loss=0.09719, over 1609238.36 frames. ], batch size: 20, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:47:07,795 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55766.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:47:31,018 INFO [train.py:901] (2/4) Epoch 7, batch 7300, loss[loss=0.2919, simple_loss=0.3585, pruned_loss=0.1127, over 8495.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3324, pruned_loss=0.09788, over 1610671.26 frames. ], batch size: 28, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:47:55,727 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.887e+02 3.402e+02 4.424e+02 1.529e+03, threshold=6.804e+02, percent-clipped=7.0 +2023-02-06 04:48:04,220 INFO [train.py:901] (2/4) Epoch 7, batch 7350, loss[loss=0.2608, simple_loss=0.3395, pruned_loss=0.09104, over 8598.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3318, pruned_loss=0.09786, over 1613129.81 frames. ], batch size: 31, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:48:15,840 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55866.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:48:26,612 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55881.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:48:27,829 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 04:48:40,221 INFO [train.py:901] (2/4) Epoch 7, batch 7400, loss[loss=0.2849, simple_loss=0.3531, pruned_loss=0.1083, over 8449.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.332, pruned_loss=0.09752, over 1611385.45 frames. ], batch size: 27, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:48:45,313 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55906.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:48:50,021 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 04:49:01,547 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 04:49:05,903 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.646e+02 3.471e+02 4.467e+02 1.348e+03, threshold=6.942e+02, percent-clipped=5.0 +2023-02-06 04:49:11,666 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:49:14,916 INFO [train.py:901] (2/4) Epoch 7, batch 7450, loss[loss=0.2275, simple_loss=0.3106, pruned_loss=0.07213, over 8601.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.331, pruned_loss=0.09673, over 1613839.30 frames. ], batch size: 31, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:49:25,938 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 04:49:28,832 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:49:29,058 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 04:49:50,666 INFO [train.py:901] (2/4) Epoch 7, batch 7500, loss[loss=0.239, simple_loss=0.3112, pruned_loss=0.08345, over 8030.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3294, pruned_loss=0.09579, over 1612568.07 frames. ], batch size: 22, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:50:17,129 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.816e+02 3.537e+02 4.737e+02 9.745e+02, threshold=7.074e+02, percent-clipped=6.0 +2023-02-06 04:50:25,618 INFO [train.py:901] (2/4) Epoch 7, batch 7550, loss[loss=0.3643, simple_loss=0.3938, pruned_loss=0.1674, over 8325.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3287, pruned_loss=0.09543, over 1611642.21 frames. ], batch size: 25, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:50:55,597 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8969, 2.9237, 3.3883, 2.0817, 1.5989, 3.4126, 0.5735, 2.1130], + device='cuda:2'), covar=tensor([0.2241, 0.2098, 0.0525, 0.3494, 0.5768, 0.0854, 0.5085, 0.2513], + device='cuda:2'), in_proj_covar=tensor([0.0145, 0.0148, 0.0088, 0.0199, 0.0237, 0.0091, 0.0150, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:50:58,642 INFO [train.py:901] (2/4) Epoch 7, batch 7600, loss[loss=0.3341, simple_loss=0.3849, pruned_loss=0.1416, over 7236.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3288, pruned_loss=0.09579, over 1610707.76 frames. ], batch size: 71, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:51:06,788 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56110.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:51:25,286 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.750e+02 3.495e+02 4.537e+02 9.121e+02, threshold=6.990e+02, percent-clipped=3.0 +2023-02-06 04:51:34,923 INFO [train.py:901] (2/4) Epoch 7, batch 7650, loss[loss=0.2128, simple_loss=0.2795, pruned_loss=0.07305, over 7523.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3271, pruned_loss=0.09489, over 1610284.96 frames. ], batch size: 18, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:51:44,426 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:52:03,621 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-06 04:52:08,618 INFO [train.py:901] (2/4) Epoch 7, batch 7700, loss[loss=0.2577, simple_loss=0.3337, pruned_loss=0.0908, over 7778.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3277, pruned_loss=0.09507, over 1610924.74 frames. ], batch size: 19, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:52:16,096 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56210.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:52:26,757 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56225.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:52:34,679 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.791e+02 3.394e+02 3.978e+02 9.035e+02, threshold=6.788e+02, percent-clipped=3.0 +2023-02-06 04:52:34,708 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 04:52:44,010 INFO [train.py:901] (2/4) Epoch 7, batch 7750, loss[loss=0.2887, simple_loss=0.3382, pruned_loss=0.1196, over 7271.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3286, pruned_loss=0.09504, over 1615016.60 frames. ], batch size: 16, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:53:11,820 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6643, 1.8676, 2.2866, 1.8920, 1.0880, 2.3880, 0.3987, 1.2662], + device='cuda:2'), covar=tensor([0.2917, 0.2285, 0.0514, 0.1825, 0.6073, 0.0549, 0.4482, 0.2030], + device='cuda:2'), in_proj_covar=tensor([0.0144, 0.0147, 0.0086, 0.0195, 0.0235, 0.0089, 0.0146, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 04:53:18,259 INFO [train.py:901] (2/4) Epoch 7, batch 7800, loss[loss=0.2347, simple_loss=0.2914, pruned_loss=0.08902, over 7691.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3291, pruned_loss=0.09543, over 1616469.64 frames. ], batch size: 18, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:53:35,707 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56325.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:53:36,356 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1296, 1.6244, 3.2280, 1.3283, 2.2144, 3.4100, 3.5399, 2.7542], + device='cuda:2'), covar=tensor([0.0977, 0.1508, 0.0415, 0.2212, 0.0927, 0.0358, 0.0428, 0.0786], + device='cuda:2'), in_proj_covar=tensor([0.0246, 0.0274, 0.0234, 0.0271, 0.0243, 0.0213, 0.0278, 0.0278], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 04:53:42,619 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.702e+02 3.307e+02 4.383e+02 8.490e+02, threshold=6.613e+02, percent-clipped=4.0 +2023-02-06 04:53:51,370 INFO [train.py:901] (2/4) Epoch 7, batch 7850, loss[loss=0.3167, simple_loss=0.3789, pruned_loss=0.1272, over 8340.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3305, pruned_loss=0.09652, over 1619589.86 frames. ], batch size: 26, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:54:24,870 INFO [train.py:901] (2/4) Epoch 7, batch 7900, loss[loss=0.2924, simple_loss=0.3607, pruned_loss=0.1121, over 8678.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3316, pruned_loss=0.09719, over 1616148.81 frames. ], batch size: 39, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:54:41,689 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0383, 2.3282, 1.8448, 2.8536, 1.4687, 1.5034, 2.0254, 2.4088], + device='cuda:2'), covar=tensor([0.0774, 0.0898, 0.1145, 0.0429, 0.1247, 0.1665, 0.1058, 0.0835], + device='cuda:2'), in_proj_covar=tensor([0.0253, 0.0239, 0.0277, 0.0221, 0.0232, 0.0269, 0.0272, 0.0234], + 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') +2023-02-06 04:54:49,422 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.758e+02 3.520e+02 4.424e+02 1.197e+03, threshold=7.039e+02, percent-clipped=9.0 +2023-02-06 04:54:58,057 INFO [train.py:901] (2/4) Epoch 7, batch 7950, loss[loss=0.2428, simple_loss=0.322, pruned_loss=0.08177, over 8289.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3309, pruned_loss=0.09733, over 1610598.23 frames. ], batch size: 23, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:55:19,787 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9567, 4.1648, 2.7599, 2.7921, 3.1287, 2.2506, 2.9195, 2.8871], + device='cuda:2'), covar=tensor([0.1526, 0.0242, 0.0776, 0.0762, 0.0631, 0.1152, 0.0961, 0.0885], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0236, 0.0312, 0.0300, 0.0307, 0.0317, 0.0336, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 04:55:19,811 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:55:31,643 INFO [train.py:901] (2/4) Epoch 7, batch 8000, loss[loss=0.3046, simple_loss=0.368, pruned_loss=0.1206, over 8594.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3317, pruned_loss=0.09769, over 1607973.59 frames. ], batch size: 39, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:55:36,302 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4382, 2.5026, 1.8287, 2.1243, 2.1328, 1.4621, 1.7850, 1.8653], + device='cuda:2'), covar=tensor([0.1148, 0.0333, 0.0817, 0.0452, 0.0492, 0.1139, 0.0803, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0239, 0.0312, 0.0301, 0.0308, 0.0317, 0.0337, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 04:55:36,307 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:55:36,855 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:55:56,103 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.809e+02 3.378e+02 4.457e+02 7.052e+02, threshold=6.755e+02, percent-clipped=1.0 +2023-02-06 04:55:59,642 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56541.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:56:03,027 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9420, 1.2514, 4.2324, 1.5390, 3.5640, 3.4200, 3.7164, 3.6082], + device='cuda:2'), covar=tensor([0.0638, 0.4521, 0.0530, 0.3355, 0.1362, 0.0872, 0.0657, 0.0746], + device='cuda:2'), in_proj_covar=tensor([0.0400, 0.0532, 0.0494, 0.0477, 0.0533, 0.0450, 0.0452, 0.0510], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 04:56:04,838 INFO [train.py:901] (2/4) Epoch 7, batch 8050, loss[loss=0.262, simple_loss=0.3237, pruned_loss=0.1001, over 7929.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3288, pruned_loss=0.09669, over 1599358.09 frames. ], batch size: 20, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:56:05,401 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 04:56:37,916 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 04:56:42,615 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:56:43,091 INFO [train.py:901] (2/4) Epoch 8, batch 0, loss[loss=0.2522, simple_loss=0.3286, pruned_loss=0.0879, over 8254.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3286, pruned_loss=0.0879, over 8254.00 frames. ], batch size: 24, lr: 9.92e-03, grad_scale: 8.0 +2023-02-06 04:56:43,091 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 04:56:54,074 INFO [train.py:935] (2/4) Epoch 8, validation: loss=0.205, simple_loss=0.3028, pruned_loss=0.05355, over 944034.00 frames. +2023-02-06 04:56:54,075 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 04:56:54,911 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3510, 4.3434, 3.9303, 1.7898, 3.8450, 3.8421, 3.9993, 3.5948], + device='cuda:2'), covar=tensor([0.0928, 0.0658, 0.1237, 0.5355, 0.0828, 0.1085, 0.1276, 0.0988], + device='cuda:2'), in_proj_covar=tensor([0.0419, 0.0327, 0.0354, 0.0451, 0.0343, 0.0322, 0.0334, 0.0287], + 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') +2023-02-06 04:57:08,603 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 04:57:10,757 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:57:22,341 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:57:28,988 INFO [train.py:901] (2/4) Epoch 8, batch 50, loss[loss=0.305, simple_loss=0.3555, pruned_loss=0.1273, over 8582.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.335, pruned_loss=0.09765, over 369156.74 frames. ], batch size: 31, lr: 9.92e-03, grad_scale: 8.0 +2023-02-06 04:57:31,765 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.831e+02 3.488e+02 4.265e+02 1.069e+03, threshold=6.975e+02, percent-clipped=2.0 +2023-02-06 04:57:39,011 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 04:57:43,126 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 04:58:03,656 INFO [train.py:901] (2/4) Epoch 8, batch 100, loss[loss=0.2643, simple_loss=0.3392, pruned_loss=0.09466, over 8133.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3309, pruned_loss=0.09658, over 647841.72 frames. ], batch size: 22, lr: 9.91e-03, grad_scale: 8.0 +2023-02-06 04:58:05,729 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 04:58:38,289 INFO [train.py:901] (2/4) Epoch 8, batch 150, loss[loss=0.2432, simple_loss=0.3156, pruned_loss=0.08542, over 8292.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3283, pruned_loss=0.0951, over 858079.46 frames. ], batch size: 23, lr: 9.91e-03, grad_scale: 8.0 +2023-02-06 04:58:40,461 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0897, 1.2358, 4.2744, 1.5109, 3.6993, 3.5862, 3.8516, 3.6680], + device='cuda:2'), covar=tensor([0.0451, 0.3967, 0.0472, 0.3203, 0.1134, 0.0813, 0.0510, 0.0694], + device='cuda:2'), in_proj_covar=tensor([0.0402, 0.0531, 0.0489, 0.0473, 0.0538, 0.0450, 0.0448, 0.0507], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 04:58:40,996 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.710e+02 3.372e+02 4.105e+02 8.611e+02, threshold=6.744e+02, percent-clipped=2.0 +2023-02-06 04:59:12,800 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 04:59:13,816 INFO [train.py:901] (2/4) Epoch 8, batch 200, loss[loss=0.2799, simple_loss=0.3453, pruned_loss=0.1073, over 8350.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3292, pruned_loss=0.0954, over 1030724.26 frames. ], batch size: 26, lr: 9.90e-03, grad_scale: 8.0 +2023-02-06 04:59:41,302 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56821.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:59:48,621 INFO [train.py:901] (2/4) Epoch 8, batch 250, loss[loss=0.2409, simple_loss=0.3279, pruned_loss=0.07691, over 8359.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3285, pruned_loss=0.0955, over 1158870.91 frames. ], batch size: 24, lr: 9.90e-03, grad_scale: 8.0 +2023-02-06 04:59:51,336 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.703e+02 3.318e+02 4.204e+02 1.022e+03, threshold=6.636e+02, percent-clipped=1.0 +2023-02-06 04:59:56,872 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 05:00:06,240 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 05:00:21,365 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56878.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:00:23,888 INFO [train.py:901] (2/4) Epoch 8, batch 300, loss[loss=0.2947, simple_loss=0.3486, pruned_loss=0.1204, over 8028.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3286, pruned_loss=0.0951, over 1260911.62 frames. ], batch size: 22, lr: 9.89e-03, grad_scale: 8.0 +2023-02-06 05:00:26,000 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56885.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 05:00:28,630 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4416, 1.2310, 1.4445, 1.1126, 0.8579, 1.2406, 1.2018, 1.0279], + device='cuda:2'), covar=tensor([0.0608, 0.1284, 0.1851, 0.1425, 0.0566, 0.1556, 0.0697, 0.0668], + device='cuda:2'), in_proj_covar=tensor([0.0112, 0.0162, 0.0201, 0.0164, 0.0112, 0.0169, 0.0124, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:00:38,046 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56903.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:00:54,958 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:00:58,776 INFO [train.py:901] (2/4) Epoch 8, batch 350, loss[loss=0.3056, simple_loss=0.3615, pruned_loss=0.1249, over 8362.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.328, pruned_loss=0.09433, over 1341574.94 frames. ], batch size: 24, lr: 9.89e-03, grad_scale: 8.0 +2023-02-06 05:01:01,452 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.612e+02 3.168e+02 3.951e+02 1.059e+03, threshold=6.336e+02, percent-clipped=3.0 +2023-02-06 05:01:02,321 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0237, 1.6047, 1.6205, 1.4015, 1.2031, 1.4802, 1.6996, 1.7640], + device='cuda:2'), covar=tensor([0.0548, 0.1172, 0.1751, 0.1331, 0.0590, 0.1466, 0.0717, 0.0516], + device='cuda:2'), in_proj_covar=tensor([0.0111, 0.0162, 0.0200, 0.0163, 0.0112, 0.0168, 0.0123, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:01:33,167 INFO [train.py:901] (2/4) Epoch 8, batch 400, loss[loss=0.2552, simple_loss=0.3108, pruned_loss=0.09984, over 7257.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3293, pruned_loss=0.09566, over 1404709.30 frames. ], batch size: 16, lr: 9.89e-03, grad_scale: 8.0 +2023-02-06 05:01:45,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57000.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 05:01:52,587 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57011.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:01:53,837 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57013.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:02:07,049 INFO [train.py:901] (2/4) Epoch 8, batch 450, loss[loss=0.273, simple_loss=0.3507, pruned_loss=0.09769, over 8641.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3297, pruned_loss=0.09552, over 1451745.84 frames. ], batch size: 49, lr: 9.88e-03, grad_scale: 8.0 +2023-02-06 05:02:10,304 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.769e+02 3.532e+02 4.551e+02 9.004e+02, threshold=7.064e+02, percent-clipped=7.0 +2023-02-06 05:02:21,405 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 +2023-02-06 05:02:33,934 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0005, 1.3322, 4.2097, 1.5721, 3.5234, 3.3768, 3.6425, 3.5994], + device='cuda:2'), covar=tensor([0.0604, 0.4491, 0.0517, 0.3138, 0.1413, 0.0906, 0.0698, 0.0767], + device='cuda:2'), in_proj_covar=tensor([0.0403, 0.0531, 0.0492, 0.0477, 0.0544, 0.0456, 0.0454, 0.0508], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 05:02:41,866 INFO [train.py:901] (2/4) Epoch 8, batch 500, loss[loss=0.2224, simple_loss=0.2979, pruned_loss=0.07348, over 8038.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3308, pruned_loss=0.09603, over 1490591.06 frames. ], batch size: 22, lr: 9.88e-03, grad_scale: 8.0 +2023-02-06 05:03:15,889 INFO [train.py:901] (2/4) Epoch 8, batch 550, loss[loss=0.2636, simple_loss=0.323, pruned_loss=0.1021, over 7185.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3281, pruned_loss=0.09464, over 1517108.03 frames. ], batch size: 16, lr: 9.87e-03, grad_scale: 8.0 +2023-02-06 05:03:18,519 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.761e+02 3.532e+02 4.192e+02 1.400e+03, threshold=7.064e+02, percent-clipped=6.0 +2023-02-06 05:03:39,525 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:03:40,892 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5473, 5.7055, 4.9631, 2.3152, 5.0248, 5.3777, 5.2764, 4.9052], + device='cuda:2'), covar=tensor([0.0633, 0.0412, 0.0915, 0.4298, 0.0666, 0.0572, 0.0990, 0.0563], + device='cuda:2'), in_proj_covar=tensor([0.0418, 0.0324, 0.0354, 0.0441, 0.0344, 0.0322, 0.0332, 0.0286], + 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') +2023-02-06 05:03:50,883 INFO [train.py:901] (2/4) Epoch 8, batch 600, loss[loss=0.255, simple_loss=0.3142, pruned_loss=0.09786, over 7823.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3259, pruned_loss=0.09296, over 1536540.69 frames. ], batch size: 20, lr: 9.87e-03, grad_scale: 8.0 +2023-02-06 05:04:03,002 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 05:04:06,350 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57204.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:13,912 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57215.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:25,748 INFO [train.py:901] (2/4) Epoch 8, batch 650, loss[loss=0.231, simple_loss=0.3036, pruned_loss=0.07922, over 7927.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3255, pruned_loss=0.09294, over 1550880.03 frames. ], batch size: 20, lr: 9.86e-03, grad_scale: 8.0 +2023-02-06 05:04:28,378 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.474e+02 3.242e+02 4.284e+02 1.059e+03, threshold=6.484e+02, percent-clipped=6.0 +2023-02-06 05:04:41,995 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57256.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 05:04:46,653 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:50,069 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8440, 3.8253, 2.3600, 2.2778, 2.9921, 1.6869, 2.4475, 2.6079], + device='cuda:2'), covar=tensor([0.1407, 0.0243, 0.0781, 0.0778, 0.0531, 0.1173, 0.0990, 0.0977], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0232, 0.0309, 0.0298, 0.0306, 0.0312, 0.0335, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 05:04:51,987 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57270.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:59,517 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57280.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:05:00,217 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57281.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 05:05:00,683 INFO [train.py:901] (2/4) Epoch 8, batch 700, loss[loss=0.2561, simple_loss=0.3267, pruned_loss=0.09282, over 8135.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.326, pruned_loss=0.09233, over 1570712.87 frames. ], batch size: 22, lr: 9.86e-03, grad_scale: 8.0 +2023-02-06 05:05:34,918 INFO [train.py:901] (2/4) Epoch 8, batch 750, loss[loss=0.2194, simple_loss=0.2835, pruned_loss=0.07766, over 7695.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3263, pruned_loss=0.09341, over 1580057.12 frames. ], batch size: 18, lr: 9.86e-03, grad_scale: 8.0 +2023-02-06 05:05:38,348 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.846e+02 3.371e+02 4.091e+02 7.333e+02, threshold=6.742e+02, percent-clipped=1.0 +2023-02-06 05:05:49,677 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 05:05:51,156 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:05:52,579 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57357.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:05:57,731 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 05:06:09,705 INFO [train.py:901] (2/4) Epoch 8, batch 800, loss[loss=0.2807, simple_loss=0.3512, pruned_loss=0.1051, over 8512.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.327, pruned_loss=0.09377, over 1590319.22 frames. ], batch size: 28, lr: 9.85e-03, grad_scale: 16.0 +2023-02-06 05:06:11,944 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:06:33,596 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57416.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:06:44,183 INFO [train.py:901] (2/4) Epoch 8, batch 850, loss[loss=0.2376, simple_loss=0.3226, pruned_loss=0.0763, over 8500.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3269, pruned_loss=0.09324, over 1598341.29 frames. ], batch size: 26, lr: 9.85e-03, grad_scale: 16.0 +2023-02-06 05:06:46,892 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.664e+02 3.287e+02 4.255e+02 8.769e+02, threshold=6.575e+02, percent-clipped=4.0 +2023-02-06 05:07:11,115 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57470.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:12,475 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:17,825 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57480.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:19,065 INFO [train.py:901] (2/4) Epoch 8, batch 900, loss[loss=0.224, simple_loss=0.3031, pruned_loss=0.07248, over 8577.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3264, pruned_loss=0.09263, over 1605720.65 frames. ], batch size: 31, lr: 9.84e-03, grad_scale: 16.0 +2023-02-06 05:07:40,918 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:52,601 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8253, 2.0895, 2.3681, 1.9210, 1.1924, 2.3618, 0.5074, 1.4687], + device='cuda:2'), covar=tensor([0.3059, 0.1603, 0.0619, 0.1896, 0.4932, 0.0722, 0.3798, 0.2289], + device='cuda:2'), in_proj_covar=tensor([0.0144, 0.0144, 0.0085, 0.0191, 0.0233, 0.0090, 0.0145, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 05:07:53,660 INFO [train.py:901] (2/4) Epoch 8, batch 950, loss[loss=0.285, simple_loss=0.3473, pruned_loss=0.1113, over 8022.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.328, pruned_loss=0.09413, over 1608993.26 frames. ], batch size: 22, lr: 9.84e-03, grad_scale: 16.0 +2023-02-06 05:07:56,420 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.713e+02 3.197e+02 4.416e+02 7.629e+02, threshold=6.394e+02, percent-clipped=6.0 +2023-02-06 05:07:56,676 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57536.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:04,585 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:13,024 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57559.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:14,313 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 05:08:14,508 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57561.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:29,125 INFO [train.py:901] (2/4) Epoch 8, batch 1000, loss[loss=0.2339, simple_loss=0.2978, pruned_loss=0.08502, over 7434.00 frames. ], tot_loss[loss=0.256, simple_loss=0.326, pruned_loss=0.09297, over 1610087.72 frames. ], batch size: 17, lr: 9.83e-03, grad_scale: 16.0 +2023-02-06 05:08:45,896 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:47,899 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 05:09:00,600 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 05:09:03,908 INFO [train.py:901] (2/4) Epoch 8, batch 1050, loss[loss=0.2401, simple_loss=0.3144, pruned_loss=0.08285, over 8193.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3276, pruned_loss=0.09368, over 1618016.59 frames. ], batch size: 23, lr: 9.83e-03, grad_scale: 16.0 +2023-02-06 05:09:06,640 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.778e+02 2.733e+02 3.382e+02 4.210e+02 1.523e+03, threshold=6.765e+02, percent-clipped=11.0 +2023-02-06 05:09:10,049 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57641.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:24,434 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57663.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:26,526 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:31,690 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:37,588 INFO [train.py:901] (2/4) Epoch 8, batch 1100, loss[loss=0.3033, simple_loss=0.3592, pruned_loss=0.1237, over 7053.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3288, pruned_loss=0.09468, over 1618003.03 frames. ], batch size: 72, lr: 9.83e-03, grad_scale: 16.0 +2023-02-06 05:10:05,401 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57721.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:08,893 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:10,257 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57728.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:10,699 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 05:10:12,767 INFO [train.py:901] (2/4) Epoch 8, batch 1150, loss[loss=0.2503, simple_loss=0.3077, pruned_loss=0.09644, over 7274.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3285, pruned_loss=0.09463, over 1615421.55 frames. ], batch size: 16, lr: 9.82e-03, grad_scale: 16.0 +2023-02-06 05:10:15,548 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.752e+02 3.349e+02 4.211e+02 1.172e+03, threshold=6.698e+02, percent-clipped=4.0 +2023-02-06 05:10:26,719 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:26,782 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:28,140 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:32,805 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:48,142 INFO [train.py:901] (2/4) Epoch 8, batch 1200, loss[loss=0.2146, simple_loss=0.2965, pruned_loss=0.06637, over 7644.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3265, pruned_loss=0.09338, over 1612820.43 frames. ], batch size: 19, lr: 9.82e-03, grad_scale: 16.0 +2023-02-06 05:11:10,444 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7381, 5.6697, 5.0097, 2.3553, 5.1543, 5.5155, 5.3585, 5.2605], + device='cuda:2'), covar=tensor([0.0439, 0.0343, 0.0690, 0.4445, 0.0597, 0.0533, 0.0776, 0.0467], + device='cuda:2'), in_proj_covar=tensor([0.0417, 0.0325, 0.0360, 0.0446, 0.0350, 0.0325, 0.0337, 0.0284], + 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') +2023-02-06 05:11:17,881 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:11:23,243 INFO [train.py:901] (2/4) Epoch 8, batch 1250, loss[loss=0.2476, simple_loss=0.3222, pruned_loss=0.08648, over 8641.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3269, pruned_loss=0.09361, over 1616308.35 frames. ], batch size: 34, lr: 9.81e-03, grad_scale: 16.0 +2023-02-06 05:11:25,902 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.817e+02 3.577e+02 4.191e+02 8.690e+02, threshold=7.155e+02, percent-clipped=5.0 +2023-02-06 05:11:41,409 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:11:53,669 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:11:58,364 INFO [train.py:901] (2/4) Epoch 8, batch 1300, loss[loss=0.2519, simple_loss=0.3221, pruned_loss=0.09089, over 7804.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3269, pruned_loss=0.09346, over 1615948.52 frames. ], batch size: 20, lr: 9.81e-03, grad_scale: 16.0 +2023-02-06 05:12:14,828 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 05:12:24,233 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57919.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:32,230 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57930.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:33,387 INFO [train.py:901] (2/4) Epoch 8, batch 1350, loss[loss=0.2511, simple_loss=0.333, pruned_loss=0.08462, over 8730.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3267, pruned_loss=0.09343, over 1617042.00 frames. ], batch size: 30, lr: 9.80e-03, grad_scale: 16.0 +2023-02-06 05:12:36,120 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.787e+02 3.281e+02 4.089e+02 1.129e+03, threshold=6.562e+02, percent-clipped=4.0 +2023-02-06 05:12:38,227 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3999, 1.4479, 1.6540, 1.3872, 1.2778, 1.5076, 1.7827, 1.6615], + device='cuda:2'), covar=tensor([0.0485, 0.1250, 0.1715, 0.1387, 0.0538, 0.1485, 0.0631, 0.0587], + device='cuda:2'), in_proj_covar=tensor([0.0109, 0.0161, 0.0199, 0.0164, 0.0112, 0.0169, 0.0123, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:12:38,234 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:41,652 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:49,717 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57955.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:01,726 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57972.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:05,180 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57977.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:08,418 INFO [train.py:901] (2/4) Epoch 8, batch 1400, loss[loss=0.2261, simple_loss=0.2924, pruned_loss=0.07986, over 7439.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3275, pruned_loss=0.09469, over 1616690.80 frames. ], batch size: 17, lr: 9.80e-03, grad_scale: 16.0 +2023-02-06 05:13:23,095 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58002.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:41,345 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 05:13:42,226 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8499, 2.7955, 3.2120, 2.2322, 1.7086, 3.3185, 0.5616, 1.9467], + device='cuda:2'), covar=tensor([0.2530, 0.1458, 0.0453, 0.2559, 0.4311, 0.0568, 0.4194, 0.2218], + device='cuda:2'), in_proj_covar=tensor([0.0142, 0.0143, 0.0083, 0.0189, 0.0230, 0.0089, 0.0143, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 05:13:43,356 INFO [train.py:901] (2/4) Epoch 8, batch 1450, loss[loss=0.2459, simple_loss=0.303, pruned_loss=0.09441, over 7705.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3282, pruned_loss=0.09502, over 1615926.84 frames. ], batch size: 18, lr: 9.80e-03, grad_scale: 16.0 +2023-02-06 05:13:46,037 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.540e+02 2.668e+02 3.298e+02 4.223e+02 1.032e+03, threshold=6.596e+02, percent-clipped=5.0 +2023-02-06 05:14:18,670 INFO [train.py:901] (2/4) Epoch 8, batch 1500, loss[loss=0.1875, simple_loss=0.2633, pruned_loss=0.05585, over 7694.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3272, pruned_loss=0.09409, over 1614425.41 frames. ], batch size: 18, lr: 9.79e-03, grad_scale: 16.0 +2023-02-06 05:14:28,078 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:14:39,241 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5258, 1.5740, 1.6807, 1.4802, 0.9859, 1.7313, 0.0962, 0.9817], + device='cuda:2'), covar=tensor([0.3062, 0.2073, 0.0735, 0.1292, 0.5457, 0.0817, 0.3511, 0.1981], + device='cuda:2'), in_proj_covar=tensor([0.0145, 0.0145, 0.0085, 0.0189, 0.0233, 0.0090, 0.0145, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 05:14:49,420 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4512, 1.9049, 2.1054, 1.1670, 2.2570, 1.3688, 0.6099, 1.6643], + device='cuda:2'), covar=tensor([0.0394, 0.0205, 0.0142, 0.0297, 0.0221, 0.0512, 0.0492, 0.0189], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0275, 0.0224, 0.0331, 0.0270, 0.0423, 0.0328, 0.0307], + device='cuda:2'), out_proj_covar=tensor([1.1037e-04, 8.3422e-05, 6.7244e-05, 9.9829e-05, 8.3318e-05, 1.3999e-04, + 1.0181e-04, 9.4163e-05], device='cuda:2') +2023-02-06 05:14:52,821 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58131.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:14:53,264 INFO [train.py:901] (2/4) Epoch 8, batch 1550, loss[loss=0.2104, simple_loss=0.2895, pruned_loss=0.06562, over 7787.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3265, pruned_loss=0.09339, over 1614897.28 frames. ], batch size: 19, lr: 9.79e-03, grad_scale: 16.0 +2023-02-06 05:14:54,854 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3749, 1.8527, 2.0249, 1.1769, 2.1936, 1.3925, 0.5486, 1.6169], + device='cuda:2'), covar=tensor([0.0381, 0.0168, 0.0122, 0.0257, 0.0190, 0.0464, 0.0438, 0.0160], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0275, 0.0225, 0.0332, 0.0271, 0.0423, 0.0329, 0.0308], + device='cuda:2'), out_proj_covar=tensor([1.1069e-04, 8.3546e-05, 6.7375e-05, 1.0023e-04, 8.3492e-05, 1.4010e-04, + 1.0201e-04, 9.4436e-05], device='cuda:2') +2023-02-06 05:14:56,007 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.601e+02 3.218e+02 3.979e+02 6.246e+02, threshold=6.435e+02, percent-clipped=0.0 +2023-02-06 05:15:08,670 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5893, 5.7902, 4.8809, 2.3740, 5.0495, 5.2848, 5.3444, 4.9812], + device='cuda:2'), covar=tensor([0.0706, 0.0442, 0.0993, 0.4478, 0.0640, 0.0638, 0.1177, 0.0597], + device='cuda:2'), in_proj_covar=tensor([0.0421, 0.0328, 0.0357, 0.0446, 0.0349, 0.0326, 0.0335, 0.0286], + 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') +2023-02-06 05:15:10,025 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58156.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:12,045 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58159.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:14,755 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:27,768 INFO [train.py:901] (2/4) Epoch 8, batch 1600, loss[loss=0.2544, simple_loss=0.3383, pruned_loss=0.08524, over 8481.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3273, pruned_loss=0.09419, over 1617621.25 frames. ], batch size: 28, lr: 9.78e-03, grad_scale: 16.0 +2023-02-06 05:15:37,318 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58195.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:47,377 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58210.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:54,047 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58220.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:16:00,075 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58228.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:16:02,613 INFO [train.py:901] (2/4) Epoch 8, batch 1650, loss[loss=0.1928, simple_loss=0.2651, pruned_loss=0.06029, over 7449.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3272, pruned_loss=0.09379, over 1615428.07 frames. ], batch size: 17, lr: 9.78e-03, grad_scale: 16.0 +2023-02-06 05:16:05,269 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.785e+02 3.241e+02 4.331e+02 1.468e+03, threshold=6.482e+02, percent-clipped=4.0 +2023-02-06 05:16:16,839 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:16:37,568 INFO [train.py:901] (2/4) Epoch 8, batch 1700, loss[loss=0.246, simple_loss=0.3114, pruned_loss=0.09025, over 7813.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3256, pruned_loss=0.09313, over 1614446.76 frames. ], batch size: 20, lr: 9.78e-03, grad_scale: 16.0 +2023-02-06 05:17:00,581 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2765, 4.2106, 3.8170, 1.7505, 3.8082, 3.6877, 3.9389, 3.4913], + device='cuda:2'), covar=tensor([0.0810, 0.0651, 0.1037, 0.4954, 0.0828, 0.0926, 0.1201, 0.0895], + device='cuda:2'), in_proj_covar=tensor([0.0419, 0.0324, 0.0355, 0.0444, 0.0349, 0.0325, 0.0335, 0.0284], + 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') +2023-02-06 05:17:11,716 INFO [train.py:901] (2/4) Epoch 8, batch 1750, loss[loss=0.3043, simple_loss=0.3622, pruned_loss=0.1232, over 8595.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3268, pruned_loss=0.09409, over 1615040.92 frames. ], batch size: 31, lr: 9.77e-03, grad_scale: 16.0 +2023-02-06 05:17:15,046 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.752e+02 3.204e+02 3.949e+02 8.384e+02, threshold=6.409e+02, percent-clipped=4.0 +2023-02-06 05:17:41,591 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 05:17:43,448 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 05:17:45,739 INFO [train.py:901] (2/4) Epoch 8, batch 1800, loss[loss=0.2588, simple_loss=0.3327, pruned_loss=0.09248, over 8439.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3269, pruned_loss=0.09405, over 1619053.19 frames. ], batch size: 27, lr: 9.77e-03, grad_scale: 16.0 +2023-02-06 05:18:06,233 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 05:18:21,313 INFO [train.py:901] (2/4) Epoch 8, batch 1850, loss[loss=0.2659, simple_loss=0.34, pruned_loss=0.09592, over 8463.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3261, pruned_loss=0.09349, over 1618291.76 frames. ], batch size: 27, lr: 9.76e-03, grad_scale: 16.0 +2023-02-06 05:18:24,003 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.956e+02 3.603e+02 4.636e+02 8.044e+02, threshold=7.207e+02, percent-clipped=5.0 +2023-02-06 05:18:45,040 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58466.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:18:55,852 INFO [train.py:901] (2/4) Epoch 8, batch 1900, loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09061, over 8506.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3256, pruned_loss=0.09291, over 1619720.76 frames. ], batch size: 26, lr: 9.76e-03, grad_scale: 16.0 +2023-02-06 05:19:02,000 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58491.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:19:10,023 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58503.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:19:12,309 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2023-02-06 05:19:12,701 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:19:19,278 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 05:19:30,166 INFO [train.py:901] (2/4) Epoch 8, batch 1950, loss[loss=0.2427, simple_loss=0.3145, pruned_loss=0.08547, over 8241.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3272, pruned_loss=0.09383, over 1621538.73 frames. ], batch size: 24, lr: 9.75e-03, grad_scale: 16.0 +2023-02-06 05:19:30,816 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 05:19:32,607 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-06 05:19:32,812 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.699e+02 3.417e+02 4.103e+02 8.210e+02, threshold=6.834e+02, percent-clipped=5.0 +2023-02-06 05:19:50,861 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 05:20:04,896 INFO [train.py:901] (2/4) Epoch 8, batch 2000, loss[loss=0.2233, simple_loss=0.3069, pruned_loss=0.0699, over 7972.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3278, pruned_loss=0.09456, over 1615137.25 frames. ], batch size: 21, lr: 9.75e-03, grad_scale: 8.0 +2023-02-06 05:20:29,412 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58618.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:20:31,392 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:20:36,253 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 05:20:39,666 INFO [train.py:901] (2/4) Epoch 8, batch 2050, loss[loss=0.2469, simple_loss=0.3155, pruned_loss=0.08919, over 8085.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3266, pruned_loss=0.09391, over 1611988.50 frames. ], batch size: 21, lr: 9.75e-03, grad_scale: 8.0 +2023-02-06 05:20:42,941 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.785e+02 3.396e+02 4.687e+02 1.585e+03, threshold=6.792e+02, percent-clipped=4.0 +2023-02-06 05:21:00,881 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1127, 1.3419, 4.2168, 1.5420, 3.6471, 3.4612, 3.8030, 3.6520], + device='cuda:2'), covar=tensor([0.0448, 0.4162, 0.0476, 0.2961, 0.1139, 0.0767, 0.0503, 0.0626], + device='cuda:2'), in_proj_covar=tensor([0.0402, 0.0524, 0.0500, 0.0472, 0.0535, 0.0452, 0.0451, 0.0506], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 05:21:13,673 INFO [train.py:901] (2/4) Epoch 8, batch 2100, loss[loss=0.2268, simple_loss=0.2974, pruned_loss=0.07811, over 7791.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.328, pruned_loss=0.09456, over 1615989.43 frames. ], batch size: 19, lr: 9.74e-03, grad_scale: 8.0 +2023-02-06 05:21:25,812 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58699.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:21:47,361 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5817, 2.0043, 2.1869, 1.0377, 2.3456, 1.4436, 0.6574, 1.6775], + device='cuda:2'), covar=tensor([0.0279, 0.0132, 0.0093, 0.0269, 0.0153, 0.0420, 0.0376, 0.0162], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0272, 0.0223, 0.0330, 0.0268, 0.0417, 0.0321, 0.0305], + device='cuda:2'), out_proj_covar=tensor([1.0852e-04, 8.2677e-05, 6.6972e-05, 9.9202e-05, 8.2319e-05, 1.3771e-04, + 9.8989e-05, 9.3451e-05], device='cuda:2') +2023-02-06 05:21:47,831 INFO [train.py:901] (2/4) Epoch 8, batch 2150, loss[loss=0.2758, simple_loss=0.3401, pruned_loss=0.1057, over 7973.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3289, pruned_loss=0.09564, over 1614943.92 frames. ], batch size: 21, lr: 9.74e-03, grad_scale: 8.0 +2023-02-06 05:21:51,082 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.839e+02 2.818e+02 3.372e+02 4.104e+02 8.704e+02, threshold=6.743e+02, percent-clipped=2.0 +2023-02-06 05:22:23,672 INFO [train.py:901] (2/4) Epoch 8, batch 2200, loss[loss=0.2966, simple_loss=0.3437, pruned_loss=0.1247, over 7527.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3276, pruned_loss=0.09432, over 1617375.48 frames. ], batch size: 73, lr: 9.73e-03, grad_scale: 8.0 +2023-02-06 05:22:31,533 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58793.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:22:58,799 INFO [train.py:901] (2/4) Epoch 8, batch 2250, loss[loss=0.2607, simple_loss=0.3295, pruned_loss=0.09594, over 7797.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3268, pruned_loss=0.09354, over 1621272.02 frames. ], batch size: 20, lr: 9.73e-03, grad_scale: 8.0 +2023-02-06 05:23:02,318 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.600e+02 3.138e+02 4.259e+02 8.800e+02, threshold=6.276e+02, percent-clipped=5.0 +2023-02-06 05:23:11,292 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0817, 1.2815, 4.2604, 1.6396, 3.6962, 3.5146, 3.7507, 3.6497], + device='cuda:2'), covar=tensor([0.0486, 0.4214, 0.0465, 0.3038, 0.1205, 0.0914, 0.0540, 0.0627], + device='cuda:2'), in_proj_covar=tensor([0.0403, 0.0528, 0.0504, 0.0473, 0.0542, 0.0456, 0.0457, 0.0510], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 05:23:21,594 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7572, 1.4151, 3.9055, 1.3698, 3.4199, 3.2403, 3.5066, 3.3648], + device='cuda:2'), covar=tensor([0.0546, 0.3869, 0.0571, 0.3028, 0.1318, 0.0810, 0.0598, 0.0730], + device='cuda:2'), in_proj_covar=tensor([0.0402, 0.0529, 0.0504, 0.0473, 0.0542, 0.0456, 0.0455, 0.0510], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 05:23:29,270 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:31,269 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:34,473 INFO [train.py:901] (2/4) Epoch 8, batch 2300, loss[loss=0.2417, simple_loss=0.3229, pruned_loss=0.08026, over 8021.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.327, pruned_loss=0.09381, over 1621757.90 frames. ], batch size: 22, lr: 9.73e-03, grad_scale: 8.0 +2023-02-06 05:23:46,517 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:48,571 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:24:03,548 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58924.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:24:09,484 INFO [train.py:901] (2/4) Epoch 8, batch 2350, loss[loss=0.2185, simple_loss=0.2915, pruned_loss=0.0727, over 7538.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3277, pruned_loss=0.09415, over 1619267.17 frames. ], batch size: 18, lr: 9.72e-03, grad_scale: 8.0 +2023-02-06 05:24:12,939 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.611e+02 3.221e+02 3.780e+02 8.999e+02, threshold=6.441e+02, percent-clipped=2.0 +2023-02-06 05:24:44,048 INFO [train.py:901] (2/4) Epoch 8, batch 2400, loss[loss=0.2807, simple_loss=0.3461, pruned_loss=0.1077, over 8510.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3276, pruned_loss=0.09426, over 1618822.57 frames. ], batch size: 26, lr: 9.72e-03, grad_scale: 8.0 +2023-02-06 05:24:44,255 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3436, 1.7546, 1.8186, 1.1404, 1.9829, 1.3411, 0.4643, 1.6631], + device='cuda:2'), covar=tensor([0.0287, 0.0150, 0.0140, 0.0210, 0.0189, 0.0431, 0.0403, 0.0136], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0273, 0.0227, 0.0332, 0.0271, 0.0422, 0.0325, 0.0306], + device='cuda:2'), out_proj_covar=tensor([1.0931e-04, 8.2739e-05, 6.8222e-05, 9.9816e-05, 8.3264e-05, 1.3915e-04, + 1.0044e-04, 9.3606e-05], device='cuda:2') +2023-02-06 05:25:18,654 INFO [train.py:901] (2/4) Epoch 8, batch 2450, loss[loss=0.2802, simple_loss=0.3651, pruned_loss=0.0977, over 8250.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3272, pruned_loss=0.0939, over 1615341.86 frames. ], batch size: 24, lr: 9.71e-03, grad_scale: 8.0 +2023-02-06 05:25:21,882 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 3.001e+02 3.706e+02 4.542e+02 9.599e+02, threshold=7.413e+02, percent-clipped=3.0 +2023-02-06 05:25:25,464 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6090, 4.6373, 4.1052, 1.9941, 4.0978, 4.0195, 4.2205, 3.8158], + device='cuda:2'), covar=tensor([0.0657, 0.0547, 0.0907, 0.4346, 0.0784, 0.0879, 0.1120, 0.0747], + device='cuda:2'), in_proj_covar=tensor([0.0423, 0.0327, 0.0351, 0.0446, 0.0346, 0.0325, 0.0337, 0.0285], + 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') +2023-02-06 05:25:26,106 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:25:52,506 INFO [train.py:901] (2/4) Epoch 8, batch 2500, loss[loss=0.2209, simple_loss=0.289, pruned_loss=0.07644, over 8086.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.327, pruned_loss=0.09393, over 1617177.71 frames. ], batch size: 21, lr: 9.71e-03, grad_scale: 8.0 +2023-02-06 05:26:27,546 INFO [train.py:901] (2/4) Epoch 8, batch 2550, loss[loss=0.266, simple_loss=0.3374, pruned_loss=0.09733, over 8569.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3265, pruned_loss=0.0939, over 1617288.29 frames. ], batch size: 49, lr: 9.71e-03, grad_scale: 8.0 +2023-02-06 05:26:29,721 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2433, 1.5981, 3.4716, 1.5173, 2.3090, 3.9161, 3.8675, 3.3159], + device='cuda:2'), covar=tensor([0.0917, 0.1466, 0.0342, 0.1850, 0.0949, 0.0246, 0.0403, 0.0559], + device='cuda:2'), in_proj_covar=tensor([0.0249, 0.0285, 0.0241, 0.0278, 0.0247, 0.0225, 0.0294, 0.0285], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 05:26:30,875 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.676e+02 3.180e+02 4.175e+02 9.807e+02, threshold=6.360e+02, percent-clipped=4.0 +2023-02-06 05:26:30,962 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:26:42,639 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3714, 2.0266, 3.1060, 2.4479, 2.7217, 2.0659, 1.6219, 1.3406], + device='cuda:2'), covar=tensor([0.2855, 0.3134, 0.0853, 0.2076, 0.1680, 0.1717, 0.1411, 0.3604], + device='cuda:2'), in_proj_covar=tensor([0.0844, 0.0793, 0.0683, 0.0787, 0.0882, 0.0735, 0.0671, 0.0719], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 05:26:42,857 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 05:26:45,981 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59158.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:27:03,040 INFO [train.py:901] (2/4) Epoch 8, batch 2600, loss[loss=0.2097, simple_loss=0.2793, pruned_loss=0.07003, over 7940.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3261, pruned_loss=0.09348, over 1616391.80 frames. ], batch size: 20, lr: 9.70e-03, grad_scale: 8.0 +2023-02-06 05:27:03,175 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59182.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:27:38,229 INFO [train.py:901] (2/4) Epoch 8, batch 2650, loss[loss=0.2105, simple_loss=0.2958, pruned_loss=0.06264, over 8133.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3259, pruned_loss=0.09323, over 1614448.97 frames. ], batch size: 22, lr: 9.70e-03, grad_scale: 8.0 +2023-02-06 05:27:41,652 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.757e+02 3.213e+02 4.207e+02 1.360e+03, threshold=6.426e+02, percent-clipped=6.0 +2023-02-06 05:27:42,696 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-06 05:27:48,555 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6818, 5.7761, 5.0335, 2.3751, 5.0859, 5.4822, 5.2712, 4.9753], + device='cuda:2'), covar=tensor([0.0569, 0.0375, 0.0823, 0.4486, 0.0572, 0.0461, 0.1020, 0.0586], + device='cuda:2'), in_proj_covar=tensor([0.0423, 0.0327, 0.0351, 0.0447, 0.0346, 0.0325, 0.0338, 0.0287], + 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') +2023-02-06 05:27:51,996 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:28:03,087 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:28:06,560 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8541, 2.1479, 1.7028, 2.4672, 1.2899, 1.5166, 1.7916, 2.2191], + device='cuda:2'), covar=tensor([0.0800, 0.0865, 0.1088, 0.0580, 0.1364, 0.1481, 0.1044, 0.0719], + device='cuda:2'), in_proj_covar=tensor([0.0249, 0.0229, 0.0265, 0.0215, 0.0232, 0.0267, 0.0268, 0.0237], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 05:28:12,227 INFO [train.py:901] (2/4) Epoch 8, batch 2700, loss[loss=0.3882, simple_loss=0.4201, pruned_loss=0.1782, over 8334.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3273, pruned_loss=0.09371, over 1616210.57 frames. ], batch size: 25, lr: 9.69e-03, grad_scale: 8.0 +2023-02-06 05:28:45,925 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 05:28:46,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1417, 1.7496, 2.6612, 2.1071, 2.3076, 1.9115, 1.4949, 1.0043], + device='cuda:2'), covar=tensor([0.2841, 0.3029, 0.0785, 0.1839, 0.1403, 0.1738, 0.1494, 0.3141], + device='cuda:2'), in_proj_covar=tensor([0.0842, 0.0796, 0.0686, 0.0790, 0.0887, 0.0735, 0.0671, 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') +2023-02-06 05:28:46,772 INFO [train.py:901] (2/4) Epoch 8, batch 2750, loss[loss=0.2648, simple_loss=0.3467, pruned_loss=0.09147, over 8360.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3277, pruned_loss=0.09391, over 1620924.09 frames. ], batch size: 24, lr: 9.69e-03, grad_scale: 8.0 +2023-02-06 05:28:50,099 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 2.846e+02 3.367e+02 4.274e+02 9.837e+02, threshold=6.735e+02, percent-clipped=6.0 +2023-02-06 05:29:07,219 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4939, 1.5553, 1.6718, 1.4674, 0.9788, 1.7982, 0.1093, 1.0810], + device='cuda:2'), covar=tensor([0.2467, 0.1618, 0.0738, 0.1788, 0.4939, 0.0598, 0.3710, 0.1991], + device='cuda:2'), in_proj_covar=tensor([0.0145, 0.0143, 0.0082, 0.0195, 0.0228, 0.0088, 0.0150, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 05:29:22,656 INFO [train.py:901] (2/4) Epoch 8, batch 2800, loss[loss=0.2819, simple_loss=0.3322, pruned_loss=0.1158, over 7544.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3276, pruned_loss=0.09345, over 1622900.83 frames. ], batch size: 18, lr: 9.69e-03, grad_scale: 8.0 +2023-02-06 05:29:23,481 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:29:44,998 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:29:56,665 INFO [train.py:901] (2/4) Epoch 8, batch 2850, loss[loss=0.2546, simple_loss=0.3259, pruned_loss=0.09166, over 8518.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3267, pruned_loss=0.09285, over 1621002.40 frames. ], batch size: 28, lr: 9.68e-03, grad_scale: 8.0 +2023-02-06 05:30:00,148 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.577e+02 2.974e+02 3.773e+02 5.956e+02, threshold=5.948e+02, percent-clipped=0.0 +2023-02-06 05:30:01,806 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:30:32,530 INFO [train.py:901] (2/4) Epoch 8, batch 2900, loss[loss=0.2058, simple_loss=0.2749, pruned_loss=0.06835, over 7718.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3272, pruned_loss=0.09342, over 1617107.49 frames. ], batch size: 18, lr: 9.68e-03, grad_scale: 8.0 +2023-02-06 05:30:51,227 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59508.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:30:59,189 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:31:03,077 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59526.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:31:04,356 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 05:31:06,043 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.85 vs. limit=5.0 +2023-02-06 05:31:07,544 INFO [train.py:901] (2/4) Epoch 8, batch 2950, loss[loss=0.2574, simple_loss=0.3376, pruned_loss=0.08857, over 8500.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3265, pruned_loss=0.09271, over 1614368.82 frames. ], batch size: 26, lr: 9.67e-03, grad_scale: 8.0 +2023-02-06 05:31:08,350 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:31:10,787 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.847e+02 3.468e+02 5.057e+02 9.591e+02, threshold=6.936e+02, percent-clipped=13.0 +2023-02-06 05:31:42,181 INFO [train.py:901] (2/4) Epoch 8, batch 3000, loss[loss=0.2292, simple_loss=0.3206, pruned_loss=0.06894, over 8288.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3277, pruned_loss=0.09342, over 1617896.57 frames. ], batch size: 23, lr: 9.67e-03, grad_scale: 8.0 +2023-02-06 05:31:42,181 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 05:31:54,428 INFO [train.py:935] (2/4) Epoch 8, validation: loss=0.2021, simple_loss=0.3001, pruned_loss=0.05199, over 944034.00 frames. +2023-02-06 05:31:54,429 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 05:32:10,778 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 05:32:29,771 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4301, 1.8541, 3.1025, 1.1998, 2.2756, 1.8269, 1.5298, 1.9524], + device='cuda:2'), covar=tensor([0.1605, 0.1982, 0.0631, 0.3497, 0.1398, 0.2519, 0.1657, 0.2095], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0479, 0.0530, 0.0558, 0.0601, 0.0535, 0.0457, 0.0591], + 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') +2023-02-06 05:32:30,901 INFO [train.py:901] (2/4) Epoch 8, batch 3050, loss[loss=0.2836, simple_loss=0.3563, pruned_loss=0.1054, over 8468.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3274, pruned_loss=0.09311, over 1619221.50 frames. ], batch size: 25, lr: 9.67e-03, grad_scale: 8.0 +2023-02-06 05:32:34,251 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.623e+02 3.324e+02 4.059e+02 7.396e+02, threshold=6.648e+02, percent-clipped=1.0 +2023-02-06 05:32:35,802 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:32:37,140 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59641.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:32:52,397 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:33:05,113 INFO [train.py:901] (2/4) Epoch 8, batch 3100, loss[loss=0.2233, simple_loss=0.3035, pruned_loss=0.07152, over 8036.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3269, pruned_loss=0.0926, over 1618841.66 frames. ], batch size: 22, lr: 9.66e-03, grad_scale: 8.0 +2023-02-06 05:33:40,028 INFO [train.py:901] (2/4) Epoch 8, batch 3150, loss[loss=0.2335, simple_loss=0.2979, pruned_loss=0.08453, over 7270.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3282, pruned_loss=0.09353, over 1620202.33 frames. ], batch size: 16, lr: 9.66e-03, grad_scale: 8.0 +2023-02-06 05:33:43,231 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.894e+02 3.427e+02 4.526e+02 8.691e+02, threshold=6.853e+02, percent-clipped=4.0 +2023-02-06 05:34:14,628 INFO [train.py:901] (2/4) Epoch 8, batch 3200, loss[loss=0.2555, simple_loss=0.3182, pruned_loss=0.09639, over 8090.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3289, pruned_loss=0.09408, over 1621935.13 frames. ], batch size: 21, lr: 9.65e-03, grad_scale: 8.0 +2023-02-06 05:34:21,173 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.52 vs. limit=5.0 +2023-02-06 05:34:30,939 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 05:34:50,981 INFO [train.py:901] (2/4) Epoch 8, batch 3250, loss[loss=0.2488, simple_loss=0.3112, pruned_loss=0.09326, over 7430.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3272, pruned_loss=0.09288, over 1620064.84 frames. ], batch size: 17, lr: 9.65e-03, grad_scale: 8.0 +2023-02-06 05:34:54,309 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.545e+02 3.201e+02 4.295e+02 9.179e+02, threshold=6.402e+02, percent-clipped=6.0 +2023-02-06 05:34:54,760 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-06 05:35:13,094 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:14,530 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59866.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:25,218 INFO [train.py:901] (2/4) Epoch 8, batch 3300, loss[loss=0.2349, simple_loss=0.3161, pruned_loss=0.07683, over 8247.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3262, pruned_loss=0.0921, over 1619160.49 frames. ], batch size: 24, lr: 9.65e-03, grad_scale: 8.0 +2023-02-06 05:35:35,306 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:41,745 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:52,456 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:53,732 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1839, 1.9289, 1.7353, 1.9741, 1.2360, 1.6971, 2.3595, 2.2747], + device='cuda:2'), covar=tensor([0.0463, 0.1100, 0.1632, 0.1213, 0.0558, 0.1491, 0.0595, 0.0529], + device='cuda:2'), in_proj_covar=tensor([0.0109, 0.0160, 0.0199, 0.0164, 0.0109, 0.0169, 0.0123, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:35:59,535 INFO [train.py:901] (2/4) Epoch 8, batch 3350, loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09917, over 8467.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3264, pruned_loss=0.09245, over 1618356.35 frames. ], batch size: 25, lr: 9.64e-03, grad_scale: 8.0 +2023-02-06 05:36:02,906 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.795e+02 3.400e+02 4.166e+02 8.824e+02, threshold=6.801e+02, percent-clipped=5.0 +2023-02-06 05:36:25,155 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3074, 1.5627, 1.5021, 1.2811, 1.0494, 1.3183, 1.6558, 1.6325], + device='cuda:2'), covar=tensor([0.0518, 0.1262, 0.1861, 0.1469, 0.0622, 0.1602, 0.0753, 0.0602], + device='cuda:2'), in_proj_covar=tensor([0.0110, 0.0162, 0.0200, 0.0165, 0.0111, 0.0170, 0.0124, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:36:31,821 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:36:33,704 INFO [train.py:901] (2/4) Epoch 8, batch 3400, loss[loss=0.2326, simple_loss=0.3138, pruned_loss=0.07571, over 8348.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3249, pruned_loss=0.0916, over 1614535.65 frames. ], batch size: 24, lr: 9.64e-03, grad_scale: 8.0 +2023-02-06 05:36:51,164 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 05:37:09,557 INFO [train.py:901] (2/4) Epoch 8, batch 3450, loss[loss=0.2442, simple_loss=0.3293, pruned_loss=0.07954, over 8362.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3256, pruned_loss=0.09197, over 1619381.96 frames. ], batch size: 24, lr: 9.63e-03, grad_scale: 8.0 +2023-02-06 05:37:12,878 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.668e+02 3.106e+02 3.891e+02 9.201e+02, threshold=6.211e+02, percent-clipped=2.0 +2023-02-06 05:37:27,911 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:36,406 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:44,252 INFO [train.py:901] (2/4) Epoch 8, batch 3500, loss[loss=0.2301, simple_loss=0.3111, pruned_loss=0.07456, over 8138.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3261, pruned_loss=0.09192, over 1616813.57 frames. ], batch size: 22, lr: 9.63e-03, grad_scale: 8.0 +2023-02-06 05:37:45,084 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60083.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:53,896 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60096.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:38:02,935 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 05:38:18,649 INFO [train.py:901] (2/4) Epoch 8, batch 3550, loss[loss=0.2864, simple_loss=0.3432, pruned_loss=0.1148, over 8507.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3264, pruned_loss=0.09235, over 1614192.05 frames. ], batch size: 26, lr: 9.63e-03, grad_scale: 8.0 +2023-02-06 05:38:22,095 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.844e+02 3.449e+02 4.512e+02 7.529e+02, threshold=6.898e+02, percent-clipped=5.0 +2023-02-06 05:38:43,017 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 +2023-02-06 05:38:54,373 INFO [train.py:901] (2/4) Epoch 8, batch 3600, loss[loss=0.2845, simple_loss=0.3438, pruned_loss=0.1126, over 8469.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3272, pruned_loss=0.09308, over 1613183.39 frames. ], batch size: 27, lr: 9.62e-03, grad_scale: 8.0 +2023-02-06 05:39:14,698 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60210.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:18,773 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60216.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:29,804 INFO [train.py:901] (2/4) Epoch 8, batch 3650, loss[loss=0.304, simple_loss=0.3526, pruned_loss=0.1277, over 7537.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3281, pruned_loss=0.09416, over 1607936.66 frames. ], batch size: 18, lr: 9.62e-03, grad_scale: 8.0 +2023-02-06 05:39:32,078 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60235.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:33,128 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.702e+02 3.457e+02 4.155e+02 9.631e+02, threshold=6.915e+02, percent-clipped=4.0 +2023-02-06 05:39:42,751 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60251.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:48,810 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60260.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:40:03,446 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 05:40:04,774 INFO [train.py:901] (2/4) Epoch 8, batch 3700, loss[loss=0.2367, simple_loss=0.3142, pruned_loss=0.07963, over 8280.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3264, pruned_loss=0.09351, over 1605223.13 frames. ], batch size: 23, lr: 9.61e-03, grad_scale: 8.0 +2023-02-06 05:40:34,569 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60325.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:40:39,115 INFO [train.py:901] (2/4) Epoch 8, batch 3750, loss[loss=0.2054, simple_loss=0.2873, pruned_loss=0.0618, over 7666.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3259, pruned_loss=0.09308, over 1606568.21 frames. ], batch size: 19, lr: 9.61e-03, grad_scale: 8.0 +2023-02-06 05:40:43,054 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.685e+02 3.295e+02 3.882e+02 8.274e+02, threshold=6.589e+02, percent-clipped=2.0 +2023-02-06 05:41:02,726 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:13,776 INFO [train.py:901] (2/4) Epoch 8, batch 3800, loss[loss=0.1937, simple_loss=0.2918, pruned_loss=0.04778, over 8343.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3267, pruned_loss=0.09349, over 1609224.87 frames. ], batch size: 24, lr: 9.61e-03, grad_scale: 8.0 +2023-02-06 05:41:28,150 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:36,467 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:45,878 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:49,067 INFO [train.py:901] (2/4) Epoch 8, batch 3850, loss[loss=0.2669, simple_loss=0.3424, pruned_loss=0.09571, over 8466.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3267, pruned_loss=0.09317, over 1609086.16 frames. ], batch size: 27, lr: 9.60e-03, grad_scale: 8.0 +2023-02-06 05:41:52,292 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.691e+02 3.271e+02 4.212e+02 1.032e+03, threshold=6.541e+02, percent-clipped=5.0 +2023-02-06 05:41:54,203 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:42:08,490 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 05:42:22,375 INFO [train.py:901] (2/4) Epoch 8, batch 3900, loss[loss=0.2373, simple_loss=0.3116, pruned_loss=0.08148, over 8200.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.327, pruned_loss=0.09406, over 1612480.54 frames. ], batch size: 23, lr: 9.60e-03, grad_scale: 8.0 +2023-02-06 05:42:46,595 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:42:55,308 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60529.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:42:57,130 INFO [train.py:901] (2/4) Epoch 8, batch 3950, loss[loss=0.223, simple_loss=0.306, pruned_loss=0.06996, over 8358.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3246, pruned_loss=0.09283, over 1608649.90 frames. ], batch size: 26, lr: 9.59e-03, grad_scale: 8.0 +2023-02-06 05:43:00,413 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.562e+02 3.362e+02 4.082e+02 8.516e+02, threshold=6.724e+02, percent-clipped=2.0 +2023-02-06 05:43:03,877 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60542.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:13,334 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60555.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:16,391 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60560.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:31,325 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:31,760 INFO [train.py:901] (2/4) Epoch 8, batch 4000, loss[loss=0.2465, simple_loss=0.3176, pruned_loss=0.08774, over 8471.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3262, pruned_loss=0.09423, over 1612200.62 frames. ], batch size: 25, lr: 9.59e-03, grad_scale: 16.0 +2023-02-06 05:43:46,944 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3582, 1.5932, 1.5623, 0.8799, 1.6930, 1.2190, 0.2913, 1.5020], + device='cuda:2'), covar=tensor([0.0259, 0.0157, 0.0147, 0.0241, 0.0175, 0.0529, 0.0423, 0.0142], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0276, 0.0226, 0.0335, 0.0267, 0.0427, 0.0329, 0.0304], + device='cuda:2'), out_proj_covar=tensor([1.0801e-04, 8.2682e-05, 6.7501e-05, 9.9941e-05, 8.1928e-05, 1.4039e-04, + 1.0076e-04, 9.2206e-05], device='cuda:2') +2023-02-06 05:43:48,273 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:59,775 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:44:06,268 INFO [train.py:901] (2/4) Epoch 8, batch 4050, loss[loss=0.2005, simple_loss=0.2843, pruned_loss=0.05837, over 7975.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3257, pruned_loss=0.09327, over 1611677.33 frames. ], batch size: 21, lr: 9.59e-03, grad_scale: 16.0 +2023-02-06 05:44:09,650 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.835e+02 3.722e+02 4.462e+02 8.493e+02, threshold=7.445e+02, percent-clipped=1.0 +2023-02-06 05:44:17,204 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:44:36,995 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60675.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:44:41,543 INFO [train.py:901] (2/4) Epoch 8, batch 4100, loss[loss=0.2423, simple_loss=0.3211, pruned_loss=0.08177, over 8481.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3256, pruned_loss=0.09342, over 1610949.20 frames. ], batch size: 29, lr: 9.58e-03, grad_scale: 16.0 +2023-02-06 05:44:51,186 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9947, 1.4644, 1.5664, 1.3418, 0.9405, 1.3716, 1.5110, 1.4968], + device='cuda:2'), covar=tensor([0.0539, 0.1178, 0.1673, 0.1335, 0.0580, 0.1443, 0.0713, 0.0598], + device='cuda:2'), in_proj_covar=tensor([0.0108, 0.0161, 0.0198, 0.0164, 0.0111, 0.0169, 0.0123, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:45:16,314 INFO [train.py:901] (2/4) Epoch 8, batch 4150, loss[loss=0.207, simple_loss=0.2856, pruned_loss=0.06424, over 7709.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3254, pruned_loss=0.09285, over 1614886.75 frames. ], batch size: 18, lr: 9.58e-03, grad_scale: 16.0 +2023-02-06 05:45:19,088 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:45:19,619 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.904e+02 3.574e+02 4.093e+02 8.234e+02, threshold=7.147e+02, percent-clipped=2.0 +2023-02-06 05:45:45,318 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:45:51,067 INFO [train.py:901] (2/4) Epoch 8, batch 4200, loss[loss=0.296, simple_loss=0.3487, pruned_loss=0.1216, over 6856.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3261, pruned_loss=0.09331, over 1611887.02 frames. ], batch size: 71, lr: 9.57e-03, grad_scale: 8.0 +2023-02-06 05:45:54,019 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:02,596 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:02,614 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:07,986 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 05:46:10,887 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:11,534 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60811.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:20,191 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60823.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:25,973 INFO [train.py:901] (2/4) Epoch 8, batch 4250, loss[loss=0.3123, simple_loss=0.3798, pruned_loss=0.1224, over 8506.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3256, pruned_loss=0.09317, over 1610403.32 frames. ], batch size: 49, lr: 9.57e-03, grad_scale: 8.0 +2023-02-06 05:46:28,902 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60836.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:30,099 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.807e+02 3.546e+02 4.515e+02 1.213e+03, threshold=7.092e+02, percent-clipped=3.0 +2023-02-06 05:46:30,813 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 05:46:56,667 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8880, 2.4390, 3.0316, 1.0907, 3.2754, 1.9971, 1.4828, 1.5947], + device='cuda:2'), covar=tensor([0.0459, 0.0231, 0.0123, 0.0423, 0.0199, 0.0433, 0.0543, 0.0305], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0276, 0.0225, 0.0337, 0.0264, 0.0424, 0.0330, 0.0304], + device='cuda:2'), out_proj_covar=tensor([1.0853e-04, 8.2391e-05, 6.6769e-05, 1.0081e-04, 8.0577e-05, 1.3922e-04, + 1.0109e-04, 9.1903e-05], device='cuda:2') +2023-02-06 05:47:01,076 INFO [train.py:901] (2/4) Epoch 8, batch 4300, loss[loss=0.2826, simple_loss=0.3534, pruned_loss=0.1059, over 8342.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3257, pruned_loss=0.09302, over 1617753.92 frames. ], batch size: 26, lr: 9.57e-03, grad_scale: 8.0 +2023-02-06 05:47:04,228 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 05:47:35,209 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:47:35,678 INFO [train.py:901] (2/4) Epoch 8, batch 4350, loss[loss=0.2598, simple_loss=0.3282, pruned_loss=0.09574, over 8464.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3265, pruned_loss=0.09364, over 1619339.59 frames. ], batch size: 29, lr: 9.56e-03, grad_scale: 8.0 +2023-02-06 05:47:39,641 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 2.751e+02 3.442e+02 4.335e+02 7.709e+02, threshold=6.884e+02, percent-clipped=1.0 +2023-02-06 05:47:40,103 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 05:47:51,989 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60956.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:47:59,289 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 05:48:10,331 INFO [train.py:901] (2/4) Epoch 8, batch 4400, loss[loss=0.241, simple_loss=0.3268, pruned_loss=0.07763, over 8353.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3248, pruned_loss=0.09253, over 1617916.38 frames. ], batch size: 26, lr: 9.56e-03, grad_scale: 8.0 +2023-02-06 05:48:34,694 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3643, 1.6676, 2.7579, 1.2053, 2.1005, 1.7348, 1.5689, 1.7182], + device='cuda:2'), covar=tensor([0.1941, 0.2286, 0.0756, 0.4135, 0.1494, 0.3071, 0.1904, 0.2307], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0486, 0.0522, 0.0559, 0.0598, 0.0541, 0.0455, 0.0593], + 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') +2023-02-06 05:48:40,683 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1442, 1.4173, 3.4679, 1.5167, 2.3146, 3.8877, 3.8829, 3.2929], + device='cuda:2'), covar=tensor([0.1003, 0.1552, 0.0340, 0.1974, 0.0859, 0.0229, 0.0394, 0.0605], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0283, 0.0245, 0.0277, 0.0253, 0.0224, 0.0290, 0.0288], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 05:48:42,527 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 05:48:45,123 INFO [train.py:901] (2/4) Epoch 8, batch 4450, loss[loss=0.3303, simple_loss=0.376, pruned_loss=0.1423, over 8501.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3247, pruned_loss=0.09248, over 1616728.04 frames. ], batch size: 26, lr: 9.55e-03, grad_scale: 8.0 +2023-02-06 05:48:49,119 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.403e+02 3.130e+02 3.948e+02 8.767e+02, threshold=6.260e+02, percent-clipped=3.0 +2023-02-06 05:49:17,374 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61079.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:49:17,968 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:49:19,295 INFO [train.py:901] (2/4) Epoch 8, batch 4500, loss[loss=0.2878, simple_loss=0.3345, pruned_loss=0.1206, over 8073.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3246, pruned_loss=0.092, over 1614067.38 frames. ], batch size: 21, lr: 9.55e-03, grad_scale: 8.0 +2023-02-06 05:49:36,537 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 05:49:53,490 INFO [train.py:901] (2/4) Epoch 8, batch 4550, loss[loss=0.264, simple_loss=0.3281, pruned_loss=0.09995, over 7930.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3255, pruned_loss=0.09303, over 1615854.36 frames. ], batch size: 20, lr: 9.55e-03, grad_scale: 8.0 +2023-02-06 05:49:58,149 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.768e+02 3.493e+02 4.645e+02 1.007e+03, threshold=6.986e+02, percent-clipped=6.0 +2023-02-06 05:50:26,108 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2716, 2.2377, 1.5520, 1.9229, 1.8139, 1.2620, 1.5957, 1.7663], + device='cuda:2'), covar=tensor([0.1193, 0.0353, 0.1073, 0.0467, 0.0580, 0.1302, 0.0840, 0.0809], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0235, 0.0313, 0.0297, 0.0305, 0.0317, 0.0340, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 05:50:29,339 INFO [train.py:901] (2/4) Epoch 8, batch 4600, loss[loss=0.2346, simple_loss=0.3089, pruned_loss=0.08012, over 7977.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3255, pruned_loss=0.09231, over 1617405.43 frames. ], batch size: 21, lr: 9.54e-03, grad_scale: 8.0 +2023-02-06 05:50:38,246 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61195.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:50:42,597 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 05:50:45,618 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6877, 2.1903, 4.8284, 2.6634, 4.3538, 4.2512, 4.5398, 4.4013], + device='cuda:2'), covar=tensor([0.0472, 0.3230, 0.0410, 0.2608, 0.0834, 0.0642, 0.0405, 0.0418], + device='cuda:2'), in_proj_covar=tensor([0.0417, 0.0532, 0.0514, 0.0484, 0.0543, 0.0461, 0.0461, 0.0510], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:2') +2023-02-06 05:51:04,224 INFO [train.py:901] (2/4) Epoch 8, batch 4650, loss[loss=0.2175, simple_loss=0.2983, pruned_loss=0.06834, over 8239.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.326, pruned_loss=0.09268, over 1617013.22 frames. ], batch size: 22, lr: 9.54e-03, grad_scale: 8.0 +2023-02-06 05:51:08,278 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 2.693e+02 3.115e+02 3.876e+02 8.832e+02, threshold=6.229e+02, percent-clipped=3.0 +2023-02-06 05:51:38,689 INFO [train.py:901] (2/4) Epoch 8, batch 4700, loss[loss=0.2706, simple_loss=0.3443, pruned_loss=0.09848, over 8667.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3267, pruned_loss=0.09268, over 1616101.42 frames. ], batch size: 39, lr: 9.54e-03, grad_scale: 8.0 +2023-02-06 05:52:13,892 INFO [train.py:901] (2/4) Epoch 8, batch 4750, loss[loss=0.252, simple_loss=0.3242, pruned_loss=0.08993, over 8029.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3267, pruned_loss=0.09246, over 1616409.94 frames. ], batch size: 22, lr: 9.53e-03, grad_scale: 8.0 +2023-02-06 05:52:14,755 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1015, 2.2412, 1.8255, 2.8616, 1.3215, 1.4281, 1.9128, 2.2650], + device='cuda:2'), covar=tensor([0.0841, 0.0857, 0.1192, 0.0433, 0.1271, 0.1906, 0.1158, 0.0926], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0231, 0.0270, 0.0217, 0.0229, 0.0265, 0.0269, 0.0236], + 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') +2023-02-06 05:52:17,850 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.870e+02 3.425e+02 4.672e+02 9.837e+02, threshold=6.850e+02, percent-clipped=8.0 +2023-02-06 05:52:36,774 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 05:52:39,420 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 05:52:48,204 INFO [train.py:901] (2/4) Epoch 8, batch 4800, loss[loss=0.2301, simple_loss=0.2996, pruned_loss=0.08032, over 7931.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.326, pruned_loss=0.09236, over 1614126.12 frames. ], batch size: 20, lr: 9.53e-03, grad_scale: 8.0 +2023-02-06 05:53:16,792 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61423.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:53:22,678 INFO [train.py:901] (2/4) Epoch 8, batch 4850, loss[loss=0.2814, simple_loss=0.3524, pruned_loss=0.1053, over 8461.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.327, pruned_loss=0.09273, over 1619032.66 frames. ], batch size: 25, lr: 9.52e-03, grad_scale: 8.0 +2023-02-06 05:53:26,652 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.780e+02 3.448e+02 4.323e+02 7.771e+02, threshold=6.895e+02, percent-clipped=1.0 +2023-02-06 05:53:28,674 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 05:53:36,187 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61451.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:53:53,262 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61476.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:53:56,852 INFO [train.py:901] (2/4) Epoch 8, batch 4900, loss[loss=0.2341, simple_loss=0.3023, pruned_loss=0.08297, over 7435.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3268, pruned_loss=0.09254, over 1620298.01 frames. ], batch size: 17, lr: 9.52e-03, grad_scale: 8.0 +2023-02-06 05:54:07,875 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 05:54:11,125 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4474, 1.5414, 2.3212, 1.1983, 1.5874, 1.6801, 1.3874, 1.4299], + device='cuda:2'), covar=tensor([0.1535, 0.1986, 0.0627, 0.3349, 0.1394, 0.2562, 0.1702, 0.1823], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0491, 0.0533, 0.0567, 0.0607, 0.0544, 0.0459, 0.0604], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:2') +2023-02-06 05:54:31,292 INFO [train.py:901] (2/4) Epoch 8, batch 4950, loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.127, over 7125.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3273, pruned_loss=0.09333, over 1619033.95 frames. ], batch size: 71, lr: 9.52e-03, grad_scale: 8.0 +2023-02-06 05:54:35,325 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.701e+02 3.325e+02 4.582e+02 7.633e+02, threshold=6.649e+02, percent-clipped=1.0 +2023-02-06 05:54:35,519 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61538.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:54:36,262 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0823, 1.2235, 1.2215, 0.4656, 1.2343, 1.0357, 0.1430, 1.1673], + device='cuda:2'), covar=tensor([0.0181, 0.0165, 0.0149, 0.0262, 0.0182, 0.0509, 0.0384, 0.0154], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0285, 0.0234, 0.0340, 0.0273, 0.0434, 0.0337, 0.0314], + device='cuda:2'), out_proj_covar=tensor([1.0832e-04, 8.5266e-05, 6.9743e-05, 1.0126e-04, 8.3255e-05, 1.4241e-04, + 1.0312e-04, 9.4617e-05], device='cuda:2') +2023-02-06 05:55:07,163 INFO [train.py:901] (2/4) Epoch 8, batch 5000, loss[loss=0.2206, simple_loss=0.31, pruned_loss=0.06561, over 8508.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3262, pruned_loss=0.09291, over 1619688.01 frames. ], batch size: 26, lr: 9.51e-03, grad_scale: 8.0 +2023-02-06 05:55:24,083 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5892, 1.9756, 2.1383, 1.0202, 2.2478, 1.4086, 0.6566, 1.7695], + device='cuda:2'), covar=tensor([0.0356, 0.0169, 0.0137, 0.0333, 0.0186, 0.0540, 0.0508, 0.0169], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0287, 0.0235, 0.0342, 0.0275, 0.0437, 0.0338, 0.0315], + device='cuda:2'), out_proj_covar=tensor([1.0891e-04, 8.5657e-05, 7.0072e-05, 1.0190e-04, 8.3875e-05, 1.4321e-04, + 1.0342e-04, 9.5080e-05], device='cuda:2') +2023-02-06 05:55:42,155 INFO [train.py:901] (2/4) Epoch 8, batch 5050, loss[loss=0.2798, simple_loss=0.3526, pruned_loss=0.1035, over 8107.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3262, pruned_loss=0.09278, over 1620368.03 frames. ], batch size: 23, lr: 9.51e-03, grad_scale: 8.0 +2023-02-06 05:55:46,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.771e+02 3.459e+02 4.924e+02 1.310e+03, threshold=6.919e+02, percent-clipped=9.0 +2023-02-06 05:56:07,700 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 05:56:17,393 INFO [train.py:901] (2/4) Epoch 8, batch 5100, loss[loss=0.2684, simple_loss=0.3256, pruned_loss=0.1056, over 8107.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3254, pruned_loss=0.09246, over 1617698.46 frames. ], batch size: 23, lr: 9.50e-03, grad_scale: 8.0 +2023-02-06 05:56:37,561 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8180, 2.0688, 2.1564, 1.9425, 1.5887, 2.1406, 2.5379, 2.1874], + device='cuda:2'), covar=tensor([0.0429, 0.0926, 0.1328, 0.1078, 0.0587, 0.1157, 0.0535, 0.0427], + device='cuda:2'), in_proj_covar=tensor([0.0108, 0.0161, 0.0200, 0.0164, 0.0112, 0.0170, 0.0123, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:56:52,939 INFO [train.py:901] (2/4) Epoch 8, batch 5150, loss[loss=0.2272, simple_loss=0.314, pruned_loss=0.07022, over 8138.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3266, pruned_loss=0.09346, over 1616730.68 frames. ], batch size: 22, lr: 9.50e-03, grad_scale: 8.0 +2023-02-06 05:56:57,119 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.584e+02 3.190e+02 4.018e+02 8.337e+02, threshold=6.381e+02, percent-clipped=2.0 +2023-02-06 05:57:06,176 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:57:11,023 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8409, 2.1212, 2.3898, 1.8680, 1.2168, 2.5459, 0.4488, 1.4832], + device='cuda:2'), covar=tensor([0.2992, 0.1848, 0.0704, 0.2268, 0.5934, 0.0538, 0.4573, 0.2336], + device='cuda:2'), in_proj_covar=tensor([0.0153, 0.0152, 0.0087, 0.0200, 0.0238, 0.0092, 0.0160, 0.0153], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 05:57:13,688 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:57:27,554 INFO [train.py:901] (2/4) Epoch 8, batch 5200, loss[loss=0.2787, simple_loss=0.3419, pruned_loss=0.1077, over 7976.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3251, pruned_loss=0.09288, over 1614254.13 frames. ], batch size: 21, lr: 9.50e-03, grad_scale: 8.0 +2023-02-06 05:57:35,725 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:57:53,737 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:58:00,556 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3722, 2.0851, 2.8828, 2.4741, 2.5655, 2.2012, 1.7458, 1.2619], + device='cuda:2'), covar=tensor([0.2871, 0.2978, 0.0740, 0.1793, 0.1548, 0.1627, 0.1582, 0.3096], + device='cuda:2'), in_proj_covar=tensor([0.0848, 0.0803, 0.0677, 0.0787, 0.0887, 0.0736, 0.0671, 0.0727], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 05:58:02,408 INFO [train.py:901] (2/4) Epoch 8, batch 5250, loss[loss=0.1887, simple_loss=0.2643, pruned_loss=0.05656, over 7416.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3253, pruned_loss=0.09312, over 1615212.43 frames. ], batch size: 17, lr: 9.49e-03, grad_scale: 8.0 +2023-02-06 05:58:06,467 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.711e+02 3.309e+02 4.013e+02 1.150e+03, threshold=6.618e+02, percent-clipped=3.0 +2023-02-06 05:58:07,805 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 05:58:12,702 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8393, 1.2420, 1.3611, 1.0917, 1.1651, 1.2220, 1.5332, 1.4533], + device='cuda:2'), covar=tensor([0.0597, 0.1751, 0.2567, 0.1879, 0.0654, 0.2152, 0.0842, 0.0717], + device='cuda:2'), in_proj_covar=tensor([0.0108, 0.0160, 0.0199, 0.0163, 0.0110, 0.0168, 0.0122, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 05:58:37,169 INFO [train.py:901] (2/4) Epoch 8, batch 5300, loss[loss=0.2522, simple_loss=0.3335, pruned_loss=0.08547, over 8198.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.325, pruned_loss=0.09237, over 1616134.63 frames. ], batch size: 23, lr: 9.49e-03, grad_scale: 8.0 +2023-02-06 05:59:08,401 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.68 vs. limit=5.0 +2023-02-06 05:59:12,097 INFO [train.py:901] (2/4) Epoch 8, batch 5350, loss[loss=0.2294, simple_loss=0.296, pruned_loss=0.08143, over 7265.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3243, pruned_loss=0.09217, over 1613605.92 frames. ], batch size: 16, lr: 9.49e-03, grad_scale: 8.0 +2023-02-06 05:59:12,256 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61932.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:59:16,964 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.596e+02 3.249e+02 3.983e+02 1.109e+03, threshold=6.498e+02, percent-clipped=6.0 +2023-02-06 05:59:47,798 INFO [train.py:901] (2/4) Epoch 8, batch 5400, loss[loss=0.2394, simple_loss=0.3169, pruned_loss=0.08099, over 8422.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3239, pruned_loss=0.09165, over 1611932.75 frames. ], batch size: 39, lr: 9.48e-03, grad_scale: 8.0 +2023-02-06 06:00:23,965 INFO [train.py:901] (2/4) Epoch 8, batch 5450, loss[loss=0.2036, simple_loss=0.277, pruned_loss=0.06507, over 7426.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3236, pruned_loss=0.09158, over 1606204.47 frames. ], batch size: 17, lr: 9.48e-03, grad_scale: 8.0 +2023-02-06 06:00:28,689 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.625e+02 3.240e+02 4.068e+02 8.471e+02, threshold=6.479e+02, percent-clipped=5.0 +2023-02-06 06:00:48,385 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8726, 2.3690, 3.8764, 2.9135, 3.2274, 2.5309, 1.9703, 1.9657], + device='cuda:2'), covar=tensor([0.2775, 0.3472, 0.0839, 0.1969, 0.1563, 0.1559, 0.1334, 0.3460], + device='cuda:2'), in_proj_covar=tensor([0.0846, 0.0803, 0.0677, 0.0788, 0.0886, 0.0738, 0.0668, 0.0726], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 06:01:00,411 INFO [train.py:901] (2/4) Epoch 8, batch 5500, loss[loss=0.2197, simple_loss=0.2912, pruned_loss=0.07405, over 7927.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3238, pruned_loss=0.09137, over 1610057.33 frames. ], batch size: 20, lr: 9.47e-03, grad_scale: 8.0 +2023-02-06 06:01:01,783 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 06:01:08,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:01:16,546 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62105.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:01:34,828 INFO [train.py:901] (2/4) Epoch 8, batch 5550, loss[loss=0.2004, simple_loss=0.2694, pruned_loss=0.06573, over 7688.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3233, pruned_loss=0.09123, over 1609192.99 frames. ], batch size: 18, lr: 9.47e-03, grad_scale: 8.0 +2023-02-06 06:01:38,597 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.724e+02 3.277e+02 4.222e+02 9.983e+02, threshold=6.553e+02, percent-clipped=5.0 +2023-02-06 06:01:40,100 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:02:09,237 INFO [train.py:901] (2/4) Epoch 8, batch 5600, loss[loss=0.2628, simple_loss=0.3345, pruned_loss=0.09557, over 8335.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3238, pruned_loss=0.09177, over 1605829.72 frames. ], batch size: 25, lr: 9.47e-03, grad_scale: 8.0 +2023-02-06 06:02:27,948 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:02:35,378 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62220.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:02:44,108 INFO [train.py:901] (2/4) Epoch 8, batch 5650, loss[loss=0.3056, simple_loss=0.363, pruned_loss=0.1241, over 8516.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3259, pruned_loss=0.09257, over 1611615.34 frames. ], batch size: 26, lr: 9.46e-03, grad_scale: 8.0 +2023-02-06 06:02:48,206 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.893e+02 3.442e+02 4.058e+02 7.819e+02, threshold=6.884e+02, percent-clipped=2.0 +2023-02-06 06:03:03,294 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 06:03:14,926 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:03:19,751 INFO [train.py:901] (2/4) Epoch 8, batch 5700, loss[loss=0.305, simple_loss=0.3594, pruned_loss=0.1253, over 6983.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3259, pruned_loss=0.09289, over 1605432.21 frames. ], batch size: 72, lr: 9.46e-03, grad_scale: 8.0 +2023-02-06 06:03:19,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8036, 2.0240, 2.3439, 1.8217, 1.2169, 2.3350, 0.4250, 1.6254], + device='cuda:2'), covar=tensor([0.3052, 0.1711, 0.0475, 0.1719, 0.5099, 0.0439, 0.4287, 0.1745], + device='cuda:2'), in_proj_covar=tensor([0.0152, 0.0152, 0.0087, 0.0199, 0.0240, 0.0091, 0.0159, 0.0151], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 06:03:24,800 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9533, 2.2856, 3.8144, 2.8842, 3.3063, 2.5142, 1.9116, 1.7995], + device='cuda:2'), covar=tensor([0.2544, 0.3379, 0.0740, 0.2055, 0.1349, 0.1667, 0.1347, 0.3638], + device='cuda:2'), in_proj_covar=tensor([0.0851, 0.0806, 0.0679, 0.0792, 0.0892, 0.0742, 0.0677, 0.0730], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 06:03:26,015 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:03:53,771 INFO [train.py:901] (2/4) Epoch 8, batch 5750, loss[loss=0.2913, simple_loss=0.3577, pruned_loss=0.1124, over 8646.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3268, pruned_loss=0.0935, over 1607268.97 frames. ], batch size: 34, lr: 9.45e-03, grad_scale: 8.0 +2023-02-06 06:03:58,446 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.702e+02 3.342e+02 4.214e+02 1.406e+03, threshold=6.684e+02, percent-clipped=3.0 +2023-02-06 06:04:07,823 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 06:04:28,273 INFO [train.py:901] (2/4) Epoch 8, batch 5800, loss[loss=0.2124, simple_loss=0.2811, pruned_loss=0.07183, over 7791.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3268, pruned_loss=0.09357, over 1605505.74 frames. ], batch size: 19, lr: 9.45e-03, grad_scale: 8.0 +2023-02-06 06:04:35,219 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:04:44,671 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0403, 3.0738, 3.1077, 2.0774, 1.7171, 3.2829, 0.7549, 2.1053], + device='cuda:2'), covar=tensor([0.2010, 0.1134, 0.0459, 0.2969, 0.5139, 0.0381, 0.4688, 0.2199], + device='cuda:2'), in_proj_covar=tensor([0.0151, 0.0153, 0.0088, 0.0200, 0.0239, 0.0092, 0.0159, 0.0153], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 06:04:48,170 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:04:59,503 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1418, 1.5786, 1.6039, 1.4928, 1.1010, 1.3983, 1.7151, 1.8307], + device='cuda:2'), covar=tensor([0.0505, 0.1208, 0.1768, 0.1349, 0.0592, 0.1529, 0.0663, 0.0532], + device='cuda:2'), in_proj_covar=tensor([0.0110, 0.0161, 0.0200, 0.0165, 0.0112, 0.0169, 0.0123, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 06:05:04,104 INFO [train.py:901] (2/4) Epoch 8, batch 5850, loss[loss=0.2205, simple_loss=0.2864, pruned_loss=0.07724, over 7525.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3264, pruned_loss=0.09347, over 1607714.31 frames. ], batch size: 18, lr: 9.45e-03, grad_scale: 8.0 +2023-02-06 06:05:08,211 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.970e+02 2.690e+02 3.286e+02 4.000e+02 6.740e+02, threshold=6.571e+02, percent-clipped=1.0 +2023-02-06 06:05:27,198 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:34,614 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62476.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:38,456 INFO [train.py:901] (2/4) Epoch 8, batch 5900, loss[loss=0.2556, simple_loss=0.3238, pruned_loss=0.0937, over 7671.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3252, pruned_loss=0.093, over 1604048.21 frames. ], batch size: 19, lr: 9.44e-03, grad_scale: 8.0 +2023-02-06 06:05:39,870 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62484.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:44,020 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:51,358 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62501.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:54,035 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7975, 1.5080, 3.5140, 1.3574, 2.1985, 3.8031, 3.8625, 3.2846], + device='cuda:2'), covar=tensor([0.1101, 0.1446, 0.0293, 0.1885, 0.0962, 0.0250, 0.0338, 0.0601], + device='cuda:2'), in_proj_covar=tensor([0.0252, 0.0284, 0.0242, 0.0274, 0.0256, 0.0226, 0.0292, 0.0287], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 06:06:13,154 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2023-02-06 06:06:13,413 INFO [train.py:901] (2/4) Epoch 8, batch 5950, loss[loss=0.3148, simple_loss=0.3604, pruned_loss=0.1346, over 6895.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3254, pruned_loss=0.09284, over 1607888.87 frames. ], batch size: 71, lr: 9.44e-03, grad_scale: 8.0 +2023-02-06 06:06:16,364 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8836, 2.6804, 1.7854, 3.9779, 2.0921, 1.4834, 2.5777, 2.8047], + device='cuda:2'), covar=tensor([0.1895, 0.1575, 0.2652, 0.0339, 0.1553, 0.2683, 0.1399, 0.1166], + device='cuda:2'), in_proj_covar=tensor([0.0245, 0.0224, 0.0264, 0.0213, 0.0226, 0.0260, 0.0262, 0.0229], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 06:06:17,413 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 2.652e+02 3.220e+02 3.904e+02 8.315e+02, threshold=6.439e+02, percent-clipped=2.0 +2023-02-06 06:06:34,791 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62563.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:06:47,816 INFO [train.py:901] (2/4) Epoch 8, batch 6000, loss[loss=0.2255, simple_loss=0.286, pruned_loss=0.08254, over 7695.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3242, pruned_loss=0.09208, over 1603640.12 frames. ], batch size: 18, lr: 9.44e-03, grad_scale: 8.0 +2023-02-06 06:06:47,816 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 06:07:00,014 INFO [train.py:935] (2/4) Epoch 8, validation: loss=0.1996, simple_loss=0.2985, pruned_loss=0.05037, over 944034.00 frames. +2023-02-06 06:07:00,015 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 06:07:12,285 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62599.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:14,968 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:33,814 INFO [train.py:901] (2/4) Epoch 8, batch 6050, loss[loss=0.2319, simple_loss=0.2988, pruned_loss=0.08249, over 7205.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3239, pruned_loss=0.09201, over 1604975.39 frames. ], batch size: 16, lr: 9.43e-03, grad_scale: 8.0 +2023-02-06 06:07:35,910 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:37,858 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.575e+02 3.268e+02 4.071e+02 9.720e+02, threshold=6.536e+02, percent-clipped=3.0 +2023-02-06 06:07:44,182 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:56,963 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2438, 1.8339, 1.9000, 1.9496, 1.0971, 2.0130, 2.4587, 2.2830], + device='cuda:2'), covar=tensor([0.0461, 0.1113, 0.1643, 0.1261, 0.0627, 0.1267, 0.0561, 0.0501], + device='cuda:2'), in_proj_covar=tensor([0.0110, 0.0161, 0.0201, 0.0165, 0.0113, 0.0169, 0.0123, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 06:08:02,399 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:08,760 INFO [train.py:901] (2/4) Epoch 8, batch 6100, loss[loss=0.2081, simple_loss=0.2881, pruned_loss=0.06401, over 7801.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3233, pruned_loss=0.09147, over 1605251.32 frames. ], batch size: 20, lr: 9.43e-03, grad_scale: 8.0 +2023-02-06 06:08:10,157 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62684.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:37,242 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 06:08:43,088 INFO [train.py:901] (2/4) Epoch 8, batch 6150, loss[loss=0.2271, simple_loss=0.2974, pruned_loss=0.07842, over 7798.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3241, pruned_loss=0.09156, over 1611687.30 frames. ], batch size: 19, lr: 9.42e-03, grad_scale: 8.0 +2023-02-06 06:08:47,084 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.667e+02 3.544e+02 4.037e+02 8.376e+02, threshold=7.087e+02, percent-clipped=5.0 +2023-02-06 06:08:55,075 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:57,067 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:09:14,631 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 +2023-02-06 06:09:17,711 INFO [train.py:901] (2/4) Epoch 8, batch 6200, loss[loss=0.2118, simple_loss=0.2905, pruned_loss=0.06652, over 6873.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3242, pruned_loss=0.09227, over 1606237.24 frames. ], batch size: 15, lr: 9.42e-03, grad_scale: 16.0 +2023-02-06 06:09:23,759 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:09:52,742 INFO [train.py:901] (2/4) Epoch 8, batch 6250, loss[loss=0.234, simple_loss=0.3029, pruned_loss=0.0826, over 7788.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3236, pruned_loss=0.09193, over 1609631.07 frames. ], batch size: 19, lr: 9.42e-03, grad_scale: 16.0 +2023-02-06 06:09:56,740 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.717e+02 3.222e+02 4.596e+02 9.217e+02, threshold=6.445e+02, percent-clipped=3.0 +2023-02-06 06:10:08,418 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62855.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:10:16,910 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:10:24,981 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62880.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:10:26,043 INFO [train.py:901] (2/4) Epoch 8, batch 6300, loss[loss=0.2295, simple_loss=0.3134, pruned_loss=0.07275, over 8358.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3237, pruned_loss=0.09204, over 1606366.74 frames. ], batch size: 24, lr: 9.41e-03, grad_scale: 16.0 +2023-02-06 06:10:28,481 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 06:10:44,098 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:11:01,294 INFO [train.py:901] (2/4) Epoch 8, batch 6350, loss[loss=0.301, simple_loss=0.3786, pruned_loss=0.1117, over 8585.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3263, pruned_loss=0.09388, over 1602742.07 frames. ], batch size: 34, lr: 9.41e-03, grad_scale: 16.0 +2023-02-06 06:11:05,344 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.824e+02 3.504e+02 4.161e+02 7.437e+02, threshold=7.007e+02, percent-clipped=2.0 +2023-02-06 06:11:12,100 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62947.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:11:22,820 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0611, 2.1947, 1.8198, 2.5391, 1.4587, 1.6424, 1.9309, 2.3071], + device='cuda:2'), covar=tensor([0.0850, 0.0898, 0.1221, 0.0501, 0.1106, 0.1457, 0.0955, 0.0768], + device='cuda:2'), in_proj_covar=tensor([0.0251, 0.0227, 0.0268, 0.0217, 0.0230, 0.0263, 0.0266, 0.0234], + 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') +2023-02-06 06:11:35,244 INFO [train.py:901] (2/4) Epoch 8, batch 6400, loss[loss=0.2989, simple_loss=0.3742, pruned_loss=0.1118, over 8467.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3259, pruned_loss=0.09325, over 1604723.50 frames. ], batch size: 25, lr: 9.41e-03, grad_scale: 16.0 +2023-02-06 06:11:52,126 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:11:54,035 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5397, 4.4913, 4.0147, 1.9058, 4.0259, 3.9571, 4.1716, 3.6564], + device='cuda:2'), covar=tensor([0.0707, 0.0554, 0.1092, 0.4737, 0.0862, 0.0889, 0.1347, 0.0764], + device='cuda:2'), in_proj_covar=tensor([0.0425, 0.0337, 0.0353, 0.0446, 0.0352, 0.0334, 0.0344, 0.0294], + 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') +2023-02-06 06:12:03,489 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:07,444 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:09,594 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:10,068 INFO [train.py:901] (2/4) Epoch 8, batch 6450, loss[loss=0.3628, simple_loss=0.3943, pruned_loss=0.1657, over 6693.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3263, pruned_loss=0.09355, over 1605975.47 frames. ], batch size: 71, lr: 9.40e-03, grad_scale: 16.0 +2023-02-06 06:12:14,130 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 2.983e+02 3.820e+02 5.218e+02 9.633e+02, threshold=7.640e+02, percent-clipped=4.0 +2023-02-06 06:12:31,362 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:45,219 INFO [train.py:901] (2/4) Epoch 8, batch 6500, loss[loss=0.2821, simple_loss=0.3604, pruned_loss=0.1019, over 8338.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3265, pruned_loss=0.09339, over 1606413.09 frames. ], batch size: 26, lr: 9.40e-03, grad_scale: 16.0 +2023-02-06 06:13:14,194 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63124.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:20,106 INFO [train.py:901] (2/4) Epoch 8, batch 6550, loss[loss=0.2737, simple_loss=0.3557, pruned_loss=0.09589, over 8244.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.327, pruned_loss=0.09328, over 1611661.52 frames. ], batch size: 24, lr: 9.40e-03, grad_scale: 16.0 +2023-02-06 06:13:22,292 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:24,224 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.645e+02 3.116e+02 3.905e+02 9.747e+02, threshold=6.232e+02, percent-clipped=3.0 +2023-02-06 06:13:24,627 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 06:13:27,749 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63143.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:31,967 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63149.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:49,398 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 06:13:54,675 INFO [train.py:901] (2/4) Epoch 8, batch 6600, loss[loss=0.2862, simple_loss=0.3623, pruned_loss=0.105, over 8637.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.327, pruned_loss=0.09377, over 1613334.55 frames. ], batch size: 34, lr: 9.39e-03, grad_scale: 16.0 +2023-02-06 06:14:07,363 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 06:14:29,747 INFO [train.py:901] (2/4) Epoch 8, batch 6650, loss[loss=0.2533, simple_loss=0.3248, pruned_loss=0.09087, over 8466.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3254, pruned_loss=0.09309, over 1607335.54 frames. ], batch size: 27, lr: 9.39e-03, grad_scale: 16.0 +2023-02-06 06:14:33,643 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 2.943e+02 3.528e+02 4.449e+02 1.178e+03, threshold=7.055e+02, percent-clipped=8.0 +2023-02-06 06:14:34,520 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:14:42,438 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:01,174 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:03,728 INFO [train.py:901] (2/4) Epoch 8, batch 6700, loss[loss=0.2275, simple_loss=0.3059, pruned_loss=0.07449, over 8484.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3237, pruned_loss=0.0918, over 1606909.84 frames. ], batch size: 25, lr: 9.38e-03, grad_scale: 16.0 +2023-02-06 06:15:07,926 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6837, 2.9977, 2.6705, 3.9571, 1.7016, 2.0297, 2.3750, 3.1706], + device='cuda:2'), covar=tensor([0.0753, 0.1001, 0.0905, 0.0258, 0.1362, 0.1530, 0.1248, 0.0847], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0225, 0.0269, 0.0217, 0.0229, 0.0262, 0.0266, 0.0232], + 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') +2023-02-06 06:15:19,259 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63303.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:23,927 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5897, 2.0454, 2.1381, 1.1783, 2.2339, 1.4590, 0.7031, 1.7808], + device='cuda:2'), covar=tensor([0.0287, 0.0162, 0.0134, 0.0284, 0.0187, 0.0478, 0.0399, 0.0133], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0284, 0.0232, 0.0338, 0.0271, 0.0434, 0.0333, 0.0311], + device='cuda:2'), out_proj_covar=tensor([1.0913e-04, 8.3731e-05, 6.8561e-05, 1.0041e-04, 8.1878e-05, 1.4112e-04, + 1.0095e-04, 9.3288e-05], device='cuda:2') +2023-02-06 06:15:29,350 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63318.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:38,594 INFO [train.py:901] (2/4) Epoch 8, batch 6750, loss[loss=0.2561, simple_loss=0.3102, pruned_loss=0.101, over 7529.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.323, pruned_loss=0.09125, over 1604523.65 frames. ], batch size: 18, lr: 9.38e-03, grad_scale: 8.0 +2023-02-06 06:15:43,298 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.603e+02 3.068e+02 3.707e+02 1.416e+03, threshold=6.136e+02, percent-clipped=3.0 +2023-02-06 06:15:46,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:13,021 INFO [train.py:901] (2/4) Epoch 8, batch 6800, loss[loss=0.2, simple_loss=0.2833, pruned_loss=0.05834, over 7648.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3236, pruned_loss=0.09129, over 1609040.78 frames. ], batch size: 19, lr: 9.38e-03, grad_scale: 8.0 +2023-02-06 06:16:19,355 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2023-02-06 06:16:20,988 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 06:16:24,489 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63399.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:40,082 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5922, 2.4658, 4.5965, 1.2450, 2.9040, 2.0485, 1.7543, 2.5695], + device='cuda:2'), covar=tensor([0.1741, 0.1961, 0.0730, 0.4081, 0.1556, 0.2860, 0.1810, 0.2529], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0487, 0.0531, 0.0566, 0.0603, 0.0535, 0.0459, 0.0598], + 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') +2023-02-06 06:16:42,731 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:47,930 INFO [train.py:901] (2/4) Epoch 8, batch 6850, loss[loss=0.3152, simple_loss=0.3642, pruned_loss=0.1331, over 6894.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3234, pruned_loss=0.09162, over 1604417.84 frames. ], batch size: 71, lr: 9.37e-03, grad_scale: 8.0 +2023-02-06 06:16:52,299 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:52,777 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.669e+02 3.418e+02 4.059e+02 7.847e+02, threshold=6.836e+02, percent-clipped=4.0 +2023-02-06 06:16:53,655 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:17:11,111 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 06:17:22,528 INFO [train.py:901] (2/4) Epoch 8, batch 6900, loss[loss=0.2234, simple_loss=0.298, pruned_loss=0.07435, over 8240.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.323, pruned_loss=0.09174, over 1606309.25 frames. ], batch size: 22, lr: 9.37e-03, grad_scale: 8.0 +2023-02-06 06:17:31,380 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 06:17:39,557 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:17:58,288 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:17:58,757 INFO [train.py:901] (2/4) Epoch 8, batch 6950, loss[loss=0.2139, simple_loss=0.3035, pruned_loss=0.06214, over 8039.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3239, pruned_loss=0.09181, over 1605843.35 frames. ], batch size: 22, lr: 9.37e-03, grad_scale: 8.0 +2023-02-06 06:18:03,570 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.671e+02 3.369e+02 4.495e+02 9.890e+02, threshold=6.738e+02, percent-clipped=4.0 +2023-02-06 06:18:09,754 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1678, 1.2538, 4.3108, 1.7766, 2.2838, 4.7950, 4.9068, 3.9431], + device='cuda:2'), covar=tensor([0.1162, 0.1931, 0.0304, 0.1881, 0.1106, 0.0244, 0.0258, 0.0735], + device='cuda:2'), in_proj_covar=tensor([0.0247, 0.0278, 0.0239, 0.0265, 0.0250, 0.0221, 0.0289, 0.0278], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 06:18:18,577 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 06:18:32,008 INFO [train.py:901] (2/4) Epoch 8, batch 7000, loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1195, over 8463.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3234, pruned_loss=0.0916, over 1607680.23 frames. ], batch size: 29, lr: 9.36e-03, grad_scale: 8.0 +2023-02-06 06:18:32,809 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:18:36,323 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:18:48,873 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63604.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:19:08,075 INFO [train.py:901] (2/4) Epoch 8, batch 7050, loss[loss=0.2082, simple_loss=0.2872, pruned_loss=0.06459, over 7647.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3244, pruned_loss=0.09168, over 1608388.54 frames. ], batch size: 19, lr: 9.36e-03, grad_scale: 8.0 +2023-02-06 06:19:12,575 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.720e+02 3.242e+02 3.930e+02 7.648e+02, threshold=6.484e+02, percent-clipped=3.0 +2023-02-06 06:19:42,436 INFO [train.py:901] (2/4) Epoch 8, batch 7100, loss[loss=0.2747, simple_loss=0.3475, pruned_loss=0.101, over 8253.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3245, pruned_loss=0.09116, over 1610041.29 frames. ], batch size: 24, lr: 9.35e-03, grad_scale: 8.0 +2023-02-06 06:19:53,533 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:20:17,191 INFO [train.py:901] (2/4) Epoch 8, batch 7150, loss[loss=0.2348, simple_loss=0.3168, pruned_loss=0.07634, over 8197.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3254, pruned_loss=0.09202, over 1613203.12 frames. ], batch size: 23, lr: 9.35e-03, grad_scale: 8.0 +2023-02-06 06:20:21,741 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.778e+02 3.549e+02 4.516e+02 1.097e+03, threshold=7.098e+02, percent-clipped=7.0 +2023-02-06 06:20:51,748 INFO [train.py:901] (2/4) Epoch 8, batch 7200, loss[loss=0.2283, simple_loss=0.309, pruned_loss=0.0738, over 8192.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3248, pruned_loss=0.09156, over 1613973.88 frames. ], batch size: 23, lr: 9.35e-03, grad_scale: 8.0 +2023-02-06 06:20:51,831 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63782.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:20:53,156 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:21:21,048 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 06:21:21,571 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5517, 1.9182, 3.1741, 1.3205, 2.2741, 1.9689, 1.6191, 1.9109], + device='cuda:2'), covar=tensor([0.1520, 0.2012, 0.0634, 0.3460, 0.1341, 0.2486, 0.1654, 0.2108], + device='cuda:2'), in_proj_covar=tensor([0.0484, 0.0488, 0.0534, 0.0568, 0.0606, 0.0542, 0.0461, 0.0603], + 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') +2023-02-06 06:21:23,973 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:21:26,392 INFO [train.py:901] (2/4) Epoch 8, batch 7250, loss[loss=0.2367, simple_loss=0.3159, pruned_loss=0.07879, over 8755.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3244, pruned_loss=0.09181, over 1611382.32 frames. ], batch size: 30, lr: 9.34e-03, grad_scale: 8.0 +2023-02-06 06:21:30,981 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.694e+02 3.202e+02 4.148e+02 8.009e+02, threshold=6.403e+02, percent-clipped=2.0 +2023-02-06 06:21:42,697 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.60 vs. limit=5.0 +2023-02-06 06:21:48,005 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63863.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:00,775 INFO [train.py:901] (2/4) Epoch 8, batch 7300, loss[loss=0.227, simple_loss=0.2985, pruned_loss=0.07777, over 8089.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3248, pruned_loss=0.09138, over 1615495.47 frames. ], batch size: 21, lr: 9.34e-03, grad_scale: 8.0 +2023-02-06 06:22:12,945 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:14,330 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:36,710 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:37,304 INFO [train.py:901] (2/4) Epoch 8, batch 7350, loss[loss=0.2179, simple_loss=0.3045, pruned_loss=0.06563, over 8079.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3237, pruned_loss=0.09032, over 1615192.18 frames. ], batch size: 21, lr: 9.34e-03, grad_scale: 8.0 +2023-02-06 06:22:38,881 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63934.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:42,238 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.551e+02 3.183e+02 3.767e+02 5.416e+02, threshold=6.365e+02, percent-clipped=0.0 +2023-02-06 06:22:48,933 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63948.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:53,925 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63954.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:23:05,956 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 06:23:10,849 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:23:12,547 INFO [train.py:901] (2/4) Epoch 8, batch 7400, loss[loss=0.2082, simple_loss=0.284, pruned_loss=0.0662, over 7925.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3242, pruned_loss=0.09048, over 1616637.52 frames. ], batch size: 20, lr: 9.33e-03, grad_scale: 8.0 +2023-02-06 06:23:22,471 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2023-02-06 06:23:24,819 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 06:23:38,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3613, 2.6077, 2.2805, 2.9854, 2.1444, 2.0496, 2.2936, 2.7541], + device='cuda:2'), covar=tensor([0.0678, 0.0701, 0.0888, 0.0410, 0.0886, 0.1090, 0.0793, 0.0604], + device='cuda:2'), in_proj_covar=tensor([0.0246, 0.0223, 0.0270, 0.0214, 0.0226, 0.0262, 0.0265, 0.0232], + 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') +2023-02-06 06:23:48,438 INFO [train.py:901] (2/4) Epoch 8, batch 7450, loss[loss=0.2406, simple_loss=0.3231, pruned_loss=0.07907, over 8679.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3234, pruned_loss=0.09092, over 1610313.65 frames. ], batch size: 34, lr: 9.33e-03, grad_scale: 8.0 +2023-02-06 06:23:53,166 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.827e+02 3.358e+02 3.935e+02 9.777e+02, threshold=6.715e+02, percent-clipped=5.0 +2023-02-06 06:23:54,690 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64041.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:23:58,105 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64046.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:24:05,494 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 06:24:10,182 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64063.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:24:23,173 INFO [train.py:901] (2/4) Epoch 8, batch 7500, loss[loss=0.236, simple_loss=0.3037, pruned_loss=0.08418, over 7810.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3249, pruned_loss=0.09164, over 1615320.73 frames. ], batch size: 20, lr: 9.33e-03, grad_scale: 8.0 +2023-02-06 06:24:26,658 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 06:24:57,465 INFO [train.py:901] (2/4) Epoch 8, batch 7550, loss[loss=0.2466, simple_loss=0.3205, pruned_loss=0.08637, over 8458.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3262, pruned_loss=0.09253, over 1613542.32 frames. ], batch size: 29, lr: 9.32e-03, grad_scale: 8.0 +2023-02-06 06:25:02,136 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 3.017e+02 3.905e+02 4.969e+02 7.546e+02, threshold=7.810e+02, percent-clipped=1.0 +2023-02-06 06:25:11,701 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:13,074 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64155.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:15,753 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9164, 1.5337, 2.1421, 1.8022, 1.9059, 1.7798, 1.4490, 0.5911], + device='cuda:2'), covar=tensor([0.3031, 0.3086, 0.0911, 0.1803, 0.1405, 0.1745, 0.1399, 0.2975], + device='cuda:2'), in_proj_covar=tensor([0.0847, 0.0808, 0.0689, 0.0803, 0.0893, 0.0746, 0.0678, 0.0723], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 06:25:24,180 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64172.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:28,950 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:30,999 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:32,166 INFO [train.py:901] (2/4) Epoch 8, batch 7600, loss[loss=0.2446, simple_loss=0.3188, pruned_loss=0.08522, over 8250.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.325, pruned_loss=0.09205, over 1610477.97 frames. ], batch size: 22, lr: 9.32e-03, grad_scale: 8.0 +2023-02-06 06:25:49,063 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64207.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:07,006 INFO [train.py:901] (2/4) Epoch 8, batch 7650, loss[loss=0.2221, simple_loss=0.3025, pruned_loss=0.0708, over 8507.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3251, pruned_loss=0.09232, over 1611357.10 frames. ], batch size: 26, lr: 9.31e-03, grad_scale: 8.0 +2023-02-06 06:26:07,157 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6541, 1.4860, 4.7517, 1.7361, 4.1846, 3.8902, 4.3124, 4.1764], + device='cuda:2'), covar=tensor([0.0406, 0.3990, 0.0503, 0.3249, 0.1131, 0.0794, 0.0472, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0426, 0.0546, 0.0536, 0.0494, 0.0561, 0.0477, 0.0471, 0.0529], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 06:26:11,834 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.778e+02 3.467e+02 5.154e+02 1.113e+03, threshold=6.933e+02, percent-clipped=3.0 +2023-02-06 06:26:19,381 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:35,467 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6877, 2.9633, 2.3460, 3.9154, 1.8385, 2.2429, 2.5195, 2.9882], + device='cuda:2'), covar=tensor([0.0716, 0.0917, 0.1072, 0.0246, 0.1328, 0.1516, 0.1295, 0.0894], + device='cuda:2'), in_proj_covar=tensor([0.0249, 0.0227, 0.0270, 0.0215, 0.0228, 0.0265, 0.0266, 0.0233], + 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') +2023-02-06 06:26:38,167 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:40,894 INFO [train.py:901] (2/4) Epoch 8, batch 7700, loss[loss=0.2373, simple_loss=0.3156, pruned_loss=0.0795, over 8427.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3254, pruned_loss=0.09212, over 1618625.15 frames. ], batch size: 49, lr: 9.31e-03, grad_scale: 8.0 +2023-02-06 06:26:45,248 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:56,487 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:08,129 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:09,995 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:10,465 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 06:27:10,608 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1647, 1.8046, 3.4432, 1.4346, 2.1780, 3.7668, 3.8302, 3.2008], + device='cuda:2'), covar=tensor([0.1016, 0.1387, 0.0348, 0.1937, 0.0967, 0.0258, 0.0431, 0.0589], + device='cuda:2'), in_proj_covar=tensor([0.0256, 0.0287, 0.0244, 0.0274, 0.0257, 0.0227, 0.0298, 0.0286], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 06:27:13,362 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:16,534 INFO [train.py:901] (2/4) Epoch 8, batch 7750, loss[loss=0.1956, simple_loss=0.272, pruned_loss=0.05961, over 7434.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3245, pruned_loss=0.09137, over 1618720.92 frames. ], batch size: 17, lr: 9.31e-03, grad_scale: 8.0 +2023-02-06 06:27:21,027 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.516e+02 3.070e+02 3.996e+02 6.859e+02, threshold=6.139e+02, percent-clipped=0.0 +2023-02-06 06:27:25,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:38,598 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:43,247 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3427, 2.6901, 2.1556, 3.8021, 1.9151, 1.8810, 2.2938, 2.7707], + device='cuda:2'), covar=tensor([0.0862, 0.0957, 0.1133, 0.0313, 0.1217, 0.1522, 0.1222, 0.0875], + device='cuda:2'), in_proj_covar=tensor([0.0247, 0.0227, 0.0268, 0.0215, 0.0224, 0.0263, 0.0264, 0.0231], + 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') +2023-02-06 06:27:51,130 INFO [train.py:901] (2/4) Epoch 8, batch 7800, loss[loss=0.2571, simple_loss=0.3317, pruned_loss=0.09119, over 8477.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3253, pruned_loss=0.09134, over 1618714.09 frames. ], batch size: 27, lr: 9.30e-03, grad_scale: 8.0 +2023-02-06 06:27:53,240 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:58,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:28:25,437 INFO [train.py:901] (2/4) Epoch 8, batch 7850, loss[loss=0.3021, simple_loss=0.3703, pruned_loss=0.117, over 8242.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3252, pruned_loss=0.09157, over 1618654.77 frames. ], batch size: 24, lr: 9.30e-03, grad_scale: 8.0 +2023-02-06 06:28:30,088 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.075e+02 2.873e+02 3.519e+02 4.505e+02 1.254e+03, threshold=7.037e+02, percent-clipped=6.0 +2023-02-06 06:28:58,104 INFO [train.py:901] (2/4) Epoch 8, batch 7900, loss[loss=0.2499, simple_loss=0.3099, pruned_loss=0.09498, over 7435.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3256, pruned_loss=0.09183, over 1618914.53 frames. ], batch size: 17, lr: 9.30e-03, grad_scale: 8.0 +2023-02-06 06:28:58,915 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64483.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:29:10,198 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64500.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:29:32,361 INFO [train.py:901] (2/4) Epoch 8, batch 7950, loss[loss=0.2376, simple_loss=0.3054, pruned_loss=0.08493, over 7432.00 frames. ], tot_loss[loss=0.254, simple_loss=0.325, pruned_loss=0.0915, over 1617812.45 frames. ], batch size: 17, lr: 9.29e-03, grad_scale: 8.0 +2023-02-06 06:29:37,072 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.689e+02 3.383e+02 4.341e+02 8.251e+02, threshold=6.766e+02, percent-clipped=4.0 +2023-02-06 06:29:40,085 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:29:56,718 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64568.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:03,508 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64578.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:06,037 INFO [train.py:901] (2/4) Epoch 8, batch 8000, loss[loss=0.227, simple_loss=0.2932, pruned_loss=0.08041, over 8055.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3247, pruned_loss=0.09174, over 1616260.31 frames. ], batch size: 20, lr: 9.29e-03, grad_scale: 8.0 +2023-02-06 06:30:10,408 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2791, 1.5222, 2.2690, 1.0857, 1.7575, 1.4988, 1.4004, 1.5451], + device='cuda:2'), covar=tensor([0.1291, 0.1623, 0.0553, 0.2863, 0.1167, 0.2004, 0.1351, 0.1659], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0480, 0.0518, 0.0554, 0.0594, 0.0529, 0.0453, 0.0591], + 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') +2023-02-06 06:30:14,259 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:20,519 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:23,054 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:23,112 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6295, 2.9854, 2.0256, 2.3658, 2.3702, 1.5247, 2.3146, 2.3991], + device='cuda:2'), covar=tensor([0.1356, 0.0354, 0.0956, 0.0618, 0.0713, 0.1358, 0.0890, 0.0813], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0235, 0.0308, 0.0293, 0.0303, 0.0318, 0.0336, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 06:30:39,930 INFO [train.py:901] (2/4) Epoch 8, batch 8050, loss[loss=0.2185, simple_loss=0.2926, pruned_loss=0.07221, over 7942.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3235, pruned_loss=0.0914, over 1604067.36 frames. ], batch size: 20, lr: 9.29e-03, grad_scale: 8.0 +2023-02-06 06:30:44,638 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.434e+02 2.955e+02 3.616e+02 6.730e+02, threshold=5.909e+02, percent-clipped=0.0 +2023-02-06 06:30:51,579 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:13,129 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 06:31:17,613 INFO [train.py:901] (2/4) Epoch 9, batch 0, loss[loss=0.2864, simple_loss=0.3456, pruned_loss=0.1136, over 8246.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3456, pruned_loss=0.1136, over 8246.00 frames. ], batch size: 24, lr: 8.79e-03, grad_scale: 8.0 +2023-02-06 06:31:17,613 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 06:31:28,851 INFO [train.py:935] (2/4) Epoch 9, validation: loss=0.1983, simple_loss=0.2974, pruned_loss=0.04961, over 944034.00 frames. +2023-02-06 06:31:28,852 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 06:31:29,665 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:35,196 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:35,991 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2023-02-06 06:31:43,422 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 06:31:56,868 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64707.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:58,428 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:02,442 INFO [train.py:901] (2/4) Epoch 9, batch 50, loss[loss=0.2364, simple_loss=0.3133, pruned_loss=0.07972, over 8321.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3281, pruned_loss=0.09302, over 368190.11 frames. ], batch size: 25, lr: 8.79e-03, grad_scale: 8.0 +2023-02-06 06:32:06,625 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64721.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:07,942 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4177, 1.8923, 3.2332, 1.1299, 2.3346, 1.7131, 1.7069, 1.8614], + device='cuda:2'), covar=tensor([0.2025, 0.2260, 0.0848, 0.4242, 0.1737, 0.3124, 0.1877, 0.2746], + device='cuda:2'), in_proj_covar=tensor([0.0475, 0.0482, 0.0522, 0.0562, 0.0595, 0.0532, 0.0457, 0.0595], + 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') +2023-02-06 06:32:16,330 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 06:32:18,980 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.818e+02 3.347e+02 4.122e+02 1.189e+03, threshold=6.695e+02, percent-clipped=9.0 +2023-02-06 06:32:30,209 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4947, 1.9379, 2.0774, 1.2049, 2.1982, 1.3041, 0.6339, 1.7030], + device='cuda:2'), covar=tensor([0.0371, 0.0190, 0.0148, 0.0341, 0.0208, 0.0617, 0.0505, 0.0185], + device='cuda:2'), in_proj_covar=tensor([0.0372, 0.0288, 0.0239, 0.0353, 0.0281, 0.0447, 0.0340, 0.0324], + device='cuda:2'), out_proj_covar=tensor([1.1223e-04, 8.4071e-05, 7.0473e-05, 1.0408e-04, 8.4099e-05, 1.4525e-04, + 1.0258e-04, 9.6760e-05], device='cuda:2') +2023-02-06 06:32:31,576 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:36,073 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64763.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:37,300 INFO [train.py:901] (2/4) Epoch 9, batch 100, loss[loss=0.2468, simple_loss=0.3297, pruned_loss=0.08195, over 8452.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3236, pruned_loss=0.0899, over 646475.57 frames. ], batch size: 25, lr: 8.78e-03, grad_scale: 8.0 +2023-02-06 06:32:41,536 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64770.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:42,074 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 06:32:49,740 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64781.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:33:11,714 INFO [train.py:901] (2/4) Epoch 9, batch 150, loss[loss=0.262, simple_loss=0.3116, pruned_loss=0.1061, over 8227.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3231, pruned_loss=0.0892, over 865017.51 frames. ], batch size: 22, lr: 8.78e-03, grad_scale: 8.0 +2023-02-06 06:33:16,751 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:33:20,060 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:33:27,774 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.577e+02 3.213e+02 3.848e+02 9.281e+02, threshold=6.425e+02, percent-clipped=3.0 +2023-02-06 06:33:45,632 INFO [train.py:901] (2/4) Epoch 9, batch 200, loss[loss=0.2694, simple_loss=0.3362, pruned_loss=0.1012, over 7810.00 frames. ], tot_loss[loss=0.25, simple_loss=0.322, pruned_loss=0.089, over 1026989.43 frames. ], batch size: 20, lr: 8.78e-03, grad_scale: 8.0 +2023-02-06 06:34:21,132 INFO [train.py:901] (2/4) Epoch 9, batch 250, loss[loss=0.2579, simple_loss=0.3146, pruned_loss=0.1006, over 7922.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3204, pruned_loss=0.08764, over 1155514.51 frames. ], batch size: 20, lr: 8.77e-03, grad_scale: 8.0 +2023-02-06 06:34:34,285 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 06:34:36,819 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 2.841e+02 3.295e+02 4.179e+02 1.029e+03, threshold=6.590e+02, percent-clipped=5.0 +2023-02-06 06:34:39,018 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64942.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:34:42,840 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 06:34:44,923 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:34:54,260 INFO [train.py:901] (2/4) Epoch 9, batch 300, loss[loss=0.2777, simple_loss=0.3464, pruned_loss=0.1045, over 8547.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3209, pruned_loss=0.0881, over 1257433.34 frames. ], batch size: 49, lr: 8.77e-03, grad_scale: 8.0 +2023-02-06 06:34:54,475 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:11,901 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:15,056 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64994.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:26,440 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65010.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:30,328 INFO [train.py:901] (2/4) Epoch 9, batch 350, loss[loss=0.2741, simple_loss=0.348, pruned_loss=0.1001, over 8033.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3197, pruned_loss=0.08826, over 1333077.91 frames. ], batch size: 22, lr: 8.77e-03, grad_scale: 8.0 +2023-02-06 06:35:31,387 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 06:35:46,488 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.570e+02 3.183e+02 3.796e+02 1.000e+03, threshold=6.367e+02, percent-clipped=4.0 +2023-02-06 06:36:03,871 INFO [train.py:901] (2/4) Epoch 9, batch 400, loss[loss=0.1848, simple_loss=0.2607, pruned_loss=0.05449, over 7424.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3205, pruned_loss=0.08869, over 1395368.89 frames. ], batch size: 17, lr: 8.76e-03, grad_scale: 8.0 +2023-02-06 06:36:03,965 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65065.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:04,748 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:09,406 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65073.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:12,870 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:30,516 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:33,062 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:37,775 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65114.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:38,389 INFO [train.py:901] (2/4) Epoch 9, batch 450, loss[loss=0.2343, simple_loss=0.3171, pruned_loss=0.07577, over 8508.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3218, pruned_loss=0.08902, over 1448163.03 frames. ], batch size: 26, lr: 8.76e-03, grad_scale: 8.0 +2023-02-06 06:36:46,195 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:56,307 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.705e+02 3.323e+02 3.920e+02 9.407e+02, threshold=6.647e+02, percent-clipped=6.0 +2023-02-06 06:37:13,469 INFO [train.py:901] (2/4) Epoch 9, batch 500, loss[loss=0.253, simple_loss=0.3265, pruned_loss=0.08976, over 8590.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3229, pruned_loss=0.08986, over 1486817.35 frames. ], batch size: 31, lr: 8.76e-03, grad_scale: 8.0 +2023-02-06 06:37:16,356 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1731, 2.0836, 1.1600, 2.9116, 1.3383, 1.0831, 2.0821, 2.1544], + device='cuda:2'), covar=tensor([0.2015, 0.1506, 0.2579, 0.0536, 0.1640, 0.2669, 0.1438, 0.1182], + device='cuda:2'), in_proj_covar=tensor([0.0246, 0.0225, 0.0269, 0.0217, 0.0225, 0.0260, 0.0268, 0.0230], + 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') +2023-02-06 06:37:23,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:35,047 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65198.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:45,912 INFO [train.py:901] (2/4) Epoch 9, batch 550, loss[loss=0.2586, simple_loss=0.3336, pruned_loss=0.09179, over 8482.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3232, pruned_loss=0.08983, over 1516932.15 frames. ], batch size: 28, lr: 8.75e-03, grad_scale: 8.0 +2023-02-06 06:37:51,952 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65222.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:52,666 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65223.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:56,718 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:38:03,161 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.829e+02 3.496e+02 4.355e+02 8.306e+02, threshold=6.991e+02, percent-clipped=2.0 +2023-02-06 06:38:21,261 INFO [train.py:901] (2/4) Epoch 9, batch 600, loss[loss=0.2802, simple_loss=0.3528, pruned_loss=0.1038, over 8545.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3231, pruned_loss=0.08969, over 1543413.61 frames. ], batch size: 31, lr: 8.75e-03, grad_scale: 8.0 +2023-02-06 06:38:38,362 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 06:38:46,522 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65303.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:38:52,418 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3713, 4.3304, 3.8909, 1.9226, 3.8654, 3.9461, 4.1259, 3.5729], + device='cuda:2'), covar=tensor([0.0693, 0.0532, 0.0907, 0.4089, 0.0768, 0.0793, 0.0939, 0.0883], + device='cuda:2'), in_proj_covar=tensor([0.0435, 0.0348, 0.0364, 0.0454, 0.0357, 0.0338, 0.0352, 0.0303], + 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') +2023-02-06 06:38:54,302 INFO [train.py:901] (2/4) Epoch 9, batch 650, loss[loss=0.2736, simple_loss=0.3493, pruned_loss=0.09892, over 8504.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3235, pruned_loss=0.08994, over 1558642.58 frames. ], batch size: 26, lr: 8.75e-03, grad_scale: 16.0 +2023-02-06 06:38:59,116 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:10,303 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:10,881 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.656e+02 3.252e+02 4.080e+02 6.220e+02, threshold=6.503e+02, percent-clipped=0.0 +2023-02-06 06:39:12,455 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3842, 1.7999, 3.1274, 1.1279, 2.1999, 1.7876, 1.5501, 1.9503], + device='cuda:2'), covar=tensor([0.1685, 0.2083, 0.0614, 0.3727, 0.1430, 0.2629, 0.1682, 0.2074], + device='cuda:2'), in_proj_covar=tensor([0.0476, 0.0490, 0.0525, 0.0565, 0.0600, 0.0537, 0.0461, 0.0600], + 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') +2023-02-06 06:39:17,065 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:19,635 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2119, 3.1107, 2.8271, 1.4182, 2.8405, 2.8521, 2.8302, 2.6386], + device='cuda:2'), covar=tensor([0.1228, 0.0930, 0.1371, 0.4630, 0.1139, 0.1285, 0.1558, 0.1174], + device='cuda:2'), in_proj_covar=tensor([0.0433, 0.0346, 0.0362, 0.0453, 0.0356, 0.0337, 0.0350, 0.0301], + 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') +2023-02-06 06:39:23,679 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8310, 5.8742, 5.0180, 2.3651, 5.1505, 5.4641, 5.3693, 5.0368], + device='cuda:2'), covar=tensor([0.0481, 0.0363, 0.0741, 0.4013, 0.0611, 0.0567, 0.0867, 0.0718], + device='cuda:2'), in_proj_covar=tensor([0.0432, 0.0345, 0.0361, 0.0451, 0.0355, 0.0336, 0.0349, 0.0300], + 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') +2023-02-06 06:39:29,082 INFO [train.py:901] (2/4) Epoch 9, batch 700, loss[loss=0.2118, simple_loss=0.277, pruned_loss=0.07332, over 7435.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3217, pruned_loss=0.08954, over 1570570.90 frames. ], batch size: 17, lr: 8.74e-03, grad_scale: 16.0 +2023-02-06 06:39:41,001 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65381.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:57,981 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:03,865 INFO [train.py:901] (2/4) Epoch 9, batch 750, loss[loss=0.2101, simple_loss=0.2959, pruned_loss=0.06215, over 8298.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.321, pruned_loss=0.0891, over 1575973.21 frames. ], batch size: 23, lr: 8.74e-03, grad_scale: 16.0 +2023-02-06 06:40:05,346 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:18,167 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65436.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:19,958 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.803e+02 3.527e+02 4.474e+02 1.505e+03, threshold=7.053e+02, percent-clipped=7.0 +2023-02-06 06:40:21,301 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 06:40:30,030 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 06:40:30,193 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65453.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:30,516 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 06:40:36,165 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65461.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:38,589 INFO [train.py:901] (2/4) Epoch 9, batch 800, loss[loss=0.2338, simple_loss=0.2981, pruned_loss=0.08478, over 7243.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3205, pruned_loss=0.0892, over 1578895.65 frames. ], batch size: 16, lr: 8.74e-03, grad_scale: 16.0 +2023-02-06 06:40:47,458 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:53,606 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65485.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:58,222 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0564, 2.4151, 1.7988, 2.9540, 1.3288, 1.4251, 1.9866, 2.4502], + device='cuda:2'), covar=tensor([0.0885, 0.0946, 0.1132, 0.0436, 0.1433, 0.1894, 0.1248, 0.0802], + device='cuda:2'), in_proj_covar=tensor([0.0247, 0.0226, 0.0270, 0.0219, 0.0226, 0.0263, 0.0266, 0.0228], + 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') +2023-02-06 06:41:05,677 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65503.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:09,189 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3557, 1.8951, 2.8855, 2.2326, 2.3730, 2.1804, 1.6497, 1.1426], + device='cuda:2'), covar=tensor([0.3073, 0.3073, 0.0781, 0.1945, 0.1622, 0.1655, 0.1442, 0.3278], + device='cuda:2'), in_proj_covar=tensor([0.0855, 0.0810, 0.0700, 0.0803, 0.0895, 0.0751, 0.0685, 0.0729], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 06:41:10,502 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:13,593 INFO [train.py:901] (2/4) Epoch 9, batch 850, loss[loss=0.2899, simple_loss=0.3504, pruned_loss=0.1147, over 8318.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3215, pruned_loss=0.08918, over 1590030.98 frames. ], batch size: 25, lr: 8.73e-03, grad_scale: 16.0 +2023-02-06 06:41:17,141 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0230, 1.2967, 1.2050, 0.4624, 1.2691, 1.0913, 0.0963, 1.1817], + device='cuda:2'), covar=tensor([0.0165, 0.0129, 0.0125, 0.0235, 0.0159, 0.0371, 0.0345, 0.0120], + device='cuda:2'), in_proj_covar=tensor([0.0365, 0.0290, 0.0241, 0.0350, 0.0282, 0.0447, 0.0336, 0.0319], + device='cuda:2'), out_proj_covar=tensor([1.0988e-04, 8.4788e-05, 7.1140e-05, 1.0286e-04, 8.4558e-05, 1.4460e-04, + 1.0110e-04, 9.5098e-05], device='cuda:2') +2023-02-06 06:41:25,164 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:29,557 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.736e+02 3.271e+02 4.209e+02 1.110e+03, threshold=6.542e+02, percent-clipped=5.0 +2023-02-06 06:41:37,089 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65550.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:41:47,587 INFO [train.py:901] (2/4) Epoch 9, batch 900, loss[loss=0.1947, simple_loss=0.2808, pruned_loss=0.05423, over 8457.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3226, pruned_loss=0.09015, over 1594618.08 frames. ], batch size: 27, lr: 8.73e-03, grad_scale: 16.0 +2023-02-06 06:42:12,162 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3553, 2.5188, 1.7072, 2.0124, 2.0481, 1.3276, 1.7521, 1.7531], + device='cuda:2'), covar=tensor([0.1376, 0.0353, 0.0952, 0.0558, 0.0587, 0.1359, 0.0896, 0.0915], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0237, 0.0312, 0.0300, 0.0304, 0.0320, 0.0341, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 06:42:23,181 INFO [train.py:901] (2/4) Epoch 9, batch 950, loss[loss=0.2024, simple_loss=0.2812, pruned_loss=0.06173, over 7538.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.323, pruned_loss=0.09016, over 1600373.44 frames. ], batch size: 18, lr: 8.73e-03, grad_scale: 16.0 +2023-02-06 06:42:39,187 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.498e+02 3.047e+02 4.041e+02 6.463e+02, threshold=6.094e+02, percent-clipped=0.0 +2023-02-06 06:42:44,699 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:42:50,379 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 06:42:56,309 INFO [train.py:901] (2/4) Epoch 9, batch 1000, loss[loss=0.2121, simple_loss=0.2982, pruned_loss=0.06303, over 8243.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3242, pruned_loss=0.09072, over 1607976.85 frames. ], batch size: 22, lr: 8.72e-03, grad_scale: 16.0 +2023-02-06 06:43:20,517 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9538, 4.1683, 2.2899, 2.7265, 2.8202, 1.8609, 2.7342, 2.8736], + device='cuda:2'), covar=tensor([0.1330, 0.0173, 0.0839, 0.0686, 0.0616, 0.1230, 0.0928, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0235, 0.0314, 0.0301, 0.0307, 0.0321, 0.0342, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 06:43:23,105 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 06:43:27,432 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:43:31,857 INFO [train.py:901] (2/4) Epoch 9, batch 1050, loss[loss=0.2621, simple_loss=0.3431, pruned_loss=0.09058, over 8605.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.324, pruned_loss=0.09091, over 1604331.94 frames. ], batch size: 34, lr: 8.72e-03, grad_scale: 16.0 +2023-02-06 06:43:35,713 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 06:43:44,970 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65734.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:43:47,995 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.877e+02 3.398e+02 4.338e+02 8.070e+02, threshold=6.796e+02, percent-clipped=6.0 +2023-02-06 06:43:57,422 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3558, 1.8814, 1.3291, 2.7754, 1.2263, 1.0940, 1.9238, 2.1081], + device='cuda:2'), covar=tensor([0.1779, 0.1369, 0.2299, 0.0507, 0.1602, 0.2462, 0.1173, 0.1046], + device='cuda:2'), in_proj_covar=tensor([0.0245, 0.0223, 0.0268, 0.0217, 0.0225, 0.0262, 0.0265, 0.0229], + 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') +2023-02-06 06:44:03,491 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65762.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:44:05,257 INFO [train.py:901] (2/4) Epoch 9, batch 1100, loss[loss=0.2348, simple_loss=0.307, pruned_loss=0.08132, over 8138.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3252, pruned_loss=0.09189, over 1609061.68 frames. ], batch size: 22, lr: 8.72e-03, grad_scale: 16.0 +2023-02-06 06:44:20,622 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:44:38,692 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:44:39,893 INFO [train.py:901] (2/4) Epoch 9, batch 1150, loss[loss=0.2431, simple_loss=0.3246, pruned_loss=0.08076, over 8522.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3235, pruned_loss=0.09082, over 1607043.49 frames. ], batch size: 28, lr: 8.71e-03, grad_scale: 16.0 +2023-02-06 06:44:44,631 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 06:44:56,767 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.552e+02 3.121e+02 3.966e+02 8.304e+02, threshold=6.242e+02, percent-clipped=2.0 +2023-02-06 06:45:09,006 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65856.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:45:14,589 INFO [train.py:901] (2/4) Epoch 9, batch 1200, loss[loss=0.2942, simple_loss=0.3383, pruned_loss=0.125, over 7921.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3225, pruned_loss=0.08996, over 1605548.22 frames. ], batch size: 20, lr: 8.71e-03, grad_scale: 16.0 +2023-02-06 06:45:33,741 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65894.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:45:47,641 INFO [train.py:901] (2/4) Epoch 9, batch 1250, loss[loss=0.2642, simple_loss=0.3322, pruned_loss=0.09811, over 7933.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.324, pruned_loss=0.0907, over 1608594.53 frames. ], batch size: 20, lr: 8.71e-03, grad_scale: 16.0 +2023-02-06 06:46:05,032 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 2.844e+02 3.477e+02 4.312e+02 8.167e+02, threshold=6.953e+02, percent-clipped=5.0 +2023-02-06 06:46:20,605 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8699, 1.8196, 2.5815, 1.3990, 2.1952, 2.8782, 2.7971, 2.5588], + device='cuda:2'), covar=tensor([0.0925, 0.1133, 0.0707, 0.1748, 0.1096, 0.0302, 0.0673, 0.0570], + device='cuda:2'), in_proj_covar=tensor([0.0253, 0.0284, 0.0244, 0.0275, 0.0259, 0.0227, 0.0301, 0.0289], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 06:46:23,824 INFO [train.py:901] (2/4) Epoch 9, batch 1300, loss[loss=0.2312, simple_loss=0.2956, pruned_loss=0.08339, over 7700.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3245, pruned_loss=0.09086, over 1611873.78 frames. ], batch size: 18, lr: 8.70e-03, grad_scale: 16.0 +2023-02-06 06:46:44,911 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 06:46:55,296 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66009.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:46:59,216 INFO [train.py:901] (2/4) Epoch 9, batch 1350, loss[loss=0.2756, simple_loss=0.3415, pruned_loss=0.1049, over 8549.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3226, pruned_loss=0.08943, over 1614014.68 frames. ], batch size: 31, lr: 8.70e-03, grad_scale: 8.0 +2023-02-06 06:47:01,386 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:47:03,664 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 06:47:17,555 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.557e+02 3.336e+02 4.233e+02 1.201e+03, threshold=6.672e+02, percent-clipped=8.0 +2023-02-06 06:47:19,777 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:47:35,101 INFO [train.py:901] (2/4) Epoch 9, batch 1400, loss[loss=0.283, simple_loss=0.3445, pruned_loss=0.1108, over 8635.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3224, pruned_loss=0.08962, over 1612468.63 frames. ], batch size: 34, lr: 8.70e-03, grad_scale: 8.0 +2023-02-06 06:48:09,442 INFO [train.py:901] (2/4) Epoch 9, batch 1450, loss[loss=0.2882, simple_loss=0.3601, pruned_loss=0.1082, over 8644.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3226, pruned_loss=0.09016, over 1611215.23 frames. ], batch size: 34, lr: 8.69e-03, grad_scale: 8.0 +2023-02-06 06:48:12,191 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 06:48:26,154 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.633e+02 3.463e+02 4.686e+02 9.003e+02, threshold=6.925e+02, percent-clipped=5.0 +2023-02-06 06:48:42,245 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:48:44,184 INFO [train.py:901] (2/4) Epoch 9, batch 1500, loss[loss=0.1987, simple_loss=0.2929, pruned_loss=0.05227, over 8459.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3211, pruned_loss=0.08902, over 1609533.31 frames. ], batch size: 25, lr: 8.69e-03, grad_scale: 8.0 +2023-02-06 06:48:52,883 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66178.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:49:08,955 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66200.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:49:12,623 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 06:49:18,740 INFO [train.py:901] (2/4) Epoch 9, batch 1550, loss[loss=0.2578, simple_loss=0.3461, pruned_loss=0.08481, over 8239.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3212, pruned_loss=0.08933, over 1608482.25 frames. ], batch size: 24, lr: 8.69e-03, grad_scale: 8.0 +2023-02-06 06:49:28,444 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9307, 1.4202, 1.4615, 1.1417, 1.0545, 1.3154, 1.6520, 1.5222], + device='cuda:2'), covar=tensor([0.0577, 0.1154, 0.1729, 0.1455, 0.0583, 0.1530, 0.0671, 0.0582], + device='cuda:2'), in_proj_covar=tensor([0.0107, 0.0158, 0.0198, 0.0163, 0.0109, 0.0167, 0.0122, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 06:49:35,629 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.559e+02 2.942e+02 3.565e+02 7.942e+02, threshold=5.885e+02, percent-clipped=2.0 +2023-02-06 06:49:36,160 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.50 vs. limit=5.0 +2023-02-06 06:49:53,203 INFO [train.py:901] (2/4) Epoch 9, batch 1600, loss[loss=0.1985, simple_loss=0.2917, pruned_loss=0.05266, over 8085.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3205, pruned_loss=0.08781, over 1614581.39 frames. ], batch size: 21, lr: 8.68e-03, grad_scale: 8.0 +2023-02-06 06:49:53,444 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66265.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:50:02,300 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:50:11,771 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66290.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:50:29,749 INFO [train.py:901] (2/4) Epoch 9, batch 1650, loss[loss=0.2761, simple_loss=0.343, pruned_loss=0.1046, over 8135.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3205, pruned_loss=0.08775, over 1612715.36 frames. ], batch size: 22, lr: 8.68e-03, grad_scale: 8.0 +2023-02-06 06:50:29,939 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66315.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:50:46,552 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.536e+02 3.360e+02 4.258e+02 7.701e+02, threshold=6.719e+02, percent-clipped=5.0 +2023-02-06 06:51:03,464 INFO [train.py:901] (2/4) Epoch 9, batch 1700, loss[loss=0.2693, simple_loss=0.3428, pruned_loss=0.09797, over 8296.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3208, pruned_loss=0.08829, over 1614815.84 frames. ], batch size: 23, lr: 8.68e-03, grad_scale: 8.0 +2023-02-06 06:51:39,896 INFO [train.py:901] (2/4) Epoch 9, batch 1750, loss[loss=0.2529, simple_loss=0.3345, pruned_loss=0.0856, over 8089.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3208, pruned_loss=0.08851, over 1617781.27 frames. ], batch size: 21, lr: 8.67e-03, grad_scale: 8.0 +2023-02-06 06:51:57,454 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.942e+02 3.542e+02 4.261e+02 7.419e+02, threshold=7.084e+02, percent-clipped=2.0 +2023-02-06 06:52:13,915 INFO [train.py:901] (2/4) Epoch 9, batch 1800, loss[loss=0.2309, simple_loss=0.3092, pruned_loss=0.07626, over 8284.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3223, pruned_loss=0.08978, over 1615426.93 frames. ], batch size: 23, lr: 8.67e-03, grad_scale: 8.0 +2023-02-06 06:52:42,186 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:52:49,000 INFO [train.py:901] (2/4) Epoch 9, batch 1850, loss[loss=0.2475, simple_loss=0.3242, pruned_loss=0.0854, over 6806.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3227, pruned_loss=0.09002, over 1617540.66 frames. ], batch size: 15, lr: 8.67e-03, grad_scale: 8.0 +2023-02-06 06:52:53,697 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66522.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:53:05,499 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66539.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:53:05,997 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.848e+02 3.228e+02 4.154e+02 1.120e+03, threshold=6.457e+02, percent-clipped=1.0 +2023-02-06 06:53:22,983 INFO [train.py:901] (2/4) Epoch 9, batch 1900, loss[loss=0.2519, simple_loss=0.3147, pruned_loss=0.09453, over 7658.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3227, pruned_loss=0.09061, over 1614695.23 frames. ], batch size: 19, lr: 8.66e-03, grad_scale: 8.0 +2023-02-06 06:53:27,130 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66571.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:53:43,851 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66596.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:53:47,185 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 06:53:56,566 INFO [train.py:901] (2/4) Epoch 9, batch 1950, loss[loss=0.2491, simple_loss=0.3223, pruned_loss=0.08795, over 8488.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3236, pruned_loss=0.09102, over 1614583.72 frames. ], batch size: 28, lr: 8.66e-03, grad_scale: 8.0 +2023-02-06 06:53:58,610 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 06:54:00,034 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:54:00,834 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:54:03,721 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 +2023-02-06 06:54:12,896 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66637.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:54:14,189 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8895, 1.9861, 1.7565, 2.5841, 1.1518, 1.3453, 1.8621, 2.0643], + device='cuda:2'), covar=tensor([0.0809, 0.0945, 0.1126, 0.0392, 0.1248, 0.1592, 0.0905, 0.0806], + device='cuda:2'), in_proj_covar=tensor([0.0245, 0.0224, 0.0266, 0.0217, 0.0222, 0.0260, 0.0262, 0.0226], + 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') +2023-02-06 06:54:14,627 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.852e+02 3.410e+02 4.369e+02 9.021e+02, threshold=6.820e+02, percent-clipped=7.0 +2023-02-06 06:54:20,063 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 06:54:32,103 INFO [train.py:901] (2/4) Epoch 9, batch 2000, loss[loss=0.2635, simple_loss=0.3273, pruned_loss=0.09988, over 8132.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3223, pruned_loss=0.09014, over 1613853.97 frames. ], batch size: 22, lr: 8.66e-03, grad_scale: 8.0 +2023-02-06 06:55:06,982 INFO [train.py:901] (2/4) Epoch 9, batch 2050, loss[loss=0.2578, simple_loss=0.325, pruned_loss=0.09528, over 8130.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3217, pruned_loss=0.08978, over 1611928.63 frames. ], batch size: 22, lr: 8.65e-03, grad_scale: 8.0 +2023-02-06 06:55:20,386 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66735.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:55:23,653 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.763e+02 3.349e+02 4.333e+02 1.017e+03, threshold=6.698e+02, percent-clipped=4.0 +2023-02-06 06:55:31,186 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66749.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:55:33,260 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5328, 2.0009, 2.2196, 1.1555, 2.2423, 1.4357, 0.6982, 1.7910], + device='cuda:2'), covar=tensor([0.0424, 0.0227, 0.0149, 0.0397, 0.0269, 0.0659, 0.0551, 0.0204], + device='cuda:2'), in_proj_covar=tensor([0.0370, 0.0298, 0.0244, 0.0360, 0.0288, 0.0455, 0.0341, 0.0328], + device='cuda:2'), out_proj_covar=tensor([1.1046e-04, 8.7178e-05, 7.1478e-05, 1.0569e-04, 8.5958e-05, 1.4669e-04, + 1.0234e-04, 9.7499e-05], device='cuda:2') +2023-02-06 06:55:42,431 INFO [train.py:901] (2/4) Epoch 9, batch 2100, loss[loss=0.248, simple_loss=0.3294, pruned_loss=0.08325, over 8236.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3221, pruned_loss=0.09047, over 1610407.45 frames. ], batch size: 22, lr: 8.65e-03, grad_scale: 8.0 +2023-02-06 06:56:17,428 INFO [train.py:901] (2/4) Epoch 9, batch 2150, loss[loss=0.2705, simple_loss=0.3499, pruned_loss=0.09556, over 8204.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3222, pruned_loss=0.0903, over 1610467.57 frames. ], batch size: 23, lr: 8.65e-03, grad_scale: 8.0 +2023-02-06 06:56:18,961 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1215, 1.7537, 3.4060, 1.4730, 2.4206, 3.8919, 3.8535, 3.3397], + device='cuda:2'), covar=tensor([0.0983, 0.1409, 0.0372, 0.1991, 0.0890, 0.0220, 0.0439, 0.0529], + device='cuda:2'), in_proj_covar=tensor([0.0252, 0.0284, 0.0244, 0.0276, 0.0259, 0.0227, 0.0302, 0.0287], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 06:56:34,545 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.810e+02 3.362e+02 4.511e+02 1.000e+03, threshold=6.724e+02, percent-clipped=7.0 +2023-02-06 06:56:53,076 INFO [train.py:901] (2/4) Epoch 9, batch 2200, loss[loss=0.2534, simple_loss=0.3169, pruned_loss=0.0949, over 7935.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3211, pruned_loss=0.08978, over 1612847.88 frames. ], batch size: 20, lr: 8.64e-03, grad_scale: 8.0 +2023-02-06 06:56:54,808 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.19 vs. limit=5.0 +2023-02-06 06:57:01,062 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:57:05,630 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66883.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:57:12,267 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66893.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:57:18,130 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:57:27,006 INFO [train.py:901] (2/4) Epoch 9, batch 2250, loss[loss=0.248, simple_loss=0.3218, pruned_loss=0.08713, over 8551.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3212, pruned_loss=0.08979, over 1614418.18 frames. ], batch size: 31, lr: 8.64e-03, grad_scale: 8.0 +2023-02-06 06:57:29,219 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66918.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:57:43,695 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.716e+02 3.375e+02 4.203e+02 7.579e+02, threshold=6.750e+02, percent-clipped=1.0 +2023-02-06 06:58:00,327 INFO [train.py:901] (2/4) Epoch 9, batch 2300, loss[loss=0.2147, simple_loss=0.3033, pruned_loss=0.06312, over 8249.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3216, pruned_loss=0.08956, over 1618902.59 frames. ], batch size: 24, lr: 8.64e-03, grad_scale: 8.0 +2023-02-06 06:58:07,510 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66975.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:58:19,737 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:58:24,436 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66998.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:58:36,110 INFO [train.py:901] (2/4) Epoch 9, batch 2350, loss[loss=0.2313, simple_loss=0.291, pruned_loss=0.08578, over 7518.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3208, pruned_loss=0.08948, over 1618322.77 frames. ], batch size: 18, lr: 8.63e-03, grad_scale: 8.0 +2023-02-06 06:58:36,992 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:58:53,558 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.755e+02 3.236e+02 4.430e+02 1.005e+03, threshold=6.472e+02, percent-clipped=3.0 +2023-02-06 06:59:10,046 INFO [train.py:901] (2/4) Epoch 9, batch 2400, loss[loss=0.2063, simple_loss=0.2692, pruned_loss=0.07166, over 7708.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3203, pruned_loss=0.08931, over 1612874.49 frames. ], batch size: 18, lr: 8.63e-03, grad_scale: 8.0 +2023-02-06 06:59:24,782 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 06:59:28,535 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67093.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:59:35,220 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67101.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:59:42,858 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 +2023-02-06 06:59:44,443 INFO [train.py:901] (2/4) Epoch 9, batch 2450, loss[loss=0.2848, simple_loss=0.3561, pruned_loss=0.1067, over 8243.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3218, pruned_loss=0.08984, over 1611793.71 frames. ], batch size: 22, lr: 8.63e-03, grad_scale: 8.0 +2023-02-06 06:59:45,250 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0111, 2.8577, 3.0234, 1.8771, 1.6229, 3.3222, 0.7374, 2.1397], + device='cuda:2'), covar=tensor([0.2290, 0.1535, 0.1241, 0.3929, 0.5650, 0.0588, 0.4659, 0.2537], + device='cuda:2'), in_proj_covar=tensor([0.0153, 0.0151, 0.0091, 0.0202, 0.0244, 0.0094, 0.0154, 0.0154], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 07:00:02,010 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.787e+02 3.467e+02 4.148e+02 8.119e+02, threshold=6.934e+02, percent-clipped=3.0 +2023-02-06 07:00:18,497 INFO [train.py:901] (2/4) Epoch 9, batch 2500, loss[loss=0.2469, simple_loss=0.3285, pruned_loss=0.08265, over 8457.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3229, pruned_loss=0.09095, over 1609966.06 frames. ], batch size: 27, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:00:48,179 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67208.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:00:52,870 INFO [train.py:901] (2/4) Epoch 9, batch 2550, loss[loss=0.2134, simple_loss=0.2977, pruned_loss=0.06451, over 8262.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3215, pruned_loss=0.08955, over 1610586.98 frames. ], batch size: 24, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:01:12,353 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.679e+02 3.405e+02 4.203e+02 8.726e+02, threshold=6.810e+02, percent-clipped=2.0 +2023-02-06 07:01:21,875 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67254.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:01:29,612 INFO [train.py:901] (2/4) Epoch 9, batch 2600, loss[loss=0.2164, simple_loss=0.2807, pruned_loss=0.07604, over 7208.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3217, pruned_loss=0.0902, over 1604429.97 frames. ], batch size: 16, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:01:39,248 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67279.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:02:03,433 INFO [train.py:901] (2/4) Epoch 9, batch 2650, loss[loss=0.2783, simple_loss=0.3444, pruned_loss=0.1061, over 8504.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3213, pruned_loss=0.08948, over 1610973.59 frames. ], batch size: 26, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:02:06,281 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:02:21,820 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.739e+02 3.376e+02 4.238e+02 9.756e+02, threshold=6.752e+02, percent-clipped=4.0 +2023-02-06 07:02:39,313 INFO [train.py:901] (2/4) Epoch 9, batch 2700, loss[loss=0.3144, simple_loss=0.3724, pruned_loss=0.1282, over 8514.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3229, pruned_loss=0.09002, over 1612140.60 frames. ], batch size: 26, lr: 8.61e-03, grad_scale: 8.0 +2023-02-06 07:02:44,769 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6958, 1.4418, 4.4528, 1.7402, 2.4683, 5.0218, 5.0064, 4.3462], + device='cuda:2'), covar=tensor([0.0936, 0.1611, 0.0257, 0.1950, 0.0919, 0.0198, 0.0368, 0.0555], + device='cuda:2'), in_proj_covar=tensor([0.0250, 0.0279, 0.0240, 0.0271, 0.0254, 0.0224, 0.0299, 0.0283], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:02:58,435 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7825, 1.8126, 2.1642, 1.5478, 1.0874, 2.1367, 0.3911, 1.3502], + device='cuda:2'), covar=tensor([0.2401, 0.1558, 0.0661, 0.2192, 0.4765, 0.0546, 0.3906, 0.2257], + device='cuda:2'), in_proj_covar=tensor([0.0155, 0.0154, 0.0092, 0.0204, 0.0243, 0.0095, 0.0153, 0.0154], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 07:03:06,866 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4414, 1.9754, 3.0538, 2.5508, 2.6019, 2.1713, 1.6493, 1.4689], + device='cuda:2'), covar=tensor([0.2939, 0.3391, 0.0928, 0.1848, 0.1610, 0.1750, 0.1507, 0.3436], + device='cuda:2'), in_proj_covar=tensor([0.0849, 0.0812, 0.0697, 0.0806, 0.0895, 0.0759, 0.0683, 0.0732], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 07:03:13,103 INFO [train.py:901] (2/4) Epoch 9, batch 2750, loss[loss=0.2897, simple_loss=0.3411, pruned_loss=0.1191, over 8086.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3229, pruned_loss=0.09031, over 1611842.52 frames. ], batch size: 21, lr: 8.61e-03, grad_scale: 8.0 +2023-02-06 07:03:26,081 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:03:29,957 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.867e+02 3.446e+02 4.196e+02 9.783e+02, threshold=6.892e+02, percent-clipped=3.0 +2023-02-06 07:03:34,268 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67445.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:03:38,822 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 07:03:48,010 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67464.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:03:48,379 INFO [train.py:901] (2/4) Epoch 9, batch 2800, loss[loss=0.3103, simple_loss=0.3773, pruned_loss=0.1217, over 8626.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3227, pruned_loss=0.09049, over 1612019.61 frames. ], batch size: 31, lr: 8.61e-03, grad_scale: 8.0 +2023-02-06 07:04:05,327 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67489.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:04:23,679 INFO [train.py:901] (2/4) Epoch 9, batch 2850, loss[loss=0.2568, simple_loss=0.3348, pruned_loss=0.08944, over 8591.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3228, pruned_loss=0.08996, over 1616263.34 frames. ], batch size: 34, lr: 8.60e-03, grad_scale: 8.0 +2023-02-06 07:04:34,544 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:04:40,478 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.765e+02 3.269e+02 4.105e+02 6.649e+02, threshold=6.538e+02, percent-clipped=0.0 +2023-02-06 07:04:53,911 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 07:04:55,077 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67560.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:04:58,190 INFO [train.py:901] (2/4) Epoch 9, batch 2900, loss[loss=0.2613, simple_loss=0.3312, pruned_loss=0.09571, over 8200.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3236, pruned_loss=0.0905, over 1615262.17 frames. ], batch size: 23, lr: 8.60e-03, grad_scale: 8.0 +2023-02-06 07:05:24,267 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 07:05:33,886 INFO [train.py:901] (2/4) Epoch 9, batch 2950, loss[loss=0.2432, simple_loss=0.3133, pruned_loss=0.08654, over 8036.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3238, pruned_loss=0.08998, over 1617887.66 frames. ], batch size: 22, lr: 8.60e-03, grad_scale: 8.0 +2023-02-06 07:05:43,266 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7254, 1.9925, 3.0174, 1.4270, 2.4763, 1.9490, 1.8415, 2.3278], + device='cuda:2'), covar=tensor([0.1270, 0.1772, 0.0560, 0.3163, 0.1067, 0.2067, 0.1231, 0.1651], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0498, 0.0530, 0.0572, 0.0605, 0.0541, 0.0462, 0.0608], + 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') +2023-02-06 07:05:51,259 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.115e+02 2.827e+02 3.390e+02 4.435e+02 7.404e+02, threshold=6.780e+02, percent-clipped=4.0 +2023-02-06 07:06:08,216 INFO [train.py:901] (2/4) Epoch 9, batch 3000, loss[loss=0.2519, simple_loss=0.3226, pruned_loss=0.09064, over 8475.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3239, pruned_loss=0.09003, over 1619202.80 frames. ], batch size: 27, lr: 8.59e-03, grad_scale: 8.0 +2023-02-06 07:06:08,217 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 07:06:13,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7210, 1.6851, 2.6921, 1.3006, 2.1365, 2.8772, 2.8988, 2.5171], + device='cuda:2'), covar=tensor([0.1139, 0.1409, 0.0419, 0.2272, 0.0837, 0.0363, 0.0560, 0.0715], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0285, 0.0249, 0.0279, 0.0262, 0.0230, 0.0309, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:06:20,342 INFO [train.py:935] (2/4) Epoch 9, validation: loss=0.1965, simple_loss=0.2957, pruned_loss=0.04864, over 944034.00 frames. +2023-02-06 07:06:20,342 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6620MB +2023-02-06 07:06:37,436 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67690.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:06:43,347 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:06:52,182 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67710.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:06:55,496 INFO [train.py:901] (2/4) Epoch 9, batch 3050, loss[loss=0.2325, simple_loss=0.3063, pruned_loss=0.07939, over 7644.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.323, pruned_loss=0.0897, over 1621148.29 frames. ], batch size: 19, lr: 8.59e-03, grad_scale: 8.0 +2023-02-06 07:06:55,703 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:07:00,567 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2138, 1.7956, 1.9678, 2.0391, 1.3489, 1.7572, 2.2569, 2.3910], + device='cuda:2'), covar=tensor([0.0385, 0.1171, 0.1616, 0.1141, 0.0542, 0.1446, 0.0593, 0.0502], + device='cuda:2'), in_proj_covar=tensor([0.0105, 0.0158, 0.0196, 0.0161, 0.0108, 0.0166, 0.0121, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 07:07:13,238 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.629e+02 3.194e+02 3.976e+02 7.575e+02, threshold=6.387e+02, percent-clipped=1.0 +2023-02-06 07:07:29,777 INFO [train.py:901] (2/4) Epoch 9, batch 3100, loss[loss=0.2483, simple_loss=0.3333, pruned_loss=0.08165, over 8442.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3226, pruned_loss=0.08884, over 1622251.00 frames. ], batch size: 27, lr: 8.59e-03, grad_scale: 8.0 +2023-02-06 07:07:57,812 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3481, 1.4191, 1.5900, 1.3099, 1.0479, 1.4054, 1.7320, 1.5747], + device='cuda:2'), covar=tensor([0.0485, 0.1279, 0.1647, 0.1403, 0.0618, 0.1466, 0.0742, 0.0611], + device='cuda:2'), in_proj_covar=tensor([0.0106, 0.0159, 0.0197, 0.0162, 0.0109, 0.0167, 0.0121, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 07:08:04,177 INFO [train.py:901] (2/4) Epoch 9, batch 3150, loss[loss=0.2418, simple_loss=0.3163, pruned_loss=0.08358, over 8324.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3232, pruned_loss=0.08978, over 1618958.35 frames. ], batch size: 26, lr: 8.58e-03, grad_scale: 8.0 +2023-02-06 07:08:05,055 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67816.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:08:21,146 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.202e+02 2.768e+02 3.401e+02 4.235e+02 8.418e+02, threshold=6.801e+02, percent-clipped=5.0 +2023-02-06 07:08:22,002 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:08:37,988 INFO [train.py:901] (2/4) Epoch 9, batch 3200, loss[loss=0.2352, simple_loss=0.3152, pruned_loss=0.07761, over 8518.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3218, pruned_loss=0.08937, over 1614162.52 frames. ], batch size: 26, lr: 8.58e-03, grad_scale: 8.0 +2023-02-06 07:08:38,989 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 07:08:44,735 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:08:47,156 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 07:09:12,192 INFO [train.py:901] (2/4) Epoch 9, batch 3250, loss[loss=0.3024, simple_loss=0.3698, pruned_loss=0.1175, over 8327.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3206, pruned_loss=0.08843, over 1611958.55 frames. ], batch size: 25, lr: 8.58e-03, grad_scale: 8.0 +2023-02-06 07:09:14,080 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 07:09:29,469 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.709e+02 3.362e+02 4.203e+02 8.128e+02, threshold=6.724e+02, percent-clipped=5.0 +2023-02-06 07:09:31,090 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3854, 1.8576, 2.7673, 1.1726, 2.0203, 1.7521, 1.5146, 1.9068], + device='cuda:2'), covar=tensor([0.1733, 0.2012, 0.0766, 0.3678, 0.1469, 0.2849, 0.1708, 0.2017], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0494, 0.0527, 0.0566, 0.0601, 0.0540, 0.0460, 0.0603], + 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') +2023-02-06 07:09:46,665 INFO [train.py:901] (2/4) Epoch 9, batch 3300, loss[loss=0.2056, simple_loss=0.2821, pruned_loss=0.06455, over 7554.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3224, pruned_loss=0.08973, over 1609215.10 frames. ], batch size: 18, lr: 8.57e-03, grad_scale: 8.0 +2023-02-06 07:10:04,401 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:10,480 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:22,535 INFO [train.py:901] (2/4) Epoch 9, batch 3350, loss[loss=0.2816, simple_loss=0.3403, pruned_loss=0.1114, over 8295.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3227, pruned_loss=0.08972, over 1610881.84 frames. ], batch size: 23, lr: 8.57e-03, grad_scale: 16.0 +2023-02-06 07:10:39,236 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.493e+02 3.108e+02 4.287e+02 1.101e+03, threshold=6.217e+02, percent-clipped=5.0 +2023-02-06 07:10:40,597 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:49,105 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:56,399 INFO [train.py:901] (2/4) Epoch 9, batch 3400, loss[loss=0.2558, simple_loss=0.3382, pruned_loss=0.08674, over 8109.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3213, pruned_loss=0.08847, over 1611127.95 frames. ], batch size: 23, lr: 8.57e-03, grad_scale: 16.0 +2023-02-06 07:11:24,319 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68105.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:11:30,825 INFO [train.py:901] (2/4) Epoch 9, batch 3450, loss[loss=0.1809, simple_loss=0.261, pruned_loss=0.05042, over 7793.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3215, pruned_loss=0.08923, over 1610741.90 frames. ], batch size: 19, lr: 8.56e-03, grad_scale: 16.0 +2023-02-06 07:11:46,279 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1837, 2.4158, 1.9294, 2.8828, 1.5301, 1.5155, 1.9721, 2.4968], + device='cuda:2'), covar=tensor([0.0739, 0.0863, 0.1102, 0.0412, 0.1191, 0.1549, 0.1175, 0.0804], + device='cuda:2'), in_proj_covar=tensor([0.0244, 0.0225, 0.0263, 0.0219, 0.0223, 0.0262, 0.0266, 0.0227], + 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') +2023-02-06 07:11:48,140 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.585e+02 3.242e+02 3.955e+02 1.617e+03, threshold=6.484e+02, percent-clipped=7.0 +2023-02-06 07:11:59,840 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:12:05,798 INFO [train.py:901] (2/4) Epoch 9, batch 3500, loss[loss=0.276, simple_loss=0.3439, pruned_loss=0.104, over 7916.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3215, pruned_loss=0.08931, over 1611611.65 frames. ], batch size: 20, lr: 8.56e-03, grad_scale: 8.0 +2023-02-06 07:12:08,692 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:12:17,930 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 07:12:40,952 INFO [train.py:901] (2/4) Epoch 9, batch 3550, loss[loss=0.2315, simple_loss=0.2981, pruned_loss=0.0825, over 5144.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3209, pruned_loss=0.08873, over 1608587.27 frames. ], batch size: 11, lr: 8.56e-03, grad_scale: 8.0 +2023-02-06 07:12:58,945 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.838e+02 3.387e+02 4.304e+02 7.616e+02, threshold=6.774e+02, percent-clipped=6.0 +2023-02-06 07:13:02,582 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68246.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:13:14,901 INFO [train.py:901] (2/4) Epoch 9, batch 3600, loss[loss=0.2978, simple_loss=0.3662, pruned_loss=0.1147, over 8187.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3208, pruned_loss=0.0884, over 1608504.40 frames. ], batch size: 23, lr: 8.56e-03, grad_scale: 8.0 +2023-02-06 07:13:19,106 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:13:32,120 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.61 vs. limit=5.0 +2023-02-06 07:13:49,586 INFO [train.py:901] (2/4) Epoch 9, batch 3650, loss[loss=0.2831, simple_loss=0.359, pruned_loss=0.1037, over 8522.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3208, pruned_loss=0.08842, over 1605101.92 frames. ], batch size: 28, lr: 8.55e-03, grad_scale: 8.0 +2023-02-06 07:14:08,230 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.637e+02 3.214e+02 4.100e+02 7.421e+02, threshold=6.428e+02, percent-clipped=2.0 +2023-02-06 07:14:09,680 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:14:18,267 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 07:14:25,009 INFO [train.py:901] (2/4) Epoch 9, batch 3700, loss[loss=0.213, simple_loss=0.2818, pruned_loss=0.07205, over 7177.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3189, pruned_loss=0.08736, over 1604204.11 frames. ], batch size: 16, lr: 8.55e-03, grad_scale: 8.0 +2023-02-06 07:14:44,968 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6993, 1.7348, 1.6391, 1.4516, 0.9560, 1.5259, 1.6293, 1.8331], + device='cuda:2'), covar=tensor([0.0569, 0.1053, 0.1597, 0.1231, 0.0560, 0.1406, 0.0628, 0.0482], + device='cuda:2'), in_proj_covar=tensor([0.0106, 0.0157, 0.0195, 0.0159, 0.0108, 0.0165, 0.0119, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 07:14:57,600 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:14:58,733 INFO [train.py:901] (2/4) Epoch 9, batch 3750, loss[loss=0.2481, simple_loss=0.3192, pruned_loss=0.08846, over 8238.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3195, pruned_loss=0.08782, over 1610390.68 frames. ], batch size: 24, lr: 8.55e-03, grad_scale: 8.0 +2023-02-06 07:15:00,178 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:06,058 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68425.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:14,905 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:16,796 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.868e+02 3.639e+02 4.960e+02 1.282e+03, threshold=7.278e+02, percent-clipped=8.0 +2023-02-06 07:15:22,898 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68449.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:15:23,629 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:29,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:33,817 INFO [train.py:901] (2/4) Epoch 9, batch 3800, loss[loss=0.257, simple_loss=0.3312, pruned_loss=0.09135, over 8447.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.319, pruned_loss=0.08734, over 1612583.68 frames. ], batch size: 49, lr: 8.54e-03, grad_scale: 8.0 +2023-02-06 07:15:37,777 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-06 07:16:07,818 INFO [train.py:901] (2/4) Epoch 9, batch 3850, loss[loss=0.2039, simple_loss=0.2819, pruned_loss=0.06292, over 7652.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3177, pruned_loss=0.08659, over 1607369.04 frames. ], batch size: 19, lr: 8.54e-03, grad_scale: 8.0 +2023-02-06 07:16:24,983 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 07:16:25,653 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.582e+02 3.048e+02 3.724e+02 6.674e+02, threshold=6.096e+02, percent-clipped=0.0 +2023-02-06 07:16:42,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68564.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:16:42,695 INFO [train.py:901] (2/4) Epoch 9, batch 3900, loss[loss=0.272, simple_loss=0.3375, pruned_loss=0.1033, over 8133.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3171, pruned_loss=0.08559, over 1612048.63 frames. ], batch size: 22, lr: 8.54e-03, grad_scale: 8.0 +2023-02-06 07:17:04,305 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68596.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:17:17,691 INFO [train.py:901] (2/4) Epoch 9, batch 3950, loss[loss=0.2488, simple_loss=0.3204, pruned_loss=0.08864, over 8246.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3163, pruned_loss=0.08519, over 1608623.48 frames. ], batch size: 22, lr: 8.53e-03, grad_scale: 8.0 +2023-02-06 07:17:35,327 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.544e+02 2.589e+02 3.045e+02 4.133e+02 1.084e+03, threshold=6.090e+02, percent-clipped=3.0 +2023-02-06 07:17:51,771 INFO [train.py:901] (2/4) Epoch 9, batch 4000, loss[loss=0.2677, simple_loss=0.3412, pruned_loss=0.0971, over 8442.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3185, pruned_loss=0.08682, over 1613292.01 frames. ], batch size: 27, lr: 8.53e-03, grad_scale: 8.0 +2023-02-06 07:18:17,944 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68703.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:25,233 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68714.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:25,677 INFO [train.py:901] (2/4) Epoch 9, batch 4050, loss[loss=0.2778, simple_loss=0.3516, pruned_loss=0.102, over 8507.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3188, pruned_loss=0.0874, over 1613317.96 frames. ], batch size: 39, lr: 8.53e-03, grad_scale: 8.0 +2023-02-06 07:18:41,427 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-06 07:18:42,621 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68739.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:43,691 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.478e+02 3.133e+02 3.692e+02 8.585e+02, threshold=6.266e+02, percent-clipped=3.0 +2023-02-06 07:18:57,741 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:58,666 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.12 vs. limit=5.0 +2023-02-06 07:19:00,367 INFO [train.py:901] (2/4) Epoch 9, batch 4100, loss[loss=0.2743, simple_loss=0.3524, pruned_loss=0.0981, over 8782.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3192, pruned_loss=0.08741, over 1615999.23 frames. ], batch size: 30, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:19:14,704 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 07:19:18,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5099, 1.4317, 4.7198, 1.7495, 4.0680, 3.9148, 4.1479, 4.0507], + device='cuda:2'), covar=tensor([0.0528, 0.4213, 0.0382, 0.3022, 0.1075, 0.0701, 0.0526, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0555, 0.0552, 0.0512, 0.0578, 0.0488, 0.0491, 0.0553], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:19:34,810 INFO [train.py:901] (2/4) Epoch 9, batch 4150, loss[loss=0.2216, simple_loss=0.2969, pruned_loss=0.07314, over 8085.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3194, pruned_loss=0.08752, over 1617175.13 frames. ], batch size: 21, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:19:38,350 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68820.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:19:52,100 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.558e+02 3.576e+02 4.352e+02 8.740e+02, threshold=7.151e+02, percent-clipped=5.0 +2023-02-06 07:19:55,806 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68845.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:20:09,485 INFO [train.py:901] (2/4) Epoch 9, batch 4200, loss[loss=0.2153, simple_loss=0.284, pruned_loss=0.07331, over 7815.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3188, pruned_loss=0.08658, over 1617774.37 frames. ], batch size: 19, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:20:17,179 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:20:23,075 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 07:20:44,846 INFO [train.py:901] (2/4) Epoch 9, batch 4250, loss[loss=0.2788, simple_loss=0.3557, pruned_loss=0.1009, over 8463.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3213, pruned_loss=0.08809, over 1617558.10 frames. ], batch size: 27, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:20:45,571 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 07:21:02,845 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68940.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:21:03,451 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.856e+02 3.701e+02 4.402e+02 9.379e+02, threshold=7.403e+02, percent-clipped=2.0 +2023-02-06 07:21:20,735 INFO [train.py:901] (2/4) Epoch 9, batch 4300, loss[loss=0.222, simple_loss=0.3003, pruned_loss=0.07187, over 7980.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3206, pruned_loss=0.08814, over 1613311.96 frames. ], batch size: 21, lr: 8.51e-03, grad_scale: 8.0 +2023-02-06 07:21:31,865 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2543, 3.0561, 3.4857, 2.1686, 1.8356, 3.5670, 0.5757, 2.2844], + device='cuda:2'), covar=tensor([0.1984, 0.1110, 0.0334, 0.2780, 0.4640, 0.0455, 0.4897, 0.2054], + device='cuda:2'), in_proj_covar=tensor([0.0158, 0.0158, 0.0091, 0.0211, 0.0248, 0.0094, 0.0159, 0.0155], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:21:54,538 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0839, 1.7089, 1.3885, 1.6125, 1.4549, 1.2837, 1.3446, 1.4183], + device='cuda:2'), covar=tensor([0.0860, 0.0335, 0.0971, 0.0459, 0.0567, 0.1092, 0.0753, 0.0683], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0236, 0.0310, 0.0294, 0.0304, 0.0319, 0.0341, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 07:21:55,035 INFO [train.py:901] (2/4) Epoch 9, batch 4350, loss[loss=0.2454, simple_loss=0.3103, pruned_loss=0.09029, over 7936.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3207, pruned_loss=0.08832, over 1613138.12 frames. ], batch size: 20, lr: 8.51e-03, grad_scale: 8.0 +2023-02-06 07:22:06,430 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9218, 6.1653, 5.2998, 2.3360, 5.3786, 5.6591, 5.6644, 5.3026], + device='cuda:2'), covar=tensor([0.0667, 0.0360, 0.0913, 0.4388, 0.0687, 0.0531, 0.0933, 0.0619], + device='cuda:2'), in_proj_covar=tensor([0.0440, 0.0348, 0.0366, 0.0456, 0.0357, 0.0342, 0.0357, 0.0309], + 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') +2023-02-06 07:22:12,971 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.761e+02 3.203e+02 3.985e+02 6.558e+02, threshold=6.405e+02, percent-clipped=0.0 +2023-02-06 07:22:16,298 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 07:22:17,725 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:22:23,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:22:23,213 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5248, 1.8743, 3.4240, 1.2228, 2.4333, 1.9413, 1.5525, 2.3619], + device='cuda:2'), covar=tensor([0.1630, 0.2194, 0.0675, 0.3752, 0.1479, 0.2693, 0.1749, 0.2070], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0491, 0.0524, 0.0561, 0.0599, 0.0538, 0.0461, 0.0600], + 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') +2023-02-06 07:22:27,153 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69061.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:22:29,684 INFO [train.py:901] (2/4) Epoch 9, batch 4400, loss[loss=0.2275, simple_loss=0.3103, pruned_loss=0.07237, over 8468.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3207, pruned_loss=0.08829, over 1611037.39 frames. ], batch size: 25, lr: 8.51e-03, grad_scale: 8.0 +2023-02-06 07:22:32,740 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 +2023-02-06 07:22:45,417 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6405, 1.9623, 2.1823, 1.0737, 2.2726, 1.5161, 0.6018, 1.9232], + device='cuda:2'), covar=tensor([0.0372, 0.0204, 0.0179, 0.0351, 0.0257, 0.0556, 0.0499, 0.0157], + device='cuda:2'), in_proj_covar=tensor([0.0365, 0.0291, 0.0244, 0.0350, 0.0283, 0.0442, 0.0335, 0.0315], + device='cuda:2'), out_proj_covar=tensor([1.0849e-04, 8.4175e-05, 7.1314e-05, 1.0202e-04, 8.3654e-05, 1.4123e-04, + 9.9800e-05, 9.2963e-05], device='cuda:2') +2023-02-06 07:22:55,805 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 07:23:04,528 INFO [train.py:901] (2/4) Epoch 9, batch 4450, loss[loss=0.1831, simple_loss=0.255, pruned_loss=0.05566, over 7221.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3193, pruned_loss=0.08723, over 1611341.08 frames. ], batch size: 16, lr: 8.50e-03, grad_scale: 8.0 +2023-02-06 07:23:16,659 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:22,376 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.746e+02 3.298e+02 3.852e+02 8.052e+02, threshold=6.596e+02, percent-clipped=4.0 +2023-02-06 07:23:33,866 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:37,255 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:39,015 INFO [train.py:901] (2/4) Epoch 9, batch 4500, loss[loss=0.2234, simple_loss=0.3009, pruned_loss=0.073, over 8239.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3194, pruned_loss=0.08699, over 1614556.74 frames. ], batch size: 22, lr: 8.50e-03, grad_scale: 8.0 +2023-02-06 07:23:49,134 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 07:23:49,291 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:53,178 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:24:12,704 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1721, 1.6911, 1.2061, 2.4270, 1.0247, 1.1718, 1.7227, 1.8001], + device='cuda:2'), covar=tensor([0.2001, 0.1363, 0.2410, 0.0556, 0.1500, 0.2214, 0.1146, 0.1211], + device='cuda:2'), in_proj_covar=tensor([0.0248, 0.0223, 0.0264, 0.0222, 0.0226, 0.0263, 0.0268, 0.0230], + 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') +2023-02-06 07:24:13,185 INFO [train.py:901] (2/4) Epoch 9, batch 4550, loss[loss=0.244, simple_loss=0.3222, pruned_loss=0.08292, over 8136.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3194, pruned_loss=0.08646, over 1615728.88 frames. ], batch size: 22, lr: 8.50e-03, grad_scale: 8.0 +2023-02-06 07:24:16,809 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8946, 1.5553, 3.3500, 1.2770, 2.3219, 3.6841, 3.6478, 3.0701], + device='cuda:2'), covar=tensor([0.1101, 0.1538, 0.0353, 0.2224, 0.0853, 0.0274, 0.0464, 0.0690], + device='cuda:2'), in_proj_covar=tensor([0.0258, 0.0290, 0.0251, 0.0280, 0.0266, 0.0235, 0.0309, 0.0291], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:24:31,949 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.522e+02 2.943e+02 3.743e+02 5.945e+02, threshold=5.886e+02, percent-clipped=0.0 +2023-02-06 07:24:33,338 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:24:47,701 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69263.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:24:48,920 INFO [train.py:901] (2/4) Epoch 9, batch 4600, loss[loss=0.2454, simple_loss=0.3133, pruned_loss=0.08881, over 8083.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3181, pruned_loss=0.08605, over 1615189.65 frames. ], batch size: 21, lr: 8.49e-03, grad_scale: 8.0 +2023-02-06 07:25:07,530 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:22,106 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69311.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:24,600 INFO [train.py:901] (2/4) Epoch 9, batch 4650, loss[loss=0.2315, simple_loss=0.3088, pruned_loss=0.0771, over 8134.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3185, pruned_loss=0.08651, over 1613938.45 frames. ], batch size: 22, lr: 8.49e-03, grad_scale: 8.0 +2023-02-06 07:25:38,791 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:40,120 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:42,602 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.666e+02 3.298e+02 3.900e+02 8.712e+02, threshold=6.595e+02, percent-clipped=8.0 +2023-02-06 07:25:58,526 INFO [train.py:901] (2/4) Epoch 9, batch 4700, loss[loss=0.2908, simple_loss=0.3642, pruned_loss=0.1087, over 8364.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.319, pruned_loss=0.08726, over 1615449.94 frames. ], batch size: 24, lr: 8.49e-03, grad_scale: 8.0 +2023-02-06 07:26:14,068 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:26,899 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69405.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:31,852 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69412.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:33,681 INFO [train.py:901] (2/4) Epoch 9, batch 4750, loss[loss=0.2374, simple_loss=0.3195, pruned_loss=0.07767, over 8464.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3182, pruned_loss=0.08661, over 1614968.56 frames. ], batch size: 29, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:26:35,878 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:43,658 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 07:26:49,115 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 07:26:50,994 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 07:26:51,644 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.564e+02 3.173e+02 4.227e+02 9.736e+02, threshold=6.346e+02, percent-clipped=4.0 +2023-02-06 07:26:53,202 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:08,536 INFO [train.py:901] (2/4) Epoch 9, batch 4800, loss[loss=0.2336, simple_loss=0.3073, pruned_loss=0.07997, over 8281.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3187, pruned_loss=0.08707, over 1611782.60 frames. ], batch size: 23, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:27:38,648 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9330, 1.3165, 6.0229, 2.2113, 5.3738, 5.0397, 5.5913, 5.5299], + device='cuda:2'), covar=tensor([0.0419, 0.4362, 0.0306, 0.2722, 0.0885, 0.0671, 0.0358, 0.0398], + device='cuda:2'), in_proj_covar=tensor([0.0447, 0.0549, 0.0545, 0.0505, 0.0576, 0.0488, 0.0481, 0.0547], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:27:41,299 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 07:27:43,205 INFO [train.py:901] (2/4) Epoch 9, batch 4850, loss[loss=0.2242, simple_loss=0.2873, pruned_loss=0.08053, over 8231.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3191, pruned_loss=0.08791, over 1608165.66 frames. ], batch size: 22, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:27:46,755 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:49,410 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:53,335 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69530.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:55,751 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 07:28:00,637 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.717e+02 3.193e+02 3.973e+02 8.915e+02, threshold=6.387e+02, percent-clipped=1.0 +2023-02-06 07:28:17,599 INFO [train.py:901] (2/4) Epoch 9, batch 4900, loss[loss=0.219, simple_loss=0.306, pruned_loss=0.06598, over 8761.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3201, pruned_loss=0.08822, over 1610629.01 frames. ], batch size: 30, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:28:33,122 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:28:47,478 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69607.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:28:53,429 INFO [train.py:901] (2/4) Epoch 9, batch 4950, loss[loss=0.3128, simple_loss=0.3751, pruned_loss=0.1253, over 6900.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3206, pruned_loss=0.0879, over 1612030.93 frames. ], batch size: 71, lr: 8.47e-03, grad_scale: 8.0 +2023-02-06 07:29:06,611 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:09,346 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:10,455 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.734e+02 3.225e+02 4.131e+02 8.295e+02, threshold=6.450e+02, percent-clipped=5.0 +2023-02-06 07:29:13,212 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:27,248 INFO [train.py:901] (2/4) Epoch 9, batch 5000, loss[loss=0.2011, simple_loss=0.2794, pruned_loss=0.06142, over 7514.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3213, pruned_loss=0.08831, over 1611244.17 frames. ], batch size: 18, lr: 8.47e-03, grad_scale: 8.0 +2023-02-06 07:29:39,129 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69682.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:46,883 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69692.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:29:53,590 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:02,816 INFO [train.py:901] (2/4) Epoch 9, batch 5050, loss[loss=0.2477, simple_loss=0.3192, pruned_loss=0.08807, over 8477.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3202, pruned_loss=0.08751, over 1611454.55 frames. ], batch size: 25, lr: 8.47e-03, grad_scale: 8.0 +2023-02-06 07:30:07,762 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69722.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:30:12,770 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:17,058 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3575, 1.8843, 3.0137, 2.2793, 2.6103, 2.0537, 1.6215, 1.3317], + device='cuda:2'), covar=tensor([0.3198, 0.3460, 0.0940, 0.2132, 0.1698, 0.1796, 0.1514, 0.3494], + device='cuda:2'), in_proj_covar=tensor([0.0858, 0.0825, 0.0702, 0.0808, 0.0911, 0.0762, 0.0689, 0.0738], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 07:30:18,144 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 07:30:20,813 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.703e+02 3.249e+02 3.895e+02 8.845e+02, threshold=6.498e+02, percent-clipped=2.0 +2023-02-06 07:30:26,987 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:30,985 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:36,720 INFO [train.py:901] (2/4) Epoch 9, batch 5100, loss[loss=0.2524, simple_loss=0.3384, pruned_loss=0.08314, over 8494.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3212, pruned_loss=0.08835, over 1614244.91 frames. ], batch size: 26, lr: 8.46e-03, grad_scale: 8.0 +2023-02-06 07:30:44,225 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69776.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:59,029 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:31:01,166 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1334, 1.6223, 3.5191, 1.5430, 2.3575, 3.8645, 3.8861, 3.2849], + device='cuda:2'), covar=tensor([0.1113, 0.1669, 0.0356, 0.2044, 0.0970, 0.0270, 0.0470, 0.0681], + device='cuda:2'), in_proj_covar=tensor([0.0254, 0.0288, 0.0249, 0.0277, 0.0265, 0.0229, 0.0305, 0.0288], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:31:01,900 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:31:11,532 INFO [train.py:901] (2/4) Epoch 9, batch 5150, loss[loss=0.2646, simple_loss=0.3316, pruned_loss=0.09876, over 8613.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3211, pruned_loss=0.08884, over 1609223.35 frames. ], batch size: 39, lr: 8.46e-03, grad_scale: 8.0 +2023-02-06 07:31:29,711 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.410e+02 3.240e+02 3.896e+02 9.119e+02, threshold=6.481e+02, percent-clipped=3.0 +2023-02-06 07:31:32,779 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69845.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:31:43,082 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 07:31:46,671 INFO [train.py:901] (2/4) Epoch 9, batch 5200, loss[loss=0.2411, simple_loss=0.321, pruned_loss=0.0806, over 8033.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3197, pruned_loss=0.08767, over 1607689.05 frames. ], batch size: 22, lr: 8.46e-03, grad_scale: 8.0 +2023-02-06 07:31:50,849 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69871.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:02,099 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4152, 1.9893, 3.0532, 2.4514, 2.7685, 2.2268, 1.7697, 1.4523], + device='cuda:2'), covar=tensor([0.3219, 0.3824, 0.0941, 0.2153, 0.1673, 0.1862, 0.1499, 0.3672], + device='cuda:2'), in_proj_covar=tensor([0.0853, 0.0820, 0.0700, 0.0807, 0.0907, 0.0758, 0.0686, 0.0735], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:32:06,754 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69895.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:11,376 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69901.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:17,773 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 07:32:20,225 INFO [train.py:901] (2/4) Epoch 9, batch 5250, loss[loss=0.2619, simple_loss=0.3246, pruned_loss=0.09959, over 8503.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3211, pruned_loss=0.08844, over 1609974.54 frames. ], batch size: 28, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:32:23,772 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69920.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:27,863 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:38,259 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.909e+02 3.504e+02 4.160e+02 7.603e+02, threshold=7.007e+02, percent-clipped=5.0 +2023-02-06 07:32:50,717 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69958.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:55,303 INFO [train.py:901] (2/4) Epoch 9, batch 5300, loss[loss=0.2573, simple_loss=0.3218, pruned_loss=0.09644, over 6850.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3204, pruned_loss=0.08792, over 1608639.89 frames. ], batch size: 72, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:33:04,978 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69978.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:33:08,282 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:33:19,169 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3215, 1.4161, 1.5432, 1.3157, 1.1074, 1.4310, 1.8601, 1.9569], + device='cuda:2'), covar=tensor([0.0440, 0.1281, 0.1804, 0.1411, 0.0592, 0.1530, 0.0639, 0.0523], + device='cuda:2'), in_proj_covar=tensor([0.0104, 0.0159, 0.0199, 0.0162, 0.0109, 0.0167, 0.0121, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 07:33:22,935 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70003.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:33:24,838 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:33:31,222 INFO [train.py:901] (2/4) Epoch 9, batch 5350, loss[loss=0.2418, simple_loss=0.3203, pruned_loss=0.08169, over 8521.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3206, pruned_loss=0.08831, over 1605489.64 frames. ], batch size: 28, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:33:36,940 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 07:33:42,007 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:33:45,286 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70036.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:33:48,397 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.965e+02 3.484e+02 4.155e+02 9.515e+02, threshold=6.968e+02, percent-clipped=2.0 +2023-02-06 07:33:57,319 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70053.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:05,194 INFO [train.py:901] (2/4) Epoch 9, batch 5400, loss[loss=0.293, simple_loss=0.3729, pruned_loss=0.1066, over 8601.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3211, pruned_loss=0.08818, over 1605699.38 frames. ], batch size: 31, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:34:14,818 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:31,571 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70101.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:40,958 INFO [train.py:901] (2/4) Epoch 9, batch 5450, loss[loss=0.2434, simple_loss=0.3333, pruned_loss=0.07674, over 8106.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3208, pruned_loss=0.08793, over 1608789.93 frames. ], batch size: 23, lr: 8.44e-03, grad_scale: 8.0 +2023-02-06 07:34:48,362 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70126.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:49,042 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70127.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:49,624 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:58,814 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.682e+02 3.191e+02 4.046e+02 1.028e+03, threshold=6.382e+02, percent-clipped=4.0 +2023-02-06 07:35:04,332 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 07:35:05,827 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70151.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:35:06,525 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70152.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:35:15,866 INFO [train.py:901] (2/4) Epoch 9, batch 5500, loss[loss=0.2183, simple_loss=0.2973, pruned_loss=0.06959, over 8246.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.32, pruned_loss=0.08735, over 1609048.84 frames. ], batch size: 24, lr: 8.44e-03, grad_scale: 16.0 +2023-02-06 07:35:31,379 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6704, 1.6690, 2.1493, 1.3161, 1.0721, 2.1527, 0.2932, 1.3081], + device='cuda:2'), covar=tensor([0.2942, 0.1810, 0.0552, 0.2787, 0.5313, 0.0471, 0.4117, 0.2306], + device='cuda:2'), in_proj_covar=tensor([0.0155, 0.0156, 0.0091, 0.0206, 0.0246, 0.0094, 0.0158, 0.0155], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:35:50,278 INFO [train.py:901] (2/4) Epoch 9, batch 5550, loss[loss=0.2823, simple_loss=0.3497, pruned_loss=0.1075, over 8098.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3196, pruned_loss=0.08777, over 1608377.11 frames. ], batch size: 23, lr: 8.44e-03, grad_scale: 16.0 +2023-02-06 07:36:07,873 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.484e+02 3.031e+02 3.937e+02 9.276e+02, threshold=6.062e+02, percent-clipped=2.0 +2023-02-06 07:36:14,988 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 07:36:24,752 INFO [train.py:901] (2/4) Epoch 9, batch 5600, loss[loss=0.2474, simple_loss=0.3242, pruned_loss=0.0853, over 8359.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3207, pruned_loss=0.08817, over 1612435.48 frames. ], batch size: 24, lr: 8.43e-03, grad_scale: 16.0 +2023-02-06 07:36:34,078 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:36:35,512 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5028, 2.0525, 3.4373, 1.2002, 2.6530, 1.9914, 1.7288, 2.3096], + device='cuda:2'), covar=tensor([0.1789, 0.1993, 0.0812, 0.3719, 0.1452, 0.2577, 0.1664, 0.2147], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0496, 0.0524, 0.0564, 0.0600, 0.0537, 0.0463, 0.0603], + 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') +2023-02-06 07:36:59,388 INFO [train.py:901] (2/4) Epoch 9, batch 5650, loss[loss=0.2733, simple_loss=0.3563, pruned_loss=0.09515, over 8192.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3203, pruned_loss=0.08814, over 1609017.05 frames. ], batch size: 23, lr: 8.43e-03, grad_scale: 8.0 +2023-02-06 07:37:08,097 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 07:37:18,047 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.609e+02 3.248e+02 4.005e+02 8.106e+02, threshold=6.497e+02, percent-clipped=5.0 +2023-02-06 07:37:32,963 INFO [train.py:901] (2/4) Epoch 9, batch 5700, loss[loss=0.2779, simple_loss=0.3606, pruned_loss=0.09755, over 8496.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3206, pruned_loss=0.08835, over 1607820.41 frames. ], batch size: 28, lr: 8.43e-03, grad_scale: 8.0 +2023-02-06 07:37:56,393 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4331, 1.9498, 3.1170, 2.5265, 2.7753, 2.2079, 1.7797, 1.3292], + device='cuda:2'), covar=tensor([0.3174, 0.3523, 0.0884, 0.2015, 0.1607, 0.1847, 0.1454, 0.3701], + device='cuda:2'), in_proj_covar=tensor([0.0861, 0.0829, 0.0707, 0.0806, 0.0905, 0.0765, 0.0686, 0.0737], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:38:02,548 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6746, 4.6890, 4.1800, 1.6809, 4.1259, 4.1925, 4.2617, 3.9384], + device='cuda:2'), covar=tensor([0.0982, 0.0693, 0.1284, 0.5835, 0.0906, 0.1016, 0.1470, 0.0839], + device='cuda:2'), in_proj_covar=tensor([0.0449, 0.0353, 0.0367, 0.0468, 0.0365, 0.0350, 0.0362, 0.0309], + 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') +2023-02-06 07:38:02,669 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70407.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:38:08,013 INFO [train.py:901] (2/4) Epoch 9, batch 5750, loss[loss=0.248, simple_loss=0.3279, pruned_loss=0.08403, over 8517.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3203, pruned_loss=0.08776, over 1615157.03 frames. ], batch size: 29, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:38:12,130 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 07:38:20,277 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70432.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:38:27,523 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.898e+02 3.376e+02 4.229e+02 8.555e+02, threshold=6.753e+02, percent-clipped=3.0 +2023-02-06 07:38:43,380 INFO [train.py:901] (2/4) Epoch 9, batch 5800, loss[loss=0.268, simple_loss=0.337, pruned_loss=0.09953, over 8500.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3202, pruned_loss=0.08737, over 1615394.41 frames. ], batch size: 26, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:38:48,295 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:38:48,580 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.70 vs. limit=5.0 +2023-02-06 07:39:18,642 INFO [train.py:901] (2/4) Epoch 9, batch 5850, loss[loss=0.1944, simple_loss=0.2681, pruned_loss=0.06039, over 7787.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3188, pruned_loss=0.08702, over 1612963.77 frames. ], batch size: 19, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:39:37,361 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.477e+02 3.501e+02 4.376e+02 8.995e+02, threshold=7.001e+02, percent-clipped=4.0 +2023-02-06 07:39:48,208 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2859, 1.9519, 3.0793, 2.5058, 2.6875, 2.2270, 1.6964, 1.3785], + device='cuda:2'), covar=tensor([0.3291, 0.3509, 0.0954, 0.1932, 0.1630, 0.1802, 0.1552, 0.3728], + device='cuda:2'), in_proj_covar=tensor([0.0853, 0.0819, 0.0703, 0.0805, 0.0900, 0.0761, 0.0683, 0.0733], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:39:53,322 INFO [train.py:901] (2/4) Epoch 9, batch 5900, loss[loss=0.2274, simple_loss=0.2982, pruned_loss=0.07827, over 6808.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3195, pruned_loss=0.08782, over 1610286.49 frames. ], batch size: 15, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:40:08,006 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:40:23,563 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.3422, 3.0098, 3.4782, 2.2982, 1.9127, 3.5624, 0.5739, 2.4138], + device='cuda:2'), covar=tensor([0.1888, 0.1654, 0.0406, 0.2642, 0.4751, 0.0439, 0.4939, 0.1791], + device='cuda:2'), in_proj_covar=tensor([0.0159, 0.0160, 0.0091, 0.0210, 0.0248, 0.0096, 0.0161, 0.0158], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:40:27,231 INFO [train.py:901] (2/4) Epoch 9, batch 5950, loss[loss=0.2103, simple_loss=0.285, pruned_loss=0.06774, over 7548.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3196, pruned_loss=0.08788, over 1608836.56 frames. ], batch size: 18, lr: 8.41e-03, grad_scale: 8.0 +2023-02-06 07:40:32,699 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:40:45,949 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.730e+02 3.193e+02 3.849e+02 7.953e+02, threshold=6.387e+02, percent-clipped=3.0 +2023-02-06 07:41:02,077 INFO [train.py:901] (2/4) Epoch 9, batch 6000, loss[loss=0.2804, simple_loss=0.3496, pruned_loss=0.1056, over 8579.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3196, pruned_loss=0.08766, over 1614547.39 frames. ], batch size: 34, lr: 8.41e-03, grad_scale: 8.0 +2023-02-06 07:41:02,078 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 07:41:08,106 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8421, 1.4247, 1.4583, 1.2532, 1.0448, 1.3559, 1.5346, 1.4734], + device='cuda:2'), covar=tensor([0.0599, 0.1412, 0.1911, 0.1519, 0.0645, 0.1684, 0.0750, 0.0638], + device='cuda:2'), in_proj_covar=tensor([0.0105, 0.0158, 0.0198, 0.0162, 0.0108, 0.0166, 0.0121, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 07:41:14,591 INFO [train.py:935] (2/4) Epoch 9, validation: loss=0.1952, simple_loss=0.2947, pruned_loss=0.0479, over 944034.00 frames. +2023-02-06 07:41:14,592 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 07:41:33,926 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0428, 1.6752, 3.1734, 1.3838, 2.2008, 3.4861, 3.4845, 2.9711], + device='cuda:2'), covar=tensor([0.1005, 0.1398, 0.0410, 0.1940, 0.0992, 0.0260, 0.0529, 0.0642], + device='cuda:2'), in_proj_covar=tensor([0.0251, 0.0286, 0.0250, 0.0273, 0.0265, 0.0229, 0.0312, 0.0290], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:41:49,711 INFO [train.py:901] (2/4) Epoch 9, batch 6050, loss[loss=0.2214, simple_loss=0.2901, pruned_loss=0.07633, over 7534.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3181, pruned_loss=0.08701, over 1612320.31 frames. ], batch size: 18, lr: 8.41e-03, grad_scale: 8.0 +2023-02-06 07:41:55,399 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.57 vs. limit=5.0 +2023-02-06 07:42:04,732 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:42:07,988 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.822e+02 3.602e+02 4.348e+02 1.269e+03, threshold=7.203e+02, percent-clipped=6.0 +2023-02-06 07:42:24,368 INFO [train.py:901] (2/4) Epoch 9, batch 6100, loss[loss=0.264, simple_loss=0.3393, pruned_loss=0.09436, over 8587.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3189, pruned_loss=0.08707, over 1613720.61 frames. ], batch size: 31, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:42:40,908 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9127, 1.6942, 3.1714, 1.4765, 2.3531, 3.4795, 3.4258, 2.9563], + device='cuda:2'), covar=tensor([0.1077, 0.1491, 0.0448, 0.1910, 0.0977, 0.0255, 0.0540, 0.0637], + device='cuda:2'), in_proj_covar=tensor([0.0251, 0.0287, 0.0250, 0.0274, 0.0267, 0.0229, 0.0311, 0.0290], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:42:42,822 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 07:43:00,373 INFO [train.py:901] (2/4) Epoch 9, batch 6150, loss[loss=0.2758, simple_loss=0.3423, pruned_loss=0.1047, over 6976.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3195, pruned_loss=0.08774, over 1610773.86 frames. ], batch size: 71, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:43:14,656 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 07:43:18,324 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.692e+02 3.232e+02 3.879e+02 7.941e+02, threshold=6.463e+02, percent-clipped=1.0 +2023-02-06 07:43:19,258 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70843.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:43:20,207 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.14 vs. limit=5.0 +2023-02-06 07:43:33,935 INFO [train.py:901] (2/4) Epoch 9, batch 6200, loss[loss=0.2697, simple_loss=0.3448, pruned_loss=0.09734, over 8202.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3199, pruned_loss=0.08749, over 1615722.23 frames. ], batch size: 23, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:43:35,020 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-06 07:43:36,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:44:05,850 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3098, 1.9056, 2.9527, 2.3081, 2.4694, 2.0295, 1.6427, 1.4153], + device='cuda:2'), covar=tensor([0.3205, 0.3487, 0.0901, 0.2071, 0.1709, 0.1911, 0.1619, 0.3439], + device='cuda:2'), in_proj_covar=tensor([0.0862, 0.0832, 0.0706, 0.0816, 0.0905, 0.0767, 0.0690, 0.0745], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 07:44:09,711 INFO [train.py:901] (2/4) Epoch 9, batch 6250, loss[loss=0.291, simple_loss=0.3567, pruned_loss=0.1127, over 8528.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3203, pruned_loss=0.08814, over 1612413.51 frames. ], batch size: 26, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:44:28,454 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.792e+02 3.423e+02 4.432e+02 1.474e+03, threshold=6.847e+02, percent-clipped=7.0 +2023-02-06 07:44:30,140 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2413, 1.4033, 2.0555, 1.1012, 1.4780, 1.5175, 1.2902, 1.4101], + device='cuda:2'), covar=tensor([0.1586, 0.2082, 0.0696, 0.3491, 0.1430, 0.2568, 0.1718, 0.1761], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0499, 0.0529, 0.0570, 0.0607, 0.0545, 0.0464, 0.0606], + 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') +2023-02-06 07:44:30,308 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 07:44:43,937 INFO [train.py:901] (2/4) Epoch 9, batch 6300, loss[loss=0.2297, simple_loss=0.3036, pruned_loss=0.0779, over 8229.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3205, pruned_loss=0.08853, over 1610319.80 frames. ], batch size: 22, lr: 8.39e-03, grad_scale: 8.0 +2023-02-06 07:45:03,882 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:16,145 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71010.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:19,448 INFO [train.py:901] (2/4) Epoch 9, batch 6350, loss[loss=0.2571, simple_loss=0.3289, pruned_loss=0.09267, over 8718.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3202, pruned_loss=0.08821, over 1613602.61 frames. ], batch size: 39, lr: 8.39e-03, grad_scale: 8.0 +2023-02-06 07:45:21,647 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:38,840 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.814e+02 3.293e+02 4.210e+02 8.338e+02, threshold=6.585e+02, percent-clipped=5.0 +2023-02-06 07:45:40,425 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5024, 2.0504, 3.3925, 1.3652, 2.5310, 1.9669, 1.7701, 2.1102], + device='cuda:2'), covar=tensor([0.1966, 0.2157, 0.0753, 0.4077, 0.1519, 0.2961, 0.1691, 0.2356], + device='cuda:2'), in_proj_covar=tensor([0.0484, 0.0500, 0.0526, 0.0572, 0.0609, 0.0545, 0.0464, 0.0608], + 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') +2023-02-06 07:45:42,976 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:54,184 INFO [train.py:901] (2/4) Epoch 9, batch 6400, loss[loss=0.2591, simple_loss=0.3407, pruned_loss=0.0888, over 8017.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3224, pruned_loss=0.08965, over 1613167.18 frames. ], batch size: 22, lr: 8.39e-03, grad_scale: 8.0 +2023-02-06 07:45:55,035 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2150, 1.2352, 3.3038, 0.9651, 2.9295, 2.7645, 3.0387, 2.9312], + device='cuda:2'), covar=tensor([0.0591, 0.3660, 0.0767, 0.3367, 0.1255, 0.1064, 0.0579, 0.0754], + device='cuda:2'), in_proj_covar=tensor([0.0447, 0.0560, 0.0554, 0.0514, 0.0584, 0.0495, 0.0485, 0.0548], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:46:28,849 INFO [train.py:901] (2/4) Epoch 9, batch 6450, loss[loss=0.2407, simple_loss=0.3108, pruned_loss=0.08531, over 7691.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3215, pruned_loss=0.08902, over 1610172.11 frames. ], batch size: 18, lr: 8.38e-03, grad_scale: 4.0 +2023-02-06 07:46:37,163 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8606, 1.5742, 3.2298, 1.4410, 2.2960, 3.5442, 3.5408, 2.9968], + device='cuda:2'), covar=tensor([0.1078, 0.1462, 0.0368, 0.1901, 0.0898, 0.0239, 0.0425, 0.0651], + device='cuda:2'), in_proj_covar=tensor([0.0251, 0.0284, 0.0247, 0.0275, 0.0263, 0.0228, 0.0306, 0.0288], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:46:48,400 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.655e+02 3.350e+02 4.272e+02 1.011e+03, threshold=6.701e+02, percent-clipped=3.0 +2023-02-06 07:47:03,136 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4551, 1.5043, 1.7264, 1.3294, 0.9015, 1.7323, 0.0963, 1.1322], + device='cuda:2'), covar=tensor([0.2615, 0.1805, 0.0594, 0.1873, 0.4987, 0.0627, 0.3892, 0.1862], + device='cuda:2'), in_proj_covar=tensor([0.0157, 0.0158, 0.0094, 0.0208, 0.0242, 0.0096, 0.0158, 0.0156], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 07:47:03,607 INFO [train.py:901] (2/4) Epoch 9, batch 6500, loss[loss=0.2847, simple_loss=0.362, pruned_loss=0.1037, over 8626.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.32, pruned_loss=0.08829, over 1607081.06 frames. ], batch size: 31, lr: 8.38e-03, grad_scale: 4.0 +2023-02-06 07:47:23,302 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0959, 2.4954, 3.0067, 1.2076, 3.1721, 1.8463, 1.4952, 1.8505], + device='cuda:2'), covar=tensor([0.0520, 0.0249, 0.0156, 0.0488, 0.0270, 0.0582, 0.0555, 0.0320], + device='cuda:2'), in_proj_covar=tensor([0.0374, 0.0300, 0.0247, 0.0356, 0.0287, 0.0446, 0.0339, 0.0323], + device='cuda:2'), out_proj_covar=tensor([1.1071e-04, 8.6471e-05, 7.1391e-05, 1.0291e-04, 8.4344e-05, 1.4103e-04, + 1.0042e-04, 9.5095e-05], device='cuda:2') +2023-02-06 07:47:37,713 INFO [train.py:901] (2/4) Epoch 9, batch 6550, loss[loss=0.227, simple_loss=0.2923, pruned_loss=0.08085, over 7450.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3201, pruned_loss=0.08771, over 1613047.18 frames. ], batch size: 17, lr: 8.38e-03, grad_scale: 4.0 +2023-02-06 07:47:50,023 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 07:47:52,884 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2975, 1.5812, 4.4385, 1.4940, 3.8431, 3.7068, 3.9669, 3.8020], + device='cuda:2'), covar=tensor([0.0488, 0.4103, 0.0506, 0.3508, 0.1161, 0.0821, 0.0567, 0.0596], + device='cuda:2'), in_proj_covar=tensor([0.0450, 0.0562, 0.0554, 0.0513, 0.0584, 0.0496, 0.0486, 0.0547], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:47:58,021 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.500e+02 3.444e+02 4.178e+02 7.414e+02, threshold=6.887e+02, percent-clipped=1.0 +2023-02-06 07:48:10,530 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 07:48:13,164 INFO [train.py:901] (2/4) Epoch 9, batch 6600, loss[loss=0.3293, simple_loss=0.3871, pruned_loss=0.1358, over 6792.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3206, pruned_loss=0.08783, over 1612156.92 frames. ], batch size: 71, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:48:47,047 INFO [train.py:901] (2/4) Epoch 9, batch 6650, loss[loss=0.2175, simple_loss=0.2975, pruned_loss=0.06871, over 8138.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3209, pruned_loss=0.08795, over 1615538.68 frames. ], batch size: 22, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:49:05,717 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.641e+02 3.214e+02 4.234e+02 1.005e+03, threshold=6.427e+02, percent-clipped=4.0 +2023-02-06 07:49:13,932 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71354.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:49:21,688 INFO [train.py:901] (2/4) Epoch 9, batch 6700, loss[loss=0.2595, simple_loss=0.316, pruned_loss=0.1015, over 7653.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3199, pruned_loss=0.08792, over 1611562.86 frames. ], batch size: 19, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:49:37,835 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7674, 1.4390, 4.0519, 1.6072, 3.1786, 3.1526, 3.5988, 3.5029], + device='cuda:2'), covar=tensor([0.1019, 0.5930, 0.1072, 0.4324, 0.2191, 0.1581, 0.1008, 0.1040], + device='cuda:2'), in_proj_covar=tensor([0.0445, 0.0555, 0.0548, 0.0512, 0.0583, 0.0494, 0.0484, 0.0545], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:49:41,042 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:49:56,537 INFO [train.py:901] (2/4) Epoch 9, batch 6750, loss[loss=0.2335, simple_loss=0.3183, pruned_loss=0.07434, over 8468.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3202, pruned_loss=0.08815, over 1612836.72 frames. ], batch size: 27, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:50:15,374 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 3.032e+02 3.821e+02 4.704e+02 1.129e+03, threshold=7.641e+02, percent-clipped=7.0 +2023-02-06 07:50:23,396 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 07:50:30,841 INFO [train.py:901] (2/4) Epoch 9, batch 6800, loss[loss=0.2936, simple_loss=0.357, pruned_loss=0.1151, over 8355.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.321, pruned_loss=0.08831, over 1613095.26 frames. ], batch size: 24, lr: 8.36e-03, grad_scale: 8.0 +2023-02-06 07:50:33,616 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:50:50,784 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-06 07:51:01,250 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:51:06,467 INFO [train.py:901] (2/4) Epoch 9, batch 6850, loss[loss=0.2242, simple_loss=0.3041, pruned_loss=0.07221, over 8520.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3196, pruned_loss=0.08726, over 1611608.95 frames. ], batch size: 31, lr: 8.36e-03, grad_scale: 8.0 +2023-02-06 07:51:14,502 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 07:51:25,243 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.628e+02 3.217e+02 4.054e+02 6.964e+02, threshold=6.433e+02, percent-clipped=0.0 +2023-02-06 07:51:40,041 INFO [train.py:901] (2/4) Epoch 9, batch 6900, loss[loss=0.3097, simple_loss=0.3687, pruned_loss=0.1254, over 8561.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3206, pruned_loss=0.08801, over 1614059.98 frames. ], batch size: 34, lr: 8.36e-03, grad_scale: 8.0 +2023-02-06 07:52:15,379 INFO [train.py:901] (2/4) Epoch 9, batch 6950, loss[loss=0.2523, simple_loss=0.3243, pruned_loss=0.09015, over 8478.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3203, pruned_loss=0.08805, over 1615604.03 frames. ], batch size: 25, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:52:23,472 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 07:52:29,649 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:52:35,520 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.622e+02 3.284e+02 3.978e+02 8.428e+02, threshold=6.567e+02, percent-clipped=2.0 +2023-02-06 07:52:50,437 INFO [train.py:901] (2/4) Epoch 9, batch 7000, loss[loss=0.2135, simple_loss=0.2892, pruned_loss=0.06894, over 7679.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3198, pruned_loss=0.08767, over 1612777.63 frames. ], batch size: 18, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:52:57,329 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1248, 1.1558, 3.3064, 0.9204, 2.8373, 2.7978, 3.0160, 2.9075], + device='cuda:2'), covar=tensor([0.0815, 0.3715, 0.0741, 0.3375, 0.1521, 0.0990, 0.0735, 0.0882], + device='cuda:2'), in_proj_covar=tensor([0.0445, 0.0552, 0.0554, 0.0511, 0.0580, 0.0492, 0.0485, 0.0543], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 07:53:06,293 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9689, 1.0716, 1.0144, 0.4941, 1.0865, 0.8693, 0.1448, 0.9903], + device='cuda:2'), covar=tensor([0.0213, 0.0185, 0.0176, 0.0285, 0.0205, 0.0482, 0.0407, 0.0168], + device='cuda:2'), in_proj_covar=tensor([0.0375, 0.0301, 0.0250, 0.0360, 0.0291, 0.0450, 0.0344, 0.0326], + device='cuda:2'), out_proj_covar=tensor([1.1063e-04, 8.6493e-05, 7.2229e-05, 1.0388e-04, 8.5340e-05, 1.4201e-04, + 1.0171e-04, 9.5733e-05], device='cuda:2') +2023-02-06 07:53:24,883 INFO [train.py:901] (2/4) Epoch 9, batch 7050, loss[loss=0.2202, simple_loss=0.2945, pruned_loss=0.07294, over 7697.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3195, pruned_loss=0.08818, over 1610283.04 frames. ], batch size: 18, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:53:32,433 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71725.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:53:41,130 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8458, 5.9484, 5.0469, 2.0901, 5.2211, 5.4374, 5.3900, 5.1491], + device='cuda:2'), covar=tensor([0.0492, 0.0360, 0.0836, 0.4480, 0.0629, 0.0620, 0.1033, 0.0667], + device='cuda:2'), in_proj_covar=tensor([0.0444, 0.0354, 0.0363, 0.0463, 0.0363, 0.0349, 0.0361, 0.0307], + 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') +2023-02-06 07:53:45,153 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.886e+02 3.338e+02 4.007e+02 6.250e+02, threshold=6.676e+02, percent-clipped=0.0 +2023-02-06 07:53:50,715 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:53:59,294 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71763.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:54:00,447 INFO [train.py:901] (2/4) Epoch 9, batch 7100, loss[loss=0.2586, simple_loss=0.3319, pruned_loss=0.09264, over 8352.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3191, pruned_loss=0.08824, over 1603433.26 frames. ], batch size: 26, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:54:05,647 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 07:54:06,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.8836, 0.8044, 0.9112, 0.8306, 0.5932, 0.9229, 0.1100, 0.7153], + device='cuda:2'), covar=tensor([0.1725, 0.1466, 0.0533, 0.1148, 0.3266, 0.0520, 0.3375, 0.1882], + device='cuda:2'), in_proj_covar=tensor([0.0157, 0.0156, 0.0092, 0.0205, 0.0242, 0.0095, 0.0155, 0.0153], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 07:54:16,149 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:54:34,522 INFO [train.py:901] (2/4) Epoch 9, batch 7150, loss[loss=0.2253, simple_loss=0.295, pruned_loss=0.07779, over 8236.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3187, pruned_loss=0.08828, over 1601700.58 frames. ], batch size: 22, lr: 8.34e-03, grad_scale: 8.0 +2023-02-06 07:54:53,688 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.29 vs. limit=5.0 +2023-02-06 07:54:54,749 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.560e+02 3.246e+02 4.043e+02 1.359e+03, threshold=6.493e+02, percent-clipped=7.0 +2023-02-06 07:55:10,759 INFO [train.py:901] (2/4) Epoch 9, batch 7200, loss[loss=0.2536, simple_loss=0.3334, pruned_loss=0.0869, over 8638.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.319, pruned_loss=0.08904, over 1598393.55 frames. ], batch size: 39, lr: 8.34e-03, grad_scale: 8.0 +2023-02-06 07:55:34,215 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 07:55:43,959 INFO [train.py:901] (2/4) Epoch 9, batch 7250, loss[loss=0.2605, simple_loss=0.344, pruned_loss=0.0885, over 8495.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3198, pruned_loss=0.08933, over 1602470.70 frames. ], batch size: 26, lr: 8.34e-03, grad_scale: 8.0 +2023-02-06 07:55:58,953 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71937.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:56:02,878 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.597e+02 3.277e+02 3.984e+02 9.565e+02, threshold=6.554e+02, percent-clipped=6.0 +2023-02-06 07:56:19,573 INFO [train.py:901] (2/4) Epoch 9, batch 7300, loss[loss=0.2433, simple_loss=0.3021, pruned_loss=0.09218, over 8082.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3188, pruned_loss=0.08858, over 1603093.08 frames. ], batch size: 21, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:56:28,839 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71978.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:56:41,715 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4448, 2.0101, 3.3362, 1.2519, 2.5048, 1.8789, 1.4865, 2.3502], + device='cuda:2'), covar=tensor([0.1652, 0.1932, 0.0716, 0.3562, 0.1382, 0.2653, 0.1725, 0.1927], + device='cuda:2'), in_proj_covar=tensor([0.0480, 0.0493, 0.0527, 0.0570, 0.0605, 0.0541, 0.0462, 0.0606], + 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') +2023-02-06 07:56:47,427 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6934, 1.3355, 1.6488, 1.2052, 0.8892, 1.4103, 1.4962, 1.4186], + device='cuda:2'), covar=tensor([0.0541, 0.1282, 0.1740, 0.1500, 0.0613, 0.1508, 0.0700, 0.0635], + device='cuda:2'), in_proj_covar=tensor([0.0103, 0.0156, 0.0195, 0.0162, 0.0108, 0.0166, 0.0119, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 07:56:54,487 INFO [train.py:901] (2/4) Epoch 9, batch 7350, loss[loss=0.2359, simple_loss=0.3194, pruned_loss=0.07618, over 8294.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3188, pruned_loss=0.08817, over 1604888.95 frames. ], batch size: 23, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:56:59,696 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.22 vs. limit=5.0 +2023-02-06 07:57:02,501 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 07:57:13,521 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.925e+02 3.749e+02 4.804e+02 1.068e+03, threshold=7.499e+02, percent-clipped=9.0 +2023-02-06 07:57:22,478 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 07:57:30,036 INFO [train.py:901] (2/4) Epoch 9, batch 7400, loss[loss=0.2744, simple_loss=0.3388, pruned_loss=0.105, over 7154.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3183, pruned_loss=0.0877, over 1608448.83 frames. ], batch size: 72, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:57:40,352 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8207, 2.8460, 3.3016, 2.2393, 1.5847, 3.3149, 0.7037, 1.8828], + device='cuda:2'), covar=tensor([0.1914, 0.1045, 0.0388, 0.2124, 0.4466, 0.0556, 0.3777, 0.1984], + device='cuda:2'), in_proj_covar=tensor([0.0156, 0.0155, 0.0091, 0.0204, 0.0240, 0.0093, 0.0152, 0.0154], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 07:57:50,403 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:58:03,869 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 07:58:05,138 INFO [train.py:901] (2/4) Epoch 9, batch 7450, loss[loss=0.2753, simple_loss=0.3395, pruned_loss=0.1055, over 8609.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3171, pruned_loss=0.08621, over 1612486.68 frames. ], batch size: 39, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:58:23,965 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.502e+02 2.410e+02 3.229e+02 3.860e+02 9.903e+02, threshold=6.459e+02, percent-clipped=1.0 +2023-02-06 07:58:26,830 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72147.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:58:38,700 INFO [train.py:901] (2/4) Epoch 9, batch 7500, loss[loss=0.2359, simple_loss=0.3146, pruned_loss=0.07859, over 7940.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3166, pruned_loss=0.08564, over 1612439.20 frames. ], batch size: 20, lr: 8.32e-03, grad_scale: 8.0 +2023-02-06 07:59:15,028 INFO [train.py:901] (2/4) Epoch 9, batch 7550, loss[loss=0.2279, simple_loss=0.3019, pruned_loss=0.07696, over 7974.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3171, pruned_loss=0.08585, over 1613698.93 frames. ], batch size: 21, lr: 8.32e-03, grad_scale: 8.0 +2023-02-06 07:59:33,695 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.819e+02 3.433e+02 4.309e+02 8.597e+02, threshold=6.865e+02, percent-clipped=4.0 +2023-02-06 07:59:48,158 INFO [train.py:901] (2/4) Epoch 9, batch 7600, loss[loss=0.2418, simple_loss=0.3225, pruned_loss=0.08058, over 8498.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3184, pruned_loss=0.08652, over 1614543.71 frames. ], batch size: 26, lr: 8.32e-03, grad_scale: 8.0 +2023-02-06 07:59:58,716 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72281.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:00:22,185 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6536, 1.5515, 2.8295, 1.2129, 2.0532, 3.0260, 3.0696, 2.6061], + device='cuda:2'), covar=tensor([0.1033, 0.1365, 0.0394, 0.2063, 0.0893, 0.0314, 0.0544, 0.0666], + device='cuda:2'), in_proj_covar=tensor([0.0260, 0.0295, 0.0255, 0.0287, 0.0271, 0.0236, 0.0319, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:00:22,667 INFO [train.py:901] (2/4) Epoch 9, batch 7650, loss[loss=0.2639, simple_loss=0.339, pruned_loss=0.09444, over 8247.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3182, pruned_loss=0.08634, over 1614354.22 frames. ], batch size: 24, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:00:42,776 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.469e+02 2.706e+02 3.178e+02 3.983e+02 6.818e+02, threshold=6.357e+02, percent-clipped=0.0 +2023-02-06 08:00:43,590 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:00:47,001 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:00:57,586 INFO [train.py:901] (2/4) Epoch 9, batch 7700, loss[loss=0.2541, simple_loss=0.3381, pruned_loss=0.08504, over 8467.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3173, pruned_loss=0.08601, over 1614275.80 frames. ], batch size: 25, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:01:02,559 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7601, 3.7933, 2.5908, 2.4123, 2.7265, 1.7685, 2.7162, 2.7211], + device='cuda:2'), covar=tensor([0.1553, 0.0262, 0.0789, 0.0829, 0.0570, 0.1387, 0.0972, 0.0981], + device='cuda:2'), in_proj_covar=tensor([0.0340, 0.0235, 0.0311, 0.0295, 0.0301, 0.0320, 0.0334, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 08:01:03,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72374.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:01:09,770 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 08:01:18,754 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72396.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:01:32,217 INFO [train.py:901] (2/4) Epoch 9, batch 7750, loss[loss=0.2131, simple_loss=0.2841, pruned_loss=0.0711, over 7218.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3166, pruned_loss=0.08589, over 1611130.91 frames. ], batch size: 16, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:01:53,035 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.670e+02 3.267e+02 4.054e+02 1.108e+03, threshold=6.534e+02, percent-clipped=5.0 +2023-02-06 08:02:08,197 INFO [train.py:901] (2/4) Epoch 9, batch 7800, loss[loss=0.2757, simple_loss=0.3369, pruned_loss=0.1073, over 8637.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3178, pruned_loss=0.08688, over 1613685.94 frames. ], batch size: 34, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:02:19,579 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7726, 4.6934, 4.1873, 1.8902, 4.2448, 4.1836, 4.2624, 3.8665], + device='cuda:2'), covar=tensor([0.0641, 0.0488, 0.1016, 0.5004, 0.0725, 0.0900, 0.1220, 0.0745], + device='cuda:2'), in_proj_covar=tensor([0.0441, 0.0350, 0.0365, 0.0462, 0.0362, 0.0345, 0.0357, 0.0307], + 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') +2023-02-06 08:02:25,493 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72491.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:02:41,268 INFO [train.py:901] (2/4) Epoch 9, batch 7850, loss[loss=0.2371, simple_loss=0.3089, pruned_loss=0.08263, over 7810.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3184, pruned_loss=0.08713, over 1612770.53 frames. ], batch size: 20, lr: 8.30e-03, grad_scale: 8.0 +2023-02-06 08:02:59,522 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.530e+02 3.201e+02 3.890e+02 8.475e+02, threshold=6.403e+02, percent-clipped=6.0 +2023-02-06 08:03:14,038 INFO [train.py:901] (2/4) Epoch 9, batch 7900, loss[loss=0.2502, simple_loss=0.3319, pruned_loss=0.08427, over 8252.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3185, pruned_loss=0.08666, over 1614882.68 frames. ], batch size: 24, lr: 8.30e-03, grad_scale: 8.0 +2023-02-06 08:03:41,418 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:03:47,068 INFO [train.py:901] (2/4) Epoch 9, batch 7950, loss[loss=0.2301, simple_loss=0.3108, pruned_loss=0.07473, over 8650.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.317, pruned_loss=0.08566, over 1614722.47 frames. ], batch size: 39, lr: 8.30e-03, grad_scale: 8.0 +2023-02-06 08:04:05,356 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.498e+02 3.176e+02 4.184e+02 8.861e+02, threshold=6.353e+02, percent-clipped=6.0 +2023-02-06 08:04:11,432 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72652.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:04:19,757 INFO [train.py:901] (2/4) Epoch 9, batch 8000, loss[loss=0.218, simple_loss=0.2784, pruned_loss=0.07882, over 7529.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3177, pruned_loss=0.08643, over 1617980.18 frames. ], batch size: 18, lr: 8.29e-03, grad_scale: 8.0 +2023-02-06 08:04:27,785 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:04:34,988 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:04:52,976 INFO [train.py:901] (2/4) Epoch 9, batch 8050, loss[loss=0.2066, simple_loss=0.2884, pruned_loss=0.06245, over 7552.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3163, pruned_loss=0.0867, over 1598197.07 frames. ], batch size: 18, lr: 8.29e-03, grad_scale: 8.0 +2023-02-06 08:05:05,108 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1294, 1.7355, 3.5120, 1.4271, 2.3955, 3.8794, 3.8555, 3.3519], + device='cuda:2'), covar=tensor([0.1021, 0.1478, 0.0321, 0.2105, 0.1015, 0.0221, 0.0440, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0251, 0.0288, 0.0248, 0.0277, 0.0263, 0.0228, 0.0308, 0.0285], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:05:11,520 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.635e+02 3.102e+02 3.711e+02 7.462e+02, threshold=6.205e+02, percent-clipped=1.0 +2023-02-06 08:05:25,732 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 08:05:31,309 INFO [train.py:901] (2/4) Epoch 10, batch 0, loss[loss=0.2745, simple_loss=0.3505, pruned_loss=0.09926, over 8469.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3505, pruned_loss=0.09926, over 8469.00 frames. ], batch size: 25, lr: 7.88e-03, grad_scale: 8.0 +2023-02-06 08:05:31,309 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 08:05:43,260 INFO [train.py:935] (2/4) Epoch 10, validation: loss=0.1954, simple_loss=0.295, pruned_loss=0.0479, over 944034.00 frames. +2023-02-06 08:05:43,261 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 08:05:57,119 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 08:06:07,500 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 08:06:17,948 INFO [train.py:901] (2/4) Epoch 10, batch 50, loss[loss=0.2896, simple_loss=0.357, pruned_loss=0.1111, over 8529.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3199, pruned_loss=0.08768, over 365087.33 frames. ], batch size: 39, lr: 7.88e-03, grad_scale: 8.0 +2023-02-06 08:06:21,761 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72803.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:06:31,220 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 08:06:49,397 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.716e+02 3.124e+02 3.887e+02 7.160e+02, threshold=6.248e+02, percent-clipped=5.0 +2023-02-06 08:06:52,292 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 08:06:52,950 INFO [train.py:901] (2/4) Epoch 10, batch 100, loss[loss=0.2738, simple_loss=0.3435, pruned_loss=0.1021, over 8188.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3174, pruned_loss=0.08673, over 646286.33 frames. ], batch size: 23, lr: 7.88e-03, grad_scale: 8.0 +2023-02-06 08:07:03,806 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72862.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:07:22,310 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72887.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:07:30,272 INFO [train.py:901] (2/4) Epoch 10, batch 150, loss[loss=0.24, simple_loss=0.3278, pruned_loss=0.07611, over 8440.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3192, pruned_loss=0.08631, over 864263.40 frames. ], batch size: 29, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:07:33,243 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9184, 1.5441, 2.2072, 1.7927, 2.0480, 1.8136, 1.4660, 0.7081], + device='cuda:2'), covar=tensor([0.3360, 0.3209, 0.0987, 0.2068, 0.1489, 0.1848, 0.1580, 0.3247], + device='cuda:2'), in_proj_covar=tensor([0.0865, 0.0828, 0.0700, 0.0817, 0.0905, 0.0763, 0.0689, 0.0744], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:07:36,683 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5952, 1.4672, 2.8103, 1.2275, 2.0963, 3.0598, 3.0482, 2.6185], + device='cuda:2'), covar=tensor([0.1100, 0.1409, 0.0403, 0.2096, 0.0824, 0.0303, 0.0600, 0.0695], + device='cuda:2'), in_proj_covar=tensor([0.0252, 0.0289, 0.0251, 0.0279, 0.0264, 0.0230, 0.0312, 0.0285], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:07:40,076 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72912.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:08:01,152 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.670e+02 3.307e+02 4.288e+02 9.841e+02, threshold=6.614e+02, percent-clipped=3.0 +2023-02-06 08:08:04,548 INFO [train.py:901] (2/4) Epoch 10, batch 200, loss[loss=0.2416, simple_loss=0.3225, pruned_loss=0.08039, over 8352.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3216, pruned_loss=0.08766, over 1033410.63 frames. ], batch size: 26, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:08:29,435 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72982.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:08:41,006 INFO [train.py:901] (2/4) Epoch 10, batch 250, loss[loss=0.2085, simple_loss=0.294, pruned_loss=0.06154, over 8027.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3207, pruned_loss=0.08779, over 1160888.87 frames. ], batch size: 22, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:08:47,838 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 08:08:56,861 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 08:09:02,432 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1101, 1.4714, 1.5731, 1.3892, 1.1341, 1.3195, 1.7447, 1.8019], + device='cuda:2'), covar=tensor([0.0572, 0.1285, 0.1736, 0.1411, 0.0585, 0.1627, 0.0700, 0.0548], + device='cuda:2'), in_proj_covar=tensor([0.0106, 0.0155, 0.0196, 0.0160, 0.0107, 0.0167, 0.0120, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 08:09:07,477 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-06 08:09:12,566 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.688e+02 3.158e+02 3.760e+02 5.735e+02, threshold=6.316e+02, percent-clipped=0.0 +2023-02-06 08:09:16,047 INFO [train.py:901] (2/4) Epoch 10, batch 300, loss[loss=0.2716, simple_loss=0.3402, pruned_loss=0.1015, over 7815.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3211, pruned_loss=0.08829, over 1258643.54 frames. ], batch size: 20, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:09:23,774 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:09:40,900 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:09:51,613 INFO [train.py:901] (2/4) Epoch 10, batch 350, loss[loss=0.2461, simple_loss=0.3161, pruned_loss=0.08805, over 8366.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3208, pruned_loss=0.08857, over 1345194.03 frames. ], batch size: 24, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:09:55,107 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5732, 1.9699, 3.4121, 1.2701, 2.5744, 1.8812, 1.5181, 2.3875], + device='cuda:2'), covar=tensor([0.1619, 0.2143, 0.0719, 0.3879, 0.1499, 0.2779, 0.1799, 0.2230], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0495, 0.0534, 0.0567, 0.0607, 0.0540, 0.0466, 0.0605], + 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') +2023-02-06 08:10:23,500 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.632e+02 3.058e+02 3.924e+02 7.931e+02, threshold=6.116e+02, percent-clipped=5.0 +2023-02-06 08:10:26,918 INFO [train.py:901] (2/4) Epoch 10, batch 400, loss[loss=0.1886, simple_loss=0.2645, pruned_loss=0.05636, over 7716.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3183, pruned_loss=0.08679, over 1404108.04 frames. ], batch size: 18, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:10:49,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1741, 3.8896, 2.7977, 2.7411, 2.8089, 2.2474, 2.8115, 2.9877], + device='cuda:2'), covar=tensor([0.1599, 0.0239, 0.0848, 0.0719, 0.0629, 0.1081, 0.1016, 0.1095], + device='cuda:2'), in_proj_covar=tensor([0.0343, 0.0231, 0.0310, 0.0297, 0.0300, 0.0318, 0.0335, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 08:11:01,343 INFO [train.py:901] (2/4) Epoch 10, batch 450, loss[loss=0.2391, simple_loss=0.306, pruned_loss=0.08611, over 7518.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3194, pruned_loss=0.08677, over 1452813.49 frames. ], batch size: 18, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:11:33,877 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.629e+02 3.140e+02 3.877e+02 8.143e+02, threshold=6.279e+02, percent-clipped=4.0 +2023-02-06 08:11:35,981 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1040, 4.1477, 3.6956, 1.9507, 3.6533, 3.7004, 3.8233, 3.2886], + device='cuda:2'), covar=tensor([0.0949, 0.0617, 0.1092, 0.4801, 0.0916, 0.0960, 0.1287, 0.1142], + device='cuda:2'), in_proj_covar=tensor([0.0461, 0.0358, 0.0375, 0.0477, 0.0373, 0.0357, 0.0367, 0.0318], + 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') +2023-02-06 08:11:37,157 INFO [train.py:901] (2/4) Epoch 10, batch 500, loss[loss=0.2203, simple_loss=0.2949, pruned_loss=0.07286, over 8144.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3173, pruned_loss=0.08618, over 1485476.37 frames. ], batch size: 22, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:11:42,548 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:12:10,508 INFO [train.py:901] (2/4) Epoch 10, batch 550, loss[loss=0.2507, simple_loss=0.3138, pruned_loss=0.0938, over 7963.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3172, pruned_loss=0.08598, over 1513114.20 frames. ], batch size: 21, lr: 7.85e-03, grad_scale: 16.0 +2023-02-06 08:12:18,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4408, 2.0165, 3.9281, 1.8773, 2.5284, 4.4670, 4.3211, 3.8421], + device='cuda:2'), covar=tensor([0.0872, 0.1246, 0.0353, 0.1672, 0.1037, 0.0178, 0.0431, 0.0520], + device='cuda:2'), in_proj_covar=tensor([0.0255, 0.0290, 0.0252, 0.0282, 0.0265, 0.0231, 0.0314, 0.0290], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:12:19,359 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73311.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:12:19,472 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5108, 2.0572, 3.1326, 2.4484, 2.8143, 2.2358, 1.7461, 1.4552], + device='cuda:2'), covar=tensor([0.3265, 0.3779, 0.0971, 0.2353, 0.1808, 0.1930, 0.1644, 0.3836], + device='cuda:2'), in_proj_covar=tensor([0.0860, 0.0831, 0.0701, 0.0817, 0.0907, 0.0763, 0.0690, 0.0744], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:12:23,928 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7763, 1.3461, 1.5022, 1.2107, 1.0580, 1.3220, 1.5668, 1.4506], + device='cuda:2'), covar=tensor([0.0551, 0.1369, 0.1849, 0.1486, 0.0620, 0.1626, 0.0728, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0105, 0.0157, 0.0197, 0.0161, 0.0108, 0.0168, 0.0120, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 08:12:29,158 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73326.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:12:41,609 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.475e+02 3.100e+02 3.629e+02 1.040e+03, threshold=6.201e+02, percent-clipped=3.0 +2023-02-06 08:12:44,822 INFO [train.py:901] (2/4) Epoch 10, batch 600, loss[loss=0.3062, simple_loss=0.3504, pruned_loss=0.131, over 7394.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3178, pruned_loss=0.08614, over 1537815.28 frames. ], batch size: 71, lr: 7.85e-03, grad_scale: 16.0 +2023-02-06 08:12:56,196 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 08:13:01,723 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73371.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:13:20,006 INFO [train.py:901] (2/4) Epoch 10, batch 650, loss[loss=0.2852, simple_loss=0.3393, pruned_loss=0.1155, over 7804.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3185, pruned_loss=0.08637, over 1560436.77 frames. ], batch size: 20, lr: 7.85e-03, grad_scale: 16.0 +2023-02-06 08:13:50,088 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73441.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:13:51,167 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.469e+02 3.040e+02 3.840e+02 6.530e+02, threshold=6.081e+02, percent-clipped=1.0 +2023-02-06 08:13:54,587 INFO [train.py:901] (2/4) Epoch 10, batch 700, loss[loss=0.2307, simple_loss=0.2951, pruned_loss=0.08312, over 7566.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.317, pruned_loss=0.08505, over 1574285.69 frames. ], batch size: 18, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:14:31,477 INFO [train.py:901] (2/4) Epoch 10, batch 750, loss[loss=0.2728, simple_loss=0.341, pruned_loss=0.1023, over 8557.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3177, pruned_loss=0.08515, over 1586210.27 frames. ], batch size: 31, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:14:45,794 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 08:14:54,750 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 08:15:02,254 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.756e+02 3.307e+02 3.958e+02 8.111e+02, threshold=6.615e+02, percent-clipped=6.0 +2023-02-06 08:15:02,466 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3098, 1.4216, 1.2764, 1.9087, 0.8137, 1.1299, 1.2663, 1.5571], + device='cuda:2'), covar=tensor([0.1026, 0.0974, 0.1256, 0.0567, 0.1259, 0.1754, 0.0944, 0.0786], + device='cuda:2'), in_proj_covar=tensor([0.0246, 0.0221, 0.0264, 0.0220, 0.0223, 0.0260, 0.0267, 0.0228], + 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') +2023-02-06 08:15:05,717 INFO [train.py:901] (2/4) Epoch 10, batch 800, loss[loss=0.2546, simple_loss=0.3255, pruned_loss=0.09184, over 8671.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3176, pruned_loss=0.08528, over 1596089.53 frames. ], batch size: 49, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:15:17,221 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7227, 1.3467, 3.9232, 1.4404, 3.4316, 3.2875, 3.5480, 3.4302], + device='cuda:2'), covar=tensor([0.0700, 0.4134, 0.0628, 0.3241, 0.1366, 0.1024, 0.0611, 0.0787], + device='cuda:2'), in_proj_covar=tensor([0.0444, 0.0556, 0.0557, 0.0510, 0.0585, 0.0496, 0.0488, 0.0552], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:15:41,988 INFO [train.py:901] (2/4) Epoch 10, batch 850, loss[loss=0.2457, simple_loss=0.3196, pruned_loss=0.0859, over 7654.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3178, pruned_loss=0.08581, over 1600983.07 frames. ], batch size: 19, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:15:49,885 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:03,006 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73627.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:13,764 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.847e+02 3.470e+02 4.482e+02 1.720e+03, threshold=6.940e+02, percent-clipped=10.0 +2023-02-06 08:16:17,263 INFO [train.py:901] (2/4) Epoch 10, batch 900, loss[loss=0.2251, simple_loss=0.3007, pruned_loss=0.07476, over 8348.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3172, pruned_loss=0.0851, over 1605144.54 frames. ], batch size: 24, lr: 7.83e-03, grad_scale: 16.0 +2023-02-06 08:16:20,226 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73652.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:22,223 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73655.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:52,131 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73697.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:16:52,440 INFO [train.py:901] (2/4) Epoch 10, batch 950, loss[loss=0.3706, simple_loss=0.4093, pruned_loss=0.166, over 8745.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3182, pruned_loss=0.0859, over 1608823.66 frames. ], batch size: 30, lr: 7.83e-03, grad_scale: 8.0 +2023-02-06 08:17:10,351 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73722.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:17:12,016 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 08:17:18,312 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 08:17:24,917 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.751e+02 3.323e+02 4.211e+02 1.163e+03, threshold=6.645e+02, percent-clipped=9.0 +2023-02-06 08:17:25,308 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 +2023-02-06 08:17:27,464 INFO [train.py:901] (2/4) Epoch 10, batch 1000, loss[loss=0.213, simple_loss=0.3023, pruned_loss=0.06179, over 8464.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3172, pruned_loss=0.08516, over 1609074.96 frames. ], batch size: 25, lr: 7.83e-03, grad_scale: 8.0 +2023-02-06 08:17:28,903 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:17:35,503 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3331, 1.6420, 1.5774, 0.8752, 1.6581, 1.2527, 0.3392, 1.5454], + device='cuda:2'), covar=tensor([0.0263, 0.0169, 0.0184, 0.0294, 0.0214, 0.0560, 0.0450, 0.0147], + device='cuda:2'), in_proj_covar=tensor([0.0370, 0.0304, 0.0256, 0.0360, 0.0287, 0.0452, 0.0340, 0.0332], + device='cuda:2'), out_proj_covar=tensor([1.0860e-04, 8.6769e-05, 7.3630e-05, 1.0359e-04, 8.3571e-05, 1.4230e-04, + 9.9960e-05, 9.7048e-05], device='cuda:2') +2023-02-06 08:17:42,250 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73770.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:17:50,797 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 08:18:00,667 INFO [train.py:901] (2/4) Epoch 10, batch 1050, loss[loss=0.2635, simple_loss=0.3264, pruned_loss=0.1003, over 8348.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3185, pruned_loss=0.08604, over 1617821.26 frames. ], batch size: 24, lr: 7.83e-03, grad_scale: 8.0 +2023-02-06 08:18:01,390 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 08:18:16,647 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2320, 1.7838, 2.7217, 2.1598, 2.3318, 2.0481, 1.6336, 1.0659], + device='cuda:2'), covar=tensor([0.3123, 0.3301, 0.0875, 0.2029, 0.1559, 0.1805, 0.1458, 0.3553], + device='cuda:2'), in_proj_covar=tensor([0.0865, 0.0829, 0.0710, 0.0821, 0.0910, 0.0771, 0.0691, 0.0748], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:18:34,182 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.634e+02 3.058e+02 3.903e+02 1.179e+03, threshold=6.116e+02, percent-clipped=2.0 +2023-02-06 08:18:36,808 INFO [train.py:901] (2/4) Epoch 10, batch 1100, loss[loss=0.2687, simple_loss=0.3377, pruned_loss=0.09983, over 8133.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.318, pruned_loss=0.08578, over 1614919.18 frames. ], batch size: 22, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:19:09,834 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 08:19:10,510 INFO [train.py:901] (2/4) Epoch 10, batch 1150, loss[loss=0.2377, simple_loss=0.3082, pruned_loss=0.08361, over 7785.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3178, pruned_loss=0.08559, over 1614944.48 frames. ], batch size: 19, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:19:23,867 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9176, 1.6095, 1.7243, 1.2656, 1.1237, 1.3686, 1.6914, 1.4566], + device='cuda:2'), covar=tensor([0.0484, 0.1128, 0.1564, 0.1362, 0.0532, 0.1403, 0.0631, 0.0569], + device='cuda:2'), in_proj_covar=tensor([0.0104, 0.0157, 0.0195, 0.0161, 0.0107, 0.0166, 0.0119, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 08:19:42,449 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.356e+02 2.791e+02 3.726e+02 1.227e+03, threshold=5.583e+02, percent-clipped=4.0 +2023-02-06 08:19:45,136 INFO [train.py:901] (2/4) Epoch 10, batch 1200, loss[loss=0.2484, simple_loss=0.3223, pruned_loss=0.08724, over 8665.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3179, pruned_loss=0.08569, over 1614261.67 frames. ], batch size: 34, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:19:48,562 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73952.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:20:20,055 INFO [train.py:901] (2/4) Epoch 10, batch 1250, loss[loss=0.2139, simple_loss=0.2886, pruned_loss=0.06957, over 7698.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3176, pruned_loss=0.08545, over 1617113.91 frames. ], batch size: 18, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:20:35,800 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7972, 2.1541, 1.6351, 2.5962, 1.2159, 1.2635, 1.8150, 2.1982], + device='cuda:2'), covar=tensor([0.0984, 0.0839, 0.1315, 0.0521, 0.1272, 0.1890, 0.1122, 0.0797], + device='cuda:2'), in_proj_covar=tensor([0.0242, 0.0219, 0.0262, 0.0219, 0.0222, 0.0257, 0.0264, 0.0224], + 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') +2023-02-06 08:20:39,877 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:20:51,572 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.519e+02 3.075e+02 3.983e+02 7.817e+02, threshold=6.150e+02, percent-clipped=4.0 +2023-02-06 08:20:54,941 INFO [train.py:901] (2/4) Epoch 10, batch 1300, loss[loss=0.2577, simple_loss=0.3335, pruned_loss=0.09096, over 8499.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.317, pruned_loss=0.08501, over 1619780.22 frames. ], batch size: 28, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:20:57,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74051.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:20:58,729 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 08:21:07,843 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74067.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:19,112 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:21,529 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.00 vs. limit=5.0 +2023-02-06 08:21:26,838 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:29,347 INFO [train.py:901] (2/4) Epoch 10, batch 1350, loss[loss=0.2044, simple_loss=0.2747, pruned_loss=0.06704, over 7249.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3178, pruned_loss=0.08537, over 1620230.24 frames. ], batch size: 16, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:21:59,534 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.773e+02 3.448e+02 4.052e+02 8.675e+02, threshold=6.895e+02, percent-clipped=5.0 +2023-02-06 08:22:02,250 INFO [train.py:901] (2/4) Epoch 10, batch 1400, loss[loss=0.2504, simple_loss=0.3024, pruned_loss=0.09917, over 8027.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3186, pruned_loss=0.08629, over 1619573.63 frames. ], batch size: 22, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:22:16,769 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 08:22:38,033 INFO [train.py:901] (2/4) Epoch 10, batch 1450, loss[loss=0.2008, simple_loss=0.2796, pruned_loss=0.06098, over 7558.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3184, pruned_loss=0.08594, over 1617982.85 frames. ], batch size: 18, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:22:41,666 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 08:22:45,869 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:23:01,296 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5355, 5.6400, 4.9311, 2.5451, 4.9479, 5.3732, 5.1697, 4.8504], + device='cuda:2'), covar=tensor([0.0712, 0.0480, 0.0889, 0.4293, 0.0747, 0.0648, 0.1121, 0.0696], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0356, 0.0377, 0.0470, 0.0372, 0.0353, 0.0361, 0.0314], + 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') +2023-02-06 08:23:01,603 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 08:23:07,379 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2172, 1.4547, 3.4763, 1.2855, 2.4098, 3.9377, 3.8933, 3.2979], + device='cuda:2'), covar=tensor([0.0907, 0.1551, 0.0302, 0.1983, 0.0916, 0.0213, 0.0387, 0.0663], + device='cuda:2'), in_proj_covar=tensor([0.0256, 0.0293, 0.0254, 0.0284, 0.0269, 0.0232, 0.0318, 0.0289], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:23:09,187 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.838e+02 2.525e+02 3.045e+02 3.954e+02 1.310e+03, threshold=6.089e+02, percent-clipped=4.0 +2023-02-06 08:23:11,852 INFO [train.py:901] (2/4) Epoch 10, batch 1500, loss[loss=0.2507, simple_loss=0.3271, pruned_loss=0.08713, over 7978.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3175, pruned_loss=0.08555, over 1615583.91 frames. ], batch size: 21, lr: 7.80e-03, grad_scale: 8.0 +2023-02-06 08:23:18,347 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74258.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:23:46,596 INFO [train.py:901] (2/4) Epoch 10, batch 1550, loss[loss=0.2521, simple_loss=0.333, pruned_loss=0.08565, over 8507.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3185, pruned_loss=0.08609, over 1616764.24 frames. ], batch size: 26, lr: 7.80e-03, grad_scale: 8.0 +2023-02-06 08:23:47,773 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 08:24:05,684 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74323.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:24:19,877 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.540e+02 3.095e+02 3.981e+02 6.537e+02, threshold=6.190e+02, percent-clipped=3.0 +2023-02-06 08:24:22,707 INFO [train.py:901] (2/4) Epoch 10, batch 1600, loss[loss=0.2093, simple_loss=0.278, pruned_loss=0.07026, over 7818.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3185, pruned_loss=0.08596, over 1618091.03 frames. ], batch size: 20, lr: 7.80e-03, grad_scale: 8.0 +2023-02-06 08:24:22,910 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74348.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:24:56,948 INFO [train.py:901] (2/4) Epoch 10, batch 1650, loss[loss=0.299, simple_loss=0.3612, pruned_loss=0.1184, over 8468.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3178, pruned_loss=0.08575, over 1616407.96 frames. ], batch size: 27, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:25:18,074 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74426.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:25:30,268 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.508e+02 3.008e+02 3.971e+02 8.483e+02, threshold=6.016e+02, percent-clipped=6.0 +2023-02-06 08:25:31,722 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3484, 1.5611, 1.3303, 1.8571, 0.8236, 1.1302, 1.4039, 1.5270], + device='cuda:2'), covar=tensor([0.0997, 0.0776, 0.1320, 0.0574, 0.1185, 0.1721, 0.0823, 0.0739], + device='cuda:2'), in_proj_covar=tensor([0.0241, 0.0216, 0.0261, 0.0217, 0.0221, 0.0255, 0.0261, 0.0222], + 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') +2023-02-06 08:25:32,863 INFO [train.py:901] (2/4) Epoch 10, batch 1700, loss[loss=0.239, simple_loss=0.3188, pruned_loss=0.07957, over 8140.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3164, pruned_loss=0.08491, over 1612401.33 frames. ], batch size: 22, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:25:44,221 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:26:00,778 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:26:05,906 INFO [train.py:901] (2/4) Epoch 10, batch 1750, loss[loss=0.2395, simple_loss=0.3098, pruned_loss=0.08458, over 8241.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.317, pruned_loss=0.08539, over 1615801.27 frames. ], batch size: 22, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:26:36,857 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74541.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:26:38,773 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.804e+02 3.517e+02 4.698e+02 1.546e+03, threshold=7.034e+02, percent-clipped=7.0 +2023-02-06 08:26:41,532 INFO [train.py:901] (2/4) Epoch 10, batch 1800, loss[loss=0.2125, simple_loss=0.2915, pruned_loss=0.06674, over 7928.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3165, pruned_loss=0.08549, over 1608356.58 frames. ], batch size: 20, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:27:14,907 INFO [train.py:901] (2/4) Epoch 10, batch 1850, loss[loss=0.2352, simple_loss=0.2976, pruned_loss=0.08644, over 7701.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.316, pruned_loss=0.08536, over 1607192.57 frames. ], batch size: 18, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:27:17,789 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74602.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:27:25,265 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74613.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:27:46,970 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 2.722e+02 3.219e+02 4.226e+02 1.097e+03, threshold=6.437e+02, percent-clipped=2.0 +2023-02-06 08:27:50,389 INFO [train.py:901] (2/4) Epoch 10, batch 1900, loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.09279, over 8488.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3167, pruned_loss=0.08564, over 1605834.17 frames. ], batch size: 28, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:28:13,876 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 08:28:25,274 INFO [train.py:901] (2/4) Epoch 10, batch 1950, loss[loss=0.2608, simple_loss=0.3175, pruned_loss=0.102, over 7701.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3149, pruned_loss=0.08467, over 1607189.51 frames. ], batch size: 18, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:28:25,938 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 08:28:38,216 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74717.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:28:43,969 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 08:28:56,744 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.465e+02 3.030e+02 3.717e+02 6.494e+02, threshold=6.060e+02, percent-clipped=3.0 +2023-02-06 08:28:59,513 INFO [train.py:901] (2/4) Epoch 10, batch 2000, loss[loss=0.3344, simple_loss=0.3772, pruned_loss=0.1459, over 6842.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3151, pruned_loss=0.08467, over 1611328.92 frames. ], batch size: 71, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:29:34,157 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:29:34,600 INFO [train.py:901] (2/4) Epoch 10, batch 2050, loss[loss=0.241, simple_loss=0.3, pruned_loss=0.09094, over 7693.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3147, pruned_loss=0.0843, over 1611943.55 frames. ], batch size: 18, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:29:36,988 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.79 vs. limit=5.0 +2023-02-06 08:29:50,349 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:29:58,215 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2359, 1.5652, 1.6474, 1.3887, 0.9980, 1.4870, 1.7824, 1.3911], + device='cuda:2'), covar=tensor([0.0488, 0.1228, 0.1592, 0.1417, 0.0603, 0.1503, 0.0655, 0.0651], + device='cuda:2'), in_proj_covar=tensor([0.0103, 0.0156, 0.0197, 0.0162, 0.0106, 0.0167, 0.0119, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 08:30:04,619 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.471e+02 3.084e+02 4.282e+02 1.276e+03, threshold=6.169e+02, percent-clipped=5.0 +2023-02-06 08:30:06,171 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1109, 1.6483, 1.3522, 1.6493, 1.4297, 1.1815, 1.1914, 1.3637], + device='cuda:2'), covar=tensor([0.0833, 0.0359, 0.1018, 0.0411, 0.0569, 0.1239, 0.0769, 0.0640], + device='cuda:2'), in_proj_covar=tensor([0.0342, 0.0232, 0.0312, 0.0296, 0.0304, 0.0317, 0.0337, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 08:30:07,353 INFO [train.py:901] (2/4) Epoch 10, batch 2100, loss[loss=0.1817, simple_loss=0.2553, pruned_loss=0.05406, over 7818.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3161, pruned_loss=0.08494, over 1612292.35 frames. ], batch size: 19, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:30:18,144 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6708, 1.6558, 2.0119, 1.6739, 1.1684, 2.1155, 0.1866, 1.2176], + device='cuda:2'), covar=tensor([0.3124, 0.2405, 0.0584, 0.1832, 0.4669, 0.0516, 0.3799, 0.1936], + device='cuda:2'), in_proj_covar=tensor([0.0162, 0.0163, 0.0092, 0.0213, 0.0254, 0.0097, 0.0161, 0.0158], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:30:21,955 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9258, 6.0983, 5.2247, 2.5187, 5.3657, 5.8179, 5.6840, 5.2667], + device='cuda:2'), covar=tensor([0.0495, 0.0419, 0.0962, 0.4303, 0.0774, 0.0450, 0.1126, 0.0576], + device='cuda:2'), in_proj_covar=tensor([0.0444, 0.0358, 0.0379, 0.0465, 0.0367, 0.0352, 0.0361, 0.0315], + 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') +2023-02-06 08:30:24,017 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5181, 1.9847, 1.9955, 1.1820, 2.2054, 1.4399, 0.6123, 1.6926], + device='cuda:2'), covar=tensor([0.0453, 0.0207, 0.0200, 0.0402, 0.0234, 0.0595, 0.0586, 0.0207], + device='cuda:2'), in_proj_covar=tensor([0.0374, 0.0306, 0.0265, 0.0364, 0.0296, 0.0452, 0.0348, 0.0332], + device='cuda:2'), out_proj_covar=tensor([1.0950e-04, 8.7103e-05, 7.5990e-05, 1.0479e-04, 8.5882e-05, 1.4168e-04, + 1.0204e-04, 9.6580e-05], device='cuda:2') +2023-02-06 08:30:43,223 INFO [train.py:901] (2/4) Epoch 10, batch 2150, loss[loss=0.2318, simple_loss=0.2997, pruned_loss=0.08199, over 7657.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3161, pruned_loss=0.0847, over 1614117.50 frames. ], batch size: 19, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:31:02,632 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:13,914 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.626e+02 3.226e+02 3.775e+02 6.882e+02, threshold=6.451e+02, percent-clipped=1.0 +2023-02-06 08:31:16,706 INFO [train.py:901] (2/4) Epoch 10, batch 2200, loss[loss=0.2456, simple_loss=0.3103, pruned_loss=0.09044, over 7984.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3165, pruned_loss=0.08452, over 1619982.10 frames. ], batch size: 21, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:31:22,705 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:33,574 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:39,396 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:50,364 INFO [train.py:901] (2/4) Epoch 10, batch 2250, loss[loss=0.2547, simple_loss=0.327, pruned_loss=0.0912, over 8460.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3158, pruned_loss=0.08436, over 1617986.03 frames. ], batch size: 27, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:31:50,575 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74998.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:32:21,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8022, 2.1969, 1.7406, 2.7170, 1.8936, 1.6406, 2.3270, 2.3643], + device='cuda:2'), covar=tensor([0.1457, 0.1021, 0.1897, 0.0571, 0.1222, 0.1797, 0.0777, 0.0730], + device='cuda:2'), in_proj_covar=tensor([0.0244, 0.0216, 0.0262, 0.0218, 0.0223, 0.0255, 0.0263, 0.0223], + 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') +2023-02-06 08:32:23,117 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.748e+02 3.468e+02 4.709e+02 1.048e+03, threshold=6.936e+02, percent-clipped=3.0 +2023-02-06 08:32:25,869 INFO [train.py:901] (2/4) Epoch 10, batch 2300, loss[loss=0.2409, simple_loss=0.3276, pruned_loss=0.07712, over 8520.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3157, pruned_loss=0.08493, over 1616302.02 frames. ], batch size: 28, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:32:42,300 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:32:59,557 INFO [train.py:901] (2/4) Epoch 10, batch 2350, loss[loss=0.2514, simple_loss=0.3243, pruned_loss=0.08924, over 8342.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3151, pruned_loss=0.08466, over 1611128.42 frames. ], batch size: 26, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:33:24,635 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1762, 1.1178, 1.2280, 1.1260, 0.8711, 1.2760, 0.0348, 0.9273], + device='cuda:2'), covar=tensor([0.2945, 0.2111, 0.0720, 0.1597, 0.4273, 0.0714, 0.3592, 0.2051], + device='cuda:2'), in_proj_covar=tensor([0.0162, 0.0162, 0.0091, 0.0212, 0.0253, 0.0097, 0.0161, 0.0157], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:33:33,018 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.526e+02 3.215e+02 4.182e+02 1.054e+03, threshold=6.430e+02, percent-clipped=5.0 +2023-02-06 08:33:35,793 INFO [train.py:901] (2/4) Epoch 10, batch 2400, loss[loss=0.2611, simple_loss=0.3305, pruned_loss=0.09584, over 8453.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3155, pruned_loss=0.08473, over 1612700.16 frames. ], batch size: 27, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:34:08,759 INFO [train.py:901] (2/4) Epoch 10, batch 2450, loss[loss=0.2584, simple_loss=0.3382, pruned_loss=0.08929, over 8486.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.316, pruned_loss=0.08509, over 1618335.48 frames. ], batch size: 28, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:34:40,830 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 2.613e+02 3.092e+02 4.227e+02 1.037e+03, threshold=6.184e+02, percent-clipped=5.0 +2023-02-06 08:34:43,537 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75246.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:34:44,708 INFO [train.py:901] (2/4) Epoch 10, batch 2500, loss[loss=0.2033, simple_loss=0.2835, pruned_loss=0.06155, over 8030.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3151, pruned_loss=0.08449, over 1615901.34 frames. ], batch size: 22, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:35:00,232 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:09,291 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 08:35:11,671 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:18,086 INFO [train.py:901] (2/4) Epoch 10, batch 2550, loss[loss=0.2394, simple_loss=0.3182, pruned_loss=0.08033, over 8197.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3147, pruned_loss=0.08449, over 1611610.82 frames. ], batch size: 23, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:35:18,200 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5928, 4.6039, 4.1256, 1.7815, 4.1582, 4.3013, 4.2070, 3.9509], + device='cuda:2'), covar=tensor([0.0711, 0.0497, 0.0989, 0.4991, 0.0737, 0.0666, 0.1069, 0.0872], + device='cuda:2'), in_proj_covar=tensor([0.0451, 0.0360, 0.0377, 0.0473, 0.0368, 0.0356, 0.0365, 0.0316], + 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') +2023-02-06 08:35:36,579 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:37,543 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 08:35:38,119 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:40,137 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5477, 2.0272, 3.5262, 1.2860, 2.4734, 2.1591, 1.6110, 2.4039], + device='cuda:2'), covar=tensor([0.1607, 0.2094, 0.0691, 0.3680, 0.1419, 0.2450, 0.1757, 0.1967], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0496, 0.0532, 0.0566, 0.0603, 0.0543, 0.0463, 0.0603], + 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') +2023-02-06 08:35:49,134 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.664e+02 3.245e+02 3.791e+02 6.757e+02, threshold=6.490e+02, percent-clipped=2.0 +2023-02-06 08:35:51,832 INFO [train.py:901] (2/4) Epoch 10, batch 2600, loss[loss=0.2278, simple_loss=0.3078, pruned_loss=0.0739, over 7966.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3138, pruned_loss=0.08431, over 1609288.01 frames. ], batch size: 21, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:35:55,418 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:36:19,400 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:36:27,351 INFO [train.py:901] (2/4) Epoch 10, batch 2650, loss[loss=0.186, simple_loss=0.259, pruned_loss=0.05653, over 7799.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3142, pruned_loss=0.08439, over 1611191.69 frames. ], batch size: 19, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:36:30,000 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 08:36:40,126 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 08:36:52,043 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9411, 1.5527, 5.9457, 2.3495, 5.4026, 5.0068, 5.5911, 5.4559], + device='cuda:2'), covar=tensor([0.0314, 0.4513, 0.0318, 0.2850, 0.0909, 0.0694, 0.0363, 0.0415], + device='cuda:2'), in_proj_covar=tensor([0.0450, 0.0559, 0.0564, 0.0514, 0.0591, 0.0504, 0.0492, 0.0554], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:36:56,747 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75441.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:36:58,551 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.723e+02 3.413e+02 4.384e+02 8.455e+02, threshold=6.827e+02, percent-clipped=3.0 +2023-02-06 08:37:01,351 INFO [train.py:901] (2/4) Epoch 10, batch 2700, loss[loss=0.1954, simple_loss=0.2698, pruned_loss=0.06054, over 7788.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3126, pruned_loss=0.0831, over 1608204.52 frames. ], batch size: 19, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:37:01,581 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6157, 1.7046, 2.1761, 1.5972, 1.0206, 2.1416, 0.3076, 1.1828], + device='cuda:2'), covar=tensor([0.2800, 0.2018, 0.0510, 0.2115, 0.5282, 0.0576, 0.3949, 0.2367], + device='cuda:2'), in_proj_covar=tensor([0.0162, 0.0163, 0.0093, 0.0215, 0.0255, 0.0098, 0.0163, 0.0159], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:37:02,240 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9105, 2.2169, 1.6734, 2.7545, 1.4443, 1.3838, 1.9748, 2.3699], + device='cuda:2'), covar=tensor([0.1032, 0.0881, 0.1435, 0.0499, 0.1287, 0.1853, 0.0951, 0.0804], + device='cuda:2'), in_proj_covar=tensor([0.0241, 0.0216, 0.0260, 0.0218, 0.0221, 0.0254, 0.0261, 0.0222], + 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') +2023-02-06 08:37:27,127 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7684, 1.8702, 2.2141, 1.7316, 1.1868, 2.2334, 0.2817, 1.3475], + device='cuda:2'), covar=tensor([0.2492, 0.1484, 0.0495, 0.1955, 0.4494, 0.0525, 0.3768, 0.1959], + device='cuda:2'), in_proj_covar=tensor([0.0162, 0.0164, 0.0093, 0.0215, 0.0256, 0.0098, 0.0164, 0.0159], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:37:37,698 INFO [train.py:901] (2/4) Epoch 10, batch 2750, loss[loss=0.2284, simple_loss=0.3013, pruned_loss=0.07779, over 7967.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3134, pruned_loss=0.08324, over 1612978.55 frames. ], batch size: 21, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:37:40,744 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-06 08:37:43,367 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 08:38:08,183 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.609e+02 3.111e+02 3.957e+02 1.084e+03, threshold=6.223e+02, percent-clipped=3.0 +2023-02-06 08:38:10,735 INFO [train.py:901] (2/4) Epoch 10, batch 2800, loss[loss=0.3058, simple_loss=0.3746, pruned_loss=0.1185, over 8189.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3141, pruned_loss=0.08377, over 1612089.07 frames. ], batch size: 23, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:38:14,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3787, 1.8050, 3.2947, 1.1645, 2.3492, 1.8739, 1.4243, 2.1353], + device='cuda:2'), covar=tensor([0.1754, 0.2225, 0.0647, 0.3867, 0.1478, 0.2772, 0.1812, 0.2218], + device='cuda:2'), in_proj_covar=tensor([0.0476, 0.0498, 0.0530, 0.0563, 0.0602, 0.0544, 0.0460, 0.0604], + 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') +2023-02-06 08:38:32,112 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 08:38:39,099 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75590.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:38:45,072 INFO [train.py:901] (2/4) Epoch 10, batch 2850, loss[loss=0.1948, simple_loss=0.2669, pruned_loss=0.06132, over 7450.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.314, pruned_loss=0.08323, over 1612498.32 frames. ], batch size: 17, lr: 7.73e-03, grad_scale: 8.0 +2023-02-06 08:39:04,152 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.50 vs. limit=5.0 +2023-02-06 08:39:09,363 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:16,272 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:17,406 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.670e+02 3.172e+02 3.749e+02 6.038e+02, threshold=6.343e+02, percent-clipped=0.0 +2023-02-06 08:39:20,058 INFO [train.py:901] (2/4) Epoch 10, batch 2900, loss[loss=0.1874, simple_loss=0.2623, pruned_loss=0.05627, over 7792.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3129, pruned_loss=0.08249, over 1607922.25 frames. ], batch size: 19, lr: 7.73e-03, grad_scale: 8.0 +2023-02-06 08:39:32,776 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:49,779 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 08:39:53,280 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75697.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:53,736 INFO [train.py:901] (2/4) Epoch 10, batch 2950, loss[loss=0.2286, simple_loss=0.3143, pruned_loss=0.07147, over 8500.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.313, pruned_loss=0.08198, over 1604657.38 frames. ], batch size: 26, lr: 7.73e-03, grad_scale: 16.0 +2023-02-06 08:39:58,781 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75705.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:59,831 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 08:40:11,379 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:40:26,686 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.480e+02 3.030e+02 3.596e+02 1.304e+03, threshold=6.060e+02, percent-clipped=4.0 +2023-02-06 08:40:28,958 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75747.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:40:29,499 INFO [train.py:901] (2/4) Epoch 10, batch 3000, loss[loss=0.2487, simple_loss=0.3107, pruned_loss=0.09341, over 7653.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.313, pruned_loss=0.08197, over 1606886.12 frames. ], batch size: 19, lr: 7.73e-03, grad_scale: 16.0 +2023-02-06 08:40:29,499 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 08:40:41,881 INFO [train.py:935] (2/4) Epoch 10, validation: loss=0.1918, simple_loss=0.2916, pruned_loss=0.04599, over 944034.00 frames. +2023-02-06 08:40:41,882 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 08:41:02,448 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 08:41:08,740 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 08:41:15,547 INFO [train.py:901] (2/4) Epoch 10, batch 3050, loss[loss=0.2253, simple_loss=0.2944, pruned_loss=0.07809, over 7550.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3128, pruned_loss=0.08218, over 1607206.19 frames. ], batch size: 18, lr: 7.72e-03, grad_scale: 16.0 +2023-02-06 08:41:35,987 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5655, 4.6156, 4.1211, 1.9141, 4.1287, 4.0573, 4.2300, 3.7919], + device='cuda:2'), covar=tensor([0.0825, 0.0615, 0.1138, 0.4930, 0.0843, 0.1034, 0.1201, 0.0920], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0361, 0.0376, 0.0468, 0.0368, 0.0354, 0.0368, 0.0313], + 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') +2023-02-06 08:41:44,322 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 08:41:47,924 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.518e+02 3.138e+02 4.468e+02 1.006e+03, threshold=6.276e+02, percent-clipped=13.0 +2023-02-06 08:41:50,031 INFO [train.py:901] (2/4) Epoch 10, batch 3100, loss[loss=0.1959, simple_loss=0.271, pruned_loss=0.06038, over 7287.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3132, pruned_loss=0.08189, over 1612377.22 frames. ], batch size: 16, lr: 7.72e-03, grad_scale: 8.0 +2023-02-06 08:42:25,562 INFO [train.py:901] (2/4) Epoch 10, batch 3150, loss[loss=0.2268, simple_loss=0.3003, pruned_loss=0.07662, over 8562.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.314, pruned_loss=0.08248, over 1617645.86 frames. ], batch size: 31, lr: 7.72e-03, grad_scale: 8.0 +2023-02-06 08:42:57,478 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.584e+02 3.323e+02 3.941e+02 8.938e+02, threshold=6.646e+02, percent-clipped=3.0 +2023-02-06 08:42:59,536 INFO [train.py:901] (2/4) Epoch 10, batch 3200, loss[loss=0.2198, simple_loss=0.2971, pruned_loss=0.07122, over 7805.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.315, pruned_loss=0.08343, over 1615059.93 frames. ], batch size: 20, lr: 7.72e-03, grad_scale: 8.0 +2023-02-06 08:43:09,324 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75961.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:43:26,966 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0804, 1.2633, 1.4373, 1.1967, 1.0242, 1.3080, 1.7019, 1.4493], + device='cuda:2'), covar=tensor([0.0564, 0.1459, 0.2034, 0.1683, 0.0665, 0.1737, 0.0768, 0.0712], + device='cuda:2'), in_proj_covar=tensor([0.0103, 0.0158, 0.0198, 0.0163, 0.0107, 0.0167, 0.0118, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 08:43:28,326 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75986.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:43:36,368 INFO [train.py:901] (2/4) Epoch 10, batch 3250, loss[loss=0.2393, simple_loss=0.3168, pruned_loss=0.08087, over 8726.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3143, pruned_loss=0.08342, over 1608165.09 frames. ], batch size: 30, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:43:41,262 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76003.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:43:57,863 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:44:05,883 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3711, 1.9538, 3.0909, 2.3771, 2.6090, 2.2486, 1.7821, 1.3747], + device='cuda:2'), covar=tensor([0.3375, 0.3726, 0.0968, 0.2250, 0.1760, 0.1835, 0.1421, 0.3888], + device='cuda:2'), in_proj_covar=tensor([0.0868, 0.0842, 0.0709, 0.0827, 0.0916, 0.0780, 0.0698, 0.0758], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:44:09,020 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.688e+02 3.300e+02 3.989e+02 9.835e+02, threshold=6.601e+02, percent-clipped=4.0 +2023-02-06 08:44:11,016 INFO [train.py:901] (2/4) Epoch 10, batch 3300, loss[loss=0.2354, simple_loss=0.3144, pruned_loss=0.07819, over 8139.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3139, pruned_loss=0.08276, over 1610899.01 frames. ], batch size: 22, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:44:47,567 INFO [train.py:901] (2/4) Epoch 10, batch 3350, loss[loss=0.2314, simple_loss=0.305, pruned_loss=0.07885, over 8325.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3141, pruned_loss=0.08307, over 1609739.53 frames. ], batch size: 25, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:45:01,648 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7763, 3.7741, 2.2474, 2.7809, 2.9090, 1.6834, 2.6955, 2.9741], + device='cuda:2'), covar=tensor([0.1543, 0.0347, 0.1004, 0.0649, 0.0671, 0.1509, 0.1026, 0.0937], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0237, 0.0314, 0.0299, 0.0310, 0.0324, 0.0341, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 08:45:18,633 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.652e+02 3.239e+02 4.192e+02 7.352e+02, threshold=6.477e+02, percent-clipped=1.0 +2023-02-06 08:45:20,669 INFO [train.py:901] (2/4) Epoch 10, batch 3400, loss[loss=0.2454, simple_loss=0.3312, pruned_loss=0.07978, over 8467.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3142, pruned_loss=0.08318, over 1607232.22 frames. ], batch size: 25, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:45:37,971 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0947, 1.4768, 1.5264, 1.4255, 1.1072, 1.3889, 1.7646, 1.7620], + device='cuda:2'), covar=tensor([0.0470, 0.1129, 0.1682, 0.1297, 0.0549, 0.1417, 0.0630, 0.0508], + device='cuda:2'), in_proj_covar=tensor([0.0102, 0.0157, 0.0197, 0.0162, 0.0107, 0.0166, 0.0117, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 08:45:55,812 INFO [train.py:901] (2/4) Epoch 10, batch 3450, loss[loss=0.2537, simple_loss=0.3371, pruned_loss=0.08513, over 8770.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3141, pruned_loss=0.08292, over 1611791.31 frames. ], batch size: 30, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:46:30,137 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.394e+02 3.045e+02 3.881e+02 9.338e+02, threshold=6.090e+02, percent-clipped=3.0 +2023-02-06 08:46:32,218 INFO [train.py:901] (2/4) Epoch 10, batch 3500, loss[loss=0.1727, simple_loss=0.25, pruned_loss=0.04771, over 7222.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3131, pruned_loss=0.08246, over 1609597.83 frames. ], batch size: 16, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:46:48,069 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 08:46:59,307 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76287.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:47:06,727 INFO [train.py:901] (2/4) Epoch 10, batch 3550, loss[loss=0.198, simple_loss=0.2778, pruned_loss=0.05911, over 7662.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3136, pruned_loss=0.08291, over 1607694.96 frames. ], batch size: 19, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:47:14,511 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9173, 1.5352, 5.9418, 2.0604, 5.3173, 4.9546, 5.5132, 5.3656], + device='cuda:2'), covar=tensor([0.0388, 0.4438, 0.0334, 0.3223, 0.0906, 0.0813, 0.0404, 0.0460], + device='cuda:2'), in_proj_covar=tensor([0.0455, 0.0566, 0.0574, 0.0525, 0.0603, 0.0512, 0.0498, 0.0563], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:47:17,269 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:47:39,378 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76341.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:47:42,025 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.725e+02 3.470e+02 4.316e+02 7.747e+02, threshold=6.941e+02, percent-clipped=6.0 +2023-02-06 08:47:44,159 INFO [train.py:901] (2/4) Epoch 10, batch 3600, loss[loss=0.2633, simple_loss=0.3347, pruned_loss=0.09595, over 8673.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.315, pruned_loss=0.08387, over 1613567.68 frames. ], batch size: 49, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:48:01,496 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.6943, 1.3327, 3.9397, 1.5173, 3.3481, 3.2278, 3.4722, 3.3568], + device='cuda:2'), covar=tensor([0.0635, 0.4390, 0.0571, 0.3245, 0.1300, 0.0935, 0.0653, 0.0792], + device='cuda:2'), in_proj_covar=tensor([0.0451, 0.0566, 0.0569, 0.0521, 0.0599, 0.0509, 0.0497, 0.0562], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:48:18,380 INFO [train.py:901] (2/4) Epoch 10, batch 3650, loss[loss=0.1609, simple_loss=0.2405, pruned_loss=0.04067, over 7703.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3132, pruned_loss=0.08305, over 1610503.06 frames. ], batch size: 18, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:48:50,989 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.530e+02 3.057e+02 3.624e+02 8.995e+02, threshold=6.114e+02, percent-clipped=3.0 +2023-02-06 08:48:52,994 INFO [train.py:901] (2/4) Epoch 10, batch 3700, loss[loss=0.1926, simple_loss=0.269, pruned_loss=0.05808, over 7788.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3145, pruned_loss=0.08413, over 1609843.74 frames. ], batch size: 19, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:48:54,939 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 08:49:28,897 INFO [train.py:901] (2/4) Epoch 10, batch 3750, loss[loss=0.2408, simple_loss=0.3217, pruned_loss=0.07992, over 8496.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3151, pruned_loss=0.08483, over 1610761.86 frames. ], batch size: 49, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:50:00,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 2.765e+02 3.416e+02 4.278e+02 1.031e+03, threshold=6.832e+02, percent-clipped=4.0 +2023-02-06 08:50:02,995 INFO [train.py:901] (2/4) Epoch 10, batch 3800, loss[loss=0.2811, simple_loss=0.3455, pruned_loss=0.1083, over 8471.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3169, pruned_loss=0.08569, over 1611622.26 frames. ], batch size: 49, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:50:38,441 INFO [train.py:901] (2/4) Epoch 10, batch 3850, loss[loss=0.2323, simple_loss=0.3115, pruned_loss=0.07652, over 8298.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3157, pruned_loss=0.08505, over 1609619.52 frames. ], batch size: 23, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:50:59,043 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 08:51:00,491 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76631.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:51:09,921 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.563e+02 3.093e+02 4.191e+02 1.151e+03, threshold=6.187e+02, percent-clipped=5.0 +2023-02-06 08:51:11,826 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 08:51:11,981 INFO [train.py:901] (2/4) Epoch 10, batch 3900, loss[loss=0.1878, simple_loss=0.2697, pruned_loss=0.0529, over 7444.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3151, pruned_loss=0.08444, over 1608523.80 frames. ], batch size: 17, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:51:17,438 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76656.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:51:38,044 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76685.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:51:48,068 INFO [train.py:901] (2/4) Epoch 10, batch 3950, loss[loss=0.2622, simple_loss=0.3338, pruned_loss=0.09531, over 8186.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3134, pruned_loss=0.08326, over 1609785.29 frames. ], batch size: 23, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:51:55,075 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5262, 1.6915, 1.9371, 1.4633, 1.0955, 2.0013, 0.1457, 1.1506], + device='cuda:2'), covar=tensor([0.2715, 0.2056, 0.0611, 0.2241, 0.4770, 0.0501, 0.3592, 0.2053], + device='cuda:2'), in_proj_covar=tensor([0.0159, 0.0161, 0.0090, 0.0208, 0.0249, 0.0098, 0.0157, 0.0156], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:52:19,558 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.430e+02 3.097e+02 3.693e+02 7.444e+02, threshold=6.193e+02, percent-clipped=4.0 +2023-02-06 08:52:20,438 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76746.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:52:20,462 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6556, 1.7539, 2.2376, 1.6542, 1.0849, 2.1839, 0.4328, 1.2782], + device='cuda:2'), covar=tensor([0.2702, 0.1769, 0.0492, 0.2426, 0.4903, 0.0553, 0.3718, 0.2429], + device='cuda:2'), in_proj_covar=tensor([0.0161, 0.0163, 0.0092, 0.0211, 0.0253, 0.0099, 0.0160, 0.0158], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:52:21,615 INFO [train.py:901] (2/4) Epoch 10, batch 4000, loss[loss=0.2059, simple_loss=0.2837, pruned_loss=0.06409, over 7818.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3126, pruned_loss=0.08261, over 1610359.01 frames. ], batch size: 20, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:52:37,416 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:52:56,079 INFO [train.py:901] (2/4) Epoch 10, batch 4050, loss[loss=0.2294, simple_loss=0.3048, pruned_loss=0.07695, over 8735.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3137, pruned_loss=0.08309, over 1612775.66 frames. ], batch size: 34, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:52:57,682 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76800.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:53:05,596 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8333, 1.9422, 1.6752, 2.3121, 1.1596, 1.3563, 1.6540, 2.0172], + device='cuda:2'), covar=tensor([0.0777, 0.0917, 0.1119, 0.0531, 0.1226, 0.1575, 0.1000, 0.0939], + device='cuda:2'), in_proj_covar=tensor([0.0236, 0.0214, 0.0256, 0.0217, 0.0220, 0.0250, 0.0260, 0.0221], + 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') +2023-02-06 08:53:29,319 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.637e+02 3.294e+02 4.061e+02 9.505e+02, threshold=6.587e+02, percent-clipped=7.0 +2023-02-06 08:53:31,234 INFO [train.py:901] (2/4) Epoch 10, batch 4100, loss[loss=0.2606, simple_loss=0.3352, pruned_loss=0.09301, over 8191.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.315, pruned_loss=0.08369, over 1614551.61 frames. ], batch size: 23, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:53:37,254 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:54:04,757 INFO [train.py:901] (2/4) Epoch 10, batch 4150, loss[loss=0.2221, simple_loss=0.3048, pruned_loss=0.06963, over 8075.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3156, pruned_loss=0.08381, over 1616839.73 frames. ], batch size: 21, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:54:38,758 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.505e+02 2.967e+02 3.617e+02 8.554e+02, threshold=5.933e+02, percent-clipped=2.0 +2023-02-06 08:54:39,010 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2888, 1.5463, 1.6492, 1.0137, 1.7069, 1.2411, 0.3304, 1.4617], + device='cuda:2'), covar=tensor([0.0328, 0.0222, 0.0167, 0.0286, 0.0208, 0.0585, 0.0535, 0.0146], + device='cuda:2'), in_proj_covar=tensor([0.0369, 0.0302, 0.0262, 0.0367, 0.0296, 0.0454, 0.0346, 0.0331], + device='cuda:2'), out_proj_covar=tensor([1.0711e-04, 8.5533e-05, 7.4512e-05, 1.0493e-04, 8.5728e-05, 1.4126e-04, + 1.0067e-04, 9.5930e-05], device='cuda:2') +2023-02-06 08:54:40,847 INFO [train.py:901] (2/4) Epoch 10, batch 4200, loss[loss=0.238, simple_loss=0.3211, pruned_loss=0.07743, over 8103.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3147, pruned_loss=0.08321, over 1619831.89 frames. ], batch size: 23, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:55:00,917 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 08:55:14,227 INFO [train.py:901] (2/4) Epoch 10, batch 4250, loss[loss=0.2441, simple_loss=0.3261, pruned_loss=0.0811, over 8344.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3147, pruned_loss=0.08344, over 1616000.18 frames. ], batch size: 26, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:55:17,220 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77002.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:55:23,798 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 08:55:31,990 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3689, 1.2834, 2.3695, 1.0897, 2.0034, 2.5420, 2.6089, 2.1331], + device='cuda:2'), covar=tensor([0.0918, 0.1096, 0.0452, 0.1907, 0.0651, 0.0372, 0.0571, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0255, 0.0292, 0.0250, 0.0280, 0.0267, 0.0232, 0.0322, 0.0284], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 08:55:34,051 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:55:34,062 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:55:46,495 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.530e+02 3.131e+02 3.743e+02 6.568e+02, threshold=6.262e+02, percent-clipped=1.0 +2023-02-06 08:55:48,445 INFO [train.py:901] (2/4) Epoch 10, batch 4300, loss[loss=0.2839, simple_loss=0.3477, pruned_loss=0.1101, over 8576.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.315, pruned_loss=0.08382, over 1618556.05 frames. ], batch size: 39, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:55:52,731 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77052.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:55:53,059 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.03 vs. limit=5.0 +2023-02-06 08:55:55,459 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77056.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:56:08,875 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6525, 2.2366, 3.5330, 2.6132, 3.0758, 2.3655, 1.7853, 1.6794], + device='cuda:2'), covar=tensor([0.3533, 0.3980, 0.1165, 0.2535, 0.1945, 0.2053, 0.1698, 0.4477], + device='cuda:2'), in_proj_covar=tensor([0.0867, 0.0844, 0.0710, 0.0826, 0.0912, 0.0773, 0.0695, 0.0750], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:56:12,966 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77081.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:56:18,180 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7755, 3.7606, 3.3605, 1.7942, 3.3039, 3.3497, 3.4173, 3.0866], + device='cuda:2'), covar=tensor([0.0944, 0.0706, 0.1105, 0.4728, 0.0947, 0.0918, 0.1353, 0.0955], + device='cuda:2'), in_proj_covar=tensor([0.0453, 0.0360, 0.0372, 0.0472, 0.0366, 0.0357, 0.0365, 0.0313], + 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') +2023-02-06 08:56:23,970 INFO [train.py:901] (2/4) Epoch 10, batch 4350, loss[loss=0.228, simple_loss=0.2964, pruned_loss=0.07976, over 7783.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3156, pruned_loss=0.08422, over 1618010.98 frames. ], batch size: 19, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:56:33,984 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77113.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:56:53,837 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 08:56:54,997 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.751e+02 3.331e+02 4.491e+02 1.022e+03, threshold=6.663e+02, percent-clipped=8.0 +2023-02-06 08:56:57,037 INFO [train.py:901] (2/4) Epoch 10, batch 4400, loss[loss=0.2136, simple_loss=0.282, pruned_loss=0.07255, over 7445.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.315, pruned_loss=0.08408, over 1611045.03 frames. ], batch size: 17, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:57:02,162 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 08:57:10,783 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.59 vs. limit=5.0 +2023-02-06 08:57:33,158 INFO [train.py:901] (2/4) Epoch 10, batch 4450, loss[loss=0.2248, simple_loss=0.3003, pruned_loss=0.07461, over 8104.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3156, pruned_loss=0.08435, over 1608891.26 frames. ], batch size: 23, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:57:35,380 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:57:36,002 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 08:57:45,673 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4378, 1.9682, 3.0756, 2.3907, 2.5994, 2.1797, 1.7974, 1.4252], + device='cuda:2'), covar=tensor([0.3407, 0.3653, 0.1024, 0.2369, 0.1922, 0.1971, 0.1601, 0.3845], + device='cuda:2'), in_proj_covar=tensor([0.0856, 0.0833, 0.0702, 0.0814, 0.0903, 0.0765, 0.0687, 0.0741], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 08:57:54,214 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:57:55,873 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.30 vs. limit=5.0 +2023-02-06 08:58:04,751 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.784e+02 3.289e+02 4.035e+02 8.452e+02, threshold=6.579e+02, percent-clipped=2.0 +2023-02-06 08:58:06,784 INFO [train.py:901] (2/4) Epoch 10, batch 4500, loss[loss=0.2569, simple_loss=0.3226, pruned_loss=0.09558, over 8276.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.315, pruned_loss=0.08392, over 1611547.43 frames. ], batch size: 23, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:58:27,560 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 08:58:38,001 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:58:41,446 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77295.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:58:43,378 INFO [train.py:901] (2/4) Epoch 10, batch 4550, loss[loss=0.196, simple_loss=0.2694, pruned_loss=0.06136, over 8051.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3134, pruned_loss=0.08317, over 1606530.79 frames. ], batch size: 20, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:58:55,662 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77316.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:59:14,841 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.639e+02 3.213e+02 4.072e+02 8.769e+02, threshold=6.426e+02, percent-clipped=3.0 +2023-02-06 08:59:16,946 INFO [train.py:901] (2/4) Epoch 10, batch 4600, loss[loss=0.2306, simple_loss=0.3079, pruned_loss=0.07667, over 8251.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3136, pruned_loss=0.08248, over 1607626.06 frames. ], batch size: 24, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:59:50,935 INFO [train.py:901] (2/4) Epoch 10, batch 4650, loss[loss=0.2052, simple_loss=0.2688, pruned_loss=0.07076, over 7791.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3141, pruned_loss=0.08361, over 1608738.46 frames. ], batch size: 19, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:00:25,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.953e+02 3.048e+02 3.591e+02 4.434e+02 8.168e+02, threshold=7.182e+02, percent-clipped=8.0 +2023-02-06 09:00:27,546 INFO [train.py:901] (2/4) Epoch 10, batch 4700, loss[loss=0.2514, simple_loss=0.3299, pruned_loss=0.08639, over 8470.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3136, pruned_loss=0.08323, over 1608833.65 frames. ], batch size: 27, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:00:33,878 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77457.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:00:40,479 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9659, 1.5051, 1.6600, 1.2871, 1.0917, 1.4002, 1.6933, 1.4077], + device='cuda:2'), covar=tensor([0.0491, 0.1185, 0.1633, 0.1410, 0.0589, 0.1517, 0.0630, 0.0646], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0194, 0.0160, 0.0106, 0.0165, 0.0118, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 09:01:02,740 INFO [train.py:901] (2/4) Epoch 10, batch 4750, loss[loss=0.2223, simple_loss=0.3067, pruned_loss=0.0689, over 8473.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3144, pruned_loss=0.08331, over 1612308.32 frames. ], batch size: 25, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:01:04,954 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.8859, 1.0518, 0.9781, 0.5851, 1.0511, 0.8153, 0.0798, 1.0172], + device='cuda:2'), covar=tensor([0.0291, 0.0225, 0.0208, 0.0337, 0.0238, 0.0642, 0.0525, 0.0187], + device='cuda:2'), in_proj_covar=tensor([0.0376, 0.0309, 0.0263, 0.0374, 0.0302, 0.0460, 0.0351, 0.0338], + device='cuda:2'), out_proj_covar=tensor([1.0908e-04, 8.7775e-05, 7.5022e-05, 1.0704e-04, 8.7344e-05, 1.4314e-04, + 1.0219e-04, 9.7788e-05], device='cuda:2') +2023-02-06 09:01:28,504 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 09:01:30,522 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 09:01:31,372 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3364, 1.2034, 1.4584, 1.1143, 0.8423, 1.2534, 1.2082, 0.9849], + device='cuda:2'), covar=tensor([0.0546, 0.1320, 0.1678, 0.1462, 0.0582, 0.1570, 0.0693, 0.0687], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0194, 0.0160, 0.0105, 0.0165, 0.0118, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 09:01:35,880 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.648e+02 3.312e+02 4.103e+02 1.054e+03, threshold=6.623e+02, percent-clipped=5.0 +2023-02-06 09:01:37,936 INFO [train.py:901] (2/4) Epoch 10, batch 4800, loss[loss=0.1921, simple_loss=0.2678, pruned_loss=0.05821, over 7809.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3143, pruned_loss=0.08339, over 1612424.44 frames. ], batch size: 20, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:01:54,063 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77572.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:01:54,106 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77572.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:01:54,626 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:04,821 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6265, 1.7679, 2.0076, 1.5896, 1.1557, 2.0501, 0.1732, 1.2167], + device='cuda:2'), covar=tensor([0.2556, 0.1627, 0.0461, 0.1717, 0.4451, 0.0490, 0.3621, 0.1953], + device='cuda:2'), in_proj_covar=tensor([0.0160, 0.0163, 0.0092, 0.0212, 0.0251, 0.0098, 0.0162, 0.0160], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:02:10,897 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77597.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:11,384 INFO [train.py:901] (2/4) Epoch 10, batch 4850, loss[loss=0.2783, simple_loss=0.3432, pruned_loss=0.1067, over 6834.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3145, pruned_loss=0.08399, over 1609685.66 frames. ], batch size: 72, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:02:16,316 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 09:02:38,005 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:38,639 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:42,021 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:46,052 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.616e+02 3.128e+02 3.870e+02 7.279e+02, threshold=6.256e+02, percent-clipped=1.0 +2023-02-06 09:02:48,054 INFO [train.py:901] (2/4) Epoch 10, batch 4900, loss[loss=0.27, simple_loss=0.3326, pruned_loss=0.1037, over 8341.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3145, pruned_loss=0.08388, over 1610623.35 frames. ], batch size: 26, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:03:15,424 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:03:20,013 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:03:21,734 INFO [train.py:901] (2/4) Epoch 10, batch 4950, loss[loss=0.2448, simple_loss=0.326, pruned_loss=0.08179, over 8208.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3148, pruned_loss=0.08382, over 1612194.90 frames. ], batch size: 23, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:03:33,221 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5152, 2.8116, 1.9680, 2.2562, 2.3093, 1.4861, 1.9795, 2.1184], + device='cuda:2'), covar=tensor([0.1322, 0.0300, 0.0900, 0.0539, 0.0579, 0.1396, 0.0954, 0.0932], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0233, 0.0310, 0.0296, 0.0303, 0.0324, 0.0337, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:03:54,515 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 2.775e+02 3.348e+02 4.012e+02 9.680e+02, threshold=6.695e+02, percent-clipped=4.0 +2023-02-06 09:03:57,178 INFO [train.py:901] (2/4) Epoch 10, batch 5000, loss[loss=0.2524, simple_loss=0.3276, pruned_loss=0.08864, over 7931.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3141, pruned_loss=0.08341, over 1607554.08 frames. ], batch size: 20, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:03:58,709 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:04:01,939 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:04:25,735 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9834, 1.6159, 2.2155, 1.7825, 1.9578, 1.8557, 1.5254, 0.7071], + device='cuda:2'), covar=tensor([0.3441, 0.3113, 0.1091, 0.2100, 0.1610, 0.1844, 0.1488, 0.3275], + device='cuda:2'), in_proj_covar=tensor([0.0882, 0.0854, 0.0721, 0.0834, 0.0930, 0.0785, 0.0701, 0.0761], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:04:30,707 INFO [train.py:901] (2/4) Epoch 10, batch 5050, loss[loss=0.2558, simple_loss=0.3234, pruned_loss=0.09411, over 7808.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3146, pruned_loss=0.08343, over 1612964.14 frames. ], batch size: 19, lr: 7.62e-03, grad_scale: 8.0 +2023-02-06 09:04:51,130 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:04:52,869 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 09:05:02,962 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.592e+02 3.051e+02 4.098e+02 9.089e+02, threshold=6.102e+02, percent-clipped=4.0 +2023-02-06 09:05:05,645 INFO [train.py:901] (2/4) Epoch 10, batch 5100, loss[loss=0.2617, simple_loss=0.3309, pruned_loss=0.09621, over 8334.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3133, pruned_loss=0.08254, over 1609225.26 frames. ], batch size: 26, lr: 7.62e-03, grad_scale: 16.0 +2023-02-06 09:05:09,144 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77853.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:05:25,006 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1837, 1.3215, 1.4931, 1.1800, 0.7030, 1.3216, 1.1805, 1.1478], + device='cuda:2'), covar=tensor([0.0526, 0.1225, 0.1609, 0.1344, 0.0526, 0.1415, 0.0648, 0.0605], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0195, 0.0160, 0.0105, 0.0165, 0.0118, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 09:05:31,888 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 09:05:34,963 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4737, 1.5243, 1.7442, 1.3811, 1.0522, 1.7181, 0.0795, 1.1075], + device='cuda:2'), covar=tensor([0.2354, 0.1816, 0.0540, 0.1724, 0.4553, 0.0711, 0.3768, 0.2077], + device='cuda:2'), in_proj_covar=tensor([0.0160, 0.0161, 0.0091, 0.0208, 0.0247, 0.0098, 0.0159, 0.0159], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:05:40,061 INFO [train.py:901] (2/4) Epoch 10, batch 5150, loss[loss=0.2111, simple_loss=0.2696, pruned_loss=0.07627, over 7549.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3129, pruned_loss=0.08235, over 1610142.96 frames. ], batch size: 18, lr: 7.62e-03, grad_scale: 16.0 +2023-02-06 09:06:11,282 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:06:11,738 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.805e+02 3.349e+02 3.898e+02 8.134e+02, threshold=6.697e+02, percent-clipped=4.0 +2023-02-06 09:06:13,803 INFO [train.py:901] (2/4) Epoch 10, batch 5200, loss[loss=0.2714, simple_loss=0.3465, pruned_loss=0.09814, over 8339.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3136, pruned_loss=0.08269, over 1612605.04 frames. ], batch size: 25, lr: 7.62e-03, grad_scale: 16.0 +2023-02-06 09:06:29,755 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:06:35,812 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77978.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:06:50,964 INFO [train.py:901] (2/4) Epoch 10, batch 5250, loss[loss=0.2376, simple_loss=0.3194, pruned_loss=0.07788, over 8548.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3141, pruned_loss=0.08316, over 1613791.48 frames. ], batch size: 39, lr: 7.61e-03, grad_scale: 16.0 +2023-02-06 09:06:56,856 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 09:06:57,793 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:00,625 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78010.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:14,858 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:17,419 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78035.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:19,992 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78039.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:23,911 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.672e+02 3.378e+02 4.041e+02 9.848e+02, threshold=6.756e+02, percent-clipped=3.0 +2023-02-06 09:07:25,979 INFO [train.py:901] (2/4) Epoch 10, batch 5300, loss[loss=0.2074, simple_loss=0.2946, pruned_loss=0.06009, over 8242.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3142, pruned_loss=0.08355, over 1607721.62 frames. ], batch size: 22, lr: 7.61e-03, grad_scale: 16.0 +2023-02-06 09:07:31,454 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0074, 1.9890, 3.9340, 1.6790, 2.5110, 4.5300, 4.5626, 3.7037], + device='cuda:2'), covar=tensor([0.1216, 0.1416, 0.0434, 0.2099, 0.1064, 0.0263, 0.0387, 0.0757], + device='cuda:2'), in_proj_covar=tensor([0.0255, 0.0292, 0.0252, 0.0282, 0.0262, 0.0229, 0.0323, 0.0286], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:07:37,478 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1587, 1.9008, 3.5496, 2.7104, 2.9326, 1.7878, 1.5567, 1.8535], + device='cuda:2'), covar=tensor([0.5471, 0.5540, 0.1100, 0.2621, 0.2578, 0.3683, 0.3026, 0.4561], + device='cuda:2'), in_proj_covar=tensor([0.0871, 0.0847, 0.0709, 0.0824, 0.0918, 0.0778, 0.0694, 0.0750], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:07:43,498 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8393, 2.1676, 1.6426, 2.5672, 1.1208, 1.2994, 1.6881, 2.0865], + device='cuda:2'), covar=tensor([0.0780, 0.0841, 0.1157, 0.0435, 0.1286, 0.1575, 0.1084, 0.0786], + device='cuda:2'), in_proj_covar=tensor([0.0242, 0.0214, 0.0258, 0.0217, 0.0222, 0.0252, 0.0262, 0.0221], + 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') +2023-02-06 09:07:46,779 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0274, 1.4578, 4.2909, 1.7025, 2.2610, 4.9550, 4.9668, 4.1891], + device='cuda:2'), covar=tensor([0.1137, 0.1726, 0.0253, 0.2076, 0.1081, 0.0178, 0.0363, 0.0587], + device='cuda:2'), in_proj_covar=tensor([0.0256, 0.0294, 0.0253, 0.0283, 0.0263, 0.0230, 0.0324, 0.0287], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:07:57,648 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:08:00,797 INFO [train.py:901] (2/4) Epoch 10, batch 5350, loss[loss=0.247, simple_loss=0.3266, pruned_loss=0.08373, over 8503.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3144, pruned_loss=0.08377, over 1609772.78 frames. ], batch size: 26, lr: 7.61e-03, grad_scale: 16.0 +2023-02-06 09:08:12,906 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4834, 2.7844, 1.7854, 2.2232, 2.2283, 1.5610, 2.0622, 2.2304], + device='cuda:2'), covar=tensor([0.1527, 0.0268, 0.1052, 0.0619, 0.0723, 0.1367, 0.0989, 0.0883], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0234, 0.0310, 0.0300, 0.0307, 0.0325, 0.0340, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:08:34,365 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.713e+02 3.238e+02 4.266e+02 6.892e+02, threshold=6.476e+02, percent-clipped=1.0 +2023-02-06 09:08:35,743 INFO [train.py:901] (2/4) Epoch 10, batch 5400, loss[loss=0.2196, simple_loss=0.2958, pruned_loss=0.07173, over 8255.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3153, pruned_loss=0.08406, over 1616278.53 frames. ], batch size: 22, lr: 7.61e-03, grad_scale: 8.0 +2023-02-06 09:08:40,010 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:08:49,864 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3769, 4.4227, 3.9565, 1.9907, 3.7143, 3.8935, 3.9477, 3.5457], + device='cuda:2'), covar=tensor([0.0854, 0.0617, 0.1159, 0.4419, 0.1125, 0.0798, 0.1314, 0.0900], + device='cuda:2'), in_proj_covar=tensor([0.0459, 0.0366, 0.0374, 0.0476, 0.0371, 0.0363, 0.0369, 0.0319], + 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') +2023-02-06 09:09:08,708 INFO [train.py:901] (2/4) Epoch 10, batch 5450, loss[loss=0.2795, simple_loss=0.331, pruned_loss=0.114, over 6553.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3155, pruned_loss=0.08466, over 1612605.73 frames. ], batch size: 71, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:09:43,437 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.697e+02 3.396e+02 4.413e+02 8.943e+02, threshold=6.791e+02, percent-clipped=7.0 +2023-02-06 09:09:44,805 INFO [train.py:901] (2/4) Epoch 10, batch 5500, loss[loss=0.2643, simple_loss=0.3383, pruned_loss=0.09519, over 8523.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3145, pruned_loss=0.08405, over 1613621.56 frames. ], batch size: 28, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:09:46,325 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:09:46,834 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 09:10:18,552 INFO [train.py:901] (2/4) Epoch 10, batch 5550, loss[loss=0.2475, simple_loss=0.3064, pruned_loss=0.09423, over 7248.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3148, pruned_loss=0.08442, over 1610405.46 frames. ], batch size: 16, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:10:52,572 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.563e+02 3.102e+02 4.076e+02 7.679e+02, threshold=6.204e+02, percent-clipped=2.0 +2023-02-06 09:10:54,679 INFO [train.py:901] (2/4) Epoch 10, batch 5600, loss[loss=0.2776, simple_loss=0.3426, pruned_loss=0.1063, over 7203.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3146, pruned_loss=0.08461, over 1610276.85 frames. ], batch size: 71, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:10:55,586 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:11:07,556 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8435, 5.9915, 5.1487, 2.5921, 5.1889, 5.6183, 5.4990, 5.2823], + device='cuda:2'), covar=tensor([0.0565, 0.0400, 0.0931, 0.4402, 0.0763, 0.0720, 0.1048, 0.0637], + device='cuda:2'), in_proj_covar=tensor([0.0454, 0.0357, 0.0371, 0.0472, 0.0366, 0.0361, 0.0368, 0.0317], + 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') +2023-02-06 09:11:12,345 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78374.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:11:28,345 INFO [train.py:901] (2/4) Epoch 10, batch 5650, loss[loss=0.2227, simple_loss=0.2998, pruned_loss=0.07274, over 8250.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3135, pruned_loss=0.08344, over 1610704.82 frames. ], batch size: 22, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:11:32,081 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6156, 4.6068, 4.1189, 1.9838, 4.0405, 4.1180, 4.2157, 3.8446], + device='cuda:2'), covar=tensor([0.0723, 0.0555, 0.1071, 0.4683, 0.0823, 0.0918, 0.1200, 0.0830], + device='cuda:2'), in_proj_covar=tensor([0.0454, 0.0357, 0.0371, 0.0473, 0.0365, 0.0361, 0.0368, 0.0317], + 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') +2023-02-06 09:11:36,902 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:11:48,314 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 09:11:54,463 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:12:02,397 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.443e+02 2.913e+02 3.480e+02 5.594e+02, threshold=5.826e+02, percent-clipped=0.0 +2023-02-06 09:12:03,748 INFO [train.py:901] (2/4) Epoch 10, batch 5700, loss[loss=0.2307, simple_loss=0.3088, pruned_loss=0.07627, over 8032.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3136, pruned_loss=0.08253, over 1617369.98 frames. ], batch size: 22, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:12:16,129 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4733, 1.3872, 2.1740, 1.1846, 1.7197, 2.3984, 2.3578, 2.1404], + device='cuda:2'), covar=tensor([0.0867, 0.1195, 0.0633, 0.1688, 0.1178, 0.0301, 0.0688, 0.0565], + device='cuda:2'), in_proj_covar=tensor([0.0257, 0.0294, 0.0255, 0.0285, 0.0269, 0.0233, 0.0326, 0.0285], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:12:25,608 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:12:36,615 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5739, 2.1473, 3.5061, 1.3427, 2.4852, 2.0408, 1.7349, 2.2819], + device='cuda:2'), covar=tensor([0.1680, 0.1894, 0.0669, 0.3962, 0.1511, 0.2699, 0.1712, 0.2247], + device='cuda:2'), in_proj_covar=tensor([0.0484, 0.0504, 0.0528, 0.0572, 0.0611, 0.0548, 0.0463, 0.0602], + 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') +2023-02-06 09:12:39,148 INFO [train.py:901] (2/4) Epoch 10, batch 5750, loss[loss=0.1992, simple_loss=0.273, pruned_loss=0.06268, over 7926.00 frames. ], tot_loss[loss=0.24, simple_loss=0.314, pruned_loss=0.08301, over 1618080.52 frames. ], batch size: 20, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:12:53,267 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 09:13:11,252 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 2.793e+02 3.478e+02 4.404e+02 1.244e+03, threshold=6.955e+02, percent-clipped=11.0 +2023-02-06 09:13:12,608 INFO [train.py:901] (2/4) Epoch 10, batch 5800, loss[loss=0.2358, simple_loss=0.3029, pruned_loss=0.08434, over 6833.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3132, pruned_loss=0.08242, over 1617939.91 frames. ], batch size: 15, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:13:29,731 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3647, 2.3854, 1.9323, 2.9286, 1.4615, 1.6395, 2.0601, 2.3689], + device='cuda:2'), covar=tensor([0.0616, 0.0825, 0.1040, 0.0381, 0.1148, 0.1421, 0.0978, 0.0811], + device='cuda:2'), in_proj_covar=tensor([0.0239, 0.0211, 0.0257, 0.0216, 0.0217, 0.0253, 0.0256, 0.0221], + 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') +2023-02-06 09:13:31,639 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78574.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:13:45,423 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:13:48,020 INFO [train.py:901] (2/4) Epoch 10, batch 5850, loss[loss=0.2514, simple_loss=0.3138, pruned_loss=0.09449, over 7654.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.314, pruned_loss=0.08266, over 1618405.15 frames. ], batch size: 19, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:14:12,863 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7247, 2.1585, 4.3699, 1.3851, 2.8874, 2.1911, 1.7769, 2.6048], + device='cuda:2'), covar=tensor([0.1637, 0.2192, 0.0748, 0.3711, 0.1614, 0.2752, 0.1726, 0.2497], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0507, 0.0530, 0.0574, 0.0614, 0.0549, 0.0465, 0.0608], + 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') +2023-02-06 09:14:16,179 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2230, 1.4541, 2.1708, 1.1319, 1.4619, 1.5160, 1.3702, 1.3943], + device='cuda:2'), covar=tensor([0.1819, 0.2194, 0.0810, 0.3854, 0.1698, 0.2952, 0.1865, 0.1996], + device='cuda:2'), in_proj_covar=tensor([0.0484, 0.0507, 0.0530, 0.0573, 0.0613, 0.0548, 0.0464, 0.0607], + 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') +2023-02-06 09:14:19,903 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.589e+02 3.164e+02 4.281e+02 9.296e+02, threshold=6.329e+02, percent-clipped=5.0 +2023-02-06 09:14:21,268 INFO [train.py:901] (2/4) Epoch 10, batch 5900, loss[loss=0.2563, simple_loss=0.3225, pruned_loss=0.09506, over 8567.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3128, pruned_loss=0.08208, over 1619376.43 frames. ], batch size: 39, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:14:43,766 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 09:14:44,475 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.15 vs. limit=5.0 +2023-02-06 09:14:57,572 INFO [train.py:901] (2/4) Epoch 10, batch 5950, loss[loss=0.2814, simple_loss=0.3488, pruned_loss=0.107, over 8183.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3124, pruned_loss=0.08157, over 1617186.67 frames. ], batch size: 23, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:15:05,427 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:15:30,052 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.430e+02 2.939e+02 3.954e+02 7.661e+02, threshold=5.878e+02, percent-clipped=3.0 +2023-02-06 09:15:31,441 INFO [train.py:901] (2/4) Epoch 10, batch 6000, loss[loss=0.2025, simple_loss=0.2736, pruned_loss=0.06571, over 6388.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3128, pruned_loss=0.08198, over 1616663.70 frames. ], batch size: 14, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:15:31,441 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 09:15:43,952 INFO [train.py:935] (2/4) Epoch 10, validation: loss=0.1914, simple_loss=0.2907, pruned_loss=0.04604, over 944034.00 frames. +2023-02-06 09:15:43,952 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 09:15:46,395 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 +2023-02-06 09:15:58,896 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 09:16:04,202 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7385, 3.6947, 3.4218, 1.7491, 3.3273, 3.3963, 3.4786, 3.2311], + device='cuda:2'), covar=tensor([0.1087, 0.0706, 0.1199, 0.5267, 0.0938, 0.1066, 0.1426, 0.1012], + device='cuda:2'), in_proj_covar=tensor([0.0447, 0.0356, 0.0365, 0.0470, 0.0360, 0.0358, 0.0364, 0.0314], + 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') +2023-02-06 09:16:18,420 INFO [train.py:901] (2/4) Epoch 10, batch 6050, loss[loss=0.1924, simple_loss=0.2785, pruned_loss=0.05313, over 8234.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3121, pruned_loss=0.08127, over 1619224.39 frames. ], batch size: 22, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:16:35,967 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:16:42,721 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3224, 2.0679, 3.1147, 2.4510, 2.7064, 2.1727, 1.7289, 1.4269], + device='cuda:2'), covar=tensor([0.3761, 0.3833, 0.1023, 0.2375, 0.1936, 0.1983, 0.1663, 0.4079], + device='cuda:2'), in_proj_covar=tensor([0.0873, 0.0846, 0.0708, 0.0819, 0.0916, 0.0775, 0.0692, 0.0745], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:16:52,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.842e+02 3.348e+02 4.641e+02 9.072e+02, threshold=6.696e+02, percent-clipped=15.0 +2023-02-06 09:16:54,184 INFO [train.py:901] (2/4) Epoch 10, batch 6100, loss[loss=0.2901, simple_loss=0.3543, pruned_loss=0.1129, over 8542.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3131, pruned_loss=0.08268, over 1617598.81 frames. ], batch size: 31, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:17:24,370 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 09:17:27,728 INFO [train.py:901] (2/4) Epoch 10, batch 6150, loss[loss=0.2254, simple_loss=0.2934, pruned_loss=0.07867, over 7658.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.312, pruned_loss=0.08225, over 1612822.65 frames. ], batch size: 19, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:17:33,873 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0548, 2.2416, 1.7020, 2.7244, 1.3125, 1.5078, 1.9044, 2.3168], + device='cuda:2'), covar=tensor([0.0766, 0.0806, 0.1161, 0.0416, 0.1292, 0.1484, 0.1073, 0.0745], + device='cuda:2'), in_proj_covar=tensor([0.0236, 0.0208, 0.0254, 0.0216, 0.0217, 0.0251, 0.0254, 0.0220], + 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') +2023-02-06 09:17:41,307 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:17:54,637 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78937.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:01,035 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.486e+02 3.076e+02 3.632e+02 7.166e+02, threshold=6.152e+02, percent-clipped=1.0 +2023-02-06 09:18:02,462 INFO [train.py:901] (2/4) Epoch 10, batch 6200, loss[loss=0.2014, simple_loss=0.279, pruned_loss=0.06189, over 7915.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.312, pruned_loss=0.08188, over 1613116.66 frames. ], batch size: 20, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:18:15,656 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:28,057 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:32,907 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:38,063 INFO [train.py:901] (2/4) Epoch 10, batch 6250, loss[loss=0.3084, simple_loss=0.3649, pruned_loss=0.1259, over 8192.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3127, pruned_loss=0.08255, over 1613948.99 frames. ], batch size: 23, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:19:01,795 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79033.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:19:10,137 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.509e+02 3.177e+02 4.128e+02 1.006e+03, threshold=6.355e+02, percent-clipped=7.0 +2023-02-06 09:19:11,556 INFO [train.py:901] (2/4) Epoch 10, batch 6300, loss[loss=0.2841, simple_loss=0.3541, pruned_loss=0.1071, over 8512.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3133, pruned_loss=0.08298, over 1615312.82 frames. ], batch size: 26, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:19:43,985 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0009, 1.7067, 3.3746, 1.4757, 2.2733, 3.6910, 3.7401, 3.2086], + device='cuda:2'), covar=tensor([0.1015, 0.1409, 0.0327, 0.1969, 0.0995, 0.0219, 0.0394, 0.0600], + device='cuda:2'), in_proj_covar=tensor([0.0255, 0.0293, 0.0253, 0.0285, 0.0267, 0.0234, 0.0325, 0.0289], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:19:47,654 INFO [train.py:901] (2/4) Epoch 10, batch 6350, loss[loss=0.2513, simple_loss=0.3041, pruned_loss=0.09924, over 7546.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3142, pruned_loss=0.08377, over 1617517.92 frames. ], batch size: 18, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:19:56,618 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7284, 2.2597, 3.5685, 2.6490, 2.9692, 2.4339, 1.8793, 1.7192], + device='cuda:2'), covar=tensor([0.3425, 0.3833, 0.1073, 0.2521, 0.2001, 0.1898, 0.1586, 0.4263], + device='cuda:2'), in_proj_covar=tensor([0.0867, 0.0844, 0.0701, 0.0817, 0.0912, 0.0773, 0.0687, 0.0743], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:20:00,245 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 09:20:01,432 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2120, 2.0012, 2.8438, 2.3199, 2.4498, 2.1149, 1.6897, 1.3719], + device='cuda:2'), covar=tensor([0.3707, 0.3588, 0.1060, 0.2315, 0.1937, 0.1955, 0.1751, 0.3767], + device='cuda:2'), in_proj_covar=tensor([0.0865, 0.0842, 0.0701, 0.0816, 0.0911, 0.0772, 0.0685, 0.0742], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:20:20,619 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.998e+02 3.636e+02 4.667e+02 1.201e+03, threshold=7.271e+02, percent-clipped=11.0 +2023-02-06 09:20:21,298 INFO [train.py:901] (2/4) Epoch 10, batch 6400, loss[loss=0.2674, simple_loss=0.3214, pruned_loss=0.1067, over 7550.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3123, pruned_loss=0.08271, over 1612711.33 frames. ], batch size: 18, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:20:54,148 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:20:57,426 INFO [train.py:901] (2/4) Epoch 10, batch 6450, loss[loss=0.23, simple_loss=0.3007, pruned_loss=0.07967, over 8246.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3113, pruned_loss=0.08203, over 1613977.85 frames. ], batch size: 22, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:21:12,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79218.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:21:27,159 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5173, 2.0474, 2.8837, 2.2936, 2.5820, 2.2502, 1.8166, 1.2497], + device='cuda:2'), covar=tensor([0.3026, 0.3439, 0.1023, 0.2303, 0.1630, 0.1804, 0.1512, 0.3667], + device='cuda:2'), in_proj_covar=tensor([0.0871, 0.0846, 0.0707, 0.0821, 0.0918, 0.0775, 0.0690, 0.0748], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:21:31,630 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.577e+02 3.130e+02 4.050e+02 7.383e+02, threshold=6.260e+02, percent-clipped=1.0 +2023-02-06 09:21:32,337 INFO [train.py:901] (2/4) Epoch 10, batch 6500, loss[loss=0.1796, simple_loss=0.2538, pruned_loss=0.05275, over 7706.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3118, pruned_loss=0.08233, over 1616611.41 frames. ], batch size: 18, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:21:49,243 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8683, 2.2250, 3.6498, 2.6093, 3.1735, 2.4095, 1.9665, 1.6027], + device='cuda:2'), covar=tensor([0.3762, 0.4468, 0.1251, 0.2829, 0.1954, 0.2170, 0.1710, 0.4697], + device='cuda:2'), in_proj_covar=tensor([0.0880, 0.0853, 0.0714, 0.0830, 0.0925, 0.0782, 0.0699, 0.0756], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:21:59,866 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79289.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:06,999 INFO [train.py:901] (2/4) Epoch 10, batch 6550, loss[loss=0.2276, simple_loss=0.295, pruned_loss=0.08006, over 8299.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3127, pruned_loss=0.08295, over 1612476.93 frames. ], batch size: 23, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:22:11,151 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79303.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:22:18,368 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79314.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:27,870 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:36,563 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 09:22:41,241 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.767e+02 3.312e+02 4.239e+02 1.073e+03, threshold=6.623e+02, percent-clipped=3.0 +2023-02-06 09:22:41,948 INFO [train.py:901] (2/4) Epoch 10, batch 6600, loss[loss=0.2222, simple_loss=0.2915, pruned_loss=0.07643, over 7809.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3124, pruned_loss=0.08282, over 1611544.06 frames. ], batch size: 20, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:22:44,159 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79351.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:53,888 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 09:23:04,709 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79382.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:23:11,915 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79393.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:23:15,195 INFO [train.py:901] (2/4) Epoch 10, batch 6650, loss[loss=0.2186, simple_loss=0.2963, pruned_loss=0.07042, over 8082.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3131, pruned_loss=0.08333, over 1611494.74 frames. ], batch size: 21, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:23:33,608 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2246, 2.2526, 1.5783, 2.0536, 1.7745, 1.3862, 1.7171, 1.7405], + device='cuda:2'), covar=tensor([0.0968, 0.0298, 0.0860, 0.0373, 0.0503, 0.1109, 0.0624, 0.0677], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0229, 0.0308, 0.0296, 0.0303, 0.0319, 0.0336, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:23:47,670 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79442.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:23:50,865 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.666e+02 3.220e+02 4.193e+02 8.839e+02, threshold=6.440e+02, percent-clipped=3.0 +2023-02-06 09:23:51,584 INFO [train.py:901] (2/4) Epoch 10, batch 6700, loss[loss=0.272, simple_loss=0.3398, pruned_loss=0.1021, over 8613.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3133, pruned_loss=0.08359, over 1612235.01 frames. ], batch size: 39, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:24:24,664 INFO [train.py:901] (2/4) Epoch 10, batch 6750, loss[loss=0.2234, simple_loss=0.3146, pruned_loss=0.0661, over 8572.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3144, pruned_loss=0.08429, over 1608100.45 frames. ], batch size: 39, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:24:28,984 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1767, 3.1764, 2.3990, 2.5405, 2.5519, 2.1535, 2.3460, 2.7569], + device='cuda:2'), covar=tensor([0.1046, 0.0246, 0.0733, 0.0576, 0.0483, 0.0904, 0.0858, 0.0803], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0229, 0.0308, 0.0296, 0.0305, 0.0320, 0.0338, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:25:00,368 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.662e+02 3.188e+02 4.113e+02 8.575e+02, threshold=6.376e+02, percent-clipped=4.0 +2023-02-06 09:25:01,059 INFO [train.py:901] (2/4) Epoch 10, batch 6800, loss[loss=0.237, simple_loss=0.3145, pruned_loss=0.07975, over 8332.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3145, pruned_loss=0.08375, over 1611328.69 frames. ], batch size: 26, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:25:11,665 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 09:25:14,137 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 09:25:19,547 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1631, 2.3420, 1.8137, 2.7970, 1.2255, 1.4496, 1.8329, 2.3643], + device='cuda:2'), covar=tensor([0.0686, 0.0765, 0.1094, 0.0426, 0.1274, 0.1552, 0.1135, 0.0805], + device='cuda:2'), in_proj_covar=tensor([0.0239, 0.0211, 0.0254, 0.0215, 0.0218, 0.0251, 0.0258, 0.0224], + 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') +2023-02-06 09:25:35,818 INFO [train.py:901] (2/4) Epoch 10, batch 6850, loss[loss=0.225, simple_loss=0.3135, pruned_loss=0.0682, over 8491.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.314, pruned_loss=0.08291, over 1612519.04 frames. ], batch size: 26, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:25:39,343 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:25:59,647 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 09:26:10,489 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.544e+02 2.480e+02 2.958e+02 3.519e+02 6.592e+02, threshold=5.916e+02, percent-clipped=1.0 +2023-02-06 09:26:10,581 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79647.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:26:11,119 INFO [train.py:901] (2/4) Epoch 10, batch 6900, loss[loss=0.2577, simple_loss=0.32, pruned_loss=0.09777, over 8077.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3153, pruned_loss=0.08426, over 1610488.26 frames. ], batch size: 21, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:26:30,985 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79675.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:26:44,402 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:26:46,399 INFO [train.py:901] (2/4) Epoch 10, batch 6950, loss[loss=0.2382, simple_loss=0.3214, pruned_loss=0.07743, over 8459.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.315, pruned_loss=0.08368, over 1609524.76 frames. ], batch size: 27, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:26:46,620 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:03,561 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79723.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:05,488 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:10,652 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 09:27:12,697 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79737.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:27:19,256 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.770e+02 3.379e+02 4.019e+02 1.115e+03, threshold=6.759e+02, percent-clipped=8.0 +2023-02-06 09:27:19,980 INFO [train.py:901] (2/4) Epoch 10, batch 7000, loss[loss=0.1901, simple_loss=0.2794, pruned_loss=0.05046, over 8200.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3155, pruned_loss=0.08385, over 1613525.04 frames. ], batch size: 23, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:27:30,364 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79762.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:27:43,460 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79780.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:55,328 INFO [train.py:901] (2/4) Epoch 10, batch 7050, loss[loss=0.2115, simple_loss=0.2849, pruned_loss=0.06901, over 7789.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3153, pruned_loss=0.08379, over 1612889.03 frames. ], batch size: 19, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:28:04,461 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:28:25,532 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:28:26,881 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2184, 1.4040, 1.7170, 1.3246, 1.1153, 1.4647, 1.8778, 1.8324], + device='cuda:2'), covar=tensor([0.0525, 0.1271, 0.1674, 0.1429, 0.0634, 0.1529, 0.0661, 0.0581], + device='cuda:2'), in_proj_covar=tensor([0.0102, 0.0152, 0.0194, 0.0159, 0.0106, 0.0164, 0.0118, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 09:28:29,361 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.704e+02 3.361e+02 4.306e+02 1.362e+03, threshold=6.722e+02, percent-clipped=5.0 +2023-02-06 09:28:30,075 INFO [train.py:901] (2/4) Epoch 10, batch 7100, loss[loss=0.2197, simple_loss=0.2932, pruned_loss=0.07311, over 7969.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3157, pruned_loss=0.08361, over 1617582.88 frames. ], batch size: 21, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:28:32,888 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79852.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:29:06,026 INFO [train.py:901] (2/4) Epoch 10, batch 7150, loss[loss=0.2923, simple_loss=0.3553, pruned_loss=0.1146, over 6988.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3142, pruned_loss=0.08299, over 1610954.83 frames. ], batch size: 72, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:29:39,472 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.702e+02 3.262e+02 4.332e+02 1.613e+03, threshold=6.525e+02, percent-clipped=3.0 +2023-02-06 09:29:39,575 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79947.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:29:40,174 INFO [train.py:901] (2/4) Epoch 10, batch 7200, loss[loss=0.259, simple_loss=0.3327, pruned_loss=0.09262, over 8105.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3141, pruned_loss=0.08282, over 1612640.31 frames. ], batch size: 23, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:30:13,866 INFO [train.py:901] (2/4) Epoch 10, batch 7250, loss[loss=0.2348, simple_loss=0.3284, pruned_loss=0.07062, over 8501.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3138, pruned_loss=0.08309, over 1612674.29 frames. ], batch size: 28, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:30:30,535 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80018.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:30:31,018 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80019.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:30:47,885 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80043.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:30:48,199 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 09:30:50,296 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.755e+02 3.243e+02 3.993e+02 1.489e+03, threshold=6.485e+02, percent-clipped=9.0 +2023-02-06 09:30:50,943 INFO [train.py:901] (2/4) Epoch 10, batch 7300, loss[loss=0.1894, simple_loss=0.2681, pruned_loss=0.05538, over 7541.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.313, pruned_loss=0.08234, over 1615661.31 frames. ], batch size: 18, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:31:00,593 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:03,353 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:04,056 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3503, 1.5610, 2.3130, 1.1487, 1.4795, 1.6057, 1.4116, 1.4414], + device='cuda:2'), covar=tensor([0.1742, 0.2039, 0.0678, 0.3682, 0.1622, 0.2999, 0.1783, 0.1901], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0507, 0.0529, 0.0568, 0.0611, 0.0550, 0.0465, 0.0603], + 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') +2023-02-06 09:31:16,362 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 09:31:20,009 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80091.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:24,180 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:24,604 INFO [train.py:901] (2/4) Epoch 10, batch 7350, loss[loss=0.2563, simple_loss=0.3241, pruned_loss=0.09426, over 8434.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3121, pruned_loss=0.08247, over 1614177.04 frames. ], batch size: 49, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:31:31,683 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80108.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:31:42,410 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:44,301 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80124.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:50,390 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80133.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:31:50,997 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:56,728 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 09:31:58,935 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7977, 1.5707, 5.8827, 1.9742, 5.2470, 4.9642, 5.4748, 5.2732], + device='cuda:2'), covar=tensor([0.0397, 0.4112, 0.0256, 0.3295, 0.0806, 0.0622, 0.0365, 0.0447], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0553, 0.0554, 0.0505, 0.0578, 0.0489, 0.0485, 0.0550], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:31:59,433 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.519e+02 3.343e+02 4.224e+02 9.659e+02, threshold=6.686e+02, percent-clipped=6.0 +2023-02-06 09:32:00,145 INFO [train.py:901] (2/4) Epoch 10, batch 7400, loss[loss=0.21, simple_loss=0.2903, pruned_loss=0.06482, over 8239.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3116, pruned_loss=0.08207, over 1606156.34 frames. ], batch size: 22, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:32:16,188 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 09:32:34,307 INFO [train.py:901] (2/4) Epoch 10, batch 7450, loss[loss=0.3052, simple_loss=0.3759, pruned_loss=0.1173, over 8647.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3114, pruned_loss=0.08168, over 1607401.66 frames. ], batch size: 39, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:32:35,892 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1746, 1.3297, 3.3090, 0.9572, 2.8500, 2.7322, 3.0014, 2.9041], + device='cuda:2'), covar=tensor([0.0687, 0.3452, 0.0744, 0.3243, 0.1353, 0.0969, 0.0677, 0.0785], + device='cuda:2'), in_proj_covar=tensor([0.0462, 0.0558, 0.0561, 0.0511, 0.0584, 0.0492, 0.0489, 0.0555], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:32:39,985 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8502, 2.1689, 4.0508, 1.5493, 2.9784, 2.2259, 1.8434, 2.6693], + device='cuda:2'), covar=tensor([0.1550, 0.2209, 0.0691, 0.3618, 0.1500, 0.2682, 0.1691, 0.2301], + device='cuda:2'), in_proj_covar=tensor([0.0488, 0.0511, 0.0532, 0.0574, 0.0614, 0.0554, 0.0467, 0.0606], + 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') +2023-02-06 09:32:54,358 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 09:33:02,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:33:09,145 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.637e+02 3.217e+02 3.901e+02 6.824e+02, threshold=6.433e+02, percent-clipped=2.0 +2023-02-06 09:33:09,858 INFO [train.py:901] (2/4) Epoch 10, batch 7500, loss[loss=0.2242, simple_loss=0.2971, pruned_loss=0.07563, over 7672.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3112, pruned_loss=0.08239, over 1601562.29 frames. ], batch size: 19, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:33:19,687 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 09:33:43,947 INFO [train.py:901] (2/4) Epoch 10, batch 7550, loss[loss=0.1995, simple_loss=0.2851, pruned_loss=0.05698, over 7659.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3128, pruned_loss=0.08255, over 1607691.06 frames. ], batch size: 19, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:33:46,243 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:33:57,962 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80318.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:34:15,033 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:34:17,459 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.399e+02 2.908e+02 3.933e+02 1.078e+03, threshold=5.816e+02, percent-clipped=3.0 +2023-02-06 09:34:18,143 INFO [train.py:901] (2/4) Epoch 10, batch 7600, loss[loss=0.3066, simple_loss=0.3864, pruned_loss=0.1134, over 8518.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.313, pruned_loss=0.08241, over 1609787.07 frames. ], batch size: 28, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:34:43,239 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3141, 1.4658, 1.3270, 1.8675, 0.7277, 1.2227, 1.2926, 1.4920], + device='cuda:2'), covar=tensor([0.0981, 0.0976, 0.1280, 0.0607, 0.1221, 0.1618, 0.0869, 0.0767], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0211, 0.0253, 0.0215, 0.0217, 0.0249, 0.0255, 0.0225], + 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') +2023-02-06 09:34:47,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2165, 1.2819, 1.5046, 1.1539, 0.8360, 1.3257, 1.3353, 0.9762], + device='cuda:2'), covar=tensor([0.0578, 0.1259, 0.1760, 0.1438, 0.0553, 0.1539, 0.0677, 0.0690], + device='cuda:2'), in_proj_covar=tensor([0.0102, 0.0154, 0.0195, 0.0160, 0.0105, 0.0164, 0.0119, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 09:34:49,208 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80390.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:34:54,343 INFO [train.py:901] (2/4) Epoch 10, batch 7650, loss[loss=0.2873, simple_loss=0.3495, pruned_loss=0.1126, over 8344.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3153, pruned_loss=0.08414, over 1613481.14 frames. ], batch size: 26, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:35:03,331 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80411.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:35:06,028 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4795, 4.4409, 4.0231, 1.9918, 3.9638, 4.1015, 4.1219, 3.7362], + device='cuda:2'), covar=tensor([0.0886, 0.0579, 0.0980, 0.5203, 0.0898, 0.0837, 0.1211, 0.0853], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0359, 0.0377, 0.0475, 0.0374, 0.0362, 0.0365, 0.0319], + 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') +2023-02-06 09:35:06,133 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80415.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:35:12,130 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6097, 1.2301, 2.7982, 1.3115, 1.9807, 2.9822, 3.0284, 2.6048], + device='cuda:2'), covar=tensor([0.1047, 0.1554, 0.0413, 0.1962, 0.0925, 0.0318, 0.0585, 0.0627], + device='cuda:2'), in_proj_covar=tensor([0.0259, 0.0294, 0.0255, 0.0285, 0.0268, 0.0232, 0.0330, 0.0288], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:35:27,280 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.621e+02 3.149e+02 3.913e+02 9.838e+02, threshold=6.298e+02, percent-clipped=6.0 +2023-02-06 09:35:27,982 INFO [train.py:901] (2/4) Epoch 10, batch 7700, loss[loss=0.2381, simple_loss=0.3152, pruned_loss=0.08043, over 8481.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3145, pruned_loss=0.08366, over 1614710.63 frames. ], batch size: 49, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:35:51,539 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:01,535 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80495.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:03,341 INFO [train.py:901] (2/4) Epoch 10, batch 7750, loss[loss=0.2019, simple_loss=0.2931, pruned_loss=0.05531, over 8246.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3144, pruned_loss=0.08342, over 1611989.69 frames. ], batch size: 24, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:36:06,751 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 09:36:12,267 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.56 vs. limit=5.0 +2023-02-06 09:36:13,276 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80512.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:18,621 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:36,346 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.713e+02 3.406e+02 4.090e+02 8.759e+02, threshold=6.812e+02, percent-clipped=3.0 +2023-02-06 09:36:37,057 INFO [train.py:901] (2/4) Epoch 10, batch 7800, loss[loss=0.2101, simple_loss=0.2873, pruned_loss=0.06649, over 7644.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3146, pruned_loss=0.08347, over 1613320.43 frames. ], batch size: 19, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:37:04,766 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1906, 3.0466, 3.5241, 2.1118, 1.7908, 3.6845, 0.5750, 2.3407], + device='cuda:2'), covar=tensor([0.2080, 0.1189, 0.0496, 0.3071, 0.4860, 0.0424, 0.4191, 0.2227], + device='cuda:2'), in_proj_covar=tensor([0.0164, 0.0163, 0.0096, 0.0211, 0.0255, 0.0101, 0.0162, 0.0161], + 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') +2023-02-06 09:37:09,858 INFO [train.py:901] (2/4) Epoch 10, batch 7850, loss[loss=0.2335, simple_loss=0.3135, pruned_loss=0.07671, over 8195.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3152, pruned_loss=0.08392, over 1611447.63 frames. ], batch size: 23, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:37:40,875 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:37:42,726 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.556e+02 3.372e+02 4.255e+02 7.191e+02, threshold=6.744e+02, percent-clipped=1.0 +2023-02-06 09:37:43,427 INFO [train.py:901] (2/4) Epoch 10, batch 7900, loss[loss=0.2279, simple_loss=0.3119, pruned_loss=0.07191, over 8321.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3151, pruned_loss=0.08357, over 1615649.19 frames. ], batch size: 25, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:38:02,367 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4344, 1.6661, 4.2462, 1.9917, 2.4121, 4.9082, 4.9020, 4.2505], + device='cuda:2'), covar=tensor([0.0931, 0.1497, 0.0290, 0.1856, 0.1044, 0.0181, 0.0357, 0.0492], + device='cuda:2'), in_proj_covar=tensor([0.0255, 0.0289, 0.0253, 0.0282, 0.0266, 0.0230, 0.0326, 0.0283], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 09:38:16,932 INFO [train.py:901] (2/4) Epoch 10, batch 7950, loss[loss=0.2071, simple_loss=0.2843, pruned_loss=0.06493, over 7944.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3139, pruned_loss=0.08271, over 1612088.04 frames. ], batch size: 20, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:38:50,733 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.443e+02 2.660e+02 3.023e+02 3.700e+02 9.606e+02, threshold=6.046e+02, percent-clipped=2.0 +2023-02-06 09:38:51,443 INFO [train.py:901] (2/4) Epoch 10, batch 8000, loss[loss=0.1811, simple_loss=0.2661, pruned_loss=0.04802, over 7808.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3136, pruned_loss=0.08268, over 1613399.43 frames. ], batch size: 20, lr: 7.48e-03, grad_scale: 8.0 +2023-02-06 09:38:55,693 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:38:56,305 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:38:59,808 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:39:20,380 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:39:24,988 INFO [train.py:901] (2/4) Epoch 10, batch 8050, loss[loss=0.2132, simple_loss=0.2882, pruned_loss=0.0691, over 7786.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3136, pruned_loss=0.08276, over 1610932.20 frames. ], batch size: 19, lr: 7.48e-03, grad_scale: 8.0 +2023-02-06 09:39:43,568 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:39:57,914 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 09:40:01,792 INFO [train.py:901] (2/4) Epoch 11, batch 0, loss[loss=0.3053, simple_loss=0.3629, pruned_loss=0.1239, over 8363.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3629, pruned_loss=0.1239, over 8363.00 frames. ], batch size: 24, lr: 7.14e-03, grad_scale: 8.0 +2023-02-06 09:40:01,793 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 09:40:13,090 INFO [train.py:935] (2/4) Epoch 11, validation: loss=0.1907, simple_loss=0.2907, pruned_loss=0.04534, over 944034.00 frames. +2023-02-06 09:40:13,091 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 09:40:23,922 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.899e+02 3.439e+02 4.416e+02 1.589e+03, threshold=6.879e+02, percent-clipped=9.0 +2023-02-06 09:40:27,458 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 09:40:30,152 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:40:39,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80870.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:40:47,948 INFO [train.py:901] (2/4) Epoch 11, batch 50, loss[loss=0.2755, simple_loss=0.3429, pruned_loss=0.104, over 8589.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3273, pruned_loss=0.08782, over 373619.91 frames. ], batch size: 31, lr: 7.14e-03, grad_scale: 8.0 +2023-02-06 09:40:52,491 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-06 09:41:03,889 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 09:41:04,383 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 09:41:24,352 INFO [train.py:901] (2/4) Epoch 11, batch 100, loss[loss=0.2029, simple_loss=0.2881, pruned_loss=0.05887, over 8233.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3216, pruned_loss=0.08551, over 654749.51 frames. ], batch size: 22, lr: 7.14e-03, grad_scale: 8.0 +2023-02-06 09:41:29,240 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 09:41:30,711 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80940.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:41:35,314 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.679e+02 3.187e+02 3.933e+02 1.063e+03, threshold=6.374e+02, percent-clipped=2.0 +2023-02-06 09:41:51,857 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80971.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:41:58,396 INFO [train.py:901] (2/4) Epoch 11, batch 150, loss[loss=0.278, simple_loss=0.3531, pruned_loss=0.1015, over 8248.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.319, pruned_loss=0.08468, over 865759.80 frames. ], batch size: 24, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:42:23,721 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:42:34,640 INFO [train.py:901] (2/4) Epoch 11, batch 200, loss[loss=0.2232, simple_loss=0.3039, pruned_loss=0.07128, over 7644.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3171, pruned_loss=0.08373, over 1031331.71 frames. ], batch size: 19, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:42:37,863 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 09:42:42,989 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81041.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:42:47,013 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.662e+02 3.186e+02 4.005e+02 8.686e+02, threshold=6.371e+02, percent-clipped=5.0 +2023-02-06 09:43:10,552 INFO [train.py:901] (2/4) Epoch 11, batch 250, loss[loss=0.2641, simple_loss=0.3427, pruned_loss=0.09274, over 8486.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3162, pruned_loss=0.08312, over 1164622.25 frames. ], batch size: 49, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:43:18,698 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 09:43:21,542 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 09:43:21,965 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 09:43:22,306 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81098.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:31,353 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 09:43:42,712 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81126.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:46,010 INFO [train.py:901] (2/4) Epoch 11, batch 300, loss[loss=0.2576, simple_loss=0.3351, pruned_loss=0.09009, over 8596.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3142, pruned_loss=0.08237, over 1260635.14 frames. ], batch size: 39, lr: 7.13e-03, grad_scale: 16.0 +2023-02-06 09:43:48,260 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:48,874 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:57,132 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.697e+02 3.136e+02 4.054e+02 9.565e+02, threshold=6.271e+02, percent-clipped=1.0 +2023-02-06 09:44:00,772 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81151.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:22,507 INFO [train.py:901] (2/4) Epoch 11, batch 350, loss[loss=0.2601, simple_loss=0.3267, pruned_loss=0.09671, over 7802.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3137, pruned_loss=0.08195, over 1340629.96 frames. ], batch size: 20, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:44:32,494 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 09:44:33,013 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81196.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:39,225 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1177, 1.7625, 2.4916, 2.0467, 2.2265, 2.0435, 1.7371, 1.0064], + device='cuda:2'), covar=tensor([0.4033, 0.3948, 0.1216, 0.2310, 0.1975, 0.2312, 0.1690, 0.4098], + device='cuda:2'), in_proj_covar=tensor([0.0892, 0.0863, 0.0720, 0.0833, 0.0935, 0.0794, 0.0704, 0.0762], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:44:44,354 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:46,356 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0436, 1.4058, 1.6764, 1.3420, 1.0982, 1.4210, 1.7531, 1.4116], + device='cuda:2'), covar=tensor([0.0466, 0.1212, 0.1629, 0.1327, 0.0565, 0.1413, 0.0667, 0.0595], + device='cuda:2'), in_proj_covar=tensor([0.0102, 0.0155, 0.0195, 0.0159, 0.0106, 0.0164, 0.0118, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 09:44:49,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:53,751 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81227.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:56,286 INFO [train.py:901] (2/4) Epoch 11, batch 400, loss[loss=0.2344, simple_loss=0.3254, pruned_loss=0.07166, over 8323.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3148, pruned_loss=0.08298, over 1401715.24 frames. ], batch size: 25, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:45:00,799 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 09:45:08,650 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.601e+02 3.216e+02 4.274e+02 6.931e+02, threshold=6.433e+02, percent-clipped=2.0 +2023-02-06 09:45:10,280 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:45:11,755 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:45:23,640 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9575, 2.5220, 2.9718, 1.2569, 3.0042, 1.7505, 1.3435, 1.7709], + device='cuda:2'), covar=tensor([0.0565, 0.0225, 0.0165, 0.0501, 0.0284, 0.0602, 0.0644, 0.0361], + device='cuda:2'), in_proj_covar=tensor([0.0373, 0.0304, 0.0259, 0.0372, 0.0293, 0.0457, 0.0345, 0.0339], + device='cuda:2'), out_proj_covar=tensor([1.0753e-04, 8.5703e-05, 7.3022e-05, 1.0561e-04, 8.3944e-05, 1.4109e-04, + 9.9503e-05, 9.6983e-05], device='cuda:2') +2023-02-06 09:45:29,082 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4186, 2.6316, 1.9250, 2.3165, 2.2175, 1.5239, 2.1689, 2.2088], + device='cuda:2'), covar=tensor([0.1433, 0.0409, 0.0968, 0.0545, 0.0611, 0.1460, 0.0967, 0.0873], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0233, 0.0308, 0.0296, 0.0303, 0.0325, 0.0338, 0.0304], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:45:32,911 INFO [train.py:901] (2/4) Epoch 11, batch 450, loss[loss=0.2331, simple_loss=0.3177, pruned_loss=0.07421, over 8518.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.315, pruned_loss=0.08289, over 1448181.48 frames. ], batch size: 49, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:45:57,086 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81317.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:45:58,457 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:46:06,250 INFO [train.py:901] (2/4) Epoch 11, batch 500, loss[loss=0.1861, simple_loss=0.2675, pruned_loss=0.0523, over 7451.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3134, pruned_loss=0.08231, over 1482592.57 frames. ], batch size: 17, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:46:07,904 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.73 vs. limit=5.0 +2023-02-06 09:46:17,539 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.501e+02 3.364e+02 4.069e+02 6.845e+02, threshold=6.728e+02, percent-clipped=2.0 +2023-02-06 09:46:40,092 INFO [train.py:901] (2/4) Epoch 11, batch 550, loss[loss=0.2323, simple_loss=0.312, pruned_loss=0.07633, over 8321.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.313, pruned_loss=0.08235, over 1513077.88 frames. ], batch size: 26, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:47:04,570 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4320, 1.9294, 2.7727, 2.1918, 2.5729, 2.2347, 1.8784, 1.1533], + device='cuda:2'), covar=tensor([0.3677, 0.3746, 0.1136, 0.2380, 0.1705, 0.2075, 0.1733, 0.4009], + device='cuda:2'), in_proj_covar=tensor([0.0886, 0.0853, 0.0717, 0.0828, 0.0927, 0.0789, 0.0701, 0.0759], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:47:15,692 INFO [train.py:901] (2/4) Epoch 11, batch 600, loss[loss=0.2388, simple_loss=0.3348, pruned_loss=0.0714, over 8479.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3126, pruned_loss=0.08217, over 1534756.00 frames. ], batch size: 29, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:47:27,354 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.633e+02 3.080e+02 3.885e+02 6.931e+02, threshold=6.160e+02, percent-clipped=1.0 +2023-02-06 09:47:27,533 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5472, 1.7497, 4.6501, 2.3915, 2.5287, 5.2996, 5.1936, 4.6987], + device='cuda:2'), covar=tensor([0.0876, 0.1434, 0.0234, 0.1533, 0.1000, 0.0160, 0.0406, 0.0478], + device='cuda:2'), in_proj_covar=tensor([0.0257, 0.0286, 0.0247, 0.0278, 0.0264, 0.0226, 0.0322, 0.0280], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 09:47:27,559 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.3637, 1.9569, 2.2729, 2.1120, 1.2502, 2.2338, 2.5316, 2.9131], + device='cuda:2'), covar=tensor([0.0372, 0.1112, 0.1544, 0.1162, 0.0588, 0.1218, 0.0588, 0.0382], + device='cuda:2'), in_proj_covar=tensor([0.0102, 0.0155, 0.0196, 0.0159, 0.0106, 0.0165, 0.0118, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 09:47:27,982 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 09:47:35,475 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 09:47:42,469 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:47:48,471 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:47:50,444 INFO [train.py:901] (2/4) Epoch 11, batch 650, loss[loss=0.264, simple_loss=0.329, pruned_loss=0.0995, over 8647.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3132, pruned_loss=0.08246, over 1557042.37 frames. ], batch size: 31, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:47:59,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:48:04,889 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5927, 2.7699, 1.9000, 2.2740, 2.1849, 1.5737, 2.1328, 2.1658], + device='cuda:2'), covar=tensor([0.1327, 0.0290, 0.0958, 0.0592, 0.0644, 0.1258, 0.0900, 0.0868], + device='cuda:2'), in_proj_covar=tensor([0.0340, 0.0228, 0.0304, 0.0295, 0.0297, 0.0318, 0.0332, 0.0300], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 09:48:08,328 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:48:26,931 INFO [train.py:901] (2/4) Epoch 11, batch 700, loss[loss=0.2516, simple_loss=0.322, pruned_loss=0.09061, over 8501.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.312, pruned_loss=0.08108, over 1569898.87 frames. ], batch size: 26, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:48:27,089 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:48:28,483 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5560, 2.3127, 4.0493, 1.1736, 2.9053, 1.9563, 1.8128, 2.3765], + device='cuda:2'), covar=tensor([0.2010, 0.2319, 0.0967, 0.4614, 0.1912, 0.3289, 0.1938, 0.3171], + device='cuda:2'), in_proj_covar=tensor([0.0480, 0.0504, 0.0526, 0.0567, 0.0609, 0.0541, 0.0466, 0.0608], + 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') +2023-02-06 09:48:38,763 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.541e+02 3.049e+02 3.626e+02 6.264e+02, threshold=6.097e+02, percent-clipped=1.0 +2023-02-06 09:49:01,394 INFO [train.py:901] (2/4) Epoch 11, batch 750, loss[loss=0.2631, simple_loss=0.3362, pruned_loss=0.09499, over 8245.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3128, pruned_loss=0.08123, over 1585233.95 frames. ], batch size: 24, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:49:07,838 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 09:49:10,210 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81593.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:49:10,586 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 09:49:21,640 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3684, 1.9007, 3.0371, 2.3517, 2.6413, 2.1959, 1.6982, 1.4321], + device='cuda:2'), covar=tensor([0.3790, 0.4413, 0.1127, 0.2463, 0.2015, 0.2226, 0.1742, 0.4136], + device='cuda:2'), in_proj_covar=tensor([0.0885, 0.0855, 0.0718, 0.0833, 0.0931, 0.0790, 0.0701, 0.0761], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:49:24,833 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 09:49:25,689 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3154, 3.0773, 2.3798, 3.8525, 1.6972, 1.9159, 2.2739, 3.0074], + device='cuda:2'), covar=tensor([0.0827, 0.0756, 0.0954, 0.0269, 0.1235, 0.1501, 0.1184, 0.0784], + device='cuda:2'), in_proj_covar=tensor([0.0239, 0.0212, 0.0254, 0.0216, 0.0216, 0.0254, 0.0257, 0.0226], + 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') +2023-02-06 09:49:29,020 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3713, 4.4655, 3.9336, 1.8878, 3.9308, 3.9261, 4.1081, 3.7041], + device='cuda:2'), covar=tensor([0.0986, 0.0617, 0.1137, 0.5192, 0.0841, 0.0845, 0.1392, 0.0876], + device='cuda:2'), in_proj_covar=tensor([0.0465, 0.0368, 0.0379, 0.0479, 0.0373, 0.0365, 0.0367, 0.0322], + 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') +2023-02-06 09:49:33,742 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 09:49:36,445 INFO [train.py:901] (2/4) Epoch 11, batch 800, loss[loss=0.2185, simple_loss=0.3036, pruned_loss=0.06669, over 8334.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3126, pruned_loss=0.08097, over 1593435.58 frames. ], batch size: 26, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:49:49,260 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.638e+02 2.494e+02 2.971e+02 3.970e+02 9.403e+02, threshold=5.941e+02, percent-clipped=2.0 +2023-02-06 09:49:58,273 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:49:59,645 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81663.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:50:11,556 INFO [train.py:901] (2/4) Epoch 11, batch 850, loss[loss=0.2616, simple_loss=0.3266, pruned_loss=0.09829, over 6700.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3134, pruned_loss=0.08171, over 1595743.27 frames. ], batch size: 71, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:50:13,988 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 09:50:16,509 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:50:39,504 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6276, 1.6911, 1.9454, 1.5512, 1.0745, 2.0730, 0.2120, 1.2386], + device='cuda:2'), covar=tensor([0.3231, 0.1821, 0.0604, 0.1938, 0.4516, 0.0544, 0.3788, 0.1914], + device='cuda:2'), in_proj_covar=tensor([0.0162, 0.0163, 0.0094, 0.0210, 0.0252, 0.0102, 0.0161, 0.0159], + 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') +2023-02-06 09:50:46,078 INFO [train.py:901] (2/4) Epoch 11, batch 900, loss[loss=0.2789, simple_loss=0.3483, pruned_loss=0.1048, over 8344.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3123, pruned_loss=0.08117, over 1601449.01 frames. ], batch size: 26, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:50:58,849 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.543e+02 3.289e+02 4.286e+02 9.063e+02, threshold=6.577e+02, percent-clipped=7.0 +2023-02-06 09:51:14,957 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 09:51:18,695 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81776.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:51:20,087 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:51:21,852 INFO [train.py:901] (2/4) Epoch 11, batch 950, loss[loss=0.2591, simple_loss=0.3308, pruned_loss=0.09369, over 8451.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3125, pruned_loss=0.08148, over 1605990.46 frames. ], batch size: 27, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:51:28,235 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4986, 2.0196, 3.0960, 2.3929, 2.8916, 2.2960, 1.9004, 1.4788], + device='cuda:2'), covar=tensor([0.3790, 0.4188, 0.1079, 0.2586, 0.1711, 0.2291, 0.1755, 0.4239], + device='cuda:2'), in_proj_covar=tensor([0.0887, 0.0856, 0.0721, 0.0833, 0.0925, 0.0793, 0.0705, 0.0760], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 09:51:51,860 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 09:51:56,016 INFO [train.py:901] (2/4) Epoch 11, batch 1000, loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.06362, over 8326.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.312, pruned_loss=0.08162, over 1609673.00 frames. ], batch size: 25, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:52:07,438 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.713e+02 3.211e+02 4.023e+02 7.481e+02, threshold=6.422e+02, percent-clipped=3.0 +2023-02-06 09:52:08,402 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81849.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:52:12,966 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1463, 2.4648, 1.9220, 2.8622, 1.3433, 1.6697, 1.9214, 2.3242], + device='cuda:2'), covar=tensor([0.0734, 0.0783, 0.0958, 0.0437, 0.1257, 0.1314, 0.1143, 0.0873], + device='cuda:2'), in_proj_covar=tensor([0.0240, 0.0213, 0.0256, 0.0218, 0.0219, 0.0254, 0.0258, 0.0225], + 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') +2023-02-06 09:52:27,193 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 09:52:27,397 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:52:31,956 INFO [train.py:901] (2/4) Epoch 11, batch 1050, loss[loss=0.2965, simple_loss=0.3521, pruned_loss=0.1204, over 8526.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3117, pruned_loss=0.08137, over 1612885.41 frames. ], batch size: 28, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:52:39,047 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 09:53:06,144 INFO [train.py:901] (2/4) Epoch 11, batch 1100, loss[loss=0.2597, simple_loss=0.3347, pruned_loss=0.09241, over 8510.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3103, pruned_loss=0.08026, over 1616757.40 frames. ], batch size: 34, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:53:18,517 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.436e+02 2.887e+02 3.709e+02 9.106e+02, threshold=5.774e+02, percent-clipped=2.0 +2023-02-06 09:53:41,608 INFO [train.py:901] (2/4) Epoch 11, batch 1150, loss[loss=0.2608, simple_loss=0.3349, pruned_loss=0.09335, over 8320.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3118, pruned_loss=0.08078, over 1625391.29 frames. ], batch size: 25, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:53:51,787 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 09:54:00,452 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:17,961 INFO [train.py:901] (2/4) Epoch 11, batch 1200, loss[loss=0.204, simple_loss=0.2957, pruned_loss=0.05618, over 8361.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3112, pruned_loss=0.0806, over 1619391.46 frames. ], batch size: 24, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:54:18,730 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:18,854 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:20,248 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82034.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:29,502 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.664e+02 3.172e+02 3.772e+02 1.117e+03, threshold=6.345e+02, percent-clipped=5.0 +2023-02-06 09:54:36,787 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:38,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:47,992 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82073.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:53,469 INFO [train.py:901] (2/4) Epoch 11, batch 1250, loss[loss=0.2696, simple_loss=0.3443, pruned_loss=0.09742, over 8187.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3109, pruned_loss=0.08027, over 1619349.85 frames. ], batch size: 23, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:55:29,356 INFO [train.py:901] (2/4) Epoch 11, batch 1300, loss[loss=0.1983, simple_loss=0.2836, pruned_loss=0.05647, over 7805.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3107, pruned_loss=0.0797, over 1619992.14 frames. ], batch size: 20, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:55:40,413 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82147.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:55:40,854 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.583e+02 3.223e+02 4.179e+02 7.623e+02, threshold=6.447e+02, percent-clipped=2.0 +2023-02-06 09:55:42,377 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2390, 1.5108, 1.2560, 2.2169, 1.0581, 1.1044, 1.6503, 1.6824], + device='cuda:2'), covar=tensor([0.1701, 0.1407, 0.2064, 0.0604, 0.1491, 0.2102, 0.0954, 0.0928], + device='cuda:2'), in_proj_covar=tensor([0.0239, 0.0214, 0.0255, 0.0217, 0.0218, 0.0254, 0.0254, 0.0222], + 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') +2023-02-06 09:56:03,694 INFO [train.py:901] (2/4) Epoch 11, batch 1350, loss[loss=0.2525, simple_loss=0.3377, pruned_loss=0.08365, over 8475.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3119, pruned_loss=0.08073, over 1621471.33 frames. ], batch size: 25, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:56:29,754 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0715, 1.4404, 1.5737, 1.3668, 1.0326, 1.3697, 1.7878, 1.5210], + device='cuda:2'), covar=tensor([0.0466, 0.1318, 0.1753, 0.1389, 0.0588, 0.1585, 0.0678, 0.0637], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0153, 0.0194, 0.0157, 0.0104, 0.0164, 0.0116, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:2') +2023-02-06 09:56:38,845 INFO [train.py:901] (2/4) Epoch 11, batch 1400, loss[loss=0.2484, simple_loss=0.3233, pruned_loss=0.08674, over 8317.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3109, pruned_loss=0.08037, over 1621159.33 frames. ], batch size: 25, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:56:51,055 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.627e+02 3.119e+02 3.954e+02 1.224e+03, threshold=6.238e+02, percent-clipped=1.0 +2023-02-06 09:57:13,607 INFO [train.py:901] (2/4) Epoch 11, batch 1450, loss[loss=0.2864, simple_loss=0.3548, pruned_loss=0.109, over 8662.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3098, pruned_loss=0.0798, over 1617190.80 frames. ], batch size: 34, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:57:17,231 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9647, 2.3150, 1.8769, 2.8147, 1.4687, 1.4467, 2.0633, 2.3541], + device='cuda:2'), covar=tensor([0.0764, 0.0879, 0.1004, 0.0450, 0.1170, 0.1568, 0.1017, 0.0874], + device='cuda:2'), in_proj_covar=tensor([0.0241, 0.0214, 0.0255, 0.0219, 0.0218, 0.0256, 0.0255, 0.0225], + 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') +2023-02-06 09:57:27,767 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 09:57:48,806 INFO [train.py:901] (2/4) Epoch 11, batch 1500, loss[loss=0.2418, simple_loss=0.3221, pruned_loss=0.0808, over 8248.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3125, pruned_loss=0.08149, over 1618748.87 frames. ], batch size: 24, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:58:01,384 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 2.743e+02 3.193e+02 4.270e+02 9.879e+02, threshold=6.387e+02, percent-clipped=7.0 +2023-02-06 09:58:02,226 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:12,521 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2677, 1.3515, 2.3339, 1.1331, 2.1320, 2.4892, 2.5599, 2.1209], + device='cuda:2'), covar=tensor([0.0969, 0.1140, 0.0457, 0.1966, 0.0636, 0.0372, 0.0646, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0260, 0.0293, 0.0254, 0.0286, 0.0272, 0.0232, 0.0330, 0.0289], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 09:58:13,198 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:24,634 INFO [train.py:901] (2/4) Epoch 11, batch 1550, loss[loss=0.2416, simple_loss=0.3296, pruned_loss=0.07681, over 8313.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3118, pruned_loss=0.08121, over 1616427.98 frames. ], batch size: 25, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 09:58:39,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82403.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:42,339 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.17 vs. limit=5.0 +2023-02-06 09:58:50,132 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:57,848 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:59,746 INFO [train.py:901] (2/4) Epoch 11, batch 1600, loss[loss=0.1824, simple_loss=0.2623, pruned_loss=0.05127, over 7442.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3102, pruned_loss=0.07981, over 1612378.75 frames. ], batch size: 17, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 09:59:06,491 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-06 09:59:11,745 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9612, 2.8552, 2.6792, 1.5686, 2.6429, 2.6140, 2.7172, 2.5293], + device='cuda:2'), covar=tensor([0.1251, 0.0964, 0.1192, 0.4248, 0.1089, 0.1312, 0.1363, 0.1091], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0375, 0.0384, 0.0482, 0.0377, 0.0369, 0.0369, 0.0325], + 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') +2023-02-06 09:59:12,992 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.328e+02 2.878e+02 3.468e+02 7.869e+02, threshold=5.757e+02, percent-clipped=2.0 +2023-02-06 09:59:18,835 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 09:59:24,125 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:59:36,397 INFO [train.py:901] (2/4) Epoch 11, batch 1650, loss[loss=0.2476, simple_loss=0.33, pruned_loss=0.08259, over 8505.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3108, pruned_loss=0.0801, over 1616454.32 frames. ], batch size: 26, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 09:59:57,314 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82511.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:00:11,602 INFO [train.py:901] (2/4) Epoch 11, batch 1700, loss[loss=0.2207, simple_loss=0.2884, pruned_loss=0.07648, over 7440.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3105, pruned_loss=0.0803, over 1612123.28 frames. ], batch size: 17, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 10:00:12,444 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:00:23,195 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.517e+02 3.185e+02 4.066e+02 8.085e+02, threshold=6.370e+02, percent-clipped=5.0 +2023-02-06 10:00:47,519 INFO [train.py:901] (2/4) Epoch 11, batch 1750, loss[loss=0.2448, simple_loss=0.3166, pruned_loss=0.0865, over 7807.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3113, pruned_loss=0.0811, over 1611233.72 frames. ], batch size: 20, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:00:52,551 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:01:23,307 INFO [train.py:901] (2/4) Epoch 11, batch 1800, loss[loss=0.2295, simple_loss=0.3109, pruned_loss=0.074, over 8373.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3115, pruned_loss=0.08126, over 1611231.51 frames. ], batch size: 24, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:01:26,965 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2989, 2.1821, 1.5643, 2.0438, 1.7708, 1.3311, 1.6649, 1.8277], + device='cuda:2'), covar=tensor([0.1196, 0.0348, 0.1071, 0.0503, 0.0606, 0.1350, 0.0851, 0.0706], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0234, 0.0314, 0.0300, 0.0300, 0.0324, 0.0341, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 10:01:35,825 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.598e+02 3.107e+02 4.193e+02 1.199e+03, threshold=6.213e+02, percent-clipped=8.0 +2023-02-06 10:01:42,099 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8261, 1.5644, 3.0949, 1.2945, 2.0413, 3.3807, 3.4014, 2.8921], + device='cuda:2'), covar=tensor([0.0996, 0.1497, 0.0356, 0.2049, 0.1028, 0.0263, 0.0436, 0.0619], + device='cuda:2'), in_proj_covar=tensor([0.0261, 0.0297, 0.0258, 0.0289, 0.0273, 0.0234, 0.0333, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 10:01:58,583 INFO [train.py:901] (2/4) Epoch 11, batch 1850, loss[loss=0.2385, simple_loss=0.3288, pruned_loss=0.07414, over 8519.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3125, pruned_loss=0.08149, over 1614410.46 frames. ], batch size: 26, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:02:10,530 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8557, 1.8718, 2.1334, 1.7418, 1.2729, 2.3014, 0.3694, 1.3334], + device='cuda:2'), covar=tensor([0.2318, 0.1943, 0.0547, 0.1922, 0.4178, 0.0557, 0.3442, 0.2107], + device='cuda:2'), in_proj_covar=tensor([0.0165, 0.0167, 0.0096, 0.0214, 0.0255, 0.0105, 0.0164, 0.0164], + 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') +2023-02-06 10:02:18,372 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:02:26,582 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82720.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:02:34,001 INFO [train.py:901] (2/4) Epoch 11, batch 1900, loss[loss=0.2115, simple_loss=0.2903, pruned_loss=0.06634, over 7981.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3121, pruned_loss=0.08169, over 1613454.14 frames. ], batch size: 21, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:02:43,684 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:02:45,519 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.442e+02 3.142e+02 3.936e+02 6.780e+02, threshold=6.284e+02, percent-clipped=1.0 +2023-02-06 10:03:08,879 INFO [train.py:901] (2/4) Epoch 11, batch 1950, loss[loss=0.2225, simple_loss=0.3026, pruned_loss=0.07118, over 7812.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3117, pruned_loss=0.08116, over 1611710.99 frames. ], batch size: 20, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:03:12,343 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 10:03:13,853 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:03:26,478 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 10:03:32,227 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:03:39,052 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82823.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:03:45,035 INFO [train.py:901] (2/4) Epoch 11, batch 2000, loss[loss=0.2204, simple_loss=0.2807, pruned_loss=0.08005, over 7798.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3122, pruned_loss=0.08154, over 1615272.63 frames. ], batch size: 19, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:03:47,028 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 10:03:56,671 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.675e+02 3.279e+02 3.987e+02 1.082e+03, threshold=6.559e+02, percent-clipped=7.0 +2023-02-06 10:03:57,937 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.98 vs. limit=5.0 +2023-02-06 10:04:01,560 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82855.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:04:16,778 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1453, 1.0530, 1.1947, 1.0766, 0.8991, 1.2354, 0.0393, 0.8860], + device='cuda:2'), covar=tensor([0.2226, 0.1932, 0.0649, 0.1310, 0.3967, 0.0765, 0.3312, 0.2059], + device='cuda:2'), in_proj_covar=tensor([0.0166, 0.0166, 0.0096, 0.0214, 0.0255, 0.0105, 0.0163, 0.0165], + 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') +2023-02-06 10:04:19,319 INFO [train.py:901] (2/4) Epoch 11, batch 2050, loss[loss=0.2242, simple_loss=0.308, pruned_loss=0.07022, over 7812.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3127, pruned_loss=0.08141, over 1619352.35 frames. ], batch size: 20, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:04:55,308 INFO [train.py:901] (2/4) Epoch 11, batch 2100, loss[loss=0.2271, simple_loss=0.3158, pruned_loss=0.0692, over 8466.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3127, pruned_loss=0.08093, over 1619487.32 frames. ], batch size: 25, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:04:55,375 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:05:07,191 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.489e+02 3.174e+02 3.706e+02 9.083e+02, threshold=6.348e+02, percent-clipped=2.0 +2023-02-06 10:05:09,477 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 10:05:22,110 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82970.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:05:29,198 INFO [train.py:901] (2/4) Epoch 11, batch 2150, loss[loss=0.2624, simple_loss=0.3317, pruned_loss=0.09653, over 8082.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.315, pruned_loss=0.08242, over 1617390.68 frames. ], batch size: 21, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:05:29,352 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82981.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:05:53,364 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8020, 2.4281, 4.6899, 1.4909, 3.3649, 2.3542, 1.8753, 2.9359], + device='cuda:2'), covar=tensor([0.1548, 0.1979, 0.0621, 0.3666, 0.1312, 0.2633, 0.1724, 0.2223], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0514, 0.0537, 0.0574, 0.0616, 0.0548, 0.0469, 0.0611], + 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') +2023-02-06 10:06:04,068 INFO [train.py:901] (2/4) Epoch 11, batch 2200, loss[loss=0.2815, simple_loss=0.3495, pruned_loss=0.1067, over 8491.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3132, pruned_loss=0.0811, over 1619190.67 frames. ], batch size: 28, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:06:15,771 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83046.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:06:16,953 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.616e+02 2.510e+02 3.092e+02 4.104e+02 1.639e+03, threshold=6.185e+02, percent-clipped=4.0 +2023-02-06 10:06:38,939 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83079.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:06:40,016 INFO [train.py:901] (2/4) Epoch 11, batch 2250, loss[loss=0.2735, simple_loss=0.342, pruned_loss=0.1025, over 8362.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3119, pruned_loss=0.0811, over 1613889.89 frames. ], batch size: 24, lr: 7.04e-03, grad_scale: 8.0 +2023-02-06 10:06:55,783 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:07:14,108 INFO [train.py:901] (2/4) Epoch 11, batch 2300, loss[loss=0.2132, simple_loss=0.2845, pruned_loss=0.07093, over 7534.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.313, pruned_loss=0.08228, over 1614129.34 frames. ], batch size: 18, lr: 7.04e-03, grad_scale: 8.0 +2023-02-06 10:07:25,647 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.480e+02 3.199e+02 4.275e+02 9.806e+02, threshold=6.398e+02, percent-clipped=6.0 +2023-02-06 10:07:48,926 INFO [train.py:901] (2/4) Epoch 11, batch 2350, loss[loss=0.2514, simple_loss=0.3233, pruned_loss=0.08974, over 8456.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3131, pruned_loss=0.08265, over 1611392.89 frames. ], batch size: 27, lr: 7.04e-03, grad_scale: 16.0 +2023-02-06 10:08:12,059 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83214.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:08:20,034 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83226.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:08:23,028 INFO [train.py:901] (2/4) Epoch 11, batch 2400, loss[loss=0.1952, simple_loss=0.2656, pruned_loss=0.06238, over 7260.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3126, pruned_loss=0.08222, over 1609499.80 frames. ], batch size: 16, lr: 7.04e-03, grad_scale: 16.0 +2023-02-06 10:08:35,098 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.638e+02 2.359e+02 2.853e+02 3.666e+02 7.740e+02, threshold=5.706e+02, percent-clipped=1.0 +2023-02-06 10:08:37,259 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83251.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:08:58,651 INFO [train.py:901] (2/4) Epoch 11, batch 2450, loss[loss=0.2781, simple_loss=0.3336, pruned_loss=0.1113, over 7928.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3117, pruned_loss=0.08128, over 1610529.66 frames. ], batch size: 20, lr: 7.04e-03, grad_scale: 16.0 +2023-02-06 10:09:13,728 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:13,795 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:29,006 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83325.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:30,518 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:33,054 INFO [train.py:901] (2/4) Epoch 11, batch 2500, loss[loss=0.2587, simple_loss=0.3268, pruned_loss=0.09529, over 8463.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3112, pruned_loss=0.08096, over 1612640.51 frames. ], batch size: 27, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:09:44,598 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.634e+02 3.143e+02 3.904e+02 7.323e+02, threshold=6.285e+02, percent-clipped=4.0 +2023-02-06 10:10:07,372 INFO [train.py:901] (2/4) Epoch 11, batch 2550, loss[loss=0.2497, simple_loss=0.3284, pruned_loss=0.08554, over 8530.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3118, pruned_loss=0.08155, over 1613607.43 frames. ], batch size: 28, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:10:43,102 INFO [train.py:901] (2/4) Epoch 11, batch 2600, loss[loss=0.2604, simple_loss=0.3276, pruned_loss=0.09663, over 7191.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3115, pruned_loss=0.08132, over 1609506.16 frames. ], batch size: 71, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:10:49,491 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:10:54,710 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.607e+02 3.192e+02 4.372e+02 8.439e+02, threshold=6.384e+02, percent-clipped=10.0 +2023-02-06 10:10:55,569 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5490, 1.5787, 4.7138, 1.9085, 4.2085, 3.8867, 4.3014, 4.1133], + device='cuda:2'), covar=tensor([0.0469, 0.4090, 0.0436, 0.3189, 0.0951, 0.0817, 0.0419, 0.0515], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0569, 0.0586, 0.0528, 0.0599, 0.0503, 0.0502, 0.0576], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:11:17,509 INFO [train.py:901] (2/4) Epoch 11, batch 2650, loss[loss=0.2301, simple_loss=0.301, pruned_loss=0.07963, over 8209.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3113, pruned_loss=0.0811, over 1610372.84 frames. ], batch size: 23, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:11:52,411 INFO [train.py:901] (2/4) Epoch 11, batch 2700, loss[loss=0.198, simple_loss=0.2833, pruned_loss=0.05637, over 8567.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.31, pruned_loss=0.08023, over 1607217.90 frames. ], batch size: 31, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:11:58,916 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.47 vs. limit=5.0 +2023-02-06 10:12:04,666 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.578e+02 3.131e+02 4.095e+02 6.916e+02, threshold=6.263e+02, percent-clipped=2.0 +2023-02-06 10:12:11,631 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83558.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:12:27,318 INFO [train.py:901] (2/4) Epoch 11, batch 2750, loss[loss=0.2539, simple_loss=0.325, pruned_loss=0.09143, over 6693.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3084, pruned_loss=0.07923, over 1603856.48 frames. ], batch size: 71, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:12:49,813 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1832, 1.1454, 3.3313, 0.9788, 2.9036, 2.8335, 3.0388, 2.9119], + device='cuda:2'), covar=tensor([0.0742, 0.3766, 0.0629, 0.3418, 0.1302, 0.0940, 0.0707, 0.0828], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0567, 0.0577, 0.0524, 0.0596, 0.0498, 0.0501, 0.0571], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:12:51,398 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 10:13:03,333 INFO [train.py:901] (2/4) Epoch 11, batch 2800, loss[loss=0.209, simple_loss=0.2858, pruned_loss=0.0661, over 8129.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3088, pruned_loss=0.07916, over 1606647.79 frames. ], batch size: 22, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:13:03,709 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 10:13:13,811 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:13:15,049 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.535e+02 3.136e+02 3.769e+02 1.201e+03, threshold=6.273e+02, percent-clipped=3.0 +2023-02-06 10:13:32,688 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83673.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:13:37,616 INFO [train.py:901] (2/4) Epoch 11, batch 2850, loss[loss=0.2981, simple_loss=0.3402, pruned_loss=0.128, over 7454.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3086, pruned_loss=0.07906, over 1607728.74 frames. ], batch size: 17, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:13:37,812 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5212, 2.7322, 1.7457, 2.1007, 2.2178, 1.5683, 2.1276, 2.1299], + device='cuda:2'), covar=tensor([0.1453, 0.0396, 0.1189, 0.0709, 0.0667, 0.1354, 0.0965, 0.1000], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0235, 0.0314, 0.0294, 0.0301, 0.0323, 0.0338, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 10:13:47,912 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83696.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:02,150 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9429, 1.5092, 1.7033, 1.3218, 1.0751, 1.4458, 1.7207, 1.6502], + device='cuda:2'), covar=tensor([0.0560, 0.1218, 0.1710, 0.1377, 0.0598, 0.1478, 0.0693, 0.0596], + device='cuda:2'), in_proj_covar=tensor([0.0102, 0.0153, 0.0194, 0.0159, 0.0105, 0.0163, 0.0117, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 10:14:05,631 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83721.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:12,966 INFO [train.py:901] (2/4) Epoch 11, batch 2900, loss[loss=0.2217, simple_loss=0.3063, pruned_loss=0.06856, over 8554.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3098, pruned_loss=0.07978, over 1609848.28 frames. ], batch size: 31, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:14:25,269 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.545e+02 3.159e+02 4.165e+02 9.643e+02, threshold=6.318e+02, percent-clipped=5.0 +2023-02-06 10:14:34,292 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:46,312 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:48,173 INFO [train.py:901] (2/4) Epoch 11, batch 2950, loss[loss=0.2457, simple_loss=0.3152, pruned_loss=0.08805, over 8190.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3104, pruned_loss=0.08027, over 1611243.22 frames. ], batch size: 23, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:14:53,616 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 10:15:19,845 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3155, 1.5300, 2.1259, 1.1689, 1.4670, 1.5771, 1.4021, 1.3945], + device='cuda:2'), covar=tensor([0.1807, 0.2106, 0.0850, 0.3743, 0.1616, 0.2874, 0.1851, 0.1992], + device='cuda:2'), in_proj_covar=tensor([0.0487, 0.0515, 0.0533, 0.0576, 0.0614, 0.0544, 0.0469, 0.0609], + 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') +2023-02-06 10:15:22,302 INFO [train.py:901] (2/4) Epoch 11, batch 3000, loss[loss=0.2977, simple_loss=0.3593, pruned_loss=0.118, over 7015.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3105, pruned_loss=0.08057, over 1609130.69 frames. ], batch size: 73, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:15:22,302 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 10:15:34,552 INFO [train.py:935] (2/4) Epoch 11, validation: loss=0.1889, simple_loss=0.2886, pruned_loss=0.04461, over 944034.00 frames. +2023-02-06 10:15:34,553 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 10:15:46,617 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.511e+02 2.977e+02 3.600e+02 5.313e+02, threshold=5.953e+02, percent-clipped=0.0 +2023-02-06 10:16:10,356 INFO [train.py:901] (2/4) Epoch 11, batch 3050, loss[loss=0.2384, simple_loss=0.3244, pruned_loss=0.07622, over 8561.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3119, pruned_loss=0.08117, over 1611738.82 frames. ], batch size: 31, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:16:43,148 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:16:44,282 INFO [train.py:901] (2/4) Epoch 11, batch 3100, loss[loss=0.2595, simple_loss=0.338, pruned_loss=0.09051, over 8618.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3116, pruned_loss=0.08119, over 1616825.53 frames. ], batch size: 39, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:16:55,416 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.748e+02 3.262e+02 3.755e+02 7.942e+02, threshold=6.525e+02, percent-clipped=1.0 +2023-02-06 10:17:00,080 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83954.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:08,759 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:18,437 INFO [train.py:901] (2/4) Epoch 11, batch 3150, loss[loss=0.2112, simple_loss=0.2966, pruned_loss=0.06288, over 8103.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3111, pruned_loss=0.08099, over 1614665.70 frames. ], batch size: 23, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:17:34,197 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84003.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:42,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7751, 2.3521, 3.3739, 2.7950, 3.1037, 2.4054, 2.0110, 1.9682], + device='cuda:2'), covar=tensor([0.3102, 0.3514, 0.1235, 0.2143, 0.1775, 0.2168, 0.1538, 0.3827], + device='cuda:2'), in_proj_covar=tensor([0.0884, 0.0865, 0.0727, 0.0837, 0.0934, 0.0793, 0.0700, 0.0763], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:17:44,276 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84017.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:53,259 INFO [train.py:901] (2/4) Epoch 11, batch 3200, loss[loss=0.1925, simple_loss=0.2713, pruned_loss=0.05681, over 7438.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3112, pruned_loss=0.08124, over 1612410.23 frames. ], batch size: 17, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:18:01,384 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:18:05,764 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.726e+02 3.369e+02 4.220e+02 9.302e+02, threshold=6.739e+02, percent-clipped=4.0 +2023-02-06 10:18:27,182 INFO [train.py:901] (2/4) Epoch 11, batch 3250, loss[loss=0.2206, simple_loss=0.2915, pruned_loss=0.07483, over 7971.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3113, pruned_loss=0.08118, over 1614632.38 frames. ], batch size: 21, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:18:50,446 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84115.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:18:55,081 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:19:01,831 INFO [train.py:901] (2/4) Epoch 11, batch 3300, loss[loss=0.2251, simple_loss=0.2983, pruned_loss=0.07597, over 8067.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3119, pruned_loss=0.08137, over 1613118.58 frames. ], batch size: 21, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:19:08,603 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8683, 1.9476, 2.4012, 1.7024, 1.3614, 2.5826, 0.5304, 1.6169], + device='cuda:2'), covar=tensor([0.2969, 0.1699, 0.0532, 0.2268, 0.3932, 0.0395, 0.3333, 0.1754], + device='cuda:2'), in_proj_covar=tensor([0.0164, 0.0167, 0.0097, 0.0215, 0.0253, 0.0104, 0.0166, 0.0165], + 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') +2023-02-06 10:19:13,378 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.729e+02 3.101e+02 4.103e+02 8.191e+02, threshold=6.202e+02, percent-clipped=3.0 +2023-02-06 10:19:35,409 INFO [train.py:901] (2/4) Epoch 11, batch 3350, loss[loss=0.2576, simple_loss=0.3528, pruned_loss=0.08114, over 8576.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3133, pruned_loss=0.08207, over 1616299.18 frames. ], batch size: 31, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:19:37,881 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 10:20:01,020 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84217.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:20:10,208 INFO [train.py:901] (2/4) Epoch 11, batch 3400, loss[loss=0.2471, simple_loss=0.3211, pruned_loss=0.08659, over 8529.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3126, pruned_loss=0.08151, over 1613566.09 frames. ], batch size: 49, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:20:15,226 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84237.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:20:23,162 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.356e+02 2.553e+02 3.068e+02 3.977e+02 7.727e+02, threshold=6.137e+02, percent-clipped=2.0 +2023-02-06 10:20:26,644 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:20:45,372 INFO [train.py:901] (2/4) Epoch 11, batch 3450, loss[loss=0.2185, simple_loss=0.3038, pruned_loss=0.06657, over 8558.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3137, pruned_loss=0.08238, over 1615402.10 frames. ], batch size: 31, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:21:06,407 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84311.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:21:20,276 INFO [train.py:901] (2/4) Epoch 11, batch 3500, loss[loss=0.2066, simple_loss=0.2743, pruned_loss=0.06948, over 8245.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3128, pruned_loss=0.08196, over 1614056.35 frames. ], batch size: 22, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:21:31,075 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:21:32,273 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.703e+02 3.166e+02 4.187e+02 8.001e+02, threshold=6.332e+02, percent-clipped=6.0 +2023-02-06 10:21:36,475 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84354.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:21:48,769 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 10:21:54,138 INFO [train.py:901] (2/4) Epoch 11, batch 3550, loss[loss=0.2503, simple_loss=0.3196, pruned_loss=0.09056, over 7810.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3126, pruned_loss=0.08188, over 1604485.19 frames. ], batch size: 20, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:22:04,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3151, 2.6139, 1.7186, 2.1970, 1.9788, 1.3037, 1.7399, 2.1207], + device='cuda:2'), covar=tensor([0.1610, 0.0386, 0.1211, 0.0609, 0.0792, 0.1775, 0.1234, 0.0841], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0232, 0.0314, 0.0291, 0.0301, 0.0319, 0.0338, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 10:22:25,881 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84426.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:22:28,969 INFO [train.py:901] (2/4) Epoch 11, batch 3600, loss[loss=0.3038, simple_loss=0.3587, pruned_loss=0.1244, over 7089.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3135, pruned_loss=0.08274, over 1603401.75 frames. ], batch size: 71, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:22:41,776 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.788e+02 3.447e+02 4.179e+02 1.001e+03, threshold=6.895e+02, percent-clipped=4.0 +2023-02-06 10:22:48,586 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:22:50,733 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:02,588 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 +2023-02-06 10:23:03,534 INFO [train.py:901] (2/4) Epoch 11, batch 3650, loss[loss=0.2185, simple_loss=0.2861, pruned_loss=0.07546, over 7645.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3121, pruned_loss=0.08167, over 1606116.86 frames. ], batch size: 19, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:23:07,613 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2690, 1.6046, 1.7113, 1.3521, 1.2621, 1.5002, 1.9889, 1.8947], + device='cuda:2'), covar=tensor([0.0433, 0.1152, 0.1652, 0.1381, 0.0524, 0.1407, 0.0586, 0.0521], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0154, 0.0191, 0.0158, 0.0105, 0.0163, 0.0117, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 10:23:11,601 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84493.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:20,289 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7228, 1.7262, 2.1404, 1.6441, 1.1787, 2.1846, 0.2747, 1.2317], + device='cuda:2'), covar=tensor([0.2770, 0.1702, 0.0513, 0.1708, 0.4443, 0.0547, 0.3616, 0.2157], + device='cuda:2'), in_proj_covar=tensor([0.0163, 0.0166, 0.0097, 0.0212, 0.0253, 0.0104, 0.0165, 0.0165], + 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') +2023-02-06 10:23:28,901 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84518.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:37,222 INFO [train.py:901] (2/4) Epoch 11, batch 3700, loss[loss=0.2143, simple_loss=0.289, pruned_loss=0.06979, over 7950.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3119, pruned_loss=0.08162, over 1608912.03 frames. ], batch size: 20, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:23:46,139 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:48,588 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 10:23:49,869 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.648e+02 3.219e+02 3.938e+02 7.332e+02, threshold=6.437e+02, percent-clipped=1.0 +2023-02-06 10:23:58,483 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84561.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:24:07,286 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84574.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:24:08,932 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 10:24:11,839 INFO [train.py:901] (2/4) Epoch 11, batch 3750, loss[loss=0.2313, simple_loss=0.3162, pruned_loss=0.07319, over 8373.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3106, pruned_loss=0.08032, over 1610796.16 frames. ], batch size: 24, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:24:23,994 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:24:46,870 INFO [train.py:901] (2/4) Epoch 11, batch 3800, loss[loss=0.2778, simple_loss=0.3447, pruned_loss=0.1055, over 8457.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3117, pruned_loss=0.08098, over 1613570.42 frames. ], batch size: 25, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:24:57,703 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7657, 1.7609, 2.1635, 1.7019, 1.1602, 2.1905, 0.4122, 1.4023], + device='cuda:2'), covar=tensor([0.2155, 0.1479, 0.0491, 0.1864, 0.4254, 0.0547, 0.3046, 0.1794], + device='cuda:2'), in_proj_covar=tensor([0.0160, 0.0164, 0.0095, 0.0210, 0.0251, 0.0103, 0.0164, 0.0163], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 10:24:58,763 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.607e+02 3.118e+02 4.251e+02 1.041e+03, threshold=6.237e+02, percent-clipped=4.0 +2023-02-06 10:25:00,934 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:18,149 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84676.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:18,379 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 10:25:21,361 INFO [train.py:901] (2/4) Epoch 11, batch 3850, loss[loss=0.2447, simple_loss=0.3214, pruned_loss=0.08397, over 7979.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3115, pruned_loss=0.08088, over 1611893.10 frames. ], batch size: 21, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:25:22,273 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84682.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:32,823 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:39,154 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3430, 1.9650, 2.9819, 2.3676, 2.7007, 2.1810, 1.7444, 1.3169], + device='cuda:2'), covar=tensor([0.3662, 0.3770, 0.1093, 0.2472, 0.1858, 0.2007, 0.1559, 0.3988], + device='cuda:2'), in_proj_covar=tensor([0.0881, 0.0862, 0.0724, 0.0834, 0.0930, 0.0794, 0.0701, 0.0762], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:25:39,748 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84707.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:43,786 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:47,049 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:51,558 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 10:25:53,839 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 10:25:55,492 INFO [train.py:901] (2/4) Epoch 11, batch 3900, loss[loss=0.2188, simple_loss=0.312, pruned_loss=0.06283, over 8185.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3111, pruned_loss=0.08083, over 1616363.04 frames. ], batch size: 23, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:26:03,707 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84743.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:26:08,291 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.619e+02 3.238e+02 3.926e+02 9.069e+02, threshold=6.476e+02, percent-clipped=5.0 +2023-02-06 10:26:30,325 INFO [train.py:901] (2/4) Epoch 11, batch 3950, loss[loss=0.2355, simple_loss=0.3131, pruned_loss=0.07892, over 7817.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.311, pruned_loss=0.08086, over 1616126.64 frames. ], batch size: 20, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:26:34,624 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2918, 1.4702, 1.6399, 1.4002, 1.0304, 1.4583, 1.8426, 1.6337], + device='cuda:2'), covar=tensor([0.0464, 0.1252, 0.1678, 0.1418, 0.0618, 0.1488, 0.0694, 0.0618], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0154, 0.0192, 0.0159, 0.0105, 0.0164, 0.0117, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 10:26:52,779 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:26:56,211 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84818.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:27:04,812 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:27:05,273 INFO [train.py:901] (2/4) Epoch 11, batch 4000, loss[loss=0.2688, simple_loss=0.3499, pruned_loss=0.09383, over 8461.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3108, pruned_loss=0.08071, over 1612080.83 frames. ], batch size: 27, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:27:17,172 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.608e+02 2.990e+02 3.694e+02 8.393e+02, threshold=5.981e+02, percent-clipped=2.0 +2023-02-06 10:27:21,548 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:27:39,658 INFO [train.py:901] (2/4) Epoch 11, batch 4050, loss[loss=0.1938, simple_loss=0.2903, pruned_loss=0.04866, over 8025.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3116, pruned_loss=0.08062, over 1614451.41 frames. ], batch size: 22, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:27:44,485 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84887.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:15,559 INFO [train.py:901] (2/4) Epoch 11, batch 4100, loss[loss=0.2114, simple_loss=0.29, pruned_loss=0.06639, over 7817.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3117, pruned_loss=0.08099, over 1610239.98 frames. ], batch size: 20, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:28:16,482 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84932.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:27,742 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.518e+02 2.978e+02 3.788e+02 7.594e+02, threshold=5.956e+02, percent-clipped=4.0 +2023-02-06 10:28:33,376 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:34,752 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6181, 1.5052, 4.8416, 1.8472, 4.2693, 4.0135, 4.3727, 4.2662], + device='cuda:2'), covar=tensor([0.0491, 0.4123, 0.0346, 0.3234, 0.0939, 0.0755, 0.0441, 0.0527], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0573, 0.0581, 0.0525, 0.0602, 0.0513, 0.0505, 0.0575], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:28:41,369 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:41,447 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:49,616 INFO [train.py:901] (2/4) Epoch 11, batch 4150, loss[loss=0.2748, simple_loss=0.3383, pruned_loss=0.1057, over 7964.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3119, pruned_loss=0.08058, over 1611245.07 frames. ], batch size: 21, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:28:58,534 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84994.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:59,078 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84995.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:29:04,353 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85002.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:29:23,824 INFO [train.py:901] (2/4) Epoch 11, batch 4200, loss[loss=0.2156, simple_loss=0.2976, pruned_loss=0.06682, over 7798.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3114, pruned_loss=0.08048, over 1612989.08 frames. ], batch size: 19, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:29:36,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.581e+02 3.261e+02 3.967e+02 9.417e+02, threshold=6.523e+02, percent-clipped=7.0 +2023-02-06 10:29:47,924 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 10:29:50,138 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85069.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:29:58,035 INFO [train.py:901] (2/4) Epoch 11, batch 4250, loss[loss=0.2363, simple_loss=0.3139, pruned_loss=0.07933, over 8683.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3107, pruned_loss=0.08022, over 1610366.94 frames. ], batch size: 34, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:30:06,850 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:10,065 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 10:30:18,077 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85110.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:25,991 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1174, 1.4205, 4.3121, 1.5613, 3.7614, 3.6024, 3.8615, 3.6992], + device='cuda:2'), covar=tensor([0.0469, 0.3798, 0.0413, 0.3254, 0.1020, 0.0788, 0.0512, 0.0607], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0570, 0.0574, 0.0522, 0.0601, 0.0513, 0.0502, 0.0572], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:30:28,615 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:32,505 INFO [train.py:901] (2/4) Epoch 11, batch 4300, loss[loss=0.2423, simple_loss=0.3233, pruned_loss=0.08067, over 8252.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3112, pruned_loss=0.08087, over 1608528.59 frames. ], batch size: 24, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:30:33,970 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:45,182 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.616e+02 3.014e+02 4.154e+02 7.931e+02, threshold=6.027e+02, percent-clipped=5.0 +2023-02-06 10:30:54,027 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:31:06,968 INFO [train.py:901] (2/4) Epoch 11, batch 4350, loss[loss=0.2396, simple_loss=0.2999, pruned_loss=0.08968, over 7218.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3117, pruned_loss=0.08074, over 1611937.39 frames. ], batch size: 16, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:31:40,295 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 10:31:41,574 INFO [train.py:901] (2/4) Epoch 11, batch 4400, loss[loss=0.2777, simple_loss=0.3384, pruned_loss=0.1086, over 7806.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3116, pruned_loss=0.08102, over 1613839.59 frames. ], batch size: 20, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:31:54,340 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.553e+02 3.172e+02 3.669e+02 6.483e+02, threshold=6.345e+02, percent-clipped=4.0 +2023-02-06 10:32:01,478 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85258.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:05,695 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 10:32:14,081 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85277.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:16,486 INFO [train.py:901] (2/4) Epoch 11, batch 4450, loss[loss=0.2279, simple_loss=0.3023, pruned_loss=0.07672, over 8092.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3107, pruned_loss=0.08075, over 1608108.23 frames. ], batch size: 21, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:32:18,773 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:22,678 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 10:32:38,776 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85313.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:41,052 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 10:32:50,691 INFO [train.py:901] (2/4) Epoch 11, batch 4500, loss[loss=0.208, simple_loss=0.2866, pruned_loss=0.06467, over 7648.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3107, pruned_loss=0.08051, over 1611865.74 frames. ], batch size: 19, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:33:03,396 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.629e+02 3.227e+02 4.085e+02 1.162e+03, threshold=6.455e+02, percent-clipped=2.0 +2023-02-06 10:33:15,850 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 10:33:16,068 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:33:26,538 INFO [train.py:901] (2/4) Epoch 11, batch 4550, loss[loss=0.2265, simple_loss=0.3056, pruned_loss=0.07369, over 8656.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3129, pruned_loss=0.08217, over 1613726.30 frames. ], batch size: 34, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:33:33,530 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85391.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:33:59,392 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:01,128 INFO [train.py:901] (2/4) Epoch 11, batch 4600, loss[loss=0.2677, simple_loss=0.3496, pruned_loss=0.09292, over 8359.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3136, pruned_loss=0.08236, over 1615802.87 frames. ], batch size: 24, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:34:11,931 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85446.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:13,795 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.573e+02 3.214e+02 4.149e+02 1.527e+03, threshold=6.427e+02, percent-clipped=2.0 +2023-02-06 10:34:14,058 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4655, 1.8110, 2.8487, 1.2168, 1.8988, 1.7983, 1.4800, 1.8578], + device='cuda:2'), covar=tensor([0.1738, 0.2043, 0.0674, 0.3919, 0.1625, 0.2835, 0.1899, 0.2111], + device='cuda:2'), in_proj_covar=tensor([0.0489, 0.0517, 0.0533, 0.0578, 0.0617, 0.0554, 0.0473, 0.0612], + 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') +2023-02-06 10:34:28,038 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:33,579 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:36,251 INFO [train.py:901] (2/4) Epoch 11, batch 4650, loss[loss=0.2692, simple_loss=0.3563, pruned_loss=0.09111, over 8415.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.314, pruned_loss=0.08286, over 1613540.75 frames. ], batch size: 49, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:34:42,455 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 10:34:50,393 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85501.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:54,609 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 10:35:11,112 INFO [train.py:901] (2/4) Epoch 11, batch 4700, loss[loss=0.2727, simple_loss=0.3455, pruned_loss=0.0999, over 8569.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3136, pruned_loss=0.0823, over 1612867.32 frames. ], batch size: 31, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:35:12,715 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:13,285 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0231, 6.0379, 5.2555, 2.6168, 5.3648, 5.6666, 5.4844, 5.3325], + device='cuda:2'), covar=tensor([0.0443, 0.0351, 0.0801, 0.4255, 0.0776, 0.0553, 0.0967, 0.0584], + device='cuda:2'), in_proj_covar=tensor([0.0472, 0.0377, 0.0379, 0.0487, 0.0377, 0.0380, 0.0376, 0.0334], + 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') +2023-02-06 10:35:22,559 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:23,054 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.812e+02 3.491e+02 4.674e+02 1.006e+03, threshold=6.983e+02, percent-clipped=9.0 +2023-02-06 10:35:30,014 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85558.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:41,262 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1653, 1.4139, 2.3884, 1.2014, 2.0864, 2.5224, 2.6265, 2.1508], + device='cuda:2'), covar=tensor([0.1013, 0.1152, 0.0433, 0.1860, 0.0632, 0.0350, 0.0606, 0.0719], + device='cuda:2'), in_proj_covar=tensor([0.0258, 0.0292, 0.0254, 0.0285, 0.0266, 0.0232, 0.0333, 0.0286], + 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') +2023-02-06 10:35:45,764 INFO [train.py:901] (2/4) Epoch 11, batch 4750, loss[loss=0.197, simple_loss=0.2782, pruned_loss=0.05788, over 7807.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3143, pruned_loss=0.08222, over 1612221.45 frames. ], batch size: 19, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:35:47,965 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85584.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:53,286 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:36:08,950 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 +2023-02-06 10:36:09,863 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 10:36:11,813 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 10:36:20,649 INFO [train.py:901] (2/4) Epoch 11, batch 4800, loss[loss=0.227, simple_loss=0.3092, pruned_loss=0.07238, over 8572.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3144, pruned_loss=0.0827, over 1612409.26 frames. ], batch size: 31, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:36:28,392 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 10:36:28,789 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85643.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:36:32,775 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.628e+02 3.255e+02 4.281e+02 8.051e+02, threshold=6.510e+02, percent-clipped=3.0 +2023-02-06 10:36:35,661 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5308, 4.4948, 4.0105, 1.8960, 3.9831, 4.1197, 4.0908, 3.8220], + device='cuda:2'), covar=tensor([0.0855, 0.0684, 0.1213, 0.5561, 0.1016, 0.1172, 0.1440, 0.1030], + device='cuda:2'), in_proj_covar=tensor([0.0473, 0.0377, 0.0379, 0.0485, 0.0376, 0.0380, 0.0375, 0.0334], + 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') +2023-02-06 10:36:55,313 INFO [train.py:901] (2/4) Epoch 11, batch 4850, loss[loss=0.2278, simple_loss=0.3131, pruned_loss=0.07125, over 8554.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3138, pruned_loss=0.0821, over 1613448.30 frames. ], batch size: 31, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:36:57,570 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85684.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:37:01,316 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 10:37:14,640 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:37:30,188 INFO [train.py:901] (2/4) Epoch 11, batch 4900, loss[loss=0.2, simple_loss=0.2768, pruned_loss=0.06157, over 7784.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.313, pruned_loss=0.08163, over 1612675.46 frames. ], batch size: 19, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:37:42,882 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.544e+02 3.151e+02 4.004e+02 8.063e+02, threshold=6.301e+02, percent-clipped=5.0 +2023-02-06 10:38:04,648 INFO [train.py:901] (2/4) Epoch 11, batch 4950, loss[loss=0.2502, simple_loss=0.3281, pruned_loss=0.08618, over 8365.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3115, pruned_loss=0.08093, over 1613919.44 frames. ], batch size: 24, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:38:10,866 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2285, 1.3525, 1.3034, 1.8986, 0.6964, 1.1353, 1.3783, 1.4692], + device='cuda:2'), covar=tensor([0.1244, 0.0937, 0.1360, 0.0674, 0.1234, 0.1608, 0.0856, 0.0805], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0216, 0.0256, 0.0217, 0.0218, 0.0253, 0.0257, 0.0222], + 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') +2023-02-06 10:38:11,407 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85790.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:26,563 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5589, 2.2465, 3.3842, 2.6522, 2.9177, 2.4399, 1.8489, 1.6487], + device='cuda:2'), covar=tensor([0.3846, 0.4031, 0.1290, 0.2540, 0.2167, 0.2083, 0.1656, 0.4288], + device='cuda:2'), in_proj_covar=tensor([0.0886, 0.0868, 0.0735, 0.0834, 0.0940, 0.0799, 0.0704, 0.0773], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:38:39,606 INFO [train.py:901] (2/4) Epoch 11, batch 5000, loss[loss=0.2866, simple_loss=0.3494, pruned_loss=0.1119, over 8492.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3118, pruned_loss=0.08081, over 1617293.46 frames. ], batch size: 29, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:38:46,525 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85840.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:49,800 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85845.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:51,924 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:52,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.585e+02 3.219e+02 4.097e+02 8.363e+02, threshold=6.438e+02, percent-clipped=6.0 +2023-02-06 10:39:03,551 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:08,908 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85873.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:13,944 INFO [train.py:901] (2/4) Epoch 11, batch 5050, loss[loss=0.2408, simple_loss=0.3116, pruned_loss=0.08502, over 8623.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3123, pruned_loss=0.08108, over 1615743.76 frames. ], batch size: 34, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:39:21,268 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85892.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:22,013 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85893.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:30,135 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:39,824 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 10:39:48,590 INFO [train.py:901] (2/4) Epoch 11, batch 5100, loss[loss=0.2129, simple_loss=0.2959, pruned_loss=0.06491, over 7655.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3119, pruned_loss=0.0803, over 1624087.62 frames. ], batch size: 19, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:39:53,825 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-06 10:40:00,818 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85948.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:40:01,252 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.570e+02 3.113e+02 3.980e+02 6.838e+02, threshold=6.226e+02, percent-clipped=2.0 +2023-02-06 10:40:02,904 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8586, 1.8875, 2.2268, 1.6429, 1.2563, 2.4349, 0.2929, 1.3127], + device='cuda:2'), covar=tensor([0.2258, 0.1704, 0.0739, 0.1880, 0.4099, 0.0451, 0.3426, 0.1914], + device='cuda:2'), in_proj_covar=tensor([0.0163, 0.0167, 0.0100, 0.0213, 0.0255, 0.0103, 0.0166, 0.0165], + 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') +2023-02-06 10:40:06,266 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0536, 1.5493, 1.5343, 1.3720, 1.0308, 1.3850, 1.8156, 1.7571], + device='cuda:2'), covar=tensor([0.0516, 0.1205, 0.1828, 0.1408, 0.0620, 0.1550, 0.0694, 0.0577], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0152, 0.0192, 0.0158, 0.0104, 0.0164, 0.0117, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 10:40:08,934 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85960.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:19,551 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85975.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:23,496 INFO [train.py:901] (2/4) Epoch 11, batch 5150, loss[loss=0.2237, simple_loss=0.2866, pruned_loss=0.08043, over 7636.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3129, pruned_loss=0.08077, over 1627022.31 frames. ], batch size: 19, lr: 6.92e-03, grad_scale: 8.0 +2023-02-06 10:40:27,617 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:38,308 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.38 vs. limit=5.0 +2023-02-06 10:40:42,270 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:51,395 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:59,434 INFO [train.py:901] (2/4) Epoch 11, batch 5200, loss[loss=0.2823, simple_loss=0.3362, pruned_loss=0.1142, over 8191.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3114, pruned_loss=0.07981, over 1621739.90 frames. ], batch size: 23, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:41:12,340 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.648e+02 3.082e+02 3.913e+02 1.007e+03, threshold=6.165e+02, percent-clipped=5.0 +2023-02-06 10:41:27,748 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:41:35,328 INFO [train.py:901] (2/4) Epoch 11, batch 5250, loss[loss=0.2299, simple_loss=0.3081, pruned_loss=0.07584, over 8660.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3119, pruned_loss=0.08008, over 1624309.05 frames. ], batch size: 34, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:41:39,724 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:41:40,967 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 10:41:50,759 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86102.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:42:10,643 INFO [train.py:901] (2/4) Epoch 11, batch 5300, loss[loss=0.2639, simple_loss=0.3355, pruned_loss=0.0961, over 8475.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3108, pruned_loss=0.07925, over 1620387.71 frames. ], batch size: 25, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:42:23,758 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.570e+02 3.118e+02 4.195e+02 8.045e+02, threshold=6.237e+02, percent-clipped=4.0 +2023-02-06 10:42:32,850 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:42:46,494 INFO [train.py:901] (2/4) Epoch 11, batch 5350, loss[loss=0.2147, simple_loss=0.2992, pruned_loss=0.06505, over 8634.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3104, pruned_loss=0.0795, over 1618211.03 frames. ], batch size: 34, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:42:50,726 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:12,428 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86216.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:14,734 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.73 vs. limit=5.0 +2023-02-06 10:43:22,276 INFO [train.py:901] (2/4) Epoch 11, batch 5400, loss[loss=0.2061, simple_loss=0.2919, pruned_loss=0.06021, over 8357.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3102, pruned_loss=0.07978, over 1616486.90 frames. ], batch size: 24, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:43:26,574 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86237.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:29,463 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86241.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:34,673 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.471e+02 3.223e+02 4.268e+02 9.619e+02, threshold=6.446e+02, percent-clipped=7.0 +2023-02-06 10:43:44,507 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:45,130 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86264.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:57,388 INFO [train.py:901] (2/4) Epoch 11, batch 5450, loss[loss=0.2312, simple_loss=0.2927, pruned_loss=0.08485, over 5152.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3103, pruned_loss=0.08014, over 1611525.11 frames. ], batch size: 11, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:44:03,061 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:05,816 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86292.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:44:25,138 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:34,080 INFO [train.py:901] (2/4) Epoch 11, batch 5500, loss[loss=0.2563, simple_loss=0.3228, pruned_loss=0.09491, over 7970.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3103, pruned_loss=0.08042, over 1611551.09 frames. ], batch size: 21, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:44:34,719 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 10:44:46,143 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.494e+02 3.013e+02 3.770e+02 8.759e+02, threshold=6.025e+02, percent-clipped=3.0 +2023-02-06 10:44:48,466 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86352.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:52,694 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:56,787 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:09,177 INFO [train.py:901] (2/4) Epoch 11, batch 5550, loss[loss=0.2295, simple_loss=0.2998, pruned_loss=0.07955, over 8293.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3107, pruned_loss=0.08074, over 1609003.29 frames. ], batch size: 23, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:45:10,785 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:27,852 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86407.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:45:33,154 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:44,334 INFO [train.py:901] (2/4) Epoch 11, batch 5600, loss[loss=0.2448, simple_loss=0.32, pruned_loss=0.0848, over 8192.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3103, pruned_loss=0.08017, over 1611592.14 frames. ], batch size: 23, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:45:44,407 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86431.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:46,522 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:57,250 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.527e+02 3.003e+02 3.802e+02 9.548e+02, threshold=6.005e+02, percent-clipped=4.0 +2023-02-06 10:46:03,255 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:06,609 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86463.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:17,347 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:18,542 INFO [train.py:901] (2/4) Epoch 11, batch 5650, loss[loss=0.2083, simple_loss=0.2912, pruned_loss=0.06269, over 8298.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3103, pruned_loss=0.08019, over 1608724.68 frames. ], batch size: 23, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:46:39,900 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 10:46:44,119 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6898, 1.6511, 2.0598, 1.5079, 1.1790, 2.1051, 0.2232, 1.2343], + device='cuda:2'), covar=tensor([0.2375, 0.1826, 0.0452, 0.1438, 0.3909, 0.0431, 0.3387, 0.1847], + device='cuda:2'), in_proj_covar=tensor([0.0159, 0.0163, 0.0096, 0.0204, 0.0245, 0.0100, 0.0158, 0.0160], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-06 10:46:52,194 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86529.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:53,988 INFO [train.py:901] (2/4) Epoch 11, batch 5700, loss[loss=0.1858, simple_loss=0.2648, pruned_loss=0.05339, over 7680.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3095, pruned_loss=0.07959, over 1605339.63 frames. ], batch size: 18, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:47:04,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:47:06,038 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.473e+02 3.032e+02 3.837e+02 8.433e+02, threshold=6.065e+02, percent-clipped=5.0 +2023-02-06 10:47:26,925 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 10:47:28,594 INFO [train.py:901] (2/4) Epoch 11, batch 5750, loss[loss=0.2491, simple_loss=0.3208, pruned_loss=0.08868, over 8028.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3089, pruned_loss=0.07934, over 1604846.75 frames. ], batch size: 22, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:47:35,590 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5163, 2.6618, 1.7859, 2.1595, 2.1361, 1.5329, 2.0762, 2.1908], + device='cuda:2'), covar=tensor([0.1322, 0.0355, 0.1092, 0.0638, 0.0670, 0.1390, 0.0954, 0.0828], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0234, 0.0315, 0.0297, 0.0304, 0.0323, 0.0341, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 10:47:40,227 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:47:42,137 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 10:47:47,610 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:47:47,750 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:48:03,809 INFO [train.py:901] (2/4) Epoch 11, batch 5800, loss[loss=0.2404, simple_loss=0.3191, pruned_loss=0.08078, over 8502.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3091, pruned_loss=0.07912, over 1606864.35 frames. ], batch size: 49, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:48:05,400 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86633.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:48:17,063 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.625e+02 3.434e+02 4.363e+02 1.044e+03, threshold=6.867e+02, percent-clipped=16.0 +2023-02-06 10:48:26,842 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86663.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:48:39,359 INFO [train.py:901] (2/4) Epoch 11, batch 5850, loss[loss=0.216, simple_loss=0.2934, pruned_loss=0.06932, over 7217.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.31, pruned_loss=0.07885, over 1610311.91 frames. ], batch size: 16, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:48:44,300 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86688.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:48:45,650 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86690.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:02,124 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:07,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86723.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:13,225 INFO [train.py:901] (2/4) Epoch 11, batch 5900, loss[loss=0.2291, simple_loss=0.31, pruned_loss=0.07409, over 8315.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3106, pruned_loss=0.07991, over 1606775.43 frames. ], batch size: 25, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:49:16,680 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86735.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:25,724 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.878e+02 2.650e+02 3.002e+02 3.837e+02 8.505e+02, threshold=6.004e+02, percent-clipped=1.0 +2023-02-06 10:49:33,352 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:48,258 INFO [train.py:901] (2/4) Epoch 11, batch 5950, loss[loss=0.2115, simple_loss=0.2831, pruned_loss=0.06996, over 7651.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3103, pruned_loss=0.07999, over 1606148.35 frames. ], batch size: 19, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:49:51,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:03,638 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:03,791 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:06,871 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:09,062 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:20,485 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:22,934 INFO [train.py:901] (2/4) Epoch 11, batch 6000, loss[loss=0.2462, simple_loss=0.3089, pruned_loss=0.09171, over 8253.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3086, pruned_loss=0.07894, over 1605126.83 frames. ], batch size: 22, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:50:22,934 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 10:50:35,333 INFO [train.py:935] (2/4) Epoch 11, validation: loss=0.1887, simple_loss=0.2887, pruned_loss=0.04439, over 944034.00 frames. +2023-02-06 10:50:35,334 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 10:50:36,203 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86832.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:47,363 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.431e+02 2.934e+02 3.566e+02 7.044e+02, threshold=5.869e+02, percent-clipped=5.0 +2023-02-06 10:51:00,724 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 10:51:10,322 INFO [train.py:901] (2/4) Epoch 11, batch 6050, loss[loss=0.2086, simple_loss=0.2861, pruned_loss=0.0656, over 7814.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3108, pruned_loss=0.0805, over 1607098.56 frames. ], batch size: 20, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:51:20,528 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86896.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:35,573 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:39,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:45,226 INFO [train.py:901] (2/4) Epoch 11, batch 6100, loss[loss=0.2682, simple_loss=0.3406, pruned_loss=0.0979, over 8496.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3104, pruned_loss=0.07972, over 1614094.24 frames. ], batch size: 29, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:51:53,495 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86942.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:58,246 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.604e+02 3.114e+02 3.901e+02 9.212e+02, threshold=6.229e+02, percent-clipped=4.0 +2023-02-06 10:52:06,857 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 10:52:19,158 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:52:20,334 INFO [train.py:901] (2/4) Epoch 11, batch 6150, loss[loss=0.2666, simple_loss=0.3293, pruned_loss=0.102, over 8573.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.31, pruned_loss=0.0788, over 1620995.98 frames. ], batch size: 31, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:52:36,926 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87004.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:52:47,749 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87020.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:52:55,887 INFO [train.py:901] (2/4) Epoch 11, batch 6200, loss[loss=0.2308, simple_loss=0.3139, pruned_loss=0.0738, over 8359.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.309, pruned_loss=0.07818, over 1616156.15 frames. ], batch size: 24, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:52:56,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0891, 1.6831, 4.3541, 1.9096, 2.4249, 4.9102, 4.9363, 4.2484], + device='cuda:2'), covar=tensor([0.1180, 0.1642, 0.0271, 0.1974, 0.1150, 0.0179, 0.0347, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0260, 0.0293, 0.0254, 0.0287, 0.0267, 0.0231, 0.0333, 0.0288], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:2') +2023-02-06 10:53:07,899 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.592e+02 3.192e+02 4.476e+02 1.804e+03, threshold=6.384e+02, percent-clipped=5.0 +2023-02-06 10:53:14,449 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:53:30,909 INFO [train.py:901] (2/4) Epoch 11, batch 6250, loss[loss=0.2638, simple_loss=0.3417, pruned_loss=0.09297, over 8286.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3091, pruned_loss=0.07867, over 1611132.56 frames. ], batch size: 23, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:54:06,553 INFO [train.py:901] (2/4) Epoch 11, batch 6300, loss[loss=0.2306, simple_loss=0.3089, pruned_loss=0.07618, over 8342.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3098, pruned_loss=0.07918, over 1614379.41 frames. ], batch size: 24, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:54:19,292 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.563e+02 3.017e+02 3.734e+02 8.364e+02, threshold=6.034e+02, percent-clipped=3.0 +2023-02-06 10:54:36,421 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87173.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:38,304 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87176.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:39,797 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:41,627 INFO [train.py:901] (2/4) Epoch 11, batch 6350, loss[loss=0.2194, simple_loss=0.3018, pruned_loss=0.06847, over 7977.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3089, pruned_loss=0.07842, over 1614592.03 frames. ], batch size: 21, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:54:53,184 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87198.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:57,269 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:58,591 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9393, 1.4997, 1.6678, 1.4388, 0.8707, 1.3896, 1.5549, 1.5307], + device='cuda:2'), covar=tensor([0.0500, 0.1187, 0.1693, 0.1301, 0.0609, 0.1486, 0.0694, 0.0611], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0158, 0.0104, 0.0164, 0.0116, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 10:55:03,603 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.61 vs. limit=5.0 +2023-02-06 10:55:16,800 INFO [train.py:901] (2/4) Epoch 11, batch 6400, loss[loss=0.2165, simple_loss=0.3034, pruned_loss=0.06485, over 8364.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3087, pruned_loss=0.07784, over 1619721.86 frames. ], batch size: 24, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:55:19,151 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4952, 1.9285, 2.9684, 2.3580, 2.6528, 2.2781, 1.8643, 1.4362], + device='cuda:2'), covar=tensor([0.3678, 0.4044, 0.1122, 0.2406, 0.1741, 0.1993, 0.1522, 0.3846], + device='cuda:2'), in_proj_covar=tensor([0.0891, 0.0871, 0.0731, 0.0843, 0.0933, 0.0798, 0.0702, 0.0766], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 10:55:23,181 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87240.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:55:28,876 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87248.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:55:29,363 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.577e+02 3.020e+02 3.786e+02 7.428e+02, threshold=6.041e+02, percent-clipped=2.0 +2023-02-06 10:55:51,534 INFO [train.py:901] (2/4) Epoch 11, batch 6450, loss[loss=0.2738, simple_loss=0.352, pruned_loss=0.09779, over 8466.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3095, pruned_loss=0.07847, over 1617260.54 frames. ], batch size: 25, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:55:59,191 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:14,180 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87313.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:27,324 INFO [train.py:901] (2/4) Epoch 11, batch 6500, loss[loss=0.2483, simple_loss=0.3183, pruned_loss=0.08912, over 8513.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3099, pruned_loss=0.0788, over 1617284.03 frames. ], batch size: 29, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:56:32,334 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:36,462 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6829, 5.6884, 5.0250, 2.2693, 5.0754, 5.4081, 5.2295, 5.1648], + device='cuda:2'), covar=tensor([0.0543, 0.0405, 0.0843, 0.4918, 0.0708, 0.0842, 0.1133, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0366, 0.0380, 0.0476, 0.0376, 0.0372, 0.0372, 0.0329], + 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') +2023-02-06 10:56:39,860 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.605e+02 3.245e+02 4.169e+02 7.875e+02, threshold=6.489e+02, percent-clipped=5.0 +2023-02-06 10:56:44,220 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:50,424 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87364.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:57:01,898 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87380.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:57:02,432 INFO [train.py:901] (2/4) Epoch 11, batch 6550, loss[loss=0.2411, simple_loss=0.3193, pruned_loss=0.08147, over 8286.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3099, pruned_loss=0.07917, over 1616310.39 frames. ], batch size: 23, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:57:17,755 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 10:57:37,081 INFO [train.py:901] (2/4) Epoch 11, batch 6600, loss[loss=0.1941, simple_loss=0.2714, pruned_loss=0.05838, over 7806.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3103, pruned_loss=0.07916, over 1621846.06 frames. ], batch size: 19, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:57:37,783 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 10:57:50,090 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.293e+02 2.790e+02 3.732e+02 8.562e+02, threshold=5.581e+02, percent-clipped=1.0 +2023-02-06 10:57:50,526 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 +2023-02-06 10:58:11,509 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:58:11,545 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87479.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:58:12,690 INFO [train.py:901] (2/4) Epoch 11, batch 6650, loss[loss=0.2044, simple_loss=0.286, pruned_loss=0.06146, over 8067.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3091, pruned_loss=0.07814, over 1622677.22 frames. ], batch size: 21, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 10:58:12,874 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1620, 2.1610, 1.9387, 2.0830, 1.1135, 1.8101, 2.2068, 2.3107], + device='cuda:2'), covar=tensor([0.0375, 0.1085, 0.1632, 0.1137, 0.0550, 0.1323, 0.0607, 0.0519], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0159, 0.0104, 0.0164, 0.0116, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:2') +2023-02-06 10:58:35,863 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 10:58:41,518 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87523.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:58:47,482 INFO [train.py:901] (2/4) Epoch 11, batch 6700, loss[loss=0.2605, simple_loss=0.3268, pruned_loss=0.09714, over 8471.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3091, pruned_loss=0.07866, over 1622799.40 frames. ], batch size: 29, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 10:58:58,351 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:58:59,467 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.493e+02 3.158e+02 4.170e+02 8.693e+02, threshold=6.316e+02, percent-clipped=8.0 +2023-02-06 10:59:16,933 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87572.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:22,975 INFO [train.py:901] (2/4) Epoch 11, batch 6750, loss[loss=0.2622, simple_loss=0.3257, pruned_loss=0.09937, over 8136.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.309, pruned_loss=0.0789, over 1617188.43 frames. ], batch size: 22, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 10:59:30,579 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:37,578 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87602.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:40,785 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:43,424 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:49,572 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8569, 3.7561, 2.3331, 2.3299, 2.6411, 1.8817, 2.2940, 2.6861], + device='cuda:2'), covar=tensor([0.1708, 0.0339, 0.1039, 0.0918, 0.0644, 0.1350, 0.1148, 0.1125], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0231, 0.0312, 0.0293, 0.0298, 0.0316, 0.0335, 0.0301], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 10:59:52,301 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:56,901 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 10:59:57,591 INFO [train.py:901] (2/4) Epoch 11, batch 6800, loss[loss=0.2055, simple_loss=0.2784, pruned_loss=0.06634, over 7514.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3089, pruned_loss=0.07952, over 1615114.76 frames. ], batch size: 18, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 11:00:01,243 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:00:10,523 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.375e+02 2.980e+02 3.798e+02 7.616e+02, threshold=5.961e+02, percent-clipped=2.0 +2023-02-06 11:00:32,377 INFO [train.py:901] (2/4) Epoch 11, batch 6850, loss[loss=0.2215, simple_loss=0.2967, pruned_loss=0.07311, over 7784.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3101, pruned_loss=0.08013, over 1612444.85 frames. ], batch size: 19, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 11:00:33,307 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5321, 2.1041, 3.4932, 1.3368, 2.4846, 1.9868, 1.6813, 2.3836], + device='cuda:2'), covar=tensor([0.1739, 0.2276, 0.0677, 0.3842, 0.1597, 0.2840, 0.1827, 0.2306], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0518, 0.0526, 0.0572, 0.0611, 0.0548, 0.0466, 0.0608], + 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') +2023-02-06 11:00:45,135 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 11:00:50,749 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87707.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:01:01,875 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87724.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:01:06,414 INFO [train.py:901] (2/4) Epoch 11, batch 6900, loss[loss=0.1918, simple_loss=0.2728, pruned_loss=0.05539, over 7647.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.31, pruned_loss=0.07977, over 1615199.87 frames. ], batch size: 19, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 11:01:10,021 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87735.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:01:19,190 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.628e+02 3.043e+02 4.130e+02 7.700e+02, threshold=6.086e+02, percent-clipped=2.0 +2023-02-06 11:01:20,945 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 11:01:26,849 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87760.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:01:41,613 INFO [train.py:901] (2/4) Epoch 11, batch 6950, loss[loss=0.2231, simple_loss=0.3084, pruned_loss=0.06893, over 8240.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3098, pruned_loss=0.0793, over 1615110.22 frames. ], batch size: 22, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:01:49,691 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-06 11:01:50,314 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 11:01:52,562 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 11:02:11,620 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87823.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:02:16,826 INFO [train.py:901] (2/4) Epoch 11, batch 7000, loss[loss=0.2193, simple_loss=0.2873, pruned_loss=0.0756, over 7709.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3087, pruned_loss=0.07899, over 1610960.86 frames. ], batch size: 18, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:02:22,314 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:02:29,502 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.548e+02 3.185e+02 4.052e+02 9.283e+02, threshold=6.369e+02, percent-clipped=6.0 +2023-02-06 11:02:41,530 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:02:51,623 INFO [train.py:901] (2/4) Epoch 11, batch 7050, loss[loss=0.2725, simple_loss=0.3473, pruned_loss=0.09881, over 8336.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3102, pruned_loss=0.07978, over 1614276.43 frames. ], batch size: 26, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:03:26,705 INFO [train.py:901] (2/4) Epoch 11, batch 7100, loss[loss=0.1944, simple_loss=0.2646, pruned_loss=0.06207, over 7403.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.31, pruned_loss=0.0797, over 1615338.83 frames. ], batch size: 17, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:03:31,613 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87938.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:36,826 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87946.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:38,767 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.732e+02 3.356e+02 4.654e+02 1.650e+03, threshold=6.712e+02, percent-clipped=12.0 +2023-02-06 11:03:40,144 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:48,305 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87963.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:51,560 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:04:00,730 INFO [train.py:901] (2/4) Epoch 11, batch 7150, loss[loss=0.2719, simple_loss=0.3343, pruned_loss=0.1047, over 8458.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3114, pruned_loss=0.08094, over 1611293.33 frames. ], batch size: 39, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:04:01,618 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:04:05,830 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87988.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:04:36,671 INFO [train.py:901] (2/4) Epoch 11, batch 7200, loss[loss=0.2495, simple_loss=0.3314, pruned_loss=0.08379, over 8581.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3114, pruned_loss=0.08093, over 1612202.01 frames. ], batch size: 31, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:04:49,439 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.591e+02 3.086e+02 3.706e+02 9.715e+02, threshold=6.172e+02, percent-clipped=2.0 +2023-02-06 11:04:57,783 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88061.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:01,282 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:11,822 INFO [train.py:901] (2/4) Epoch 11, batch 7250, loss[loss=0.216, simple_loss=0.2951, pruned_loss=0.06843, over 8464.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3106, pruned_loss=0.0802, over 1609564.21 frames. ], batch size: 49, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:05:12,645 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:21,492 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:37,917 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88118.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:39,337 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:40,258 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-06 11:05:46,909 INFO [train.py:901] (2/4) Epoch 11, batch 7300, loss[loss=0.2384, simple_loss=0.318, pruned_loss=0.07939, over 8500.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3101, pruned_loss=0.07962, over 1609023.09 frames. ], batch size: 28, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:06:00,690 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.414e+02 2.958e+02 3.757e+02 7.369e+02, threshold=5.915e+02, percent-clipped=2.0 +2023-02-06 11:06:10,197 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 11:06:22,820 INFO [train.py:901] (2/4) Epoch 11, batch 7350, loss[loss=0.1819, simple_loss=0.2578, pruned_loss=0.05297, over 7705.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3102, pruned_loss=0.0797, over 1606891.59 frames. ], batch size: 18, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:06:32,250 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 11:06:32,470 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88194.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:06:50,351 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:06:51,485 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 11:06:58,211 INFO [train.py:901] (2/4) Epoch 11, batch 7400, loss[loss=0.2622, simple_loss=0.3365, pruned_loss=0.09398, over 8341.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3105, pruned_loss=0.07947, over 1607029.45 frames. ], batch size: 26, lr: 6.84e-03, grad_scale: 16.0 +2023-02-06 11:07:03,169 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88238.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:07:11,892 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.577e+02 3.074e+02 3.691e+02 9.024e+02, threshold=6.148e+02, percent-clipped=4.0 +2023-02-06 11:07:20,894 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:07:26,470 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 11:07:31,795 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7262, 2.3229, 4.4721, 1.4621, 3.0869, 2.2646, 1.9345, 2.8530], + device='cuda:2'), covar=tensor([0.1567, 0.2224, 0.0588, 0.3840, 0.1494, 0.2742, 0.1669, 0.2260], + device='cuda:2'), in_proj_covar=tensor([0.0487, 0.0522, 0.0529, 0.0576, 0.0618, 0.0558, 0.0472, 0.0618], + 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') +2023-02-06 11:07:32,884 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 11:07:33,599 INFO [train.py:901] (2/4) Epoch 11, batch 7450, loss[loss=0.1996, simple_loss=0.2764, pruned_loss=0.0614, over 7759.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3093, pruned_loss=0.07902, over 1607976.19 frames. ], batch size: 19, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:07:52,585 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88309.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:07:58,569 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88317.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:01,847 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:07,738 INFO [train.py:901] (2/4) Epoch 11, batch 7500, loss[loss=0.2262, simple_loss=0.299, pruned_loss=0.07667, over 7906.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.309, pruned_loss=0.07882, over 1607992.46 frames. ], batch size: 20, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:08:13,326 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:15,898 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:19,203 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:20,935 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.827e+02 3.509e+02 4.304e+02 1.282e+03, threshold=7.018e+02, percent-clipped=8.0 +2023-02-06 11:08:30,002 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:42,589 INFO [train.py:901] (2/4) Epoch 11, batch 7550, loss[loss=0.2403, simple_loss=0.3229, pruned_loss=0.0789, over 8362.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3105, pruned_loss=0.07964, over 1609899.34 frames. ], batch size: 26, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:09:17,376 INFO [train.py:901] (2/4) Epoch 11, batch 7600, loss[loss=0.255, simple_loss=0.321, pruned_loss=0.09453, over 7307.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3106, pruned_loss=0.08023, over 1611199.59 frames. ], batch size: 72, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:09:31,032 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.441e+02 2.975e+02 3.888e+02 6.138e+02, threshold=5.951e+02, percent-clipped=0.0 +2023-02-06 11:09:39,150 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:09:49,837 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0772, 2.9960, 2.1758, 4.2563, 1.5119, 2.0030, 2.1950, 2.9919], + device='cuda:2'), covar=tensor([0.1029, 0.0697, 0.1145, 0.0233, 0.1156, 0.1366, 0.1071, 0.0716], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0213, 0.0256, 0.0216, 0.0220, 0.0255, 0.0256, 0.0222], + 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') +2023-02-06 11:09:51,498 INFO [train.py:901] (2/4) Epoch 11, batch 7650, loss[loss=0.1819, simple_loss=0.2636, pruned_loss=0.05008, over 7225.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3111, pruned_loss=0.07991, over 1614313.51 frames. ], batch size: 16, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:10:24,205 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-06 11:10:24,769 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 +2023-02-06 11:10:26,441 INFO [train.py:901] (2/4) Epoch 11, batch 7700, loss[loss=0.2447, simple_loss=0.3223, pruned_loss=0.08353, over 8024.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3107, pruned_loss=0.07975, over 1614737.86 frames. ], batch size: 22, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:10:39,149 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 11:10:39,701 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.472e+02 3.053e+02 3.571e+02 8.603e+02, threshold=6.105e+02, percent-clipped=3.0 +2023-02-06 11:10:58,418 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88577.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:11:00,812 INFO [train.py:901] (2/4) Epoch 11, batch 7750, loss[loss=0.1729, simple_loss=0.2494, pruned_loss=0.0482, over 7427.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3102, pruned_loss=0.07984, over 1616422.33 frames. ], batch size: 17, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:11:36,340 INFO [train.py:901] (2/4) Epoch 11, batch 7800, loss[loss=0.2222, simple_loss=0.2983, pruned_loss=0.07309, over 7975.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3111, pruned_loss=0.08027, over 1610961.15 frames. ], batch size: 21, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:11:37,767 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6505, 2.8516, 1.9156, 2.2847, 2.3206, 1.5892, 2.2124, 2.2488], + device='cuda:2'), covar=tensor([0.1458, 0.0351, 0.1112, 0.0671, 0.0621, 0.1430, 0.1002, 0.0986], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0236, 0.0318, 0.0299, 0.0302, 0.0323, 0.0339, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:11:48,833 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.685e+02 3.345e+02 4.152e+02 1.012e+03, threshold=6.690e+02, percent-clipped=6.0 +2023-02-06 11:11:50,872 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88653.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:11:51,150 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 11:12:09,486 INFO [train.py:901] (2/4) Epoch 11, batch 7850, loss[loss=0.2377, simple_loss=0.3047, pruned_loss=0.08535, over 7923.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3109, pruned_loss=0.0807, over 1610416.92 frames. ], batch size: 20, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:12:33,306 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0775, 1.7302, 2.5131, 1.9709, 2.3000, 1.9186, 1.5659, 1.0225], + device='cuda:2'), covar=tensor([0.4077, 0.3842, 0.1141, 0.2532, 0.1787, 0.2406, 0.1840, 0.3706], + device='cuda:2'), in_proj_covar=tensor([0.0886, 0.0865, 0.0727, 0.0842, 0.0925, 0.0798, 0.0701, 0.0763], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:12:42,909 INFO [train.py:901] (2/4) Epoch 11, batch 7900, loss[loss=0.1805, simple_loss=0.265, pruned_loss=0.048, over 7807.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3103, pruned_loss=0.0801, over 1608520.95 frames. ], batch size: 19, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:12:55,418 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.490e+02 3.060e+02 3.735e+02 6.734e+02, threshold=6.120e+02, percent-clipped=1.0 +2023-02-06 11:13:07,255 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88768.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:13:15,813 INFO [train.py:901] (2/4) Epoch 11, batch 7950, loss[loss=0.2342, simple_loss=0.3128, pruned_loss=0.07781, over 8520.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3114, pruned_loss=0.0803, over 1613840.56 frames. ], batch size: 28, lr: 6.81e-03, grad_scale: 16.0 +2023-02-06 11:13:46,952 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3886, 2.7556, 3.1202, 1.4699, 3.4161, 1.9076, 1.5480, 2.0065], + device='cuda:2'), covar=tensor([0.0533, 0.0234, 0.0206, 0.0485, 0.0282, 0.0582, 0.0580, 0.0336], + device='cuda:2'), in_proj_covar=tensor([0.0386, 0.0322, 0.0270, 0.0376, 0.0308, 0.0468, 0.0353, 0.0347], + device='cuda:2'), out_proj_covar=tensor([1.0990e-04, 8.9445e-05, 7.5304e-05, 1.0534e-04, 8.7000e-05, 1.4238e-04, + 1.0043e-04, 9.8118e-05], device='cuda:2') +2023-02-06 11:13:49,364 INFO [train.py:901] (2/4) Epoch 11, batch 8000, loss[loss=0.2166, simple_loss=0.28, pruned_loss=0.07659, over 7699.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3098, pruned_loss=0.07947, over 1605800.79 frames. ], batch size: 18, lr: 6.81e-03, grad_scale: 16.0 +2023-02-06 11:13:50,909 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:14:02,010 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.603e+02 3.071e+02 3.730e+02 8.421e+02, threshold=6.141e+02, percent-clipped=3.0 +2023-02-06 11:14:07,241 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88858.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:14:22,236 INFO [train.py:901] (2/4) Epoch 11, batch 8050, loss[loss=0.1923, simple_loss=0.2607, pruned_loss=0.06197, over 7226.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3068, pruned_loss=0.07819, over 1592394.89 frames. ], batch size: 16, lr: 6.81e-03, grad_scale: 16.0 +2023-02-06 11:14:41,733 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9960, 2.4017, 1.9312, 3.0288, 1.3319, 1.5821, 1.8301, 2.4927], + device='cuda:2'), covar=tensor([0.0775, 0.0795, 0.0970, 0.0332, 0.1162, 0.1435, 0.1043, 0.0674], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0216, 0.0256, 0.0215, 0.0219, 0.0253, 0.0258, 0.0222], + 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') +2023-02-06 11:14:54,231 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 11:14:58,686 INFO [train.py:901] (2/4) Epoch 12, batch 0, loss[loss=0.2166, simple_loss=0.2942, pruned_loss=0.06946, over 8038.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2942, pruned_loss=0.06946, over 8038.00 frames. ], batch size: 22, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:14:58,687 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 11:15:09,781 INFO [train.py:935] (2/4) Epoch 12, validation: loss=0.1897, simple_loss=0.2896, pruned_loss=0.04486, over 944034.00 frames. +2023-02-06 11:15:09,782 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 11:15:23,303 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 11:15:35,201 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.690e+02 3.540e+02 4.339e+02 7.249e+02, threshold=7.080e+02, percent-clipped=5.0 +2023-02-06 11:15:44,686 INFO [train.py:901] (2/4) Epoch 12, batch 50, loss[loss=0.2658, simple_loss=0.3426, pruned_loss=0.09449, over 8444.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3155, pruned_loss=0.08072, over 366782.36 frames. ], batch size: 27, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:15:57,434 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 11:16:19,045 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 11:16:19,755 INFO [train.py:901] (2/4) Epoch 12, batch 100, loss[loss=0.2075, simple_loss=0.2887, pruned_loss=0.06314, over 7971.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3125, pruned_loss=0.08022, over 646310.94 frames. ], batch size: 21, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:16:26,516 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89024.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:16:33,361 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.50 vs. limit=5.0 +2023-02-06 11:16:40,637 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:16:43,468 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89049.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:16:43,913 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.771e+02 3.256e+02 4.152e+02 1.357e+03, threshold=6.512e+02, percent-clipped=1.0 +2023-02-06 11:16:54,731 INFO [train.py:901] (2/4) Epoch 12, batch 150, loss[loss=0.234, simple_loss=0.3124, pruned_loss=0.0778, over 7648.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3101, pruned_loss=0.07862, over 859261.58 frames. ], batch size: 19, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:17:24,723 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 11:17:29,009 INFO [train.py:901] (2/4) Epoch 12, batch 200, loss[loss=0.2317, simple_loss=0.3064, pruned_loss=0.07851, over 7548.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3106, pruned_loss=0.07968, over 1026211.66 frames. ], batch size: 18, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:17:53,943 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 2.712e+02 3.423e+02 4.383e+02 1.008e+03, threshold=6.845e+02, percent-clipped=3.0 +2023-02-06 11:18:03,547 INFO [train.py:901] (2/4) Epoch 12, batch 250, loss[loss=0.221, simple_loss=0.2991, pruned_loss=0.07145, over 8510.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3113, pruned_loss=0.07958, over 1162040.51 frames. ], batch size: 26, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:18:13,231 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 11:18:17,197 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.33 vs. limit=5.0 +2023-02-06 11:18:20,886 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89187.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:18:22,844 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 11:18:40,040 INFO [train.py:901] (2/4) Epoch 12, batch 300, loss[loss=0.1816, simple_loss=0.2666, pruned_loss=0.04824, over 7806.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3118, pruned_loss=0.07972, over 1264953.02 frames. ], batch size: 20, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:19:05,001 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.536e+02 3.052e+02 3.921e+02 6.584e+02, threshold=6.103e+02, percent-clipped=0.0 +2023-02-06 11:19:14,500 INFO [train.py:901] (2/4) Epoch 12, batch 350, loss[loss=0.2412, simple_loss=0.3213, pruned_loss=0.08062, over 8317.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3108, pruned_loss=0.07992, over 1343850.90 frames. ], batch size: 25, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:19:49,367 INFO [train.py:901] (2/4) Epoch 12, batch 400, loss[loss=0.2616, simple_loss=0.3295, pruned_loss=0.09679, over 8617.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3103, pruned_loss=0.07991, over 1403050.63 frames. ], batch size: 39, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:20:14,265 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.417e+02 2.965e+02 3.513e+02 5.511e+02, threshold=5.929e+02, percent-clipped=0.0 +2023-02-06 11:20:24,218 INFO [train.py:901] (2/4) Epoch 12, batch 450, loss[loss=0.2268, simple_loss=0.3118, pruned_loss=0.07087, over 8360.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3096, pruned_loss=0.0793, over 1452148.55 frames. ], batch size: 24, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:20:25,042 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1173, 1.0779, 4.2985, 1.5538, 3.7510, 3.5478, 3.8357, 3.6895], + device='cuda:2'), covar=tensor([0.0529, 0.4630, 0.0418, 0.3389, 0.1016, 0.0842, 0.0554, 0.0667], + device='cuda:2'), in_proj_covar=tensor([0.0490, 0.0571, 0.0579, 0.0529, 0.0606, 0.0511, 0.0507, 0.0575], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:20:40,988 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:20:43,195 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:20:58,861 INFO [train.py:901] (2/4) Epoch 12, batch 500, loss[loss=0.2395, simple_loss=0.314, pruned_loss=0.08249, over 8485.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3103, pruned_loss=0.07917, over 1489783.61 frames. ], batch size: 28, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:21:19,376 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:21:22,222 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5374, 1.8663, 3.4222, 1.3297, 2.5759, 2.0078, 1.5955, 2.4302], + device='cuda:2'), covar=tensor([0.1642, 0.2440, 0.0687, 0.3913, 0.1508, 0.2739, 0.1827, 0.2156], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0521, 0.0534, 0.0580, 0.0619, 0.0555, 0.0473, 0.0612], + 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') +2023-02-06 11:21:24,101 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.539e+02 3.031e+02 3.696e+02 8.346e+02, threshold=6.063e+02, percent-clipped=3.0 +2023-02-06 11:21:29,672 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89457.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:21:34,346 INFO [train.py:901] (2/4) Epoch 12, batch 550, loss[loss=0.2174, simple_loss=0.3016, pruned_loss=0.06658, over 8245.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3101, pruned_loss=0.07971, over 1516795.21 frames. ], batch size: 22, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:21:53,123 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7725, 2.1966, 1.7775, 2.7268, 1.1524, 1.5517, 1.7324, 2.2617], + device='cuda:2'), covar=tensor([0.0895, 0.0826, 0.1041, 0.0417, 0.1309, 0.1494, 0.1131, 0.0766], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0213, 0.0255, 0.0214, 0.0218, 0.0251, 0.0257, 0.0220], + 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') +2023-02-06 11:22:02,587 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:22:09,198 INFO [train.py:901] (2/4) Epoch 12, batch 600, loss[loss=0.2565, simple_loss=0.313, pruned_loss=0.09996, over 7798.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3109, pruned_loss=0.08027, over 1543311.56 frames. ], batch size: 20, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:22:17,835 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:22:19,216 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0189, 1.7512, 3.0837, 1.2931, 2.1344, 3.2861, 3.4623, 2.8498], + device='cuda:2'), covar=tensor([0.0972, 0.1379, 0.0387, 0.2155, 0.0972, 0.0297, 0.0505, 0.0621], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0298, 0.0262, 0.0294, 0.0273, 0.0237, 0.0344, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:22:20,461 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:22:26,583 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 11:22:34,513 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.630e+02 3.047e+02 3.733e+02 1.036e+03, threshold=6.094e+02, percent-clipped=2.0 +2023-02-06 11:22:44,043 INFO [train.py:901] (2/4) Epoch 12, batch 650, loss[loss=0.2416, simple_loss=0.3215, pruned_loss=0.08082, over 8286.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3113, pruned_loss=0.08, over 1562947.09 frames. ], batch size: 23, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:23:18,860 INFO [train.py:901] (2/4) Epoch 12, batch 700, loss[loss=0.287, simple_loss=0.3565, pruned_loss=0.1087, over 8446.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3116, pruned_loss=0.08032, over 1575915.38 frames. ], batch size: 27, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:23:40,443 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:23:43,633 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.821e+02 3.296e+02 4.031e+02 9.579e+02, threshold=6.593e+02, percent-clipped=5.0 +2023-02-06 11:23:53,838 INFO [train.py:901] (2/4) Epoch 12, batch 750, loss[loss=0.217, simple_loss=0.2869, pruned_loss=0.07357, over 7659.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3115, pruned_loss=0.0804, over 1583965.00 frames. ], batch size: 19, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:24:02,096 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3363, 1.4694, 4.4533, 2.0411, 2.4567, 5.1291, 5.1549, 4.4783], + device='cuda:2'), covar=tensor([0.1057, 0.1858, 0.0261, 0.1851, 0.1129, 0.0203, 0.0372, 0.0585], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0300, 0.0264, 0.0296, 0.0276, 0.0239, 0.0347, 0.0293], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:24:04,276 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 +2023-02-06 11:24:11,508 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 11:24:17,639 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:24:20,256 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 11:24:28,186 INFO [train.py:901] (2/4) Epoch 12, batch 800, loss[loss=0.2293, simple_loss=0.3204, pruned_loss=0.06914, over 8492.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.311, pruned_loss=0.08024, over 1589993.38 frames. ], batch size: 29, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:24:32,437 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0000, 2.7754, 3.4411, 2.0348, 1.7656, 3.4287, 0.6746, 2.1560], + device='cuda:2'), covar=tensor([0.1847, 0.1308, 0.0339, 0.2427, 0.3718, 0.0318, 0.3867, 0.2017], + device='cuda:2'), in_proj_covar=tensor([0.0170, 0.0171, 0.0103, 0.0218, 0.0259, 0.0107, 0.0167, 0.0169], + 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') +2023-02-06 11:24:43,682 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:24:53,390 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.628e+02 3.285e+02 4.121e+02 9.349e+02, threshold=6.571e+02, percent-clipped=6.0 +2023-02-06 11:24:59,659 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:25:02,837 INFO [train.py:901] (2/4) Epoch 12, batch 850, loss[loss=0.232, simple_loss=0.3072, pruned_loss=0.07838, over 7980.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3097, pruned_loss=0.07923, over 1594926.61 frames. ], batch size: 21, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:25:18,006 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:25:19,284 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:25:28,670 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89801.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:25:34,027 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3615, 1.6529, 1.6558, 0.9248, 1.6958, 1.2495, 0.2785, 1.5678], + device='cuda:2'), covar=tensor([0.0333, 0.0224, 0.0168, 0.0347, 0.0285, 0.0673, 0.0587, 0.0189], + device='cuda:2'), in_proj_covar=tensor([0.0391, 0.0330, 0.0275, 0.0384, 0.0317, 0.0474, 0.0356, 0.0353], + device='cuda:2'), out_proj_covar=tensor([1.1120e-04, 9.1706e-05, 7.6565e-05, 1.0750e-04, 8.9654e-05, 1.4412e-04, + 1.0104e-04, 9.9689e-05], device='cuda:2') +2023-02-06 11:25:37,801 INFO [train.py:901] (2/4) Epoch 12, batch 900, loss[loss=0.2165, simple_loss=0.2939, pruned_loss=0.06958, over 7978.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3101, pruned_loss=0.07884, over 1604277.72 frames. ], batch size: 21, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:26:03,292 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.446e+02 3.021e+02 3.729e+02 6.397e+02, threshold=6.041e+02, percent-clipped=0.0 +2023-02-06 11:26:03,490 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:11,307 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6048, 1.8878, 1.5920, 2.2895, 1.0003, 1.3928, 1.4741, 1.8623], + device='cuda:2'), covar=tensor([0.0856, 0.0826, 0.0979, 0.0418, 0.1199, 0.1486, 0.1031, 0.0761], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0213, 0.0255, 0.0215, 0.0218, 0.0252, 0.0258, 0.0218], + 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') +2023-02-06 11:26:11,795 INFO [train.py:901] (2/4) Epoch 12, batch 950, loss[loss=0.2267, simple_loss=0.2999, pruned_loss=0.07678, over 8030.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3094, pruned_loss=0.07859, over 1604868.98 frames. ], batch size: 22, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:26:16,404 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89871.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:24,521 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89883.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:33,230 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7528, 2.2426, 3.6921, 2.4765, 3.1468, 2.4367, 2.0400, 1.7255], + device='cuda:2'), covar=tensor([0.3813, 0.4272, 0.1178, 0.2922, 0.1968, 0.2186, 0.1588, 0.4468], + device='cuda:2'), in_proj_covar=tensor([0.0895, 0.0876, 0.0743, 0.0853, 0.0936, 0.0806, 0.0708, 0.0772], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:26:38,570 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:38,612 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:39,087 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 11:26:46,279 INFO [train.py:901] (2/4) Epoch 12, batch 1000, loss[loss=0.2129, simple_loss=0.2801, pruned_loss=0.07288, over 7420.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3102, pruned_loss=0.07879, over 1609511.14 frames. ], batch size: 17, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:26:47,816 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89916.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:26:48,419 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9439, 2.8620, 3.5215, 2.0191, 1.6545, 3.6618, 0.6037, 2.1063], + device='cuda:2'), covar=tensor([0.2507, 0.1254, 0.0387, 0.2788, 0.4494, 0.0285, 0.3965, 0.2236], + device='cuda:2'), in_proj_covar=tensor([0.0170, 0.0172, 0.0103, 0.0218, 0.0259, 0.0107, 0.0166, 0.0169], + 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') +2023-02-06 11:26:55,680 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:27:11,383 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.648e+02 3.254e+02 4.081e+02 9.414e+02, threshold=6.507e+02, percent-clipped=7.0 +2023-02-06 11:27:11,408 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 11:27:13,883 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 11:27:20,836 INFO [train.py:901] (2/4) Epoch 12, batch 1050, loss[loss=0.2964, simple_loss=0.362, pruned_loss=0.1155, over 8490.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3102, pruned_loss=0.07876, over 1610328.59 frames. ], batch size: 28, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:27:22,384 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0677, 1.3748, 6.1377, 2.2005, 5.5044, 5.2512, 5.8017, 5.5569], + device='cuda:2'), covar=tensor([0.0361, 0.4507, 0.0333, 0.3100, 0.0850, 0.0736, 0.0305, 0.0408], + device='cuda:2'), in_proj_covar=tensor([0.0501, 0.0586, 0.0593, 0.0543, 0.0620, 0.0528, 0.0520, 0.0592], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:27:24,339 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 11:27:35,860 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89986.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:27:56,235 INFO [train.py:901] (2/4) Epoch 12, batch 1100, loss[loss=0.2943, simple_loss=0.3594, pruned_loss=0.1146, over 8328.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3102, pruned_loss=0.07903, over 1612662.42 frames. ], batch size: 49, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:28:04,632 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1908, 2.1776, 1.5813, 1.9480, 1.8046, 1.2853, 1.7221, 1.7198], + device='cuda:2'), covar=tensor([0.1207, 0.0366, 0.1129, 0.0505, 0.0636, 0.1457, 0.0755, 0.0730], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0240, 0.0316, 0.0297, 0.0301, 0.0323, 0.0336, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:28:16,467 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:28:22,907 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.536e+02 3.046e+02 3.976e+02 6.882e+02, threshold=6.092e+02, percent-clipped=1.0 +2023-02-06 11:28:31,018 INFO [train.py:901] (2/4) Epoch 12, batch 1150, loss[loss=0.226, simple_loss=0.2919, pruned_loss=0.0801, over 7697.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3105, pruned_loss=0.07904, over 1615925.56 frames. ], batch size: 18, lr: 6.48e-03, grad_scale: 4.0 +2023-02-06 11:28:34,457 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 11:29:01,100 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:05,604 INFO [train.py:901] (2/4) Epoch 12, batch 1200, loss[loss=0.2226, simple_loss=0.2937, pruned_loss=0.07578, over 7811.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3107, pruned_loss=0.07957, over 1614369.46 frames. ], batch size: 20, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:29:19,570 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:24,287 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1723, 1.0342, 1.2694, 1.1273, 0.9227, 1.2957, 0.0421, 0.9136], + device='cuda:2'), covar=tensor([0.2056, 0.1799, 0.0599, 0.1310, 0.3839, 0.0606, 0.3015, 0.1717], + device='cuda:2'), in_proj_covar=tensor([0.0168, 0.0171, 0.0103, 0.0218, 0.0258, 0.0107, 0.0165, 0.0168], + 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') +2023-02-06 11:29:30,358 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90148.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:32,963 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.449e+02 3.099e+02 4.282e+02 6.791e+02, threshold=6.197e+02, percent-clipped=4.0 +2023-02-06 11:29:36,556 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:37,305 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90158.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:41,830 INFO [train.py:901] (2/4) Epoch 12, batch 1250, loss[loss=0.2288, simple_loss=0.3004, pruned_loss=0.07866, over 7631.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3103, pruned_loss=0.07919, over 1613744.95 frames. ], batch size: 19, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:29:47,513 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90172.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:29:53,674 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0672, 1.6405, 1.3788, 1.5902, 1.3649, 1.2190, 1.2789, 1.4443], + device='cuda:2'), covar=tensor([0.1033, 0.0436, 0.1173, 0.0535, 0.0636, 0.1292, 0.0840, 0.0665], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0240, 0.0318, 0.0300, 0.0302, 0.0325, 0.0339, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:29:55,720 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:04,776 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1217, 2.8828, 3.4701, 2.3903, 1.9116, 3.4733, 0.7979, 2.1428], + device='cuda:2'), covar=tensor([0.1864, 0.1429, 0.0402, 0.2128, 0.3537, 0.0465, 0.3360, 0.1948], + device='cuda:2'), in_proj_covar=tensor([0.0169, 0.0172, 0.0104, 0.0218, 0.0258, 0.0107, 0.0165, 0.0169], + 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') +2023-02-06 11:30:05,421 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90197.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:30:17,139 INFO [train.py:901] (2/4) Epoch 12, batch 1300, loss[loss=0.2548, simple_loss=0.3327, pruned_loss=0.0885, over 8523.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.31, pruned_loss=0.07872, over 1614304.31 frames. ], batch size: 26, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:30:26,145 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90227.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:37,502 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90242.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:43,700 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:44,956 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.408e+02 3.209e+02 4.069e+02 1.568e+03, threshold=6.418e+02, percent-clipped=9.0 +2023-02-06 11:30:53,233 INFO [train.py:901] (2/4) Epoch 12, batch 1350, loss[loss=0.3337, simple_loss=0.3755, pruned_loss=0.1459, over 6879.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.31, pruned_loss=0.07858, over 1616226.09 frames. ], batch size: 71, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:30:55,572 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90267.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:31:19,418 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 11:31:28,605 INFO [train.py:901] (2/4) Epoch 12, batch 1400, loss[loss=0.1898, simple_loss=0.2751, pruned_loss=0.05225, over 8080.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3095, pruned_loss=0.0783, over 1618472.62 frames. ], batch size: 21, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:31:47,914 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:31:54,627 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.398e+02 2.808e+02 3.540e+02 8.131e+02, threshold=5.617e+02, percent-clipped=1.0 +2023-02-06 11:32:03,611 INFO [train.py:901] (2/4) Epoch 12, batch 1450, loss[loss=0.1829, simple_loss=0.2779, pruned_loss=0.04396, over 8246.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3077, pruned_loss=0.07702, over 1618237.58 frames. ], batch size: 24, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:32:07,896 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90369.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:32:08,443 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 11:32:38,149 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:32:38,609 INFO [train.py:901] (2/4) Epoch 12, batch 1500, loss[loss=0.2349, simple_loss=0.3195, pruned_loss=0.07512, over 8335.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3078, pruned_loss=0.07719, over 1618916.62 frames. ], batch size: 26, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:32:50,009 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6913, 1.9804, 2.2103, 1.2450, 2.3517, 1.4210, 0.6466, 1.8811], + device='cuda:2'), covar=tensor([0.0432, 0.0222, 0.0187, 0.0441, 0.0238, 0.0653, 0.0603, 0.0221], + device='cuda:2'), in_proj_covar=tensor([0.0395, 0.0332, 0.0278, 0.0388, 0.0317, 0.0474, 0.0359, 0.0356], + device='cuda:2'), out_proj_covar=tensor([1.1232e-04, 9.2314e-05, 7.7557e-05, 1.0856e-04, 8.9250e-05, 1.4397e-04, + 1.0217e-04, 1.0033e-04], device='cuda:2') +2023-02-06 11:32:55,142 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:33:04,310 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 2.462e+02 2.993e+02 3.898e+02 9.256e+02, threshold=5.985e+02, percent-clipped=2.0 +2023-02-06 11:33:12,499 INFO [train.py:901] (2/4) Epoch 12, batch 1550, loss[loss=0.2391, simple_loss=0.305, pruned_loss=0.08661, over 7973.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3081, pruned_loss=0.0773, over 1617283.35 frames. ], batch size: 21, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:33:15,003 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 11:33:21,457 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:33:33,067 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90492.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:33:48,745 INFO [train.py:901] (2/4) Epoch 12, batch 1600, loss[loss=0.2857, simple_loss=0.3534, pruned_loss=0.109, over 8360.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3065, pruned_loss=0.07667, over 1616111.92 frames. ], batch size: 26, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:33:48,905 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90514.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:15,464 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.712e+02 3.378e+02 4.197e+02 8.231e+02, threshold=6.755e+02, percent-clipped=6.0 +2023-02-06 11:34:23,535 INFO [train.py:901] (2/4) Epoch 12, batch 1650, loss[loss=0.2154, simple_loss=0.3008, pruned_loss=0.06503, over 8018.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3071, pruned_loss=0.07681, over 1617692.94 frames. ], batch size: 22, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:34:43,459 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:47,049 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:53,818 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:58,216 INFO [train.py:901] (2/4) Epoch 12, batch 1700, loss[loss=0.2379, simple_loss=0.3024, pruned_loss=0.0867, over 7915.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3075, pruned_loss=0.07691, over 1621207.16 frames. ], batch size: 20, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:35:04,324 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:35:19,239 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3520, 1.4687, 1.3377, 1.9003, 0.7176, 1.1822, 1.3037, 1.5626], + device='cuda:2'), covar=tensor([0.0850, 0.0850, 0.1119, 0.0512, 0.1259, 0.1523, 0.0816, 0.0726], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0213, 0.0255, 0.0216, 0.0218, 0.0251, 0.0256, 0.0219], + 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') +2023-02-06 11:35:24,539 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.486e+02 2.952e+02 3.646e+02 6.764e+02, threshold=5.904e+02, percent-clipped=1.0 +2023-02-06 11:35:33,331 INFO [train.py:901] (2/4) Epoch 12, batch 1750, loss[loss=0.203, simple_loss=0.2775, pruned_loss=0.06427, over 7802.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3074, pruned_loss=0.0766, over 1623468.86 frames. ], batch size: 20, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:36:04,031 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:36:07,401 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:36:08,038 INFO [train.py:901] (2/4) Epoch 12, batch 1800, loss[loss=0.2563, simple_loss=0.3342, pruned_loss=0.0892, over 8522.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3073, pruned_loss=0.07688, over 1617585.11 frames. ], batch size: 26, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:36:35,309 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.625e+02 3.119e+02 3.569e+02 7.012e+02, threshold=6.239e+02, percent-clipped=2.0 +2023-02-06 11:36:43,315 INFO [train.py:901] (2/4) Epoch 12, batch 1850, loss[loss=0.2595, simple_loss=0.337, pruned_loss=0.09099, over 8479.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3064, pruned_loss=0.07658, over 1612422.38 frames. ], batch size: 28, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:36:50,561 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9202, 1.5123, 6.0247, 2.2396, 5.3817, 5.0860, 5.6315, 5.3962], + device='cuda:2'), covar=tensor([0.0410, 0.4619, 0.0332, 0.3207, 0.0948, 0.0683, 0.0378, 0.0465], + device='cuda:2'), in_proj_covar=tensor([0.0503, 0.0590, 0.0598, 0.0541, 0.0618, 0.0526, 0.0524, 0.0592], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:37:17,707 INFO [train.py:901] (2/4) Epoch 12, batch 1900, loss[loss=0.1957, simple_loss=0.2683, pruned_loss=0.06156, over 7537.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3057, pruned_loss=0.07667, over 1611332.68 frames. ], batch size: 18, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:37:22,472 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90821.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:27,316 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:31,347 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2826, 1.8955, 2.6866, 2.0889, 2.4609, 2.1821, 1.8201, 1.1308], + device='cuda:2'), covar=tensor([0.3723, 0.3768, 0.1214, 0.2933, 0.1999, 0.2126, 0.1586, 0.4410], + device='cuda:2'), in_proj_covar=tensor([0.0889, 0.0874, 0.0740, 0.0852, 0.0937, 0.0809, 0.0708, 0.0773], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:37:36,340 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 11:37:44,436 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.569e+02 3.031e+02 3.632e+02 7.649e+02, threshold=6.063e+02, percent-clipped=2.0 +2023-02-06 11:37:47,237 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 11:37:48,678 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90858.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:52,168 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90863.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:52,633 INFO [train.py:901] (2/4) Epoch 12, batch 1950, loss[loss=0.2234, simple_loss=0.302, pruned_loss=0.0724, over 7660.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3037, pruned_loss=0.07574, over 1605171.34 frames. ], batch size: 19, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:37:55,457 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:59,339 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 11:38:10,330 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:38:19,026 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 11:38:27,977 INFO [train.py:901] (2/4) Epoch 12, batch 2000, loss[loss=0.2429, simple_loss=0.3196, pruned_loss=0.08315, over 8608.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3049, pruned_loss=0.07605, over 1609057.69 frames. ], batch size: 31, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:38:43,401 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:38:54,924 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.641e+02 3.163e+02 4.034e+02 9.087e+02, threshold=6.326e+02, percent-clipped=9.0 +2023-02-06 11:39:02,895 INFO [train.py:901] (2/4) Epoch 12, batch 2050, loss[loss=0.2174, simple_loss=0.2991, pruned_loss=0.06783, over 8535.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3065, pruned_loss=0.0774, over 1609114.21 frames. ], batch size: 26, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:39:03,761 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:39:09,923 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:39:10,140 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 +2023-02-06 11:39:21,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:39:38,681 INFO [train.py:901] (2/4) Epoch 12, batch 2100, loss[loss=0.2497, simple_loss=0.3183, pruned_loss=0.09055, over 8313.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3087, pruned_loss=0.07828, over 1613411.46 frames. ], batch size: 25, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:40:04,173 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 2.659e+02 3.265e+02 4.247e+02 8.349e+02, threshold=6.531e+02, percent-clipped=2.0 +2023-02-06 11:40:12,108 INFO [train.py:901] (2/4) Epoch 12, batch 2150, loss[loss=0.2016, simple_loss=0.2763, pruned_loss=0.06341, over 7454.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3083, pruned_loss=0.07827, over 1613347.96 frames. ], batch size: 17, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:40:26,821 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:40:30,728 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0587, 1.4926, 1.7635, 1.3672, 0.9376, 1.5623, 1.6981, 1.5090], + device='cuda:2'), covar=tensor([0.0462, 0.1179, 0.1568, 0.1330, 0.0556, 0.1402, 0.0644, 0.0592], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0152, 0.0190, 0.0159, 0.0103, 0.0162, 0.0116, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 11:40:44,051 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:40:47,222 INFO [train.py:901] (2/4) Epoch 12, batch 2200, loss[loss=0.2003, simple_loss=0.2829, pruned_loss=0.05883, over 7249.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.308, pruned_loss=0.07843, over 1616233.86 frames. ], batch size: 16, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:40:51,509 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1999, 1.5417, 3.3605, 1.5249, 2.9839, 2.8564, 3.0693, 2.9628], + device='cuda:2'), covar=tensor([0.0685, 0.3230, 0.0759, 0.3033, 0.1013, 0.0787, 0.0587, 0.0654], + device='cuda:2'), in_proj_covar=tensor([0.0496, 0.0577, 0.0587, 0.0534, 0.0612, 0.0521, 0.0517, 0.0580], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:41:13,716 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.751e+02 3.546e+02 4.173e+02 9.054e+02, threshold=7.092e+02, percent-clipped=3.0 +2023-02-06 11:41:21,753 INFO [train.py:901] (2/4) Epoch 12, batch 2250, loss[loss=0.1903, simple_loss=0.2666, pruned_loss=0.05698, over 7537.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3082, pruned_loss=0.07886, over 1613178.80 frames. ], batch size: 18, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:41:41,137 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91192.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:41:54,522 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91211.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:41:56,402 INFO [train.py:901] (2/4) Epoch 12, batch 2300, loss[loss=0.1648, simple_loss=0.249, pruned_loss=0.04032, over 7453.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3074, pruned_loss=0.07825, over 1610197.50 frames. ], batch size: 17, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:41:58,498 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91217.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:42:07,356 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:42:11,983 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7072, 1.7059, 2.2743, 1.6525, 1.1005, 2.3063, 0.4004, 1.3284], + device='cuda:2'), covar=tensor([0.2479, 0.1651, 0.0404, 0.1792, 0.4131, 0.0400, 0.3282, 0.2048], + device='cuda:2'), in_proj_covar=tensor([0.0169, 0.0174, 0.0104, 0.0219, 0.0255, 0.0107, 0.0167, 0.0170], + 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') +2023-02-06 11:42:23,426 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.635e+02 3.142e+02 4.194e+02 9.102e+02, threshold=6.284e+02, percent-clipped=2.0 +2023-02-06 11:42:24,990 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:42:31,681 INFO [train.py:901] (2/4) Epoch 12, batch 2350, loss[loss=0.2319, simple_loss=0.315, pruned_loss=0.07435, over 8234.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3069, pruned_loss=0.07757, over 1612747.56 frames. ], batch size: 22, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:42:57,738 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91303.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:43:05,700 INFO [train.py:901] (2/4) Epoch 12, batch 2400, loss[loss=0.2385, simple_loss=0.3129, pruned_loss=0.08203, over 8492.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3069, pruned_loss=0.07809, over 1612145.16 frames. ], batch size: 26, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:43:14,304 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:43:31,008 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1662, 3.0512, 2.8422, 1.4698, 2.7576, 2.8350, 2.7889, 2.6434], + device='cuda:2'), covar=tensor([0.1333, 0.0973, 0.1314, 0.5255, 0.1257, 0.1372, 0.1708, 0.1336], + device='cuda:2'), in_proj_covar=tensor([0.0459, 0.0371, 0.0383, 0.0478, 0.0376, 0.0380, 0.0373, 0.0328], + 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') +2023-02-06 11:43:32,245 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.547e+02 3.046e+02 3.774e+02 7.420e+02, threshold=6.092e+02, percent-clipped=3.0 +2023-02-06 11:43:41,054 INFO [train.py:901] (2/4) Epoch 12, batch 2450, loss[loss=0.2557, simple_loss=0.3252, pruned_loss=0.09314, over 7805.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3077, pruned_loss=0.0788, over 1614845.72 frames. ], batch size: 20, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:44:15,046 INFO [train.py:901] (2/4) Epoch 12, batch 2500, loss[loss=0.1969, simple_loss=0.2846, pruned_loss=0.05463, over 7962.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3061, pruned_loss=0.07817, over 1610632.90 frames. ], batch size: 21, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:44:41,740 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.576e+02 3.186e+02 4.386e+02 8.083e+02, threshold=6.372e+02, percent-clipped=11.0 +2023-02-06 11:44:50,280 INFO [train.py:901] (2/4) Epoch 12, batch 2550, loss[loss=0.2098, simple_loss=0.2816, pruned_loss=0.06905, over 7916.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3078, pruned_loss=0.07905, over 1612919.40 frames. ], batch size: 20, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:45:03,017 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6637, 1.5410, 2.8649, 1.3691, 2.0778, 3.1069, 3.1618, 2.6030], + device='cuda:2'), covar=tensor([0.1108, 0.1447, 0.0361, 0.2018, 0.0865, 0.0271, 0.0459, 0.0745], + device='cuda:2'), in_proj_covar=tensor([0.0267, 0.0298, 0.0262, 0.0295, 0.0279, 0.0236, 0.0350, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:45:24,403 INFO [train.py:901] (2/4) Epoch 12, batch 2600, loss[loss=0.2053, simple_loss=0.2892, pruned_loss=0.06068, over 7922.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3085, pruned_loss=0.07942, over 1615172.20 frames. ], batch size: 20, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:45:33,951 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.25 vs. limit=5.0 +2023-02-06 11:45:50,013 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.880e+02 3.430e+02 4.544e+02 8.443e+02, threshold=6.860e+02, percent-clipped=9.0 +2023-02-06 11:45:56,362 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91560.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:45:58,890 INFO [train.py:901] (2/4) Epoch 12, batch 2650, loss[loss=0.2142, simple_loss=0.3038, pruned_loss=0.06229, over 7813.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3104, pruned_loss=0.08065, over 1615852.69 frames. ], batch size: 20, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:46:11,932 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91582.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:46:18,126 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2023-02-06 11:46:29,369 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:46:33,854 INFO [train.py:901] (2/4) Epoch 12, batch 2700, loss[loss=0.2118, simple_loss=0.2913, pruned_loss=0.06613, over 8523.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3091, pruned_loss=0.07948, over 1619292.86 frames. ], batch size: 28, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:46:55,396 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6553, 1.9295, 2.0350, 1.0413, 2.2106, 1.5050, 0.4991, 1.7157], + device='cuda:2'), covar=tensor([0.0403, 0.0226, 0.0167, 0.0399, 0.0251, 0.0593, 0.0591, 0.0211], + device='cuda:2'), in_proj_covar=tensor([0.0401, 0.0334, 0.0283, 0.0388, 0.0321, 0.0477, 0.0363, 0.0360], + device='cuda:2'), out_proj_covar=tensor([1.1393e-04, 9.2547e-05, 7.8851e-05, 1.0843e-04, 9.0469e-05, 1.4417e-04, + 1.0303e-04, 1.0155e-04], device='cuda:2') +2023-02-06 11:46:55,932 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:46:59,291 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.691e+02 3.205e+02 3.908e+02 7.628e+02, threshold=6.410e+02, percent-clipped=2.0 +2023-02-06 11:47:08,034 INFO [train.py:901] (2/4) Epoch 12, batch 2750, loss[loss=0.2415, simple_loss=0.3193, pruned_loss=0.0818, over 8135.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3083, pruned_loss=0.07834, over 1618658.85 frames. ], batch size: 22, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:47:43,505 INFO [train.py:901] (2/4) Epoch 12, batch 2800, loss[loss=0.2024, simple_loss=0.2877, pruned_loss=0.05861, over 7927.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3083, pruned_loss=0.07793, over 1615843.62 frames. ], batch size: 20, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:48:08,848 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.634e+02 3.181e+02 3.784e+02 9.192e+02, threshold=6.362e+02, percent-clipped=3.0 +2023-02-06 11:48:15,801 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91762.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:48:16,915 INFO [train.py:901] (2/4) Epoch 12, batch 2850, loss[loss=0.242, simple_loss=0.3244, pruned_loss=0.07981, over 8466.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3097, pruned_loss=0.07804, over 1618276.21 frames. ], batch size: 25, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:48:52,946 INFO [train.py:901] (2/4) Epoch 12, batch 2900, loss[loss=0.2009, simple_loss=0.2782, pruned_loss=0.06185, over 7810.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3071, pruned_loss=0.07709, over 1611617.14 frames. ], batch size: 20, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:49:14,741 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 11:49:18,830 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.538e+02 3.175e+02 3.875e+02 8.885e+02, threshold=6.349e+02, percent-clipped=4.0 +2023-02-06 11:49:22,151 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 11:49:26,753 INFO [train.py:901] (2/4) Epoch 12, batch 2950, loss[loss=0.2505, simple_loss=0.3301, pruned_loss=0.08547, over 8541.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3088, pruned_loss=0.07817, over 1611262.73 frames. ], batch size: 39, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:49:30,885 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4298, 4.3856, 3.9407, 1.8581, 3.9318, 4.0216, 3.9850, 3.7504], + device='cuda:2'), covar=tensor([0.0621, 0.0523, 0.0940, 0.4646, 0.0731, 0.0934, 0.1102, 0.0833], + device='cuda:2'), in_proj_covar=tensor([0.0461, 0.0374, 0.0385, 0.0483, 0.0376, 0.0381, 0.0375, 0.0328], + 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') +2023-02-06 11:49:40,299 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 11:49:54,045 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91904.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:49:57,393 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:50:00,613 INFO [train.py:901] (2/4) Epoch 12, batch 3000, loss[loss=0.1792, simple_loss=0.2536, pruned_loss=0.05244, over 7229.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3088, pruned_loss=0.07886, over 1611175.57 frames. ], batch size: 16, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:50:00,613 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 11:50:12,187 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4363, 1.7318, 2.6991, 1.2878, 2.0173, 1.8243, 1.5596, 1.9032], + device='cuda:2'), covar=tensor([0.1635, 0.2508, 0.0768, 0.4173, 0.1734, 0.2879, 0.1968, 0.1943], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0524, 0.0532, 0.0581, 0.0616, 0.0556, 0.0471, 0.0610], + 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') +2023-02-06 11:50:13,634 INFO [train.py:935] (2/4) Epoch 12, validation: loss=0.1868, simple_loss=0.2871, pruned_loss=0.04323, over 944034.00 frames. +2023-02-06 11:50:13,635 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 11:50:39,691 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-06 11:50:40,667 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.361e+02 2.883e+02 3.802e+02 7.578e+02, threshold=5.767e+02, percent-clipped=3.0 +2023-02-06 11:50:49,090 INFO [train.py:901] (2/4) Epoch 12, batch 3050, loss[loss=0.2445, simple_loss=0.3406, pruned_loss=0.07425, over 8281.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3095, pruned_loss=0.07913, over 1613052.17 frames. ], batch size: 23, lr: 6.41e-03, grad_scale: 8.0 +2023-02-06 11:50:56,130 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:51:14,061 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91999.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:51:25,121 INFO [train.py:901] (2/4) Epoch 12, batch 3100, loss[loss=0.231, simple_loss=0.3099, pruned_loss=0.07602, over 8378.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3103, pruned_loss=0.07975, over 1613214.33 frames. ], batch size: 49, lr: 6.41e-03, grad_scale: 8.0 +2023-02-06 11:51:28,117 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:51:28,799 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92019.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:51:45,787 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:51:51,733 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.663e+02 3.347e+02 4.142e+02 7.838e+02, threshold=6.695e+02, percent-clipped=5.0 +2023-02-06 11:52:01,126 INFO [train.py:901] (2/4) Epoch 12, batch 3150, loss[loss=0.2443, simple_loss=0.3129, pruned_loss=0.08787, over 8075.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3096, pruned_loss=0.0792, over 1610462.39 frames. ], batch size: 21, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:52:10,914 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4873, 1.5028, 4.7070, 1.7430, 4.1046, 3.8548, 4.2508, 4.0752], + device='cuda:2'), covar=tensor([0.0570, 0.4412, 0.0458, 0.3575, 0.1202, 0.0952, 0.0578, 0.0676], + device='cuda:2'), in_proj_covar=tensor([0.0498, 0.0576, 0.0590, 0.0537, 0.0617, 0.0525, 0.0520, 0.0584], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 11:52:35,760 INFO [train.py:901] (2/4) Epoch 12, batch 3200, loss[loss=0.2645, simple_loss=0.3364, pruned_loss=0.0963, over 8488.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3096, pruned_loss=0.07905, over 1613652.57 frames. ], batch size: 39, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:53:02,007 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.674e+02 3.226e+02 3.971e+02 7.397e+02, threshold=6.453e+02, percent-clipped=3.0 +2023-02-06 11:53:10,357 INFO [train.py:901] (2/4) Epoch 12, batch 3250, loss[loss=0.2256, simple_loss=0.3118, pruned_loss=0.06975, over 8510.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3094, pruned_loss=0.07907, over 1613000.75 frames. ], batch size: 28, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:53:46,149 INFO [train.py:901] (2/4) Epoch 12, batch 3300, loss[loss=0.1896, simple_loss=0.2738, pruned_loss=0.05266, over 7908.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3084, pruned_loss=0.07856, over 1612146.48 frames. ], batch size: 20, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:53:55,659 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4362, 1.8166, 1.8447, 1.0480, 1.8800, 1.3600, 0.4129, 1.5197], + device='cuda:2'), covar=tensor([0.0430, 0.0249, 0.0211, 0.0431, 0.0314, 0.0743, 0.0697, 0.0234], + device='cuda:2'), in_proj_covar=tensor([0.0395, 0.0328, 0.0282, 0.0389, 0.0317, 0.0476, 0.0358, 0.0356], + device='cuda:2'), out_proj_covar=tensor([1.1205e-04, 9.0791e-05, 7.8236e-05, 1.0855e-04, 8.9083e-05, 1.4390e-04, + 1.0156e-04, 1.0017e-04], device='cuda:2') +2023-02-06 11:54:11,039 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.357e+02 2.935e+02 3.680e+02 6.719e+02, threshold=5.870e+02, percent-clipped=1.0 +2023-02-06 11:54:11,766 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:54:19,107 INFO [train.py:901] (2/4) Epoch 12, batch 3350, loss[loss=0.2253, simple_loss=0.3046, pruned_loss=0.073, over 8337.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3078, pruned_loss=0.07808, over 1609408.88 frames. ], batch size: 26, lr: 6.40e-03, grad_scale: 16.0 +2023-02-06 11:54:27,348 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92275.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:54:36,860 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4267, 4.4514, 3.9558, 1.9575, 4.0174, 3.9911, 4.0446, 3.6895], + device='cuda:2'), covar=tensor([0.0868, 0.0597, 0.1167, 0.4885, 0.0772, 0.0872, 0.1391, 0.0866], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0379, 0.0388, 0.0489, 0.0382, 0.0382, 0.0378, 0.0333], + 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') +2023-02-06 11:54:45,176 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92300.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:54:55,335 INFO [train.py:901] (2/4) Epoch 12, batch 3400, loss[loss=0.2156, simple_loss=0.3065, pruned_loss=0.06231, over 8351.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3081, pruned_loss=0.07809, over 1612597.64 frames. ], batch size: 24, lr: 6.40e-03, grad_scale: 16.0 +2023-02-06 11:54:57,464 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92317.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:55:15,992 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92343.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:55:21,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.364e+02 2.893e+02 3.659e+02 6.777e+02, threshold=5.785e+02, percent-clipped=2.0 +2023-02-06 11:55:29,928 INFO [train.py:901] (2/4) Epoch 12, batch 3450, loss[loss=0.2258, simple_loss=0.3054, pruned_loss=0.07317, over 8440.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3084, pruned_loss=0.07838, over 1608769.25 frames. ], batch size: 27, lr: 6.40e-03, grad_scale: 16.0 +2023-02-06 11:55:32,861 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92368.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:55:46,609 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 +2023-02-06 11:56:04,898 INFO [train.py:901] (2/4) Epoch 12, batch 3500, loss[loss=0.2286, simple_loss=0.3046, pruned_loss=0.07629, over 7528.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.307, pruned_loss=0.07764, over 1608981.28 frames. ], batch size: 18, lr: 6.40e-03, grad_scale: 8.0 +2023-02-06 11:56:18,395 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92432.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:56:29,130 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 11:56:33,046 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.438e+02 2.928e+02 3.742e+02 8.211e+02, threshold=5.856e+02, percent-clipped=5.0 +2023-02-06 11:56:36,517 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92458.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:56:40,250 INFO [train.py:901] (2/4) Epoch 12, batch 3550, loss[loss=0.1958, simple_loss=0.2739, pruned_loss=0.05886, over 7718.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.307, pruned_loss=0.07775, over 1605250.53 frames. ], batch size: 18, lr: 6.40e-03, grad_scale: 8.0 +2023-02-06 11:56:48,396 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6250, 1.5783, 2.2207, 1.6804, 1.1199, 2.2238, 0.2609, 1.2846], + device='cuda:2'), covar=tensor([0.2834, 0.2322, 0.0549, 0.1917, 0.4623, 0.0542, 0.3549, 0.1920], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0175, 0.0104, 0.0219, 0.0256, 0.0108, 0.0165, 0.0168], + 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') +2023-02-06 11:57:14,696 INFO [train.py:901] (2/4) Epoch 12, batch 3600, loss[loss=0.2597, simple_loss=0.3309, pruned_loss=0.09423, over 7808.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3084, pruned_loss=0.07798, over 1615567.30 frames. ], batch size: 20, lr: 6.40e-03, grad_scale: 8.0 +2023-02-06 11:57:42,455 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.628e+02 3.055e+02 4.234e+02 9.851e+02, threshold=6.109e+02, percent-clipped=7.0 +2023-02-06 11:57:50,882 INFO [train.py:901] (2/4) Epoch 12, batch 3650, loss[loss=0.1972, simple_loss=0.277, pruned_loss=0.05873, over 7546.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3082, pruned_loss=0.0774, over 1613393.32 frames. ], batch size: 18, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:58:01,981 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 +2023-02-06 11:58:05,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9827, 1.4722, 3.2277, 1.4812, 2.2913, 3.5858, 3.6067, 3.0324], + device='cuda:2'), covar=tensor([0.1041, 0.1585, 0.0348, 0.2039, 0.0997, 0.0242, 0.0468, 0.0597], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0302, 0.0265, 0.0298, 0.0281, 0.0241, 0.0355, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 11:58:19,599 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.37 vs. limit=5.0 +2023-02-06 11:58:23,912 INFO [train.py:901] (2/4) Epoch 12, batch 3700, loss[loss=0.2453, simple_loss=0.333, pruned_loss=0.07878, over 8240.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3096, pruned_loss=0.07881, over 1614679.52 frames. ], batch size: 24, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:58:28,536 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 11:58:30,693 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92624.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:58:44,255 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92643.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:58:48,449 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:58:50,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.476e+02 3.116e+02 4.152e+02 8.400e+02, threshold=6.233e+02, percent-clipped=9.0 +2023-02-06 11:58:59,710 INFO [train.py:901] (2/4) Epoch 12, batch 3750, loss[loss=0.2725, simple_loss=0.3227, pruned_loss=0.1112, over 6723.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3094, pruned_loss=0.07905, over 1610772.98 frames. ], batch size: 71, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:59:17,078 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:59:20,025 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 11:59:28,421 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 11:59:32,388 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4187, 1.8186, 1.9216, 0.9541, 1.9983, 1.2907, 0.4569, 1.6123], + device='cuda:2'), covar=tensor([0.0470, 0.0232, 0.0194, 0.0455, 0.0266, 0.0750, 0.0637, 0.0212], + device='cuda:2'), in_proj_covar=tensor([0.0407, 0.0336, 0.0285, 0.0399, 0.0327, 0.0488, 0.0364, 0.0366], + device='cuda:2'), out_proj_covar=tensor([1.1541e-04, 9.3007e-05, 7.8854e-05, 1.1123e-04, 9.1677e-05, 1.4725e-04, + 1.0333e-04, 1.0285e-04], device='cuda:2') +2023-02-06 11:59:34,463 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:59:34,936 INFO [train.py:901] (2/4) Epoch 12, batch 3800, loss[loss=0.209, simple_loss=0.274, pruned_loss=0.07194, over 7700.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3084, pruned_loss=0.07813, over 1611163.97 frames. ], batch size: 18, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:59:35,190 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92714.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:59:43,939 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0624, 2.2238, 1.9052, 2.5113, 1.8363, 1.7958, 2.0002, 2.3805], + device='cuda:2'), covar=tensor([0.0662, 0.0752, 0.0906, 0.0436, 0.0913, 0.1123, 0.0796, 0.0565], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0212, 0.0258, 0.0217, 0.0217, 0.0252, 0.0258, 0.0218], + 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') +2023-02-06 11:59:52,026 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:59:52,779 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92739.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:00:02,135 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.558e+02 2.972e+02 3.756e+02 9.318e+02, threshold=5.944e+02, percent-clipped=5.0 +2023-02-06 12:00:03,335 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 12:00:09,491 INFO [train.py:901] (2/4) Epoch 12, batch 3850, loss[loss=0.2542, simple_loss=0.3239, pruned_loss=0.09225, over 8659.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3095, pruned_loss=0.0785, over 1616801.48 frames. ], batch size: 34, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 12:00:33,553 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 12:00:45,108 INFO [train.py:901] (2/4) Epoch 12, batch 3900, loss[loss=0.19, simple_loss=0.2625, pruned_loss=0.05877, over 7793.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3082, pruned_loss=0.07779, over 1617620.13 frames. ], batch size: 19, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 12:01:08,875 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92849.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:01:11,295 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.538e+02 2.989e+02 3.922e+02 7.912e+02, threshold=5.979e+02, percent-clipped=3.0 +2023-02-06 12:01:18,298 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3390, 1.5599, 2.2540, 1.1571, 1.5423, 1.6440, 1.3746, 1.4389], + device='cuda:2'), covar=tensor([0.1772, 0.2283, 0.0726, 0.3984, 0.1621, 0.2976, 0.1924, 0.1954], + device='cuda:2'), in_proj_covar=tensor([0.0492, 0.0529, 0.0539, 0.0583, 0.0622, 0.0564, 0.0477, 0.0615], + 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') +2023-02-06 12:01:19,448 INFO [train.py:901] (2/4) Epoch 12, batch 3950, loss[loss=0.234, simple_loss=0.3214, pruned_loss=0.07333, over 8286.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3079, pruned_loss=0.0774, over 1617176.12 frames. ], batch size: 23, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:01:54,545 INFO [train.py:901] (2/4) Epoch 12, batch 4000, loss[loss=0.1888, simple_loss=0.2721, pruned_loss=0.05272, over 7810.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3082, pruned_loss=0.07766, over 1616131.60 frames. ], batch size: 20, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:01:56,841 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:18,348 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92949.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:20,892 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.373e+02 3.059e+02 3.649e+02 8.513e+02, threshold=6.118e+02, percent-clipped=6.0 +2023-02-06 12:02:28,371 INFO [train.py:901] (2/4) Epoch 12, batch 4050, loss[loss=0.2758, simple_loss=0.35, pruned_loss=0.1008, over 8518.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3092, pruned_loss=0.07834, over 1619705.72 frames. ], batch size: 26, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:02:44,203 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:48,266 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:57,121 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1879, 1.0988, 1.2505, 1.1297, 0.9631, 1.3024, 0.0239, 0.8615], + device='cuda:2'), covar=tensor([0.2473, 0.1843, 0.0573, 0.1148, 0.3588, 0.0631, 0.2918, 0.1581], + device='cuda:2'), in_proj_covar=tensor([0.0170, 0.0174, 0.0103, 0.0216, 0.0253, 0.0107, 0.0162, 0.0168], + 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') +2023-02-06 12:03:03,722 INFO [train.py:901] (2/4) Epoch 12, batch 4100, loss[loss=0.2391, simple_loss=0.3204, pruned_loss=0.07888, over 8473.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3086, pruned_loss=0.07851, over 1616771.12 frames. ], batch size: 25, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:03:04,569 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8789, 1.5574, 1.8159, 1.4319, 1.0029, 1.5462, 2.3224, 2.1142], + device='cuda:2'), covar=tensor([0.0446, 0.1310, 0.1732, 0.1413, 0.0646, 0.1484, 0.0634, 0.0548], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0153, 0.0194, 0.0160, 0.0104, 0.0164, 0.0117, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 12:03:13,908 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:03:21,021 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 12:03:30,614 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.418e+02 3.048e+02 3.757e+02 7.047e+02, threshold=6.097e+02, percent-clipped=3.0 +2023-02-06 12:03:31,427 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:03:37,937 INFO [train.py:901] (2/4) Epoch 12, batch 4150, loss[loss=0.2681, simple_loss=0.3322, pruned_loss=0.102, over 8711.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3081, pruned_loss=0.07792, over 1620220.95 frames. ], batch size: 34, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:03:50,876 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:04:04,576 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93102.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:04:05,526 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 12:04:12,237 INFO [train.py:901] (2/4) Epoch 12, batch 4200, loss[loss=0.2126, simple_loss=0.291, pruned_loss=0.06712, over 7920.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.308, pruned_loss=0.07719, over 1621003.13 frames. ], batch size: 20, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:04:13,197 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6981, 2.1648, 3.4701, 2.4948, 3.0134, 2.3868, 2.0095, 1.6288], + device='cuda:2'), covar=tensor([0.4120, 0.4413, 0.1283, 0.3065, 0.2125, 0.2470, 0.1842, 0.4753], + device='cuda:2'), in_proj_covar=tensor([0.0885, 0.0864, 0.0720, 0.0845, 0.0927, 0.0796, 0.0699, 0.0758], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:04:24,931 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 12:04:26,409 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:04:29,265 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0160, 2.7758, 3.4804, 2.0101, 1.8112, 3.4035, 0.8244, 1.9317], + device='cuda:2'), covar=tensor([0.2284, 0.1592, 0.0469, 0.2452, 0.3812, 0.0401, 0.3399, 0.2189], + device='cuda:2'), in_proj_covar=tensor([0.0169, 0.0172, 0.0101, 0.0214, 0.0252, 0.0107, 0.0161, 0.0167], + 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') +2023-02-06 12:04:40,122 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.563e+02 2.943e+02 3.717e+02 8.503e+02, threshold=5.885e+02, percent-clipped=3.0 +2023-02-06 12:04:47,442 INFO [train.py:901] (2/4) Epoch 12, batch 4250, loss[loss=0.2197, simple_loss=0.288, pruned_loss=0.07566, over 7919.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3094, pruned_loss=0.07788, over 1623942.37 frames. ], batch size: 20, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:04:48,815 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 12:04:53,107 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2023-02-06 12:05:06,771 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:05:09,368 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93197.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:05:21,310 INFO [train.py:901] (2/4) Epoch 12, batch 4300, loss[loss=0.2569, simple_loss=0.3339, pruned_loss=0.08991, over 8759.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3096, pruned_loss=0.07883, over 1611086.18 frames. ], batch size: 30, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:05:22,138 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0792, 1.4016, 4.3320, 1.5285, 3.7984, 3.6247, 3.8777, 3.7208], + device='cuda:2'), covar=tensor([0.0572, 0.4114, 0.0478, 0.3290, 0.1122, 0.0809, 0.0577, 0.0633], + device='cuda:2'), in_proj_covar=tensor([0.0501, 0.0580, 0.0594, 0.0537, 0.0615, 0.0527, 0.0519, 0.0581], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:05:48,576 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.718e+02 3.236e+02 4.116e+02 1.260e+03, threshold=6.473e+02, percent-clipped=7.0 +2023-02-06 12:05:54,515 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93261.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:05:56,509 INFO [train.py:901] (2/4) Epoch 12, batch 4350, loss[loss=0.2421, simple_loss=0.3135, pruned_loss=0.08536, over 8079.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3091, pruned_loss=0.07849, over 1609775.57 frames. ], batch size: 21, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:06:15,790 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93293.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:06:16,408 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 12:06:25,941 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93308.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:06:29,909 INFO [train.py:901] (2/4) Epoch 12, batch 4400, loss[loss=0.2248, simple_loss=0.2892, pruned_loss=0.08019, over 7810.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3083, pruned_loss=0.07796, over 1611218.75 frames. ], batch size: 20, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:06:42,311 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7050, 2.1240, 3.3100, 1.5035, 2.4295, 2.1004, 1.7661, 2.2493], + device='cuda:2'), covar=tensor([0.1553, 0.1977, 0.0699, 0.3579, 0.1418, 0.2538, 0.1638, 0.1951], + device='cuda:2'), in_proj_covar=tensor([0.0488, 0.0525, 0.0535, 0.0579, 0.0618, 0.0555, 0.0473, 0.0611], + 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') +2023-02-06 12:06:46,155 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93337.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:06:58,347 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.551e+02 2.995e+02 3.715e+02 7.484e+02, threshold=5.990e+02, percent-clipped=1.0 +2023-02-06 12:06:58,372 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 12:07:01,790 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93358.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:05,646 INFO [train.py:901] (2/4) Epoch 12, batch 4450, loss[loss=0.24, simple_loss=0.3175, pruned_loss=0.08131, over 8494.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3074, pruned_loss=0.07748, over 1612977.88 frames. ], batch size: 26, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:07:11,752 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93372.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:12,587 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4897, 2.0102, 3.0260, 2.3914, 2.8126, 2.2579, 1.8944, 1.4207], + device='cuda:2'), covar=tensor([0.3683, 0.3943, 0.1155, 0.2424, 0.1633, 0.2068, 0.1532, 0.4078], + device='cuda:2'), in_proj_covar=tensor([0.0886, 0.0868, 0.0725, 0.0847, 0.0927, 0.0798, 0.0700, 0.0762], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:07:14,525 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:19,337 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:29,362 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:36,071 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93408.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:39,922 INFO [train.py:901] (2/4) Epoch 12, batch 4500, loss[loss=0.1941, simple_loss=0.2744, pruned_loss=0.0569, over 7644.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3071, pruned_loss=0.07764, over 1612418.07 frames. ], batch size: 19, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:07:50,713 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 12:08:06,002 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93452.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:06,470 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.576e+02 3.193e+02 4.187e+02 6.619e+02, threshold=6.386e+02, percent-clipped=4.0 +2023-02-06 12:08:06,708 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93453.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:13,859 INFO [train.py:901] (2/4) Epoch 12, batch 4550, loss[loss=0.2393, simple_loss=0.3161, pruned_loss=0.08127, over 8703.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3067, pruned_loss=0.07683, over 1619483.32 frames. ], batch size: 39, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:08:24,173 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:24,999 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:31,901 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:49,674 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:50,191 INFO [train.py:901] (2/4) Epoch 12, batch 4600, loss[loss=0.1823, simple_loss=0.2665, pruned_loss=0.04904, over 8133.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3063, pruned_loss=0.07649, over 1620399.22 frames. ], batch size: 22, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:09:16,605 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.455e+02 3.020e+02 4.052e+02 9.299e+02, threshold=6.041e+02, percent-clipped=5.0 +2023-02-06 12:09:24,858 INFO [train.py:901] (2/4) Epoch 12, batch 4650, loss[loss=0.2285, simple_loss=0.3132, pruned_loss=0.07196, over 8465.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3063, pruned_loss=0.0767, over 1614675.97 frames. ], batch size: 25, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:09:25,072 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:09:41,900 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93589.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:09:45,160 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:09:45,361 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 12:09:47,208 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5305, 1.9036, 1.9882, 1.1826, 2.0929, 1.3696, 0.5188, 1.7134], + device='cuda:2'), covar=tensor([0.0436, 0.0254, 0.0182, 0.0418, 0.0305, 0.0722, 0.0660, 0.0213], + device='cuda:2'), in_proj_covar=tensor([0.0398, 0.0332, 0.0277, 0.0394, 0.0322, 0.0481, 0.0361, 0.0363], + device='cuda:2'), out_proj_covar=tensor([1.1275e-04, 9.1637e-05, 7.6628e-05, 1.0989e-04, 9.0315e-05, 1.4519e-04, + 1.0213e-04, 1.0190e-04], device='cuda:2') +2023-02-06 12:09:59,871 INFO [train.py:901] (2/4) Epoch 12, batch 4700, loss[loss=0.1899, simple_loss=0.2723, pruned_loss=0.05373, over 7770.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.305, pruned_loss=0.07586, over 1609993.55 frames. ], batch size: 19, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:10:12,706 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:20,658 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9996, 6.0896, 5.2252, 2.5248, 5.4501, 5.6126, 5.5237, 5.2887], + device='cuda:2'), covar=tensor([0.0630, 0.0494, 0.1072, 0.4393, 0.0889, 0.0767, 0.1236, 0.0635], + device='cuda:2'), in_proj_covar=tensor([0.0472, 0.0389, 0.0395, 0.0493, 0.0387, 0.0390, 0.0384, 0.0336], + 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') +2023-02-06 12:10:25,038 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-06 12:10:26,588 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.370e+02 2.939e+02 3.568e+02 8.447e+02, threshold=5.879e+02, percent-clipped=4.0 +2023-02-06 12:10:29,424 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93657.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:33,899 INFO [train.py:901] (2/4) Epoch 12, batch 4750, loss[loss=0.2354, simple_loss=0.3139, pruned_loss=0.07842, over 8252.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3061, pruned_loss=0.0772, over 1607721.80 frames. ], batch size: 24, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:10:34,127 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:36,644 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4338, 1.6794, 4.5791, 2.0543, 2.4728, 5.1793, 5.1450, 4.5240], + device='cuda:2'), covar=tensor([0.0970, 0.1673, 0.0233, 0.1767, 0.1151, 0.0153, 0.0405, 0.0497], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0300, 0.0264, 0.0295, 0.0278, 0.0240, 0.0354, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:10:47,983 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 12:10:51,744 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93689.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:56,585 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6303, 2.0796, 2.1610, 1.2227, 2.3084, 1.3733, 0.7347, 1.8280], + device='cuda:2'), covar=tensor([0.0442, 0.0215, 0.0146, 0.0455, 0.0274, 0.0690, 0.0623, 0.0226], + device='cuda:2'), in_proj_covar=tensor([0.0398, 0.0330, 0.0277, 0.0394, 0.0324, 0.0480, 0.0359, 0.0360], + device='cuda:2'), out_proj_covar=tensor([1.1261e-04, 9.0855e-05, 7.6483e-05, 1.0976e-04, 9.0692e-05, 1.4513e-04, + 1.0152e-04, 1.0112e-04], device='cuda:2') +2023-02-06 12:11:00,460 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 12:11:02,495 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 12:11:04,761 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:09,472 INFO [train.py:901] (2/4) Epoch 12, batch 4800, loss[loss=0.226, simple_loss=0.3059, pruned_loss=0.073, over 8238.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3072, pruned_loss=0.07734, over 1613102.80 frames. ], batch size: 22, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:11:23,311 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:30,151 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93743.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:36,832 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.519e+02 2.967e+02 3.635e+02 7.460e+02, threshold=5.934e+02, percent-clipped=2.0 +2023-02-06 12:11:44,129 INFO [train.py:901] (2/4) Epoch 12, batch 4850, loss[loss=0.2572, simple_loss=0.3365, pruned_loss=0.08896, over 8638.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.308, pruned_loss=0.07807, over 1618037.05 frames. ], batch size: 34, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:11:47,087 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93768.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:47,773 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:53,086 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 12:11:53,201 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1926, 4.1094, 3.7404, 1.9440, 3.6530, 3.7695, 3.7408, 3.4549], + device='cuda:2'), covar=tensor([0.0764, 0.0604, 0.0975, 0.4808, 0.0941, 0.1019, 0.1349, 0.0867], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0383, 0.0391, 0.0491, 0.0385, 0.0388, 0.0385, 0.0335], + 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') +2023-02-06 12:12:00,548 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:12:04,713 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:12:18,628 INFO [train.py:901] (2/4) Epoch 12, batch 4900, loss[loss=0.2167, simple_loss=0.2907, pruned_loss=0.07129, over 7789.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3079, pruned_loss=0.07809, over 1616314.10 frames. ], batch size: 19, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:12:42,679 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:12:45,716 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.389e+02 2.920e+02 3.679e+02 7.315e+02, threshold=5.841e+02, percent-clipped=3.0 +2023-02-06 12:12:53,898 INFO [train.py:901] (2/4) Epoch 12, batch 4950, loss[loss=0.225, simple_loss=0.3059, pruned_loss=0.072, over 8101.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3067, pruned_loss=0.07761, over 1613462.44 frames. ], batch size: 23, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:13:00,013 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93873.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:13:20,756 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5453, 1.4125, 4.7349, 1.7871, 4.1955, 4.0290, 4.3327, 4.2647], + device='cuda:2'), covar=tensor([0.0510, 0.4108, 0.0418, 0.3199, 0.0981, 0.0821, 0.0484, 0.0484], + device='cuda:2'), in_proj_covar=tensor([0.0497, 0.0569, 0.0584, 0.0528, 0.0603, 0.0516, 0.0506, 0.0569], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:13:27,054 INFO [train.py:901] (2/4) Epoch 12, batch 5000, loss[loss=0.2461, simple_loss=0.3228, pruned_loss=0.08472, over 8465.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.307, pruned_loss=0.07795, over 1617586.90 frames. ], batch size: 27, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:13:55,391 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.574e+02 3.082e+02 3.748e+02 7.333e+02, threshold=6.165e+02, percent-clipped=4.0 +2023-02-06 12:13:56,535 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 12:14:02,955 INFO [train.py:901] (2/4) Epoch 12, batch 5050, loss[loss=0.2593, simple_loss=0.3231, pruned_loss=0.09776, over 8482.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3056, pruned_loss=0.07694, over 1615060.03 frames. ], batch size: 28, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:14:03,320 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.24 vs. limit=5.0 +2023-02-06 12:14:27,661 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:14:31,132 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 12:14:38,576 INFO [train.py:901] (2/4) Epoch 12, batch 5100, loss[loss=0.2214, simple_loss=0.29, pruned_loss=0.07646, over 7555.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3063, pruned_loss=0.07713, over 1613431.66 frames. ], batch size: 18, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:15:02,451 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 12:15:04,423 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 12:15:05,336 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.576e+02 2.962e+02 4.029e+02 5.912e+02, threshold=5.924e+02, percent-clipped=0.0 +2023-02-06 12:15:10,262 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0682, 3.9493, 3.6555, 2.0668, 3.6280, 3.6135, 3.7364, 3.2995], + device='cuda:2'), covar=tensor([0.0813, 0.0660, 0.0900, 0.4234, 0.0881, 0.1003, 0.1104, 0.1012], + device='cuda:2'), in_proj_covar=tensor([0.0463, 0.0379, 0.0389, 0.0486, 0.0382, 0.0386, 0.0380, 0.0334], + 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') +2023-02-06 12:15:13,501 INFO [train.py:901] (2/4) Epoch 12, batch 5150, loss[loss=0.2167, simple_loss=0.292, pruned_loss=0.07074, over 8236.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3054, pruned_loss=0.07687, over 1609043.19 frames. ], batch size: 22, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:15:47,476 INFO [train.py:901] (2/4) Epoch 12, batch 5200, loss[loss=0.2066, simple_loss=0.2825, pruned_loss=0.06534, over 7527.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.305, pruned_loss=0.07692, over 1606582.24 frames. ], batch size: 18, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:15:59,921 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:16:04,815 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9526, 1.1012, 3.0910, 1.0212, 2.6771, 2.6062, 2.7955, 2.7162], + device='cuda:2'), covar=tensor([0.0770, 0.3859, 0.0839, 0.3517, 0.1401, 0.0993, 0.0725, 0.0830], + device='cuda:2'), in_proj_covar=tensor([0.0496, 0.0573, 0.0584, 0.0529, 0.0608, 0.0519, 0.0512, 0.0570], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:16:14,659 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.571e+02 3.074e+02 4.467e+02 8.286e+02, threshold=6.149e+02, percent-clipped=7.0 +2023-02-06 12:16:21,913 INFO [train.py:901] (2/4) Epoch 12, batch 5250, loss[loss=0.2187, simple_loss=0.293, pruned_loss=0.07216, over 7931.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3059, pruned_loss=0.07711, over 1605964.77 frames. ], batch size: 20, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:16:25,905 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 12:16:57,676 INFO [train.py:901] (2/4) Epoch 12, batch 5300, loss[loss=0.2282, simple_loss=0.3039, pruned_loss=0.07621, over 8509.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3061, pruned_loss=0.07653, over 1610223.94 frames. ], batch size: 39, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:17:08,946 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7626, 4.6313, 4.1224, 1.9410, 4.1939, 4.2919, 4.2648, 3.8955], + device='cuda:2'), covar=tensor([0.0618, 0.0614, 0.1159, 0.4796, 0.0848, 0.0789, 0.1500, 0.0841], + device='cuda:2'), in_proj_covar=tensor([0.0465, 0.0380, 0.0390, 0.0483, 0.0379, 0.0385, 0.0379, 0.0333], + 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') +2023-02-06 12:17:15,483 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94241.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:17:19,508 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:17:23,401 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.628e+02 3.237e+02 4.138e+02 9.258e+02, threshold=6.473e+02, percent-clipped=5.0 +2023-02-06 12:17:31,599 INFO [train.py:901] (2/4) Epoch 12, batch 5350, loss[loss=0.2504, simple_loss=0.323, pruned_loss=0.08887, over 8637.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3063, pruned_loss=0.07707, over 1607852.74 frames. ], batch size: 34, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:18:05,213 INFO [train.py:901] (2/4) Epoch 12, batch 5400, loss[loss=0.2091, simple_loss=0.2877, pruned_loss=0.06521, over 7796.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3067, pruned_loss=0.07745, over 1609830.45 frames. ], batch size: 19, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:18:12,846 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9160, 2.1387, 3.3355, 1.7240, 2.7744, 2.2271, 2.0570, 2.6433], + device='cuda:2'), covar=tensor([0.1322, 0.1865, 0.0521, 0.3052, 0.1087, 0.2100, 0.1454, 0.1642], + device='cuda:2'), in_proj_covar=tensor([0.0491, 0.0524, 0.0535, 0.0586, 0.0615, 0.0558, 0.0478, 0.0616], + 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') +2023-02-06 12:18:25,500 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:18:32,249 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.485e+02 2.978e+02 4.110e+02 9.009e+02, threshold=5.957e+02, percent-clipped=6.0 +2023-02-06 12:18:39,975 INFO [train.py:901] (2/4) Epoch 12, batch 5450, loss[loss=0.1965, simple_loss=0.2828, pruned_loss=0.05512, over 8203.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.308, pruned_loss=0.0779, over 1608679.30 frames. ], batch size: 23, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:18:41,431 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:18:47,004 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 12:19:12,398 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 12:19:15,766 INFO [train.py:901] (2/4) Epoch 12, batch 5500, loss[loss=0.2626, simple_loss=0.3268, pruned_loss=0.09916, over 8339.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3079, pruned_loss=0.07794, over 1611289.68 frames. ], batch size: 26, lr: 6.33e-03, grad_scale: 16.0 +2023-02-06 12:19:23,448 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1111, 1.5554, 1.6666, 1.4552, 1.0376, 1.4472, 1.8729, 1.5099], + device='cuda:2'), covar=tensor([0.0431, 0.1087, 0.1455, 0.1184, 0.0539, 0.1288, 0.0520, 0.0562], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0152, 0.0192, 0.0157, 0.0103, 0.0162, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 12:19:43,163 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.661e+02 2.406e+02 2.798e+02 3.361e+02 6.650e+02, threshold=5.597e+02, percent-clipped=1.0 +2023-02-06 12:19:45,384 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:19:48,697 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9440, 1.8213, 2.3103, 1.7296, 1.3186, 2.4045, 0.4905, 1.4743], + device='cuda:2'), covar=tensor([0.2161, 0.1753, 0.0520, 0.2014, 0.4484, 0.0545, 0.3620, 0.2044], + device='cuda:2'), in_proj_covar=tensor([0.0167, 0.0170, 0.0101, 0.0216, 0.0255, 0.0105, 0.0164, 0.0165], + 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') +2023-02-06 12:19:49,114 INFO [train.py:901] (2/4) Epoch 12, batch 5550, loss[loss=0.2006, simple_loss=0.2822, pruned_loss=0.05951, over 7528.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3086, pruned_loss=0.07825, over 1609367.41 frames. ], batch size: 18, lr: 6.33e-03, grad_scale: 4.0 +2023-02-06 12:20:16,681 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94503.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:20:24,041 INFO [train.py:901] (2/4) Epoch 12, batch 5600, loss[loss=0.2537, simple_loss=0.3297, pruned_loss=0.0889, over 8579.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3082, pruned_loss=0.07803, over 1609963.56 frames. ], batch size: 31, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:20:35,258 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94528.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:20:54,482 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.675e+02 3.313e+02 4.214e+02 1.006e+03, threshold=6.626e+02, percent-clipped=7.0 +2023-02-06 12:21:00,660 INFO [train.py:901] (2/4) Epoch 12, batch 5650, loss[loss=0.2253, simple_loss=0.318, pruned_loss=0.06631, over 8461.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3061, pruned_loss=0.07658, over 1605517.94 frames. ], batch size: 25, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:21:15,256 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94585.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:21:21,314 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 12:21:35,757 INFO [train.py:901] (2/4) Epoch 12, batch 5700, loss[loss=0.2294, simple_loss=0.302, pruned_loss=0.0784, over 7915.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.306, pruned_loss=0.07631, over 1609627.49 frames. ], batch size: 20, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:21:46,150 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4004, 1.7797, 1.8511, 0.9590, 1.9880, 1.3048, 0.3688, 1.5767], + device='cuda:2'), covar=tensor([0.0395, 0.0240, 0.0192, 0.0409, 0.0250, 0.0708, 0.0603, 0.0204], + device='cuda:2'), in_proj_covar=tensor([0.0394, 0.0327, 0.0276, 0.0386, 0.0318, 0.0474, 0.0356, 0.0356], + device='cuda:2'), out_proj_covar=tensor([1.1137e-04, 8.9839e-05, 7.6204e-05, 1.0733e-04, 8.8825e-05, 1.4288e-04, + 1.0048e-04, 9.9711e-05], device='cuda:2') +2023-02-06 12:22:04,729 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.327e+02 3.036e+02 3.801e+02 7.493e+02, threshold=6.072e+02, percent-clipped=2.0 +2023-02-06 12:22:06,921 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7719, 5.9872, 5.1413, 2.2300, 5.1729, 5.5895, 5.4106, 5.1945], + device='cuda:2'), covar=tensor([0.0591, 0.0417, 0.0935, 0.5043, 0.0836, 0.0685, 0.1095, 0.0567], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0378, 0.0391, 0.0484, 0.0381, 0.0384, 0.0381, 0.0333], + 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') +2023-02-06 12:22:10,799 INFO [train.py:901] (2/4) Epoch 12, batch 5750, loss[loss=0.2162, simple_loss=0.3045, pruned_loss=0.06395, over 8337.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3065, pruned_loss=0.0766, over 1613053.09 frames. ], batch size: 26, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:22:26,193 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 12:22:35,728 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94700.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:22:42,476 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94710.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:22:45,146 INFO [train.py:901] (2/4) Epoch 12, batch 5800, loss[loss=0.2498, simple_loss=0.3265, pruned_loss=0.08659, over 8459.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3076, pruned_loss=0.07743, over 1612733.55 frames. ], batch size: 25, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:22:45,333 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94714.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:22:46,398 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4834, 4.4205, 4.0105, 1.7555, 3.9771, 4.0556, 4.0413, 3.6365], + device='cuda:2'), covar=tensor([0.0657, 0.0515, 0.0989, 0.4921, 0.0810, 0.0835, 0.1119, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0378, 0.0389, 0.0484, 0.0381, 0.0385, 0.0381, 0.0334], + 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') +2023-02-06 12:23:02,552 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94739.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:23:13,644 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.575e+02 3.288e+02 4.021e+02 7.847e+02, threshold=6.576e+02, percent-clipped=2.0 +2023-02-06 12:23:19,973 INFO [train.py:901] (2/4) Epoch 12, batch 5850, loss[loss=0.2522, simple_loss=0.3344, pruned_loss=0.08496, over 8354.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3077, pruned_loss=0.078, over 1615886.61 frames. ], batch size: 24, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:23:22,899 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9706, 1.9454, 2.3995, 1.7737, 1.2736, 2.4209, 0.5347, 1.4435], + device='cuda:2'), covar=tensor([0.2316, 0.1521, 0.0505, 0.2006, 0.4104, 0.0501, 0.2965, 0.1870], + device='cuda:2'), in_proj_covar=tensor([0.0168, 0.0171, 0.0101, 0.0215, 0.0254, 0.0106, 0.0164, 0.0165], + 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') +2023-02-06 12:23:54,270 INFO [train.py:901] (2/4) Epoch 12, batch 5900, loss[loss=0.2079, simple_loss=0.2898, pruned_loss=0.06307, over 8086.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3063, pruned_loss=0.07716, over 1616536.05 frames. ], batch size: 21, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:24:01,764 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:24:07,292 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 12:24:17,126 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4919, 2.1261, 3.2976, 1.2528, 2.2756, 1.7876, 1.8285, 2.0350], + device='cuda:2'), covar=tensor([0.2026, 0.2344, 0.0897, 0.4394, 0.1978, 0.3387, 0.1960, 0.2900], + device='cuda:2'), in_proj_covar=tensor([0.0490, 0.0524, 0.0534, 0.0582, 0.0615, 0.0555, 0.0477, 0.0616], + 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') +2023-02-06 12:24:22,275 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.604e+02 3.248e+02 4.213e+02 6.479e+02, threshold=6.496e+02, percent-clipped=0.0 +2023-02-06 12:24:26,530 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0077, 1.5246, 1.7006, 1.3946, 0.8739, 1.4944, 1.6337, 1.4691], + device='cuda:2'), covar=tensor([0.0463, 0.1136, 0.1637, 0.1344, 0.0596, 0.1414, 0.0643, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0151, 0.0191, 0.0157, 0.0103, 0.0161, 0.0114, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 12:24:28,377 INFO [train.py:901] (2/4) Epoch 12, batch 5950, loss[loss=0.2371, simple_loss=0.3153, pruned_loss=0.07951, over 8499.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3058, pruned_loss=0.07648, over 1617382.62 frames. ], batch size: 26, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:25:03,795 INFO [train.py:901] (2/4) Epoch 12, batch 6000, loss[loss=0.2276, simple_loss=0.2992, pruned_loss=0.07798, over 7694.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3061, pruned_loss=0.07712, over 1616280.19 frames. ], batch size: 18, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:25:03,796 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 12:25:16,950 INFO [train.py:935] (2/4) Epoch 12, validation: loss=0.1862, simple_loss=0.286, pruned_loss=0.04318, over 944034.00 frames. +2023-02-06 12:25:16,950 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 12:25:44,731 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.442e+02 2.970e+02 3.787e+02 9.017e+02, threshold=5.940e+02, percent-clipped=3.0 +2023-02-06 12:25:45,513 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94956.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:25:50,750 INFO [train.py:901] (2/4) Epoch 12, batch 6050, loss[loss=0.2113, simple_loss=0.298, pruned_loss=0.06227, over 8131.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3065, pruned_loss=0.07735, over 1614570.30 frames. ], batch size: 22, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:26:02,541 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94981.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:26:07,069 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8620, 3.7705, 3.5061, 1.7283, 3.4215, 3.4591, 3.4572, 3.0607], + device='cuda:2'), covar=tensor([0.0899, 0.0680, 0.1024, 0.5027, 0.0983, 0.1237, 0.1381, 0.1120], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0384, 0.0392, 0.0490, 0.0384, 0.0392, 0.0382, 0.0337], + 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') +2023-02-06 12:26:25,544 INFO [train.py:901] (2/4) Epoch 12, batch 6100, loss[loss=0.2265, simple_loss=0.3077, pruned_loss=0.07266, over 8477.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3065, pruned_loss=0.07693, over 1617485.18 frames. ], batch size: 27, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:26:54,014 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 12:26:54,670 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.590e+02 3.216e+02 4.301e+02 8.648e+02, threshold=6.433e+02, percent-clipped=2.0 +2023-02-06 12:27:00,769 INFO [train.py:901] (2/4) Epoch 12, batch 6150, loss[loss=0.2135, simple_loss=0.2952, pruned_loss=0.06591, over 8094.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3065, pruned_loss=0.07701, over 1613693.60 frames. ], batch size: 21, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:27:12,233 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95081.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:27:29,634 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95106.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:27:34,664 INFO [train.py:901] (2/4) Epoch 12, batch 6200, loss[loss=0.1998, simple_loss=0.2676, pruned_loss=0.066, over 7799.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3075, pruned_loss=0.07775, over 1614864.19 frames. ], batch size: 19, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:27:41,088 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95123.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:27:45,840 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6279, 4.6544, 4.1890, 2.0635, 4.1635, 4.3509, 4.2072, 3.8081], + device='cuda:2'), covar=tensor([0.0724, 0.0579, 0.1079, 0.4659, 0.0853, 0.0870, 0.1450, 0.0826], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0382, 0.0388, 0.0487, 0.0383, 0.0388, 0.0382, 0.0334], + 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') +2023-02-06 12:27:57,990 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6387, 2.3905, 4.5409, 2.0059, 2.6429, 5.2388, 5.1305, 4.5834], + device='cuda:2'), covar=tensor([0.0913, 0.1203, 0.0259, 0.1817, 0.0910, 0.0155, 0.0418, 0.0484], + device='cuda:2'), in_proj_covar=tensor([0.0266, 0.0297, 0.0260, 0.0289, 0.0274, 0.0236, 0.0347, 0.0288], + 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') +2023-02-06 12:28:04,337 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.519e+02 2.980e+02 3.798e+02 7.393e+02, threshold=5.961e+02, percent-clipped=2.0 +2023-02-06 12:28:10,299 INFO [train.py:901] (2/4) Epoch 12, batch 6250, loss[loss=0.2364, simple_loss=0.3026, pruned_loss=0.08508, over 7655.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3085, pruned_loss=0.07826, over 1620125.78 frames. ], batch size: 19, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:28:21,987 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-06 12:28:43,818 INFO [train.py:901] (2/4) Epoch 12, batch 6300, loss[loss=0.2284, simple_loss=0.3061, pruned_loss=0.07536, over 8641.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3086, pruned_loss=0.07852, over 1621359.34 frames. ], batch size: 49, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:29:13,408 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.657e+02 3.224e+02 4.358e+02 1.571e+03, threshold=6.448e+02, percent-clipped=5.0 +2023-02-06 12:29:20,987 INFO [train.py:901] (2/4) Epoch 12, batch 6350, loss[loss=0.2187, simple_loss=0.2962, pruned_loss=0.07056, over 8321.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.308, pruned_loss=0.07813, over 1618174.83 frames. ], batch size: 26, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:29:30,726 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:29:36,281 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:29:55,395 INFO [train.py:901] (2/4) Epoch 12, batch 6400, loss[loss=0.217, simple_loss=0.2821, pruned_loss=0.07595, over 7789.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3071, pruned_loss=0.07763, over 1619563.99 frames. ], batch size: 19, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:30:23,572 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.403e+02 2.937e+02 3.904e+02 6.682e+02, threshold=5.874e+02, percent-clipped=3.0 +2023-02-06 12:30:25,096 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95357.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:30:29,662 INFO [train.py:901] (2/4) Epoch 12, batch 6450, loss[loss=0.1939, simple_loss=0.2679, pruned_loss=0.05997, over 7540.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3066, pruned_loss=0.07711, over 1619710.46 frames. ], batch size: 18, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:31:05,016 INFO [train.py:901] (2/4) Epoch 12, batch 6500, loss[loss=0.234, simple_loss=0.325, pruned_loss=0.07148, over 8374.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3072, pruned_loss=0.07713, over 1621407.53 frames. ], batch size: 24, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:31:14,610 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5037, 1.8198, 3.0447, 1.3042, 2.0189, 1.9292, 1.6159, 1.9289], + device='cuda:2'), covar=tensor([0.1763, 0.2432, 0.0733, 0.4152, 0.1896, 0.2835, 0.1852, 0.2336], + device='cuda:2'), in_proj_covar=tensor([0.0489, 0.0527, 0.0533, 0.0580, 0.0618, 0.0555, 0.0473, 0.0611], + 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') +2023-02-06 12:31:31,888 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.431e+02 2.857e+02 3.846e+02 1.801e+03, threshold=5.713e+02, percent-clipped=8.0 +2023-02-06 12:31:35,003 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-06 12:31:37,942 INFO [train.py:901] (2/4) Epoch 12, batch 6550, loss[loss=0.1951, simple_loss=0.2844, pruned_loss=0.05283, over 8028.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.307, pruned_loss=0.07692, over 1620387.03 frames. ], batch size: 22, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:31:40,442 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.40 vs. limit=5.0 +2023-02-06 12:31:40,701 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95467.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:31:59,688 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0171, 1.7097, 6.1075, 2.3015, 5.5384, 5.0987, 5.7165, 5.5526], + device='cuda:2'), covar=tensor([0.0360, 0.4188, 0.0263, 0.2954, 0.0697, 0.0679, 0.0354, 0.0347], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0581, 0.0594, 0.0544, 0.0622, 0.0532, 0.0520, 0.0582], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:32:06,852 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 12:32:13,517 INFO [train.py:901] (2/4) Epoch 12, batch 6600, loss[loss=0.263, simple_loss=0.3352, pruned_loss=0.09542, over 8447.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3077, pruned_loss=0.07725, over 1620168.02 frames. ], batch size: 27, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:32:25,469 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 12:32:25,781 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 12:32:40,270 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.522e+02 3.078e+02 3.913e+02 8.021e+02, threshold=6.157e+02, percent-clipped=7.0 +2023-02-06 12:32:43,801 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5661, 1.6708, 2.1569, 1.3067, 1.0669, 2.2329, 0.2036, 1.2528], + device='cuda:2'), covar=tensor([0.2982, 0.1544, 0.0496, 0.2630, 0.4355, 0.0408, 0.3225, 0.1990], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0173, 0.0103, 0.0216, 0.0256, 0.0108, 0.0163, 0.0166], + 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') +2023-02-06 12:32:46,205 INFO [train.py:901] (2/4) Epoch 12, batch 6650, loss[loss=0.2144, simple_loss=0.3013, pruned_loss=0.06369, over 8100.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3092, pruned_loss=0.0784, over 1615807.52 frames. ], batch size: 23, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:32:59,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95582.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:33:15,570 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 +2023-02-06 12:33:21,228 INFO [train.py:901] (2/4) Epoch 12, batch 6700, loss[loss=0.2217, simple_loss=0.2952, pruned_loss=0.07406, over 7548.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3085, pruned_loss=0.07871, over 1609121.07 frames. ], batch size: 18, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:33:27,481 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:33:33,686 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:33:35,025 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3410, 1.4752, 4.3285, 1.6667, 2.2245, 5.0094, 5.0134, 4.3093], + device='cuda:2'), covar=tensor([0.1056, 0.1777, 0.0290, 0.2072, 0.1197, 0.0164, 0.0420, 0.0549], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0297, 0.0261, 0.0291, 0.0272, 0.0236, 0.0347, 0.0288], + 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') +2023-02-06 12:33:37,107 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:33:50,488 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 2.656e+02 3.142e+02 4.011e+02 7.522e+02, threshold=6.284e+02, percent-clipped=4.0 +2023-02-06 12:33:56,566 INFO [train.py:901] (2/4) Epoch 12, batch 6750, loss[loss=0.2511, simple_loss=0.3201, pruned_loss=0.09103, over 7804.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3083, pruned_loss=0.07854, over 1606303.62 frames. ], batch size: 20, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:34:22,170 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95701.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:34:30,620 INFO [train.py:901] (2/4) Epoch 12, batch 6800, loss[loss=0.2063, simple_loss=0.2755, pruned_loss=0.06856, over 7526.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3072, pruned_loss=0.078, over 1608578.54 frames. ], batch size: 18, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:34:40,704 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 12:34:47,120 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:34:53,258 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:35:00,263 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.510e+02 2.822e+02 3.564e+02 9.162e+02, threshold=5.644e+02, percent-clipped=3.0 +2023-02-06 12:35:06,292 INFO [train.py:901] (2/4) Epoch 12, batch 6850, loss[loss=0.2671, simple_loss=0.3298, pruned_loss=0.1022, over 6881.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3075, pruned_loss=0.07833, over 1607029.86 frames. ], batch size: 71, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:35:19,154 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1013, 2.7394, 3.2258, 1.2326, 3.1953, 1.7559, 1.4635, 2.0527], + device='cuda:2'), covar=tensor([0.0612, 0.0208, 0.0174, 0.0575, 0.0345, 0.0698, 0.0634, 0.0403], + device='cuda:2'), in_proj_covar=tensor([0.0394, 0.0329, 0.0282, 0.0393, 0.0325, 0.0483, 0.0360, 0.0360], + device='cuda:2'), out_proj_covar=tensor([1.1132e-04, 8.9943e-05, 7.7810e-05, 1.0918e-04, 9.0896e-05, 1.4543e-04, + 1.0159e-04, 1.0083e-04], device='cuda:2') +2023-02-06 12:35:25,953 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.27 vs. limit=5.0 +2023-02-06 12:35:26,864 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 12:35:40,354 INFO [train.py:901] (2/4) Epoch 12, batch 6900, loss[loss=0.221, simple_loss=0.3004, pruned_loss=0.07075, over 8142.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3095, pruned_loss=0.0792, over 1611346.38 frames. ], batch size: 22, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:35:41,922 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95816.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:35:43,309 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7751, 1.6473, 2.8207, 2.2284, 2.4326, 1.6104, 1.3365, 1.2535], + device='cuda:2'), covar=tensor([0.6114, 0.5365, 0.1301, 0.2755, 0.2316, 0.3620, 0.2925, 0.4514], + device='cuda:2'), in_proj_covar=tensor([0.0888, 0.0878, 0.0733, 0.0858, 0.0934, 0.0806, 0.0705, 0.0769], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:35:48,561 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5575, 2.4053, 4.3442, 1.4223, 3.1906, 2.2285, 1.8347, 2.9250], + device='cuda:2'), covar=tensor([0.1710, 0.2103, 0.0620, 0.3955, 0.1423, 0.2717, 0.1736, 0.2230], + device='cuda:2'), in_proj_covar=tensor([0.0487, 0.0525, 0.0530, 0.0579, 0.0615, 0.0554, 0.0475, 0.0609], + 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') +2023-02-06 12:35:51,216 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95830.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:35:57,286 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95838.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:36:08,021 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.693e+02 3.422e+02 4.342e+02 1.062e+03, threshold=6.843e+02, percent-clipped=12.0 +2023-02-06 12:36:14,243 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95863.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:36:14,701 INFO [train.py:901] (2/4) Epoch 12, batch 6950, loss[loss=0.2459, simple_loss=0.3145, pruned_loss=0.08862, over 8017.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.309, pruned_loss=0.0788, over 1613408.10 frames. ], batch size: 22, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:36:15,459 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:36:34,588 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 12:36:48,736 INFO [train.py:901] (2/4) Epoch 12, batch 7000, loss[loss=0.2099, simple_loss=0.2948, pruned_loss=0.06256, over 8127.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3088, pruned_loss=0.07799, over 1617973.45 frames. ], batch size: 22, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:37:17,431 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.501e+02 3.116e+02 3.850e+02 8.001e+02, threshold=6.232e+02, percent-clipped=2.0 +2023-02-06 12:37:19,590 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.40 vs. limit=5.0 +2023-02-06 12:37:23,289 INFO [train.py:901] (2/4) Epoch 12, batch 7050, loss[loss=0.2128, simple_loss=0.2861, pruned_loss=0.06976, over 7653.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3083, pruned_loss=0.07803, over 1611377.52 frames. ], batch size: 19, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:37:24,101 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1163, 1.3870, 4.2711, 1.5978, 3.7154, 3.5445, 3.8495, 3.7120], + device='cuda:2'), covar=tensor([0.0562, 0.4412, 0.0568, 0.3704, 0.1221, 0.0885, 0.0605, 0.0669], + device='cuda:2'), in_proj_covar=tensor([0.0504, 0.0580, 0.0592, 0.0543, 0.0618, 0.0530, 0.0521, 0.0582], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:37:34,013 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:44,077 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:45,562 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.82 vs. limit=5.0 +2023-02-06 12:37:47,417 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95998.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:50,531 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:58,502 INFO [train.py:901] (2/4) Epoch 12, batch 7100, loss[loss=0.1902, simple_loss=0.2839, pruned_loss=0.04831, over 8199.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.308, pruned_loss=0.07828, over 1606762.99 frames. ], batch size: 23, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:38:01,407 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:07,006 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:26,973 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.539e+02 3.029e+02 4.080e+02 8.783e+02, threshold=6.058e+02, percent-clipped=4.0 +2023-02-06 12:38:33,146 INFO [train.py:901] (2/4) Epoch 12, batch 7150, loss[loss=0.2649, simple_loss=0.3273, pruned_loss=0.1012, over 8321.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3065, pruned_loss=0.07725, over 1610460.34 frames. ], batch size: 49, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:38:38,700 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:54,775 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:56,840 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:58,153 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96098.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:39:08,660 INFO [train.py:901] (2/4) Epoch 12, batch 7200, loss[loss=0.2078, simple_loss=0.281, pruned_loss=0.06726, over 8091.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3072, pruned_loss=0.07739, over 1612185.76 frames. ], batch size: 21, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:39:36,231 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.446e+02 3.002e+02 3.633e+02 6.248e+02, threshold=6.005e+02, percent-clipped=1.0 +2023-02-06 12:39:42,858 INFO [train.py:901] (2/4) Epoch 12, batch 7250, loss[loss=0.2307, simple_loss=0.3067, pruned_loss=0.07734, over 7973.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.308, pruned_loss=0.07759, over 1613433.54 frames. ], batch size: 21, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:39:49,511 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96174.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:39:55,437 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:39:56,079 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8560, 3.8523, 3.5136, 1.7639, 3.3842, 3.3334, 3.5981, 3.0875], + device='cuda:2'), covar=tensor([0.0960, 0.0631, 0.1057, 0.4158, 0.1007, 0.1107, 0.1155, 0.1086], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0381, 0.0393, 0.0484, 0.0385, 0.0385, 0.0382, 0.0334], + 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') +2023-02-06 12:40:14,150 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:40:17,295 INFO [train.py:901] (2/4) Epoch 12, batch 7300, loss[loss=0.2732, simple_loss=0.3445, pruned_loss=0.1009, over 8334.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3083, pruned_loss=0.07786, over 1614429.09 frames. ], batch size: 26, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:40:44,971 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6510, 2.0834, 3.5420, 1.3459, 2.8382, 2.1694, 1.7237, 2.5893], + device='cuda:2'), covar=tensor([0.1613, 0.2105, 0.0608, 0.3907, 0.1301, 0.2621, 0.1762, 0.1856], + device='cuda:2'), in_proj_covar=tensor([0.0488, 0.0526, 0.0535, 0.0578, 0.0620, 0.0556, 0.0475, 0.0610], + 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') +2023-02-06 12:40:45,403 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.567e+02 3.297e+02 4.044e+02 1.170e+03, threshold=6.593e+02, percent-clipped=7.0 +2023-02-06 12:40:51,434 INFO [train.py:901] (2/4) Epoch 12, batch 7350, loss[loss=0.214, simple_loss=0.2865, pruned_loss=0.07072, over 7426.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3089, pruned_loss=0.07813, over 1613298.64 frames. ], batch size: 17, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:41:09,298 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96289.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:41:15,844 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 12:41:26,502 INFO [train.py:901] (2/4) Epoch 12, batch 7400, loss[loss=0.2766, simple_loss=0.34, pruned_loss=0.1066, over 7653.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3073, pruned_loss=0.07713, over 1612278.39 frames. ], batch size: 19, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:41:33,419 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:36,479 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 12:41:40,810 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7392, 4.6382, 4.1788, 2.7588, 4.1010, 4.1858, 4.4382, 3.6851], + device='cuda:2'), covar=tensor([0.0558, 0.0427, 0.0834, 0.3601, 0.0713, 0.0895, 0.0835, 0.0926], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0381, 0.0390, 0.0483, 0.0381, 0.0384, 0.0380, 0.0333], + 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') +2023-02-06 12:41:42,988 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7510, 1.1342, 3.9012, 1.3879, 3.3824, 3.2595, 3.4993, 3.4233], + device='cuda:2'), covar=tensor([0.0654, 0.5082, 0.0679, 0.3906, 0.1361, 0.1005, 0.0690, 0.0775], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0589, 0.0604, 0.0549, 0.0627, 0.0535, 0.0530, 0.0592], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 12:41:47,014 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:47,084 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:52,327 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96350.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:55,371 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.475e+02 3.182e+02 4.307e+02 9.281e+02, threshold=6.365e+02, percent-clipped=3.0 +2023-02-06 12:42:01,559 INFO [train.py:901] (2/4) Epoch 12, batch 7450, loss[loss=0.2253, simple_loss=0.308, pruned_loss=0.07131, over 8609.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3082, pruned_loss=0.07765, over 1614898.69 frames. ], batch size: 34, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:42:09,243 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96375.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:42:15,390 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 12:42:31,443 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96408.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:42:35,248 INFO [train.py:901] (2/4) Epoch 12, batch 7500, loss[loss=0.2547, simple_loss=0.3249, pruned_loss=0.09223, over 7132.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3086, pruned_loss=0.07799, over 1611408.43 frames. ], batch size: 71, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:42:54,778 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96442.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:43:02,270 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3904, 2.7265, 1.9179, 2.2424, 2.1241, 1.5139, 1.8999, 1.9640], + device='cuda:2'), covar=tensor([0.1341, 0.0328, 0.1028, 0.0557, 0.0662, 0.1373, 0.0896, 0.0930], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0240, 0.0323, 0.0301, 0.0306, 0.0325, 0.0343, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:43:03,968 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.684e+02 3.354e+02 4.069e+02 8.964e+02, threshold=6.707e+02, percent-clipped=7.0 +2023-02-06 12:43:05,483 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96457.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:43:09,785 INFO [train.py:901] (2/4) Epoch 12, batch 7550, loss[loss=0.2855, simple_loss=0.354, pruned_loss=0.1085, over 8286.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.308, pruned_loss=0.07778, over 1609316.75 frames. ], batch size: 23, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:43:15,330 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0569, 2.5299, 3.3832, 1.9855, 1.7813, 3.3113, 0.6790, 1.9837], + device='cuda:2'), covar=tensor([0.1739, 0.1543, 0.0382, 0.2331, 0.4153, 0.0653, 0.3505, 0.2052], + device='cuda:2'), in_proj_covar=tensor([0.0169, 0.0173, 0.0103, 0.0216, 0.0254, 0.0109, 0.0164, 0.0166], + 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') +2023-02-06 12:43:42,988 INFO [train.py:901] (2/4) Epoch 12, batch 7600, loss[loss=0.2516, simple_loss=0.3152, pruned_loss=0.09403, over 8399.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3091, pruned_loss=0.0782, over 1610852.70 frames. ], batch size: 49, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:43:52,501 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:05,571 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96545.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:44:11,798 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.706e+02 3.173e+02 4.121e+02 9.971e+02, threshold=6.345e+02, percent-clipped=8.0 +2023-02-06 12:44:13,334 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:18,541 INFO [train.py:901] (2/4) Epoch 12, batch 7650, loss[loss=0.2041, simple_loss=0.2806, pruned_loss=0.06382, over 8190.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3082, pruned_loss=0.07775, over 1613061.19 frames. ], batch size: 23, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:44:23,310 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96570.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:44:29,847 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:47,076 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:53,025 INFO [train.py:901] (2/4) Epoch 12, batch 7700, loss[loss=0.2379, simple_loss=0.3265, pruned_loss=0.07469, over 8584.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3075, pruned_loss=0.07743, over 1607431.97 frames. ], batch size: 34, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:45:02,659 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1752, 1.1113, 1.2358, 1.1060, 0.9319, 1.2823, 0.0770, 0.8486], + device='cuda:2'), covar=tensor([0.2116, 0.1524, 0.0649, 0.1230, 0.3667, 0.0608, 0.3007, 0.1694], + device='cuda:2'), in_proj_covar=tensor([0.0169, 0.0172, 0.0103, 0.0215, 0.0255, 0.0109, 0.0163, 0.0166], + 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') +2023-02-06 12:45:12,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:45:21,241 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.521e+02 3.004e+02 3.630e+02 7.905e+02, threshold=6.007e+02, percent-clipped=3.0 +2023-02-06 12:45:23,905 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 12:45:27,882 INFO [train.py:901] (2/4) Epoch 12, batch 7750, loss[loss=0.2534, simple_loss=0.3273, pruned_loss=0.08977, over 8294.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3055, pruned_loss=0.07596, over 1608399.58 frames. ], batch size: 23, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:45:42,857 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:02,099 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:02,580 INFO [train.py:901] (2/4) Epoch 12, batch 7800, loss[loss=0.2094, simple_loss=0.2797, pruned_loss=0.06955, over 8096.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3047, pruned_loss=0.07555, over 1611961.70 frames. ], batch size: 21, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:46:15,975 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 +2023-02-06 12:46:19,264 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:23,937 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6967, 1.7593, 4.2229, 1.8317, 2.3773, 4.9663, 4.9351, 4.0407], + device='cuda:2'), covar=tensor([0.1035, 0.1727, 0.0373, 0.2192, 0.1268, 0.0236, 0.0411, 0.0699], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0299, 0.0263, 0.0291, 0.0274, 0.0238, 0.0354, 0.0289], + 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') +2023-02-06 12:46:28,440 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96752.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:30,315 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.519e+02 3.193e+02 4.174e+02 8.059e+02, threshold=6.386e+02, percent-clipped=4.0 +2023-02-06 12:46:31,382 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 12:46:36,557 INFO [train.py:901] (2/4) Epoch 12, batch 7850, loss[loss=0.2292, simple_loss=0.2995, pruned_loss=0.0794, over 8224.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3052, pruned_loss=0.07612, over 1612766.37 frames. ], batch size: 22, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:46:56,860 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:01,901 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:10,009 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:10,492 INFO [train.py:901] (2/4) Epoch 12, batch 7900, loss[loss=0.2099, simple_loss=0.2856, pruned_loss=0.06714, over 8142.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3036, pruned_loss=0.07568, over 1602721.17 frames. ], batch size: 22, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:47:27,488 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96838.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:38,783 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.595e+02 3.080e+02 3.878e+02 8.124e+02, threshold=6.160e+02, percent-clipped=3.0 +2023-02-06 12:47:44,784 INFO [train.py:901] (2/4) Epoch 12, batch 7950, loss[loss=0.1988, simple_loss=0.2892, pruned_loss=0.05417, over 7662.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3027, pruned_loss=0.07556, over 1599012.98 frames. ], batch size: 19, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:47:47,002 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:48:07,428 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96898.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:48:17,695 INFO [train.py:901] (2/4) Epoch 12, batch 8000, loss[loss=0.2549, simple_loss=0.3408, pruned_loss=0.08451, over 8246.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3024, pruned_loss=0.07502, over 1602445.70 frames. ], batch size: 24, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:48:23,689 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96923.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:48:45,040 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.377e+02 3.280e+02 4.266e+02 7.100e+02, threshold=6.559e+02, percent-clipped=4.0 +2023-02-06 12:48:50,058 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8930, 1.7588, 2.8661, 1.2308, 2.2465, 3.0661, 3.1153, 2.6320], + device='cuda:2'), covar=tensor([0.0927, 0.1195, 0.0337, 0.2078, 0.0713, 0.0292, 0.0531, 0.0590], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0302, 0.0263, 0.0292, 0.0274, 0.0239, 0.0357, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:48:51,258 INFO [train.py:901] (2/4) Epoch 12, batch 8050, loss[loss=0.2217, simple_loss=0.2916, pruned_loss=0.07586, over 7532.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3028, pruned_loss=0.07548, over 1598822.76 frames. ], batch size: 18, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:49:24,720 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 12:49:29,795 INFO [train.py:901] (2/4) Epoch 13, batch 0, loss[loss=0.2219, simple_loss=0.3053, pruned_loss=0.0692, over 8078.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3053, pruned_loss=0.0692, over 8078.00 frames. ], batch size: 21, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:49:29,796 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 12:49:40,738 INFO [train.py:935] (2/4) Epoch 13, validation: loss=0.1867, simple_loss=0.2865, pruned_loss=0.04345, over 944034.00 frames. +2023-02-06 12:49:40,739 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 12:49:41,812 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 12:49:55,386 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 12:49:55,520 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:50:14,914 INFO [train.py:901] (2/4) Epoch 13, batch 50, loss[loss=0.1955, simple_loss=0.2833, pruned_loss=0.05388, over 7954.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3094, pruned_loss=0.07578, over 369955.89 frames. ], batch size: 21, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:50:20,337 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.833e+02 3.357e+02 4.758e+02 6.927e+02, threshold=6.715e+02, percent-clipped=2.0 +2023-02-06 12:50:21,933 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:50:29,185 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 12:50:41,098 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:50:50,958 INFO [train.py:901] (2/4) Epoch 13, batch 100, loss[loss=0.2149, simple_loss=0.3015, pruned_loss=0.06411, over 8475.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3088, pruned_loss=0.07501, over 650175.92 frames. ], batch size: 25, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:50:52,983 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 12:51:09,080 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97123.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:51:18,983 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97138.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:51:24,694 INFO [train.py:901] (2/4) Epoch 13, batch 150, loss[loss=0.2562, simple_loss=0.336, pruned_loss=0.08819, over 8186.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3097, pruned_loss=0.07684, over 866872.64 frames. ], batch size: 23, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:51:25,592 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97148.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:51:30,102 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.477e+02 2.848e+02 3.342e+02 7.997e+02, threshold=5.696e+02, percent-clipped=2.0 +2023-02-06 12:51:39,289 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 12:51:42,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9432, 1.5890, 3.5014, 1.5163, 2.3711, 3.8665, 3.8908, 3.3747], + device='cuda:2'), covar=tensor([0.1153, 0.1556, 0.0308, 0.1993, 0.0976, 0.0237, 0.0468, 0.0563], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0302, 0.0265, 0.0293, 0.0275, 0.0239, 0.0359, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:51:58,442 INFO [train.py:901] (2/4) Epoch 13, batch 200, loss[loss=0.2046, simple_loss=0.281, pruned_loss=0.06408, over 8079.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3086, pruned_loss=0.07619, over 1039274.49 frames. ], batch size: 21, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:52:09,220 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0126, 2.1885, 1.8589, 2.7297, 1.3658, 1.7329, 1.8757, 2.2848], + device='cuda:2'), covar=tensor([0.0649, 0.0891, 0.0916, 0.0363, 0.1148, 0.1295, 0.0966, 0.0745], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0211, 0.0251, 0.0214, 0.0213, 0.0251, 0.0254, 0.0218], + 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') +2023-02-06 12:52:18,287 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3167, 1.3481, 2.2666, 1.1065, 2.0557, 2.3693, 2.5555, 1.8960], + device='cuda:2'), covar=tensor([0.1176, 0.1370, 0.0571, 0.2235, 0.0846, 0.0534, 0.0706, 0.1079], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0301, 0.0263, 0.0292, 0.0274, 0.0238, 0.0356, 0.0291], + 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') +2023-02-06 12:52:23,162 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 12:52:33,409 INFO [train.py:901] (2/4) Epoch 13, batch 250, loss[loss=0.2288, simple_loss=0.3094, pruned_loss=0.0741, over 8507.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3082, pruned_loss=0.0768, over 1167011.71 frames. ], batch size: 26, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:52:37,624 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:52:38,766 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.455e+02 3.117e+02 3.819e+02 7.824e+02, threshold=6.233e+02, percent-clipped=7.0 +2023-02-06 12:52:46,027 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 12:52:53,685 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 12:52:54,018 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:52:54,530 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 12:53:06,364 INFO [train.py:901] (2/4) Epoch 13, batch 300, loss[loss=0.1947, simple_loss=0.2571, pruned_loss=0.06618, over 7529.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3087, pruned_loss=0.07792, over 1268010.69 frames. ], batch size: 18, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:53:06,751 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 12:53:34,632 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.51 vs. limit=5.0 +2023-02-06 12:53:41,548 INFO [train.py:901] (2/4) Epoch 13, batch 350, loss[loss=0.2509, simple_loss=0.324, pruned_loss=0.08894, over 8655.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3086, pruned_loss=0.07788, over 1347548.39 frames. ], batch size: 34, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:53:46,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.701e+02 2.508e+02 3.076e+02 3.709e+02 6.548e+02, threshold=6.153e+02, percent-clipped=1.0 +2023-02-06 12:53:50,474 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97360.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:53:51,766 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97362.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:53:53,813 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97365.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:53:55,087 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2479, 4.2281, 3.8247, 1.9622, 3.7816, 3.6788, 3.8310, 3.3730], + device='cuda:2'), covar=tensor([0.0895, 0.0654, 0.1116, 0.4843, 0.1033, 0.1082, 0.1398, 0.1174], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0385, 0.0392, 0.0488, 0.0387, 0.0387, 0.0381, 0.0338], + 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') +2023-02-06 12:54:11,898 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:54:15,163 INFO [train.py:901] (2/4) Epoch 13, batch 400, loss[loss=0.2506, simple_loss=0.3216, pruned_loss=0.08983, over 8547.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3067, pruned_loss=0.07706, over 1402026.82 frames. ], batch size: 31, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:54:39,465 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7364, 5.6911, 5.0878, 2.5054, 5.1271, 5.4801, 5.2699, 5.1943], + device='cuda:2'), covar=tensor([0.0488, 0.0376, 0.0819, 0.4514, 0.0639, 0.0848, 0.0876, 0.0696], + device='cuda:2'), in_proj_covar=tensor([0.0468, 0.0380, 0.0390, 0.0485, 0.0383, 0.0386, 0.0377, 0.0335], + 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') +2023-02-06 12:54:51,246 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2043, 1.6528, 1.4343, 1.7115, 1.3886, 1.2760, 1.3284, 1.4479], + device='cuda:2'), covar=tensor([0.0836, 0.0385, 0.1061, 0.0428, 0.0613, 0.1164, 0.0774, 0.0585], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0235, 0.0318, 0.0300, 0.0301, 0.0321, 0.0339, 0.0300], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:54:51,730 INFO [train.py:901] (2/4) Epoch 13, batch 450, loss[loss=0.2625, simple_loss=0.3204, pruned_loss=0.1023, over 7777.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3056, pruned_loss=0.07646, over 1445987.16 frames. ], batch size: 19, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:54:57,107 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.318e+02 2.836e+02 3.756e+02 7.381e+02, threshold=5.672e+02, percent-clipped=3.0 +2023-02-06 12:54:58,931 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 12:55:13,265 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:55:26,766 INFO [train.py:901] (2/4) Epoch 13, batch 500, loss[loss=0.2302, simple_loss=0.3002, pruned_loss=0.08011, over 8125.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3049, pruned_loss=0.07588, over 1482259.73 frames. ], batch size: 22, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:55:35,500 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97509.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:55:52,968 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97534.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:56:01,767 INFO [train.py:901] (2/4) Epoch 13, batch 550, loss[loss=0.2581, simple_loss=0.3296, pruned_loss=0.09331, over 8108.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3056, pruned_loss=0.07627, over 1514706.77 frames. ], batch size: 23, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:56:07,720 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.533e+02 3.037e+02 3.770e+02 9.997e+02, threshold=6.074e+02, percent-clipped=4.0 +2023-02-06 12:56:20,095 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:56:36,757 INFO [train.py:901] (2/4) Epoch 13, batch 600, loss[loss=0.3007, simple_loss=0.3638, pruned_loss=0.1188, over 6940.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3065, pruned_loss=0.07675, over 1539792.73 frames. ], batch size: 71, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:56:53,703 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:56:54,431 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2175, 3.1753, 2.9030, 1.4487, 2.8987, 2.9580, 2.9001, 2.7429], + device='cuda:2'), covar=tensor([0.1245, 0.0921, 0.1405, 0.4847, 0.1199, 0.1236, 0.1602, 0.1144], + device='cuda:2'), in_proj_covar=tensor([0.0470, 0.0385, 0.0391, 0.0488, 0.0385, 0.0389, 0.0382, 0.0339], + 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') +2023-02-06 12:56:55,693 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 12:56:55,856 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8668, 1.5649, 1.6552, 1.4545, 1.1467, 1.5415, 1.7373, 1.7317], + device='cuda:2'), covar=tensor([0.0498, 0.0904, 0.1312, 0.1084, 0.0596, 0.1115, 0.0644, 0.0421], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0151, 0.0191, 0.0157, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 12:56:59,736 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6389, 2.6006, 1.9479, 2.2554, 2.0384, 1.4088, 1.9212, 2.1304], + device='cuda:2'), covar=tensor([0.1407, 0.0364, 0.1117, 0.0618, 0.0764, 0.1520, 0.1008, 0.1005], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0238, 0.0324, 0.0304, 0.0306, 0.0327, 0.0345, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:57:09,686 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7992, 3.7749, 3.4358, 1.9248, 3.2746, 3.4841, 3.4905, 3.1413], + device='cuda:2'), covar=tensor([0.0988, 0.0716, 0.1224, 0.4671, 0.1126, 0.1131, 0.1230, 0.1084], + device='cuda:2'), in_proj_covar=tensor([0.0467, 0.0385, 0.0391, 0.0487, 0.0384, 0.0388, 0.0380, 0.0338], + 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') +2023-02-06 12:57:10,250 INFO [train.py:901] (2/4) Epoch 13, batch 650, loss[loss=0.1914, simple_loss=0.2858, pruned_loss=0.04847, over 8354.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3062, pruned_loss=0.07655, over 1557620.38 frames. ], batch size: 24, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:57:16,274 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.537e+02 2.925e+02 3.842e+02 7.324e+02, threshold=5.850e+02, percent-clipped=4.0 +2023-02-06 12:57:37,850 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3351, 2.1317, 1.7632, 1.9615, 1.7160, 1.3324, 1.5934, 1.7509], + device='cuda:2'), covar=tensor([0.1127, 0.0344, 0.0995, 0.0476, 0.0677, 0.1292, 0.0859, 0.0659], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0238, 0.0323, 0.0303, 0.0306, 0.0327, 0.0345, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:57:42,540 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97692.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:57:45,737 INFO [train.py:901] (2/4) Epoch 13, batch 700, loss[loss=0.2311, simple_loss=0.3143, pruned_loss=0.07395, over 8252.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3061, pruned_loss=0.07654, over 1571068.10 frames. ], batch size: 24, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:57:50,154 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.68 vs. limit=5.0 +2023-02-06 12:57:51,260 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:57:54,513 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:57:59,103 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 12:58:10,715 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:12,618 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:13,391 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:19,711 INFO [train.py:901] (2/4) Epoch 13, batch 750, loss[loss=0.2015, simple_loss=0.2661, pruned_loss=0.06845, over 7542.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3048, pruned_loss=0.0761, over 1579235.75 frames. ], batch size: 18, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:58:25,053 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.478e+02 2.997e+02 3.995e+02 8.399e+02, threshold=5.994e+02, percent-clipped=5.0 +2023-02-06 12:58:27,374 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97758.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:39,715 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 12:58:49,002 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 12:58:51,769 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4053, 2.2477, 3.2890, 2.0657, 2.7745, 3.6880, 3.7544, 2.9897], + device='cuda:2'), covar=tensor([0.1058, 0.1426, 0.0762, 0.1967, 0.1526, 0.0371, 0.0654, 0.0861], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0302, 0.0267, 0.0294, 0.0278, 0.0242, 0.0359, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 12:58:54,236 INFO [train.py:901] (2/4) Epoch 13, batch 800, loss[loss=0.2103, simple_loss=0.2779, pruned_loss=0.07134, over 7718.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3048, pruned_loss=0.07605, over 1585230.34 frames. ], batch size: 18, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:59:10,091 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:59:14,043 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:59:29,475 INFO [train.py:901] (2/4) Epoch 13, batch 850, loss[loss=0.2031, simple_loss=0.2893, pruned_loss=0.05847, over 8111.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3055, pruned_loss=0.07615, over 1591669.23 frames. ], batch size: 23, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:59:32,320 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:59:35,501 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.627e+02 3.254e+02 4.246e+02 9.834e+02, threshold=6.507e+02, percent-clipped=8.0 +2023-02-06 13:00:03,797 INFO [train.py:901] (2/4) Epoch 13, batch 900, loss[loss=0.2728, simple_loss=0.3541, pruned_loss=0.09574, over 8667.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3049, pruned_loss=0.07556, over 1593755.34 frames. ], batch size: 34, lr: 5.98e-03, grad_scale: 8.0 +2023-02-06 13:00:05,040 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 13:00:09,703 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97906.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:00:18,212 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:00:39,215 INFO [train.py:901] (2/4) Epoch 13, batch 950, loss[loss=0.2427, simple_loss=0.303, pruned_loss=0.09123, over 7791.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3046, pruned_loss=0.07531, over 1597839.40 frames. ], batch size: 19, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:00:45,306 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.593e+02 3.202e+02 4.020e+02 7.231e+02, threshold=6.403e+02, percent-clipped=2.0 +2023-02-06 13:00:55,372 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9915, 1.7935, 3.4231, 1.4564, 2.2976, 3.8166, 3.8901, 3.2734], + device='cuda:2'), covar=tensor([0.1022, 0.1341, 0.0333, 0.1985, 0.1019, 0.0221, 0.0447, 0.0568], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0300, 0.0266, 0.0292, 0.0276, 0.0241, 0.0358, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:01:08,765 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 13:01:11,125 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:01:13,712 INFO [train.py:901] (2/4) Epoch 13, batch 1000, loss[loss=0.2047, simple_loss=0.2875, pruned_loss=0.06091, over 8338.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3053, pruned_loss=0.0757, over 1596558.28 frames. ], batch size: 25, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:01:29,781 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:01:39,964 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:01:43,178 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98036.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:01:44,386 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 13:01:50,454 INFO [train.py:901] (2/4) Epoch 13, batch 1050, loss[loss=0.1939, simple_loss=0.2664, pruned_loss=0.06068, over 7687.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3052, pruned_loss=0.07568, over 1602182.15 frames. ], batch size: 18, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:01:56,533 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.707e+02 2.365e+02 2.893e+02 3.782e+02 5.594e+02, threshold=5.785e+02, percent-clipped=0.0 +2023-02-06 13:01:57,233 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 13:02:10,160 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98075.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:13,458 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:24,727 INFO [train.py:901] (2/4) Epoch 13, batch 1100, loss[loss=0.186, simple_loss=0.2635, pruned_loss=0.0543, over 7772.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3044, pruned_loss=0.0753, over 1605430.88 frames. ], batch size: 19, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:02:26,972 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98100.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:30,281 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98105.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:31,648 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:43,619 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98124.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:49,148 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:59,486 INFO [train.py:901] (2/4) Epoch 13, batch 1150, loss[loss=0.2001, simple_loss=0.2917, pruned_loss=0.0543, over 8590.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3036, pruned_loss=0.07428, over 1609482.47 frames. ], batch size: 31, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:03:02,368 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98151.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:03:05,420 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 13:03:06,082 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.603e+02 3.101e+02 3.825e+02 7.832e+02, threshold=6.203e+02, percent-clipped=6.0 +2023-02-06 13:03:14,956 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1204, 1.2755, 1.1852, 0.6791, 1.2497, 1.0130, 0.1084, 1.2533], + device='cuda:2'), covar=tensor([0.0276, 0.0241, 0.0223, 0.0357, 0.0257, 0.0620, 0.0494, 0.0200], + device='cuda:2'), in_proj_covar=tensor([0.0402, 0.0337, 0.0289, 0.0397, 0.0325, 0.0484, 0.0360, 0.0362], + device='cuda:2'), out_proj_covar=tensor([1.1293e-04, 9.2114e-05, 7.9625e-05, 1.0987e-04, 9.0458e-05, 1.4467e-04, + 1.0148e-04, 1.0106e-04], device='cuda:2') +2023-02-06 13:03:34,172 INFO [train.py:901] (2/4) Epoch 13, batch 1200, loss[loss=0.2248, simple_loss=0.3099, pruned_loss=0.06989, over 8686.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3047, pruned_loss=0.07488, over 1614449.76 frames. ], batch size: 34, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:03:46,029 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 13:04:06,223 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3935, 1.6306, 4.5479, 2.1326, 4.0443, 3.8926, 4.1529, 4.0138], + device='cuda:2'), covar=tensor([0.0505, 0.4036, 0.0476, 0.3235, 0.0945, 0.0781, 0.0479, 0.0551], + device='cuda:2'), in_proj_covar=tensor([0.0510, 0.0587, 0.0598, 0.0544, 0.0618, 0.0531, 0.0522, 0.0579], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:04:08,147 INFO [train.py:901] (2/4) Epoch 13, batch 1250, loss[loss=0.2179, simple_loss=0.2919, pruned_loss=0.07191, over 8311.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3046, pruned_loss=0.07489, over 1617121.56 frames. ], batch size: 25, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:04:10,208 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:04:14,076 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.568e+02 3.066e+02 4.053e+02 1.440e+03, threshold=6.132e+02, percent-clipped=8.0 +2023-02-06 13:04:36,958 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:04:43,429 INFO [train.py:901] (2/4) Epoch 13, batch 1300, loss[loss=0.2517, simple_loss=0.3102, pruned_loss=0.09663, over 7654.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3048, pruned_loss=0.07485, over 1619714.21 frames. ], batch size: 19, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:04:44,913 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4783, 1.9311, 2.9227, 2.3570, 2.5877, 2.2338, 1.8794, 1.2985], + device='cuda:2'), covar=tensor([0.3948, 0.4415, 0.1222, 0.2408, 0.1842, 0.2304, 0.1848, 0.4270], + device='cuda:2'), in_proj_covar=tensor([0.0893, 0.0883, 0.0741, 0.0861, 0.0939, 0.0810, 0.0706, 0.0771], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:04:54,299 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8161, 1.6936, 1.8415, 1.5613, 1.1584, 1.5925, 2.1894, 1.9729], + device='cuda:2'), covar=tensor([0.0424, 0.1181, 0.1629, 0.1371, 0.0593, 0.1503, 0.0645, 0.0614], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0153, 0.0193, 0.0158, 0.0102, 0.0164, 0.0116, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:04:54,329 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98313.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:05:12,874 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8737, 3.8630, 2.4459, 2.5415, 2.7701, 2.0317, 2.3868, 2.9276], + device='cuda:2'), covar=tensor([0.1537, 0.0301, 0.0945, 0.0832, 0.0629, 0.1239, 0.1175, 0.1045], + device='cuda:2'), in_proj_covar=tensor([0.0338, 0.0231, 0.0313, 0.0295, 0.0296, 0.0317, 0.0334, 0.0298], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:05:16,645 INFO [train.py:901] (2/4) Epoch 13, batch 1350, loss[loss=0.1825, simple_loss=0.2526, pruned_loss=0.05617, over 7693.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3057, pruned_loss=0.07566, over 1621870.92 frames. ], batch size: 18, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:05:23,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.492e+02 3.102e+02 3.697e+02 5.327e+02, threshold=6.205e+02, percent-clipped=0.0 +2023-02-06 13:05:29,572 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98365.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:05:52,228 INFO [train.py:901] (2/4) Epoch 13, batch 1400, loss[loss=0.2355, simple_loss=0.3235, pruned_loss=0.07376, over 8466.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3047, pruned_loss=0.07462, over 1622270.73 frames. ], batch size: 29, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:05:59,398 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98407.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:06:14,070 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:06:16,852 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98432.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:06:26,901 INFO [train.py:901] (2/4) Epoch 13, batch 1450, loss[loss=0.2333, simple_loss=0.3239, pruned_loss=0.07131, over 8241.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3041, pruned_loss=0.07469, over 1619080.83 frames. ], batch size: 24, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:06:32,939 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.293e+02 2.812e+02 3.491e+02 8.118e+02, threshold=5.625e+02, percent-clipped=1.0 +2023-02-06 13:06:34,304 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 13:06:41,232 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98468.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:06:59,729 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 +2023-02-06 13:07:02,157 INFO [train.py:901] (2/4) Epoch 13, batch 1500, loss[loss=0.2463, simple_loss=0.3348, pruned_loss=0.0789, over 8194.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3056, pruned_loss=0.07538, over 1621112.13 frames. ], batch size: 23, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:07:37,383 INFO [train.py:901] (2/4) Epoch 13, batch 1550, loss[loss=0.1808, simple_loss=0.2579, pruned_loss=0.05182, over 7689.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3056, pruned_loss=0.07548, over 1616702.34 frames. ], batch size: 18, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:07:43,378 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.583e+02 3.208e+02 4.119e+02 6.608e+02, threshold=6.417e+02, percent-clipped=3.0 +2023-02-06 13:07:46,952 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98561.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:02,057 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:05,867 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1413, 2.8286, 3.5014, 2.2075, 1.9658, 3.6364, 0.7757, 2.2225], + device='cuda:2'), covar=tensor([0.1963, 0.1394, 0.0420, 0.2252, 0.3681, 0.0399, 0.3240, 0.1738], + device='cuda:2'), in_proj_covar=tensor([0.0172, 0.0175, 0.0104, 0.0222, 0.0262, 0.0112, 0.0164, 0.0168], + 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') +2023-02-06 13:08:06,458 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1452, 4.1194, 3.7419, 1.7478, 3.5581, 3.6327, 3.7979, 3.3487], + device='cuda:2'), covar=tensor([0.0819, 0.0649, 0.1124, 0.4962, 0.1015, 0.1105, 0.1311, 0.0996], + device='cuda:2'), in_proj_covar=tensor([0.0470, 0.0387, 0.0396, 0.0489, 0.0387, 0.0392, 0.0382, 0.0341], + 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') +2023-02-06 13:08:11,761 INFO [train.py:901] (2/4) Epoch 13, batch 1600, loss[loss=0.2314, simple_loss=0.3123, pruned_loss=0.07525, over 8455.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.305, pruned_loss=0.07526, over 1617954.25 frames. ], batch size: 29, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:08:29,745 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:46,809 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:47,298 INFO [train.py:901] (2/4) Epoch 13, batch 1650, loss[loss=0.2509, simple_loss=0.3142, pruned_loss=0.09377, over 7796.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3036, pruned_loss=0.07478, over 1612220.02 frames. ], batch size: 20, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:08:53,412 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.462e+02 2.942e+02 3.707e+02 8.113e+02, threshold=5.885e+02, percent-clipped=6.0 +2023-02-06 13:09:20,791 INFO [train.py:901] (2/4) Epoch 13, batch 1700, loss[loss=0.2053, simple_loss=0.2989, pruned_loss=0.05583, over 8241.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.304, pruned_loss=0.07461, over 1608248.65 frames. ], batch size: 24, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:09:45,855 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9617, 2.1104, 1.7762, 2.7601, 1.2365, 1.6246, 1.8937, 2.1880], + device='cuda:2'), covar=tensor([0.0737, 0.0845, 0.1021, 0.0409, 0.1208, 0.1325, 0.0962, 0.0833], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0213, 0.0254, 0.0215, 0.0215, 0.0253, 0.0257, 0.0219], + 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') +2023-02-06 13:09:57,388 INFO [train.py:901] (2/4) Epoch 13, batch 1750, loss[loss=0.2162, simple_loss=0.2975, pruned_loss=0.06744, over 8026.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3052, pruned_loss=0.07511, over 1608386.09 frames. ], batch size: 22, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:10:03,432 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.663e+02 3.311e+02 3.905e+02 7.561e+02, threshold=6.622e+02, percent-clipped=6.0 +2023-02-06 13:10:15,124 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98772.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:10:23,960 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9851, 1.4648, 6.1291, 2.1536, 5.5436, 5.2403, 5.7533, 5.5799], + device='cuda:2'), covar=tensor([0.0388, 0.4287, 0.0337, 0.3219, 0.0831, 0.0776, 0.0350, 0.0416], + device='cuda:2'), in_proj_covar=tensor([0.0509, 0.0583, 0.0596, 0.0549, 0.0619, 0.0531, 0.0521, 0.0584], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:10:31,882 INFO [train.py:901] (2/4) Epoch 13, batch 1800, loss[loss=0.2444, simple_loss=0.3352, pruned_loss=0.07685, over 8460.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3038, pruned_loss=0.07386, over 1608435.54 frames. ], batch size: 25, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:11:01,378 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:06,993 INFO [train.py:901] (2/4) Epoch 13, batch 1850, loss[loss=0.2621, simple_loss=0.3395, pruned_loss=0.09237, over 8201.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.303, pruned_loss=0.07362, over 1608980.97 frames. ], batch size: 23, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:11:13,499 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.419e+02 2.914e+02 4.067e+02 1.078e+03, threshold=5.828e+02, percent-clipped=2.0 +2023-02-06 13:11:18,920 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:29,415 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8422, 2.7815, 2.0857, 2.2376, 2.2013, 1.7204, 2.2063, 2.3290], + device='cuda:2'), covar=tensor([0.1243, 0.0330, 0.0870, 0.0624, 0.0612, 0.1250, 0.0820, 0.0853], + device='cuda:2'), in_proj_covar=tensor([0.0343, 0.0235, 0.0316, 0.0299, 0.0300, 0.0323, 0.0339, 0.0301], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:11:34,563 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98887.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:41,190 INFO [train.py:901] (2/4) Epoch 13, batch 1900, loss[loss=0.1871, simple_loss=0.2745, pruned_loss=0.04982, over 7973.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3033, pruned_loss=0.07392, over 1612330.30 frames. ], batch size: 21, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:11:46,607 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:48,678 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98908.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:51,826 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98913.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:12:11,193 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 13:12:15,141 INFO [train.py:901] (2/4) Epoch 13, batch 1950, loss[loss=0.1968, simple_loss=0.2729, pruned_loss=0.06034, over 7799.00 frames. ], tot_loss[loss=0.225, simple_loss=0.303, pruned_loss=0.07348, over 1613256.93 frames. ], batch size: 20, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:12:21,304 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.445e+02 3.079e+02 3.874e+02 6.986e+02, threshold=6.158e+02, percent-clipped=4.0 +2023-02-06 13:12:23,960 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 13:12:35,737 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7287, 5.8084, 5.0572, 2.5773, 4.9522, 5.5218, 5.3064, 5.3060], + device='cuda:2'), covar=tensor([0.0523, 0.0464, 0.0982, 0.4142, 0.0783, 0.0735, 0.1198, 0.0578], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0390, 0.0396, 0.0492, 0.0388, 0.0392, 0.0384, 0.0341], + 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') +2023-02-06 13:12:44,582 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 13:12:45,369 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98989.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:12:50,408 INFO [train.py:901] (2/4) Epoch 13, batch 2000, loss[loss=0.2267, simple_loss=0.2971, pruned_loss=0.07814, over 7548.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3032, pruned_loss=0.07375, over 1615603.07 frames. ], batch size: 18, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:13:06,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:13:24,249 INFO [train.py:901] (2/4) Epoch 13, batch 2050, loss[loss=0.2728, simple_loss=0.3378, pruned_loss=0.1039, over 8508.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3039, pruned_loss=0.07428, over 1615421.45 frames. ], batch size: 49, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:13:30,110 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.527e+02 3.267e+02 4.166e+02 9.227e+02, threshold=6.535e+02, percent-clipped=8.0 +2023-02-06 13:13:39,428 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99069.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:13:43,692 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 13:13:51,524 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1535, 1.3063, 3.3015, 1.1098, 2.8895, 2.7788, 3.0028, 2.9124], + device='cuda:2'), covar=tensor([0.0842, 0.3881, 0.0885, 0.3625, 0.1584, 0.1198, 0.0756, 0.0922], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0581, 0.0594, 0.0550, 0.0624, 0.0534, 0.0524, 0.0587], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:13:52,406 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 13:13:58,777 INFO [train.py:901] (2/4) Epoch 13, batch 2100, loss[loss=0.246, simple_loss=0.3289, pruned_loss=0.0816, over 8488.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3038, pruned_loss=0.07481, over 1615867.29 frames. ], batch size: 29, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:14:27,699 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9946, 1.6493, 6.0889, 2.2763, 5.5008, 5.0911, 5.6696, 5.5545], + device='cuda:2'), covar=tensor([0.0496, 0.4429, 0.0314, 0.3169, 0.0968, 0.0848, 0.0422, 0.0517], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0583, 0.0596, 0.0550, 0.0624, 0.0534, 0.0523, 0.0587], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:14:31,277 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99143.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:14:33,890 INFO [train.py:901] (2/4) Epoch 13, batch 2150, loss[loss=0.2402, simple_loss=0.3232, pruned_loss=0.07865, over 8249.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3037, pruned_loss=0.07463, over 1615398.40 frames. ], batch size: 24, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:14:39,911 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.393e+02 2.767e+02 3.323e+02 5.467e+02, threshold=5.533e+02, percent-clipped=0.0 +2023-02-06 13:14:48,048 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99168.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:15:07,838 INFO [train.py:901] (2/4) Epoch 13, batch 2200, loss[loss=0.2307, simple_loss=0.315, pruned_loss=0.07319, over 8544.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3042, pruned_loss=0.07529, over 1614718.94 frames. ], batch size: 31, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:15:27,755 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0915, 1.8116, 3.3405, 1.5093, 2.3467, 3.7156, 3.7118, 3.1726], + device='cuda:2'), covar=tensor([0.0987, 0.1277, 0.0367, 0.2056, 0.1002, 0.0230, 0.0531, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0273, 0.0303, 0.0270, 0.0295, 0.0280, 0.0241, 0.0361, 0.0298], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:15:43,035 INFO [train.py:901] (2/4) Epoch 13, batch 2250, loss[loss=0.2559, simple_loss=0.3162, pruned_loss=0.09778, over 7185.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3053, pruned_loss=0.07589, over 1614566.69 frames. ], batch size: 16, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:15:43,152 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:15:46,454 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:15:48,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.623e+02 3.377e+02 4.135e+02 6.545e+02, threshold=6.753e+02, percent-clipped=6.0 +2023-02-06 13:15:49,683 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99257.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:02,576 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:09,209 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:16,492 INFO [train.py:901] (2/4) Epoch 13, batch 2300, loss[loss=0.194, simple_loss=0.288, pruned_loss=0.05, over 8128.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3049, pruned_loss=0.07512, over 1618130.95 frames. ], batch size: 22, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:16:19,366 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:43,130 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99333.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:49,184 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:52,409 INFO [train.py:901] (2/4) Epoch 13, batch 2350, loss[loss=0.1916, simple_loss=0.2696, pruned_loss=0.05678, over 7441.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3045, pruned_loss=0.0752, over 1618776.52 frames. ], batch size: 17, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:16:58,394 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.470e+02 3.074e+02 3.865e+02 1.080e+03, threshold=6.149e+02, percent-clipped=3.0 +2023-02-06 13:17:06,764 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99367.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:17:10,178 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99372.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:17:12,142 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5482, 1.5979, 4.7519, 1.7543, 4.2045, 3.9957, 4.3062, 4.1782], + device='cuda:2'), covar=tensor([0.0467, 0.4095, 0.0408, 0.3635, 0.1022, 0.0838, 0.0459, 0.0533], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0596, 0.0613, 0.0564, 0.0637, 0.0546, 0.0538, 0.0601], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:17:22,906 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4843, 1.8012, 1.8707, 1.1257, 2.0319, 1.4319, 0.3781, 1.7938], + device='cuda:2'), covar=tensor([0.0362, 0.0236, 0.0195, 0.0369, 0.0240, 0.0657, 0.0559, 0.0169], + device='cuda:2'), in_proj_covar=tensor([0.0401, 0.0336, 0.0292, 0.0402, 0.0325, 0.0487, 0.0361, 0.0367], + device='cuda:2'), out_proj_covar=tensor([1.1254e-04, 9.1758e-05, 8.0184e-05, 1.1101e-04, 9.0365e-05, 1.4522e-04, + 1.0168e-04, 1.0207e-04], device='cuda:2') +2023-02-06 13:17:26,591 INFO [train.py:901] (2/4) Epoch 13, batch 2400, loss[loss=0.2212, simple_loss=0.3101, pruned_loss=0.06613, over 8031.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3037, pruned_loss=0.07463, over 1619631.30 frames. ], batch size: 22, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:17:37,519 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:17:54,054 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4340, 1.9876, 2.8745, 2.2859, 2.7148, 2.3021, 1.9627, 1.3865], + device='cuda:2'), covar=tensor([0.4350, 0.4358, 0.1447, 0.2982, 0.2161, 0.2354, 0.1698, 0.4720], + device='cuda:2'), in_proj_covar=tensor([0.0896, 0.0887, 0.0740, 0.0862, 0.0942, 0.0813, 0.0705, 0.0775], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:17:55,999 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:18:01,834 INFO [train.py:901] (2/4) Epoch 13, batch 2450, loss[loss=0.2115, simple_loss=0.2878, pruned_loss=0.06765, over 7540.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3039, pruned_loss=0.07459, over 1619634.18 frames. ], batch size: 18, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:18:02,725 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99448.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:18:08,608 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.541e+02 3.157e+02 3.793e+02 6.756e+02, threshold=6.314e+02, percent-clipped=3.0 +2023-02-06 13:18:29,566 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99486.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:18:36,513 INFO [train.py:901] (2/4) Epoch 13, batch 2500, loss[loss=0.219, simple_loss=0.2959, pruned_loss=0.07105, over 8362.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.304, pruned_loss=0.07455, over 1616600.50 frames. ], batch size: 49, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:18:45,253 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7656, 1.7552, 2.5019, 1.7675, 1.2564, 2.4169, 0.4021, 1.4164], + device='cuda:2'), covar=tensor([0.2708, 0.1877, 0.0411, 0.2193, 0.4284, 0.0572, 0.3372, 0.2109], + device='cuda:2'), in_proj_covar=tensor([0.0172, 0.0176, 0.0104, 0.0220, 0.0257, 0.0111, 0.0165, 0.0167], + 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') +2023-02-06 13:18:57,698 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99528.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:19:10,746 INFO [train.py:901] (2/4) Epoch 13, batch 2550, loss[loss=0.231, simple_loss=0.315, pruned_loss=0.07348, over 8258.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3052, pruned_loss=0.07547, over 1615268.96 frames. ], batch size: 24, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:19:17,198 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.420e+02 2.977e+02 3.875e+02 7.325e+02, threshold=5.954e+02, percent-clipped=4.0 +2023-02-06 13:19:19,408 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2739, 1.9505, 2.9236, 2.2485, 2.6890, 2.0848, 1.7439, 1.4583], + device='cuda:2'), covar=tensor([0.4550, 0.4568, 0.1373, 0.3410, 0.2228, 0.2557, 0.1848, 0.4748], + device='cuda:2'), in_proj_covar=tensor([0.0893, 0.0887, 0.0738, 0.0862, 0.0943, 0.0812, 0.0703, 0.0774], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:19:20,015 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6148, 1.7120, 2.3142, 1.4758, 1.0917, 2.2388, 0.2779, 1.2545], + device='cuda:2'), covar=tensor([0.2554, 0.1804, 0.0421, 0.2366, 0.4379, 0.0462, 0.3391, 0.2124], + device='cuda:2'), in_proj_covar=tensor([0.0172, 0.0176, 0.0103, 0.0219, 0.0257, 0.0110, 0.0165, 0.0167], + 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') +2023-02-06 13:19:21,125 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3746, 2.1852, 1.6454, 1.9309, 1.9019, 1.3526, 1.6034, 1.7337], + device='cuda:2'), covar=tensor([0.1116, 0.0342, 0.0991, 0.0487, 0.0581, 0.1280, 0.0887, 0.0764], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0240, 0.0320, 0.0300, 0.0301, 0.0326, 0.0344, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:19:37,747 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99586.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:19:41,075 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99591.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:19:45,636 INFO [train.py:901] (2/4) Epoch 13, batch 2600, loss[loss=0.2102, simple_loss=0.3004, pruned_loss=0.05999, over 8030.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3054, pruned_loss=0.07569, over 1615545.15 frames. ], batch size: 22, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:19:52,601 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:03,510 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:06,925 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99628.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:08,161 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:19,553 INFO [train.py:901] (2/4) Epoch 13, batch 2650, loss[loss=0.2577, simple_loss=0.3302, pruned_loss=0.09257, over 8448.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3063, pruned_loss=0.07604, over 1620844.57 frames. ], batch size: 27, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:20:20,456 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:23,637 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99653.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:25,436 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.509e+02 2.403e+02 3.099e+02 4.031e+02 8.160e+02, threshold=6.198e+02, percent-clipped=1.0 +2023-02-06 13:20:47,384 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:54,457 INFO [train.py:901] (2/4) Epoch 13, batch 2700, loss[loss=0.191, simple_loss=0.2753, pruned_loss=0.05333, over 7819.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3058, pruned_loss=0.07552, over 1624585.69 frames. ], batch size: 20, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:20:55,213 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:59,158 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0549, 1.3906, 1.6435, 1.3623, 0.9553, 1.4303, 1.7039, 1.6517], + device='cuda:2'), covar=tensor([0.0465, 0.1343, 0.1748, 0.1378, 0.0610, 0.1571, 0.0667, 0.0661], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0192, 0.0158, 0.0102, 0.0163, 0.0115, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:20:59,205 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:00,476 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99706.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:12,260 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 13:21:16,801 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99729.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:27,687 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:28,872 INFO [train.py:901] (2/4) Epoch 13, batch 2750, loss[loss=0.1914, simple_loss=0.2699, pruned_loss=0.05647, over 7544.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3052, pruned_loss=0.07528, over 1617529.13 frames. ], batch size: 18, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:21:34,775 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.508e+02 3.194e+02 3.866e+02 8.318e+02, threshold=6.387e+02, percent-clipped=3.0 +2023-02-06 13:21:52,456 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99782.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:53,985 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:22:03,192 INFO [train.py:901] (2/4) Epoch 13, batch 2800, loss[loss=0.228, simple_loss=0.3067, pruned_loss=0.07467, over 8352.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3064, pruned_loss=0.07658, over 1617220.99 frames. ], batch size: 26, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:22:06,163 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:22:11,445 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99809.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:22:22,820 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0474, 2.7076, 3.6929, 1.9869, 1.8327, 3.5274, 0.5915, 2.1280], + device='cuda:2'), covar=tensor([0.1427, 0.1455, 0.0219, 0.2675, 0.3630, 0.0429, 0.3510, 0.1515], + device='cuda:2'), in_proj_covar=tensor([0.0174, 0.0179, 0.0105, 0.0223, 0.0260, 0.0113, 0.0166, 0.0169], + 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') +2023-02-06 13:22:26,105 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99830.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:22:37,608 INFO [train.py:901] (2/4) Epoch 13, batch 2850, loss[loss=0.2465, simple_loss=0.3187, pruned_loss=0.08715, over 8357.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3055, pruned_loss=0.07603, over 1615158.73 frames. ], batch size: 24, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:22:43,878 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.486e+02 2.909e+02 3.673e+02 9.445e+02, threshold=5.818e+02, percent-clipped=3.0 +2023-02-06 13:22:56,571 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4538, 2.0033, 3.2850, 1.2994, 2.4960, 1.9107, 1.6042, 2.3098], + device='cuda:2'), covar=tensor([0.1799, 0.2132, 0.0717, 0.4140, 0.1617, 0.2966, 0.1978, 0.2201], + device='cuda:2'), in_proj_covar=tensor([0.0490, 0.0530, 0.0538, 0.0589, 0.0624, 0.0559, 0.0482, 0.0613], + 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') +2023-02-06 13:23:02,725 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 +2023-02-06 13:23:11,489 INFO [train.py:901] (2/4) Epoch 13, batch 2900, loss[loss=0.1692, simple_loss=0.2555, pruned_loss=0.04145, over 7428.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3072, pruned_loss=0.07723, over 1615931.73 frames. ], batch size: 17, lr: 5.92e-03, grad_scale: 16.0 +2023-02-06 13:23:11,689 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:23:34,892 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99930.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:23:43,670 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5033, 1.7210, 2.8315, 1.3225, 2.1063, 1.9212, 1.5774, 1.9529], + device='cuda:2'), covar=tensor([0.1723, 0.2171, 0.0597, 0.3968, 0.1441, 0.2835, 0.1896, 0.1902], + device='cuda:2'), in_proj_covar=tensor([0.0493, 0.0535, 0.0542, 0.0594, 0.0629, 0.0565, 0.0487, 0.0617], + 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') +2023-02-06 13:23:44,992 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99945.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:23:46,141 INFO [train.py:901] (2/4) Epoch 13, batch 2950, loss[loss=0.2007, simple_loss=0.278, pruned_loss=0.06168, over 7932.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3068, pruned_loss=0.07666, over 1621412.82 frames. ], batch size: 20, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:23:48,969 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:23:52,247 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 13:23:52,906 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.762e+02 3.280e+02 4.150e+02 8.176e+02, threshold=6.560e+02, percent-clipped=12.0 +2023-02-06 13:23:57,303 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99962.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:24:14,478 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:24:21,166 INFO [train.py:901] (2/4) Epoch 13, batch 3000, loss[loss=0.2039, simple_loss=0.2901, pruned_loss=0.05888, over 7968.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3064, pruned_loss=0.07591, over 1622045.81 frames. ], batch size: 21, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:24:21,167 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 13:24:33,565 INFO [train.py:935] (2/4) Epoch 13, validation: loss=0.1841, simple_loss=0.2841, pruned_loss=0.04204, over 944034.00 frames. +2023-02-06 13:24:33,565 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 13:24:37,724 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:24:46,103 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.51 vs. limit=5.0 +2023-02-06 13:24:54,613 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100025.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:24:55,300 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:05,695 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:07,832 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:09,023 INFO [train.py:901] (2/4) Epoch 13, batch 3050, loss[loss=0.2026, simple_loss=0.2732, pruned_loss=0.06604, over 6807.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3055, pruned_loss=0.07517, over 1621495.22 frames. ], batch size: 15, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:25:15,849 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.522e+02 3.008e+02 4.207e+02 1.157e+03, threshold=6.017e+02, percent-clipped=6.0 +2023-02-06 13:25:16,795 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:22,892 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:34,604 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:40,054 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4721, 1.4013, 2.3758, 1.1840, 2.0998, 2.5460, 2.6478, 2.1428], + device='cuda:2'), covar=tensor([0.0937, 0.1169, 0.0416, 0.1972, 0.0722, 0.0359, 0.0659, 0.0766], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0299, 0.0266, 0.0292, 0.0278, 0.0241, 0.0357, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:25:44,732 INFO [train.py:901] (2/4) Epoch 13, batch 3100, loss[loss=0.2118, simple_loss=0.2923, pruned_loss=0.0657, over 7965.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3048, pruned_loss=0.07497, over 1615897.12 frames. ], batch size: 21, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:26:02,703 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9416, 2.1441, 1.8553, 2.8140, 1.2764, 1.6047, 1.8733, 2.3563], + device='cuda:2'), covar=tensor([0.0766, 0.0819, 0.0954, 0.0401, 0.1158, 0.1384, 0.1004, 0.0670], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0211, 0.0255, 0.0214, 0.0214, 0.0251, 0.0260, 0.0220], + 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') +2023-02-06 13:26:11,764 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 13:26:19,854 INFO [train.py:901] (2/4) Epoch 13, batch 3150, loss[loss=0.1982, simple_loss=0.268, pruned_loss=0.06413, over 7708.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3049, pruned_loss=0.07573, over 1613616.94 frames. ], batch size: 18, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:26:20,929 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.17 vs. limit=5.0 +2023-02-06 13:26:22,136 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6626, 2.3529, 3.5086, 2.6452, 3.1925, 2.5747, 2.1387, 1.9853], + device='cuda:2'), covar=tensor([0.4330, 0.4804, 0.1370, 0.3218, 0.2214, 0.2402, 0.1832, 0.4619], + device='cuda:2'), in_proj_covar=tensor([0.0892, 0.0887, 0.0740, 0.0859, 0.0943, 0.0814, 0.0706, 0.0774], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:26:23,028 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 13:26:24,135 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:26:25,933 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.491e+02 3.036e+02 4.077e+02 6.258e+02, threshold=6.072e+02, percent-clipped=1.0 +2023-02-06 13:26:26,824 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:26:29,083 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.96 vs. limit=5.0 +2023-02-06 13:26:41,741 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:26:55,156 INFO [train.py:901] (2/4) Epoch 13, batch 3200, loss[loss=0.2043, simple_loss=0.2757, pruned_loss=0.06642, over 7430.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3046, pruned_loss=0.07545, over 1611175.17 frames. ], batch size: 17, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:26:58,256 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100201.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:27:13,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1540, 1.4140, 1.6057, 1.3127, 0.9914, 1.3192, 1.7952, 1.8823], + device='cuda:2'), covar=tensor([0.0489, 0.1231, 0.1743, 0.1378, 0.0587, 0.1492, 0.0647, 0.0534], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0191, 0.0158, 0.0102, 0.0163, 0.0114, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:27:15,356 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100226.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:27:29,407 INFO [train.py:901] (2/4) Epoch 13, batch 3250, loss[loss=0.2329, simple_loss=0.2917, pruned_loss=0.08708, over 7804.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3049, pruned_loss=0.0761, over 1608792.23 frames. ], batch size: 19, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:27:35,465 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.430e+02 2.991e+02 3.670e+02 7.489e+02, threshold=5.982e+02, percent-clipped=4.0 +2023-02-06 13:27:35,623 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:04,523 INFO [train.py:901] (2/4) Epoch 13, batch 3300, loss[loss=0.2267, simple_loss=0.2996, pruned_loss=0.0769, over 7916.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3062, pruned_loss=0.07646, over 1613009.49 frames. ], batch size: 20, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:28:07,521 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:22,244 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:24,890 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:27,059 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.77 vs. limit=5.0 +2023-02-06 13:28:39,552 INFO [train.py:901] (2/4) Epoch 13, batch 3350, loss[loss=0.2233, simple_loss=0.3116, pruned_loss=0.06748, over 8287.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3055, pruned_loss=0.07565, over 1617392.65 frames. ], batch size: 23, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:28:39,767 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:45,562 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.599e+02 3.166e+02 3.997e+02 7.990e+02, threshold=6.333e+02, percent-clipped=6.0 +2023-02-06 13:28:54,364 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100369.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:29:09,754 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5511, 1.9679, 2.2530, 1.1934, 2.1834, 1.3542, 0.6892, 1.8664], + device='cuda:2'), covar=tensor([0.0475, 0.0232, 0.0160, 0.0467, 0.0315, 0.0678, 0.0621, 0.0245], + device='cuda:2'), in_proj_covar=tensor([0.0398, 0.0341, 0.0292, 0.0399, 0.0327, 0.0490, 0.0362, 0.0366], + device='cuda:2'), out_proj_covar=tensor([1.1149e-04, 9.3206e-05, 8.0129e-05, 1.1026e-04, 9.0678e-05, 1.4605e-04, + 1.0167e-04, 1.0158e-04], device='cuda:2') +2023-02-06 13:29:13,503 INFO [train.py:901] (2/4) Epoch 13, batch 3400, loss[loss=0.2183, simple_loss=0.2836, pruned_loss=0.07653, over 7538.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.305, pruned_loss=0.07567, over 1611279.17 frames. ], batch size: 18, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:29:25,260 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:29:29,919 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100420.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:29:42,784 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:29:43,513 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7671, 1.8803, 2.4428, 1.7384, 1.3423, 2.4556, 0.4517, 1.4467], + device='cuda:2'), covar=tensor([0.2444, 0.1773, 0.0520, 0.1957, 0.3929, 0.0519, 0.3091, 0.2035], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0176, 0.0106, 0.0218, 0.0257, 0.0112, 0.0165, 0.0169], + 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') +2023-02-06 13:29:48,799 INFO [train.py:901] (2/4) Epoch 13, batch 3450, loss[loss=0.2169, simple_loss=0.3026, pruned_loss=0.0656, over 8158.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3049, pruned_loss=0.07528, over 1613643.09 frames. ], batch size: 48, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:29:52,569 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 13:29:54,814 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.523e+02 3.011e+02 4.006e+02 7.808e+02, threshold=6.023e+02, percent-clipped=2.0 +2023-02-06 13:30:13,839 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 13:30:14,319 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100484.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:30:23,421 INFO [train.py:901] (2/4) Epoch 13, batch 3500, loss[loss=0.256, simple_loss=0.3321, pruned_loss=0.08997, over 8325.00 frames. ], tot_loss[loss=0.228, simple_loss=0.305, pruned_loss=0.07552, over 1616203.71 frames. ], batch size: 26, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:30:26,289 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5114, 1.4897, 1.7823, 1.4178, 1.1750, 1.8184, 0.1950, 1.1290], + device='cuda:2'), covar=tensor([0.2676, 0.1627, 0.0474, 0.1315, 0.3571, 0.0529, 0.2684, 0.1604], + device='cuda:2'), in_proj_covar=tensor([0.0172, 0.0175, 0.0106, 0.0217, 0.0254, 0.0111, 0.0164, 0.0169], + 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') +2023-02-06 13:30:45,388 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 13:30:53,033 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 13:30:58,420 INFO [train.py:901] (2/4) Epoch 13, batch 3550, loss[loss=0.2415, simple_loss=0.3242, pruned_loss=0.07941, over 8621.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3041, pruned_loss=0.07489, over 1616270.07 frames. ], batch size: 34, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:31:03,295 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6607, 1.7538, 2.2280, 1.7524, 1.3488, 2.2063, 0.6982, 1.5456], + device='cuda:2'), covar=tensor([0.2748, 0.1227, 0.0591, 0.1525, 0.3113, 0.0648, 0.2521, 0.1582], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0175, 0.0106, 0.0217, 0.0253, 0.0111, 0.0164, 0.0169], + 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') +2023-02-06 13:31:04,446 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.564e+02 3.091e+02 3.906e+02 9.185e+02, threshold=6.182e+02, percent-clipped=3.0 +2023-02-06 13:31:12,643 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2900, 2.7003, 2.3056, 3.6884, 1.6757, 1.9026, 2.2037, 2.7755], + device='cuda:2'), covar=tensor([0.0720, 0.0754, 0.0893, 0.0282, 0.1159, 0.1339, 0.1104, 0.0699], + device='cuda:2'), in_proj_covar=tensor([0.0236, 0.0211, 0.0256, 0.0213, 0.0215, 0.0253, 0.0261, 0.0220], + 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') +2023-02-06 13:31:33,167 INFO [train.py:901] (2/4) Epoch 13, batch 3600, loss[loss=0.2521, simple_loss=0.318, pruned_loss=0.09313, over 8611.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3035, pruned_loss=0.07475, over 1612663.44 frames. ], batch size: 34, lr: 5.89e-03, grad_scale: 16.0 +2023-02-06 13:31:35,334 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:32:08,079 INFO [train.py:901] (2/4) Epoch 13, batch 3650, loss[loss=0.2836, simple_loss=0.3554, pruned_loss=0.1059, over 8504.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.303, pruned_loss=0.07431, over 1614863.61 frames. ], batch size: 26, lr: 5.89e-03, grad_scale: 16.0 +2023-02-06 13:32:14,109 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.362e+02 3.080e+02 3.827e+02 7.938e+02, threshold=6.161e+02, percent-clipped=3.0 +2023-02-06 13:32:40,076 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4709, 1.8155, 3.1220, 1.3009, 2.1656, 1.8887, 1.5453, 2.0361], + device='cuda:2'), covar=tensor([0.1831, 0.2410, 0.0684, 0.4122, 0.1673, 0.2995, 0.2038, 0.2137], + device='cuda:2'), in_proj_covar=tensor([0.0490, 0.0531, 0.0534, 0.0587, 0.0623, 0.0556, 0.0481, 0.0611], + 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') +2023-02-06 13:32:43,148 INFO [train.py:901] (2/4) Epoch 13, batch 3700, loss[loss=0.1756, simple_loss=0.2483, pruned_loss=0.05146, over 7708.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3031, pruned_loss=0.07415, over 1616228.53 frames. ], batch size: 18, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:32:55,275 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:32:57,155 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 13:33:13,283 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100740.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:33:17,842 INFO [train.py:901] (2/4) Epoch 13, batch 3750, loss[loss=0.2399, simple_loss=0.3256, pruned_loss=0.07704, over 8643.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3022, pruned_loss=0.07357, over 1615394.42 frames. ], batch size: 49, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:33:18,685 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100748.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:33:24,691 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.543e+02 3.029e+02 3.909e+02 6.778e+02, threshold=6.059e+02, percent-clipped=2.0 +2023-02-06 13:33:30,142 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:33:30,989 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100765.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:33:52,960 INFO [train.py:901] (2/4) Epoch 13, batch 3800, loss[loss=0.1736, simple_loss=0.2547, pruned_loss=0.04622, over 7925.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3027, pruned_loss=0.07435, over 1615356.83 frames. ], batch size: 20, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:34:27,589 INFO [train.py:901] (2/4) Epoch 13, batch 3850, loss[loss=0.1982, simple_loss=0.2666, pruned_loss=0.0649, over 7786.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3032, pruned_loss=0.07471, over 1612115.44 frames. ], batch size: 19, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:34:34,506 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 2.830e+02 3.312e+02 3.730e+02 7.453e+02, threshold=6.624e+02, percent-clipped=3.0 +2023-02-06 13:34:50,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100879.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:35:01,752 INFO [train.py:901] (2/4) Epoch 13, batch 3900, loss[loss=0.1987, simple_loss=0.2797, pruned_loss=0.05884, over 8130.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3034, pruned_loss=0.07496, over 1612936.46 frames. ], batch size: 22, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:35:01,759 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 13:35:33,289 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6729, 2.3476, 3.4247, 2.6149, 3.0836, 2.4523, 2.0791, 1.7658], + device='cuda:2'), covar=tensor([0.3917, 0.4305, 0.1305, 0.2939, 0.2008, 0.2255, 0.1595, 0.4644], + device='cuda:2'), in_proj_covar=tensor([0.0891, 0.0887, 0.0734, 0.0862, 0.0941, 0.0813, 0.0705, 0.0776], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:35:37,124 INFO [train.py:901] (2/4) Epoch 13, batch 3950, loss[loss=0.2155, simple_loss=0.2827, pruned_loss=0.07417, over 7518.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3037, pruned_loss=0.07511, over 1611980.58 frames. ], batch size: 18, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:35:44,007 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.489e+02 3.011e+02 3.855e+02 9.802e+02, threshold=6.021e+02, percent-clipped=2.0 +2023-02-06 13:35:54,139 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100971.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:36:10,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2612, 1.2578, 3.3467, 1.0426, 2.9407, 2.7989, 3.0405, 2.9938], + device='cuda:2'), covar=tensor([0.0662, 0.3709, 0.0760, 0.3588, 0.1355, 0.1053, 0.0684, 0.0772], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0588, 0.0612, 0.0551, 0.0630, 0.0543, 0.0533, 0.0593], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 13:36:11,900 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100996.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:36:12,364 INFO [train.py:901] (2/4) Epoch 13, batch 4000, loss[loss=0.2348, simple_loss=0.3109, pruned_loss=0.07939, over 8506.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3032, pruned_loss=0.07526, over 1608714.16 frames. ], batch size: 26, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:36:20,495 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.95 vs. limit=5.0 +2023-02-06 13:36:47,690 INFO [train.py:901] (2/4) Epoch 13, batch 4050, loss[loss=0.1972, simple_loss=0.2852, pruned_loss=0.05461, over 8133.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3026, pruned_loss=0.07492, over 1604664.83 frames. ], batch size: 22, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:36:54,253 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.645e+02 3.184e+02 3.816e+02 9.518e+02, threshold=6.368e+02, percent-clipped=3.0 +2023-02-06 13:37:18,352 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101092.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:37:21,623 INFO [train.py:901] (2/4) Epoch 13, batch 4100, loss[loss=0.2381, simple_loss=0.3221, pruned_loss=0.07705, over 8571.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3037, pruned_loss=0.07528, over 1608660.21 frames. ], batch size: 34, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:37:41,407 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:37:47,945 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:37:55,650 INFO [train.py:901] (2/4) Epoch 13, batch 4150, loss[loss=0.1706, simple_loss=0.2682, pruned_loss=0.03646, over 7799.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3035, pruned_loss=0.07504, over 1610120.46 frames. ], batch size: 19, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:38:02,997 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.783e+02 3.401e+02 4.642e+02 1.010e+03, threshold=6.803e+02, percent-clipped=7.0 +2023-02-06 13:38:05,275 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101160.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:38:05,884 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5684, 5.6496, 4.9119, 2.2034, 5.0007, 5.2723, 5.2231, 4.9877], + device='cuda:2'), covar=tensor([0.0547, 0.0362, 0.0817, 0.4630, 0.0618, 0.0684, 0.1003, 0.0660], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0391, 0.0401, 0.0500, 0.0394, 0.0395, 0.0389, 0.0344], + 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') +2023-02-06 13:38:06,616 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0951, 1.4135, 1.5459, 1.2806, 0.8912, 1.3517, 1.7392, 1.5998], + device='cuda:2'), covar=tensor([0.0560, 0.1678, 0.2339, 0.1712, 0.0706, 0.1964, 0.0753, 0.0682], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0192, 0.0157, 0.0102, 0.0164, 0.0116, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:38:30,242 INFO [train.py:901] (2/4) Epoch 13, batch 4200, loss[loss=0.2207, simple_loss=0.3106, pruned_loss=0.06538, over 8458.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3033, pruned_loss=0.07432, over 1615742.18 frames. ], batch size: 25, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:38:37,998 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101207.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:38:39,766 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 13:38:54,601 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 13:39:05,859 INFO [train.py:901] (2/4) Epoch 13, batch 4250, loss[loss=0.2031, simple_loss=0.2813, pruned_loss=0.06249, over 7967.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3032, pruned_loss=0.0746, over 1612577.19 frames. ], batch size: 21, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:39:12,480 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 2.514e+02 3.154e+02 3.992e+02 7.648e+02, threshold=6.307e+02, percent-clipped=3.0 +2023-02-06 13:39:16,498 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 13:39:40,012 INFO [train.py:901] (2/4) Epoch 13, batch 4300, loss[loss=0.1977, simple_loss=0.2819, pruned_loss=0.0568, over 8243.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3037, pruned_loss=0.07457, over 1615183.65 frames. ], batch size: 22, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:39:58,266 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 13:40:01,018 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 +2023-02-06 13:40:14,581 INFO [train.py:901] (2/4) Epoch 13, batch 4350, loss[loss=0.2255, simple_loss=0.3031, pruned_loss=0.07396, over 7971.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3046, pruned_loss=0.07523, over 1614805.91 frames. ], batch size: 21, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:40:14,815 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6117, 1.6364, 2.0048, 1.5413, 1.2408, 2.0731, 0.3680, 1.3489], + device='cuda:2'), covar=tensor([0.2029, 0.1535, 0.0456, 0.1366, 0.3361, 0.0427, 0.2763, 0.1596], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0176, 0.0107, 0.0219, 0.0256, 0.0111, 0.0164, 0.0170], + 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') +2023-02-06 13:40:21,341 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.682e+02 3.184e+02 4.441e+02 9.358e+02, threshold=6.368e+02, percent-clipped=11.0 +2023-02-06 13:40:49,428 INFO [train.py:901] (2/4) Epoch 13, batch 4400, loss[loss=0.2088, simple_loss=0.2772, pruned_loss=0.07019, over 7410.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3051, pruned_loss=0.07565, over 1615925.75 frames. ], batch size: 17, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:40:49,433 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 13:40:57,009 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6699, 1.4037, 1.6021, 1.2618, 0.8531, 1.3406, 1.3754, 1.3011], + device='cuda:2'), covar=tensor([0.0519, 0.1299, 0.1700, 0.1403, 0.0613, 0.1561, 0.0723, 0.0643], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0151, 0.0189, 0.0157, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:41:23,440 INFO [train.py:901] (2/4) Epoch 13, batch 4450, loss[loss=0.2154, simple_loss=0.2874, pruned_loss=0.07171, over 8070.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.305, pruned_loss=0.07597, over 1611191.11 frames. ], batch size: 21, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:41:28,823 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 13:41:30,660 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.700e+02 3.319e+02 4.103e+02 1.285e+03, threshold=6.638e+02, percent-clipped=3.0 +2023-02-06 13:41:34,863 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101463.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:41:38,576 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:41:52,070 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5146, 2.8521, 1.9869, 2.2212, 2.3576, 1.6338, 2.1081, 2.2273], + device='cuda:2'), covar=tensor([0.1378, 0.0300, 0.0956, 0.0648, 0.0583, 0.1281, 0.0879, 0.0819], + device='cuda:2'), in_proj_covar=tensor([0.0342, 0.0233, 0.0313, 0.0294, 0.0298, 0.0319, 0.0338, 0.0301], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:41:52,075 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101488.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:41:57,886 INFO [train.py:901] (2/4) Epoch 13, batch 4500, loss[loss=0.2657, simple_loss=0.332, pruned_loss=0.09974, over 8317.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3056, pruned_loss=0.07625, over 1612634.90 frames. ], batch size: 25, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:42:14,781 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2208, 1.2318, 1.6362, 1.1734, 0.6628, 1.3120, 1.1468, 1.1188], + device='cuda:2'), covar=tensor([0.0522, 0.1303, 0.1555, 0.1364, 0.0548, 0.1488, 0.0643, 0.0629], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0151, 0.0189, 0.0156, 0.0101, 0.0161, 0.0114, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:42:22,208 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 13:42:33,098 INFO [train.py:901] (2/4) Epoch 13, batch 4550, loss[loss=0.2626, simple_loss=0.3108, pruned_loss=0.1071, over 7803.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3044, pruned_loss=0.0758, over 1609606.59 frames. ], batch size: 20, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:42:39,881 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.403e+02 2.986e+02 3.546e+02 6.918e+02, threshold=5.973e+02, percent-clipped=1.0 +2023-02-06 13:42:58,212 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1680, 4.1469, 3.7147, 1.9949, 3.6249, 3.7699, 3.8666, 3.5059], + device='cuda:2'), covar=tensor([0.0885, 0.0566, 0.1064, 0.4469, 0.0955, 0.1046, 0.1200, 0.1000], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0384, 0.0389, 0.0488, 0.0383, 0.0387, 0.0378, 0.0336], + 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') +2023-02-06 13:42:58,931 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101584.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:43:08,196 INFO [train.py:901] (2/4) Epoch 13, batch 4600, loss[loss=0.244, simple_loss=0.3292, pruned_loss=0.07943, over 8481.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3048, pruned_loss=0.07575, over 1611047.65 frames. ], batch size: 29, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:43:17,792 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1207, 1.1734, 1.1017, 1.4892, 0.5772, 0.9951, 1.0398, 1.2483], + device='cuda:2'), covar=tensor([0.0753, 0.0714, 0.0951, 0.0507, 0.1015, 0.1158, 0.0681, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0210, 0.0254, 0.0212, 0.0214, 0.0250, 0.0256, 0.0216], + 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') +2023-02-06 13:43:42,600 INFO [train.py:901] (2/4) Epoch 13, batch 4650, loss[loss=0.2346, simple_loss=0.3101, pruned_loss=0.07951, over 8301.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3057, pruned_loss=0.07687, over 1609855.19 frames. ], batch size: 23, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:43:49,456 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.500e+02 2.989e+02 3.844e+02 7.619e+02, threshold=5.978e+02, percent-clipped=4.0 +2023-02-06 13:44:17,683 INFO [train.py:901] (2/4) Epoch 13, batch 4700, loss[loss=0.2483, simple_loss=0.3278, pruned_loss=0.08439, over 8459.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3066, pruned_loss=0.07704, over 1613876.86 frames. ], batch size: 27, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:44:21,756 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:44:52,084 INFO [train.py:901] (2/4) Epoch 13, batch 4750, loss[loss=0.1995, simple_loss=0.2734, pruned_loss=0.06281, over 7540.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3047, pruned_loss=0.07588, over 1613356.74 frames. ], batch size: 18, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:44:59,493 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.558e+02 3.081e+02 3.778e+02 8.564e+02, threshold=6.162e+02, percent-clipped=2.0 +2023-02-06 13:45:21,968 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 13:45:24,382 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 13:45:27,098 INFO [train.py:901] (2/4) Epoch 13, batch 4800, loss[loss=0.2467, simple_loss=0.3231, pruned_loss=0.08511, over 8603.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3047, pruned_loss=0.07543, over 1614918.34 frames. ], batch size: 31, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:45:32,218 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0766, 2.5259, 3.1304, 1.2121, 3.2014, 1.6943, 1.5131, 1.9556], + device='cuda:2'), covar=tensor([0.0623, 0.0259, 0.0180, 0.0652, 0.0304, 0.0733, 0.0733, 0.0450], + device='cuda:2'), in_proj_covar=tensor([0.0396, 0.0334, 0.0287, 0.0393, 0.0325, 0.0482, 0.0360, 0.0364], + device='cuda:2'), out_proj_covar=tensor([1.1075e-04, 9.1230e-05, 7.8297e-05, 1.0815e-04, 9.0008e-05, 1.4335e-04, + 1.0114e-04, 1.0107e-04], device='cuda:2') +2023-02-06 13:45:38,839 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5050, 1.8441, 1.9792, 1.0875, 2.0056, 1.3564, 0.3858, 1.6506], + device='cuda:2'), covar=tensor([0.0356, 0.0224, 0.0165, 0.0352, 0.0250, 0.0594, 0.0551, 0.0192], + device='cuda:2'), in_proj_covar=tensor([0.0397, 0.0335, 0.0287, 0.0394, 0.0327, 0.0483, 0.0361, 0.0365], + device='cuda:2'), out_proj_covar=tensor([1.1113e-04, 9.1418e-05, 7.8489e-05, 1.0848e-04, 9.0429e-05, 1.4374e-04, + 1.0132e-04, 1.0128e-04], device='cuda:2') +2023-02-06 13:45:57,602 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101840.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:46:01,953 INFO [train.py:901] (2/4) Epoch 13, batch 4850, loss[loss=0.2658, simple_loss=0.3388, pruned_loss=0.09639, over 8239.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3068, pruned_loss=0.07659, over 1612734.28 frames. ], batch size: 24, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:46:08,644 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.589e+02 3.137e+02 3.918e+02 7.572e+02, threshold=6.274e+02, percent-clipped=4.0 +2023-02-06 13:46:14,081 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 13:46:14,940 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:46:36,896 INFO [train.py:901] (2/4) Epoch 13, batch 4900, loss[loss=0.2328, simple_loss=0.311, pruned_loss=0.07732, over 8245.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3059, pruned_loss=0.07635, over 1613067.79 frames. ], batch size: 22, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:47:06,387 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101938.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:47:12,292 INFO [train.py:901] (2/4) Epoch 13, batch 4950, loss[loss=0.2275, simple_loss=0.3116, pruned_loss=0.07172, over 8345.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3061, pruned_loss=0.07645, over 1612778.51 frames. ], batch size: 26, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:47:15,205 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3085, 1.4701, 1.3885, 1.8957, 0.8032, 1.1501, 1.2892, 1.4572], + device='cuda:2'), covar=tensor([0.1008, 0.0874, 0.1151, 0.0529, 0.1228, 0.1576, 0.0851, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0211, 0.0253, 0.0212, 0.0215, 0.0252, 0.0254, 0.0217], + 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') +2023-02-06 13:47:19,104 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.432e+02 3.023e+02 3.670e+02 7.494e+02, threshold=6.046e+02, percent-clipped=3.0 +2023-02-06 13:47:20,638 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1455, 1.5516, 1.6877, 1.4978, 0.9145, 1.5678, 1.6689, 1.7123], + device='cuda:2'), covar=tensor([0.0492, 0.1198, 0.1654, 0.1363, 0.0605, 0.1425, 0.0703, 0.0577], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0191, 0.0157, 0.0102, 0.0163, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:47:46,757 INFO [train.py:901] (2/4) Epoch 13, batch 5000, loss[loss=0.1707, simple_loss=0.2612, pruned_loss=0.0401, over 8289.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3051, pruned_loss=0.07497, over 1610976.04 frames. ], batch size: 23, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:47:52,333 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 13:47:53,374 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:48:16,623 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1179, 1.5359, 1.7380, 1.4092, 1.0044, 1.5069, 1.7595, 1.6311], + device='cuda:2'), covar=tensor([0.0466, 0.1226, 0.1646, 0.1363, 0.0591, 0.1470, 0.0648, 0.0630], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0158, 0.0102, 0.0163, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 13:48:22,479 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102046.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:48:23,083 INFO [train.py:901] (2/4) Epoch 13, batch 5050, loss[loss=0.1939, simple_loss=0.2766, pruned_loss=0.05561, over 7549.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3053, pruned_loss=0.07556, over 1607838.64 frames. ], batch size: 18, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:48:29,968 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.626e+02 3.300e+02 4.185e+02 9.088e+02, threshold=6.599e+02, percent-clipped=3.0 +2023-02-06 13:48:54,017 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 13:48:56,650 INFO [train.py:901] (2/4) Epoch 13, batch 5100, loss[loss=0.2251, simple_loss=0.3065, pruned_loss=0.07181, over 8237.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.304, pruned_loss=0.07466, over 1609319.37 frames. ], batch size: 22, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:49:31,560 INFO [train.py:901] (2/4) Epoch 13, batch 5150, loss[loss=0.2257, simple_loss=0.3016, pruned_loss=0.07497, over 8455.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3042, pruned_loss=0.07478, over 1613731.76 frames. ], batch size: 25, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:49:38,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.413e+02 2.853e+02 3.425e+02 7.647e+02, threshold=5.706e+02, percent-clipped=3.0 +2023-02-06 13:49:41,790 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:50:06,644 INFO [train.py:901] (2/4) Epoch 13, batch 5200, loss[loss=0.2498, simple_loss=0.3383, pruned_loss=0.08069, over 8359.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3052, pruned_loss=0.07547, over 1614039.10 frames. ], batch size: 24, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:50:41,855 INFO [train.py:901] (2/4) Epoch 13, batch 5250, loss[loss=0.2471, simple_loss=0.3206, pruned_loss=0.08676, over 8486.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3044, pruned_loss=0.07486, over 1614422.22 frames. ], batch size: 49, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:50:48,577 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.565e+02 3.047e+02 3.925e+02 1.157e+03, threshold=6.094e+02, percent-clipped=6.0 +2023-02-06 13:50:51,019 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.87 vs. limit=5.0 +2023-02-06 13:50:53,901 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 13:51:06,839 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102282.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:51:16,475 INFO [train.py:901] (2/4) Epoch 13, batch 5300, loss[loss=0.1913, simple_loss=0.2674, pruned_loss=0.05758, over 7925.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.303, pruned_loss=0.07412, over 1614187.40 frames. ], batch size: 20, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:51:51,015 INFO [train.py:901] (2/4) Epoch 13, batch 5350, loss[loss=0.2141, simple_loss=0.2972, pruned_loss=0.06552, over 8087.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3018, pruned_loss=0.07331, over 1614217.62 frames. ], batch size: 21, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:51:52,483 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:51:52,611 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5094, 2.7625, 1.8562, 2.1713, 2.2184, 1.5444, 2.0018, 2.1561], + device='cuda:2'), covar=tensor([0.1465, 0.0389, 0.1149, 0.0619, 0.0659, 0.1366, 0.1022, 0.0929], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0236, 0.0318, 0.0297, 0.0298, 0.0324, 0.0345, 0.0304], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:51:57,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.535e+02 3.049e+02 3.805e+02 7.372e+02, threshold=6.098e+02, percent-clipped=2.0 +2023-02-06 13:52:08,203 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 +2023-02-06 13:52:26,065 INFO [train.py:901] (2/4) Epoch 13, batch 5400, loss[loss=0.2615, simple_loss=0.3186, pruned_loss=0.1022, over 6881.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3012, pruned_loss=0.07357, over 1609175.80 frames. ], batch size: 71, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:52:26,257 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102397.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:52:40,299 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:52:56,878 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102442.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:53:00,140 INFO [train.py:901] (2/4) Epoch 13, batch 5450, loss[loss=0.2047, simple_loss=0.2848, pruned_loss=0.06234, over 8084.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3035, pruned_loss=0.07508, over 1613241.94 frames. ], batch size: 21, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:53:07,658 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.724e+02 3.222e+02 3.900e+02 7.023e+02, threshold=6.444e+02, percent-clipped=3.0 +2023-02-06 13:53:12,593 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:53:34,976 INFO [train.py:901] (2/4) Epoch 13, batch 5500, loss[loss=0.207, simple_loss=0.2955, pruned_loss=0.05924, over 8255.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3034, pruned_loss=0.07483, over 1616796.59 frames. ], batch size: 24, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:53:41,583 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 13:54:03,097 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0903, 1.6516, 3.3318, 1.5372, 2.2561, 3.8270, 3.9064, 3.2138], + device='cuda:2'), covar=tensor([0.1056, 0.1568, 0.0409, 0.2142, 0.1133, 0.0228, 0.0511, 0.0622], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0300, 0.0265, 0.0296, 0.0277, 0.0237, 0.0360, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:54:09,466 INFO [train.py:901] (2/4) Epoch 13, batch 5550, loss[loss=0.2162, simple_loss=0.2914, pruned_loss=0.07049, over 7664.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3036, pruned_loss=0.07469, over 1615696.85 frames. ], batch size: 19, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:54:15,938 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.373e+02 2.279e+02 3.010e+02 3.933e+02 6.976e+02, threshold=6.019e+02, percent-clipped=1.0 +2023-02-06 13:54:43,200 INFO [train.py:901] (2/4) Epoch 13, batch 5600, loss[loss=0.2239, simple_loss=0.3101, pruned_loss=0.06884, over 7978.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3025, pruned_loss=0.07434, over 1612719.58 frames. ], batch size: 21, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:54:44,703 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2119, 1.8741, 3.2753, 1.7348, 2.3946, 3.6815, 3.6363, 3.2201], + device='cuda:2'), covar=tensor([0.1027, 0.1470, 0.0453, 0.1933, 0.1221, 0.0218, 0.0634, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0302, 0.0266, 0.0296, 0.0278, 0.0239, 0.0360, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 13:55:04,664 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5180, 1.8095, 1.9610, 1.0785, 2.0017, 1.4345, 0.4049, 1.7537], + device='cuda:2'), covar=tensor([0.0338, 0.0242, 0.0164, 0.0387, 0.0268, 0.0632, 0.0599, 0.0169], + device='cuda:2'), in_proj_covar=tensor([0.0406, 0.0341, 0.0293, 0.0401, 0.0332, 0.0490, 0.0368, 0.0373], + device='cuda:2'), out_proj_covar=tensor([1.1347e-04, 9.3025e-05, 7.9922e-05, 1.1031e-04, 9.1562e-05, 1.4528e-04, + 1.0327e-04, 1.0324e-04], device='cuda:2') +2023-02-06 13:55:18,286 INFO [train.py:901] (2/4) Epoch 13, batch 5650, loss[loss=0.2339, simple_loss=0.3194, pruned_loss=0.07414, over 8131.00 frames. ], tot_loss[loss=0.225, simple_loss=0.302, pruned_loss=0.07399, over 1610809.42 frames. ], batch size: 22, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:55:22,472 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102653.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:55:24,861 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.730e+02 3.267e+02 4.266e+02 8.129e+02, threshold=6.534e+02, percent-clipped=5.0 +2023-02-06 13:55:39,201 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102678.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:55:43,635 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 13:55:52,506 INFO [train.py:901] (2/4) Epoch 13, batch 5700, loss[loss=0.2168, simple_loss=0.2955, pruned_loss=0.06907, over 7826.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3031, pruned_loss=0.07437, over 1609567.95 frames. ], batch size: 20, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:56:08,738 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102720.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:56:25,982 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:56:27,756 INFO [train.py:901] (2/4) Epoch 13, batch 5750, loss[loss=0.2227, simple_loss=0.3074, pruned_loss=0.06901, over 8374.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3042, pruned_loss=0.07494, over 1612773.75 frames. ], batch size: 24, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:56:34,421 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.514e+02 3.075e+02 4.012e+02 7.214e+02, threshold=6.150e+02, percent-clipped=2.0 +2023-02-06 13:56:47,297 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 13:56:57,966 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-06 13:57:01,390 INFO [train.py:901] (2/4) Epoch 13, batch 5800, loss[loss=0.2623, simple_loss=0.3288, pruned_loss=0.09792, over 8484.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3034, pruned_loss=0.07459, over 1609873.49 frames. ], batch size: 25, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:57:28,953 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3878, 1.7287, 1.8138, 0.9839, 1.8714, 1.2408, 0.3742, 1.5387], + device='cuda:2'), covar=tensor([0.0503, 0.0305, 0.0234, 0.0513, 0.0341, 0.0846, 0.0766, 0.0281], + device='cuda:2'), in_proj_covar=tensor([0.0407, 0.0342, 0.0295, 0.0403, 0.0333, 0.0492, 0.0369, 0.0374], + device='cuda:2'), out_proj_covar=tensor([1.1376e-04, 9.3217e-05, 8.0403e-05, 1.1081e-04, 9.1968e-05, 1.4587e-04, + 1.0361e-04, 1.0345e-04], device='cuda:2') +2023-02-06 13:57:36,623 INFO [train.py:901] (2/4) Epoch 13, batch 5850, loss[loss=0.2272, simple_loss=0.3119, pruned_loss=0.07125, over 8463.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3038, pruned_loss=0.07455, over 1611777.94 frames. ], batch size: 29, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:57:36,816 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9560, 2.0848, 1.9603, 2.8273, 1.4364, 1.5489, 2.0606, 2.2893], + device='cuda:2'), covar=tensor([0.0743, 0.0951, 0.0994, 0.0402, 0.1144, 0.1480, 0.0846, 0.0773], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0210, 0.0254, 0.0212, 0.0216, 0.0254, 0.0256, 0.0218], + 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') +2023-02-06 13:57:43,173 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.218e+02 2.874e+02 3.517e+02 7.476e+02, threshold=5.748e+02, percent-clipped=3.0 +2023-02-06 13:57:52,588 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102869.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:58:11,324 INFO [train.py:901] (2/4) Epoch 13, batch 5900, loss[loss=0.2954, simple_loss=0.347, pruned_loss=0.1218, over 6843.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3038, pruned_loss=0.07486, over 1605677.61 frames. ], batch size: 72, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:58:16,843 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:58:46,272 INFO [train.py:901] (2/4) Epoch 13, batch 5950, loss[loss=0.2068, simple_loss=0.2905, pruned_loss=0.06156, over 8239.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3055, pruned_loss=0.07586, over 1609937.39 frames. ], batch size: 22, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:58:52,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 2.537e+02 3.124e+02 4.010e+02 1.248e+03, threshold=6.247e+02, percent-clipped=9.0 +2023-02-06 13:59:21,495 INFO [train.py:901] (2/4) Epoch 13, batch 6000, loss[loss=0.2811, simple_loss=0.3396, pruned_loss=0.1113, over 7134.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3057, pruned_loss=0.07595, over 1608600.08 frames. ], batch size: 71, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:59:21,495 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 13:59:36,606 INFO [train.py:935] (2/4) Epoch 13, validation: loss=0.1836, simple_loss=0.2836, pruned_loss=0.04176, over 944034.00 frames. +2023-02-06 13:59:36,607 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 14:00:11,030 INFO [train.py:901] (2/4) Epoch 13, batch 6050, loss[loss=0.1972, simple_loss=0.2702, pruned_loss=0.06206, over 7658.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3055, pruned_loss=0.07574, over 1610425.12 frames. ], batch size: 19, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:00:14,465 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4436, 1.5418, 1.8171, 1.1784, 0.9657, 1.7977, 0.0816, 1.1459], + device='cuda:2'), covar=tensor([0.2178, 0.1438, 0.0472, 0.1660, 0.3754, 0.0420, 0.2559, 0.1505], + device='cuda:2'), in_proj_covar=tensor([0.0172, 0.0173, 0.0106, 0.0218, 0.0254, 0.0110, 0.0164, 0.0170], + 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') +2023-02-06 14:00:18,307 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.701e+02 2.480e+02 3.014e+02 3.999e+02 8.436e+02, threshold=6.027e+02, percent-clipped=4.0 +2023-02-06 14:00:21,130 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2082, 4.1070, 3.7820, 1.8337, 3.6432, 3.7625, 3.7861, 3.4137], + device='cuda:2'), covar=tensor([0.0830, 0.0688, 0.1046, 0.5022, 0.0980, 0.0897, 0.1450, 0.1050], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0391, 0.0397, 0.0495, 0.0391, 0.0392, 0.0389, 0.0343], + 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') +2023-02-06 14:00:45,637 INFO [train.py:901] (2/4) Epoch 13, batch 6100, loss[loss=0.2304, simple_loss=0.3083, pruned_loss=0.07623, over 8330.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3061, pruned_loss=0.07614, over 1611535.14 frames. ], batch size: 26, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:00:58,590 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103116.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:01:14,163 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 14:01:19,578 INFO [train.py:901] (2/4) Epoch 13, batch 6150, loss[loss=0.1817, simple_loss=0.2603, pruned_loss=0.05158, over 7221.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3053, pruned_loss=0.07556, over 1612622.72 frames. ], batch size: 16, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:01:21,558 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:01:26,199 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.434e+02 3.117e+02 4.172e+02 7.466e+02, threshold=6.235e+02, percent-clipped=2.0 +2023-02-06 14:01:46,193 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5078, 1.4210, 2.8112, 1.2294, 1.9101, 2.9860, 3.1020, 2.5107], + device='cuda:2'), covar=tensor([0.1279, 0.1627, 0.0420, 0.2272, 0.1067, 0.0342, 0.0694, 0.0760], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0303, 0.0265, 0.0297, 0.0278, 0.0239, 0.0361, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:01:51,716 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0418, 1.6584, 3.3604, 1.4072, 2.2324, 3.6828, 3.6924, 3.1061], + device='cuda:2'), covar=tensor([0.1098, 0.1532, 0.0401, 0.2087, 0.1114, 0.0220, 0.0517, 0.0601], + device='cuda:2'), in_proj_covar=tensor([0.0267, 0.0301, 0.0264, 0.0296, 0.0277, 0.0239, 0.0359, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:01:55,078 INFO [train.py:901] (2/4) Epoch 13, batch 6200, loss[loss=0.1892, simple_loss=0.2699, pruned_loss=0.05425, over 8079.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3044, pruned_loss=0.075, over 1616789.92 frames. ], batch size: 21, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:02:06,143 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:02:30,996 INFO [train.py:901] (2/4) Epoch 13, batch 6250, loss[loss=0.1726, simple_loss=0.252, pruned_loss=0.04658, over 7812.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3039, pruned_loss=0.07493, over 1612004.81 frames. ], batch size: 20, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:02:32,362 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103249.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:02:37,850 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.482e+02 2.950e+02 3.630e+02 6.819e+02, threshold=5.900e+02, percent-clipped=4.0 +2023-02-06 14:03:06,056 INFO [train.py:901] (2/4) Epoch 13, batch 6300, loss[loss=0.1979, simple_loss=0.2625, pruned_loss=0.0667, over 7280.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3034, pruned_loss=0.07495, over 1614237.69 frames. ], batch size: 16, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:03:27,935 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:03:40,545 INFO [train.py:901] (2/4) Epoch 13, batch 6350, loss[loss=0.219, simple_loss=0.3004, pruned_loss=0.06875, over 8553.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3041, pruned_loss=0.07508, over 1614348.86 frames. ], batch size: 31, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:03:48,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.547e+02 3.093e+02 3.716e+02 8.603e+02, threshold=6.185e+02, percent-clipped=3.0 +2023-02-06 14:03:52,985 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:04:06,352 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103384.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:04:14,877 INFO [train.py:901] (2/4) Epoch 13, batch 6400, loss[loss=0.2456, simple_loss=0.3067, pruned_loss=0.09227, over 8243.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3035, pruned_loss=0.07459, over 1618989.30 frames. ], batch size: 22, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:04:49,451 INFO [train.py:901] (2/4) Epoch 13, batch 6450, loss[loss=0.21, simple_loss=0.2816, pruned_loss=0.06923, over 7648.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3034, pruned_loss=0.0742, over 1617054.05 frames. ], batch size: 19, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:04:56,168 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.528e+02 3.186e+02 3.863e+02 6.544e+02, threshold=6.372e+02, percent-clipped=1.0 +2023-02-06 14:04:58,241 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103460.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:05:13,645 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2416, 1.2197, 1.5279, 1.1549, 0.6901, 1.3109, 1.1977, 1.2003], + device='cuda:2'), covar=tensor([0.0528, 0.1258, 0.1677, 0.1390, 0.0573, 0.1484, 0.0638, 0.0626], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0150, 0.0189, 0.0155, 0.0100, 0.0161, 0.0113, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:05:22,247 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:05:24,240 INFO [train.py:901] (2/4) Epoch 13, batch 6500, loss[loss=0.2211, simple_loss=0.2985, pruned_loss=0.07185, over 8159.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3025, pruned_loss=0.07423, over 1615140.00 frames. ], batch size: 48, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:05:58,666 INFO [train.py:901] (2/4) Epoch 13, batch 6550, loss[loss=0.1978, simple_loss=0.2818, pruned_loss=0.05693, over 7934.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3032, pruned_loss=0.07446, over 1615895.73 frames. ], batch size: 20, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:06:05,487 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.442e+02 3.089e+02 4.027e+02 9.292e+02, threshold=6.177e+02, percent-clipped=8.0 +2023-02-06 14:06:12,684 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5199, 5.5948, 4.9976, 2.4799, 4.8696, 5.3161, 5.2035, 5.0680], + device='cuda:2'), covar=tensor([0.0555, 0.0400, 0.0848, 0.4383, 0.0680, 0.0659, 0.0938, 0.0637], + device='cuda:2'), in_proj_covar=tensor([0.0476, 0.0389, 0.0399, 0.0498, 0.0394, 0.0392, 0.0385, 0.0343], + 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') +2023-02-06 14:06:18,302 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103575.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:24,184 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 14:06:24,395 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103584.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:34,118 INFO [train.py:901] (2/4) Epoch 13, batch 6600, loss[loss=0.1991, simple_loss=0.2785, pruned_loss=0.05984, over 7925.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3023, pruned_loss=0.07375, over 1618188.86 frames. ], batch size: 20, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:06:36,427 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6377, 1.1954, 1.4500, 1.1199, 0.7893, 1.2223, 1.4744, 1.3720], + device='cuda:2'), covar=tensor([0.0546, 0.1314, 0.1895, 0.1519, 0.0621, 0.1626, 0.0715, 0.0697], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0150, 0.0189, 0.0154, 0.0100, 0.0161, 0.0114, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:06:37,031 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0121, 1.3598, 1.5589, 1.2566, 0.8300, 1.3585, 1.6425, 1.6594], + device='cuda:2'), covar=tensor([0.0483, 0.1197, 0.1657, 0.1404, 0.0622, 0.1500, 0.0687, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0150, 0.0189, 0.0154, 0.0100, 0.0161, 0.0114, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:06:42,485 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:42,507 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:43,683 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 14:06:49,829 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:07:02,402 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6339, 2.2575, 3.6400, 2.6525, 3.1274, 2.4964, 2.0747, 1.8613], + device='cuda:2'), covar=tensor([0.4115, 0.4517, 0.1331, 0.3241, 0.2200, 0.2364, 0.1733, 0.4915], + device='cuda:2'), in_proj_covar=tensor([0.0896, 0.0887, 0.0742, 0.0866, 0.0945, 0.0822, 0.0706, 0.0774], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:07:07,869 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:07:09,047 INFO [train.py:901] (2/4) Epoch 13, batch 6650, loss[loss=0.2201, simple_loss=0.2952, pruned_loss=0.07255, over 7814.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3033, pruned_loss=0.0743, over 1615442.15 frames. ], batch size: 19, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:07:16,587 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.459e+02 2.800e+02 3.637e+02 6.016e+02, threshold=5.600e+02, percent-clipped=0.0 +2023-02-06 14:07:43,986 INFO [train.py:901] (2/4) Epoch 13, batch 6700, loss[loss=0.2316, simple_loss=0.2906, pruned_loss=0.08631, over 7805.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3038, pruned_loss=0.07468, over 1612000.30 frames. ], batch size: 19, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:08:06,486 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103728.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:08:18,549 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 14:08:19,449 INFO [train.py:901] (2/4) Epoch 13, batch 6750, loss[loss=0.2214, simple_loss=0.2865, pruned_loss=0.07811, over 7241.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3038, pruned_loss=0.07437, over 1616514.16 frames. ], batch size: 16, lr: 5.80e-03, grad_scale: 16.0 +2023-02-06 14:08:24,987 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3831, 2.9781, 2.4095, 3.9285, 1.7465, 2.2584, 2.4112, 2.9175], + device='cuda:2'), covar=tensor([0.0789, 0.0821, 0.0933, 0.0273, 0.1219, 0.1264, 0.1104, 0.0919], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0205, 0.0250, 0.0208, 0.0213, 0.0249, 0.0252, 0.0213], + 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') +2023-02-06 14:08:26,126 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.551e+02 3.234e+02 3.983e+02 1.044e+03, threshold=6.469e+02, percent-clipped=6.0 +2023-02-06 14:08:26,989 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103758.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:08:44,328 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103783.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:08:53,551 INFO [train.py:901] (2/4) Epoch 13, batch 6800, loss[loss=0.2401, simple_loss=0.3034, pruned_loss=0.08838, over 7239.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3025, pruned_loss=0.07345, over 1610201.13 frames. ], batch size: 16, lr: 5.80e-03, grad_scale: 16.0 +2023-02-06 14:08:58,344 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 14:09:17,167 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103831.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:09:19,856 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7937, 1.8793, 1.6915, 2.3511, 1.0490, 1.5116, 1.6444, 1.9374], + device='cuda:2'), covar=tensor([0.0692, 0.0753, 0.0961, 0.0423, 0.1093, 0.1402, 0.0814, 0.0691], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0205, 0.0251, 0.0208, 0.0213, 0.0249, 0.0253, 0.0213], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 14:09:25,231 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103843.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:09:27,543 INFO [train.py:901] (2/4) Epoch 13, batch 6850, loss[loss=0.2477, simple_loss=0.3122, pruned_loss=0.09161, over 7704.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3016, pruned_loss=0.07334, over 1609690.38 frames. ], batch size: 18, lr: 5.80e-03, grad_scale: 16.0 +2023-02-06 14:09:33,732 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:09:34,154 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 2.670e+02 3.153e+02 3.957e+02 9.275e+02, threshold=6.306e+02, percent-clipped=2.0 +2023-02-06 14:09:40,249 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:09:44,804 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 14:09:46,353 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0183, 1.8392, 2.4765, 2.0362, 2.3540, 2.0454, 1.7221, 1.1536], + device='cuda:2'), covar=tensor([0.4531, 0.3908, 0.1390, 0.2580, 0.1864, 0.2481, 0.1805, 0.3919], + device='cuda:2'), in_proj_covar=tensor([0.0899, 0.0887, 0.0742, 0.0866, 0.0948, 0.0822, 0.0705, 0.0777], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:09:57,532 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:10:00,231 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103894.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:10:02,110 INFO [train.py:901] (2/4) Epoch 13, batch 6900, loss[loss=0.1722, simple_loss=0.2524, pruned_loss=0.04601, over 7686.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3021, pruned_loss=0.07391, over 1607562.86 frames. ], batch size: 18, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:10:35,839 INFO [train.py:901] (2/4) Epoch 13, batch 6950, loss[loss=0.2544, simple_loss=0.3179, pruned_loss=0.09539, over 8442.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3022, pruned_loss=0.07429, over 1608306.50 frames. ], batch size: 27, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:10:43,033 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.539e+02 3.074e+02 3.917e+02 9.810e+02, threshold=6.147e+02, percent-clipped=9.0 +2023-02-06 14:10:53,184 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 14:11:00,841 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4555, 1.9419, 3.2058, 1.2906, 2.3105, 1.8525, 1.6029, 2.1039], + device='cuda:2'), covar=tensor([0.1812, 0.2242, 0.0695, 0.4045, 0.1697, 0.3104, 0.2006, 0.2413], + device='cuda:2'), in_proj_covar=tensor([0.0496, 0.0537, 0.0531, 0.0588, 0.0620, 0.0559, 0.0483, 0.0613], + 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') +2023-02-06 14:11:09,727 INFO [train.py:901] (2/4) Epoch 13, batch 7000, loss[loss=0.2305, simple_loss=0.3115, pruned_loss=0.07477, over 8648.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3019, pruned_loss=0.07421, over 1607167.29 frames. ], batch size: 34, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:11:18,201 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104008.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:11:19,231 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 14:11:23,553 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:11:35,613 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9301, 2.1461, 1.6380, 2.5638, 1.3958, 1.4277, 1.8636, 2.1194], + device='cuda:2'), covar=tensor([0.0715, 0.0751, 0.1039, 0.0420, 0.1101, 0.1441, 0.0894, 0.0702], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0208, 0.0252, 0.0210, 0.0215, 0.0252, 0.0256, 0.0215], + 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') +2023-02-06 14:11:43,224 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.85 vs. limit=5.0 +2023-02-06 14:11:44,818 INFO [train.py:901] (2/4) Epoch 13, batch 7050, loss[loss=0.2957, simple_loss=0.3705, pruned_loss=0.1104, over 8617.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3021, pruned_loss=0.07394, over 1613320.93 frames. ], batch size: 31, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:11:52,798 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.385e+02 2.879e+02 3.637e+02 6.044e+02, threshold=5.759e+02, percent-clipped=0.0 +2023-02-06 14:11:58,856 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104067.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:12:01,820 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 14:12:18,830 INFO [train.py:901] (2/4) Epoch 13, batch 7100, loss[loss=0.207, simple_loss=0.2913, pruned_loss=0.06133, over 8255.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3017, pruned_loss=0.07394, over 1611607.77 frames. ], batch size: 24, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:12:21,101 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104099.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:12:22,946 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104102.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:12:37,509 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104124.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:12:39,432 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104127.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:12:46,609 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 14:12:47,798 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 14:12:52,701 INFO [train.py:901] (2/4) Epoch 13, batch 7150, loss[loss=0.2211, simple_loss=0.3073, pruned_loss=0.0675, over 8499.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3036, pruned_loss=0.07522, over 1611018.67 frames. ], batch size: 28, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:13:00,084 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.541e+02 2.991e+02 4.071e+02 7.912e+02, threshold=5.982e+02, percent-clipped=4.0 +2023-02-06 14:13:27,754 INFO [train.py:901] (2/4) Epoch 13, batch 7200, loss[loss=0.1989, simple_loss=0.2731, pruned_loss=0.06236, over 7203.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3041, pruned_loss=0.07537, over 1614409.03 frames. ], batch size: 16, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:13:42,139 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104217.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:13:55,692 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104238.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:13:58,507 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104242.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:14:01,682 INFO [train.py:901] (2/4) Epoch 13, batch 7250, loss[loss=0.296, simple_loss=0.3501, pruned_loss=0.121, over 6740.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3037, pruned_loss=0.07532, over 1605997.30 frames. ], batch size: 72, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:14:09,619 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.469e+02 3.063e+02 3.939e+02 8.277e+02, threshold=6.126e+02, percent-clipped=7.0 +2023-02-06 14:14:37,319 INFO [train.py:901] (2/4) Epoch 13, batch 7300, loss[loss=0.2023, simple_loss=0.2824, pruned_loss=0.06111, over 7194.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3035, pruned_loss=0.07512, over 1606902.85 frames. ], batch size: 16, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:14:41,423 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1621, 1.4904, 4.3835, 1.9241, 2.4374, 5.1048, 5.0474, 4.3595], + device='cuda:2'), covar=tensor([0.1197, 0.1772, 0.0282, 0.1990, 0.1176, 0.0170, 0.0494, 0.0572], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0302, 0.0264, 0.0295, 0.0280, 0.0240, 0.0361, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:15:11,548 INFO [train.py:901] (2/4) Epoch 13, batch 7350, loss[loss=0.2069, simple_loss=0.2926, pruned_loss=0.06061, over 7633.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3033, pruned_loss=0.07487, over 1607602.39 frames. ], batch size: 19, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:15:14,977 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104352.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:15:15,787 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:15:19,027 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 14:15:19,099 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.439e+02 3.043e+02 3.823e+02 6.373e+02, threshold=6.086e+02, percent-clipped=2.0 +2023-02-06 14:15:19,897 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:15:30,134 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4850, 1.9116, 3.4795, 1.3171, 2.4430, 1.8831, 1.5562, 2.4448], + device='cuda:2'), covar=tensor([0.2007, 0.2455, 0.0753, 0.4240, 0.1713, 0.3141, 0.2152, 0.2180], + device='cuda:2'), in_proj_covar=tensor([0.0500, 0.0543, 0.0535, 0.0595, 0.0623, 0.0563, 0.0490, 0.0619], + 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') +2023-02-06 14:15:32,076 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7584, 2.2923, 3.4963, 2.6169, 3.0185, 2.5306, 2.1056, 1.8337], + device='cuda:2'), covar=tensor([0.3971, 0.4711, 0.1382, 0.2984, 0.2262, 0.2380, 0.1700, 0.4844], + device='cuda:2'), in_proj_covar=tensor([0.0886, 0.0885, 0.0744, 0.0862, 0.0941, 0.0819, 0.0704, 0.0776], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:15:33,232 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 14:15:46,158 INFO [train.py:901] (2/4) Epoch 13, batch 7400, loss[loss=0.2079, simple_loss=0.2908, pruned_loss=0.06244, over 8195.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3022, pruned_loss=0.074, over 1607418.62 frames. ], batch size: 23, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:15:53,248 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 14:15:56,727 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104411.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:12,371 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0321, 1.5422, 3.3241, 1.4273, 2.2141, 3.7440, 3.7791, 3.1262], + device='cuda:2'), covar=tensor([0.1134, 0.1650, 0.0387, 0.2069, 0.1157, 0.0205, 0.0456, 0.0632], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0302, 0.0264, 0.0294, 0.0279, 0.0239, 0.0360, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:16:21,096 INFO [train.py:901] (2/4) Epoch 13, batch 7450, loss[loss=0.2383, simple_loss=0.3192, pruned_loss=0.07871, over 8460.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3028, pruned_loss=0.07462, over 1606742.06 frames. ], batch size: 27, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:16:29,258 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.488e+02 2.494e+02 3.000e+02 3.814e+02 1.100e+03, threshold=5.999e+02, percent-clipped=4.0 +2023-02-06 14:16:33,245 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 14:16:33,404 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8550, 1.8564, 4.4123, 1.9551, 2.4229, 5.1396, 5.0561, 4.3685], + device='cuda:2'), covar=tensor([0.0910, 0.1569, 0.0258, 0.1990, 0.1187, 0.0143, 0.0387, 0.0562], + device='cuda:2'), in_proj_covar=tensor([0.0268, 0.0299, 0.0263, 0.0292, 0.0277, 0.0237, 0.0357, 0.0292], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:16:35,445 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104467.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:39,562 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104473.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:40,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:55,893 INFO [train.py:901] (2/4) Epoch 13, batch 7500, loss[loss=0.2334, simple_loss=0.3229, pruned_loss=0.07194, over 8258.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3034, pruned_loss=0.07528, over 1607282.98 frames. ], batch size: 24, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:16:56,753 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104498.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:56,772 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104498.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:17:14,578 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104523.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:17:16,624 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104526.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:17:30,650 INFO [train.py:901] (2/4) Epoch 13, batch 7550, loss[loss=0.2136, simple_loss=0.2781, pruned_loss=0.0746, over 7777.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3049, pruned_loss=0.07619, over 1606004.99 frames. ], batch size: 19, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:17:37,881 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.482e+02 3.042e+02 4.105e+02 9.709e+02, threshold=6.085e+02, percent-clipped=7.0 +2023-02-06 14:18:05,105 INFO [train.py:901] (2/4) Epoch 13, batch 7600, loss[loss=0.2087, simple_loss=0.3001, pruned_loss=0.05869, over 8332.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3044, pruned_loss=0.0755, over 1609254.16 frames. ], batch size: 25, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:18:13,225 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:18:30,785 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:18:40,157 INFO [train.py:901] (2/4) Epoch 13, batch 7650, loss[loss=0.2243, simple_loss=0.2958, pruned_loss=0.07638, over 7906.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3033, pruned_loss=0.07489, over 1606225.12 frames. ], batch size: 20, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:18:47,609 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.638e+02 3.280e+02 4.340e+02 1.130e+03, threshold=6.560e+02, percent-clipped=9.0 +2023-02-06 14:19:14,835 INFO [train.py:901] (2/4) Epoch 13, batch 7700, loss[loss=0.2271, simple_loss=0.3121, pruned_loss=0.07107, over 8338.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3026, pruned_loss=0.0746, over 1605752.52 frames. ], batch size: 25, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:19:33,198 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104723.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:37,761 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 14:19:37,980 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:49,824 INFO [train.py:901] (2/4) Epoch 13, batch 7750, loss[loss=0.269, simple_loss=0.3371, pruned_loss=0.1005, over 8668.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3031, pruned_loss=0.07479, over 1604255.26 frames. ], batch size: 49, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:19:50,602 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104748.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:55,987 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:57,815 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 2.530e+02 2.944e+02 3.392e+02 9.198e+02, threshold=5.888e+02, percent-clipped=3.0 +2023-02-06 14:20:10,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1822, 1.2050, 2.3033, 1.1024, 1.9494, 2.4987, 2.6207, 2.1052], + device='cuda:2'), covar=tensor([0.1191, 0.1421, 0.0508, 0.2208, 0.0841, 0.0404, 0.0796, 0.0823], + device='cuda:2'), in_proj_covar=tensor([0.0269, 0.0304, 0.0263, 0.0293, 0.0278, 0.0238, 0.0360, 0.0294], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:20:14,010 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104782.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:20:20,306 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 14:20:24,829 INFO [train.py:901] (2/4) Epoch 13, batch 7800, loss[loss=0.2091, simple_loss=0.2975, pruned_loss=0.06031, over 8497.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3007, pruned_loss=0.07278, over 1609019.08 frames. ], batch size: 26, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:20:31,891 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:20:52,229 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9757, 1.5098, 1.6431, 1.4219, 0.9842, 1.5052, 1.7184, 1.5740], + device='cuda:2'), covar=tensor([0.0500, 0.1145, 0.1538, 0.1299, 0.0595, 0.1377, 0.0635, 0.0563], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0149, 0.0188, 0.0155, 0.0100, 0.0160, 0.0112, 0.0136], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:20:57,821 INFO [train.py:901] (2/4) Epoch 13, batch 7850, loss[loss=0.1935, simple_loss=0.274, pruned_loss=0.05651, over 7815.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3015, pruned_loss=0.07325, over 1607474.90 frames. ], batch size: 20, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:21:05,227 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.619e+02 3.074e+02 4.074e+02 1.012e+03, threshold=6.148e+02, percent-clipped=5.0 +2023-02-06 14:21:30,904 INFO [train.py:901] (2/4) Epoch 13, batch 7900, loss[loss=0.1881, simple_loss=0.2677, pruned_loss=0.05423, over 7534.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3024, pruned_loss=0.07372, over 1607790.91 frames. ], batch size: 18, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:22:04,219 INFO [train.py:901] (2/4) Epoch 13, batch 7950, loss[loss=0.2468, simple_loss=0.323, pruned_loss=0.08533, over 8339.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3027, pruned_loss=0.07355, over 1608254.36 frames. ], batch size: 25, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:22:11,301 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.521e+02 3.027e+02 3.866e+02 6.555e+02, threshold=6.053e+02, percent-clipped=2.0 +2023-02-06 14:22:37,760 INFO [train.py:901] (2/4) Epoch 13, batch 8000, loss[loss=0.2199, simple_loss=0.2861, pruned_loss=0.07685, over 8081.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3038, pruned_loss=0.07436, over 1603966.89 frames. ], batch size: 21, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:23:10,562 INFO [train.py:901] (2/4) Epoch 13, batch 8050, loss[loss=0.2232, simple_loss=0.2862, pruned_loss=0.08008, over 7925.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3033, pruned_loss=0.07494, over 1596860.94 frames. ], batch size: 20, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:23:18,070 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.417e+02 2.946e+02 3.621e+02 6.025e+02, threshold=5.892e+02, percent-clipped=0.0 +2023-02-06 14:23:23,619 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9567, 3.9175, 2.4403, 2.7493, 3.0817, 2.4109, 2.8256, 3.1208], + device='cuda:2'), covar=tensor([0.1701, 0.0305, 0.1119, 0.0824, 0.0627, 0.1316, 0.1055, 0.0989], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0236, 0.0318, 0.0296, 0.0298, 0.0321, 0.0338, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:23:50,198 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 14:23:54,152 INFO [train.py:901] (2/4) Epoch 14, batch 0, loss[loss=0.2055, simple_loss=0.2811, pruned_loss=0.06502, over 8085.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2811, pruned_loss=0.06502, over 8085.00 frames. ], batch size: 21, lr: 5.56e-03, grad_scale: 8.0 +2023-02-06 14:23:54,152 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 14:24:01,292 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5383, 1.7647, 2.6126, 1.3037, 1.9883, 1.8453, 1.5494, 1.9024], + device='cuda:2'), covar=tensor([0.1744, 0.2448, 0.0878, 0.4174, 0.1742, 0.3060, 0.2134, 0.2035], + device='cuda:2'), in_proj_covar=tensor([0.0499, 0.0545, 0.0539, 0.0600, 0.0623, 0.0566, 0.0493, 0.0619], + 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') +2023-02-06 14:24:05,197 INFO [train.py:935] (2/4) Epoch 14, validation: loss=0.184, simple_loss=0.2839, pruned_loss=0.04201, over 944034.00 frames. +2023-02-06 14:24:05,197 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 14:24:21,244 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 14:24:38,588 INFO [train.py:901] (2/4) Epoch 14, batch 50, loss[loss=0.2323, simple_loss=0.3111, pruned_loss=0.07677, over 8347.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.309, pruned_loss=0.07643, over 370019.40 frames. ], batch size: 26, lr: 5.56e-03, grad_scale: 8.0 +2023-02-06 14:24:54,770 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 14:24:58,160 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.684e+02 3.092e+02 3.835e+02 7.852e+02, threshold=6.183e+02, percent-clipped=3.0 +2023-02-06 14:25:14,421 INFO [train.py:901] (2/4) Epoch 14, batch 100, loss[loss=0.2358, simple_loss=0.3168, pruned_loss=0.07736, over 8491.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3055, pruned_loss=0.07428, over 647109.44 frames. ], batch size: 26, lr: 5.56e-03, grad_scale: 8.0 +2023-02-06 14:25:17,792 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 14:25:48,648 INFO [train.py:901] (2/4) Epoch 14, batch 150, loss[loss=0.1824, simple_loss=0.2706, pruned_loss=0.04708, over 7931.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3065, pruned_loss=0.07534, over 858909.02 frames. ], batch size: 20, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:26:08,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.384e+02 2.990e+02 3.742e+02 5.781e+02, threshold=5.980e+02, percent-clipped=0.0 +2023-02-06 14:26:23,058 INFO [train.py:901] (2/4) Epoch 14, batch 200, loss[loss=0.2304, simple_loss=0.3085, pruned_loss=0.07613, over 8627.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3062, pruned_loss=0.07535, over 1030769.11 frames. ], batch size: 49, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:26:58,947 INFO [train.py:901] (2/4) Epoch 14, batch 250, loss[loss=0.2162, simple_loss=0.3057, pruned_loss=0.06333, over 8113.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3056, pruned_loss=0.07552, over 1159766.86 frames. ], batch size: 23, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:27:07,613 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 14:27:15,937 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 14:27:18,043 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.546e+02 3.157e+02 4.204e+02 9.163e+02, threshold=6.313e+02, percent-clipped=6.0 +2023-02-06 14:27:33,656 INFO [train.py:901] (2/4) Epoch 14, batch 300, loss[loss=0.2404, simple_loss=0.3249, pruned_loss=0.07792, over 8715.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.305, pruned_loss=0.07542, over 1259129.51 frames. ], batch size: 34, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:27:52,149 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5455, 4.4620, 3.9973, 1.8553, 3.9434, 4.1569, 4.0478, 3.8187], + device='cuda:2'), covar=tensor([0.0757, 0.0629, 0.1137, 0.5291, 0.0872, 0.0871, 0.1341, 0.0923], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0391, 0.0402, 0.0502, 0.0394, 0.0393, 0.0382, 0.0344], + 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') +2023-02-06 14:27:52,820 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:27:52,839 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7203, 1.4031, 2.7988, 1.3671, 2.0749, 3.0286, 3.1209, 2.5606], + device='cuda:2'), covar=tensor([0.1081, 0.1546, 0.0404, 0.1991, 0.0902, 0.0312, 0.0666, 0.0680], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0304, 0.0266, 0.0294, 0.0280, 0.0242, 0.0363, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:28:09,626 INFO [train.py:901] (2/4) Epoch 14, batch 350, loss[loss=0.2129, simple_loss=0.273, pruned_loss=0.0764, over 7252.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3041, pruned_loss=0.07426, over 1338583.95 frames. ], batch size: 16, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:28:28,598 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.437e+02 2.818e+02 3.446e+02 5.751e+02, threshold=5.636e+02, percent-clipped=0.0 +2023-02-06 14:28:43,599 INFO [train.py:901] (2/4) Epoch 14, batch 400, loss[loss=0.2557, simple_loss=0.3329, pruned_loss=0.08922, over 8291.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.305, pruned_loss=0.07463, over 1403742.19 frames. ], batch size: 23, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:29:00,989 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:29:04,408 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2457, 2.6517, 2.1077, 3.6319, 1.7025, 1.8961, 2.4121, 2.9734], + device='cuda:2'), covar=tensor([0.0772, 0.0831, 0.0959, 0.0324, 0.1131, 0.1334, 0.0995, 0.0749], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0207, 0.0252, 0.0212, 0.0213, 0.0254, 0.0259, 0.0216], + 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') +2023-02-06 14:29:13,227 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:29:13,562 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 14:29:20,748 INFO [train.py:901] (2/4) Epoch 14, batch 450, loss[loss=0.2509, simple_loss=0.3339, pruned_loss=0.08395, over 8332.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3041, pruned_loss=0.07363, over 1456075.99 frames. ], batch size: 26, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:29:40,047 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.497e+02 2.804e+02 3.770e+02 6.336e+02, threshold=5.609e+02, percent-clipped=1.0 +2023-02-06 14:29:55,211 INFO [train.py:901] (2/4) Epoch 14, batch 500, loss[loss=0.2248, simple_loss=0.307, pruned_loss=0.0713, over 8323.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3043, pruned_loss=0.07418, over 1491129.82 frames. ], batch size: 25, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:30:22,228 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9278, 1.5557, 2.1102, 1.8470, 1.9947, 1.8821, 1.6324, 0.7751], + device='cuda:2'), covar=tensor([0.4681, 0.4233, 0.1509, 0.2769, 0.2099, 0.2446, 0.1795, 0.4246], + device='cuda:2'), in_proj_covar=tensor([0.0894, 0.0887, 0.0743, 0.0868, 0.0947, 0.0819, 0.0704, 0.0778], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:30:29,395 INFO [train.py:901] (2/4) Epoch 14, batch 550, loss[loss=0.2565, simple_loss=0.3349, pruned_loss=0.0891, over 8629.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3042, pruned_loss=0.07408, over 1518922.39 frames. ], batch size: 49, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:30:50,293 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.442e+02 2.933e+02 3.700e+02 8.163e+02, threshold=5.867e+02, percent-clipped=3.0 +2023-02-06 14:30:56,676 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3355, 2.1626, 1.5258, 1.8915, 1.7209, 1.3687, 1.5503, 1.6797], + device='cuda:2'), covar=tensor([0.1302, 0.0416, 0.1275, 0.0578, 0.0751, 0.1462, 0.1063, 0.0938], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0236, 0.0318, 0.0295, 0.0299, 0.0320, 0.0339, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:31:05,191 INFO [train.py:901] (2/4) Epoch 14, batch 600, loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08528, over 8448.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.3037, pruned_loss=0.07337, over 1544779.58 frames. ], batch size: 25, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:31:18,468 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 14:31:39,813 INFO [train.py:901] (2/4) Epoch 14, batch 650, loss[loss=0.2032, simple_loss=0.2874, pruned_loss=0.05946, over 8261.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3037, pruned_loss=0.07358, over 1559565.02 frames. ], batch size: 24, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:31:44,178 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7278, 3.0274, 2.6982, 4.1241, 1.6513, 2.2557, 2.5682, 3.2670], + device='cuda:2'), covar=tensor([0.0665, 0.0839, 0.0869, 0.0271, 0.1237, 0.1333, 0.1011, 0.0755], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0205, 0.0252, 0.0211, 0.0213, 0.0253, 0.0258, 0.0217], + 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') +2023-02-06 14:31:54,393 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:32:01,338 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.402e+02 3.000e+02 3.711e+02 7.109e+02, threshold=6.000e+02, percent-clipped=4.0 +2023-02-06 14:32:17,353 INFO [train.py:901] (2/4) Epoch 14, batch 700, loss[loss=0.2242, simple_loss=0.3144, pruned_loss=0.06705, over 8473.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3013, pruned_loss=0.07247, over 1569768.59 frames. ], batch size: 25, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:32:37,870 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:32:42,742 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105817.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:32:51,388 INFO [train.py:901] (2/4) Epoch 14, batch 750, loss[loss=0.2319, simple_loss=0.3254, pruned_loss=0.06915, over 8287.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3021, pruned_loss=0.07288, over 1581915.17 frames. ], batch size: 23, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:33:03,873 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:33:06,461 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 14:33:06,989 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 14:33:11,294 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.459e+02 2.898e+02 3.725e+02 7.154e+02, threshold=5.796e+02, percent-clipped=4.0 +2023-02-06 14:33:12,182 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9656, 1.6426, 1.7330, 1.5100, 1.0765, 1.5789, 1.7099, 1.7214], + device='cuda:2'), covar=tensor([0.0502, 0.1104, 0.1560, 0.1344, 0.0613, 0.1371, 0.0669, 0.0577], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0157, 0.0101, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:33:15,491 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:33:16,040 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 14:33:16,240 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:33:18,463 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.68 vs. limit=5.0 +2023-02-06 14:33:26,359 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 14:33:27,173 INFO [train.py:901] (2/4) Epoch 14, batch 800, loss[loss=0.2451, simple_loss=0.3248, pruned_loss=0.08271, over 8131.00 frames. ], tot_loss[loss=0.224, simple_loss=0.302, pruned_loss=0.07303, over 1589067.98 frames. ], batch size: 22, lr: 5.54e-03, grad_scale: 16.0 +2023-02-06 14:33:27,419 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5505, 1.7633, 2.7585, 1.3133, 1.9156, 1.8579, 1.6114, 1.8681], + device='cuda:2'), covar=tensor([0.1763, 0.2249, 0.0813, 0.4042, 0.1801, 0.2966, 0.1823, 0.2121], + device='cuda:2'), in_proj_covar=tensor([0.0498, 0.0543, 0.0535, 0.0598, 0.0625, 0.0562, 0.0492, 0.0619], + 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') +2023-02-06 14:34:02,179 INFO [train.py:901] (2/4) Epoch 14, batch 850, loss[loss=0.2696, simple_loss=0.3425, pruned_loss=0.09834, over 8466.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3023, pruned_loss=0.07325, over 1598067.68 frames. ], batch size: 25, lr: 5.54e-03, grad_scale: 16.0 +2023-02-06 14:34:20,960 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.477e+02 2.961e+02 4.061e+02 6.411e+02, threshold=5.921e+02, percent-clipped=4.0 +2023-02-06 14:34:24,619 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105963.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:34:36,587 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:34:37,082 INFO [train.py:901] (2/4) Epoch 14, batch 900, loss[loss=0.2469, simple_loss=0.3288, pruned_loss=0.08247, over 8452.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.304, pruned_loss=0.07437, over 1604300.32 frames. ], batch size: 29, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:35:02,679 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1357, 2.4917, 3.0631, 1.7334, 3.1801, 1.7112, 1.3972, 2.0230], + device='cuda:2'), covar=tensor([0.0668, 0.0286, 0.0168, 0.0573, 0.0333, 0.0750, 0.0700, 0.0472], + device='cuda:2'), in_proj_covar=tensor([0.0409, 0.0344, 0.0299, 0.0405, 0.0333, 0.0493, 0.0367, 0.0376], + device='cuda:2'), out_proj_covar=tensor([1.1399e-04, 9.3554e-05, 8.1063e-05, 1.1089e-04, 9.1507e-05, 1.4540e-04, + 1.0260e-04, 1.0377e-04], device='cuda:2') +2023-02-06 14:35:14,900 INFO [train.py:901] (2/4) Epoch 14, batch 950, loss[loss=0.2115, simple_loss=0.2861, pruned_loss=0.06842, over 8538.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3036, pruned_loss=0.07418, over 1602530.72 frames. ], batch size: 28, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:35:33,973 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.617e+02 3.202e+02 4.119e+02 6.844e+02, threshold=6.403e+02, percent-clipped=3.0 +2023-02-06 14:35:38,945 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 14:35:49,296 INFO [train.py:901] (2/4) Epoch 14, batch 1000, loss[loss=0.1869, simple_loss=0.2608, pruned_loss=0.05653, over 7548.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3023, pruned_loss=0.07318, over 1601169.47 frames. ], batch size: 18, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:36:00,600 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:14,287 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 14:36:20,059 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106121.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:26,960 INFO [train.py:901] (2/4) Epoch 14, batch 1050, loss[loss=0.1661, simple_loss=0.2469, pruned_loss=0.04268, over 7439.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3018, pruned_loss=0.07272, over 1604591.56 frames. ], batch size: 17, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:36:26,968 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 14:36:37,984 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:43,496 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:46,252 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.417e+02 2.951e+02 3.593e+02 9.096e+02, threshold=5.903e+02, percent-clipped=2.0 +2023-02-06 14:36:48,430 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:48,710 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 14:37:01,621 INFO [train.py:901] (2/4) Epoch 14, batch 1100, loss[loss=0.2064, simple_loss=0.2918, pruned_loss=0.06055, over 8245.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3026, pruned_loss=0.07322, over 1611321.63 frames. ], batch size: 24, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:37:27,046 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9151, 1.6108, 2.0609, 1.8525, 1.9699, 1.8752, 1.6201, 0.8031], + device='cuda:2'), covar=tensor([0.4736, 0.4090, 0.1504, 0.2715, 0.2079, 0.2528, 0.1787, 0.4086], + device='cuda:2'), in_proj_covar=tensor([0.0899, 0.0893, 0.0747, 0.0871, 0.0955, 0.0826, 0.0707, 0.0782], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:37:29,713 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:35,898 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 14:37:38,717 INFO [train.py:901] (2/4) Epoch 14, batch 1150, loss[loss=0.2163, simple_loss=0.3025, pruned_loss=0.06504, over 8187.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3028, pruned_loss=0.07323, over 1613318.87 frames. ], batch size: 23, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:37:42,398 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106235.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:46,996 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6479, 1.4047, 4.7786, 1.7840, 4.2401, 3.9400, 4.3053, 4.1724], + device='cuda:2'), covar=tensor([0.0442, 0.4627, 0.0460, 0.3695, 0.0980, 0.0879, 0.0509, 0.0558], + device='cuda:2'), in_proj_covar=tensor([0.0525, 0.0588, 0.0621, 0.0557, 0.0636, 0.0541, 0.0531, 0.0600], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:37:49,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106244.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:53,384 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 14:37:58,398 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.475e+02 3.133e+02 3.919e+02 6.906e+02, threshold=6.266e+02, percent-clipped=3.0 +2023-02-06 14:37:59,996 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106260.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:06,201 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106269.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:07,700 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2999, 1.8990, 2.7342, 2.2075, 2.5822, 2.1793, 1.8689, 1.2014], + device='cuda:2'), covar=tensor([0.4579, 0.4388, 0.1363, 0.3006, 0.2039, 0.2637, 0.1729, 0.4808], + device='cuda:2'), in_proj_covar=tensor([0.0897, 0.0890, 0.0742, 0.0867, 0.0950, 0.0821, 0.0704, 0.0778], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:38:10,869 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:13,354 INFO [train.py:901] (2/4) Epoch 14, batch 1200, loss[loss=0.2202, simple_loss=0.2933, pruned_loss=0.07358, over 8255.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.303, pruned_loss=0.0737, over 1614924.53 frames. ], batch size: 24, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:38:17,553 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:47,945 INFO [train.py:901] (2/4) Epoch 14, batch 1250, loss[loss=0.203, simple_loss=0.2888, pruned_loss=0.05866, over 8565.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3044, pruned_loss=0.0746, over 1617133.82 frames. ], batch size: 31, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:38:51,204 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 14:38:54,191 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5800, 4.5385, 4.0764, 2.0092, 4.0195, 4.1893, 4.1816, 3.8268], + device='cuda:2'), covar=tensor([0.0666, 0.0583, 0.1051, 0.5049, 0.0803, 0.0842, 0.1159, 0.0847], + device='cuda:2'), in_proj_covar=tensor([0.0474, 0.0393, 0.0399, 0.0498, 0.0392, 0.0394, 0.0381, 0.0341], + 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') +2023-02-06 14:39:05,943 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106354.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:39:08,463 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.539e+02 3.303e+02 4.386e+02 1.450e+03, threshold=6.607e+02, percent-clipped=4.0 +2023-02-06 14:39:24,632 INFO [train.py:901] (2/4) Epoch 14, batch 1300, loss[loss=0.2113, simple_loss=0.2828, pruned_loss=0.06993, over 7558.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3027, pruned_loss=0.0737, over 1613526.22 frames. ], batch size: 18, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:39:58,995 INFO [train.py:901] (2/4) Epoch 14, batch 1350, loss[loss=0.2139, simple_loss=0.3075, pruned_loss=0.06012, over 8357.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3019, pruned_loss=0.07322, over 1614510.47 frames. ], batch size: 24, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:40:05,435 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:40:17,483 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-06 14:40:19,202 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.550e+02 3.060e+02 3.665e+02 8.767e+02, threshold=6.121e+02, percent-clipped=1.0 +2023-02-06 14:40:29,788 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:40:35,138 INFO [train.py:901] (2/4) Epoch 14, batch 1400, loss[loss=0.2308, simple_loss=0.3058, pruned_loss=0.0779, over 8342.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3017, pruned_loss=0.07311, over 1617659.91 frames. ], batch size: 49, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:41:07,398 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106525.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:10,541 INFO [train.py:901] (2/4) Epoch 14, batch 1450, loss[loss=0.2598, simple_loss=0.3348, pruned_loss=0.09237, over 8592.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3021, pruned_loss=0.07344, over 1619661.31 frames. ], batch size: 31, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:41:11,263 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 14:41:12,184 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:24,664 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106550.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:27,432 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106554.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:29,529 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:29,959 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.546e+02 3.123e+02 4.151e+02 8.254e+02, threshold=6.246e+02, percent-clipped=6.0 +2023-02-06 14:41:47,572 INFO [train.py:901] (2/4) Epoch 14, batch 1500, loss[loss=0.2464, simple_loss=0.3196, pruned_loss=0.08655, over 8030.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3042, pruned_loss=0.07475, over 1619013.94 frames. ], batch size: 22, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:42:22,568 INFO [train.py:901] (2/4) Epoch 14, batch 1550, loss[loss=0.2268, simple_loss=0.3087, pruned_loss=0.07243, over 8323.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3049, pruned_loss=0.07505, over 1618815.54 frames. ], batch size: 25, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:42:22,651 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:42:41,334 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.593e+02 3.196e+02 4.114e+02 8.054e+02, threshold=6.391e+02, percent-clipped=4.0 +2023-02-06 14:42:52,899 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0999, 1.6568, 3.4969, 1.4849, 2.3805, 3.7350, 3.8793, 3.2416], + device='cuda:2'), covar=tensor([0.1057, 0.1564, 0.0291, 0.2015, 0.0968, 0.0243, 0.0396, 0.0591], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0303, 0.0263, 0.0293, 0.0281, 0.0242, 0.0362, 0.0293], + 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') +2023-02-06 14:42:56,718 INFO [train.py:901] (2/4) Epoch 14, batch 1600, loss[loss=0.2815, simple_loss=0.3435, pruned_loss=0.1098, over 8354.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3053, pruned_loss=0.07533, over 1618715.86 frames. ], batch size: 26, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:43:10,339 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:43:32,335 INFO [train.py:901] (2/4) Epoch 14, batch 1650, loss[loss=0.2074, simple_loss=0.2845, pruned_loss=0.06518, over 8080.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3048, pruned_loss=0.07503, over 1614847.85 frames. ], batch size: 21, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:43:35,230 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6613, 1.5704, 1.9662, 1.6943, 0.9730, 1.7221, 2.2619, 1.8692], + device='cuda:2'), covar=tensor([0.0456, 0.1216, 0.1592, 0.1302, 0.0608, 0.1407, 0.0594, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0157, 0.0101, 0.0162, 0.0113, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:43:42,581 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:43:51,907 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.493e+02 3.038e+02 4.078e+02 1.080e+03, threshold=6.076e+02, percent-clipped=3.0 +2023-02-06 14:43:52,348 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.16 vs. limit=5.0 +2023-02-06 14:43:53,531 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5954, 1.9837, 2.1191, 1.1827, 2.2290, 1.4785, 0.4675, 1.8236], + device='cuda:2'), covar=tensor([0.0395, 0.0244, 0.0167, 0.0415, 0.0254, 0.0625, 0.0607, 0.0213], + device='cuda:2'), in_proj_covar=tensor([0.0412, 0.0349, 0.0304, 0.0406, 0.0338, 0.0492, 0.0370, 0.0375], + device='cuda:2'), out_proj_covar=tensor([1.1478e-04, 9.4922e-05, 8.2619e-05, 1.1079e-04, 9.2894e-05, 1.4505e-04, + 1.0328e-04, 1.0334e-04], device='cuda:2') +2023-02-06 14:44:06,421 INFO [train.py:901] (2/4) Epoch 14, batch 1700, loss[loss=0.2152, simple_loss=0.2978, pruned_loss=0.06625, over 7807.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3038, pruned_loss=0.07448, over 1609962.39 frames. ], batch size: 20, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:44:28,260 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:44:31,566 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:44:33,569 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106816.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:44:43,437 INFO [train.py:901] (2/4) Epoch 14, batch 1750, loss[loss=0.2206, simple_loss=0.2993, pruned_loss=0.07093, over 8026.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3037, pruned_loss=0.07448, over 1610041.57 frames. ], batch size: 22, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:44:47,857 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106835.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:45:04,129 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.358e+02 2.865e+02 3.554e+02 7.426e+02, threshold=5.730e+02, percent-clipped=3.0 +2023-02-06 14:45:18,429 INFO [train.py:901] (2/4) Epoch 14, batch 1800, loss[loss=0.2075, simple_loss=0.2871, pruned_loss=0.0639, over 8250.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3043, pruned_loss=0.0748, over 1609732.79 frames. ], batch size: 24, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:45:54,581 INFO [train.py:901] (2/4) Epoch 14, batch 1850, loss[loss=0.2086, simple_loss=0.2777, pruned_loss=0.06971, over 7668.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3044, pruned_loss=0.07518, over 1610353.41 frames. ], batch size: 19, lr: 5.51e-03, grad_scale: 4.0 +2023-02-06 14:45:55,508 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:46:16,023 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.570e+02 3.068e+02 3.847e+02 1.325e+03, threshold=6.136e+02, percent-clipped=4.0 +2023-02-06 14:46:29,524 INFO [train.py:901] (2/4) Epoch 14, batch 1900, loss[loss=0.1903, simple_loss=0.2666, pruned_loss=0.05705, over 7832.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3028, pruned_loss=0.07461, over 1605749.04 frames. ], batch size: 20, lr: 5.51e-03, grad_scale: 4.0 +2023-02-06 14:46:37,375 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 14:46:43,881 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:46:47,076 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 14:46:59,782 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 14:47:01,263 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:47:03,727 INFO [train.py:901] (2/4) Epoch 14, batch 1950, loss[loss=0.1785, simple_loss=0.2503, pruned_loss=0.05329, over 7431.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3022, pruned_loss=0.07444, over 1604496.04 frames. ], batch size: 17, lr: 5.51e-03, grad_scale: 4.0 +2023-02-06 14:47:19,867 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 14:47:26,055 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.386e+02 2.840e+02 3.483e+02 6.138e+02, threshold=5.681e+02, percent-clipped=1.0 +2023-02-06 14:47:31,988 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107069.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:47:39,093 INFO [train.py:901] (2/4) Epoch 14, batch 2000, loss[loss=0.2394, simple_loss=0.3052, pruned_loss=0.0868, over 6389.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.3023, pruned_loss=0.07407, over 1603396.45 frames. ], batch size: 14, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:47:48,656 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:48:12,604 INFO [train.py:901] (2/4) Epoch 14, batch 2050, loss[loss=0.2095, simple_loss=0.2954, pruned_loss=0.06175, over 8340.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3027, pruned_loss=0.07418, over 1610141.93 frames. ], batch size: 26, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:48:33,216 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107158.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:48:34,368 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.411e+02 3.055e+02 3.713e+02 7.642e+02, threshold=6.109e+02, percent-clipped=4.0 +2023-02-06 14:48:42,098 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107170.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:48:49,407 INFO [train.py:901] (2/4) Epoch 14, batch 2100, loss[loss=0.2581, simple_loss=0.3451, pruned_loss=0.08562, over 8185.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3026, pruned_loss=0.07365, over 1609901.00 frames. ], batch size: 23, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:48:54,367 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107187.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:49:11,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:49:21,509 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4850, 1.8874, 3.4267, 1.3599, 2.4440, 2.0617, 1.6210, 2.3887], + device='cuda:2'), covar=tensor([0.2125, 0.2661, 0.0588, 0.4406, 0.1597, 0.2945, 0.2175, 0.2135], + device='cuda:2'), in_proj_covar=tensor([0.0498, 0.0544, 0.0537, 0.0594, 0.0621, 0.0562, 0.0488, 0.0616], + 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') +2023-02-06 14:49:23,201 INFO [train.py:901] (2/4) Epoch 14, batch 2150, loss[loss=0.233, simple_loss=0.3229, pruned_loss=0.07149, over 8341.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3023, pruned_loss=0.07301, over 1610425.27 frames. ], batch size: 25, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:49:44,431 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.615e+02 3.041e+02 3.823e+02 8.460e+02, threshold=6.081e+02, percent-clipped=1.0 +2023-02-06 14:49:58,901 INFO [train.py:901] (2/4) Epoch 14, batch 2200, loss[loss=0.2256, simple_loss=0.3034, pruned_loss=0.07387, over 8354.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3014, pruned_loss=0.0724, over 1611244.57 frames. ], batch size: 26, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:50:34,527 INFO [train.py:901] (2/4) Epoch 14, batch 2250, loss[loss=0.1735, simple_loss=0.2608, pruned_loss=0.04307, over 7906.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3019, pruned_loss=0.07246, over 1615117.35 frames. ], batch size: 20, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:50:54,554 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.599e+02 3.319e+02 4.071e+02 1.027e+03, threshold=6.637e+02, percent-clipped=7.0 +2023-02-06 14:51:08,893 INFO [train.py:901] (2/4) Epoch 14, batch 2300, loss[loss=0.2331, simple_loss=0.3198, pruned_loss=0.07319, over 8543.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3026, pruned_loss=0.07298, over 1616583.74 frames. ], batch size: 31, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:51:44,730 INFO [train.py:901] (2/4) Epoch 14, batch 2350, loss[loss=0.2065, simple_loss=0.2736, pruned_loss=0.06968, over 7698.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3006, pruned_loss=0.07229, over 1613611.30 frames. ], batch size: 18, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:52:00,477 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8909, 1.6109, 2.0499, 1.8074, 1.8711, 1.8603, 1.5882, 0.7732], + device='cuda:2'), covar=tensor([0.4090, 0.3803, 0.1385, 0.2350, 0.1936, 0.2198, 0.1535, 0.3747], + device='cuda:2'), in_proj_covar=tensor([0.0892, 0.0893, 0.0745, 0.0866, 0.0953, 0.0822, 0.0709, 0.0776], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:52:04,939 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.358e+02 2.889e+02 3.449e+02 7.134e+02, threshold=5.779e+02, percent-clipped=1.0 +2023-02-06 14:52:18,373 INFO [train.py:901] (2/4) Epoch 14, batch 2400, loss[loss=0.198, simple_loss=0.272, pruned_loss=0.06199, over 7805.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3004, pruned_loss=0.07221, over 1613064.84 frames. ], batch size: 20, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:52:28,437 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 14:52:34,432 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107502.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:52:43,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107514.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:52:53,748 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107528.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:52:54,996 INFO [train.py:901] (2/4) Epoch 14, batch 2450, loss[loss=0.2295, simple_loss=0.3159, pruned_loss=0.07152, over 8362.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3012, pruned_loss=0.07273, over 1614549.00 frames. ], batch size: 24, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:53:16,528 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.477e+02 3.089e+02 4.011e+02 1.178e+03, threshold=6.179e+02, percent-clipped=8.0 +2023-02-06 14:53:29,783 INFO [train.py:901] (2/4) Epoch 14, batch 2500, loss[loss=0.2028, simple_loss=0.2811, pruned_loss=0.06222, over 7548.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3004, pruned_loss=0.07238, over 1615511.53 frames. ], batch size: 18, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:53:54,962 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5090, 1.4674, 1.7818, 1.4033, 1.1313, 1.7905, 0.1953, 1.2380], + device='cuda:2'), covar=tensor([0.2197, 0.1486, 0.0532, 0.1048, 0.3339, 0.0501, 0.2465, 0.1433], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0177, 0.0108, 0.0218, 0.0261, 0.0112, 0.0162, 0.0174], + 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') +2023-02-06 14:53:55,571 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107617.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:54:03,470 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107629.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:54:03,973 INFO [train.py:901] (2/4) Epoch 14, batch 2550, loss[loss=0.2188, simple_loss=0.3128, pruned_loss=0.06236, over 8035.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3001, pruned_loss=0.07219, over 1614822.42 frames. ], batch size: 22, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:54:12,187 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7390, 1.3149, 1.7403, 1.2148, 0.9782, 1.4566, 2.2281, 2.2602], + device='cuda:2'), covar=tensor([0.0471, 0.1767, 0.2334, 0.1821, 0.0691, 0.2076, 0.0725, 0.0623], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0156, 0.0100, 0.0162, 0.0113, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 14:54:17,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7860, 2.0032, 2.1985, 1.2712, 2.2381, 1.5190, 0.9071, 1.9328], + device='cuda:2'), covar=tensor([0.0451, 0.0253, 0.0184, 0.0453, 0.0336, 0.0597, 0.0622, 0.0238], + device='cuda:2'), in_proj_covar=tensor([0.0413, 0.0353, 0.0309, 0.0409, 0.0341, 0.0499, 0.0375, 0.0381], + device='cuda:2'), out_proj_covar=tensor([1.1511e-04, 9.5815e-05, 8.3996e-05, 1.1152e-04, 9.3479e-05, 1.4706e-04, + 1.0461e-04, 1.0489e-04], device='cuda:2') +2023-02-06 14:54:26,274 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.642e+02 3.253e+02 4.518e+02 1.030e+03, threshold=6.506e+02, percent-clipped=5.0 +2023-02-06 14:54:37,781 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:54:39,572 INFO [train.py:901] (2/4) Epoch 14, batch 2600, loss[loss=0.2389, simple_loss=0.3105, pruned_loss=0.0836, over 8467.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.301, pruned_loss=0.07295, over 1618745.03 frames. ], batch size: 27, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:54:45,902 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9141, 2.4442, 3.3941, 1.9957, 1.7375, 3.3594, 0.5886, 2.0000], + device='cuda:2'), covar=tensor([0.2048, 0.1701, 0.0371, 0.2358, 0.3600, 0.0283, 0.3213, 0.1884], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0178, 0.0107, 0.0219, 0.0261, 0.0112, 0.0163, 0.0175], + 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') +2023-02-06 14:55:12,835 INFO [train.py:901] (2/4) Epoch 14, batch 2650, loss[loss=0.2309, simple_loss=0.3168, pruned_loss=0.07245, over 8721.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3017, pruned_loss=0.07335, over 1620715.57 frames. ], batch size: 30, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:55:30,855 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:55:34,865 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.443e+02 2.980e+02 3.881e+02 9.981e+02, threshold=5.960e+02, percent-clipped=6.0 +2023-02-06 14:55:38,610 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8893, 2.1927, 2.3313, 1.4396, 2.4275, 1.7111, 0.8025, 2.1190], + device='cuda:2'), covar=tensor([0.0423, 0.0231, 0.0178, 0.0405, 0.0279, 0.0644, 0.0582, 0.0222], + device='cuda:2'), in_proj_covar=tensor([0.0415, 0.0353, 0.0308, 0.0411, 0.0341, 0.0499, 0.0376, 0.0381], + device='cuda:2'), out_proj_covar=tensor([1.1548e-04, 9.5921e-05, 8.3857e-05, 1.1200e-04, 9.3445e-05, 1.4695e-04, + 1.0487e-04, 1.0510e-04], device='cuda:2') +2023-02-06 14:55:49,934 INFO [train.py:901] (2/4) Epoch 14, batch 2700, loss[loss=0.2161, simple_loss=0.3074, pruned_loss=0.06235, over 8566.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3009, pruned_loss=0.07284, over 1619317.39 frames. ], batch size: 31, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:56:04,454 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5782, 2.8675, 1.8722, 2.3147, 2.3889, 1.5435, 2.0960, 2.2193], + device='cuda:2'), covar=tensor([0.1527, 0.0310, 0.1054, 0.0670, 0.0648, 0.1336, 0.1073, 0.0948], + device='cuda:2'), in_proj_covar=tensor([0.0343, 0.0230, 0.0316, 0.0295, 0.0296, 0.0319, 0.0336, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:56:23,700 INFO [train.py:901] (2/4) Epoch 14, batch 2750, loss[loss=0.2139, simple_loss=0.2835, pruned_loss=0.07217, over 7705.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3011, pruned_loss=0.0728, over 1618431.04 frames. ], batch size: 18, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:56:44,689 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.404e+02 2.918e+02 3.592e+02 1.217e+03, threshold=5.837e+02, percent-clipped=4.0 +2023-02-06 14:56:53,331 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107872.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:56:54,144 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107873.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:56:59,450 INFO [train.py:901] (2/4) Epoch 14, batch 2800, loss[loss=0.224, simple_loss=0.2998, pruned_loss=0.07414, over 7968.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.302, pruned_loss=0.07284, over 1619109.62 frames. ], batch size: 21, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:57:03,899 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107885.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:57:12,675 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107898.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:57:20,776 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107910.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:57:33,814 INFO [train.py:901] (2/4) Epoch 14, batch 2850, loss[loss=0.2645, simple_loss=0.3479, pruned_loss=0.09053, over 8356.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.302, pruned_loss=0.07279, over 1617487.17 frames. ], batch size: 24, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:57:40,958 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2878, 1.9273, 2.8180, 2.2265, 2.7610, 2.1673, 1.7997, 1.4603], + device='cuda:2'), covar=tensor([0.4265, 0.4226, 0.1357, 0.2911, 0.1788, 0.2358, 0.1701, 0.4257], + device='cuda:2'), in_proj_covar=tensor([0.0898, 0.0897, 0.0741, 0.0867, 0.0952, 0.0827, 0.0711, 0.0780], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 14:57:54,103 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.478e+02 3.087e+02 3.919e+02 8.173e+02, threshold=6.173e+02, percent-clipped=5.0 +2023-02-06 14:58:08,144 INFO [train.py:901] (2/4) Epoch 14, batch 2900, loss[loss=0.2374, simple_loss=0.3018, pruned_loss=0.08643, over 7804.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3018, pruned_loss=0.07282, over 1615953.67 frames. ], batch size: 20, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:58:12,786 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:58:27,879 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 14:58:38,640 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:58:44,575 INFO [train.py:901] (2/4) Epoch 14, batch 2950, loss[loss=0.1949, simple_loss=0.2626, pruned_loss=0.06356, over 7787.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3009, pruned_loss=0.07259, over 1612515.31 frames. ], batch size: 19, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:58:55,741 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 14:59:04,838 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.696e+02 3.199e+02 4.019e+02 8.231e+02, threshold=6.398e+02, percent-clipped=3.0 +2023-02-06 14:59:05,061 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9432, 3.6625, 2.2135, 2.8521, 2.9009, 2.0221, 2.5885, 2.7574], + device='cuda:2'), covar=tensor([0.1647, 0.0313, 0.1094, 0.0692, 0.0633, 0.1389, 0.1101, 0.1102], + device='cuda:2'), in_proj_covar=tensor([0.0342, 0.0228, 0.0318, 0.0296, 0.0295, 0.0320, 0.0339, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 14:59:18,172 INFO [train.py:901] (2/4) Epoch 14, batch 3000, loss[loss=0.2366, simple_loss=0.3141, pruned_loss=0.07951, over 8494.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3022, pruned_loss=0.07318, over 1617278.59 frames. ], batch size: 26, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:59:18,173 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 14:59:30,508 INFO [train.py:935] (2/4) Epoch 14, validation: loss=0.1827, simple_loss=0.283, pruned_loss=0.04121, over 944034.00 frames. +2023-02-06 14:59:30,510 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 14:59:37,800 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0814, 1.2668, 1.2583, 0.5996, 1.2313, 0.9657, 0.0767, 1.1973], + device='cuda:2'), covar=tensor([0.0304, 0.0286, 0.0243, 0.0442, 0.0340, 0.0791, 0.0636, 0.0265], + device='cuda:2'), in_proj_covar=tensor([0.0418, 0.0355, 0.0309, 0.0411, 0.0342, 0.0500, 0.0378, 0.0382], + device='cuda:2'), out_proj_covar=tensor([1.1639e-04, 9.6309e-05, 8.3979e-05, 1.1187e-04, 9.3452e-05, 1.4703e-04, + 1.0569e-04, 1.0526e-04], device='cuda:2') +2023-02-06 14:59:43,701 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108099.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:59:50,130 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 15:00:05,760 INFO [train.py:901] (2/4) Epoch 14, batch 3050, loss[loss=0.2159, simple_loss=0.2915, pruned_loss=0.07013, over 7928.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3028, pruned_loss=0.0735, over 1623117.20 frames. ], batch size: 20, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:00:10,682 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:00:28,082 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.670e+02 3.118e+02 3.835e+02 7.160e+02, threshold=6.236e+02, percent-clipped=1.0 +2023-02-06 15:00:41,676 INFO [train.py:901] (2/4) Epoch 14, batch 3100, loss[loss=0.2176, simple_loss=0.2996, pruned_loss=0.06782, over 8085.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3024, pruned_loss=0.07297, over 1623809.99 frames. ], batch size: 21, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:01:04,692 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:01:15,954 INFO [train.py:901] (2/4) Epoch 14, batch 3150, loss[loss=0.1715, simple_loss=0.2427, pruned_loss=0.05009, over 7787.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3027, pruned_loss=0.07338, over 1619705.20 frames. ], batch size: 19, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:01:24,690 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:01:37,159 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.570e+02 3.163e+02 4.155e+02 7.848e+02, threshold=6.326e+02, percent-clipped=5.0 +2023-02-06 15:01:43,487 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:01:46,163 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7226, 1.3294, 4.8094, 1.8322, 4.2669, 3.9615, 4.3015, 4.2157], + device='cuda:2'), covar=tensor([0.0467, 0.4828, 0.0426, 0.3669, 0.0979, 0.0794, 0.0557, 0.0534], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0600, 0.0615, 0.0565, 0.0643, 0.0550, 0.0535, 0.0604], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 15:01:51,504 INFO [train.py:901] (2/4) Epoch 14, batch 3200, loss[loss=0.1677, simple_loss=0.2482, pruned_loss=0.04363, over 6842.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3018, pruned_loss=0.07248, over 1619784.88 frames. ], batch size: 15, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:02:25,764 INFO [train.py:901] (2/4) Epoch 14, batch 3250, loss[loss=0.1914, simple_loss=0.2771, pruned_loss=0.05283, over 8254.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3017, pruned_loss=0.0723, over 1619777.73 frames. ], batch size: 22, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:02:35,710 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:02:47,016 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.638e+02 3.239e+02 4.086e+02 1.012e+03, threshold=6.478e+02, percent-clipped=4.0 +2023-02-06 15:03:02,194 INFO [train.py:901] (2/4) Epoch 14, batch 3300, loss[loss=0.2583, simple_loss=0.3403, pruned_loss=0.08814, over 8355.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3021, pruned_loss=0.07234, over 1620406.53 frames. ], batch size: 24, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:03:11,256 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:03:17,375 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4296, 2.0780, 3.3796, 1.2873, 2.3638, 1.9253, 1.6671, 2.3700], + device='cuda:2'), covar=tensor([0.1823, 0.2190, 0.0775, 0.3952, 0.1759, 0.2794, 0.1919, 0.2352], + device='cuda:2'), in_proj_covar=tensor([0.0499, 0.0548, 0.0543, 0.0595, 0.0621, 0.0561, 0.0492, 0.0618], + 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') +2023-02-06 15:03:27,911 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:03:36,472 INFO [train.py:901] (2/4) Epoch 14, batch 3350, loss[loss=0.2286, simple_loss=0.2907, pruned_loss=0.08329, over 7821.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3034, pruned_loss=0.0733, over 1620561.72 frames. ], batch size: 20, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:03:49,806 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:03:57,194 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.656e+02 3.299e+02 4.467e+02 8.781e+02, threshold=6.597e+02, percent-clipped=5.0 +2023-02-06 15:04:04,262 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108470.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:04:10,805 INFO [train.py:901] (2/4) Epoch 14, batch 3400, loss[loss=0.1819, simple_loss=0.2665, pruned_loss=0.0486, over 7674.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3034, pruned_loss=0.07336, over 1618134.31 frames. ], batch size: 19, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:04:22,161 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108495.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:04:46,705 INFO [train.py:901] (2/4) Epoch 14, batch 3450, loss[loss=0.2158, simple_loss=0.292, pruned_loss=0.06978, over 8568.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3038, pruned_loss=0.07398, over 1619530.43 frames. ], batch size: 39, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:05:07,907 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.481e+02 3.055e+02 3.627e+02 7.933e+02, threshold=6.110e+02, percent-clipped=3.0 +2023-02-06 15:05:21,982 INFO [train.py:901] (2/4) Epoch 14, batch 3500, loss[loss=0.2295, simple_loss=0.3129, pruned_loss=0.07304, over 8335.00 frames. ], tot_loss[loss=0.227, simple_loss=0.305, pruned_loss=0.07452, over 1624583.17 frames. ], batch size: 25, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:05:29,138 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 15:05:31,991 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:05:57,712 INFO [train.py:901] (2/4) Epoch 14, batch 3550, loss[loss=0.2346, simple_loss=0.3109, pruned_loss=0.07913, over 8531.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3037, pruned_loss=0.07433, over 1618429.39 frames. ], batch size: 28, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:06:17,919 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.417e+02 3.151e+02 4.175e+02 8.210e+02, threshold=6.301e+02, percent-clipped=3.0 +2023-02-06 15:06:31,408 INFO [train.py:901] (2/4) Epoch 14, batch 3600, loss[loss=0.1877, simple_loss=0.2694, pruned_loss=0.05293, over 7696.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3025, pruned_loss=0.07425, over 1613737.33 frames. ], batch size: 18, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:06:36,123 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108687.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:07:07,032 INFO [train.py:901] (2/4) Epoch 14, batch 3650, loss[loss=0.1941, simple_loss=0.277, pruned_loss=0.0556, over 8079.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3037, pruned_loss=0.07439, over 1612985.45 frames. ], batch size: 21, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:07:27,798 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.655e+02 3.191e+02 3.880e+02 8.243e+02, threshold=6.382e+02, percent-clipped=2.0 +2023-02-06 15:07:30,605 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 15:07:41,526 INFO [train.py:901] (2/4) Epoch 14, batch 3700, loss[loss=0.2715, simple_loss=0.3406, pruned_loss=0.1012, over 8337.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.303, pruned_loss=0.07373, over 1616097.99 frames. ], batch size: 26, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:07:47,698 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0129, 1.2272, 1.2213, 0.5848, 1.2201, 1.0159, 0.0555, 1.1426], + device='cuda:2'), covar=tensor([0.0296, 0.0278, 0.0240, 0.0391, 0.0294, 0.0707, 0.0576, 0.0233], + device='cuda:2'), in_proj_covar=tensor([0.0414, 0.0349, 0.0306, 0.0406, 0.0338, 0.0494, 0.0372, 0.0377], + device='cuda:2'), out_proj_covar=tensor([1.1507e-04, 9.4579e-05, 8.2963e-05, 1.1041e-04, 9.2260e-05, 1.4516e-04, + 1.0364e-04, 1.0377e-04], device='cuda:2') +2023-02-06 15:07:50,764 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:07:56,397 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:08:01,073 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108809.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:08:15,867 INFO [train.py:901] (2/4) Epoch 14, batch 3750, loss[loss=0.2236, simple_loss=0.2975, pruned_loss=0.07482, over 7918.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3025, pruned_loss=0.07284, over 1619200.71 frames. ], batch size: 20, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:08:37,485 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.414e+02 2.846e+02 3.664e+02 8.039e+02, threshold=5.692e+02, percent-clipped=5.0 +2023-02-06 15:08:37,680 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7441, 1.5706, 3.1313, 1.4707, 2.2495, 3.3709, 3.4685, 2.9078], + device='cuda:2'), covar=tensor([0.1096, 0.1521, 0.0335, 0.1887, 0.0891, 0.0277, 0.0480, 0.0630], + device='cuda:2'), in_proj_covar=tensor([0.0272, 0.0303, 0.0267, 0.0297, 0.0283, 0.0245, 0.0366, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:08:51,938 INFO [train.py:901] (2/4) Epoch 14, batch 3800, loss[loss=0.2324, simple_loss=0.3151, pruned_loss=0.0749, over 8293.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.302, pruned_loss=0.07312, over 1614866.38 frames. ], batch size: 23, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:09:12,245 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:09:21,318 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1602, 1.7094, 4.3537, 2.0099, 2.3488, 4.9276, 4.9073, 4.3091], + device='cuda:2'), covar=tensor([0.1152, 0.1659, 0.0269, 0.1924, 0.1326, 0.0188, 0.0368, 0.0523], + device='cuda:2'), in_proj_covar=tensor([0.0272, 0.0303, 0.0266, 0.0296, 0.0283, 0.0246, 0.0367, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:09:26,876 INFO [train.py:901] (2/4) Epoch 14, batch 3850, loss[loss=0.2124, simple_loss=0.2953, pruned_loss=0.06477, over 8484.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3023, pruned_loss=0.07341, over 1619612.57 frames. ], batch size: 29, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:09:33,935 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:09:35,483 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6506, 2.2087, 3.5344, 2.6409, 3.1952, 2.4932, 2.1720, 1.8519], + device='cuda:2'), covar=tensor([0.4087, 0.4478, 0.1275, 0.3188, 0.2247, 0.2467, 0.1736, 0.4846], + device='cuda:2'), in_proj_covar=tensor([0.0892, 0.0896, 0.0741, 0.0869, 0.0948, 0.0825, 0.0708, 0.0777], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 15:09:35,959 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 15:09:49,057 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.574e+02 3.020e+02 4.517e+02 9.725e+02, threshold=6.039e+02, percent-clipped=15.0 +2023-02-06 15:09:51,340 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9929, 2.3945, 1.8431, 2.8801, 1.3428, 1.4984, 2.0411, 2.5714], + device='cuda:2'), covar=tensor([0.0786, 0.0824, 0.1021, 0.0408, 0.1186, 0.1659, 0.1019, 0.0699], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0207, 0.0252, 0.0214, 0.0216, 0.0252, 0.0257, 0.0214], + 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') +2023-02-06 15:10:04,279 INFO [train.py:901] (2/4) Epoch 14, batch 3900, loss[loss=0.1867, simple_loss=0.2696, pruned_loss=0.05187, over 8082.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3016, pruned_loss=0.07287, over 1616235.40 frames. ], batch size: 21, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:10:07,763 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1861, 1.1526, 4.5762, 1.7545, 3.6350, 3.5730, 4.0727, 4.0419], + device='cuda:2'), covar=tensor([0.1088, 0.6498, 0.0803, 0.4578, 0.1860, 0.1439, 0.0974, 0.0894], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0608, 0.0623, 0.0570, 0.0645, 0.0552, 0.0543, 0.0606], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 15:10:26,821 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1647, 4.1321, 3.7821, 2.4742, 3.8028, 3.7862, 3.8781, 3.4481], + device='cuda:2'), covar=tensor([0.0861, 0.0611, 0.0959, 0.3944, 0.0881, 0.1147, 0.1125, 0.0981], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0396, 0.0404, 0.0497, 0.0395, 0.0401, 0.0385, 0.0345], + 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') +2023-02-06 15:10:39,023 INFO [train.py:901] (2/4) Epoch 14, batch 3950, loss[loss=0.2171, simple_loss=0.3079, pruned_loss=0.06313, over 8503.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.3013, pruned_loss=0.07281, over 1613281.24 frames. ], batch size: 28, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:10:40,197 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.83 vs. limit=5.0 +2023-02-06 15:10:56,307 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:10:56,956 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109055.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 15:10:58,501 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.55 vs. limit=5.0 +2023-02-06 15:10:59,097 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:11:00,248 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.458e+02 2.966e+02 3.777e+02 8.079e+02, threshold=5.932e+02, percent-clipped=4.0 +2023-02-06 15:11:14,744 INFO [train.py:901] (2/4) Epoch 14, batch 4000, loss[loss=0.2482, simple_loss=0.3288, pruned_loss=0.08386, over 8532.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3019, pruned_loss=0.07349, over 1613022.38 frames. ], batch size: 49, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:11:17,674 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109083.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:11:23,849 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1619, 1.3914, 1.5666, 1.3346, 0.8271, 1.4184, 1.7879, 1.6537], + device='cuda:2'), covar=tensor([0.0462, 0.1322, 0.1728, 0.1443, 0.0628, 0.1521, 0.0698, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0157, 0.0101, 0.0162, 0.0113, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 15:11:43,039 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7395, 5.8351, 5.0479, 2.3899, 5.0745, 5.6102, 5.3231, 5.3220], + device='cuda:2'), covar=tensor([0.0508, 0.0371, 0.0901, 0.4594, 0.0715, 0.0738, 0.0960, 0.0533], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0399, 0.0407, 0.0500, 0.0399, 0.0402, 0.0388, 0.0348], + 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') +2023-02-06 15:11:50,547 INFO [train.py:901] (2/4) Epoch 14, batch 4050, loss[loss=0.201, simple_loss=0.2719, pruned_loss=0.06504, over 7514.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3009, pruned_loss=0.07266, over 1610144.28 frames. ], batch size: 18, lr: 5.45e-03, grad_scale: 16.0 +2023-02-06 15:12:06,803 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:12:11,650 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.362e+02 2.684e+02 3.543e+02 7.215e+02, threshold=5.369e+02, percent-clipped=4.0 +2023-02-06 15:12:16,034 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:12:26,284 INFO [train.py:901] (2/4) Epoch 14, batch 4100, loss[loss=0.2753, simple_loss=0.3419, pruned_loss=0.1043, over 8526.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3024, pruned_loss=0.07388, over 1608149.70 frames. ], batch size: 29, lr: 5.45e-03, grad_scale: 16.0 +2023-02-06 15:12:33,976 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109190.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:12:50,502 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:13:02,498 INFO [train.py:901] (2/4) Epoch 14, batch 4150, loss[loss=0.2422, simple_loss=0.3153, pruned_loss=0.08451, over 8644.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3022, pruned_loss=0.0737, over 1609939.10 frames. ], batch size: 34, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:13:10,242 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1547, 3.8651, 2.2018, 2.6088, 2.9403, 2.1523, 2.6538, 2.9731], + device='cuda:2'), covar=tensor([0.1491, 0.0271, 0.0988, 0.0721, 0.0599, 0.1186, 0.1041, 0.1020], + device='cuda:2'), in_proj_covar=tensor([0.0344, 0.0231, 0.0320, 0.0294, 0.0297, 0.0324, 0.0342, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:13:23,981 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.677e+02 3.078e+02 3.893e+02 8.547e+02, threshold=6.157e+02, percent-clipped=10.0 +2023-02-06 15:13:28,944 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:13:35,717 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 15:13:37,136 INFO [train.py:901] (2/4) Epoch 14, batch 4200, loss[loss=0.1901, simple_loss=0.2653, pruned_loss=0.05746, over 7795.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3011, pruned_loss=0.07268, over 1611535.65 frames. ], batch size: 19, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:13:59,673 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109310.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:00,996 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:01,568 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 15:14:14,549 INFO [train.py:901] (2/4) Epoch 14, batch 4250, loss[loss=0.2198, simple_loss=0.3048, pruned_loss=0.0674, over 8447.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3008, pruned_loss=0.07234, over 1615369.13 frames. ], batch size: 49, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:14:18,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109335.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:18,815 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:28,518 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8916, 1.5761, 2.0576, 1.8073, 1.9634, 1.8674, 1.6228, 0.7657], + device='cuda:2'), covar=tensor([0.4847, 0.4175, 0.1533, 0.2820, 0.2014, 0.2344, 0.1765, 0.4295], + device='cuda:2'), in_proj_covar=tensor([0.0896, 0.0899, 0.0743, 0.0873, 0.0950, 0.0827, 0.0712, 0.0781], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 15:14:32,592 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2199, 1.5748, 3.5283, 1.4442, 2.2085, 3.7880, 3.8632, 3.1674], + device='cuda:2'), covar=tensor([0.0997, 0.1619, 0.0284, 0.2207, 0.1127, 0.0235, 0.0523, 0.0601], + device='cuda:2'), in_proj_covar=tensor([0.0272, 0.0302, 0.0267, 0.0296, 0.0283, 0.0246, 0.0366, 0.0293], + 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') +2023-02-06 15:14:35,684 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.488e+02 3.016e+02 3.845e+02 8.299e+02, threshold=6.033e+02, percent-clipped=4.0 +2023-02-06 15:14:48,162 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6964, 1.1163, 1.4334, 1.0689, 0.9165, 1.2286, 1.4985, 1.3535], + device='cuda:2'), covar=tensor([0.0594, 0.1978, 0.2484, 0.1878, 0.0694, 0.2183, 0.0801, 0.0780], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0156, 0.0101, 0.0162, 0.0113, 0.0137], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 15:14:48,667 INFO [train.py:901] (2/4) Epoch 14, batch 4300, loss[loss=0.2452, simple_loss=0.3245, pruned_loss=0.08297, over 8238.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3027, pruned_loss=0.07373, over 1613770.74 frames. ], batch size: 24, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:15:01,894 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109399.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:15:24,493 INFO [train.py:901] (2/4) Epoch 14, batch 4350, loss[loss=0.1774, simple_loss=0.2531, pruned_loss=0.05083, over 7427.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3021, pruned_loss=0.07327, over 1613942.87 frames. ], batch size: 17, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:15:27,404 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8956, 3.8300, 3.4731, 1.7112, 3.4402, 3.4244, 3.4022, 3.1475], + device='cuda:2'), covar=tensor([0.0868, 0.0706, 0.1310, 0.4718, 0.0948, 0.1020, 0.1706, 0.1031], + device='cuda:2'), in_proj_covar=tensor([0.0476, 0.0392, 0.0400, 0.0491, 0.0392, 0.0396, 0.0383, 0.0343], + 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') +2023-02-06 15:15:34,081 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 15:15:47,273 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.678e+02 3.270e+02 4.253e+02 1.326e+03, threshold=6.540e+02, percent-clipped=8.0 +2023-02-06 15:16:00,544 INFO [train.py:901] (2/4) Epoch 14, batch 4400, loss[loss=0.2588, simple_loss=0.329, pruned_loss=0.09426, over 8181.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3029, pruned_loss=0.07356, over 1618026.58 frames. ], batch size: 23, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:16:15,765 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 15:16:24,166 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109514.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:16:31,893 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:16:35,898 INFO [train.py:901] (2/4) Epoch 14, batch 4450, loss[loss=0.2015, simple_loss=0.2745, pruned_loss=0.06427, over 7707.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3021, pruned_loss=0.073, over 1616330.95 frames. ], batch size: 18, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:16:49,975 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:16:55,367 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:16:58,634 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.456e+02 2.864e+02 3.608e+02 1.087e+03, threshold=5.728e+02, percent-clipped=4.0 +2023-02-06 15:17:11,127 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 15:17:12,486 INFO [train.py:901] (2/4) Epoch 14, batch 4500, loss[loss=0.1883, simple_loss=0.2627, pruned_loss=0.05697, over 7422.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.3015, pruned_loss=0.07274, over 1612479.71 frames. ], batch size: 17, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:17:24,283 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109597.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:17:47,002 INFO [train.py:901] (2/4) Epoch 14, batch 4550, loss[loss=0.1712, simple_loss=0.2446, pruned_loss=0.04894, over 7658.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3022, pruned_loss=0.07324, over 1612976.47 frames. ], batch size: 19, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:18:05,732 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109656.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:09,031 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.640e+02 3.232e+02 4.162e+02 9.021e+02, threshold=6.464e+02, percent-clipped=8.0 +2023-02-06 15:18:16,734 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109671.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:21,366 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:23,259 INFO [train.py:901] (2/4) Epoch 14, batch 4600, loss[loss=0.2046, simple_loss=0.2902, pruned_loss=0.05951, over 8100.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.3008, pruned_loss=0.07214, over 1616035.69 frames. ], batch size: 23, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:18:23,331 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:42,475 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 15:18:56,947 INFO [train.py:901] (2/4) Epoch 14, batch 4650, loss[loss=0.2325, simple_loss=0.3109, pruned_loss=0.0771, over 7978.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3003, pruned_loss=0.07213, over 1612555.91 frames. ], batch size: 21, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:19:08,007 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0606, 2.3688, 2.7199, 1.4731, 2.7690, 1.6948, 1.6485, 2.0260], + device='cuda:2'), covar=tensor([0.0509, 0.0284, 0.0205, 0.0508, 0.0299, 0.0636, 0.0546, 0.0356], + device='cuda:2'), in_proj_covar=tensor([0.0415, 0.0353, 0.0304, 0.0409, 0.0341, 0.0498, 0.0373, 0.0379], + device='cuda:2'), out_proj_covar=tensor([1.1533e-04, 9.5424e-05, 8.2226e-05, 1.1137e-04, 9.2971e-05, 1.4660e-04, + 1.0396e-04, 1.0414e-04], device='cuda:2') +2023-02-06 15:19:18,374 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 15:19:18,715 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.556e+02 3.032e+02 3.907e+02 9.020e+02, threshold=6.065e+02, percent-clipped=4.0 +2023-02-06 15:19:24,820 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109770.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:19:25,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:19:31,312 INFO [train.py:901] (2/4) Epoch 14, batch 4700, loss[loss=0.2478, simple_loss=0.3228, pruned_loss=0.08635, over 8361.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.301, pruned_loss=0.07284, over 1611319.28 frames. ], batch size: 24, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:19:39,520 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8867, 2.4438, 2.9422, 1.3546, 3.1320, 1.5857, 1.4499, 1.7906], + device='cuda:2'), covar=tensor([0.0731, 0.0359, 0.0263, 0.0636, 0.0408, 0.0829, 0.0780, 0.0502], + device='cuda:2'), in_proj_covar=tensor([0.0414, 0.0352, 0.0303, 0.0408, 0.0340, 0.0497, 0.0372, 0.0378], + device='cuda:2'), out_proj_covar=tensor([1.1512e-04, 9.5139e-05, 8.1990e-05, 1.1119e-04, 9.2708e-05, 1.4612e-04, + 1.0354e-04, 1.0378e-04], device='cuda:2') +2023-02-06 15:19:42,824 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:19:42,848 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109795.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:20:06,516 INFO [train.py:901] (2/4) Epoch 14, batch 4750, loss[loss=0.2152, simple_loss=0.2902, pruned_loss=0.07005, over 8105.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3019, pruned_loss=0.07308, over 1614292.49 frames. ], batch size: 23, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:20:10,482 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 15:20:12,424 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 15:20:26,961 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.643e+02 3.166e+02 4.371e+02 1.104e+03, threshold=6.332e+02, percent-clipped=5.0 +2023-02-06 15:20:40,300 INFO [train.py:901] (2/4) Epoch 14, batch 4800, loss[loss=0.2219, simple_loss=0.2917, pruned_loss=0.07605, over 8151.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3007, pruned_loss=0.07296, over 1608700.57 frames. ], batch size: 22, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:20:50,358 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0353, 2.2713, 1.8614, 2.9469, 1.2543, 1.6257, 2.2371, 2.3976], + device='cuda:2'), covar=tensor([0.0740, 0.0840, 0.0895, 0.0344, 0.1177, 0.1347, 0.0881, 0.0768], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0207, 0.0252, 0.0211, 0.0213, 0.0250, 0.0256, 0.0214], + 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') +2023-02-06 15:21:03,196 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 15:21:14,265 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:21:16,039 INFO [train.py:901] (2/4) Epoch 14, batch 4850, loss[loss=0.2311, simple_loss=0.3135, pruned_loss=0.07433, over 8510.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.2999, pruned_loss=0.07256, over 1604504.80 frames. ], batch size: 26, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:21:23,509 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109941.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:21:31,141 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109952.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:21:37,038 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.400e+02 2.854e+02 3.344e+02 7.947e+02, threshold=5.708e+02, percent-clipped=2.0 +2023-02-06 15:21:49,880 INFO [train.py:901] (2/4) Epoch 14, batch 4900, loss[loss=0.2518, simple_loss=0.33, pruned_loss=0.08685, over 8252.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3023, pruned_loss=0.07359, over 1606680.04 frames. ], batch size: 24, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:22:19,457 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:24,322 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:26,032 INFO [train.py:901] (2/4) Epoch 14, batch 4950, loss[loss=0.2608, simple_loss=0.3403, pruned_loss=0.09065, over 8515.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3007, pruned_loss=0.07269, over 1609177.58 frames. ], batch size: 28, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:22:30,963 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110035.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:39,018 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5744, 1.9811, 3.1114, 1.3837, 2.3115, 2.0631, 1.6957, 2.2013], + device='cuda:2'), covar=tensor([0.1876, 0.2445, 0.0943, 0.4069, 0.1786, 0.2950, 0.2010, 0.2362], + device='cuda:2'), in_proj_covar=tensor([0.0499, 0.0547, 0.0539, 0.0593, 0.0619, 0.0563, 0.0486, 0.0616], + 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') +2023-02-06 15:22:41,738 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110051.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:42,425 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110052.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:45,152 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110056.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:48,332 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.585e+02 3.180e+02 4.032e+02 7.448e+02, threshold=6.360e+02, percent-clipped=3.0 +2023-02-06 15:22:49,157 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:58,260 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110076.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:00,777 INFO [train.py:901] (2/4) Epoch 14, batch 5000, loss[loss=0.1973, simple_loss=0.2792, pruned_loss=0.05769, over 7975.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3007, pruned_loss=0.07263, over 1610435.70 frames. ], batch size: 21, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:23:31,467 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5251, 1.3659, 2.3327, 1.1901, 2.1011, 2.5130, 2.6586, 2.0952], + device='cuda:2'), covar=tensor([0.0949, 0.1220, 0.0439, 0.2020, 0.0733, 0.0346, 0.0605, 0.0735], + device='cuda:2'), in_proj_covar=tensor([0.0270, 0.0302, 0.0266, 0.0294, 0.0281, 0.0243, 0.0366, 0.0292], + 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') +2023-02-06 15:23:33,487 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:34,635 INFO [train.py:901] (2/4) Epoch 14, batch 5050, loss[loss=0.2801, simple_loss=0.3518, pruned_loss=0.1042, over 8823.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3019, pruned_loss=0.07357, over 1606224.76 frames. ], batch size: 40, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:23:36,750 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:38,741 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:41,650 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 15:23:43,174 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 15:23:57,271 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.889e+02 3.601e+02 4.263e+02 9.587e+02, threshold=7.203e+02, percent-clipped=6.0 +2023-02-06 15:24:09,949 INFO [train.py:901] (2/4) Epoch 14, batch 5100, loss[loss=0.2096, simple_loss=0.3019, pruned_loss=0.05868, over 8471.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3021, pruned_loss=0.07368, over 1608595.08 frames. ], batch size: 25, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:24:21,133 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5066, 2.2265, 2.8231, 1.7780, 1.7068, 2.8284, 1.0662, 2.0667], + device='cuda:2'), covar=tensor([0.2020, 0.1335, 0.0494, 0.2316, 0.3427, 0.0488, 0.2581, 0.1847], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0175, 0.0107, 0.0217, 0.0262, 0.0113, 0.0162, 0.0174], + 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') +2023-02-06 15:24:42,795 INFO [train.py:901] (2/4) Epoch 14, batch 5150, loss[loss=0.2787, simple_loss=0.3486, pruned_loss=0.1044, over 8482.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3018, pruned_loss=0.07401, over 1606455.60 frames. ], batch size: 29, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:24:44,028 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.21 vs. limit=5.0 +2023-02-06 15:25:05,057 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.455e+02 3.012e+02 3.817e+02 9.599e+02, threshold=6.024e+02, percent-clipped=2.0 +2023-02-06 15:25:19,324 INFO [train.py:901] (2/4) Epoch 14, batch 5200, loss[loss=0.2363, simple_loss=0.3101, pruned_loss=0.08123, over 8358.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3032, pruned_loss=0.0745, over 1609611.47 frames. ], batch size: 24, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:25:33,778 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:39,480 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 15:25:41,059 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:47,126 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110321.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:53,166 INFO [train.py:901] (2/4) Epoch 14, batch 5250, loss[loss=0.1848, simple_loss=0.2587, pruned_loss=0.05543, over 7640.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3022, pruned_loss=0.07421, over 1607142.10 frames. ], batch size: 19, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:25:57,544 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:58,346 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110337.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:15,514 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.532e+02 3.204e+02 3.879e+02 8.466e+02, threshold=6.409e+02, percent-clipped=5.0 +2023-02-06 15:26:28,847 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:29,481 INFO [train.py:901] (2/4) Epoch 14, batch 5300, loss[loss=0.2026, simple_loss=0.2916, pruned_loss=0.05679, over 7920.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3031, pruned_loss=0.07441, over 1608368.76 frames. ], batch size: 20, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:26:37,624 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 15:26:39,495 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:48,920 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:56,331 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:27:03,328 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 +2023-02-06 15:27:05,003 INFO [train.py:901] (2/4) Epoch 14, batch 5350, loss[loss=0.234, simple_loss=0.3141, pruned_loss=0.07695, over 8505.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3034, pruned_loss=0.07404, over 1614058.04 frames. ], batch size: 26, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:27:25,505 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.452e+02 3.047e+02 3.791e+02 6.566e+02, threshold=6.094e+02, percent-clipped=2.0 +2023-02-06 15:27:32,803 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:27:36,872 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:27:38,768 INFO [train.py:901] (2/4) Epoch 14, batch 5400, loss[loss=0.3003, simple_loss=0.3582, pruned_loss=0.1213, over 8606.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3048, pruned_loss=0.07481, over 1612452.78 frames. ], batch size: 31, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:27:48,332 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:28:08,457 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:28:14,356 INFO [train.py:901] (2/4) Epoch 14, batch 5450, loss[loss=0.2507, simple_loss=0.3217, pruned_loss=0.08978, over 6966.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3045, pruned_loss=0.07473, over 1612980.91 frames. ], batch size: 71, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:28:30,484 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 15:28:34,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.661e+02 2.429e+02 2.846e+02 3.589e+02 7.640e+02, threshold=5.692e+02, percent-clipped=1.0 +2023-02-06 15:28:39,769 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3037, 2.6333, 1.7557, 2.1303, 2.2123, 1.5002, 1.9069, 2.0079], + device='cuda:2'), covar=tensor([0.1539, 0.0360, 0.1106, 0.0603, 0.0670, 0.1495, 0.1062, 0.0891], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0235, 0.0322, 0.0297, 0.0298, 0.0328, 0.0344, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:28:47,542 INFO [train.py:901] (2/4) Epoch 14, batch 5500, loss[loss=0.2474, simple_loss=0.3268, pruned_loss=0.08397, over 8522.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3045, pruned_loss=0.07488, over 1614129.24 frames. ], batch size: 49, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:28:52,322 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:28:55,815 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:01,840 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7530, 1.8291, 2.3490, 1.6182, 1.2454, 2.3906, 0.3232, 1.3779], + device='cuda:2'), covar=tensor([0.2579, 0.1672, 0.0475, 0.2430, 0.4118, 0.0440, 0.3129, 0.1911], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0176, 0.0108, 0.0219, 0.0263, 0.0113, 0.0162, 0.0174], + 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') +2023-02-06 15:29:07,790 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0336, 1.5197, 3.6209, 1.5144, 2.2721, 3.9644, 3.9699, 3.4112], + device='cuda:2'), covar=tensor([0.1045, 0.1681, 0.0277, 0.2055, 0.1070, 0.0195, 0.0405, 0.0535], + device='cuda:2'), in_proj_covar=tensor([0.0271, 0.0304, 0.0268, 0.0295, 0.0283, 0.0243, 0.0368, 0.0290], + 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') +2023-02-06 15:29:23,307 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110629.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:23,833 INFO [train.py:901] (2/4) Epoch 14, batch 5550, loss[loss=0.2575, simple_loss=0.3276, pruned_loss=0.0937, over 8133.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3029, pruned_loss=0.07411, over 1612866.20 frames. ], batch size: 22, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:29:25,763 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.04 vs. limit=5.0 +2023-02-06 15:29:28,288 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.87 vs. limit=5.0 +2023-02-06 15:29:34,006 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:35,490 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4385, 2.7319, 1.9086, 2.1894, 2.3137, 1.5785, 2.1089, 2.0066], + device='cuda:2'), covar=tensor([0.1290, 0.0327, 0.1042, 0.0572, 0.0599, 0.1375, 0.0849, 0.0824], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0235, 0.0323, 0.0298, 0.0298, 0.0329, 0.0345, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:29:44,459 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.692e+02 2.421e+02 3.120e+02 3.692e+02 1.093e+03, threshold=6.240e+02, percent-clipped=9.0 +2023-02-06 15:29:47,079 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110665.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:56,916 INFO [train.py:901] (2/4) Epoch 14, batch 5600, loss[loss=0.2355, simple_loss=0.3228, pruned_loss=0.07407, over 8505.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3024, pruned_loss=0.07379, over 1612545.20 frames. ], batch size: 26, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:29:56,987 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:30,944 INFO [train.py:901] (2/4) Epoch 14, batch 5650, loss[loss=0.1878, simple_loss=0.2568, pruned_loss=0.05947, over 7677.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3014, pruned_loss=0.07336, over 1612775.57 frames. ], batch size: 18, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:30:33,059 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 15:30:34,099 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 +2023-02-06 15:30:47,217 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:49,879 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:54,030 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:54,498 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.502e+02 3.092e+02 3.638e+02 5.778e+02, threshold=6.185e+02, percent-clipped=0.0 +2023-02-06 15:31:02,499 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 15:31:04,322 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110775.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:05,717 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110777.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:07,494 INFO [train.py:901] (2/4) Epoch 14, batch 5700, loss[loss=0.202, simple_loss=0.2899, pruned_loss=0.05703, over 8341.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3005, pruned_loss=0.07267, over 1610202.17 frames. ], batch size: 26, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:31:07,673 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110780.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:15,732 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1133, 4.0852, 3.6710, 1.8493, 3.6353, 3.8048, 3.7127, 3.5639], + device='cuda:2'), covar=tensor([0.0823, 0.0646, 0.0965, 0.4517, 0.0868, 0.0968, 0.1234, 0.0856], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0397, 0.0396, 0.0496, 0.0392, 0.0399, 0.0385, 0.0343], + 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') +2023-02-06 15:31:17,912 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:22,872 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:40,945 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 15:31:41,630 INFO [train.py:901] (2/4) Epoch 14, batch 5750, loss[loss=0.2095, simple_loss=0.2948, pruned_loss=0.06209, over 8367.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3006, pruned_loss=0.07246, over 1610883.77 frames. ], batch size: 24, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:31:51,510 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110843.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:54,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:04,400 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.447e+02 3.019e+02 3.853e+02 7.521e+02, threshold=6.038e+02, percent-clipped=3.0 +2023-02-06 15:32:10,141 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:14,047 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110873.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:18,667 INFO [train.py:901] (2/4) Epoch 14, batch 5800, loss[loss=0.192, simple_loss=0.274, pruned_loss=0.05496, over 8247.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3016, pruned_loss=0.07325, over 1616803.69 frames. ], batch size: 22, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:32:22,951 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:53,260 INFO [train.py:901] (2/4) Epoch 14, batch 5850, loss[loss=0.2393, simple_loss=0.3062, pruned_loss=0.08618, over 7973.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3018, pruned_loss=0.07304, over 1618278.59 frames. ], batch size: 21, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:33:02,463 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 +2023-02-06 15:33:14,131 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.498e+02 3.098e+02 4.112e+02 1.106e+03, threshold=6.195e+02, percent-clipped=10.0 +2023-02-06 15:33:22,882 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:33:26,431 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110976.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:33:28,980 INFO [train.py:901] (2/4) Epoch 14, batch 5900, loss[loss=0.3363, simple_loss=0.378, pruned_loss=0.1473, over 6898.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.303, pruned_loss=0.07346, over 1613665.63 frames. ], batch size: 72, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:33:38,588 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4506, 1.8055, 1.8885, 1.1879, 1.9364, 1.3112, 0.4604, 1.7484], + device='cuda:2'), covar=tensor([0.0410, 0.0245, 0.0208, 0.0376, 0.0297, 0.0680, 0.0636, 0.0177], + device='cuda:2'), in_proj_covar=tensor([0.0415, 0.0358, 0.0305, 0.0406, 0.0345, 0.0499, 0.0372, 0.0377], + device='cuda:2'), out_proj_covar=tensor([1.1526e-04, 9.6830e-05, 8.2572e-05, 1.1029e-04, 9.4174e-05, 1.4647e-04, + 1.0367e-04, 1.0325e-04], device='cuda:2') +2023-02-06 15:33:54,412 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:03,738 INFO [train.py:901] (2/4) Epoch 14, batch 5950, loss[loss=0.2409, simple_loss=0.3174, pruned_loss=0.08221, over 8631.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3025, pruned_loss=0.07321, over 1618576.14 frames. ], batch size: 31, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:34:07,780 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111036.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:11,075 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111041.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:17,723 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111051.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:24,231 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.648e+02 3.047e+02 4.016e+02 7.772e+02, threshold=6.093e+02, percent-clipped=5.0 +2023-02-06 15:34:24,459 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111061.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:34,723 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111076.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:38,000 INFO [train.py:901] (2/4) Epoch 14, batch 6000, loss[loss=0.2377, simple_loss=0.3196, pruned_loss=0.07793, over 8470.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3033, pruned_loss=0.0737, over 1618511.69 frames. ], batch size: 25, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:34:38,000 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 15:34:50,552 INFO [train.py:935] (2/4) Epoch 14, validation: loss=0.1818, simple_loss=0.2816, pruned_loss=0.04094, over 944034.00 frames. +2023-02-06 15:34:50,553 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 15:34:51,683 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 15:34:56,292 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111088.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:35:03,696 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111098.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:35:11,766 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 15:35:27,227 INFO [train.py:901] (2/4) Epoch 14, batch 6050, loss[loss=0.2018, simple_loss=0.2868, pruned_loss=0.05841, over 8288.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3025, pruned_loss=0.07359, over 1615501.19 frames. ], batch size: 23, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:35:49,432 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.432e+02 2.876e+02 3.526e+02 5.542e+02, threshold=5.752e+02, percent-clipped=0.0 +2023-02-06 15:35:59,326 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 15:36:01,589 INFO [train.py:901] (2/4) Epoch 14, batch 6100, loss[loss=0.2043, simple_loss=0.2751, pruned_loss=0.0667, over 7771.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3031, pruned_loss=0.07442, over 1616970.20 frames. ], batch size: 19, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:36:15,951 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 15:36:24,937 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:36:37,116 INFO [train.py:901] (2/4) Epoch 14, batch 6150, loss[loss=0.1845, simple_loss=0.2705, pruned_loss=0.04923, over 7654.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3021, pruned_loss=0.07379, over 1618171.24 frames. ], batch size: 19, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:36:37,874 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111230.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:36:39,287 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8239, 3.7709, 3.4121, 1.7415, 3.3540, 3.4510, 3.4019, 3.1553], + device='cuda:2'), covar=tensor([0.0965, 0.0695, 0.1157, 0.4650, 0.1000, 0.1068, 0.1454, 0.1035], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0402, 0.0405, 0.0503, 0.0396, 0.0403, 0.0389, 0.0350], + 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') +2023-02-06 15:36:46,727 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:36:59,331 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.474e+02 3.213e+02 4.029e+02 8.079e+02, threshold=6.426e+02, percent-clipped=5.0 +2023-02-06 15:37:11,858 INFO [train.py:901] (2/4) Epoch 14, batch 6200, loss[loss=0.2089, simple_loss=0.2954, pruned_loss=0.06123, over 8073.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3014, pruned_loss=0.07381, over 1613288.82 frames. ], batch size: 21, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:37:20,717 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2658, 2.1841, 1.7499, 1.9769, 1.8856, 1.3467, 1.6407, 1.7220], + device='cuda:2'), covar=tensor([0.1194, 0.0376, 0.1016, 0.0507, 0.0614, 0.1415, 0.0903, 0.0737], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0232, 0.0322, 0.0293, 0.0296, 0.0326, 0.0340, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:37:38,514 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111320.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:37:45,102 INFO [train.py:901] (2/4) Epoch 14, batch 6250, loss[loss=0.1793, simple_loss=0.2655, pruned_loss=0.04652, over 5186.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3001, pruned_loss=0.07309, over 1608351.29 frames. ], batch size: 11, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:37:55,176 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:37:55,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111345.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:38:08,443 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.286e+02 2.818e+02 3.691e+02 1.208e+03, threshold=5.637e+02, percent-clipped=2.0 +2023-02-06 15:38:13,369 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111369.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:38:20,509 INFO [train.py:901] (2/4) Epoch 14, batch 6300, loss[loss=0.2385, simple_loss=0.3117, pruned_loss=0.0826, over 8143.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3007, pruned_loss=0.07322, over 1607726.22 frames. ], batch size: 22, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:38:28,035 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3484, 2.4024, 1.6654, 2.0985, 2.0928, 1.3991, 1.7724, 1.9210], + device='cuda:2'), covar=tensor([0.1393, 0.0367, 0.1141, 0.0503, 0.0675, 0.1392, 0.0967, 0.0835], + device='cuda:2'), in_proj_covar=tensor([0.0346, 0.0232, 0.0322, 0.0293, 0.0297, 0.0327, 0.0342, 0.0307], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:38:39,596 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7021, 1.5315, 4.8924, 1.9330, 4.3716, 4.1392, 4.4645, 4.3219], + device='cuda:2'), covar=tensor([0.0462, 0.4157, 0.0433, 0.3349, 0.0943, 0.0786, 0.0424, 0.0552], + device='cuda:2'), in_proj_covar=tensor([0.0540, 0.0608, 0.0630, 0.0573, 0.0645, 0.0555, 0.0544, 0.0609], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 15:38:55,226 INFO [train.py:901] (2/4) Epoch 14, batch 6350, loss[loss=0.1979, simple_loss=0.2831, pruned_loss=0.05632, over 7967.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.301, pruned_loss=0.07301, over 1608138.53 frames. ], batch size: 21, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:38:58,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:39:05,851 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7389, 1.5583, 1.8298, 1.5805, 0.9912, 1.7251, 2.1481, 1.8944], + device='cuda:2'), covar=tensor([0.0408, 0.1192, 0.1569, 0.1302, 0.0605, 0.1367, 0.0608, 0.0587], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0191, 0.0157, 0.0101, 0.0162, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 15:39:17,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.340e+02 2.891e+02 3.552e+02 9.934e+02, threshold=5.783e+02, percent-clipped=8.0 +2023-02-06 15:39:23,084 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:39:30,948 INFO [train.py:901] (2/4) Epoch 14, batch 6400, loss[loss=0.2107, simple_loss=0.2931, pruned_loss=0.06415, over 8790.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3002, pruned_loss=0.07214, over 1614862.51 frames. ], batch size: 30, lr: 5.40e-03, grad_scale: 8.0 +2023-02-06 15:39:32,188 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 15:39:35,205 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111486.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:39:40,669 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:40:05,246 INFO [train.py:901] (2/4) Epoch 14, batch 6450, loss[loss=0.1965, simple_loss=0.2856, pruned_loss=0.05375, over 8342.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2988, pruned_loss=0.07094, over 1617220.19 frames. ], batch size: 26, lr: 5.40e-03, grad_scale: 8.0 +2023-02-06 15:40:26,309 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.418e+02 3.184e+02 3.807e+02 1.482e+03, threshold=6.367e+02, percent-clipped=8.0 +2023-02-06 15:40:39,067 INFO [train.py:901] (2/4) Epoch 14, batch 6500, loss[loss=0.1983, simple_loss=0.2831, pruned_loss=0.05676, over 8533.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.3003, pruned_loss=0.07233, over 1613397.39 frames. ], batch size: 28, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:40:44,372 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:40:55,099 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111601.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:41:12,084 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111626.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:41:14,591 INFO [train.py:901] (2/4) Epoch 14, batch 6550, loss[loss=0.1718, simple_loss=0.2547, pruned_loss=0.04445, over 7421.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3016, pruned_loss=0.07311, over 1612097.05 frames. ], batch size: 17, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:41:24,427 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 15:41:35,844 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.515e+02 3.055e+02 3.900e+02 7.605e+02, threshold=6.110e+02, percent-clipped=3.0 +2023-02-06 15:41:43,362 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 15:41:47,951 INFO [train.py:901] (2/4) Epoch 14, batch 6600, loss[loss=0.2117, simple_loss=0.2805, pruned_loss=0.07145, over 7690.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3008, pruned_loss=0.0728, over 1608594.81 frames. ], batch size: 18, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:41:55,650 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:42:03,625 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:42:14,400 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111716.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:42:23,467 INFO [train.py:901] (2/4) Epoch 14, batch 6650, loss[loss=0.2221, simple_loss=0.3051, pruned_loss=0.06951, over 8024.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3001, pruned_loss=0.0723, over 1608744.81 frames. ], batch size: 22, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:42:45,230 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.391e+02 3.105e+02 3.860e+02 7.189e+02, threshold=6.209e+02, percent-clipped=3.0 +2023-02-06 15:42:48,090 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9464, 2.1737, 1.9518, 2.8507, 1.3558, 1.5464, 1.8693, 2.3657], + device='cuda:2'), covar=tensor([0.0844, 0.0991, 0.0985, 0.0387, 0.1295, 0.1621, 0.1072, 0.0837], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0207, 0.0256, 0.0215, 0.0216, 0.0253, 0.0261, 0.0217], + 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') +2023-02-06 15:42:49,656 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 15:42:51,480 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6988, 1.9312, 2.1781, 1.3317, 2.3222, 1.4259, 0.7507, 1.8243], + device='cuda:2'), covar=tensor([0.0530, 0.0293, 0.0223, 0.0497, 0.0306, 0.0711, 0.0717, 0.0276], + device='cuda:2'), in_proj_covar=tensor([0.0409, 0.0354, 0.0302, 0.0405, 0.0339, 0.0493, 0.0369, 0.0376], + device='cuda:2'), out_proj_covar=tensor([1.1356e-04, 9.5802e-05, 8.1746e-05, 1.1006e-04, 9.2653e-05, 1.4440e-04, + 1.0266e-04, 1.0282e-04], device='cuda:2') +2023-02-06 15:42:57,305 INFO [train.py:901] (2/4) Epoch 14, batch 6700, loss[loss=0.1881, simple_loss=0.2649, pruned_loss=0.0557, over 7548.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3001, pruned_loss=0.07285, over 1604091.79 frames. ], batch size: 18, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:43:16,619 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111809.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:43:32,517 INFO [train.py:901] (2/4) Epoch 14, batch 6750, loss[loss=0.2311, simple_loss=0.3033, pruned_loss=0.07948, over 8105.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3, pruned_loss=0.07242, over 1610011.22 frames. ], batch size: 23, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:43:32,588 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:43:54,028 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.559e+02 3.020e+02 4.182e+02 1.269e+03, threshold=6.039e+02, percent-clipped=6.0 +2023-02-06 15:44:02,381 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 15:44:02,883 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 15:44:07,192 INFO [train.py:901] (2/4) Epoch 14, batch 6800, loss[loss=0.2253, simple_loss=0.2968, pruned_loss=0.07687, over 7914.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.3001, pruned_loss=0.07205, over 1616495.93 frames. ], batch size: 20, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:44:40,417 INFO [train.py:901] (2/4) Epoch 14, batch 6850, loss[loss=0.2305, simple_loss=0.315, pruned_loss=0.07299, over 8031.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3014, pruned_loss=0.07258, over 1616152.70 frames. ], batch size: 22, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:44:51,217 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 15:44:52,739 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111945.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:45:01,425 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111958.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:45:03,874 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.542e+02 3.126e+02 4.226e+02 8.027e+02, threshold=6.251e+02, percent-clipped=7.0 +2023-02-06 15:45:16,594 INFO [train.py:901] (2/4) Epoch 14, batch 6900, loss[loss=0.2249, simple_loss=0.3023, pruned_loss=0.07373, over 8251.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2994, pruned_loss=0.07196, over 1608652.58 frames. ], batch size: 24, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:45:18,705 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:45:50,580 INFO [train.py:901] (2/4) Epoch 14, batch 6950, loss[loss=0.2038, simple_loss=0.2661, pruned_loss=0.07077, over 7441.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2997, pruned_loss=0.07224, over 1607697.68 frames. ], batch size: 17, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:45:58,677 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 15:46:13,915 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.419e+02 2.987e+02 3.531e+02 6.552e+02, threshold=5.974e+02, percent-clipped=1.0 +2023-02-06 15:46:25,965 INFO [train.py:901] (2/4) Epoch 14, batch 7000, loss[loss=0.1639, simple_loss=0.251, pruned_loss=0.03834, over 7223.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2998, pruned_loss=0.07175, over 1609274.01 frames. ], batch size: 16, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:46:28,777 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:46:28,796 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0966, 1.5641, 4.5188, 2.0376, 2.3851, 5.1807, 5.2182, 4.5196], + device='cuda:2'), covar=tensor([0.1113, 0.1714, 0.0225, 0.1810, 0.1052, 0.0156, 0.0323, 0.0534], + device='cuda:2'), in_proj_covar=tensor([0.0277, 0.0306, 0.0268, 0.0298, 0.0284, 0.0244, 0.0369, 0.0295], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:46:41,424 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7593, 3.7746, 3.4494, 1.9582, 3.3278, 3.3549, 3.4547, 3.1977], + device='cuda:2'), covar=tensor([0.1060, 0.0727, 0.1091, 0.4503, 0.1060, 0.1197, 0.1368, 0.0932], + device='cuda:2'), in_proj_covar=tensor([0.0485, 0.0397, 0.0403, 0.0500, 0.0396, 0.0400, 0.0390, 0.0346], + 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') +2023-02-06 15:46:59,912 INFO [train.py:901] (2/4) Epoch 14, batch 7050, loss[loss=0.2539, simple_loss=0.328, pruned_loss=0.08994, over 8673.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3003, pruned_loss=0.07203, over 1613049.36 frames. ], batch size: 39, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:47:15,925 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112153.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 15:47:21,792 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.608e+02 3.153e+02 4.211e+02 1.237e+03, threshold=6.307e+02, percent-clipped=12.0 +2023-02-06 15:47:35,265 INFO [train.py:901] (2/4) Epoch 14, batch 7100, loss[loss=0.1714, simple_loss=0.2523, pruned_loss=0.04522, over 8091.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2992, pruned_loss=0.07139, over 1609149.57 frames. ], batch size: 21, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:47:50,289 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:47:59,152 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 15:48:07,015 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1050, 3.9361, 2.5203, 2.8951, 2.8132, 1.9670, 2.7918, 3.1552], + device='cuda:2'), covar=tensor([0.1559, 0.0285, 0.0854, 0.0692, 0.0668, 0.1331, 0.1036, 0.0942], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0234, 0.0323, 0.0298, 0.0301, 0.0329, 0.0345, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:48:07,692 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112226.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:48:10,129 INFO [train.py:901] (2/4) Epoch 14, batch 7150, loss[loss=0.2392, simple_loss=0.3157, pruned_loss=0.08134, over 8606.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2995, pruned_loss=0.07136, over 1612406.01 frames. ], batch size: 39, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:48:16,274 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7036, 1.2922, 4.8841, 1.7893, 4.3531, 4.0897, 4.3625, 4.2921], + device='cuda:2'), covar=tensor([0.0476, 0.4456, 0.0411, 0.3531, 0.0968, 0.0837, 0.0521, 0.0529], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0599, 0.0625, 0.0566, 0.0638, 0.0550, 0.0542, 0.0603], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 15:48:31,549 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.371e+02 2.859e+02 3.664e+02 7.587e+02, threshold=5.717e+02, percent-clipped=3.0 +2023-02-06 15:48:35,312 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 15:48:35,576 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112268.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:48:43,202 INFO [train.py:901] (2/4) Epoch 14, batch 7200, loss[loss=0.2326, simple_loss=0.3142, pruned_loss=0.07552, over 8461.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2996, pruned_loss=0.07142, over 1610862.05 frames. ], batch size: 25, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:49:18,426 INFO [train.py:901] (2/4) Epoch 14, batch 7250, loss[loss=0.2097, simple_loss=0.2857, pruned_loss=0.06683, over 8328.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3014, pruned_loss=0.07211, over 1611572.46 frames. ], batch size: 25, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:49:39,910 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.708e+02 3.212e+02 3.989e+02 8.387e+02, threshold=6.424e+02, percent-clipped=5.0 +2023-02-06 15:49:52,049 INFO [train.py:901] (2/4) Epoch 14, batch 7300, loss[loss=0.2496, simple_loss=0.3361, pruned_loss=0.08161, over 8624.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3006, pruned_loss=0.07151, over 1609803.69 frames. ], batch size: 34, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:49:54,808 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8481, 1.7833, 2.5314, 1.6363, 1.1534, 2.5984, 0.4512, 1.2808], + device='cuda:2'), covar=tensor([0.2490, 0.1520, 0.0361, 0.1813, 0.3992, 0.0272, 0.3014, 0.2027], + device='cuda:2'), in_proj_covar=tensor([0.0172, 0.0175, 0.0107, 0.0215, 0.0256, 0.0111, 0.0161, 0.0173], + 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') +2023-02-06 15:50:11,933 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112407.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:50:26,718 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:50:28,007 INFO [train.py:901] (2/4) Epoch 14, batch 7350, loss[loss=0.1954, simple_loss=0.2749, pruned_loss=0.05791, over 7975.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3019, pruned_loss=0.07236, over 1612087.15 frames. ], batch size: 21, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:50:40,030 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 15:50:49,906 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.462e+02 2.972e+02 3.682e+02 1.093e+03, threshold=5.943e+02, percent-clipped=5.0 +2023-02-06 15:50:59,967 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 15:51:02,056 INFO [train.py:901] (2/4) Epoch 14, batch 7400, loss[loss=0.2585, simple_loss=0.337, pruned_loss=0.09003, over 8109.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3024, pruned_loss=0.07219, over 1611157.00 frames. ], batch size: 23, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:51:32,634 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112524.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:51:37,190 INFO [train.py:901] (2/4) Epoch 14, batch 7450, loss[loss=0.2451, simple_loss=0.3368, pruned_loss=0.07668, over 8845.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3033, pruned_loss=0.07271, over 1611645.02 frames. ], batch size: 32, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:51:41,775 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 15:51:46,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:51:50,987 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112549.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:51:53,939 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 15:51:59,931 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.546e+02 2.473e+02 3.112e+02 3.710e+02 6.215e+02, threshold=6.224e+02, percent-clipped=1.0 +2023-02-06 15:52:13,215 INFO [train.py:901] (2/4) Epoch 14, batch 7500, loss[loss=0.2169, simple_loss=0.3028, pruned_loss=0.06552, over 8466.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3025, pruned_loss=0.07251, over 1609421.54 frames. ], batch size: 25, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:52:13,390 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:52:23,790 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:52:47,695 INFO [train.py:901] (2/4) Epoch 14, batch 7550, loss[loss=0.2416, simple_loss=0.3178, pruned_loss=0.08272, over 8565.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3027, pruned_loss=0.07285, over 1610236.16 frames. ], batch size: 39, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:52:51,298 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:53:08,103 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5376, 1.3266, 2.2698, 1.0955, 1.9996, 2.3941, 2.5485, 2.0721], + device='cuda:2'), covar=tensor([0.0955, 0.1278, 0.0492, 0.2237, 0.0809, 0.0415, 0.0680, 0.0744], + device='cuda:2'), in_proj_covar=tensor([0.0278, 0.0308, 0.0270, 0.0301, 0.0288, 0.0248, 0.0372, 0.0296], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 15:53:11,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.402e+02 2.890e+02 3.643e+02 7.164e+02, threshold=5.781e+02, percent-clipped=3.0 +2023-02-06 15:53:23,622 INFO [train.py:901] (2/4) Epoch 14, batch 7600, loss[loss=0.2198, simple_loss=0.3013, pruned_loss=0.06917, over 8538.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3037, pruned_loss=0.07343, over 1613415.07 frames. ], batch size: 28, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:53:57,102 INFO [train.py:901] (2/4) Epoch 14, batch 7650, loss[loss=0.2684, simple_loss=0.3326, pruned_loss=0.1021, over 8568.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3027, pruned_loss=0.07309, over 1612348.33 frames. ], batch size: 39, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:54:11,112 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:54:18,473 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.546e+02 2.941e+02 3.649e+02 7.123e+02, threshold=5.882e+02, percent-clipped=5.0 +2023-02-06 15:54:32,364 INFO [train.py:901] (2/4) Epoch 14, batch 7700, loss[loss=0.2082, simple_loss=0.2998, pruned_loss=0.05837, over 8469.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3012, pruned_loss=0.07264, over 1609723.67 frames. ], batch size: 25, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:54:44,371 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 15:54:45,583 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:54:52,888 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 15:55:03,126 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:55:06,911 INFO [train.py:901] (2/4) Epoch 14, batch 7750, loss[loss=0.1942, simple_loss=0.2854, pruned_loss=0.05152, over 8288.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3009, pruned_loss=0.07279, over 1605249.07 frames. ], batch size: 23, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:55:28,269 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.453e+02 3.172e+02 4.245e+02 8.131e+02, threshold=6.343e+02, percent-clipped=10.0 +2023-02-06 15:55:30,967 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112866.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:55:41,028 INFO [train.py:901] (2/4) Epoch 14, batch 7800, loss[loss=0.2391, simple_loss=0.3203, pruned_loss=0.07893, over 8509.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3, pruned_loss=0.07243, over 1606562.68 frames. ], batch size: 28, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:56:02,573 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 15:56:12,172 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112924.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:16,236 INFO [train.py:901] (2/4) Epoch 14, batch 7850, loss[loss=0.2429, simple_loss=0.309, pruned_loss=0.08842, over 8096.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3017, pruned_loss=0.07319, over 1609043.40 frames. ], batch size: 21, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:56:22,358 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:22,423 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:37,784 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.504e+02 3.067e+02 3.726e+02 7.698e+02, threshold=6.135e+02, percent-clipped=2.0 +2023-02-06 15:56:49,064 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:49,646 INFO [train.py:901] (2/4) Epoch 14, batch 7900, loss[loss=0.1935, simple_loss=0.2818, pruned_loss=0.05265, over 8029.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3026, pruned_loss=0.07342, over 1613743.61 frames. ], batch size: 22, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:57:22,233 INFO [train.py:901] (2/4) Epoch 14, batch 7950, loss[loss=0.2275, simple_loss=0.3075, pruned_loss=0.07378, over 8021.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3021, pruned_loss=0.07323, over 1613144.88 frames. ], batch size: 22, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:57:28,306 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113039.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:57:38,047 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:57:43,248 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.654e+02 3.191e+02 4.041e+02 1.304e+03, threshold=6.382e+02, percent-clipped=5.0 +2023-02-06 15:57:55,257 INFO [train.py:901] (2/4) Epoch 14, batch 8000, loss[loss=0.2317, simple_loss=0.3035, pruned_loss=0.07996, over 7653.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3009, pruned_loss=0.07279, over 1612647.81 frames. ], batch size: 19, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:58:04,703 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:58:23,457 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:58:28,557 INFO [train.py:901] (2/4) Epoch 14, batch 8050, loss[loss=0.1948, simple_loss=0.2717, pruned_loss=0.05901, over 7246.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3003, pruned_loss=0.07364, over 1587135.20 frames. ], batch size: 16, lr: 5.36e-03, grad_scale: 16.0 +2023-02-06 15:58:40,129 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113147.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:58:49,769 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.323e+02 2.856e+02 3.288e+02 8.076e+02, threshold=5.712e+02, percent-clipped=1.0 +2023-02-06 15:59:01,309 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 15:59:06,257 INFO [train.py:901] (2/4) Epoch 15, batch 0, loss[loss=0.2079, simple_loss=0.2896, pruned_loss=0.06314, over 7814.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2896, pruned_loss=0.06314, over 7814.00 frames. ], batch size: 20, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 15:59:06,257 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 15:59:17,270 INFO [train.py:935] (2/4) Epoch 15, validation: loss=0.1825, simple_loss=0.283, pruned_loss=0.04098, over 944034.00 frames. +2023-02-06 15:59:17,270 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 15:59:32,303 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 15:59:48,307 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 +2023-02-06 15:59:49,089 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 15:59:51,453 INFO [train.py:901] (2/4) Epoch 15, batch 50, loss[loss=0.197, simple_loss=0.2768, pruned_loss=0.05863, over 7930.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3026, pruned_loss=0.07113, over 365608.40 frames. ], batch size: 20, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:00:08,691 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 16:00:21,174 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:00:27,786 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.549e+02 3.077e+02 3.582e+02 9.445e+02, threshold=6.153e+02, percent-clipped=5.0 +2023-02-06 16:00:28,491 INFO [train.py:901] (2/4) Epoch 15, batch 100, loss[loss=0.2298, simple_loss=0.312, pruned_loss=0.07384, over 8282.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3025, pruned_loss=0.07272, over 639345.74 frames. ], batch size: 23, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:00:29,902 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 16:00:39,305 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7139, 5.8051, 5.0449, 2.2919, 5.0237, 5.4679, 5.4138, 5.3265], + device='cuda:2'), covar=tensor([0.0660, 0.0522, 0.1046, 0.4778, 0.0847, 0.1014, 0.1129, 0.0661], + device='cuda:2'), in_proj_covar=tensor([0.0484, 0.0404, 0.0403, 0.0503, 0.0402, 0.0400, 0.0393, 0.0352], + 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') +2023-02-06 16:00:41,981 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:00:50,245 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:00,329 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113310.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:02,086 INFO [train.py:901] (2/4) Epoch 15, batch 150, loss[loss=0.1908, simple_loss=0.2668, pruned_loss=0.05742, over 7433.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3016, pruned_loss=0.0723, over 854900.90 frames. ], batch size: 17, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:01:06,864 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113320.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:17,347 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113335.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:28,773 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113350.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:32,820 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6480, 1.9703, 2.2013, 1.1957, 2.2776, 1.3747, 0.6950, 1.8896], + device='cuda:2'), covar=tensor([0.0619, 0.0287, 0.0228, 0.0555, 0.0325, 0.0818, 0.0722, 0.0289], + device='cuda:2'), in_proj_covar=tensor([0.0416, 0.0358, 0.0304, 0.0408, 0.0343, 0.0497, 0.0370, 0.0380], + device='cuda:2'), out_proj_covar=tensor([1.1544e-04, 9.6722e-05, 8.2157e-05, 1.1080e-04, 9.3424e-05, 1.4533e-04, + 1.0253e-04, 1.0406e-04], device='cuda:2') +2023-02-06 16:01:37,319 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 2.511e+02 3.032e+02 4.146e+02 1.005e+03, threshold=6.064e+02, percent-clipped=3.0 +2023-02-06 16:01:38,016 INFO [train.py:901] (2/4) Epoch 15, batch 200, loss[loss=0.2026, simple_loss=0.2891, pruned_loss=0.05806, over 8360.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3004, pruned_loss=0.07163, over 1027307.89 frames. ], batch size: 24, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:01:46,247 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113375.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:02:01,346 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:02:11,078 INFO [train.py:901] (2/4) Epoch 15, batch 250, loss[loss=0.2239, simple_loss=0.3084, pruned_loss=0.06966, over 8301.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.304, pruned_loss=0.0741, over 1161453.26 frames. ], batch size: 23, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:02:19,373 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 16:02:28,581 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 16:02:43,772 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.666e+02 3.062e+02 4.026e+02 8.735e+02, threshold=6.124e+02, percent-clipped=4.0 +2023-02-06 16:02:44,412 INFO [train.py:901] (2/4) Epoch 15, batch 300, loss[loss=0.2678, simple_loss=0.3464, pruned_loss=0.09459, over 8585.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3037, pruned_loss=0.07383, over 1262894.82 frames. ], batch size: 31, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:03:15,226 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 16:03:19,321 INFO [train.py:901] (2/4) Epoch 15, batch 350, loss[loss=0.2829, simple_loss=0.3447, pruned_loss=0.1105, over 6608.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3036, pruned_loss=0.07404, over 1341678.86 frames. ], batch size: 71, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:03:52,044 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.415e+02 3.115e+02 3.728e+02 6.919e+02, threshold=6.229e+02, percent-clipped=2.0 +2023-02-06 16:03:52,739 INFO [train.py:901] (2/4) Epoch 15, batch 400, loss[loss=0.2279, simple_loss=0.3078, pruned_loss=0.07395, over 8203.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3026, pruned_loss=0.07348, over 1403570.50 frames. ], batch size: 23, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:04:17,489 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113596.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:04:28,817 INFO [train.py:901] (2/4) Epoch 15, batch 450, loss[loss=0.2204, simple_loss=0.2922, pruned_loss=0.07426, over 7652.00 frames. ], tot_loss[loss=0.225, simple_loss=0.303, pruned_loss=0.07354, over 1451751.51 frames. ], batch size: 19, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:04:30,186 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1341, 1.5555, 1.7845, 1.3996, 0.9524, 1.6175, 1.9086, 1.6459], + device='cuda:2'), covar=tensor([0.0478, 0.1143, 0.1531, 0.1313, 0.0590, 0.1341, 0.0631, 0.0622], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0155, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 16:04:43,286 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:04:56,091 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113654.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:01,031 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.616e+02 3.268e+02 4.141e+02 9.119e+02, threshold=6.536e+02, percent-clipped=2.0 +2023-02-06 16:05:01,749 INFO [train.py:901] (2/4) Epoch 15, batch 500, loss[loss=0.239, simple_loss=0.3124, pruned_loss=0.08284, over 6805.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3018, pruned_loss=0.07321, over 1486059.21 frames. ], batch size: 71, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:05:02,562 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:12,317 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113679.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:35,420 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113711.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:36,634 INFO [train.py:901] (2/4) Epoch 15, batch 550, loss[loss=0.2506, simple_loss=0.3186, pruned_loss=0.09129, over 8036.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3017, pruned_loss=0.07329, over 1514411.76 frames. ], batch size: 22, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:06:09,824 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.516e+02 3.119e+02 4.209e+02 9.524e+02, threshold=6.239e+02, percent-clipped=4.0 +2023-02-06 16:06:10,533 INFO [train.py:901] (2/4) Epoch 15, batch 600, loss[loss=0.1858, simple_loss=0.2648, pruned_loss=0.05336, over 7436.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3017, pruned_loss=0.07292, over 1535817.22 frames. ], batch size: 17, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:06:17,857 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9834, 1.4728, 3.3205, 1.4908, 2.2782, 3.5952, 3.6573, 3.1079], + device='cuda:2'), covar=tensor([0.0968, 0.1663, 0.0327, 0.1985, 0.1041, 0.0261, 0.0595, 0.0574], + device='cuda:2'), in_proj_covar=tensor([0.0278, 0.0308, 0.0273, 0.0301, 0.0288, 0.0247, 0.0377, 0.0299], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:06:24,220 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 16:06:44,312 INFO [train.py:901] (2/4) Epoch 15, batch 650, loss[loss=0.2164, simple_loss=0.2968, pruned_loss=0.06802, over 8497.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3028, pruned_loss=0.07299, over 1558817.20 frames. ], batch size: 26, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:06:53,836 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6905, 3.0969, 2.4006, 4.1227, 1.7858, 2.1494, 2.6148, 3.1485], + device='cuda:2'), covar=tensor([0.0671, 0.0789, 0.0923, 0.0192, 0.1112, 0.1270, 0.0935, 0.0742], + device='cuda:2'), in_proj_covar=tensor([0.0226, 0.0201, 0.0247, 0.0207, 0.0208, 0.0244, 0.0249, 0.0208], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-06 16:07:19,220 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 2.270e+02 2.767e+02 3.649e+02 9.673e+02, threshold=5.535e+02, percent-clipped=4.0 +2023-02-06 16:07:19,888 INFO [train.py:901] (2/4) Epoch 15, batch 700, loss[loss=0.3217, simple_loss=0.3656, pruned_loss=0.1389, over 6960.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3026, pruned_loss=0.07277, over 1572423.19 frames. ], batch size: 71, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:07:53,444 INFO [train.py:901] (2/4) Epoch 15, batch 750, loss[loss=0.2406, simple_loss=0.3125, pruned_loss=0.08434, over 8196.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3028, pruned_loss=0.07289, over 1584211.49 frames. ], batch size: 23, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:08:11,189 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 16:08:20,441 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 16:08:29,183 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 2.237e+02 2.791e+02 3.511e+02 6.350e+02, threshold=5.582e+02, percent-clipped=4.0 +2023-02-06 16:08:29,879 INFO [train.py:901] (2/4) Epoch 15, batch 800, loss[loss=0.2036, simple_loss=0.2982, pruned_loss=0.05446, over 8292.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3004, pruned_loss=0.07175, over 1585391.12 frames. ], batch size: 23, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:08:32,849 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:08:40,773 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:08:50,093 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113992.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:09:01,983 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114008.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:09:05,302 INFO [train.py:901] (2/4) Epoch 15, batch 850, loss[loss=0.1866, simple_loss=0.2799, pruned_loss=0.04665, over 8334.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3004, pruned_loss=0.07146, over 1594222.37 frames. ], batch size: 26, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:09:12,128 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0595, 1.2480, 1.2106, 0.5235, 1.2184, 1.0099, 0.0467, 1.1874], + device='cuda:2'), covar=tensor([0.0334, 0.0277, 0.0238, 0.0450, 0.0335, 0.0736, 0.0630, 0.0240], + device='cuda:2'), in_proj_covar=tensor([0.0419, 0.0359, 0.0305, 0.0411, 0.0342, 0.0499, 0.0370, 0.0381], + device='cuda:2'), out_proj_covar=tensor([1.1633e-04, 9.6711e-05, 8.2020e-05, 1.1161e-04, 9.3094e-05, 1.4561e-04, + 1.0267e-04, 1.0432e-04], device='cuda:2') +2023-02-06 16:09:32,139 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2691, 1.2558, 1.4266, 1.2028, 0.7061, 1.2849, 1.1507, 1.0085], + device='cuda:2'), covar=tensor([0.0550, 0.1170, 0.1570, 0.1293, 0.0595, 0.1360, 0.0683, 0.0632], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0156, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 16:09:39,413 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.394e+02 2.826e+02 3.443e+02 6.296e+02, threshold=5.653e+02, percent-clipped=1.0 +2023-02-06 16:09:40,788 INFO [train.py:901] (2/4) Epoch 15, batch 900, loss[loss=0.2129, simple_loss=0.3061, pruned_loss=0.05986, over 8352.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3012, pruned_loss=0.0721, over 1598913.29 frames. ], batch size: 24, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:09:45,805 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 16:09:54,016 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4226, 2.0018, 2.8781, 2.2804, 2.7716, 2.2801, 1.9779, 1.3718], + device='cuda:2'), covar=tensor([0.4208, 0.4430, 0.1427, 0.2832, 0.1964, 0.2438, 0.1680, 0.4806], + device='cuda:2'), in_proj_covar=tensor([0.0890, 0.0903, 0.0742, 0.0872, 0.0938, 0.0828, 0.0711, 0.0781], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 16:09:58,192 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-06 16:10:02,636 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:10:15,159 INFO [train.py:901] (2/4) Epoch 15, batch 950, loss[loss=0.2492, simple_loss=0.3267, pruned_loss=0.0859, over 8362.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3005, pruned_loss=0.07201, over 1600080.13 frames. ], batch size: 24, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:10:21,911 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114123.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:10:36,821 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114145.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:10:38,931 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4415, 1.3746, 1.6751, 1.3112, 1.0699, 1.6959, 0.1640, 1.1361], + device='cuda:2'), covar=tensor([0.1995, 0.1689, 0.0554, 0.1165, 0.3417, 0.0488, 0.2793, 0.1659], + device='cuda:2'), in_proj_covar=tensor([0.0174, 0.0179, 0.0111, 0.0217, 0.0259, 0.0114, 0.0165, 0.0175], + 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') +2023-02-06 16:10:39,440 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 16:10:45,954 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4435, 2.7443, 1.9397, 2.1884, 2.2590, 1.4576, 2.0590, 2.0636], + device='cuda:2'), covar=tensor([0.1633, 0.0349, 0.0995, 0.0612, 0.0628, 0.1428, 0.0936, 0.0985], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0233, 0.0327, 0.0300, 0.0303, 0.0327, 0.0344, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:10:49,208 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.449e+02 2.913e+02 3.851e+02 8.356e+02, threshold=5.826e+02, percent-clipped=3.0 +2023-02-06 16:10:49,925 INFO [train.py:901] (2/4) Epoch 15, batch 1000, loss[loss=0.1945, simple_loss=0.2872, pruned_loss=0.05086, over 8292.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.3009, pruned_loss=0.07205, over 1602414.33 frames. ], batch size: 23, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:11:00,772 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9705, 4.0179, 2.5561, 2.7252, 2.8625, 2.1425, 2.8410, 2.9742], + device='cuda:2'), covar=tensor([0.1601, 0.0256, 0.0898, 0.0760, 0.0705, 0.1171, 0.0889, 0.1020], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0233, 0.0326, 0.0299, 0.0303, 0.0327, 0.0343, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:11:14,220 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 16:11:25,588 INFO [train.py:901] (2/4) Epoch 15, batch 1050, loss[loss=0.2054, simple_loss=0.2737, pruned_loss=0.06852, over 7660.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3014, pruned_loss=0.07272, over 1609882.32 frames. ], batch size: 19, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:11:25,602 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 16:11:57,606 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.504e+02 3.058e+02 3.938e+02 1.189e+03, threshold=6.116e+02, percent-clipped=4.0 +2023-02-06 16:11:58,328 INFO [train.py:901] (2/4) Epoch 15, batch 1100, loss[loss=0.2319, simple_loss=0.3244, pruned_loss=0.06971, over 8362.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3005, pruned_loss=0.07271, over 1610518.29 frames. ], batch size: 24, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:12:11,299 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 16:12:26,183 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-06 16:12:32,998 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 16:12:33,900 INFO [train.py:901] (2/4) Epoch 15, batch 1150, loss[loss=0.1889, simple_loss=0.2824, pruned_loss=0.04766, over 7800.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2998, pruned_loss=0.07211, over 1613428.14 frames. ], batch size: 20, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:12:38,620 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 16:12:59,538 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:07,354 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.463e+02 3.139e+02 3.955e+02 6.139e+02, threshold=6.277e+02, percent-clipped=1.0 +2023-02-06 16:13:07,982 INFO [train.py:901] (2/4) Epoch 15, batch 1200, loss[loss=0.2032, simple_loss=0.2673, pruned_loss=0.0696, over 7682.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2997, pruned_loss=0.07208, over 1611205.77 frames. ], batch size: 18, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:13:16,141 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114375.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:18,728 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114379.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:36,375 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114404.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:42,788 INFO [train.py:901] (2/4) Epoch 15, batch 1250, loss[loss=0.2297, simple_loss=0.3119, pruned_loss=0.07373, over 8456.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2992, pruned_loss=0.07151, over 1615119.29 frames. ], batch size: 27, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:14:16,861 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.591e+02 3.148e+02 4.129e+02 1.085e+03, threshold=6.295e+02, percent-clipped=6.0 +2023-02-06 16:14:17,473 INFO [train.py:901] (2/4) Epoch 15, batch 1300, loss[loss=0.243, simple_loss=0.3263, pruned_loss=0.07986, over 8560.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2998, pruned_loss=0.07206, over 1616011.64 frames. ], batch size: 31, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:14:35,221 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114489.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:14:51,240 INFO [train.py:901] (2/4) Epoch 15, batch 1350, loss[loss=0.2311, simple_loss=0.3101, pruned_loss=0.07604, over 8285.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.3, pruned_loss=0.07156, over 1617962.03 frames. ], batch size: 23, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:15:08,701 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1053, 4.1509, 3.7312, 1.5738, 3.5648, 3.7359, 3.7772, 3.4701], + device='cuda:2'), covar=tensor([0.0908, 0.0627, 0.1051, 0.5569, 0.0955, 0.0981, 0.1289, 0.1001], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0398, 0.0399, 0.0498, 0.0395, 0.0398, 0.0381, 0.0348], + 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') +2023-02-06 16:15:12,373 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 16:15:26,456 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.434e+02 2.903e+02 3.628e+02 5.826e+02, threshold=5.807e+02, percent-clipped=0.0 +2023-02-06 16:15:27,128 INFO [train.py:901] (2/4) Epoch 15, batch 1400, loss[loss=0.2546, simple_loss=0.3331, pruned_loss=0.08803, over 8646.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2994, pruned_loss=0.07091, over 1618164.81 frames. ], batch size: 34, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:15:54,551 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114604.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:16:00,677 INFO [train.py:901] (2/4) Epoch 15, batch 1450, loss[loss=0.2007, simple_loss=0.2706, pruned_loss=0.06545, over 7450.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2998, pruned_loss=0.07104, over 1614879.39 frames. ], batch size: 17, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:16:08,831 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 16:16:36,177 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.414e+02 3.068e+02 3.744e+02 6.619e+02, threshold=6.136e+02, percent-clipped=3.0 +2023-02-06 16:16:36,887 INFO [train.py:901] (2/4) Epoch 15, batch 1500, loss[loss=0.1811, simple_loss=0.2577, pruned_loss=0.05225, over 7702.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.3004, pruned_loss=0.07091, over 1616345.52 frames. ], batch size: 18, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:16:48,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3541, 1.3278, 4.5299, 1.6676, 3.9587, 3.7987, 4.0643, 3.9393], + device='cuda:2'), covar=tensor([0.0630, 0.4973, 0.0548, 0.4061, 0.1261, 0.0885, 0.0618, 0.0677], + device='cuda:2'), in_proj_covar=tensor([0.0549, 0.0609, 0.0634, 0.0576, 0.0651, 0.0554, 0.0548, 0.0614], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 16:16:58,918 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4401, 1.8095, 2.7922, 1.2796, 2.0224, 1.9072, 1.5061, 1.9806], + device='cuda:2'), covar=tensor([0.1747, 0.2197, 0.0710, 0.4020, 0.1529, 0.2778, 0.2008, 0.1932], + device='cuda:2'), in_proj_covar=tensor([0.0498, 0.0551, 0.0537, 0.0601, 0.0622, 0.0565, 0.0493, 0.0621], + 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') +2023-02-06 16:17:00,203 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5074, 1.7948, 1.9102, 1.1910, 1.9762, 1.4185, 0.4084, 1.6535], + device='cuda:2'), covar=tensor([0.0393, 0.0280, 0.0200, 0.0333, 0.0296, 0.0662, 0.0612, 0.0210], + device='cuda:2'), in_proj_covar=tensor([0.0415, 0.0354, 0.0302, 0.0406, 0.0338, 0.0493, 0.0366, 0.0378], + device='cuda:2'), out_proj_covar=tensor([1.1498e-04, 9.5558e-05, 8.1249e-05, 1.1026e-04, 9.1970e-05, 1.4356e-04, + 1.0150e-04, 1.0324e-04], device='cuda:2') +2023-02-06 16:17:09,892 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 16:17:11,507 INFO [train.py:901] (2/4) Epoch 15, batch 1550, loss[loss=0.2224, simple_loss=0.3089, pruned_loss=0.06798, over 8361.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.3009, pruned_loss=0.07102, over 1621741.90 frames. ], batch size: 24, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:17:26,183 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114734.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:17:45,709 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.278e+02 2.828e+02 3.736e+02 6.971e+02, threshold=5.655e+02, percent-clipped=1.0 +2023-02-06 16:17:46,443 INFO [train.py:901] (2/4) Epoch 15, batch 1600, loss[loss=0.2812, simple_loss=0.3373, pruned_loss=0.1125, over 7513.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.3001, pruned_loss=0.07045, over 1619220.95 frames. ], batch size: 72, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:18:22,446 INFO [train.py:901] (2/4) Epoch 15, batch 1650, loss[loss=0.1745, simple_loss=0.255, pruned_loss=0.047, over 7650.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2982, pruned_loss=0.0698, over 1617162.47 frames. ], batch size: 19, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:18:55,126 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114860.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:18:56,274 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.429e+02 2.845e+02 3.384e+02 6.803e+02, threshold=5.691e+02, percent-clipped=1.0 +2023-02-06 16:18:56,970 INFO [train.py:901] (2/4) Epoch 15, batch 1700, loss[loss=0.2125, simple_loss=0.2775, pruned_loss=0.07374, over 7437.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2979, pruned_loss=0.07, over 1616043.54 frames. ], batch size: 17, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:19:12,811 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114885.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:19:16,114 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114889.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:19:32,921 INFO [train.py:901] (2/4) Epoch 15, batch 1750, loss[loss=0.2141, simple_loss=0.2837, pruned_loss=0.0722, over 7444.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2985, pruned_loss=0.07018, over 1619452.72 frames. ], batch size: 17, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:19:45,254 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7419, 1.7529, 4.8926, 1.9673, 4.3796, 4.0626, 4.4484, 4.3579], + device='cuda:2'), covar=tensor([0.0480, 0.4221, 0.0379, 0.3555, 0.0905, 0.0852, 0.0487, 0.0506], + device='cuda:2'), in_proj_covar=tensor([0.0553, 0.0611, 0.0639, 0.0581, 0.0653, 0.0560, 0.0553, 0.0618], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 16:20:06,955 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.463e+02 2.435e+02 3.025e+02 3.758e+02 7.531e+02, threshold=6.050e+02, percent-clipped=3.0 +2023-02-06 16:20:07,575 INFO [train.py:901] (2/4) Epoch 15, batch 1800, loss[loss=0.2319, simple_loss=0.3076, pruned_loss=0.07811, over 8466.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2968, pruned_loss=0.06925, over 1613108.19 frames. ], batch size: 28, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:20:43,782 INFO [train.py:901] (2/4) Epoch 15, batch 1850, loss[loss=0.205, simple_loss=0.2945, pruned_loss=0.05773, over 8508.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2984, pruned_loss=0.06998, over 1615640.53 frames. ], batch size: 26, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:20:48,473 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 16:20:49,523 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0510, 2.4023, 2.6850, 1.3245, 2.8103, 1.5963, 1.5274, 1.9132], + device='cuda:2'), covar=tensor([0.0668, 0.0313, 0.0233, 0.0589, 0.0312, 0.0668, 0.0728, 0.0445], + device='cuda:2'), in_proj_covar=tensor([0.0420, 0.0358, 0.0306, 0.0410, 0.0341, 0.0495, 0.0369, 0.0381], + device='cuda:2'), out_proj_covar=tensor([1.1640e-04, 9.6495e-05, 8.2128e-05, 1.1120e-04, 9.2733e-05, 1.4421e-04, + 1.0221e-04, 1.0411e-04], device='cuda:2') +2023-02-06 16:20:59,009 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7595, 1.6949, 2.2577, 1.6453, 1.2298, 2.2921, 0.4198, 1.3895], + device='cuda:2'), covar=tensor([0.2509, 0.1798, 0.0448, 0.1758, 0.3841, 0.0462, 0.2903, 0.1867], + device='cuda:2'), in_proj_covar=tensor([0.0174, 0.0180, 0.0112, 0.0219, 0.0262, 0.0117, 0.0166, 0.0177], + 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') +2023-02-06 16:21:05,807 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1353, 1.6684, 3.4596, 1.4309, 2.5314, 3.7839, 3.8905, 3.2940], + device='cuda:2'), covar=tensor([0.1017, 0.1650, 0.0346, 0.2147, 0.0988, 0.0212, 0.0472, 0.0531], + device='cuda:2'), in_proj_covar=tensor([0.0276, 0.0303, 0.0267, 0.0296, 0.0282, 0.0244, 0.0371, 0.0295], + 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') +2023-02-06 16:21:17,801 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.659e+02 3.189e+02 4.139e+02 1.250e+03, threshold=6.379e+02, percent-clipped=4.0 +2023-02-06 16:21:18,502 INFO [train.py:901] (2/4) Epoch 15, batch 1900, loss[loss=0.2107, simple_loss=0.2919, pruned_loss=0.06475, over 8335.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2982, pruned_loss=0.06957, over 1619264.28 frames. ], batch size: 26, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:21:28,842 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:21:46,756 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:21:50,087 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 16:21:53,561 INFO [train.py:901] (2/4) Epoch 15, batch 1950, loss[loss=0.2637, simple_loss=0.3336, pruned_loss=0.09695, over 8339.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2999, pruned_loss=0.07001, over 1622798.71 frames. ], batch size: 48, lr: 5.13e-03, grad_scale: 32.0 +2023-02-06 16:22:04,496 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 16:22:23,220 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 16:22:28,420 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.421e+02 3.112e+02 3.916e+02 6.433e+02, threshold=6.224e+02, percent-clipped=1.0 +2023-02-06 16:22:29,125 INFO [train.py:901] (2/4) Epoch 15, batch 2000, loss[loss=0.2097, simple_loss=0.2963, pruned_loss=0.06158, over 8348.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.3009, pruned_loss=0.07085, over 1622133.75 frames. ], batch size: 24, lr: 5.13e-03, grad_scale: 32.0 +2023-02-06 16:22:49,747 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:23:03,503 INFO [train.py:901] (2/4) Epoch 15, batch 2050, loss[loss=0.2247, simple_loss=0.2918, pruned_loss=0.07876, over 7262.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2995, pruned_loss=0.07031, over 1618680.10 frames. ], batch size: 16, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:23:18,037 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115233.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:23:39,484 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.382e+02 2.963e+02 3.753e+02 6.860e+02, threshold=5.925e+02, percent-clipped=2.0 +2023-02-06 16:23:39,504 INFO [train.py:901] (2/4) Epoch 15, batch 2100, loss[loss=0.1684, simple_loss=0.2512, pruned_loss=0.04277, over 7720.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2985, pruned_loss=0.0699, over 1614923.72 frames. ], batch size: 18, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:23:51,102 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9091, 2.6104, 3.4420, 1.9846, 1.7517, 3.5134, 0.5568, 2.0736], + device='cuda:2'), covar=tensor([0.1812, 0.1447, 0.0347, 0.2464, 0.3886, 0.0356, 0.3328, 0.1789], + device='cuda:2'), in_proj_covar=tensor([0.0175, 0.0181, 0.0112, 0.0220, 0.0264, 0.0117, 0.0166, 0.0178], + 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') +2023-02-06 16:24:13,866 INFO [train.py:901] (2/4) Epoch 15, batch 2150, loss[loss=0.2255, simple_loss=0.3159, pruned_loss=0.06757, over 8449.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2997, pruned_loss=0.07098, over 1614006.80 frames. ], batch size: 27, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:24:37,892 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115348.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:24:49,119 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.532e+02 3.093e+02 4.065e+02 1.254e+03, threshold=6.185e+02, percent-clipped=7.0 +2023-02-06 16:24:49,140 INFO [train.py:901] (2/4) Epoch 15, batch 2200, loss[loss=0.1924, simple_loss=0.2786, pruned_loss=0.05311, over 8248.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2988, pruned_loss=0.07012, over 1612349.23 frames. ], batch size: 22, lr: 5.12e-03, grad_scale: 16.0 +2023-02-06 16:25:07,044 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:07,718 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:24,252 INFO [train.py:901] (2/4) Epoch 15, batch 2250, loss[loss=0.2105, simple_loss=0.298, pruned_loss=0.06156, over 8649.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2995, pruned_loss=0.0711, over 1611822.57 frames. ], batch size: 27, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:25:31,073 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2664, 2.7611, 3.1820, 1.7788, 3.3612, 1.9243, 1.4903, 2.1653], + device='cuda:2'), covar=tensor([0.0735, 0.0298, 0.0203, 0.0597, 0.0334, 0.0792, 0.0790, 0.0437], + device='cuda:2'), in_proj_covar=tensor([0.0424, 0.0362, 0.0309, 0.0417, 0.0346, 0.0504, 0.0377, 0.0388], + device='cuda:2'), out_proj_covar=tensor([1.1748e-04, 9.7619e-05, 8.3126e-05, 1.1313e-04, 9.4066e-05, 1.4695e-04, + 1.0433e-04, 1.0588e-04], device='cuda:2') +2023-02-06 16:25:48,106 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115448.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:48,941 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115449.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:58,282 INFO [train.py:901] (2/4) Epoch 15, batch 2300, loss[loss=0.2421, simple_loss=0.3249, pruned_loss=0.07965, over 8191.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3008, pruned_loss=0.07186, over 1618362.69 frames. ], batch size: 23, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:25:58,953 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 2.502e+02 3.175e+02 3.927e+02 9.067e+02, threshold=6.350e+02, percent-clipped=5.0 +2023-02-06 16:26:07,458 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5462, 4.4913, 4.1325, 2.1432, 4.0376, 4.1274, 4.1280, 3.9272], + device='cuda:2'), covar=tensor([0.0659, 0.0489, 0.0921, 0.4292, 0.0813, 0.0965, 0.1156, 0.0834], + device='cuda:2'), in_proj_covar=tensor([0.0491, 0.0405, 0.0407, 0.0506, 0.0405, 0.0405, 0.0392, 0.0353], + 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') +2023-02-06 16:26:07,559 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:26:16,645 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9675, 2.3755, 1.9980, 2.9607, 1.5291, 1.5729, 2.1785, 2.4999], + device='cuda:2'), covar=tensor([0.0853, 0.0781, 0.0928, 0.0428, 0.1167, 0.1524, 0.0900, 0.0816], + device='cuda:2'), in_proj_covar=tensor([0.0227, 0.0202, 0.0245, 0.0209, 0.0209, 0.0246, 0.0250, 0.0210], + 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') +2023-02-06 16:26:17,334 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9644, 1.5813, 3.2428, 1.3265, 2.2147, 3.4238, 3.6408, 2.9890], + device='cuda:2'), covar=tensor([0.1126, 0.1761, 0.0342, 0.2275, 0.1085, 0.0295, 0.0476, 0.0621], + device='cuda:2'), in_proj_covar=tensor([0.0276, 0.0306, 0.0271, 0.0297, 0.0286, 0.0247, 0.0376, 0.0299], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:26:34,675 INFO [train.py:901] (2/4) Epoch 15, batch 2350, loss[loss=0.2034, simple_loss=0.2845, pruned_loss=0.06113, over 8319.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3005, pruned_loss=0.07149, over 1620567.73 frames. ], batch size: 25, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:27:02,129 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0594, 1.2221, 1.1586, 0.6679, 1.1894, 0.9929, 0.1219, 1.1862], + device='cuda:2'), covar=tensor([0.0318, 0.0270, 0.0225, 0.0399, 0.0322, 0.0761, 0.0630, 0.0222], + device='cuda:2'), in_proj_covar=tensor([0.0420, 0.0358, 0.0308, 0.0413, 0.0344, 0.0500, 0.0373, 0.0384], + device='cuda:2'), out_proj_covar=tensor([1.1646e-04, 9.6316e-05, 8.2875e-05, 1.1209e-04, 9.3527e-05, 1.4545e-04, + 1.0325e-04, 1.0486e-04], device='cuda:2') +2023-02-06 16:27:03,472 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6999, 1.7787, 1.6025, 1.9040, 1.2977, 1.5104, 1.6944, 1.8889], + device='cuda:2'), covar=tensor([0.0588, 0.0686, 0.0785, 0.0604, 0.0898, 0.0953, 0.0655, 0.0539], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0203, 0.0247, 0.0210, 0.0209, 0.0247, 0.0251, 0.0211], + 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') +2023-02-06 16:27:09,319 INFO [train.py:901] (2/4) Epoch 15, batch 2400, loss[loss=0.2616, simple_loss=0.3342, pruned_loss=0.09452, over 8593.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3009, pruned_loss=0.07213, over 1620202.42 frames. ], batch size: 39, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:27:09,505 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115563.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:27:10,002 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.542e+02 3.047e+02 3.524e+02 9.073e+02, threshold=6.095e+02, percent-clipped=1.0 +2023-02-06 16:27:32,718 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7326, 2.8023, 2.0442, 2.3399, 2.3293, 1.7563, 2.1882, 2.3614], + device='cuda:2'), covar=tensor([0.1406, 0.0280, 0.0940, 0.0618, 0.0619, 0.1276, 0.0871, 0.0823], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0230, 0.0327, 0.0303, 0.0302, 0.0332, 0.0345, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:27:39,713 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115604.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:27:45,572 INFO [train.py:901] (2/4) Epoch 15, batch 2450, loss[loss=0.1798, simple_loss=0.255, pruned_loss=0.05231, over 7224.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3, pruned_loss=0.0718, over 1616667.18 frames. ], batch size: 16, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:27:56,565 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115629.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:27:59,314 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4161, 1.9693, 3.1295, 1.2629, 2.2326, 1.8444, 1.5660, 2.2039], + device='cuda:2'), covar=tensor([0.1926, 0.2329, 0.0848, 0.4319, 0.1861, 0.3139, 0.2057, 0.2358], + device='cuda:2'), in_proj_covar=tensor([0.0494, 0.0547, 0.0537, 0.0600, 0.0619, 0.0563, 0.0492, 0.0615], + 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') +2023-02-06 16:28:11,489 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1862, 1.2089, 1.4999, 1.1350, 0.6472, 1.2659, 1.1100, 0.9326], + device='cuda:2'), covar=tensor([0.0552, 0.1337, 0.1670, 0.1465, 0.0598, 0.1560, 0.0732, 0.0717], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0157, 0.0102, 0.0161, 0.0115, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 16:28:19,898 INFO [train.py:901] (2/4) Epoch 15, batch 2500, loss[loss=0.2299, simple_loss=0.2954, pruned_loss=0.08223, over 7926.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3002, pruned_loss=0.07171, over 1617651.89 frames. ], batch size: 20, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:28:20,557 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.367e+02 2.686e+02 3.697e+02 9.165e+02, threshold=5.372e+02, percent-clipped=5.0 +2023-02-06 16:28:25,817 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.28 vs. limit=5.0 +2023-02-06 16:28:48,720 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9556, 1.8947, 2.5016, 1.6491, 1.2847, 2.4690, 0.3211, 1.4651], + device='cuda:2'), covar=tensor([0.2345, 0.1584, 0.0349, 0.1983, 0.3853, 0.0451, 0.3038, 0.1827], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0177, 0.0110, 0.0214, 0.0258, 0.0115, 0.0163, 0.0174], + 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') +2023-02-06 16:28:49,386 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4561, 1.4093, 1.7370, 1.2570, 0.9249, 1.7381, 0.1145, 1.1778], + device='cuda:2'), covar=tensor([0.2324, 0.1677, 0.0567, 0.1572, 0.3961, 0.0505, 0.2967, 0.1578], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0177, 0.0110, 0.0214, 0.0258, 0.0115, 0.0163, 0.0174], + 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') +2023-02-06 16:28:55,234 INFO [train.py:901] (2/4) Epoch 15, batch 2550, loss[loss=0.2138, simple_loss=0.2952, pruned_loss=0.06622, over 8096.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3, pruned_loss=0.07202, over 1617096.88 frames. ], batch size: 23, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:29:08,940 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115732.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:29:09,205 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.31 vs. limit=5.0 +2023-02-06 16:29:09,605 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:29:13,049 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 16:29:30,405 INFO [train.py:901] (2/4) Epoch 15, batch 2600, loss[loss=0.235, simple_loss=0.3104, pruned_loss=0.07979, over 8583.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2983, pruned_loss=0.0712, over 1611312.57 frames. ], batch size: 31, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:29:31,072 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.427e+02 3.148e+02 3.839e+02 8.607e+02, threshold=6.296e+02, percent-clipped=3.0 +2023-02-06 16:29:49,239 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0653, 2.6935, 3.4797, 1.9790, 1.7813, 3.5944, 0.6180, 2.0394], + device='cuda:2'), covar=tensor([0.1481, 0.1108, 0.0313, 0.2284, 0.3498, 0.0346, 0.2933, 0.1554], + device='cuda:2'), in_proj_covar=tensor([0.0170, 0.0176, 0.0110, 0.0213, 0.0257, 0.0115, 0.0162, 0.0174], + 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') +2023-02-06 16:29:59,661 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 16:30:04,163 INFO [train.py:901] (2/4) Epoch 15, batch 2650, loss[loss=0.2274, simple_loss=0.286, pruned_loss=0.08441, over 7220.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2979, pruned_loss=0.07087, over 1612627.06 frames. ], batch size: 16, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:30:08,492 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:27,405 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115844.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:29,373 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115847.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:30,055 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:39,731 INFO [train.py:901] (2/4) Epoch 15, batch 2700, loss[loss=0.2115, simple_loss=0.2888, pruned_loss=0.06717, over 7509.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2969, pruned_loss=0.07022, over 1611842.10 frames. ], batch size: 18, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:30:40,393 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.337e+02 2.718e+02 3.606e+02 6.832e+02, threshold=5.436e+02, percent-clipped=3.0 +2023-02-06 16:30:45,208 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1054, 1.2478, 4.3061, 1.6783, 3.7909, 3.5928, 3.8475, 3.7007], + device='cuda:2'), covar=tensor([0.0586, 0.4667, 0.0557, 0.3619, 0.1153, 0.0978, 0.0585, 0.0714], + device='cuda:2'), in_proj_covar=tensor([0.0557, 0.0613, 0.0639, 0.0583, 0.0656, 0.0564, 0.0557, 0.0620], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 16:30:50,043 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0952, 2.5657, 2.9075, 1.3813, 2.9812, 1.7099, 1.4732, 2.0661], + device='cuda:2'), covar=tensor([0.0775, 0.0354, 0.0210, 0.0676, 0.0362, 0.0853, 0.0868, 0.0505], + device='cuda:2'), in_proj_covar=tensor([0.0429, 0.0366, 0.0314, 0.0422, 0.0353, 0.0510, 0.0379, 0.0392], + device='cuda:2'), out_proj_covar=tensor([1.1883e-04, 9.8599e-05, 8.4276e-05, 1.1444e-04, 9.6013e-05, 1.4856e-04, + 1.0471e-04, 1.0695e-04], device='cuda:2') +2023-02-06 16:31:12,009 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115910.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 16:31:13,863 INFO [train.py:901] (2/4) Epoch 15, batch 2750, loss[loss=0.2327, simple_loss=0.3185, pruned_loss=0.07339, over 8494.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2974, pruned_loss=0.07043, over 1608296.50 frames. ], batch size: 26, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:31:19,885 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 16:31:29,674 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 16:31:49,502 INFO [train.py:901] (2/4) Epoch 15, batch 2800, loss[loss=0.244, simple_loss=0.323, pruned_loss=0.08255, over 8356.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2974, pruned_loss=0.07048, over 1608385.47 frames. ], batch size: 24, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:31:50,150 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.517e+02 2.986e+02 3.677e+02 9.071e+02, threshold=5.972e+02, percent-clipped=5.0 +2023-02-06 16:32:24,937 INFO [train.py:901] (2/4) Epoch 15, batch 2850, loss[loss=0.2243, simple_loss=0.3014, pruned_loss=0.07356, over 8462.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2976, pruned_loss=0.07065, over 1607901.48 frames. ], batch size: 27, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:32:38,094 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:00,834 INFO [train.py:901] (2/4) Epoch 15, batch 2900, loss[loss=0.1786, simple_loss=0.2587, pruned_loss=0.0493, over 8232.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2982, pruned_loss=0.0706, over 1610845.03 frames. ], batch size: 22, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:33:01,416 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.452e+02 2.959e+02 3.782e+02 6.842e+02, threshold=5.917e+02, percent-clipped=3.0 +2023-02-06 16:33:04,305 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6243, 2.7019, 1.9506, 2.2864, 2.2439, 1.4666, 2.1214, 2.1751], + device='cuda:2'), covar=tensor([0.1630, 0.0386, 0.1122, 0.0660, 0.0744, 0.1550, 0.1018, 0.1011], + device='cuda:2'), in_proj_covar=tensor([0.0345, 0.0228, 0.0321, 0.0299, 0.0297, 0.0326, 0.0340, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:33:29,119 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:29,807 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:35,327 INFO [train.py:901] (2/4) Epoch 15, batch 2950, loss[loss=0.2243, simple_loss=0.3077, pruned_loss=0.0705, over 8494.00 frames. ], tot_loss[loss=0.221, simple_loss=0.3, pruned_loss=0.071, over 1619342.89 frames. ], batch size: 26, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:33:36,701 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 16:33:42,119 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116123.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:45,602 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:46,299 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:49,703 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:58,563 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9343, 2.6313, 3.7399, 1.6126, 1.6207, 3.7123, 0.5371, 1.8980], + device='cuda:2'), covar=tensor([0.2254, 0.1264, 0.0233, 0.2971, 0.3847, 0.0293, 0.3282, 0.2076], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0178, 0.0111, 0.0216, 0.0260, 0.0116, 0.0164, 0.0176], + 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') +2023-02-06 16:34:08,976 INFO [train.py:901] (2/4) Epoch 15, batch 3000, loss[loss=0.2164, simple_loss=0.2855, pruned_loss=0.07359, over 7229.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2989, pruned_loss=0.07031, over 1614405.13 frames. ], batch size: 16, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:34:08,977 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 16:34:21,683 INFO [train.py:935] (2/4) Epoch 15, validation: loss=0.1808, simple_loss=0.2809, pruned_loss=0.04034, over 944034.00 frames. +2023-02-06 16:34:21,684 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 16:34:22,357 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.534e+02 3.127e+02 3.845e+02 7.463e+02, threshold=6.253e+02, percent-clipped=8.0 +2023-02-06 16:34:55,002 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 16:34:57,897 INFO [train.py:901] (2/4) Epoch 15, batch 3050, loss[loss=0.1865, simple_loss=0.2766, pruned_loss=0.04818, over 8346.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2977, pruned_loss=0.0696, over 1615874.79 frames. ], batch size: 26, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:35:07,481 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7584, 2.1170, 3.4650, 1.4680, 2.4786, 2.1815, 1.7693, 2.5118], + device='cuda:2'), covar=tensor([0.1587, 0.2145, 0.0572, 0.3907, 0.1597, 0.2672, 0.1792, 0.2050], + device='cuda:2'), in_proj_covar=tensor([0.0497, 0.0553, 0.0541, 0.0600, 0.0624, 0.0567, 0.0495, 0.0619], + 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') +2023-02-06 16:35:26,163 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116254.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 16:35:31,947 INFO [train.py:901] (2/4) Epoch 15, batch 3100, loss[loss=0.2199, simple_loss=0.2898, pruned_loss=0.07498, over 7804.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2984, pruned_loss=0.07001, over 1615695.73 frames. ], batch size: 19, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:35:32,570 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.573e+02 3.095e+02 3.865e+02 1.142e+03, threshold=6.190e+02, percent-clipped=3.0 +2023-02-06 16:35:40,327 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 16:35:55,812 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 16:36:06,923 INFO [train.py:901] (2/4) Epoch 15, batch 3150, loss[loss=0.2471, simple_loss=0.3189, pruned_loss=0.0876, over 8240.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2989, pruned_loss=0.07069, over 1611805.96 frames. ], batch size: 22, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:36:20,579 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8055, 1.6172, 1.9956, 1.7077, 1.7737, 1.8710, 1.6107, 0.8030], + device='cuda:2'), covar=tensor([0.4552, 0.3694, 0.1389, 0.2705, 0.2084, 0.2395, 0.1770, 0.3909], + device='cuda:2'), in_proj_covar=tensor([0.0904, 0.0913, 0.0747, 0.0884, 0.0945, 0.0840, 0.0719, 0.0791], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 16:36:27,181 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116341.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:36:41,970 INFO [train.py:901] (2/4) Epoch 15, batch 3200, loss[loss=0.26, simple_loss=0.3386, pruned_loss=0.09073, over 8311.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3002, pruned_loss=0.0716, over 1615095.54 frames. ], batch size: 25, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:36:42,889 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7477, 1.6192, 2.1287, 1.4559, 1.2206, 2.0798, 0.2022, 1.2463], + device='cuda:2'), covar=tensor([0.1857, 0.1640, 0.0361, 0.1497, 0.3399, 0.0444, 0.2729, 0.1553], + device='cuda:2'), in_proj_covar=tensor([0.0171, 0.0176, 0.0109, 0.0213, 0.0256, 0.0114, 0.0161, 0.0174], + 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') +2023-02-06 16:36:43,343 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.524e+02 3.304e+02 3.942e+02 1.206e+03, threshold=6.608e+02, percent-clipped=2.0 +2023-02-06 16:36:46,745 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116369.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 16:36:51,223 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:37:16,502 INFO [train.py:901] (2/4) Epoch 15, batch 3250, loss[loss=0.1855, simple_loss=0.2589, pruned_loss=0.05611, over 7791.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3, pruned_loss=0.07175, over 1612645.89 frames. ], batch size: 19, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:37:17,465 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4001, 1.6748, 1.6886, 0.9288, 1.7198, 1.2746, 0.2797, 1.6313], + device='cuda:2'), covar=tensor([0.0360, 0.0257, 0.0230, 0.0386, 0.0281, 0.0748, 0.0634, 0.0203], + device='cuda:2'), in_proj_covar=tensor([0.0422, 0.0361, 0.0311, 0.0416, 0.0347, 0.0504, 0.0373, 0.0384], + device='cuda:2'), out_proj_covar=tensor([1.1662e-04, 9.7238e-05, 8.3609e-05, 1.1273e-04, 9.4500e-05, 1.4688e-04, + 1.0307e-04, 1.0473e-04], device='cuda:2') +2023-02-06 16:37:52,495 INFO [train.py:901] (2/4) Epoch 15, batch 3300, loss[loss=0.2168, simple_loss=0.2979, pruned_loss=0.06785, over 8290.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2997, pruned_loss=0.07144, over 1612408.76 frames. ], batch size: 48, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:37:53,156 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.388e+02 2.875e+02 3.716e+02 9.209e+02, threshold=5.750e+02, percent-clipped=3.0 +2023-02-06 16:37:53,298 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:37:55,235 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116467.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:38:02,541 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:38:03,353 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7047, 1.9014, 1.5714, 2.3127, 0.9651, 1.4450, 1.6489, 1.8708], + device='cuda:2'), covar=tensor([0.0815, 0.0822, 0.1088, 0.0462, 0.1188, 0.1475, 0.0894, 0.0763], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0203, 0.0250, 0.0212, 0.0211, 0.0249, 0.0257, 0.0214], + 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') +2023-02-06 16:38:08,602 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1373, 2.1798, 4.3188, 2.5416, 3.8983, 3.6805, 3.9815, 3.8858], + device='cuda:2'), covar=tensor([0.0642, 0.3550, 0.0691, 0.3231, 0.0877, 0.0863, 0.0566, 0.0553], + device='cuda:2'), in_proj_covar=tensor([0.0554, 0.0608, 0.0633, 0.0577, 0.0654, 0.0560, 0.0551, 0.0609], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 16:38:12,020 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116491.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:38:26,446 INFO [train.py:901] (2/4) Epoch 15, batch 3350, loss[loss=0.2285, simple_loss=0.3028, pruned_loss=0.07709, over 7689.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3007, pruned_loss=0.07189, over 1613292.06 frames. ], batch size: 18, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:38:33,230 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116523.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:39:00,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7631, 2.1053, 1.5503, 2.5808, 1.1125, 1.4215, 1.7153, 1.9841], + device='cuda:2'), covar=tensor([0.0777, 0.0760, 0.1091, 0.0428, 0.1172, 0.1422, 0.0987, 0.0812], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0200, 0.0248, 0.0210, 0.0209, 0.0246, 0.0253, 0.0212], + 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') +2023-02-06 16:39:02,059 INFO [train.py:901] (2/4) Epoch 15, batch 3400, loss[loss=0.2823, simple_loss=0.3312, pruned_loss=0.1167, over 6855.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2999, pruned_loss=0.0716, over 1610486.10 frames. ], batch size: 71, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:39:02,726 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.566e+02 3.149e+02 4.104e+02 8.501e+02, threshold=6.298e+02, percent-clipped=7.0 +2023-02-06 16:39:14,871 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116582.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:39:22,204 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116593.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:39:36,209 INFO [train.py:901] (2/4) Epoch 15, batch 3450, loss[loss=0.2626, simple_loss=0.315, pruned_loss=0.1051, over 8082.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2996, pruned_loss=0.07163, over 1611742.53 frames. ], batch size: 21, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:39:44,405 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116625.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 16:39:51,716 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:40:01,116 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116650.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 16:40:10,185 INFO [train.py:901] (2/4) Epoch 15, batch 3500, loss[loss=0.1881, simple_loss=0.2721, pruned_loss=0.052, over 7963.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.3004, pruned_loss=0.07139, over 1618215.99 frames. ], batch size: 21, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:40:10,857 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.398e+02 2.936e+02 3.935e+02 9.560e+02, threshold=5.871e+02, percent-clipped=3.0 +2023-02-06 16:40:26,417 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116685.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:40:35,611 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 16:40:44,884 INFO [train.py:901] (2/4) Epoch 15, batch 3550, loss[loss=0.1869, simple_loss=0.257, pruned_loss=0.05843, over 7232.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.3, pruned_loss=0.07077, over 1620585.83 frames. ], batch size: 16, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:40:50,943 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:08,579 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116747.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:19,405 INFO [train.py:901] (2/4) Epoch 15, batch 3600, loss[loss=0.2208, simple_loss=0.3021, pruned_loss=0.06975, over 8586.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2999, pruned_loss=0.07083, over 1622823.98 frames. ], batch size: 31, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:41:20,115 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.627e+02 3.005e+02 3.918e+02 8.490e+02, threshold=6.010e+02, percent-clipped=4.0 +2023-02-06 16:41:25,802 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:47,427 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116800.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:52,897 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116808.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:56,175 INFO [train.py:901] (2/4) Epoch 15, batch 3650, loss[loss=0.252, simple_loss=0.3329, pruned_loss=0.08557, over 8139.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2991, pruned_loss=0.07048, over 1617756.01 frames. ], batch size: 22, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:42:00,908 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:13,597 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116838.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:21,037 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116849.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:30,275 INFO [train.py:901] (2/4) Epoch 15, batch 3700, loss[loss=0.246, simple_loss=0.2997, pruned_loss=0.09613, over 7525.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2988, pruned_loss=0.07081, over 1614282.19 frames. ], batch size: 18, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:42:30,504 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:30,956 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.301e+02 2.797e+02 3.414e+02 8.630e+02, threshold=5.595e+02, percent-clipped=3.0 +2023-02-06 16:42:33,143 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:36,563 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 16:42:38,088 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:47,930 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2023-02-06 16:43:06,644 INFO [train.py:901] (2/4) Epoch 15, batch 3750, loss[loss=0.2477, simple_loss=0.3288, pruned_loss=0.08331, over 8469.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2981, pruned_loss=0.07032, over 1609509.02 frames. ], batch size: 29, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:43:13,670 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116923.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:43:23,720 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8412, 2.4804, 4.4279, 1.6292, 3.1103, 2.2792, 2.0364, 2.5191], + device='cuda:2'), covar=tensor([0.1638, 0.2013, 0.0636, 0.3762, 0.1469, 0.2698, 0.1648, 0.2499], + device='cuda:2'), in_proj_covar=tensor([0.0499, 0.0553, 0.0542, 0.0602, 0.0624, 0.0565, 0.0497, 0.0622], + 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') +2023-02-06 16:43:25,968 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 16:43:40,810 INFO [train.py:901] (2/4) Epoch 15, batch 3800, loss[loss=0.2435, simple_loss=0.3188, pruned_loss=0.08407, over 7229.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2993, pruned_loss=0.07128, over 1606366.98 frames. ], batch size: 16, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:43:41,462 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.512e+02 2.989e+02 3.697e+02 7.171e+02, threshold=5.977e+02, percent-clipped=7.0 +2023-02-06 16:43:45,791 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 +2023-02-06 16:43:52,420 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116980.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:43:53,917 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:44:15,607 INFO [train.py:901] (2/4) Epoch 15, batch 3850, loss[loss=0.2375, simple_loss=0.3061, pruned_loss=0.08448, over 8671.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2991, pruned_loss=0.07108, over 1611306.49 frames. ], batch size: 34, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:44:23,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4689, 1.7824, 3.0316, 1.2486, 2.0932, 1.9098, 1.5166, 2.1193], + device='cuda:2'), covar=tensor([0.1891, 0.2526, 0.0783, 0.4270, 0.1802, 0.3007, 0.2146, 0.2225], + device='cuda:2'), in_proj_covar=tensor([0.0500, 0.0554, 0.0540, 0.0602, 0.0625, 0.0565, 0.0496, 0.0622], + 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') +2023-02-06 16:44:42,555 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 16:44:46,205 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117056.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:44:50,964 INFO [train.py:901] (2/4) Epoch 15, batch 3900, loss[loss=0.244, simple_loss=0.3108, pruned_loss=0.08864, over 6958.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2995, pruned_loss=0.07088, over 1613538.16 frames. ], batch size: 71, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:44:51,618 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.460e+02 2.428e+02 3.027e+02 3.797e+02 6.654e+02, threshold=6.053e+02, percent-clipped=2.0 +2023-02-06 16:44:53,042 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:45:03,695 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117081.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:45:12,951 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:45:24,921 INFO [train.py:901] (2/4) Epoch 15, batch 3950, loss[loss=0.2182, simple_loss=0.3053, pruned_loss=0.06551, over 8507.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3004, pruned_loss=0.0715, over 1612795.54 frames. ], batch size: 28, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:46:01,086 INFO [train.py:901] (2/4) Epoch 15, batch 4000, loss[loss=0.203, simple_loss=0.2899, pruned_loss=0.05807, over 8475.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2998, pruned_loss=0.0713, over 1610355.29 frames. ], batch size: 27, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:46:01,785 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.473e+02 2.992e+02 3.534e+02 5.115e+02, threshold=5.984e+02, percent-clipped=0.0 +2023-02-06 16:46:01,905 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117164.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:12,553 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117179.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:13,874 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117181.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:29,850 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117204.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:35,577 INFO [train.py:901] (2/4) Epoch 15, batch 4050, loss[loss=0.2316, simple_loss=0.3061, pruned_loss=0.0785, over 8327.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2994, pruned_loss=0.07185, over 1609218.05 frames. ], batch size: 25, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:46:53,306 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117238.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:53,910 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:47:11,627 INFO [train.py:901] (2/4) Epoch 15, batch 4100, loss[loss=0.2512, simple_loss=0.3255, pruned_loss=0.08842, over 8321.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2997, pruned_loss=0.07185, over 1610113.23 frames. ], batch size: 25, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:47:11,816 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:47:12,278 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.506e+02 3.096e+02 3.742e+02 9.544e+02, threshold=6.191e+02, percent-clipped=4.0 +2023-02-06 16:47:20,186 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4230, 1.6493, 1.6582, 1.1822, 1.7437, 1.3097, 0.2351, 1.6213], + device='cuda:2'), covar=tensor([0.0330, 0.0275, 0.0220, 0.0297, 0.0282, 0.0681, 0.0644, 0.0181], + device='cuda:2'), in_proj_covar=tensor([0.0424, 0.0366, 0.0314, 0.0422, 0.0349, 0.0510, 0.0377, 0.0387], + device='cuda:2'), out_proj_covar=tensor([1.1707e-04, 9.8413e-05, 8.4152e-05, 1.1460e-04, 9.4907e-05, 1.4848e-04, + 1.0405e-04, 1.0521e-04], device='cuda:2') +2023-02-06 16:47:22,938 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117279.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:47:46,643 INFO [train.py:901] (2/4) Epoch 15, batch 4150, loss[loss=0.1883, simple_loss=0.2766, pruned_loss=0.04998, over 8287.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2994, pruned_loss=0.07135, over 1613154.67 frames. ], batch size: 23, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:47:51,112 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 +2023-02-06 16:48:10,097 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:48:12,980 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117351.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:48:21,502 INFO [train.py:901] (2/4) Epoch 15, batch 4200, loss[loss=0.2289, simple_loss=0.3118, pruned_loss=0.07298, over 8185.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2995, pruned_loss=0.07108, over 1612060.58 frames. ], batch size: 23, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:48:22,820 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.404e+02 2.907e+02 3.383e+02 1.073e+03, threshold=5.814e+02, percent-clipped=1.0 +2023-02-06 16:48:31,968 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:48:40,572 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 16:48:41,600 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 +2023-02-06 16:48:45,700 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.38 vs. limit=5.0 +2023-02-06 16:48:57,040 INFO [train.py:901] (2/4) Epoch 15, batch 4250, loss[loss=0.233, simple_loss=0.3039, pruned_loss=0.08111, over 8456.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2992, pruned_loss=0.07104, over 1610264.14 frames. ], batch size: 27, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:49:03,723 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 16:49:14,101 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117437.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:49:30,903 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:49:31,426 INFO [train.py:901] (2/4) Epoch 15, batch 4300, loss[loss=0.1806, simple_loss=0.2589, pruned_loss=0.05117, over 7819.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2994, pruned_loss=0.07119, over 1611711.17 frames. ], batch size: 20, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:49:32,090 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.479e+02 3.115e+02 3.892e+02 7.815e+02, threshold=6.229e+02, percent-clipped=5.0 +2023-02-06 16:49:37,253 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 16:50:07,598 INFO [train.py:901] (2/4) Epoch 15, batch 4350, loss[loss=0.2546, simple_loss=0.3374, pruned_loss=0.08596, over 8433.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2978, pruned_loss=0.07047, over 1608217.98 frames. ], batch size: 27, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:50:23,039 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117535.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:50:25,369 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 +2023-02-06 16:50:36,325 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 16:50:40,555 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117560.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:50:42,367 INFO [train.py:901] (2/4) Epoch 15, batch 4400, loss[loss=0.1705, simple_loss=0.2441, pruned_loss=0.04839, over 7433.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2973, pruned_loss=0.07022, over 1604965.42 frames. ], batch size: 17, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:50:43,039 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.383e+02 3.124e+02 3.901e+02 9.506e+02, threshold=6.248e+02, percent-clipped=7.0 +2023-02-06 16:50:55,790 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:50:55,919 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1873, 1.7042, 1.4700, 1.5628, 1.4896, 1.3333, 1.3585, 1.3757], + device='cuda:2'), covar=tensor([0.0997, 0.0450, 0.1096, 0.0543, 0.0674, 0.1211, 0.0818, 0.0692], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0232, 0.0327, 0.0303, 0.0302, 0.0332, 0.0347, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:51:17,967 INFO [train.py:901] (2/4) Epoch 15, batch 4450, loss[loss=0.2145, simple_loss=0.2866, pruned_loss=0.07125, over 7705.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2984, pruned_loss=0.07063, over 1610574.17 frames. ], batch size: 18, lr: 5.07e-03, grad_scale: 16.0 +2023-02-06 16:51:17,985 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 16:51:38,328 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.49 vs. limit=5.0 +2023-02-06 16:51:52,101 INFO [train.py:901] (2/4) Epoch 15, batch 4500, loss[loss=0.2393, simple_loss=0.3069, pruned_loss=0.08578, over 8079.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2981, pruned_loss=0.07052, over 1611786.58 frames. ], batch size: 21, lr: 5.07e-03, grad_scale: 16.0 +2023-02-06 16:51:52,740 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.480e+02 2.963e+02 4.043e+02 1.091e+03, threshold=5.927e+02, percent-clipped=5.0 +2023-02-06 16:52:11,215 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:52:11,859 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 16:52:16,205 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:52:27,010 INFO [train.py:901] (2/4) Epoch 15, batch 4550, loss[loss=0.1927, simple_loss=0.2694, pruned_loss=0.05804, over 7801.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2975, pruned_loss=0.07032, over 1612768.49 frames. ], batch size: 20, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:53:02,114 INFO [train.py:901] (2/4) Epoch 15, batch 4600, loss[loss=0.2369, simple_loss=0.3118, pruned_loss=0.08101, over 8460.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2975, pruned_loss=0.07009, over 1610241.85 frames. ], batch size: 27, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:53:03,483 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.311e+02 2.848e+02 3.671e+02 5.923e+02, threshold=5.697e+02, percent-clipped=0.0 +2023-02-06 16:53:27,493 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7684, 4.6815, 4.3113, 3.0128, 4.2149, 4.3004, 4.4117, 4.0565], + device='cuda:2'), covar=tensor([0.0633, 0.0516, 0.1021, 0.3371, 0.0763, 0.0941, 0.1265, 0.0726], + device='cuda:2'), in_proj_covar=tensor([0.0486, 0.0403, 0.0407, 0.0505, 0.0398, 0.0407, 0.0389, 0.0350], + 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') +2023-02-06 16:53:31,670 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:53:35,833 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.63 vs. limit=5.0 +2023-02-06 16:53:36,051 INFO [train.py:901] (2/4) Epoch 15, batch 4650, loss[loss=0.2209, simple_loss=0.3014, pruned_loss=0.07015, over 8281.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2979, pruned_loss=0.07087, over 1611500.50 frames. ], batch size: 23, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:53:44,243 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7048, 1.3314, 3.4019, 1.5457, 2.3526, 3.7921, 3.8953, 3.2091], + device='cuda:2'), covar=tensor([0.1304, 0.1888, 0.0359, 0.2179, 0.1079, 0.0240, 0.0510, 0.0632], + device='cuda:2'), in_proj_covar=tensor([0.0277, 0.0306, 0.0270, 0.0300, 0.0288, 0.0246, 0.0376, 0.0298], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:53:57,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2284, 1.2387, 1.5604, 1.1950, 0.7326, 1.2958, 1.2235, 1.1282], + device='cuda:2'), covar=tensor([0.0551, 0.1310, 0.1625, 0.1421, 0.0578, 0.1487, 0.0660, 0.0646], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0156, 0.0101, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 16:54:11,638 INFO [train.py:901] (2/4) Epoch 15, batch 4700, loss[loss=0.1971, simple_loss=0.2733, pruned_loss=0.06049, over 7814.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2986, pruned_loss=0.0712, over 1612819.09 frames. ], batch size: 20, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:54:12,892 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.509e+02 3.109e+02 4.231e+02 8.316e+02, threshold=6.217e+02, percent-clipped=12.0 +2023-02-06 16:54:46,541 INFO [train.py:901] (2/4) Epoch 15, batch 4750, loss[loss=0.1954, simple_loss=0.2701, pruned_loss=0.06029, over 7655.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2995, pruned_loss=0.07217, over 1613568.48 frames. ], batch size: 19, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:55:11,979 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 16:55:15,276 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 16:55:16,043 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117954.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:55:22,504 INFO [train.py:901] (2/4) Epoch 15, batch 4800, loss[loss=0.2428, simple_loss=0.3298, pruned_loss=0.07787, over 8617.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2985, pruned_loss=0.07166, over 1610109.35 frames. ], batch size: 31, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:55:23,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.482e+02 3.121e+02 4.555e+02 1.692e+03, threshold=6.242e+02, percent-clipped=8.0 +2023-02-06 16:55:33,838 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:55:38,254 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7038, 4.6804, 4.2105, 2.0450, 4.2153, 4.2122, 4.3159, 4.0541], + device='cuda:2'), covar=tensor([0.0767, 0.0551, 0.1080, 0.4399, 0.0654, 0.0947, 0.1185, 0.0732], + device='cuda:2'), in_proj_covar=tensor([0.0482, 0.0400, 0.0405, 0.0499, 0.0396, 0.0403, 0.0384, 0.0348], + 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') +2023-02-06 16:55:38,472 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 16:55:57,733 INFO [train.py:901] (2/4) Epoch 15, batch 4850, loss[loss=0.2092, simple_loss=0.2763, pruned_loss=0.07106, over 8079.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2986, pruned_loss=0.07196, over 1607249.34 frames. ], batch size: 21, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:56:07,039 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 16:56:29,211 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:56:31,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:56:32,449 INFO [train.py:901] (2/4) Epoch 15, batch 4900, loss[loss=0.2021, simple_loss=0.2718, pruned_loss=0.06622, over 7655.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2972, pruned_loss=0.07078, over 1609531.92 frames. ], batch size: 19, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:56:33,728 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.453e+02 2.951e+02 3.688e+02 9.605e+02, threshold=5.903e+02, percent-clipped=5.0 +2023-02-06 16:56:50,257 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:57:07,602 INFO [train.py:901] (2/4) Epoch 15, batch 4950, loss[loss=0.2099, simple_loss=0.289, pruned_loss=0.06547, over 8452.00 frames. ], tot_loss[loss=0.219, simple_loss=0.297, pruned_loss=0.07051, over 1609984.61 frames. ], batch size: 27, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:57:29,397 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8894, 1.6793, 3.3849, 1.5373, 2.2517, 3.7341, 3.7505, 3.2017], + device='cuda:2'), covar=tensor([0.1154, 0.1545, 0.0356, 0.1986, 0.1094, 0.0229, 0.0518, 0.0552], + device='cuda:2'), in_proj_covar=tensor([0.0274, 0.0303, 0.0268, 0.0299, 0.0286, 0.0245, 0.0373, 0.0297], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 16:57:42,120 INFO [train.py:901] (2/4) Epoch 15, batch 5000, loss[loss=0.2047, simple_loss=0.2726, pruned_loss=0.06845, over 7653.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2983, pruned_loss=0.07116, over 1615691.53 frames. ], batch size: 19, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:57:43,379 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.421e+02 2.910e+02 3.813e+02 6.624e+02, threshold=5.820e+02, percent-clipped=4.0 +2023-02-06 16:57:58,599 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:58:17,604 INFO [train.py:901] (2/4) Epoch 15, batch 5050, loss[loss=0.2189, simple_loss=0.3013, pruned_loss=0.06826, over 8667.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2977, pruned_loss=0.07078, over 1613922.35 frames. ], batch size: 39, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:58:43,428 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 16:58:46,024 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2023-02-06 16:58:46,883 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.86 vs. limit=5.0 +2023-02-06 16:58:52,546 INFO [train.py:901] (2/4) Epoch 15, batch 5100, loss[loss=0.2273, simple_loss=0.2867, pruned_loss=0.08397, over 7793.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2981, pruned_loss=0.07113, over 1614292.88 frames. ], batch size: 19, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:58:53,819 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.591e+02 3.125e+02 3.877e+02 7.785e+02, threshold=6.249e+02, percent-clipped=4.0 +2023-02-06 16:59:09,115 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:59:23,816 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118307.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:59:27,781 INFO [train.py:901] (2/4) Epoch 15, batch 5150, loss[loss=0.2291, simple_loss=0.2979, pruned_loss=0.08013, over 8480.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2975, pruned_loss=0.07078, over 1614974.87 frames. ], batch size: 29, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:59:51,101 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:00:02,425 INFO [train.py:901] (2/4) Epoch 15, batch 5200, loss[loss=0.1969, simple_loss=0.2842, pruned_loss=0.05487, over 8472.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2993, pruned_loss=0.07185, over 1617200.87 frames. ], batch size: 27, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:00:03,700 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.390e+02 2.269e+02 2.811e+02 3.673e+02 9.088e+02, threshold=5.623e+02, percent-clipped=2.0 +2023-02-06 17:00:10,163 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4385, 1.5556, 2.1734, 1.3297, 1.5372, 1.7023, 1.4962, 1.4759], + device='cuda:2'), covar=tensor([0.1795, 0.2339, 0.0811, 0.3853, 0.1746, 0.3066, 0.2041, 0.1984], + device='cuda:2'), in_proj_covar=tensor([0.0497, 0.0552, 0.0535, 0.0604, 0.0625, 0.0564, 0.0494, 0.0616], + 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') +2023-02-06 17:00:29,678 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:00:37,913 INFO [train.py:901] (2/4) Epoch 15, batch 5250, loss[loss=0.2118, simple_loss=0.2913, pruned_loss=0.06612, over 8469.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2999, pruned_loss=0.07184, over 1612837.62 frames. ], batch size: 29, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:00:46,145 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 17:01:12,973 INFO [train.py:901] (2/4) Epoch 15, batch 5300, loss[loss=0.1981, simple_loss=0.2897, pruned_loss=0.05324, over 8295.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2998, pruned_loss=0.07134, over 1606779.21 frames. ], batch size: 23, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:01:14,342 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.534e+02 2.995e+02 3.765e+02 8.916e+02, threshold=5.991e+02, percent-clipped=4.0 +2023-02-06 17:01:47,930 INFO [train.py:901] (2/4) Epoch 15, batch 5350, loss[loss=0.2099, simple_loss=0.2898, pruned_loss=0.06499, over 8607.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3001, pruned_loss=0.07183, over 1606158.57 frames. ], batch size: 39, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:01:50,885 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:02:01,053 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118530.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:02:11,013 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2023-02-06 17:02:24,467 INFO [train.py:901] (2/4) Epoch 15, batch 5400, loss[loss=0.2067, simple_loss=0.2836, pruned_loss=0.06491, over 7454.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3004, pruned_loss=0.07174, over 1608526.77 frames. ], batch size: 17, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:02:25,794 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.478e+02 2.903e+02 3.717e+02 8.291e+02, threshold=5.806e+02, percent-clipped=5.0 +2023-02-06 17:02:27,340 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1893, 1.3725, 1.2647, 1.8401, 0.7582, 1.0805, 1.3528, 1.4576], + device='cuda:2'), covar=tensor([0.1096, 0.0891, 0.1270, 0.0547, 0.1178, 0.1699, 0.0785, 0.0770], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0204, 0.0249, 0.0212, 0.0211, 0.0246, 0.0253, 0.0213], + 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') +2023-02-06 17:02:58,969 INFO [train.py:901] (2/4) Epoch 15, batch 5450, loss[loss=0.2045, simple_loss=0.2906, pruned_loss=0.05922, over 8469.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3002, pruned_loss=0.07181, over 1607960.10 frames. ], batch size: 25, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:03:11,221 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:21,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:24,941 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:26,157 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:31,677 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3057, 1.3992, 1.3313, 1.7680, 0.7443, 1.1465, 1.2014, 1.3948], + device='cuda:2'), covar=tensor([0.0940, 0.0895, 0.1102, 0.0570, 0.1220, 0.1590, 0.0956, 0.0864], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0204, 0.0250, 0.0213, 0.0213, 0.0248, 0.0255, 0.0214], + 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') +2023-02-06 17:03:34,862 INFO [train.py:901] (2/4) Epoch 15, batch 5500, loss[loss=0.2852, simple_loss=0.3416, pruned_loss=0.1144, over 8604.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2992, pruned_loss=0.07117, over 1611341.71 frames. ], batch size: 31, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:03:36,244 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.592e+02 3.113e+02 3.610e+02 8.755e+02, threshold=6.227e+02, percent-clipped=2.0 +2023-02-06 17:03:38,406 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 17:03:54,400 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:04:09,096 INFO [train.py:901] (2/4) Epoch 15, batch 5550, loss[loss=0.217, simple_loss=0.3068, pruned_loss=0.0636, over 8468.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2992, pruned_loss=0.07133, over 1612328.01 frames. ], batch size: 25, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:04:32,307 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118746.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:04:37,106 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3176, 1.3081, 4.5071, 1.6719, 3.9366, 3.7373, 4.0669, 3.9300], + device='cuda:2'), covar=tensor([0.0674, 0.4981, 0.0525, 0.3841, 0.1193, 0.1028, 0.0567, 0.0751], + device='cuda:2'), in_proj_covar=tensor([0.0551, 0.0599, 0.0632, 0.0572, 0.0649, 0.0552, 0.0547, 0.0610], + device='cuda:2'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:04:44,988 INFO [train.py:901] (2/4) Epoch 15, batch 5600, loss[loss=0.2283, simple_loss=0.3071, pruned_loss=0.07477, over 8289.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2999, pruned_loss=0.07145, over 1612811.24 frames. ], batch size: 23, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:04:46,300 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.537e+02 3.218e+02 3.925e+02 9.216e+02, threshold=6.435e+02, percent-clipped=4.0 +2023-02-06 17:04:47,204 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118766.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:04:52,579 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:05:09,076 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:05:14,505 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:05:19,155 INFO [train.py:901] (2/4) Epoch 15, batch 5650, loss[loss=0.1938, simple_loss=0.2869, pruned_loss=0.05033, over 8292.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2991, pruned_loss=0.07095, over 1611454.96 frames. ], batch size: 23, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:05:43,436 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 17:05:53,489 INFO [train.py:901] (2/4) Epoch 15, batch 5700, loss[loss=0.1851, simple_loss=0.2723, pruned_loss=0.04896, over 7530.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2981, pruned_loss=0.07042, over 1610590.10 frames. ], batch size: 18, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:05:54,817 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.491e+02 2.972e+02 3.726e+02 7.690e+02, threshold=5.944e+02, percent-clipped=5.0 +2023-02-06 17:05:56,526 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 17:06:21,155 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118901.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:06:28,617 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5069, 2.8938, 2.3523, 4.0248, 1.5136, 2.0855, 2.1837, 3.1358], + device='cuda:2'), covar=tensor([0.0667, 0.0805, 0.0886, 0.0248, 0.1246, 0.1283, 0.1144, 0.0738], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0202, 0.0249, 0.0212, 0.0212, 0.0246, 0.0254, 0.0212], + 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') +2023-02-06 17:06:29,109 INFO [train.py:901] (2/4) Epoch 15, batch 5750, loss[loss=0.1846, simple_loss=0.2673, pruned_loss=0.05093, over 8254.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2984, pruned_loss=0.07097, over 1606867.48 frames. ], batch size: 22, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:06:31,384 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0692, 2.2549, 1.8169, 2.8166, 1.1684, 1.5565, 1.8653, 2.2951], + device='cuda:2'), covar=tensor([0.0696, 0.0802, 0.0973, 0.0366, 0.1217, 0.1350, 0.1010, 0.0690], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0203, 0.0249, 0.0212, 0.0212, 0.0246, 0.0254, 0.0212], + 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') +2023-02-06 17:06:38,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:06:46,382 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 17:06:54,645 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8226, 5.9244, 5.1983, 2.2512, 5.2908, 5.6956, 5.4865, 5.3252], + device='cuda:2'), covar=tensor([0.0696, 0.0497, 0.1175, 0.4682, 0.0737, 0.0824, 0.1110, 0.0636], + device='cuda:2'), in_proj_covar=tensor([0.0486, 0.0400, 0.0407, 0.0502, 0.0395, 0.0408, 0.0387, 0.0353], + 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') +2023-02-06 17:07:03,693 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2910, 1.2830, 1.6356, 1.2249, 0.7108, 1.3332, 1.2530, 1.2425], + device='cuda:2'), covar=tensor([0.0534, 0.1278, 0.1560, 0.1399, 0.0563, 0.1493, 0.0676, 0.0644], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0156, 0.0102, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 17:07:04,189 INFO [train.py:901] (2/4) Epoch 15, batch 5800, loss[loss=0.205, simple_loss=0.2771, pruned_loss=0.06643, over 7445.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2993, pruned_loss=0.07132, over 1609465.31 frames. ], batch size: 17, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:07:05,529 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.317e+02 2.944e+02 4.100e+02 6.996e+02, threshold=5.887e+02, percent-clipped=4.0 +2023-02-06 17:07:26,178 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:07:32,110 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119002.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:07:39,875 INFO [train.py:901] (2/4) Epoch 15, batch 5850, loss[loss=0.1919, simple_loss=0.2828, pruned_loss=0.05048, over 8029.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2996, pruned_loss=0.07102, over 1612946.28 frames. ], batch size: 22, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:07:46,175 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:07:49,456 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:02,803 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:13,676 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:14,145 INFO [train.py:901] (2/4) Epoch 15, batch 5900, loss[loss=0.1853, simple_loss=0.2608, pruned_loss=0.05487, over 7807.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2986, pruned_loss=0.07049, over 1609490.31 frames. ], batch size: 20, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:08:15,366 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.486e+02 2.938e+02 3.942e+02 7.909e+02, threshold=5.877e+02, percent-clipped=6.0 +2023-02-06 17:08:27,002 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-06 17:08:30,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:34,141 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:44,790 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119108.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:46,719 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119111.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:47,873 INFO [train.py:901] (2/4) Epoch 15, batch 5950, loss[loss=0.2391, simple_loss=0.3207, pruned_loss=0.07876, over 8358.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2992, pruned_loss=0.07069, over 1615175.08 frames. ], batch size: 24, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:09:22,903 INFO [train.py:901] (2/4) Epoch 15, batch 6000, loss[loss=0.2375, simple_loss=0.3108, pruned_loss=0.08213, over 8366.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2987, pruned_loss=0.07027, over 1611782.42 frames. ], batch size: 24, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:09:22,903 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 17:09:30,930 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6924, 1.6260, 2.7898, 1.3990, 2.0171, 2.9215, 3.0533, 2.5217], + device='cuda:2'), covar=tensor([0.1079, 0.1442, 0.0334, 0.2066, 0.0919, 0.0331, 0.0615, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0276, 0.0306, 0.0268, 0.0298, 0.0286, 0.0247, 0.0374, 0.0298], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:09:35,678 INFO [train.py:935] (2/4) Epoch 15, validation: loss=0.181, simple_loss=0.2808, pruned_loss=0.04056, over 944034.00 frames. +2023-02-06 17:09:35,679 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 17:09:37,100 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.469e+02 2.578e+02 3.120e+02 3.956e+02 1.218e+03, threshold=6.240e+02, percent-clipped=5.0 +2023-02-06 17:09:38,918 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 17:10:10,487 INFO [train.py:901] (2/4) Epoch 15, batch 6050, loss[loss=0.2253, simple_loss=0.2927, pruned_loss=0.07893, over 7525.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2977, pruned_loss=0.0699, over 1613129.37 frames. ], batch size: 18, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:10:32,944 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.27 vs. limit=5.0 +2023-02-06 17:10:44,323 INFO [train.py:901] (2/4) Epoch 15, batch 6100, loss[loss=0.1757, simple_loss=0.2534, pruned_loss=0.04898, over 8078.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2967, pruned_loss=0.06944, over 1609957.50 frames. ], batch size: 21, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:10:45,656 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.463e+02 3.114e+02 4.132e+02 8.492e+02, threshold=6.229e+02, percent-clipped=7.0 +2023-02-06 17:11:00,074 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4845, 2.7722, 1.7648, 2.1883, 2.3028, 1.5426, 2.0115, 2.1283], + device='cuda:2'), covar=tensor([0.1561, 0.0388, 0.1322, 0.0729, 0.0706, 0.1602, 0.1108, 0.0931], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0234, 0.0329, 0.0304, 0.0305, 0.0335, 0.0351, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:11:18,256 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 17:11:20,322 INFO [train.py:901] (2/4) Epoch 15, batch 6150, loss[loss=0.2425, simple_loss=0.3148, pruned_loss=0.08511, over 7647.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2949, pruned_loss=0.06901, over 1603017.95 frames. ], batch size: 19, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:11:54,707 INFO [train.py:901] (2/4) Epoch 15, batch 6200, loss[loss=0.2508, simple_loss=0.3254, pruned_loss=0.08805, over 8597.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2958, pruned_loss=0.0692, over 1605004.50 frames. ], batch size: 31, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:11:55,641 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:11:56,081 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.347e+02 3.204e+02 3.871e+02 7.576e+02, threshold=6.408e+02, percent-clipped=2.0 +2023-02-06 17:12:14,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:12:30,378 INFO [train.py:901] (2/4) Epoch 15, batch 6250, loss[loss=0.2862, simple_loss=0.3402, pruned_loss=0.1161, over 8678.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.297, pruned_loss=0.06994, over 1608211.66 frames. ], batch size: 49, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:12:47,167 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119437.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:12:59,471 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119455.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:13:04,842 INFO [train.py:901] (2/4) Epoch 15, batch 6300, loss[loss=0.1806, simple_loss=0.2512, pruned_loss=0.05502, over 7545.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2968, pruned_loss=0.07009, over 1606895.92 frames. ], batch size: 18, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:13:06,138 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.517e+02 3.087e+02 3.932e+02 1.134e+03, threshold=6.173e+02, percent-clipped=3.0 +2023-02-06 17:13:41,043 INFO [train.py:901] (2/4) Epoch 15, batch 6350, loss[loss=0.2082, simple_loss=0.2912, pruned_loss=0.06259, over 7927.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2973, pruned_loss=0.07018, over 1610182.02 frames. ], batch size: 20, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:13:53,347 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-06 17:13:53,782 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:14:03,580 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 17:14:07,837 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119552.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:14:15,082 INFO [train.py:901] (2/4) Epoch 15, batch 6400, loss[loss=0.1642, simple_loss=0.2537, pruned_loss=0.03733, over 7974.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2968, pruned_loss=0.06989, over 1612495.64 frames. ], batch size: 21, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:14:16,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.423e+02 3.023e+02 3.752e+02 7.818e+02, threshold=6.047e+02, percent-clipped=4.0 +2023-02-06 17:14:20,027 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119570.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:14:49,975 INFO [train.py:901] (2/4) Epoch 15, batch 6450, loss[loss=0.2119, simple_loss=0.3007, pruned_loss=0.06152, over 8368.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2964, pruned_loss=0.06992, over 1608516.12 frames. ], batch size: 24, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:15:24,235 INFO [train.py:901] (2/4) Epoch 15, batch 6500, loss[loss=0.2177, simple_loss=0.3054, pruned_loss=0.06504, over 8711.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2969, pruned_loss=0.06982, over 1614493.26 frames. ], batch size: 34, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:15:25,561 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.561e+02 2.888e+02 3.578e+02 6.995e+02, threshold=5.776e+02, percent-clipped=4.0 +2023-02-06 17:15:38,170 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 17:15:38,582 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:15:53,105 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-06 17:15:58,707 INFO [train.py:901] (2/4) Epoch 15, batch 6550, loss[loss=0.2486, simple_loss=0.319, pruned_loss=0.08906, over 8463.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2964, pruned_loss=0.06935, over 1614172.12 frames. ], batch size: 27, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:16:23,985 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 17:16:27,862 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9264, 3.9328, 2.2456, 2.6493, 2.7857, 1.9691, 2.6152, 2.9356], + device='cuda:2'), covar=tensor([0.1608, 0.0275, 0.1171, 0.0823, 0.0745, 0.1393, 0.1053, 0.0965], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0232, 0.0328, 0.0304, 0.0302, 0.0334, 0.0346, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:16:29,728 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 17:16:34,455 INFO [train.py:901] (2/4) Epoch 15, batch 6600, loss[loss=0.1865, simple_loss=0.2642, pruned_loss=0.05444, over 7529.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2966, pruned_loss=0.06919, over 1616560.06 frames. ], batch size: 18, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:16:35,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.456e+02 2.938e+02 3.854e+02 9.901e+02, threshold=5.877e+02, percent-clipped=5.0 +2023-02-06 17:16:47,907 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 17:16:51,495 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 +2023-02-06 17:17:05,454 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119808.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:08,628 INFO [train.py:901] (2/4) Epoch 15, batch 6650, loss[loss=0.2047, simple_loss=0.287, pruned_loss=0.0612, over 8359.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2969, pruned_loss=0.06944, over 1617093.55 frames. ], batch size: 26, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:17:17,684 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119826.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:22,306 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:31,209 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 17:17:36,362 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:44,400 INFO [train.py:901] (2/4) Epoch 15, batch 6700, loss[loss=0.2237, simple_loss=0.3121, pruned_loss=0.06764, over 8322.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2971, pruned_loss=0.06925, over 1618265.06 frames. ], batch size: 25, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:17:45,753 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.601e+02 2.951e+02 3.516e+02 8.618e+02, threshold=5.902e+02, percent-clipped=2.0 +2023-02-06 17:17:53,452 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:18:19,554 INFO [train.py:901] (2/4) Epoch 15, batch 6750, loss[loss=0.2811, simple_loss=0.346, pruned_loss=0.1081, over 8631.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2987, pruned_loss=0.07053, over 1618592.51 frames. ], batch size: 34, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:18:33,558 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6275, 2.6369, 1.7106, 2.3488, 2.3200, 1.3857, 2.1000, 2.1400], + device='cuda:2'), covar=tensor([0.1420, 0.0375, 0.1209, 0.0588, 0.0663, 0.1560, 0.1008, 0.0945], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0232, 0.0326, 0.0304, 0.0302, 0.0332, 0.0345, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:18:55,244 INFO [train.py:901] (2/4) Epoch 15, batch 6800, loss[loss=0.2767, simple_loss=0.3332, pruned_loss=0.1101, over 7964.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2977, pruned_loss=0.06984, over 1616775.29 frames. ], batch size: 21, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:18:57,360 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.559e+02 3.032e+02 3.835e+02 7.300e+02, threshold=6.064e+02, percent-clipped=2.0 +2023-02-06 17:19:03,572 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 17:19:15,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:19:32,101 INFO [train.py:901] (2/4) Epoch 15, batch 6850, loss[loss=0.2512, simple_loss=0.3296, pruned_loss=0.08637, over 8249.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2992, pruned_loss=0.07043, over 1619359.54 frames. ], batch size: 22, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:19:41,653 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=120027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:19:49,109 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 17:19:53,397 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 17:20:06,210 INFO [train.py:901] (2/4) Epoch 15, batch 6900, loss[loss=0.2187, simple_loss=0.3033, pruned_loss=0.0671, over 8723.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2981, pruned_loss=0.07, over 1614660.24 frames. ], batch size: 49, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:20:07,530 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.397e+02 2.973e+02 3.506e+02 9.980e+02, threshold=5.947e+02, percent-clipped=2.0 +2023-02-06 17:20:40,404 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1899, 2.1737, 1.5464, 1.8508, 1.7972, 1.3260, 1.5672, 1.5979], + device='cuda:2'), covar=tensor([0.1364, 0.0377, 0.1194, 0.0606, 0.0720, 0.1407, 0.1021, 0.0811], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0232, 0.0326, 0.0305, 0.0303, 0.0331, 0.0345, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:20:42,259 INFO [train.py:901] (2/4) Epoch 15, batch 6950, loss[loss=0.2426, simple_loss=0.3156, pruned_loss=0.08484, over 8621.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2972, pruned_loss=0.06949, over 1610134.01 frames. ], batch size: 34, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:20:59,737 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1517, 1.6324, 3.3522, 1.6676, 2.3667, 3.7538, 3.7945, 3.2029], + device='cuda:2'), covar=tensor([0.1100, 0.1718, 0.0373, 0.2005, 0.1153, 0.0232, 0.0526, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0310, 0.0275, 0.0304, 0.0290, 0.0252, 0.0383, 0.0304], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:21:02,420 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:21:03,577 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 17:21:16,268 INFO [train.py:901] (2/4) Epoch 15, batch 7000, loss[loss=0.2529, simple_loss=0.3241, pruned_loss=0.09081, over 8140.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2966, pruned_loss=0.069, over 1610734.47 frames. ], batch size: 22, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:21:17,612 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.303e+02 2.879e+02 3.620e+02 6.461e+02, threshold=5.757e+02, percent-clipped=3.0 +2023-02-06 17:21:51,887 INFO [train.py:901] (2/4) Epoch 15, batch 7050, loss[loss=0.2215, simple_loss=0.3118, pruned_loss=0.0656, over 8460.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2963, pruned_loss=0.06923, over 1605996.84 frames. ], batch size: 25, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:22:15,023 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:22:26,158 INFO [train.py:901] (2/4) Epoch 15, batch 7100, loss[loss=0.2056, simple_loss=0.2892, pruned_loss=0.06097, over 8606.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2976, pruned_loss=0.07009, over 1607280.97 frames. ], batch size: 49, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:22:27,479 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.428e+02 3.078e+02 4.147e+02 9.225e+02, threshold=6.156e+02, percent-clipped=10.0 +2023-02-06 17:22:32,371 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120272.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:22:37,703 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.1285, 1.8179, 6.1754, 2.3440, 5.5677, 5.2871, 5.7877, 5.6661], + device='cuda:2'), covar=tensor([0.0399, 0.4111, 0.0254, 0.3028, 0.0795, 0.0710, 0.0409, 0.0409], + device='cuda:2'), in_proj_covar=tensor([0.0559, 0.0611, 0.0636, 0.0582, 0.0657, 0.0563, 0.0555, 0.0616], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:23:00,844 INFO [train.py:901] (2/4) Epoch 15, batch 7150, loss[loss=0.2838, simple_loss=0.3359, pruned_loss=0.1158, over 6893.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2969, pruned_loss=0.07008, over 1604164.07 frames. ], batch size: 71, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:23:22,065 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1932, 1.2146, 3.3193, 1.0390, 2.9197, 2.7542, 3.0301, 2.9058], + device='cuda:2'), covar=tensor([0.0849, 0.4236, 0.0845, 0.4063, 0.1503, 0.1149, 0.0735, 0.0940], + device='cuda:2'), in_proj_covar=tensor([0.0567, 0.0618, 0.0642, 0.0590, 0.0666, 0.0567, 0.0560, 0.0620], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:23:22,132 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9982, 1.6640, 1.3740, 1.4961, 1.3599, 1.2049, 1.2125, 1.2672], + device='cuda:2'), covar=tensor([0.1121, 0.0434, 0.1220, 0.0559, 0.0708, 0.1384, 0.0913, 0.0769], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0232, 0.0324, 0.0305, 0.0302, 0.0330, 0.0344, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:23:35,464 INFO [train.py:901] (2/4) Epoch 15, batch 7200, loss[loss=0.2528, simple_loss=0.3207, pruned_loss=0.09247, over 8433.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2971, pruned_loss=0.07021, over 1604415.52 frames. ], batch size: 49, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:23:36,812 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.418e+02 2.853e+02 3.692e+02 6.645e+02, threshold=5.707e+02, percent-clipped=2.0 +2023-02-06 17:23:44,759 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 17:23:56,966 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 17:24:00,215 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:24:08,687 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 17:24:10,194 INFO [train.py:901] (2/4) Epoch 15, batch 7250, loss[loss=0.1964, simple_loss=0.2799, pruned_loss=0.05641, over 8029.00 frames. ], tot_loss[loss=0.218, simple_loss=0.297, pruned_loss=0.06953, over 1608627.83 frames. ], batch size: 22, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:24:17,175 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0704, 1.4980, 1.7290, 1.4440, 0.9003, 1.5794, 1.7335, 1.5343], + device='cuda:2'), covar=tensor([0.0481, 0.1220, 0.1677, 0.1356, 0.0584, 0.1383, 0.0652, 0.0631], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0157, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 17:24:17,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120423.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:24:45,987 INFO [train.py:901] (2/4) Epoch 15, batch 7300, loss[loss=0.206, simple_loss=0.2962, pruned_loss=0.05786, over 8587.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2972, pruned_loss=0.06978, over 1611479.23 frames. ], batch size: 34, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:24:47,338 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.423e+02 2.925e+02 3.483e+02 5.889e+02, threshold=5.849e+02, percent-clipped=3.0 +2023-02-06 17:24:55,830 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.73 vs. limit=5.0 +2023-02-06 17:25:20,536 INFO [train.py:901] (2/4) Epoch 15, batch 7350, loss[loss=0.175, simple_loss=0.256, pruned_loss=0.04704, over 7657.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2983, pruned_loss=0.07009, over 1613743.44 frames. ], batch size: 19, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:25:45,380 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 17:25:56,238 INFO [train.py:901] (2/4) Epoch 15, batch 7400, loss[loss=0.1748, simple_loss=0.2574, pruned_loss=0.04615, over 7782.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.299, pruned_loss=0.0704, over 1618854.00 frames. ], batch size: 19, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:25:57,545 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.487e+02 3.190e+02 4.160e+02 9.613e+02, threshold=6.380e+02, percent-clipped=9.0 +2023-02-06 17:26:04,620 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 17:26:30,896 INFO [train.py:901] (2/4) Epoch 15, batch 7450, loss[loss=0.2357, simple_loss=0.3069, pruned_loss=0.0822, over 8520.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2985, pruned_loss=0.07055, over 1616467.88 frames. ], batch size: 28, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:26:42,790 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 17:26:45,071 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7199, 1.9536, 1.6564, 2.3039, 1.1126, 1.4362, 1.6979, 1.9692], + device='cuda:2'), covar=tensor([0.0782, 0.0712, 0.1012, 0.0465, 0.1120, 0.1358, 0.0791, 0.0695], + device='cuda:2'), in_proj_covar=tensor([0.0237, 0.0206, 0.0251, 0.0217, 0.0215, 0.0254, 0.0258, 0.0216], + 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') +2023-02-06 17:27:06,387 INFO [train.py:901] (2/4) Epoch 15, batch 7500, loss[loss=0.2309, simple_loss=0.3137, pruned_loss=0.07403, over 8504.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2983, pruned_loss=0.07035, over 1616887.42 frames. ], batch size: 28, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:27:07,745 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.388e+02 2.853e+02 3.831e+02 7.536e+02, threshold=5.707e+02, percent-clipped=4.0 +2023-02-06 17:27:27,373 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120694.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:27:40,251 INFO [train.py:901] (2/4) Epoch 15, batch 7550, loss[loss=0.2294, simple_loss=0.3192, pruned_loss=0.06982, over 8245.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2993, pruned_loss=0.07092, over 1615687.63 frames. ], batch size: 24, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:28:14,829 INFO [train.py:901] (2/4) Epoch 15, batch 7600, loss[loss=0.2073, simple_loss=0.3005, pruned_loss=0.05706, over 8357.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2988, pruned_loss=0.07072, over 1615962.44 frames. ], batch size: 24, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:28:16,200 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.426e+02 3.048e+02 3.965e+02 8.844e+02, threshold=6.096e+02, percent-clipped=6.0 +2023-02-06 17:28:42,985 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.42 vs. limit=5.0 +2023-02-06 17:28:50,147 INFO [train.py:901] (2/4) Epoch 15, batch 7650, loss[loss=0.2223, simple_loss=0.2959, pruned_loss=0.0743, over 7801.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2983, pruned_loss=0.07018, over 1618742.29 frames. ], batch size: 20, lr: 5.01e-03, grad_scale: 8.0 +2023-02-06 17:29:25,339 INFO [train.py:901] (2/4) Epoch 15, batch 7700, loss[loss=0.2329, simple_loss=0.3131, pruned_loss=0.07634, over 8486.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.299, pruned_loss=0.07036, over 1622166.97 frames. ], batch size: 26, lr: 5.01e-03, grad_scale: 8.0 +2023-02-06 17:29:27,393 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.773e+02 2.503e+02 3.087e+02 4.175e+02 9.539e+02, threshold=6.174e+02, percent-clipped=7.0 +2023-02-06 17:29:32,696 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 17:29:46,994 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 +2023-02-06 17:29:52,773 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 17:30:01,556 INFO [train.py:901] (2/4) Epoch 15, batch 7750, loss[loss=0.2287, simple_loss=0.3124, pruned_loss=0.07244, over 8472.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2993, pruned_loss=0.0706, over 1620028.87 frames. ], batch size: 29, lr: 5.01e-03, grad_scale: 8.0 +2023-02-06 17:30:36,085 INFO [train.py:901] (2/4) Epoch 15, batch 7800, loss[loss=0.2148, simple_loss=0.2795, pruned_loss=0.07505, over 7230.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2992, pruned_loss=0.07077, over 1619732.82 frames. ], batch size: 16, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:30:38,108 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.376e+02 2.783e+02 3.266e+02 5.993e+02, threshold=5.565e+02, percent-clipped=0.0 +2023-02-06 17:31:09,466 INFO [train.py:901] (2/4) Epoch 15, batch 7850, loss[loss=0.2044, simple_loss=0.2971, pruned_loss=0.05588, over 8314.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.3001, pruned_loss=0.0713, over 1618622.03 frames. ], batch size: 25, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:31:14,873 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:31:26,042 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121038.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:31:42,563 INFO [train.py:901] (2/4) Epoch 15, batch 7900, loss[loss=0.2275, simple_loss=0.3151, pruned_loss=0.06996, over 8555.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3008, pruned_loss=0.07182, over 1621260.30 frames. ], batch size: 34, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:31:44,514 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.419e+02 3.139e+02 4.114e+02 1.036e+03, threshold=6.279e+02, percent-clipped=8.0 +2023-02-06 17:32:15,975 INFO [train.py:901] (2/4) Epoch 15, batch 7950, loss[loss=0.2081, simple_loss=0.2966, pruned_loss=0.05985, over 8510.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3006, pruned_loss=0.0717, over 1615328.90 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:32:42,023 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:32:48,355 INFO [train.py:901] (2/4) Epoch 15, batch 8000, loss[loss=0.2286, simple_loss=0.3204, pruned_loss=0.06837, over 8492.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3006, pruned_loss=0.072, over 1617664.97 frames. ], batch size: 26, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:32:50,388 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.460e+02 2.992e+02 3.696e+02 7.694e+02, threshold=5.984e+02, percent-clipped=2.0 +2023-02-06 17:33:01,524 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121182.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:33:22,838 INFO [train.py:901] (2/4) Epoch 15, batch 8050, loss[loss=0.2252, simple_loss=0.2981, pruned_loss=0.07616, over 8090.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2986, pruned_loss=0.07135, over 1603599.81 frames. ], batch size: 21, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:33:55,752 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 17:34:00,925 INFO [train.py:901] (2/4) Epoch 16, batch 0, loss[loss=0.2302, simple_loss=0.3024, pruned_loss=0.07899, over 8286.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3024, pruned_loss=0.07899, over 8286.00 frames. ], batch size: 23, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:34:00,925 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 17:34:10,746 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6787, 1.6783, 2.6650, 1.3632, 2.0772, 2.8583, 2.9148, 2.5272], + device='cuda:2'), covar=tensor([0.1193, 0.1447, 0.0415, 0.2216, 0.0931, 0.0344, 0.0743, 0.0616], + device='cuda:2'), in_proj_covar=tensor([0.0280, 0.0309, 0.0271, 0.0301, 0.0288, 0.0248, 0.0377, 0.0297], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:34:11,911 INFO [train.py:935] (2/4) Epoch 16, validation: loss=0.1795, simple_loss=0.2801, pruned_loss=0.03944, over 944034.00 frames. +2023-02-06 17:34:11,912 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 17:34:24,910 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.543e+02 3.194e+02 4.084e+02 8.334e+02, threshold=6.389e+02, percent-clipped=7.0 +2023-02-06 17:34:26,235 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 17:34:47,406 INFO [train.py:901] (2/4) Epoch 16, batch 50, loss[loss=0.2182, simple_loss=0.3083, pruned_loss=0.06401, over 8254.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3028, pruned_loss=0.07288, over 365253.25 frames. ], batch size: 24, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:34:55,517 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5707, 1.9058, 3.1936, 1.3531, 2.2885, 1.9741, 1.5539, 2.3199], + device='cuda:2'), covar=tensor([0.1855, 0.2435, 0.0725, 0.4311, 0.1706, 0.3034, 0.2293, 0.2147], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0559, 0.0541, 0.0612, 0.0634, 0.0577, 0.0505, 0.0627], + 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') +2023-02-06 17:35:02,271 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 17:35:09,789 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121329.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 17:35:20,725 INFO [train.py:901] (2/4) Epoch 16, batch 100, loss[loss=0.2326, simple_loss=0.295, pruned_loss=0.08513, over 7789.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3015, pruned_loss=0.07111, over 643269.61 frames. ], batch size: 19, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:35:24,731 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 17:35:33,287 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121365.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:35:33,868 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.470e+02 2.913e+02 3.674e+02 6.203e+02, threshold=5.826e+02, percent-clipped=0.0 +2023-02-06 17:35:53,850 INFO [train.py:901] (2/4) Epoch 16, batch 150, loss[loss=0.208, simple_loss=0.294, pruned_loss=0.06097, over 8032.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3029, pruned_loss=0.07232, over 861258.18 frames. ], batch size: 22, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:36:04,297 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:11,281 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-06 17:36:15,655 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121425.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:20,976 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7808, 1.7646, 1.9283, 1.6589, 0.9711, 1.5923, 2.1485, 1.9568], + device='cuda:2'), covar=tensor([0.0434, 0.1146, 0.1575, 0.1250, 0.0611, 0.1378, 0.0632, 0.0614], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0156, 0.0100, 0.0161, 0.0113, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 17:36:21,676 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:30,076 INFO [train.py:901] (2/4) Epoch 16, batch 200, loss[loss=0.1978, simple_loss=0.2624, pruned_loss=0.06656, over 7662.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3017, pruned_loss=0.07222, over 1027426.61 frames. ], batch size: 19, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:36:43,676 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.385e+02 2.940e+02 3.661e+02 7.455e+02, threshold=5.881e+02, percent-clipped=4.0 +2023-02-06 17:36:53,373 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121480.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:37:04,016 INFO [train.py:901] (2/4) Epoch 16, batch 250, loss[loss=0.2303, simple_loss=0.3028, pruned_loss=0.07892, over 8148.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3018, pruned_loss=0.07268, over 1157609.57 frames. ], batch size: 22, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:37:18,646 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 17:37:24,826 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121526.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:37:28,166 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 17:37:36,145 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 17:37:39,707 INFO [train.py:901] (2/4) Epoch 16, batch 300, loss[loss=0.2238, simple_loss=0.3001, pruned_loss=0.07376, over 7808.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3007, pruned_loss=0.07239, over 1258040.72 frames. ], batch size: 20, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:37:54,064 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.529e+02 3.079e+02 3.820e+02 7.739e+02, threshold=6.158e+02, percent-clipped=5.0 +2023-02-06 17:38:01,960 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2672, 1.9125, 2.6666, 2.1249, 2.4581, 2.1737, 1.8626, 1.3282], + device='cuda:2'), covar=tensor([0.4586, 0.4366, 0.1535, 0.3344, 0.2448, 0.2637, 0.1848, 0.4762], + device='cuda:2'), in_proj_covar=tensor([0.0911, 0.0921, 0.0756, 0.0890, 0.0955, 0.0840, 0.0722, 0.0799], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:38:09,460 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6677, 1.4527, 1.8024, 1.4296, 0.9443, 1.4958, 2.0337, 1.7332], + device='cuda:2'), covar=tensor([0.0447, 0.1355, 0.1712, 0.1519, 0.0653, 0.1533, 0.0695, 0.0685], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0157, 0.0100, 0.0162, 0.0113, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 17:38:14,587 INFO [train.py:901] (2/4) Epoch 16, batch 350, loss[loss=0.2664, simple_loss=0.3422, pruned_loss=0.09535, over 8578.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3004, pruned_loss=0.07173, over 1341557.57 frames. ], batch size: 31, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:38:15,674 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2023-02-06 17:38:45,977 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121641.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:38:49,789 INFO [train.py:901] (2/4) Epoch 16, batch 400, loss[loss=0.2164, simple_loss=0.301, pruned_loss=0.06592, over 8317.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3014, pruned_loss=0.07196, over 1405871.30 frames. ], batch size: 25, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:38:52,983 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 17:39:04,290 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.467e+02 3.087e+02 3.761e+02 6.357e+02, threshold=6.175e+02, percent-clipped=1.0 +2023-02-06 17:39:09,178 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121673.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 17:39:17,981 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1497, 1.4313, 4.3287, 1.5924, 3.8064, 3.6048, 3.9087, 3.7745], + device='cuda:2'), covar=tensor([0.0548, 0.4500, 0.0534, 0.3919, 0.1134, 0.0928, 0.0541, 0.0682], + device='cuda:2'), in_proj_covar=tensor([0.0557, 0.0611, 0.0637, 0.0582, 0.0656, 0.0564, 0.0556, 0.0615], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:39:24,231 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 17:39:25,132 INFO [train.py:901] (2/4) Epoch 16, batch 450, loss[loss=0.2477, simple_loss=0.3008, pruned_loss=0.09726, over 7539.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3008, pruned_loss=0.0718, over 1449688.25 frames. ], batch size: 18, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:39:48,309 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4756, 1.7724, 2.7201, 1.3114, 2.0505, 1.8509, 1.5257, 2.0006], + device='cuda:2'), covar=tensor([0.1425, 0.1988, 0.0603, 0.3401, 0.1481, 0.2341, 0.1624, 0.1917], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0559, 0.0542, 0.0610, 0.0632, 0.0577, 0.0504, 0.0626], + 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') +2023-02-06 17:39:52,402 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:39:59,042 INFO [train.py:901] (2/4) Epoch 16, batch 500, loss[loss=0.2326, simple_loss=0.3013, pruned_loss=0.08195, over 7922.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3019, pruned_loss=0.07229, over 1488569.79 frames. ], batch size: 20, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:40:10,166 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4173, 1.4899, 4.5897, 1.6488, 4.1103, 3.8514, 4.1250, 4.0352], + device='cuda:2'), covar=tensor([0.0448, 0.4234, 0.0451, 0.3845, 0.0893, 0.0842, 0.0506, 0.0552], + device='cuda:2'), in_proj_covar=tensor([0.0554, 0.0609, 0.0635, 0.0582, 0.0653, 0.0563, 0.0557, 0.0613], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:40:10,954 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:40:14,769 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.435e+02 2.838e+02 3.555e+02 6.989e+02, threshold=5.677e+02, percent-clipped=1.0 +2023-02-06 17:40:17,014 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:40:29,898 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121788.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 17:40:35,812 INFO [train.py:901] (2/4) Epoch 16, batch 550, loss[loss=0.1879, simple_loss=0.2735, pruned_loss=0.05118, over 8020.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3016, pruned_loss=0.07162, over 1522172.03 frames. ], batch size: 22, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:40:52,141 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8041, 1.7438, 2.3791, 1.6157, 1.2238, 2.4259, 0.3872, 1.3874], + device='cuda:2'), covar=tensor([0.2250, 0.1520, 0.0442, 0.1650, 0.3478, 0.0455, 0.2759, 0.1670], + device='cuda:2'), in_proj_covar=tensor([0.0176, 0.0181, 0.0112, 0.0214, 0.0257, 0.0116, 0.0163, 0.0178], + 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') +2023-02-06 17:40:58,378 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 +2023-02-06 17:41:09,174 INFO [train.py:901] (2/4) Epoch 16, batch 600, loss[loss=0.2395, simple_loss=0.3195, pruned_loss=0.07976, over 8030.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3008, pruned_loss=0.07128, over 1543608.49 frames. ], batch size: 22, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:41:22,430 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.425e+02 3.086e+02 4.175e+02 1.417e+03, threshold=6.173e+02, percent-clipped=9.0 +2023-02-06 17:41:26,595 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 17:41:36,805 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121884.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:41:44,735 INFO [train.py:901] (2/4) Epoch 16, batch 650, loss[loss=0.2349, simple_loss=0.3114, pruned_loss=0.07924, over 8511.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.3, pruned_loss=0.07033, over 1559644.27 frames. ], batch size: 28, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:41:45,637 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:42:02,833 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:42:05,542 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5693, 2.3337, 3.4831, 2.6202, 3.2338, 2.4740, 2.1107, 1.7736], + device='cuda:2'), covar=tensor([0.4759, 0.4858, 0.1616, 0.3325, 0.2295, 0.2733, 0.1825, 0.5324], + device='cuda:2'), in_proj_covar=tensor([0.0912, 0.0923, 0.0759, 0.0890, 0.0956, 0.0843, 0.0721, 0.0797], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:42:18,664 INFO [train.py:901] (2/4) Epoch 16, batch 700, loss[loss=0.2155, simple_loss=0.2814, pruned_loss=0.07479, over 8238.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.3004, pruned_loss=0.07017, over 1578042.63 frames. ], batch size: 22, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:42:32,096 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 2.496e+02 2.978e+02 3.542e+02 1.118e+03, threshold=5.957e+02, percent-clipped=1.0 +2023-02-06 17:42:33,592 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121968.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:42:53,716 INFO [train.py:901] (2/4) Epoch 16, batch 750, loss[loss=0.2468, simple_loss=0.3138, pruned_loss=0.08989, over 8549.00 frames. ], tot_loss[loss=0.22, simple_loss=0.3, pruned_loss=0.06995, over 1589828.73 frames. ], batch size: 49, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:43:07,454 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1701, 3.0679, 2.8138, 1.6159, 2.8120, 2.8595, 2.8160, 2.7486], + device='cuda:2'), covar=tensor([0.1299, 0.0920, 0.1567, 0.4748, 0.1232, 0.1248, 0.1709, 0.1163], + device='cuda:2'), in_proj_covar=tensor([0.0496, 0.0410, 0.0413, 0.0514, 0.0403, 0.0410, 0.0396, 0.0358], + 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') +2023-02-06 17:43:14,269 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 17:43:23,869 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 17:43:28,677 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122044.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 17:43:29,776 INFO [train.py:901] (2/4) Epoch 16, batch 800, loss[loss=0.2202, simple_loss=0.2783, pruned_loss=0.08108, over 7683.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2995, pruned_loss=0.06997, over 1598389.26 frames. ], batch size: 18, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:43:43,079 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.422e+02 2.925e+02 3.576e+02 6.712e+02, threshold=5.851e+02, percent-clipped=2.0 +2023-02-06 17:43:45,456 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122069.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 17:44:03,133 INFO [train.py:901] (2/4) Epoch 16, batch 850, loss[loss=0.2231, simple_loss=0.287, pruned_loss=0.07963, over 7809.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2996, pruned_loss=0.07021, over 1604084.13 frames. ], batch size: 20, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:44:07,377 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7691, 1.9643, 1.7393, 2.3108, 1.1228, 1.4742, 1.6312, 2.0776], + device='cuda:2'), covar=tensor([0.0713, 0.0724, 0.0868, 0.0418, 0.1066, 0.1345, 0.0759, 0.0623], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0203, 0.0248, 0.0214, 0.0213, 0.0251, 0.0255, 0.0214], + 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') +2023-02-06 17:44:31,621 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:44:34,939 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:44:38,759 INFO [train.py:901] (2/4) Epoch 16, batch 900, loss[loss=0.1983, simple_loss=0.2777, pruned_loss=0.05951, over 7802.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2987, pruned_loss=0.06983, over 1607804.45 frames. ], batch size: 19, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:44:52,336 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:44:52,800 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.482e+02 3.085e+02 4.013e+02 7.148e+02, threshold=6.170e+02, percent-clipped=4.0 +2023-02-06 17:45:12,883 INFO [train.py:901] (2/4) Epoch 16, batch 950, loss[loss=0.2685, simple_loss=0.3345, pruned_loss=0.1012, over 8472.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2984, pruned_loss=0.06948, over 1610872.95 frames. ], batch size: 25, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:45:24,862 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:45:40,113 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 17:45:49,036 INFO [train.py:901] (2/4) Epoch 16, batch 1000, loss[loss=0.2369, simple_loss=0.3167, pruned_loss=0.07853, over 8199.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2984, pruned_loss=0.06934, over 1614142.85 frames. ], batch size: 23, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:46:03,407 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.462e+02 3.004e+02 3.600e+02 8.525e+02, threshold=6.009e+02, percent-clipped=4.0 +2023-02-06 17:46:14,178 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 17:46:23,685 INFO [train.py:901] (2/4) Epoch 16, batch 1050, loss[loss=0.185, simple_loss=0.2761, pruned_loss=0.04697, over 7954.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2993, pruned_loss=0.0698, over 1617503.28 frames. ], batch size: 21, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:46:26,447 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 17:46:34,620 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:46:51,300 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122337.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:46:57,870 INFO [train.py:901] (2/4) Epoch 16, batch 1100, loss[loss=0.2017, simple_loss=0.2734, pruned_loss=0.06497, over 7973.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2993, pruned_loss=0.0699, over 1617256.60 frames. ], batch size: 21, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:47:12,614 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.699e+02 3.204e+02 3.982e+02 8.590e+02, threshold=6.408e+02, percent-clipped=5.0 +2023-02-06 17:47:33,533 INFO [train.py:901] (2/4) Epoch 16, batch 1150, loss[loss=0.2222, simple_loss=0.2962, pruned_loss=0.07407, over 8139.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2986, pruned_loss=0.06916, over 1618483.50 frames. ], batch size: 22, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:47:38,292 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 17:47:42,503 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3327, 1.4368, 4.5151, 1.6563, 3.9947, 3.7021, 4.0816, 3.9508], + device='cuda:2'), covar=tensor([0.0541, 0.4644, 0.0548, 0.3991, 0.1076, 0.1017, 0.0592, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0557, 0.0610, 0.0643, 0.0588, 0.0658, 0.0565, 0.0562, 0.0622], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:47:43,906 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7076, 1.6721, 2.0861, 1.5902, 1.2324, 2.1303, 0.3286, 1.2485], + device='cuda:2'), covar=tensor([0.1810, 0.1349, 0.0437, 0.1214, 0.3315, 0.0454, 0.2460, 0.1845], + device='cuda:2'), in_proj_covar=tensor([0.0174, 0.0179, 0.0111, 0.0210, 0.0254, 0.0115, 0.0160, 0.0177], + 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') +2023-02-06 17:47:54,849 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:48:07,550 INFO [train.py:901] (2/4) Epoch 16, batch 1200, loss[loss=0.2025, simple_loss=0.2956, pruned_loss=0.0547, over 8465.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2984, pruned_loss=0.06899, over 1617512.51 frames. ], batch size: 25, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:48:21,991 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.417e+02 3.007e+02 3.779e+02 1.089e+03, threshold=6.013e+02, percent-clipped=2.0 +2023-02-06 17:48:31,789 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:48:33,529 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 17:48:36,683 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122486.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:48:43,455 INFO [train.py:901] (2/4) Epoch 16, batch 1250, loss[loss=0.2302, simple_loss=0.3038, pruned_loss=0.07829, over 8280.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2979, pruned_loss=0.06893, over 1618646.42 frames. ], batch size: 23, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:49:19,088 INFO [train.py:901] (2/4) Epoch 16, batch 1300, loss[loss=0.2173, simple_loss=0.2975, pruned_loss=0.06856, over 7659.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2983, pruned_loss=0.0689, over 1617254.86 frames. ], batch size: 19, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:49:26,971 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:49:33,313 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.571e+02 3.105e+02 3.703e+02 6.719e+02, threshold=6.210e+02, percent-clipped=4.0 +2023-02-06 17:49:44,500 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9920, 3.4800, 2.1117, 2.7016, 2.4299, 1.7032, 2.5947, 2.9117], + device='cuda:2'), covar=tensor([0.1614, 0.0389, 0.1261, 0.0703, 0.0919, 0.1719, 0.1193, 0.0972], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0237, 0.0330, 0.0306, 0.0304, 0.0330, 0.0346, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:49:52,142 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6727, 1.5401, 2.0441, 1.5579, 1.1996, 2.0656, 0.2517, 1.1897], + device='cuda:2'), covar=tensor([0.1923, 0.1789, 0.0424, 0.1180, 0.3334, 0.0456, 0.2614, 0.1648], + device='cuda:2'), in_proj_covar=tensor([0.0176, 0.0182, 0.0112, 0.0214, 0.0257, 0.0117, 0.0162, 0.0178], + 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') +2023-02-06 17:49:54,914 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:49:56,063 INFO [train.py:901] (2/4) Epoch 16, batch 1350, loss[loss=0.2263, simple_loss=0.3075, pruned_loss=0.0725, over 7821.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2982, pruned_loss=0.06886, over 1620193.92 frames. ], batch size: 20, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:50:31,463 INFO [train.py:901] (2/4) Epoch 16, batch 1400, loss[loss=0.2502, simple_loss=0.328, pruned_loss=0.08624, over 8359.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2967, pruned_loss=0.06818, over 1617950.03 frames. ], batch size: 24, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:50:45,931 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.319e+02 2.799e+02 3.491e+02 7.123e+02, threshold=5.597e+02, percent-clipped=1.0 +2023-02-06 17:50:49,437 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:50:50,210 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 17:50:55,476 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122681.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:50:56,973 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:51:05,778 INFO [train.py:901] (2/4) Epoch 16, batch 1450, loss[loss=0.2365, simple_loss=0.3057, pruned_loss=0.0836, over 8023.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2964, pruned_loss=0.06815, over 1615944.52 frames. ], batch size: 22, lr: 4.81e-03, grad_scale: 4.0 +2023-02-06 17:51:12,699 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 17:51:15,653 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:51:42,668 INFO [train.py:901] (2/4) Epoch 16, batch 1500, loss[loss=0.2485, simple_loss=0.3248, pruned_loss=0.08614, over 8198.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2973, pruned_loss=0.06899, over 1618635.72 frames. ], batch size: 23, lr: 4.81e-03, grad_scale: 4.0 +2023-02-06 17:51:56,867 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 2.515e+02 3.024e+02 4.111e+02 8.238e+02, threshold=6.047e+02, percent-clipped=9.0 +2023-02-06 17:52:16,413 INFO [train.py:901] (2/4) Epoch 16, batch 1550, loss[loss=0.1954, simple_loss=0.2665, pruned_loss=0.06219, over 7441.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.297, pruned_loss=0.06904, over 1618300.23 frames. ], batch size: 17, lr: 4.81e-03, grad_scale: 4.0 +2023-02-06 17:52:16,630 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122796.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:52:41,665 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:52:47,997 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4052, 1.5825, 2.8568, 1.3690, 2.1289, 1.8860, 1.4403, 2.0340], + device='cuda:2'), covar=tensor([0.1987, 0.2497, 0.0689, 0.4265, 0.1469, 0.2997, 0.2235, 0.1848], + device='cuda:2'), in_proj_covar=tensor([0.0504, 0.0558, 0.0539, 0.0609, 0.0627, 0.0570, 0.0501, 0.0618], + 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') +2023-02-06 17:52:52,395 INFO [train.py:901] (2/4) Epoch 16, batch 1600, loss[loss=0.2258, simple_loss=0.3078, pruned_loss=0.07194, over 8532.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2966, pruned_loss=0.06853, over 1619898.25 frames. ], batch size: 28, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:52:55,435 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122850.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:53:07,575 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 17:53:07,650 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.387e+02 2.555e+02 3.178e+02 4.067e+02 1.179e+03, threshold=6.355e+02, percent-clipped=12.0 +2023-02-06 17:53:13,369 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:53:23,678 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:53:27,666 INFO [train.py:901] (2/4) Epoch 16, batch 1650, loss[loss=0.2131, simple_loss=0.2911, pruned_loss=0.06752, over 8084.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2956, pruned_loss=0.06818, over 1611373.36 frames. ], batch size: 21, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:53:29,433 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.03 vs. limit=5.0 +2023-02-06 17:53:49,585 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122928.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:54:02,350 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122945.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:54:02,864 INFO [train.py:901] (2/4) Epoch 16, batch 1700, loss[loss=0.234, simple_loss=0.3164, pruned_loss=0.07584, over 8021.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2968, pruned_loss=0.0688, over 1616590.50 frames. ], batch size: 22, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:54:08,415 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122953.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:54:09,136 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4984, 1.8064, 2.6997, 1.3927, 1.8887, 1.8473, 1.5523, 1.8743], + device='cuda:2'), covar=tensor([0.1971, 0.2494, 0.0767, 0.4528, 0.1806, 0.3296, 0.2245, 0.2215], + device='cuda:2'), in_proj_covar=tensor([0.0503, 0.0557, 0.0539, 0.0608, 0.0627, 0.0570, 0.0500, 0.0617], + 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') +2023-02-06 17:54:17,614 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.342e+02 2.881e+02 3.479e+02 7.679e+02, threshold=5.763e+02, percent-clipped=3.0 +2023-02-06 17:54:21,196 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9948, 1.7627, 3.3604, 1.5351, 2.3309, 3.6486, 3.6784, 3.0818], + device='cuda:2'), covar=tensor([0.1135, 0.1595, 0.0380, 0.1989, 0.1140, 0.0254, 0.0547, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0279, 0.0308, 0.0270, 0.0297, 0.0290, 0.0248, 0.0379, 0.0297], + 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') +2023-02-06 17:54:38,071 INFO [train.py:901] (2/4) Epoch 16, batch 1750, loss[loss=0.1955, simple_loss=0.2651, pruned_loss=0.06294, over 7696.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2952, pruned_loss=0.06821, over 1615830.59 frames. ], batch size: 18, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:54:42,733 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 17:55:01,805 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4073, 2.4150, 1.5878, 2.2130, 2.0715, 1.2279, 1.9250, 2.1071], + device='cuda:2'), covar=tensor([0.1759, 0.0510, 0.1557, 0.0755, 0.0889, 0.2008, 0.1289, 0.1062], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0234, 0.0327, 0.0303, 0.0302, 0.0330, 0.0344, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 17:55:12,114 INFO [train.py:901] (2/4) Epoch 16, batch 1800, loss[loss=0.2008, simple_loss=0.2837, pruned_loss=0.05893, over 8220.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2947, pruned_loss=0.06825, over 1616863.61 frames. ], batch size: 22, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:55:16,362 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123052.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:55:18,921 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2520, 1.5807, 1.7806, 1.4784, 1.1569, 1.5629, 1.8000, 1.8783], + device='cuda:2'), covar=tensor([0.0461, 0.1156, 0.1597, 0.1340, 0.0563, 0.1439, 0.0657, 0.0578], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0156, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 17:55:25,835 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9810, 1.9226, 2.5796, 1.5052, 1.3936, 2.5471, 0.4237, 1.4702], + device='cuda:2'), covar=tensor([0.1862, 0.1415, 0.0313, 0.2025, 0.3239, 0.0342, 0.2504, 0.1817], + device='cuda:2'), in_proj_covar=tensor([0.0173, 0.0179, 0.0111, 0.0211, 0.0254, 0.0115, 0.0161, 0.0176], + 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') +2023-02-06 17:55:27,695 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.489e+02 2.922e+02 3.750e+02 7.056e+02, threshold=5.843e+02, percent-clipped=4.0 +2023-02-06 17:55:35,411 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123077.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:55:48,781 INFO [train.py:901] (2/4) Epoch 16, batch 1850, loss[loss=0.2667, simple_loss=0.3363, pruned_loss=0.09857, over 7182.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2945, pruned_loss=0.06815, over 1613344.11 frames. ], batch size: 71, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:56:22,235 INFO [train.py:901] (2/4) Epoch 16, batch 1900, loss[loss=0.2518, simple_loss=0.3399, pruned_loss=0.08181, over 8342.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2958, pruned_loss=0.06854, over 1616530.46 frames. ], batch size: 26, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:56:36,278 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.569e+02 3.077e+02 4.069e+02 9.708e+02, threshold=6.154e+02, percent-clipped=7.0 +2023-02-06 17:56:57,741 INFO [train.py:901] (2/4) Epoch 16, batch 1950, loss[loss=0.2219, simple_loss=0.2806, pruned_loss=0.08163, over 7792.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.296, pruned_loss=0.06919, over 1615572.45 frames. ], batch size: 19, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:56:59,131 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 17:57:01,385 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:57:11,361 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 17:57:18,813 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123226.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:57:24,081 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123234.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:57:30,728 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 17:57:32,102 INFO [train.py:901] (2/4) Epoch 16, batch 2000, loss[loss=0.1605, simple_loss=0.2494, pruned_loss=0.03583, over 7783.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2957, pruned_loss=0.06881, over 1613090.46 frames. ], batch size: 19, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:57:46,356 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.202e+02 2.631e+02 3.355e+02 6.225e+02, threshold=5.263e+02, percent-clipped=1.0 +2023-02-06 17:58:05,870 INFO [train.py:901] (2/4) Epoch 16, batch 2050, loss[loss=0.219, simple_loss=0.3034, pruned_loss=0.06727, over 8474.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2944, pruned_loss=0.06835, over 1610066.92 frames. ], batch size: 25, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:58:31,864 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123332.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:58:41,130 INFO [train.py:901] (2/4) Epoch 16, batch 2100, loss[loss=0.182, simple_loss=0.2585, pruned_loss=0.05272, over 7688.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2949, pruned_loss=0.06863, over 1611329.32 frames. ], batch size: 18, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:58:43,352 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:58:54,988 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.773e+02 2.517e+02 3.000e+02 3.631e+02 1.037e+03, threshold=6.000e+02, percent-clipped=6.0 +2023-02-06 17:59:14,279 INFO [train.py:901] (2/4) Epoch 16, batch 2150, loss[loss=0.1867, simple_loss=0.2619, pruned_loss=0.05574, over 7311.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.297, pruned_loss=0.07003, over 1609985.44 frames. ], batch size: 16, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:59:19,816 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3808, 2.0513, 2.8538, 2.2971, 2.8771, 2.3560, 2.0455, 1.4435], + device='cuda:2'), covar=tensor([0.4892, 0.4604, 0.1564, 0.3089, 0.2144, 0.2482, 0.1855, 0.5103], + device='cuda:2'), in_proj_covar=tensor([0.0920, 0.0928, 0.0763, 0.0900, 0.0964, 0.0846, 0.0720, 0.0803], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 17:59:22,908 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2383, 3.1978, 2.9665, 1.5560, 2.9563, 2.8645, 2.9088, 2.7160], + device='cuda:2'), covar=tensor([0.1347, 0.0899, 0.1345, 0.4461, 0.1154, 0.1231, 0.1585, 0.1264], + device='cuda:2'), in_proj_covar=tensor([0.0493, 0.0410, 0.0410, 0.0511, 0.0404, 0.0410, 0.0397, 0.0358], + 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') +2023-02-06 17:59:50,132 INFO [train.py:901] (2/4) Epoch 16, batch 2200, loss[loss=0.2842, simple_loss=0.3444, pruned_loss=0.112, over 8459.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2987, pruned_loss=0.07095, over 1617102.11 frames. ], batch size: 25, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 17:59:50,271 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123446.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:00:04,135 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.694e+02 3.295e+02 4.036e+02 1.292e+03, threshold=6.590e+02, percent-clipped=6.0 +2023-02-06 18:00:23,391 INFO [train.py:901] (2/4) Epoch 16, batch 2250, loss[loss=0.2181, simple_loss=0.292, pruned_loss=0.07211, over 8563.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2974, pruned_loss=0.07045, over 1614150.96 frames. ], batch size: 39, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:00:46,379 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-06 18:00:58,618 INFO [train.py:901] (2/4) Epoch 16, batch 2300, loss[loss=0.2231, simple_loss=0.29, pruned_loss=0.07814, over 7797.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2977, pruned_loss=0.07013, over 1612979.23 frames. ], batch size: 19, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:01:13,226 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.374e+02 2.935e+02 3.719e+02 2.594e+03, threshold=5.871e+02, percent-clipped=2.0 +2023-02-06 18:01:22,765 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4901, 1.8010, 1.8062, 1.1810, 1.9534, 1.4981, 0.4826, 1.6961], + device='cuda:2'), covar=tensor([0.0416, 0.0280, 0.0217, 0.0411, 0.0266, 0.0607, 0.0636, 0.0216], + device='cuda:2'), in_proj_covar=tensor([0.0423, 0.0364, 0.0313, 0.0416, 0.0350, 0.0507, 0.0372, 0.0389], + device='cuda:2'), out_proj_covar=tensor([1.1640e-04, 9.7522e-05, 8.3455e-05, 1.1214e-04, 9.4408e-05, 1.4701e-04, + 1.0201e-04, 1.0529e-04], device='cuda:2') +2023-02-06 18:01:32,631 INFO [train.py:901] (2/4) Epoch 16, batch 2350, loss[loss=0.2476, simple_loss=0.317, pruned_loss=0.0891, over 8138.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2981, pruned_loss=0.07041, over 1614097.08 frames. ], batch size: 22, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:01:38,851 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:01:55,683 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:02:06,131 INFO [train.py:901] (2/4) Epoch 16, batch 2400, loss[loss=0.2248, simple_loss=0.305, pruned_loss=0.07227, over 8335.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.298, pruned_loss=0.07028, over 1615266.69 frames. ], batch size: 25, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:02:22,333 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.321e+02 3.011e+02 3.485e+02 7.740e+02, threshold=6.021e+02, percent-clipped=5.0 +2023-02-06 18:02:28,429 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123676.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:02:42,479 INFO [train.py:901] (2/4) Epoch 16, batch 2450, loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.09177, over 8565.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2977, pruned_loss=0.06978, over 1617294.45 frames. ], batch size: 39, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:03:15,342 INFO [train.py:901] (2/4) Epoch 16, batch 2500, loss[loss=0.2038, simple_loss=0.2738, pruned_loss=0.06688, over 7162.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2987, pruned_loss=0.07102, over 1614643.88 frames. ], batch size: 16, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:03:25,454 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0300, 1.7129, 2.2809, 1.6795, 1.1653, 1.8910, 2.1332, 2.2429], + device='cuda:2'), covar=tensor([0.0427, 0.1173, 0.1505, 0.1297, 0.0570, 0.1347, 0.0599, 0.0532], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0157, 0.0100, 0.0163, 0.0114, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 18:03:29,367 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.388e+02 3.009e+02 3.987e+02 1.163e+03, threshold=6.019e+02, percent-clipped=7.0 +2023-02-06 18:03:46,948 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123790.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:03:47,745 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:03:50,911 INFO [train.py:901] (2/4) Epoch 16, batch 2550, loss[loss=0.2255, simple_loss=0.3033, pruned_loss=0.07383, over 8323.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2985, pruned_loss=0.07115, over 1614396.12 frames. ], batch size: 25, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:04:24,936 INFO [train.py:901] (2/4) Epoch 16, batch 2600, loss[loss=0.2512, simple_loss=0.3076, pruned_loss=0.09744, over 7786.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2982, pruned_loss=0.07128, over 1609391.95 frames. ], batch size: 19, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:04:38,925 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.447e+02 2.814e+02 3.524e+02 5.517e+02, threshold=5.629e+02, percent-clipped=0.0 +2023-02-06 18:04:54,348 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:04:58,762 INFO [train.py:901] (2/4) Epoch 16, batch 2650, loss[loss=0.2406, simple_loss=0.3237, pruned_loss=0.07874, over 8503.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2978, pruned_loss=0.07092, over 1607558.67 frames. ], batch size: 28, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:05:06,361 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:05:34,191 INFO [train.py:901] (2/4) Epoch 16, batch 2700, loss[loss=0.2191, simple_loss=0.2978, pruned_loss=0.07025, over 8648.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2986, pruned_loss=0.07103, over 1613524.55 frames. ], batch size: 34, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:05:48,205 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.455e+02 3.188e+02 4.135e+02 8.908e+02, threshold=6.377e+02, percent-clipped=7.0 +2023-02-06 18:06:07,595 INFO [train.py:901] (2/4) Epoch 16, batch 2750, loss[loss=0.2273, simple_loss=0.3165, pruned_loss=0.06906, over 8654.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.3001, pruned_loss=0.07151, over 1615255.47 frames. ], batch size: 39, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:06:45,090 INFO [train.py:901] (2/4) Epoch 16, batch 2800, loss[loss=0.2464, simple_loss=0.3285, pruned_loss=0.08211, over 8539.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2988, pruned_loss=0.07051, over 1615751.78 frames. ], batch size: 49, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:06:46,013 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:06:50,849 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:06:59,429 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.578e+02 3.039e+02 4.001e+02 1.196e+03, threshold=6.079e+02, percent-clipped=5.0 +2023-02-06 18:07:03,063 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:07:08,263 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:07:18,991 INFO [train.py:901] (2/4) Epoch 16, batch 2850, loss[loss=0.2255, simple_loss=0.3055, pruned_loss=0.07278, over 8655.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.299, pruned_loss=0.07041, over 1618130.30 frames. ], batch size: 34, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:07:22,868 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-06 18:07:55,341 INFO [train.py:901] (2/4) Epoch 16, batch 2900, loss[loss=0.1911, simple_loss=0.278, pruned_loss=0.05207, over 7806.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2979, pruned_loss=0.06972, over 1615042.01 frames. ], batch size: 20, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:08:06,284 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:08:06,322 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9119, 1.6538, 2.0948, 1.8371, 1.9231, 1.9267, 1.6753, 0.7772], + device='cuda:2'), covar=tensor([0.5179, 0.4409, 0.1636, 0.2748, 0.2056, 0.2673, 0.1884, 0.4526], + device='cuda:2'), in_proj_covar=tensor([0.0914, 0.0924, 0.0758, 0.0894, 0.0961, 0.0848, 0.0723, 0.0797], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:08:10,038 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.456e+02 3.206e+02 4.387e+02 8.191e+02, threshold=6.412e+02, percent-clipped=4.0 +2023-02-06 18:08:22,886 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:08:29,509 INFO [train.py:901] (2/4) Epoch 16, batch 2950, loss[loss=0.2445, simple_loss=0.3241, pruned_loss=0.08245, over 8182.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2981, pruned_loss=0.06952, over 1618233.05 frames. ], batch size: 23, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:08:29,747 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4038, 1.6242, 1.6325, 1.0578, 1.7628, 1.3100, 0.2676, 1.6161], + device='cuda:2'), covar=tensor([0.0362, 0.0260, 0.0224, 0.0387, 0.0288, 0.0753, 0.0668, 0.0213], + device='cuda:2'), in_proj_covar=tensor([0.0425, 0.0363, 0.0311, 0.0417, 0.0350, 0.0508, 0.0371, 0.0387], + device='cuda:2'), out_proj_covar=tensor([1.1693e-04, 9.7248e-05, 8.2672e-05, 1.1219e-04, 9.4535e-05, 1.4726e-04, + 1.0189e-04, 1.0494e-04], device='cuda:2') +2023-02-06 18:08:35,643 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 18:08:55,344 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124234.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:09:01,457 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.79 vs. limit=5.0 +2023-02-06 18:09:03,841 INFO [train.py:901] (2/4) Epoch 16, batch 3000, loss[loss=0.1771, simple_loss=0.2596, pruned_loss=0.0473, over 7420.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.297, pruned_loss=0.0692, over 1612766.25 frames. ], batch size: 17, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:09:03,841 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 18:09:16,271 INFO [train.py:935] (2/4) Epoch 16, validation: loss=0.1794, simple_loss=0.2796, pruned_loss=0.03958, over 944034.00 frames. +2023-02-06 18:09:16,271 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 18:09:32,700 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.393e+02 2.939e+02 3.627e+02 1.404e+03, threshold=5.877e+02, percent-clipped=2.0 +2023-02-06 18:09:49,112 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124290.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:09:52,909 INFO [train.py:901] (2/4) Epoch 16, batch 3050, loss[loss=0.2247, simple_loss=0.3189, pruned_loss=0.06524, over 8524.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2977, pruned_loss=0.06986, over 1605905.14 frames. ], batch size: 49, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:10:25,958 INFO [train.py:901] (2/4) Epoch 16, batch 3100, loss[loss=0.2019, simple_loss=0.2907, pruned_loss=0.0566, over 8490.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2975, pruned_loss=0.07001, over 1608265.74 frames. ], batch size: 28, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:10:28,055 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:10:39,329 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124366.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:10:39,809 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.464e+02 2.975e+02 4.095e+02 1.383e+03, threshold=5.950e+02, percent-clipped=6.0 +2023-02-06 18:11:01,463 INFO [train.py:901] (2/4) Epoch 16, batch 3150, loss[loss=0.2145, simple_loss=0.3088, pruned_loss=0.06015, over 8361.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2981, pruned_loss=0.07046, over 1608938.99 frames. ], batch size: 24, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:11:02,940 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:11:21,449 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:11:36,587 INFO [train.py:901] (2/4) Epoch 16, batch 3200, loss[loss=0.246, simple_loss=0.3304, pruned_loss=0.08085, over 8495.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2985, pruned_loss=0.07118, over 1609093.82 frames. ], batch size: 26, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:11:39,398 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:11:50,425 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.552e+02 3.102e+02 3.772e+02 6.284e+02, threshold=6.205e+02, percent-clipped=3.0 +2023-02-06 18:12:09,966 INFO [train.py:901] (2/4) Epoch 16, batch 3250, loss[loss=0.2406, simple_loss=0.3095, pruned_loss=0.08584, over 7922.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2987, pruned_loss=0.07116, over 1606507.59 frames. ], batch size: 20, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:12:23,020 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:12:40,936 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:12:45,404 INFO [train.py:901] (2/4) Epoch 16, batch 3300, loss[loss=0.2227, simple_loss=0.3092, pruned_loss=0.06812, over 8099.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2974, pruned_loss=0.07007, over 1606400.79 frames. ], batch size: 23, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:12:59,413 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.425e+02 2.919e+02 3.659e+02 6.879e+02, threshold=5.837e+02, percent-clipped=1.0 +2023-02-06 18:13:18,843 INFO [train.py:901] (2/4) Epoch 16, batch 3350, loss[loss=0.1864, simple_loss=0.275, pruned_loss=0.04887, over 8277.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2961, pruned_loss=0.06894, over 1611022.55 frames. ], batch size: 23, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:13:25,354 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:13:40,969 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 +2023-02-06 18:13:43,500 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:13:46,714 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:13:54,713 INFO [train.py:901] (2/4) Epoch 16, batch 3400, loss[loss=0.1966, simple_loss=0.2764, pruned_loss=0.05842, over 7940.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2951, pruned_loss=0.06827, over 1612200.87 frames. ], batch size: 20, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:14:01,615 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124656.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:14:08,830 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.420e+02 3.011e+02 3.525e+02 7.222e+02, threshold=6.022e+02, percent-clipped=3.0 +2023-02-06 18:14:18,573 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124681.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:14:28,869 INFO [train.py:901] (2/4) Epoch 16, batch 3450, loss[loss=0.2034, simple_loss=0.2842, pruned_loss=0.06133, over 8336.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.297, pruned_loss=0.06932, over 1613338.16 frames. ], batch size: 25, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:14:38,586 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124710.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:14:48,191 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2185, 3.1128, 2.9070, 1.6035, 2.8608, 2.9475, 2.8540, 2.7838], + device='cuda:2'), covar=tensor([0.1156, 0.0864, 0.1376, 0.4423, 0.1161, 0.1190, 0.1597, 0.1109], + device='cuda:2'), in_proj_covar=tensor([0.0489, 0.0408, 0.0410, 0.0511, 0.0401, 0.0408, 0.0396, 0.0355], + 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') +2023-02-06 18:15:04,129 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3559, 2.6414, 3.1768, 1.4943, 3.2180, 1.9320, 1.5376, 2.1387], + device='cuda:2'), covar=tensor([0.0650, 0.0308, 0.0171, 0.0640, 0.0370, 0.0668, 0.0804, 0.0472], + device='cuda:2'), in_proj_covar=tensor([0.0424, 0.0363, 0.0310, 0.0415, 0.0351, 0.0508, 0.0371, 0.0387], + device='cuda:2'), out_proj_covar=tensor([1.1649e-04, 9.7225e-05, 8.2362e-05, 1.1162e-04, 9.4818e-05, 1.4724e-04, + 1.0156e-04, 1.0472e-04], device='cuda:2') +2023-02-06 18:15:05,352 INFO [train.py:901] (2/4) Epoch 16, batch 3500, loss[loss=0.2622, simple_loss=0.3331, pruned_loss=0.09569, over 8335.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2969, pruned_loss=0.06912, over 1611693.58 frames. ], batch size: 49, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:15:07,646 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124749.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:09,702 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6577, 4.6433, 4.1657, 1.8920, 4.1689, 4.1224, 4.2273, 3.8283], + device='cuda:2'), covar=tensor([0.0693, 0.0534, 0.1126, 0.4457, 0.0796, 0.0847, 0.1199, 0.0811], + device='cuda:2'), in_proj_covar=tensor([0.0487, 0.0406, 0.0409, 0.0509, 0.0400, 0.0407, 0.0395, 0.0354], + 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') +2023-02-06 18:15:20,549 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.534e+02 3.082e+02 3.894e+02 7.146e+02, threshold=6.164e+02, percent-clipped=3.0 +2023-02-06 18:15:22,095 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:38,262 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 18:15:38,975 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:39,092 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:39,719 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:40,174 INFO [train.py:901] (2/4) Epoch 16, batch 3550, loss[loss=0.1902, simple_loss=0.2787, pruned_loss=0.05088, over 8452.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2973, pruned_loss=0.06943, over 1614943.40 frames. ], batch size: 25, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:15:56,877 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124820.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:16:00,143 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124825.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:16:04,089 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6193, 1.2784, 4.8319, 1.8695, 4.3015, 4.0914, 4.3727, 4.2051], + device='cuda:2'), covar=tensor([0.0553, 0.4966, 0.0424, 0.3900, 0.0984, 0.0864, 0.0523, 0.0639], + device='cuda:2'), in_proj_covar=tensor([0.0561, 0.0613, 0.0637, 0.0589, 0.0662, 0.0569, 0.0561, 0.0625], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:16:08,892 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124838.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:16:14,228 INFO [train.py:901] (2/4) Epoch 16, batch 3600, loss[loss=0.1902, simple_loss=0.2829, pruned_loss=0.04877, over 8510.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2975, pruned_loss=0.06954, over 1612605.44 frames. ], batch size: 26, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:16:30,802 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.338e+02 2.977e+02 3.463e+02 8.977e+02, threshold=5.954e+02, percent-clipped=2.0 +2023-02-06 18:16:50,928 INFO [train.py:901] (2/4) Epoch 16, batch 3650, loss[loss=0.2579, simple_loss=0.3314, pruned_loss=0.09223, over 8497.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2984, pruned_loss=0.0695, over 1613908.47 frames. ], batch size: 26, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:16:59,900 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:17:24,963 INFO [train.py:901] (2/4) Epoch 16, batch 3700, loss[loss=0.225, simple_loss=0.3024, pruned_loss=0.07381, over 8033.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2979, pruned_loss=0.06926, over 1614458.88 frames. ], batch size: 22, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:17:38,855 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 18:17:40,140 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.880e+02 2.643e+02 3.299e+02 4.315e+02 1.525e+03, threshold=6.598e+02, percent-clipped=10.0 +2023-02-06 18:18:01,618 INFO [train.py:901] (2/4) Epoch 16, batch 3750, loss[loss=0.235, simple_loss=0.3139, pruned_loss=0.07807, over 8471.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2975, pruned_loss=0.06956, over 1615463.63 frames. ], batch size: 25, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:18:04,451 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125000.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:07,883 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:20,985 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125025.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:24,555 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125030.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:35,166 INFO [train.py:901] (2/4) Epoch 16, batch 3800, loss[loss=0.2103, simple_loss=0.296, pruned_loss=0.06233, over 8471.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2958, pruned_loss=0.06876, over 1612657.02 frames. ], batch size: 29, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:18:49,275 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.284e+02 2.854e+02 3.651e+02 7.015e+02, threshold=5.709e+02, percent-clipped=3.0 +2023-02-06 18:18:58,965 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125081.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:19:10,722 INFO [train.py:901] (2/4) Epoch 16, batch 3850, loss[loss=0.2254, simple_loss=0.3022, pruned_loss=0.07433, over 8098.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2966, pruned_loss=0.06934, over 1612798.41 frames. ], batch size: 23, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:19:18,424 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125106.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:19:24,393 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125115.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:41,038 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:45,073 INFO [train.py:901] (2/4) Epoch 16, batch 3900, loss[loss=0.1835, simple_loss=0.2604, pruned_loss=0.05331, over 6832.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2957, pruned_loss=0.06852, over 1610703.88 frames. ], batch size: 15, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:19:45,092 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 18:19:45,230 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:58,195 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:59,306 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.507e+02 2.888e+02 3.601e+02 7.393e+02, threshold=5.777e+02, percent-clipped=3.0 +2023-02-06 18:20:09,607 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125182.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:20:15,224 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125190.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:20:19,099 INFO [train.py:901] (2/4) Epoch 16, batch 3950, loss[loss=0.2441, simple_loss=0.3316, pruned_loss=0.07825, over 8233.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2952, pruned_loss=0.0678, over 1609766.31 frames. ], batch size: 22, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:20:55,574 INFO [train.py:901] (2/4) Epoch 16, batch 4000, loss[loss=0.1888, simple_loss=0.2707, pruned_loss=0.05343, over 7157.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2956, pruned_loss=0.0682, over 1611447.64 frames. ], batch size: 16, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:21:08,118 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9739, 1.4904, 1.5850, 1.3733, 0.8952, 1.4512, 1.6357, 1.5585], + device='cuda:2'), covar=tensor([0.0517, 0.1278, 0.1725, 0.1410, 0.0613, 0.1519, 0.0724, 0.0646], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0156, 0.0100, 0.0162, 0.0113, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 18:21:09,904 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.424e+02 2.747e+02 3.530e+02 7.172e+02, threshold=5.495e+02, percent-clipped=3.0 +2023-02-06 18:21:22,770 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8331, 2.9475, 2.2923, 4.0068, 1.8417, 1.9607, 2.3122, 3.0560], + device='cuda:2'), covar=tensor([0.0569, 0.0789, 0.0841, 0.0210, 0.1105, 0.1354, 0.1091, 0.0735], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0203, 0.0251, 0.0212, 0.0210, 0.0248, 0.0256, 0.0214], + 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') +2023-02-06 18:21:29,134 INFO [train.py:901] (2/4) Epoch 16, batch 4050, loss[loss=0.2582, simple_loss=0.3269, pruned_loss=0.09475, over 8624.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2963, pruned_loss=0.06913, over 1612054.75 frames. ], batch size: 39, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:21:29,951 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125297.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:21:41,117 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5962, 4.5230, 4.0826, 2.2321, 4.0271, 4.1490, 4.2062, 3.8815], + device='cuda:2'), covar=tensor([0.0626, 0.0552, 0.0887, 0.4425, 0.0756, 0.1062, 0.1108, 0.0859], + device='cuda:2'), in_proj_covar=tensor([0.0495, 0.0412, 0.0410, 0.0516, 0.0405, 0.0411, 0.0402, 0.0358], + 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') +2023-02-06 18:22:05,146 INFO [train.py:901] (2/4) Epoch 16, batch 4100, loss[loss=0.1824, simple_loss=0.2817, pruned_loss=0.04152, over 8364.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2964, pruned_loss=0.06889, over 1612404.25 frames. ], batch size: 24, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:22:19,363 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.458e+02 2.941e+02 3.398e+02 7.943e+02, threshold=5.881e+02, percent-clipped=6.0 +2023-02-06 18:22:22,353 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125371.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:24,216 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125374.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:26,972 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1061, 1.8016, 2.4309, 1.9758, 2.2002, 2.0995, 1.8294, 1.0142], + device='cuda:2'), covar=tensor([0.5035, 0.4411, 0.1561, 0.3259, 0.2223, 0.2745, 0.1965, 0.4654], + device='cuda:2'), in_proj_covar=tensor([0.0911, 0.0917, 0.0755, 0.0893, 0.0957, 0.0846, 0.0716, 0.0791], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:22:37,727 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125394.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:38,920 INFO [train.py:901] (2/4) Epoch 16, batch 4150, loss[loss=0.2339, simple_loss=0.3102, pruned_loss=0.07875, over 8135.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2975, pruned_loss=0.0694, over 1617214.55 frames. ], batch size: 22, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:22:39,125 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125396.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:39,142 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125396.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:51,749 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5935, 1.3532, 2.3780, 1.3036, 2.1113, 2.5404, 2.6526, 2.1597], + device='cuda:2'), covar=tensor([0.0877, 0.1296, 0.0431, 0.1811, 0.0718, 0.0360, 0.0550, 0.0666], + device='cuda:2'), in_proj_covar=tensor([0.0277, 0.0307, 0.0273, 0.0297, 0.0289, 0.0248, 0.0381, 0.0295], + 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') +2023-02-06 18:22:55,761 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125421.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:23:09,081 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:23:14,198 INFO [train.py:901] (2/4) Epoch 16, batch 4200, loss[loss=0.2633, simple_loss=0.3432, pruned_loss=0.09165, over 8649.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2969, pruned_loss=0.06906, over 1616551.69 frames. ], batch size: 34, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:23:29,128 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.508e+02 2.881e+02 3.373e+02 7.881e+02, threshold=5.761e+02, percent-clipped=2.0 +2023-02-06 18:23:39,904 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 18:23:44,591 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:23:48,645 INFO [train.py:901] (2/4) Epoch 16, batch 4250, loss[loss=0.1647, simple_loss=0.2464, pruned_loss=0.04151, over 7645.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2965, pruned_loss=0.06854, over 1619265.74 frames. ], batch size: 19, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:24:01,577 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 18:24:09,735 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2188, 1.9903, 2.7845, 2.2629, 2.6907, 2.2832, 1.9691, 1.4961], + device='cuda:2'), covar=tensor([0.4986, 0.4549, 0.1689, 0.3194, 0.2252, 0.2547, 0.1758, 0.4945], + device='cuda:2'), in_proj_covar=tensor([0.0913, 0.0921, 0.0758, 0.0898, 0.0961, 0.0849, 0.0720, 0.0795], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:24:11,903 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 18:24:23,119 INFO [train.py:901] (2/4) Epoch 16, batch 4300, loss[loss=0.1978, simple_loss=0.2823, pruned_loss=0.05662, over 8477.00 frames. ], tot_loss[loss=0.217, simple_loss=0.297, pruned_loss=0.06853, over 1619052.17 frames. ], batch size: 28, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:24:28,632 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125553.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:24:38,405 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.516e+02 3.115e+02 4.119e+02 8.810e+02, threshold=6.231e+02, percent-clipped=6.0 +2023-02-06 18:24:46,748 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125578.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:24:52,794 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5057, 2.5069, 1.7984, 2.2682, 2.1554, 1.4583, 1.9594, 2.0188], + device='cuda:2'), covar=tensor([0.1520, 0.0352, 0.1198, 0.0628, 0.0716, 0.1563, 0.1075, 0.1169], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0232, 0.0327, 0.0304, 0.0300, 0.0333, 0.0346, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 18:24:58,864 INFO [train.py:901] (2/4) Epoch 16, batch 4350, loss[loss=0.2263, simple_loss=0.3123, pruned_loss=0.0702, over 8708.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2972, pruned_loss=0.06873, over 1619224.80 frames. ], batch size: 39, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:25:02,855 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 18:25:05,371 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:25:16,538 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:25:33,360 INFO [train.py:901] (2/4) Epoch 16, batch 4400, loss[loss=0.1926, simple_loss=0.2701, pruned_loss=0.05755, over 7776.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2969, pruned_loss=0.06868, over 1617595.29 frames. ], batch size: 19, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:25:34,015 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 18:25:48,651 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.489e+02 3.156e+02 3.927e+02 6.760e+02, threshold=6.312e+02, percent-clipped=2.0 +2023-02-06 18:26:09,564 INFO [train.py:901] (2/4) Epoch 16, batch 4450, loss[loss=0.3044, simple_loss=0.3466, pruned_loss=0.1311, over 7060.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2973, pruned_loss=0.06879, over 1617468.29 frames. ], batch size: 72, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:26:14,216 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 18:26:23,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2908, 2.6324, 3.0840, 1.6439, 3.4061, 1.8513, 1.4332, 2.1722], + device='cuda:2'), covar=tensor([0.0663, 0.0308, 0.0222, 0.0638, 0.0325, 0.0792, 0.0838, 0.0425], + device='cuda:2'), in_proj_covar=tensor([0.0430, 0.0361, 0.0314, 0.0420, 0.0354, 0.0511, 0.0372, 0.0391], + device='cuda:2'), out_proj_covar=tensor([1.1825e-04, 9.6388e-05, 8.3468e-05, 1.1270e-04, 9.5577e-05, 1.4802e-04, + 1.0204e-04, 1.0591e-04], device='cuda:2') +2023-02-06 18:26:24,218 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:26:38,220 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:26:41,742 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125743.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:26:43,638 INFO [train.py:901] (2/4) Epoch 16, batch 4500, loss[loss=0.1946, simple_loss=0.2662, pruned_loss=0.06152, over 7930.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2979, pruned_loss=0.0691, over 1619462.74 frames. ], batch size: 20, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:26:57,815 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.361e+02 2.740e+02 3.373e+02 6.169e+02, threshold=5.479e+02, percent-clipped=0.0 +2023-02-06 18:27:04,067 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 18:27:10,712 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125783.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:27:19,272 INFO [train.py:901] (2/4) Epoch 16, batch 4550, loss[loss=0.1578, simple_loss=0.2446, pruned_loss=0.03546, over 7433.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2973, pruned_loss=0.06869, over 1617818.20 frames. ], batch size: 17, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:27:20,087 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1953, 1.4214, 3.3561, 0.9749, 2.9419, 2.8124, 3.0475, 2.9068], + device='cuda:2'), covar=tensor([0.0826, 0.3744, 0.0830, 0.3958, 0.1546, 0.1187, 0.0750, 0.0957], + device='cuda:2'), in_proj_covar=tensor([0.0560, 0.0619, 0.0641, 0.0590, 0.0660, 0.0573, 0.0563, 0.0630], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:27:45,706 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:27:54,464 INFO [train.py:901] (2/4) Epoch 16, batch 4600, loss[loss=0.2108, simple_loss=0.2874, pruned_loss=0.06706, over 7927.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2967, pruned_loss=0.0686, over 1614633.26 frames. ], batch size: 20, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:27:59,460 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125853.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:03,905 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.83 vs. limit=5.0 +2023-02-06 18:28:05,122 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:08,980 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.490e+02 3.040e+02 3.897e+02 1.241e+03, threshold=6.080e+02, percent-clipped=8.0 +2023-02-06 18:28:22,164 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:30,018 INFO [train.py:901] (2/4) Epoch 16, batch 4650, loss[loss=0.2284, simple_loss=0.3169, pruned_loss=0.06994, over 8320.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2971, pruned_loss=0.06908, over 1617634.52 frames. ], batch size: 25, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:28:31,587 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125898.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:29:06,106 INFO [train.py:901] (2/4) Epoch 16, batch 4700, loss[loss=0.2127, simple_loss=0.2827, pruned_loss=0.07137, over 7529.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2962, pruned_loss=0.06897, over 1611418.45 frames. ], batch size: 18, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:29:18,700 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 18:29:18,975 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:29:20,236 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.580e+02 3.138e+02 4.127e+02 1.212e+03, threshold=6.277e+02, percent-clipped=5.0 +2023-02-06 18:29:39,829 INFO [train.py:901] (2/4) Epoch 16, batch 4750, loss[loss=0.1861, simple_loss=0.2635, pruned_loss=0.05434, over 7919.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2977, pruned_loss=0.06956, over 1615463.51 frames. ], batch size: 20, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:29:55,981 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:29:58,748 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7831, 1.7694, 2.4001, 1.6860, 1.3191, 2.3202, 0.3324, 1.5479], + device='cuda:2'), covar=tensor([0.2040, 0.1428, 0.0356, 0.1713, 0.3197, 0.0438, 0.2675, 0.1814], + device='cuda:2'), in_proj_covar=tensor([0.0175, 0.0181, 0.0114, 0.0214, 0.0260, 0.0118, 0.0164, 0.0177], + 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') +2023-02-06 18:30:11,196 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 18:30:13,729 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 18:30:16,332 INFO [train.py:901] (2/4) Epoch 16, batch 4800, loss[loss=0.2484, simple_loss=0.3235, pruned_loss=0.08661, over 8813.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2971, pruned_loss=0.06898, over 1618089.15 frames. ], batch size: 32, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:30:31,319 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.301e+02 2.788e+02 3.330e+02 6.705e+02, threshold=5.575e+02, percent-clipped=2.0 +2023-02-06 18:30:37,043 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7186, 2.0509, 1.6068, 2.6651, 1.2899, 1.3094, 1.7116, 2.2439], + device='cuda:2'), covar=tensor([0.0968, 0.0940, 0.1192, 0.0465, 0.1107, 0.1667, 0.1042, 0.0746], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0204, 0.0250, 0.0214, 0.0211, 0.0249, 0.0257, 0.0213], + 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') +2023-02-06 18:30:40,398 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:45,001 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:46,459 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126089.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:51,047 INFO [train.py:901] (2/4) Epoch 16, batch 4850, loss[loss=0.2335, simple_loss=0.3066, pruned_loss=0.08019, over 8505.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2972, pruned_loss=0.06926, over 1619255.76 frames. ], batch size: 28, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:30:59,316 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6937, 2.0208, 2.1615, 1.3819, 2.2427, 1.4879, 0.6336, 1.9475], + device='cuda:2'), covar=tensor([0.0499, 0.0278, 0.0211, 0.0510, 0.0317, 0.0806, 0.0722, 0.0237], + device='cuda:2'), in_proj_covar=tensor([0.0427, 0.0362, 0.0314, 0.0421, 0.0353, 0.0511, 0.0371, 0.0390], + device='cuda:2'), out_proj_covar=tensor([1.1717e-04, 9.6436e-05, 8.3497e-05, 1.1313e-04, 9.5198e-05, 1.4780e-04, + 1.0152e-04, 1.0561e-04], device='cuda:2') +2023-02-06 18:30:59,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:01,797 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 18:31:03,308 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126114.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:19,049 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:26,988 INFO [train.py:901] (2/4) Epoch 16, batch 4900, loss[loss=0.2085, simple_loss=0.3072, pruned_loss=0.05493, over 8024.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2964, pruned_loss=0.06855, over 1620328.87 frames. ], batch size: 22, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:31:32,562 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:41,755 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.384e+02 3.140e+02 3.836e+02 7.587e+02, threshold=6.281e+02, percent-clipped=5.0 +2023-02-06 18:31:50,105 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126179.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:32:01,581 INFO [train.py:901] (2/4) Epoch 16, batch 4950, loss[loss=0.2297, simple_loss=0.3084, pruned_loss=0.0755, over 8561.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2963, pruned_loss=0.06902, over 1615426.55 frames. ], batch size: 31, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:32:06,040 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126202.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:32:35,842 INFO [train.py:901] (2/4) Epoch 16, batch 5000, loss[loss=0.2129, simple_loss=0.2979, pruned_loss=0.06393, over 8357.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2962, pruned_loss=0.06908, over 1612590.59 frames. ], batch size: 24, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:32:37,271 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5960, 1.6912, 4.8206, 1.8334, 4.2495, 3.9778, 4.2982, 4.1904], + device='cuda:2'), covar=tensor([0.0586, 0.4059, 0.0475, 0.3624, 0.0999, 0.0947, 0.0578, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0564, 0.0622, 0.0647, 0.0593, 0.0669, 0.0576, 0.0565, 0.0633], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:32:50,295 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.421e+02 2.802e+02 3.540e+02 7.456e+02, threshold=5.603e+02, percent-clipped=2.0 +2023-02-06 18:33:10,450 INFO [train.py:901] (2/4) Epoch 16, batch 5050, loss[loss=0.2253, simple_loss=0.3161, pruned_loss=0.06722, over 8371.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2969, pruned_loss=0.0694, over 1611624.60 frames. ], batch size: 24, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:33:38,225 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:33:39,105 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 18:33:41,485 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 18:33:44,853 INFO [train.py:901] (2/4) Epoch 16, batch 5100, loss[loss=0.2494, simple_loss=0.3095, pruned_loss=0.09466, over 7785.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2962, pruned_loss=0.06944, over 1611426.12 frames. ], batch size: 19, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:33:55,180 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126360.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:33:55,977 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126361.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:34:01,121 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.574e+02 2.967e+02 3.773e+02 8.448e+02, threshold=5.934e+02, percent-clipped=7.0 +2023-02-06 18:34:20,689 INFO [train.py:901] (2/4) Epoch 16, batch 5150, loss[loss=0.2043, simple_loss=0.2796, pruned_loss=0.06448, over 7254.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2955, pruned_loss=0.06901, over 1607791.06 frames. ], batch size: 16, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:34:22,777 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:34:54,893 INFO [train.py:901] (2/4) Epoch 16, batch 5200, loss[loss=0.1666, simple_loss=0.2496, pruned_loss=0.04179, over 7702.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2958, pruned_loss=0.06906, over 1607957.12 frames. ], batch size: 18, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:35:03,390 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:10,023 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.450e+02 2.961e+02 4.009e+02 9.502e+02, threshold=5.923e+02, percent-clipped=8.0 +2023-02-06 18:35:10,199 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126468.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:15,127 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126475.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:21,844 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126483.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:30,964 INFO [train.py:901] (2/4) Epoch 16, batch 5250, loss[loss=0.2638, simple_loss=0.3179, pruned_loss=0.1049, over 7713.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.296, pruned_loss=0.06912, over 1604506.47 frames. ], batch size: 18, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:35:39,842 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 18:36:05,590 INFO [train.py:901] (2/4) Epoch 16, batch 5300, loss[loss=0.238, simple_loss=0.3216, pruned_loss=0.07716, over 8332.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2957, pruned_loss=0.06893, over 1606946.01 frames. ], batch size: 25, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:36:20,886 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.415e+02 2.951e+02 3.953e+02 1.148e+03, threshold=5.902e+02, percent-clipped=4.0 +2023-02-06 18:36:41,565 INFO [train.py:901] (2/4) Epoch 16, batch 5350, loss[loss=0.2141, simple_loss=0.2997, pruned_loss=0.0643, over 8099.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2964, pruned_loss=0.06907, over 1608621.77 frames. ], batch size: 23, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:37:16,921 INFO [train.py:901] (2/4) Epoch 16, batch 5400, loss[loss=0.1988, simple_loss=0.2982, pruned_loss=0.04975, over 8373.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2969, pruned_loss=0.06928, over 1607948.96 frames. ], batch size: 24, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:37:32,196 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.413e+02 2.875e+02 3.758e+02 9.843e+02, threshold=5.751e+02, percent-clipped=6.0 +2023-02-06 18:37:36,633 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5625, 2.3115, 3.3808, 2.6250, 3.1644, 2.4808, 2.1904, 1.7585], + device='cuda:2'), covar=tensor([0.4739, 0.4472, 0.1606, 0.3441, 0.2260, 0.2844, 0.1799, 0.5238], + device='cuda:2'), in_proj_covar=tensor([0.0907, 0.0914, 0.0755, 0.0891, 0.0956, 0.0840, 0.0716, 0.0794], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:37:51,432 INFO [train.py:901] (2/4) Epoch 16, batch 5450, loss[loss=0.2086, simple_loss=0.3065, pruned_loss=0.05532, over 8747.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2956, pruned_loss=0.06856, over 1603081.44 frames. ], batch size: 30, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:38:17,602 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126731.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:38:24,942 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126742.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:38:28,282 INFO [train.py:901] (2/4) Epoch 16, batch 5500, loss[loss=0.2679, simple_loss=0.3429, pruned_loss=0.09642, over 8505.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2968, pruned_loss=0.06929, over 1609196.14 frames. ], batch size: 28, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:38:28,990 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 18:38:35,356 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:38:44,227 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.451e+02 2.886e+02 3.496e+02 8.391e+02, threshold=5.772e+02, percent-clipped=4.0 +2023-02-06 18:39:02,248 INFO [train.py:901] (2/4) Epoch 16, batch 5550, loss[loss=0.2905, simple_loss=0.3639, pruned_loss=0.1085, over 8463.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2964, pruned_loss=0.06896, over 1608458.08 frames. ], batch size: 27, lr: 4.73e-03, grad_scale: 4.0 +2023-02-06 18:39:13,434 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:39:30,334 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126834.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:39:38,324 INFO [train.py:901] (2/4) Epoch 16, batch 5600, loss[loss=0.2243, simple_loss=0.3023, pruned_loss=0.07315, over 8333.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2964, pruned_loss=0.06882, over 1611841.61 frames. ], batch size: 25, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:39:45,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:39:54,360 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.374e+02 2.959e+02 4.088e+02 8.002e+02, threshold=5.917e+02, percent-clipped=4.0 +2023-02-06 18:40:09,605 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1877, 1.9371, 2.6355, 2.1796, 2.6208, 2.1906, 1.9214, 1.3718], + device='cuda:2'), covar=tensor([0.5005, 0.4416, 0.1568, 0.3404, 0.2261, 0.2581, 0.1764, 0.4779], + device='cuda:2'), in_proj_covar=tensor([0.0916, 0.0924, 0.0762, 0.0897, 0.0961, 0.0846, 0.0721, 0.0797], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:40:12,816 INFO [train.py:901] (2/4) Epoch 16, batch 5650, loss[loss=0.2091, simple_loss=0.2997, pruned_loss=0.05924, over 8470.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2962, pruned_loss=0.0687, over 1612822.88 frames. ], batch size: 29, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:40:33,382 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 18:40:33,506 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:40:48,558 INFO [train.py:901] (2/4) Epoch 16, batch 5700, loss[loss=0.2275, simple_loss=0.3145, pruned_loss=0.07026, over 8246.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2958, pruned_loss=0.06856, over 1614400.50 frames. ], batch size: 24, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:41:04,180 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.583e+02 3.205e+02 4.543e+02 7.570e+02, threshold=6.410e+02, percent-clipped=11.0 +2023-02-06 18:41:14,338 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3568, 1.9328, 2.7509, 2.2314, 2.6458, 2.2917, 2.0086, 1.4021], + device='cuda:2'), covar=tensor([0.4799, 0.4557, 0.1543, 0.3119, 0.2156, 0.2583, 0.1735, 0.4671], + device='cuda:2'), in_proj_covar=tensor([0.0914, 0.0926, 0.0765, 0.0897, 0.0963, 0.0847, 0.0723, 0.0800], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:41:17,568 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4906, 2.7186, 1.8336, 2.1427, 2.1539, 1.6487, 2.0346, 2.1165], + device='cuda:2'), covar=tensor([0.1632, 0.0397, 0.1135, 0.0733, 0.0713, 0.1404, 0.1020, 0.1106], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0234, 0.0327, 0.0304, 0.0301, 0.0336, 0.0346, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 18:41:22,817 INFO [train.py:901] (2/4) Epoch 16, batch 5750, loss[loss=0.2369, simple_loss=0.3213, pruned_loss=0.07624, over 8316.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2951, pruned_loss=0.06815, over 1613468.03 frames. ], batch size: 25, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:41:39,583 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 18:41:56,554 INFO [train.py:901] (2/4) Epoch 16, batch 5800, loss[loss=0.2709, simple_loss=0.339, pruned_loss=0.1015, over 8510.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2957, pruned_loss=0.06892, over 1613317.67 frames. ], batch size: 49, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:42:14,367 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.425e+02 2.951e+02 3.537e+02 6.549e+02, threshold=5.902e+02, percent-clipped=1.0 +2023-02-06 18:42:27,503 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 18:42:33,216 INFO [train.py:901] (2/4) Epoch 16, batch 5850, loss[loss=0.2822, simple_loss=0.3439, pruned_loss=0.1102, over 8651.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2969, pruned_loss=0.06955, over 1617642.58 frames. ], batch size: 49, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:42:45,190 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127113.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:02,065 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127138.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:07,130 INFO [train.py:901] (2/4) Epoch 16, batch 5900, loss[loss=0.2082, simple_loss=0.2988, pruned_loss=0.05878, over 8578.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2967, pruned_loss=0.06942, over 1614499.40 frames. ], batch size: 34, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:43:09,451 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4284, 2.1165, 2.8232, 2.3318, 2.7605, 2.3748, 2.0354, 1.5166], + device='cuda:2'), covar=tensor([0.4477, 0.4385, 0.1487, 0.3172, 0.2068, 0.2698, 0.1664, 0.4691], + device='cuda:2'), in_proj_covar=tensor([0.0911, 0.0921, 0.0764, 0.0893, 0.0959, 0.0842, 0.0720, 0.0796], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:43:12,285 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8720, 1.6144, 2.1005, 1.8059, 2.0323, 1.8904, 1.6210, 0.7463], + device='cuda:2'), covar=tensor([0.5249, 0.4689, 0.1616, 0.2799, 0.1973, 0.2778, 0.2081, 0.4441], + device='cuda:2'), in_proj_covar=tensor([0.0911, 0.0922, 0.0764, 0.0894, 0.0960, 0.0843, 0.0721, 0.0796], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:43:23,002 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.337e+02 2.920e+02 3.581e+02 1.365e+03, threshold=5.840e+02, percent-clipped=5.0 +2023-02-06 18:43:30,612 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:34,123 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:42,721 INFO [train.py:901] (2/4) Epoch 16, batch 5950, loss[loss=0.2246, simple_loss=0.3025, pruned_loss=0.07332, over 8257.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.297, pruned_loss=0.06922, over 1618689.47 frames. ], batch size: 24, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:43:51,308 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127208.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:44:17,704 INFO [train.py:901] (2/4) Epoch 16, batch 6000, loss[loss=0.1978, simple_loss=0.2712, pruned_loss=0.06221, over 7523.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.297, pruned_loss=0.06928, over 1613812.30 frames. ], batch size: 18, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:44:17,704 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 18:44:29,968 INFO [train.py:935] (2/4) Epoch 16, validation: loss=0.1793, simple_loss=0.2799, pruned_loss=0.03935, over 944034.00 frames. +2023-02-06 18:44:29,969 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 18:44:44,469 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127267.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:44:45,663 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.282e+02 2.976e+02 3.659e+02 8.304e+02, threshold=5.951e+02, percent-clipped=2.0 +2023-02-06 18:44:47,121 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3214, 1.8768, 2.9192, 1.2293, 2.0194, 1.7565, 1.5725, 1.9603], + device='cuda:2'), covar=tensor([0.2163, 0.2593, 0.0866, 0.4798, 0.2042, 0.3332, 0.2419, 0.2549], + device='cuda:2'), in_proj_covar=tensor([0.0502, 0.0554, 0.0538, 0.0609, 0.0622, 0.0562, 0.0498, 0.0616], + 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') +2023-02-06 18:45:01,834 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127293.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:45:03,656 INFO [train.py:901] (2/4) Epoch 16, batch 6050, loss[loss=0.2904, simple_loss=0.3651, pruned_loss=0.1079, over 8487.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2988, pruned_loss=0.0704, over 1617356.17 frames. ], batch size: 49, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:45:39,301 INFO [train.py:901] (2/4) Epoch 16, batch 6100, loss[loss=0.2365, simple_loss=0.3289, pruned_loss=0.072, over 8026.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2972, pruned_loss=0.06931, over 1611569.46 frames. ], batch size: 22, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:45:55,485 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.555e+02 2.947e+02 3.627e+02 8.036e+02, threshold=5.895e+02, percent-clipped=1.0 +2023-02-06 18:45:59,284 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 18:46:09,095 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 18:46:13,726 INFO [train.py:901] (2/4) Epoch 16, batch 6150, loss[loss=0.2065, simple_loss=0.2896, pruned_loss=0.06165, over 8249.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2965, pruned_loss=0.06918, over 1609469.16 frames. ], batch size: 24, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:46:28,955 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8230, 5.9028, 5.1445, 2.4190, 5.2041, 5.6266, 5.4133, 5.2227], + device='cuda:2'), covar=tensor([0.0456, 0.0365, 0.0928, 0.4302, 0.0679, 0.0602, 0.1072, 0.0535], + device='cuda:2'), in_proj_covar=tensor([0.0500, 0.0413, 0.0415, 0.0518, 0.0407, 0.0415, 0.0407, 0.0358], + 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') +2023-02-06 18:46:48,822 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127445.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:46:49,324 INFO [train.py:901] (2/4) Epoch 16, batch 6200, loss[loss=0.2441, simple_loss=0.3333, pruned_loss=0.07744, over 8327.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2972, pruned_loss=0.069, over 1614018.99 frames. ], batch size: 25, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:46:52,901 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7901, 1.4005, 1.5509, 1.1915, 0.8840, 1.3908, 1.5526, 1.4015], + device='cuda:2'), covar=tensor([0.0532, 0.1308, 0.1765, 0.1531, 0.0634, 0.1566, 0.0712, 0.0660], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0190, 0.0156, 0.0100, 0.0162, 0.0113, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 18:47:04,719 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.656e+02 3.320e+02 4.256e+02 8.643e+02, threshold=6.639e+02, percent-clipped=4.0 +2023-02-06 18:47:23,432 INFO [train.py:901] (2/4) Epoch 16, batch 6250, loss[loss=0.2629, simple_loss=0.343, pruned_loss=0.09143, over 8466.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2961, pruned_loss=0.06861, over 1615121.30 frames. ], batch size: 29, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:47:57,839 INFO [train.py:901] (2/4) Epoch 16, batch 6300, loss[loss=0.214, simple_loss=0.29, pruned_loss=0.06896, over 7821.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2973, pruned_loss=0.06904, over 1616802.70 frames. ], batch size: 20, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:47:58,580 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:47:59,971 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:48:08,819 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4211, 2.2172, 3.1528, 2.4190, 3.0311, 2.3973, 2.1696, 1.7885], + device='cuda:2'), covar=tensor([0.4678, 0.4746, 0.1749, 0.3448, 0.2360, 0.2623, 0.1698, 0.4826], + device='cuda:2'), in_proj_covar=tensor([0.0913, 0.0921, 0.0766, 0.0896, 0.0964, 0.0846, 0.0722, 0.0796], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:48:13,475 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1462, 1.3183, 1.2877, 0.6737, 1.2972, 1.0854, 0.1247, 1.2985], + device='cuda:2'), covar=tensor([0.0254, 0.0214, 0.0195, 0.0326, 0.0247, 0.0509, 0.0468, 0.0193], + device='cuda:2'), in_proj_covar=tensor([0.0420, 0.0362, 0.0311, 0.0416, 0.0349, 0.0507, 0.0367, 0.0389], + device='cuda:2'), out_proj_covar=tensor([1.1525e-04, 9.6553e-05, 8.2719e-05, 1.1171e-04, 9.3875e-05, 1.4659e-04, + 1.0041e-04, 1.0499e-04], device='cuda:2') +2023-02-06 18:48:14,532 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.653e+02 3.258e+02 3.936e+02 6.732e+02, threshold=6.516e+02, percent-clipped=2.0 +2023-02-06 18:48:17,991 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127574.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:48:32,713 INFO [train.py:901] (2/4) Epoch 16, batch 6350, loss[loss=0.2097, simple_loss=0.3001, pruned_loss=0.05964, over 8340.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2965, pruned_loss=0.06853, over 1614162.62 frames. ], batch size: 25, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:48:43,690 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:49:01,232 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0125, 1.6278, 1.2411, 1.5577, 1.3302, 1.0378, 1.2587, 1.3169], + device='cuda:2'), covar=tensor([0.1073, 0.0479, 0.1296, 0.0536, 0.0790, 0.1690, 0.1000, 0.0767], + device='cuda:2'), in_proj_covar=tensor([0.0347, 0.0233, 0.0323, 0.0300, 0.0297, 0.0330, 0.0340, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 18:49:03,417 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-06 18:49:07,014 INFO [train.py:901] (2/4) Epoch 16, batch 6400, loss[loss=0.198, simple_loss=0.279, pruned_loss=0.05851, over 7933.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2974, pruned_loss=0.06895, over 1614931.28 frames. ], batch size: 20, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:49:24,188 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.402e+02 3.034e+02 3.710e+02 8.847e+02, threshold=6.069e+02, percent-clipped=1.0 +2023-02-06 18:49:32,563 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:49:43,214 INFO [train.py:901] (2/4) Epoch 16, batch 6450, loss[loss=0.2743, simple_loss=0.348, pruned_loss=0.1003, over 8623.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2972, pruned_loss=0.06855, over 1618309.61 frames. ], batch size: 34, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:50:04,134 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:50:17,017 INFO [train.py:901] (2/4) Epoch 16, batch 6500, loss[loss=0.2733, simple_loss=0.3405, pruned_loss=0.103, over 8509.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2988, pruned_loss=0.06971, over 1619198.49 frames. ], batch size: 26, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:50:32,622 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.427e+02 3.150e+02 4.006e+02 1.604e+03, threshold=6.301e+02, percent-clipped=4.0 +2023-02-06 18:50:48,409 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127789.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:50:52,953 INFO [train.py:901] (2/4) Epoch 16, batch 6550, loss[loss=0.2284, simple_loss=0.3069, pruned_loss=0.07497, over 8599.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2994, pruned_loss=0.07014, over 1624708.51 frames. ], batch size: 31, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:50:53,788 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:51:17,239 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 18:51:27,450 INFO [train.py:901] (2/4) Epoch 16, batch 6600, loss[loss=0.2409, simple_loss=0.3065, pruned_loss=0.08768, over 7923.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2988, pruned_loss=0.06986, over 1622182.30 frames. ], batch size: 20, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:51:36,812 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 18:51:42,286 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:51:42,776 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.405e+02 2.899e+02 3.574e+02 1.034e+03, threshold=5.799e+02, percent-clipped=3.0 +2023-02-06 18:51:57,075 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:51:57,611 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127891.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:52:00,745 INFO [train.py:901] (2/4) Epoch 16, batch 6650, loss[loss=0.2047, simple_loss=0.2921, pruned_loss=0.05864, over 8467.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2978, pruned_loss=0.06945, over 1618172.35 frames. ], batch size: 25, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:52:07,581 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127904.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:52:36,161 INFO [train.py:901] (2/4) Epoch 16, batch 6700, loss[loss=0.222, simple_loss=0.2908, pruned_loss=0.07658, over 8088.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2977, pruned_loss=0.0692, over 1618877.29 frames. ], batch size: 21, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:52:41,772 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1865, 1.1007, 1.2847, 1.0556, 0.9073, 1.2972, 0.0556, 0.9104], + device='cuda:2'), covar=tensor([0.1984, 0.1513, 0.0529, 0.0988, 0.3235, 0.0621, 0.2650, 0.1448], + device='cuda:2'), in_proj_covar=tensor([0.0174, 0.0184, 0.0114, 0.0212, 0.0259, 0.0118, 0.0165, 0.0179], + 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') +2023-02-06 18:52:52,429 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.543e+02 2.898e+02 3.564e+02 8.195e+02, threshold=5.796e+02, percent-clipped=3.0 +2023-02-06 18:53:01,215 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:10,437 INFO [train.py:901] (2/4) Epoch 16, batch 6750, loss[loss=0.2677, simple_loss=0.3483, pruned_loss=0.09359, over 8369.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2971, pruned_loss=0.06901, over 1613305.59 frames. ], batch size: 24, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:53:15,087 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3724, 1.4360, 1.3696, 1.8080, 0.7605, 1.2011, 1.3079, 1.4933], + device='cuda:2'), covar=tensor([0.0859, 0.0753, 0.1017, 0.0499, 0.1096, 0.1351, 0.0783, 0.0691], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0201, 0.0247, 0.0211, 0.0208, 0.0247, 0.0252, 0.0211], + 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') +2023-02-06 18:53:15,265 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 18:53:18,488 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:19,176 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:32,877 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128024.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:47,637 INFO [train.py:901] (2/4) Epoch 16, batch 6800, loss[loss=0.2104, simple_loss=0.2973, pruned_loss=0.06169, over 8265.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.297, pruned_loss=0.069, over 1610856.87 frames. ], batch size: 24, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:53:51,039 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 18:54:04,017 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.604e+02 3.143e+02 4.008e+02 8.483e+02, threshold=6.287e+02, percent-clipped=3.0 +2023-02-06 18:54:22,242 INFO [train.py:901] (2/4) Epoch 16, batch 6850, loss[loss=0.2336, simple_loss=0.3183, pruned_loss=0.07444, over 8287.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.298, pruned_loss=0.06948, over 1615406.97 frames. ], batch size: 23, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:54:40,675 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 18:54:53,393 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128139.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:54:54,707 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128141.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:54:56,195 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8318, 2.6682, 3.7221, 1.8279, 1.6488, 3.6385, 0.6702, 2.1522], + device='cuda:2'), covar=tensor([0.1592, 0.1176, 0.0334, 0.2558, 0.3517, 0.0370, 0.2884, 0.1737], + device='cuda:2'), in_proj_covar=tensor([0.0175, 0.0183, 0.0114, 0.0212, 0.0258, 0.0118, 0.0164, 0.0178], + 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') +2023-02-06 18:54:58,083 INFO [train.py:901] (2/4) Epoch 16, batch 6900, loss[loss=0.1776, simple_loss=0.2577, pruned_loss=0.04879, over 7705.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2975, pruned_loss=0.06904, over 1615264.25 frames. ], batch size: 18, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:55:08,128 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128160.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:55:14,249 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.605e+02 3.172e+02 3.868e+02 9.306e+02, threshold=6.344e+02, percent-clipped=5.0 +2023-02-06 18:55:25,857 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128185.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:55:32,783 INFO [train.py:901] (2/4) Epoch 16, batch 6950, loss[loss=0.191, simple_loss=0.2753, pruned_loss=0.05339, over 7922.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2981, pruned_loss=0.06937, over 1611707.47 frames. ], batch size: 20, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:55:34,511 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 18:55:43,517 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:55:48,019 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 18:55:58,348 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128234.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:07,200 INFO [train.py:901] (2/4) Epoch 16, batch 7000, loss[loss=0.2338, simple_loss=0.3131, pruned_loss=0.07725, over 8729.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2969, pruned_loss=0.06903, over 1608558.15 frames. ], batch size: 34, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:56:15,592 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:19,638 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:23,999 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.690e+02 3.457e+02 5.056e+02 8.270e+02, threshold=6.915e+02, percent-clipped=6.0 +2023-02-06 18:56:36,186 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:42,574 INFO [train.py:901] (2/4) Epoch 16, batch 7050, loss[loss=0.2502, simple_loss=0.3219, pruned_loss=0.0893, over 8356.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2971, pruned_loss=0.06876, over 1608333.00 frames. ], batch size: 24, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:57:03,869 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:11,323 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:16,463 INFO [train.py:901] (2/4) Epoch 16, batch 7100, loss[loss=0.2461, simple_loss=0.3184, pruned_loss=0.08689, over 8513.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2964, pruned_loss=0.06851, over 1606006.71 frames. ], batch size: 26, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:57:18,647 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:33,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.456e+02 3.083e+02 3.766e+02 8.441e+02, threshold=6.166e+02, percent-clipped=2.0 +2023-02-06 18:57:52,038 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128395.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:52,510 INFO [train.py:901] (2/4) Epoch 16, batch 7150, loss[loss=0.185, simple_loss=0.2728, pruned_loss=0.0486, over 8078.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.296, pruned_loss=0.06777, over 1608336.94 frames. ], batch size: 21, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:58:08,475 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4858, 2.5875, 1.8652, 2.1163, 2.1382, 1.4433, 1.8990, 2.0902], + device='cuda:2'), covar=tensor([0.1475, 0.0338, 0.1098, 0.0677, 0.0700, 0.1639, 0.1038, 0.1036], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0233, 0.0325, 0.0301, 0.0302, 0.0330, 0.0342, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 18:58:09,864 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128420.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:58:17,227 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128431.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:58:27,263 INFO [train.py:901] (2/4) Epoch 16, batch 7200, loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09676, over 7091.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.297, pruned_loss=0.06829, over 1608031.59 frames. ], batch size: 71, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:58:42,535 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.378e+02 2.905e+02 3.370e+02 6.119e+02, threshold=5.810e+02, percent-clipped=0.0 +2023-02-06 18:59:02,791 INFO [train.py:901] (2/4) Epoch 16, batch 7250, loss[loss=0.1795, simple_loss=0.2524, pruned_loss=0.05332, over 7538.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2984, pruned_loss=0.06928, over 1613442.42 frames. ], batch size: 18, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:59:13,767 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128512.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:59:26,470 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6691, 2.3269, 4.7831, 2.9351, 4.3618, 4.1657, 4.5029, 4.4273], + device='cuda:2'), covar=tensor([0.0460, 0.3732, 0.0555, 0.2916, 0.0875, 0.0792, 0.0459, 0.0490], + device='cuda:2'), in_proj_covar=tensor([0.0561, 0.0623, 0.0643, 0.0591, 0.0670, 0.0573, 0.0564, 0.0632], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 18:59:31,351 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128537.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:59:32,908 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 18:59:37,227 INFO [train.py:901] (2/4) Epoch 16, batch 7300, loss[loss=0.1895, simple_loss=0.2842, pruned_loss=0.04739, over 8258.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2973, pruned_loss=0.06877, over 1614411.87 frames. ], batch size: 24, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:59:52,609 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.470e+02 2.980e+02 3.722e+02 1.252e+03, threshold=5.960e+02, percent-clipped=4.0 +2023-02-06 19:00:02,293 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:12,483 INFO [train.py:901] (2/4) Epoch 16, batch 7350, loss[loss=0.1879, simple_loss=0.262, pruned_loss=0.05684, over 7705.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2981, pruned_loss=0.06936, over 1614729.75 frames. ], batch size: 18, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:00:19,519 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:21,501 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:31,403 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 19:00:36,116 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:47,434 INFO [train.py:901] (2/4) Epoch 16, batch 7400, loss[loss=0.1666, simple_loss=0.2456, pruned_loss=0.04378, over 7702.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2978, pruned_loss=0.06962, over 1615189.43 frames. ], batch size: 18, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:00:49,532 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 19:01:00,966 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8788, 1.7080, 2.0768, 1.6743, 1.0099, 1.9641, 2.3179, 2.3976], + device='cuda:2'), covar=tensor([0.0451, 0.1198, 0.1611, 0.1360, 0.0578, 0.1321, 0.0601, 0.0530], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0156, 0.0100, 0.0161, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 19:01:02,831 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.610e+02 3.305e+02 3.788e+02 1.058e+03, threshold=6.610e+02, percent-clipped=7.0 +2023-02-06 19:01:11,762 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128682.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:01:21,078 INFO [train.py:901] (2/4) Epoch 16, batch 7450, loss[loss=0.2224, simple_loss=0.3052, pruned_loss=0.06981, over 8352.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2984, pruned_loss=0.06977, over 1617289.38 frames. ], batch size: 24, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:01:30,645 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 19:01:52,238 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6880, 5.7371, 5.0309, 2.2740, 5.0141, 5.4384, 5.3561, 5.1385], + device='cuda:2'), covar=tensor([0.0549, 0.0401, 0.0860, 0.4835, 0.0784, 0.0725, 0.0954, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0503, 0.0411, 0.0416, 0.0516, 0.0403, 0.0412, 0.0405, 0.0357], + 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') +2023-02-06 19:01:56,807 INFO [train.py:901] (2/4) Epoch 16, batch 7500, loss[loss=0.2115, simple_loss=0.2798, pruned_loss=0.07156, over 7224.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2974, pruned_loss=0.06919, over 1618493.50 frames. ], batch size: 16, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:02:12,299 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 19:02:13,138 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.417e+02 2.923e+02 3.614e+02 6.549e+02, threshold=5.847e+02, percent-clipped=0.0 +2023-02-06 19:02:17,186 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128775.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:02:31,138 INFO [train.py:901] (2/4) Epoch 16, batch 7550, loss[loss=0.1847, simple_loss=0.281, pruned_loss=0.04425, over 8478.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2967, pruned_loss=0.06862, over 1617991.73 frames. ], batch size: 25, lr: 4.69e-03, grad_scale: 16.0 +2023-02-06 19:02:32,027 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:02:33,327 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:02:49,735 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9301, 1.7276, 2.8895, 2.2502, 2.5887, 1.8621, 1.5609, 1.3161], + device='cuda:2'), covar=tensor([0.6491, 0.6144, 0.1709, 0.3315, 0.2535, 0.3975, 0.2864, 0.5228], + device='cuda:2'), in_proj_covar=tensor([0.0910, 0.0921, 0.0758, 0.0897, 0.0958, 0.0848, 0.0721, 0.0794], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:03:04,845 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5901, 1.5735, 2.2353, 1.5230, 1.1772, 2.1122, 0.4965, 1.3135], + device='cuda:2'), covar=tensor([0.2057, 0.1401, 0.0360, 0.1369, 0.3197, 0.0496, 0.2437, 0.1588], + device='cuda:2'), in_proj_covar=tensor([0.0177, 0.0185, 0.0115, 0.0215, 0.0261, 0.0120, 0.0165, 0.0179], + 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') +2023-02-06 19:03:07,387 INFO [train.py:901] (2/4) Epoch 16, batch 7600, loss[loss=0.2379, simple_loss=0.3148, pruned_loss=0.0805, over 8354.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2969, pruned_loss=0.06887, over 1616450.26 frames. ], batch size: 49, lr: 4.69e-03, grad_scale: 16.0 +2023-02-06 19:03:23,267 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.439e+02 3.123e+02 4.017e+02 8.994e+02, threshold=6.245e+02, percent-clipped=5.0 +2023-02-06 19:03:38,583 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:03:42,463 INFO [train.py:901] (2/4) Epoch 16, batch 7650, loss[loss=0.2296, simple_loss=0.3067, pruned_loss=0.07629, over 8191.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2963, pruned_loss=0.06816, over 1617860.00 frames. ], batch size: 23, lr: 4.69e-03, grad_scale: 16.0 +2023-02-06 19:03:50,523 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128908.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:04:08,892 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8976, 3.8563, 3.5190, 1.6058, 3.4784, 3.4419, 3.4578, 3.1332], + device='cuda:2'), covar=tensor([0.0909, 0.0601, 0.1239, 0.4549, 0.0978, 0.1113, 0.1381, 0.0909], + device='cuda:2'), in_proj_covar=tensor([0.0497, 0.0406, 0.0411, 0.0508, 0.0400, 0.0408, 0.0400, 0.0353], + 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') +2023-02-06 19:04:17,494 INFO [train.py:901] (2/4) Epoch 16, batch 7700, loss[loss=0.2108, simple_loss=0.3068, pruned_loss=0.05744, over 8501.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2943, pruned_loss=0.06755, over 1613433.07 frames. ], batch size: 28, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:04:34,543 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.361e+02 3.016e+02 3.880e+02 7.767e+02, threshold=6.032e+02, percent-clipped=3.0 +2023-02-06 19:04:42,141 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 19:04:49,724 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6209, 2.0215, 3.2687, 1.3681, 2.6281, 2.1259, 1.7264, 2.3439], + device='cuda:2'), covar=tensor([0.1847, 0.2393, 0.0883, 0.4476, 0.1568, 0.2998, 0.2064, 0.2277], + device='cuda:2'), in_proj_covar=tensor([0.0504, 0.0561, 0.0544, 0.0612, 0.0628, 0.0570, 0.0502, 0.0617], + 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') +2023-02-06 19:04:52,875 INFO [train.py:901] (2/4) Epoch 16, batch 7750, loss[loss=0.1922, simple_loss=0.2741, pruned_loss=0.05516, over 7805.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2941, pruned_loss=0.06712, over 1605487.02 frames. ], batch size: 20, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:04:56,371 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3515, 1.5275, 2.1197, 1.2177, 1.4707, 1.6410, 1.4197, 1.3650], + device='cuda:2'), covar=tensor([0.1910, 0.2505, 0.0913, 0.4348, 0.1932, 0.3273, 0.2246, 0.2170], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0562, 0.0545, 0.0613, 0.0629, 0.0571, 0.0503, 0.0618], + 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') +2023-02-06 19:05:21,211 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6043, 1.8473, 1.9584, 1.1939, 2.0739, 1.4724, 0.5800, 1.7505], + device='cuda:2'), covar=tensor([0.0485, 0.0285, 0.0233, 0.0501, 0.0354, 0.0774, 0.0741, 0.0266], + device='cuda:2'), in_proj_covar=tensor([0.0426, 0.0364, 0.0314, 0.0421, 0.0350, 0.0510, 0.0373, 0.0393], + device='cuda:2'), out_proj_covar=tensor([1.1664e-04, 9.6988e-05, 8.3333e-05, 1.1296e-04, 9.4072e-05, 1.4731e-04, + 1.0217e-04, 1.0608e-04], device='cuda:2') +2023-02-06 19:05:24,790 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 19:05:26,242 INFO [train.py:901] (2/4) Epoch 16, batch 7800, loss[loss=0.25, simple_loss=0.3369, pruned_loss=0.08153, over 8324.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2957, pruned_loss=0.06795, over 1610394.30 frames. ], batch size: 25, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:05:28,998 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5205, 2.7059, 1.8679, 2.3102, 2.2952, 1.5317, 1.9883, 2.2636], + device='cuda:2'), covar=tensor([0.1507, 0.0337, 0.1116, 0.0651, 0.0670, 0.1542, 0.1014, 0.0965], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0234, 0.0327, 0.0300, 0.0299, 0.0332, 0.0341, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:05:31,086 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129053.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:05:41,941 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.423e+02 2.949e+02 3.975e+02 9.373e+02, threshold=5.898e+02, percent-clipped=5.0 +2023-02-06 19:05:45,472 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3470, 1.3561, 2.2985, 1.1146, 2.0708, 2.4501, 2.5765, 1.9516], + device='cuda:2'), covar=tensor([0.1158, 0.1390, 0.0537, 0.2303, 0.0883, 0.0470, 0.0758, 0.0920], + device='cuda:2'), in_proj_covar=tensor([0.0281, 0.0309, 0.0273, 0.0301, 0.0293, 0.0251, 0.0385, 0.0297], + 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') +2023-02-06 19:05:48,216 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:06:00,670 INFO [train.py:901] (2/4) Epoch 16, batch 7850, loss[loss=0.2747, simple_loss=0.3398, pruned_loss=0.1048, over 7091.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.296, pruned_loss=0.06836, over 1606854.59 frames. ], batch size: 71, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:06:09,640 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1866, 4.1498, 3.7659, 1.9500, 3.6609, 3.7447, 3.8322, 3.4702], + device='cuda:2'), covar=tensor([0.0799, 0.0571, 0.1080, 0.4404, 0.0857, 0.0956, 0.1172, 0.0829], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0413, 0.0420, 0.0516, 0.0407, 0.0415, 0.0407, 0.0360], + 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') +2023-02-06 19:06:32,322 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129143.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:06:34,266 INFO [train.py:901] (2/4) Epoch 16, batch 7900, loss[loss=0.2199, simple_loss=0.2969, pruned_loss=0.07141, over 8572.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2949, pruned_loss=0.06754, over 1603015.41 frames. ], batch size: 34, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:06:34,503 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:06:51,032 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.289e+02 2.786e+02 3.620e+02 6.776e+02, threshold=5.572e+02, percent-clipped=2.0 +2023-02-06 19:06:51,877 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129171.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:07:08,434 INFO [train.py:901] (2/4) Epoch 16, batch 7950, loss[loss=0.2358, simple_loss=0.3149, pruned_loss=0.07831, over 8800.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2951, pruned_loss=0.06754, over 1604341.66 frames. ], batch size: 40, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:07:42,916 INFO [train.py:901] (2/4) Epoch 16, batch 8000, loss[loss=0.2289, simple_loss=0.3117, pruned_loss=0.07307, over 8766.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2957, pruned_loss=0.06742, over 1611966.79 frames. ], batch size: 30, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:07:47,178 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129252.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:07:48,169 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.39 vs. limit=5.0 +2023-02-06 19:07:51,183 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129258.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:07:59,068 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.511e+02 2.964e+02 3.601e+02 8.820e+02, threshold=5.927e+02, percent-clipped=6.0 +2023-02-06 19:08:16,586 INFO [train.py:901] (2/4) Epoch 16, batch 8050, loss[loss=0.1991, simple_loss=0.2743, pruned_loss=0.06192, over 7554.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2946, pruned_loss=0.06837, over 1595521.39 frames. ], batch size: 18, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:08:52,753 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 19:08:56,567 INFO [train.py:901] (2/4) Epoch 17, batch 0, loss[loss=0.2165, simple_loss=0.3116, pruned_loss=0.06071, over 8180.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.3116, pruned_loss=0.06071, over 8180.00 frames. ], batch size: 23, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:08:56,567 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 19:09:04,438 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5238, 1.7924, 2.6773, 1.3606, 1.9870, 1.7836, 1.6368, 1.9044], + device='cuda:2'), covar=tensor([0.1694, 0.2352, 0.0741, 0.4194, 0.1743, 0.3058, 0.2083, 0.2293], + device='cuda:2'), in_proj_covar=tensor([0.0504, 0.0561, 0.0543, 0.0612, 0.0631, 0.0570, 0.0501, 0.0620], + 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') +2023-02-06 19:09:07,561 INFO [train.py:935] (2/4) Epoch 17, validation: loss=0.1792, simple_loss=0.2794, pruned_loss=0.03944, over 944034.00 frames. +2023-02-06 19:09:07,562 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 19:09:19,459 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 19:09:21,129 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 19:09:29,750 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2338, 1.9814, 2.6159, 2.1498, 2.3910, 2.1583, 1.9561, 1.3298], + device='cuda:2'), covar=tensor([0.4450, 0.4228, 0.1530, 0.2919, 0.2149, 0.2705, 0.1728, 0.4719], + device='cuda:2'), in_proj_covar=tensor([0.0906, 0.0918, 0.0758, 0.0892, 0.0954, 0.0845, 0.0719, 0.0792], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:09:33,856 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129367.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:09:35,633 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.551e+02 3.127e+02 3.678e+02 8.568e+02, threshold=6.254e+02, percent-clipped=4.0 +2023-02-06 19:09:41,816 INFO [train.py:901] (2/4) Epoch 17, batch 50, loss[loss=0.2059, simple_loss=0.2848, pruned_loss=0.0635, over 8454.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.3006, pruned_loss=0.07051, over 367479.17 frames. ], batch size: 27, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:09:54,006 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 19:10:18,436 INFO [train.py:901] (2/4) Epoch 17, batch 100, loss[loss=0.2165, simple_loss=0.2958, pruned_loss=0.0686, over 8492.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2988, pruned_loss=0.06967, over 646979.90 frames. ], batch size: 28, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:10:18,444 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 19:10:19,953 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129431.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:10:23,384 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5297, 1.7543, 2.6111, 1.4234, 1.9028, 1.9357, 1.5470, 1.9446], + device='cuda:2'), covar=tensor([0.1837, 0.2518, 0.0888, 0.4182, 0.1832, 0.2963, 0.2270, 0.2138], + device='cuda:2'), in_proj_covar=tensor([0.0507, 0.0563, 0.0546, 0.0613, 0.0633, 0.0573, 0.0505, 0.0621], + 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') +2023-02-06 19:10:46,024 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.489e+02 3.062e+02 3.657e+02 7.822e+02, threshold=6.124e+02, percent-clipped=4.0 +2023-02-06 19:10:52,181 INFO [train.py:901] (2/4) Epoch 17, batch 150, loss[loss=0.2182, simple_loss=0.296, pruned_loss=0.07022, over 8347.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2995, pruned_loss=0.06975, over 863003.91 frames. ], batch size: 24, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:11:18,263 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129514.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:11:29,025 INFO [train.py:901] (2/4) Epoch 17, batch 200, loss[loss=0.228, simple_loss=0.2986, pruned_loss=0.07876, over 8498.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2987, pruned_loss=0.06876, over 1034679.06 frames. ], batch size: 28, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:11:36,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:11:57,077 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.455e+02 2.902e+02 3.926e+02 7.649e+02, threshold=5.804e+02, percent-clipped=5.0 +2023-02-06 19:12:03,433 INFO [train.py:901] (2/4) Epoch 17, batch 250, loss[loss=0.2239, simple_loss=0.3017, pruned_loss=0.07308, over 8311.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.298, pruned_loss=0.06851, over 1167431.71 frames. ], batch size: 49, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:12:09,653 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 19:12:11,839 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4871, 2.7357, 1.8101, 2.2300, 2.2070, 1.6109, 2.1074, 2.2681], + device='cuda:2'), covar=tensor([0.1521, 0.0340, 0.1139, 0.0634, 0.0745, 0.1410, 0.0959, 0.0866], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0236, 0.0330, 0.0303, 0.0301, 0.0337, 0.0344, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:12:18,387 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 19:12:33,734 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129623.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:12:38,236 INFO [train.py:901] (2/4) Epoch 17, batch 300, loss[loss=0.2228, simple_loss=0.3046, pruned_loss=0.07045, over 8518.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.298, pruned_loss=0.06827, over 1269824.61 frames. ], batch size: 31, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:12:39,078 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:12:53,545 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129648.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:13:08,148 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.453e+02 3.064e+02 3.747e+02 1.027e+03, threshold=6.129e+02, percent-clipped=5.0 +2023-02-06 19:13:14,344 INFO [train.py:901] (2/4) Epoch 17, batch 350, loss[loss=0.2268, simple_loss=0.2994, pruned_loss=0.07705, over 8203.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2965, pruned_loss=0.06826, over 1345144.53 frames. ], batch size: 23, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:13:47,840 INFO [train.py:901] (2/4) Epoch 17, batch 400, loss[loss=0.2439, simple_loss=0.3152, pruned_loss=0.0863, over 8285.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2965, pruned_loss=0.06822, over 1405011.40 frames. ], batch size: 23, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:14:18,001 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.355e+02 2.898e+02 3.830e+02 8.224e+02, threshold=5.797e+02, percent-clipped=7.0 +2023-02-06 19:14:21,454 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129775.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:14:24,071 INFO [train.py:901] (2/4) Epoch 17, batch 450, loss[loss=0.1814, simple_loss=0.2616, pruned_loss=0.05055, over 8086.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2974, pruned_loss=0.0689, over 1453156.63 frames. ], batch size: 21, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:14:34,957 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3020, 2.0130, 2.8393, 2.3303, 2.6695, 2.2548, 2.0244, 1.4054], + device='cuda:2'), covar=tensor([0.4505, 0.4366, 0.1450, 0.3196, 0.2250, 0.2774, 0.1884, 0.4849], + device='cuda:2'), in_proj_covar=tensor([0.0911, 0.0922, 0.0763, 0.0897, 0.0956, 0.0848, 0.0721, 0.0796], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:14:44,505 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129809.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:14:58,027 INFO [train.py:901] (2/4) Epoch 17, batch 500, loss[loss=0.212, simple_loss=0.2971, pruned_loss=0.06349, over 8244.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2958, pruned_loss=0.06834, over 1488556.74 frames. ], batch size: 24, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:15:03,870 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 19:15:05,062 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8607, 1.8806, 2.3881, 1.5499, 1.2914, 2.3768, 0.3375, 1.4206], + device='cuda:2'), covar=tensor([0.2135, 0.1430, 0.0417, 0.1579, 0.3568, 0.0434, 0.2730, 0.1668], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0186, 0.0115, 0.0218, 0.0266, 0.0121, 0.0166, 0.0178], + 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') +2023-02-06 19:15:28,005 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.377e+02 2.910e+02 3.862e+02 1.132e+03, threshold=5.820e+02, percent-clipped=8.0 +2023-02-06 19:15:35,661 INFO [train.py:901] (2/4) Epoch 17, batch 550, loss[loss=0.2626, simple_loss=0.3218, pruned_loss=0.1016, over 7805.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2971, pruned_loss=0.06908, over 1518420.20 frames. ], batch size: 20, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:15:43,354 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129890.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:15:47,845 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 19:16:10,112 INFO [train.py:901] (2/4) Epoch 17, batch 600, loss[loss=0.1844, simple_loss=0.2577, pruned_loss=0.05553, over 7788.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2968, pruned_loss=0.069, over 1538713.00 frames. ], batch size: 19, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:16:19,712 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 19:16:38,512 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 2.576e+02 2.936e+02 3.639e+02 7.352e+02, threshold=5.872e+02, percent-clipped=2.0 +2023-02-06 19:16:41,365 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:16:44,753 INFO [train.py:901] (2/4) Epoch 17, batch 650, loss[loss=0.2431, simple_loss=0.3152, pruned_loss=0.08552, over 8024.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2955, pruned_loss=0.06816, over 1551801.47 frames. ], batch size: 22, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:17:05,620 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1094, 2.1593, 1.6092, 1.8750, 1.7330, 1.3720, 1.5782, 1.7126], + device='cuda:2'), covar=tensor([0.1353, 0.0343, 0.1171, 0.0496, 0.0715, 0.1449, 0.0981, 0.0842], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0233, 0.0328, 0.0301, 0.0300, 0.0332, 0.0342, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:17:16,415 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:17:17,844 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:17:23,793 INFO [train.py:901] (2/4) Epoch 17, batch 700, loss[loss=0.2408, simple_loss=0.3041, pruned_loss=0.08873, over 7978.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2953, pruned_loss=0.06801, over 1564690.61 frames. ], batch size: 21, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:17:24,267 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-06 19:17:51,871 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.350e+02 2.811e+02 3.683e+02 1.098e+03, threshold=5.622e+02, percent-clipped=6.0 +2023-02-06 19:17:58,281 INFO [train.py:901] (2/4) Epoch 17, batch 750, loss[loss=0.1764, simple_loss=0.2497, pruned_loss=0.05152, over 7428.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2949, pruned_loss=0.06777, over 1572520.69 frames. ], batch size: 17, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:18:05,544 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130089.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:18:08,226 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 19:18:10,422 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8639, 1.4196, 1.6993, 1.2463, 0.9335, 1.3996, 1.7655, 1.3798], + device='cuda:2'), covar=tensor([0.0515, 0.1279, 0.1591, 0.1481, 0.0603, 0.1538, 0.0680, 0.0662], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0157, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 19:18:19,420 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 19:18:30,498 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7927, 4.7156, 4.2114, 2.0995, 4.2247, 4.3979, 4.3595, 4.1271], + device='cuda:2'), covar=tensor([0.0577, 0.0502, 0.1003, 0.4657, 0.0827, 0.0909, 0.1248, 0.0651], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0416, 0.0419, 0.0513, 0.0408, 0.0417, 0.0405, 0.0359], + 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') +2023-02-06 19:18:36,027 INFO [train.py:901] (2/4) Epoch 17, batch 800, loss[loss=0.2071, simple_loss=0.2901, pruned_loss=0.06199, over 8294.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2947, pruned_loss=0.06778, over 1582719.67 frames. ], batch size: 23, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:18:41,743 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1797, 1.0835, 1.2653, 1.1209, 0.9250, 1.2933, 0.0798, 1.0481], + device='cuda:2'), covar=tensor([0.1868, 0.1311, 0.0511, 0.0987, 0.3075, 0.0591, 0.2277, 0.1489], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0186, 0.0116, 0.0218, 0.0265, 0.0122, 0.0167, 0.0181], + 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') +2023-02-06 19:18:45,902 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1511, 1.5796, 4.3496, 1.6139, 3.8522, 3.6661, 4.0106, 3.8587], + device='cuda:2'), covar=tensor([0.0562, 0.3965, 0.0505, 0.3634, 0.1110, 0.0869, 0.0543, 0.0605], + device='cuda:2'), in_proj_covar=tensor([0.0562, 0.0621, 0.0643, 0.0589, 0.0671, 0.0575, 0.0566, 0.0626], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:18:48,062 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130146.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:18:52,763 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:19:04,223 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.363e+02 2.676e+02 3.408e+02 8.560e+02, threshold=5.353e+02, percent-clipped=3.0 +2023-02-06 19:19:05,144 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130171.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:19:10,475 INFO [train.py:901] (2/4) Epoch 17, batch 850, loss[loss=0.2071, simple_loss=0.2916, pruned_loss=0.06127, over 8438.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2961, pruned_loss=0.0681, over 1590834.25 frames. ], batch size: 29, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:19:47,569 INFO [train.py:901] (2/4) Epoch 17, batch 900, loss[loss=0.2322, simple_loss=0.3146, pruned_loss=0.07492, over 8250.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2962, pruned_loss=0.06786, over 1600212.91 frames. ], batch size: 24, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:20:12,542 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3979, 1.5000, 4.4813, 2.0039, 2.5520, 5.0946, 5.1065, 4.3492], + device='cuda:2'), covar=tensor([0.1033, 0.1804, 0.0258, 0.1931, 0.1086, 0.0142, 0.0293, 0.0551], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0309, 0.0275, 0.0303, 0.0295, 0.0254, 0.0389, 0.0299], + 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') +2023-02-06 19:20:15,373 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:20:16,505 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.489e+02 3.023e+02 3.878e+02 8.176e+02, threshold=6.045e+02, percent-clipped=7.0 +2023-02-06 19:20:22,802 INFO [train.py:901] (2/4) Epoch 17, batch 950, loss[loss=0.1859, simple_loss=0.2597, pruned_loss=0.056, over 7257.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2975, pruned_loss=0.06876, over 1604749.44 frames. ], batch size: 16, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:20:43,392 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 19:20:57,181 INFO [train.py:901] (2/4) Epoch 17, batch 1000, loss[loss=0.2244, simple_loss=0.307, pruned_loss=0.07096, over 8332.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2973, pruned_loss=0.0686, over 1605554.09 frames. ], batch size: 25, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:21:04,779 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3528, 4.3178, 3.8506, 2.1350, 3.8153, 3.9580, 3.9532, 3.6679], + device='cuda:2'), covar=tensor([0.0763, 0.0552, 0.1073, 0.4546, 0.0871, 0.1049, 0.1119, 0.0883], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0412, 0.0415, 0.0510, 0.0405, 0.0413, 0.0401, 0.0356], + 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') +2023-02-06 19:21:09,239 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130345.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:15,351 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8106, 5.9404, 5.0850, 2.3867, 5.2303, 5.6322, 5.5489, 5.3494], + device='cuda:2'), covar=tensor([0.0647, 0.0407, 0.0914, 0.4513, 0.0732, 0.0721, 0.1139, 0.0587], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0413, 0.0417, 0.0511, 0.0407, 0.0415, 0.0402, 0.0357], + 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') +2023-02-06 19:21:20,026 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 19:21:21,980 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130362.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:23,330 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:26,305 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6038, 1.7218, 2.0892, 1.4726, 1.2597, 2.1006, 0.2428, 1.3720], + device='cuda:2'), covar=tensor([0.2263, 0.1317, 0.0434, 0.1401, 0.3204, 0.0530, 0.2560, 0.1288], + device='cuda:2'), in_proj_covar=tensor([0.0177, 0.0184, 0.0115, 0.0216, 0.0263, 0.0120, 0.0167, 0.0178], + 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') +2023-02-06 19:21:27,490 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.680e+02 3.059e+02 3.924e+02 8.380e+02, threshold=6.118e+02, percent-clipped=2.0 +2023-02-06 19:21:27,757 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130370.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:33,156 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 19:21:33,840 INFO [train.py:901] (2/4) Epoch 17, batch 1050, loss[loss=0.2015, simple_loss=0.3022, pruned_loss=0.05042, over 8104.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2968, pruned_loss=0.06784, over 1609125.28 frames. ], batch size: 23, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:21:49,934 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:08,449 INFO [train.py:901] (2/4) Epoch 17, batch 1100, loss[loss=0.2027, simple_loss=0.2756, pruned_loss=0.06494, over 7246.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2957, pruned_loss=0.06757, over 1609953.44 frames. ], batch size: 16, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:22:23,041 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:27,218 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130456.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:35,459 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0026, 3.6174, 2.3013, 2.7764, 2.6859, 2.1588, 2.7130, 2.9451], + device='cuda:2'), covar=tensor([0.1650, 0.0314, 0.0993, 0.0732, 0.0766, 0.1239, 0.0998, 0.0944], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0231, 0.0325, 0.0298, 0.0297, 0.0328, 0.0339, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:22:38,667 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.545e+02 2.978e+02 3.676e+02 6.168e+02, threshold=5.956e+02, percent-clipped=1.0 +2023-02-06 19:22:44,137 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:45,337 INFO [train.py:901] (2/4) Epoch 17, batch 1150, loss[loss=0.2094, simple_loss=0.3004, pruned_loss=0.05922, over 8488.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2946, pruned_loss=0.06698, over 1613458.55 frames. ], batch size: 28, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:22:45,516 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:45,962 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 19:23:16,145 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:23:19,425 INFO [train.py:901] (2/4) Epoch 17, batch 1200, loss[loss=0.2263, simple_loss=0.3099, pruned_loss=0.07137, over 8460.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2949, pruned_loss=0.06726, over 1613898.25 frames. ], batch size: 27, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:23:26,513 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0761, 1.4413, 1.7269, 1.4035, 1.0027, 1.4727, 1.8880, 1.6823], + device='cuda:2'), covar=tensor([0.0527, 0.1192, 0.1657, 0.1360, 0.0609, 0.1477, 0.0652, 0.0589], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0191, 0.0157, 0.0100, 0.0163, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 19:23:33,401 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:23:45,147 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130566.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:23:47,782 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.314e+02 2.862e+02 3.617e+02 1.013e+03, threshold=5.724e+02, percent-clipped=2.0 +2023-02-06 19:23:53,885 INFO [train.py:901] (2/4) Epoch 17, batch 1250, loss[loss=0.2171, simple_loss=0.2819, pruned_loss=0.07613, over 7535.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2943, pruned_loss=0.06727, over 1607607.08 frames. ], batch size: 18, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:23:57,478 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:24:16,790 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2023-02-06 19:24:30,849 INFO [train.py:901] (2/4) Epoch 17, batch 1300, loss[loss=0.2848, simple_loss=0.349, pruned_loss=0.1103, over 8425.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2948, pruned_loss=0.06768, over 1612541.39 frames. ], batch size: 49, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:24:59,337 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.380e+02 3.126e+02 3.675e+02 7.509e+02, threshold=6.253e+02, percent-clipped=2.0 +2023-02-06 19:25:05,686 INFO [train.py:901] (2/4) Epoch 17, batch 1350, loss[loss=0.2133, simple_loss=0.2957, pruned_loss=0.06541, over 8488.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2944, pruned_loss=0.06726, over 1615319.89 frames. ], batch size: 29, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:25:16,157 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6667, 1.9699, 3.4774, 1.4069, 2.5530, 2.2340, 1.7072, 2.5651], + device='cuda:2'), covar=tensor([0.1816, 0.2345, 0.0771, 0.4231, 0.1701, 0.2752, 0.2062, 0.2112], + device='cuda:2'), in_proj_covar=tensor([0.0509, 0.0564, 0.0545, 0.0613, 0.0635, 0.0571, 0.0506, 0.0619], + 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') +2023-02-06 19:25:29,182 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2744, 1.8774, 2.4513, 2.0321, 2.2822, 2.2666, 2.0167, 1.1389], + device='cuda:2'), covar=tensor([0.4738, 0.4287, 0.1664, 0.3255, 0.2290, 0.2571, 0.1804, 0.4693], + device='cuda:2'), in_proj_covar=tensor([0.0921, 0.0930, 0.0780, 0.0905, 0.0967, 0.0856, 0.0729, 0.0804], + 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') +2023-02-06 19:25:43,052 INFO [train.py:901] (2/4) Epoch 17, batch 1400, loss[loss=0.2359, simple_loss=0.3166, pruned_loss=0.07764, over 7926.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06723, over 1617961.89 frames. ], batch size: 20, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:25:46,010 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:25:47,361 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130735.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:25:54,825 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130746.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:03,100 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130758.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:04,418 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:05,097 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6107, 2.0709, 4.2282, 1.4324, 2.9580, 2.2974, 1.6932, 2.8012], + device='cuda:2'), covar=tensor([0.1960, 0.2683, 0.0905, 0.4525, 0.1781, 0.3093, 0.2244, 0.2494], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0569, 0.0550, 0.0618, 0.0639, 0.0576, 0.0510, 0.0624], + 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') +2023-02-06 19:26:11,026 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 2.607e+02 3.260e+02 4.191e+02 1.113e+03, threshold=6.520e+02, percent-clipped=3.0 +2023-02-06 19:26:17,373 INFO [train.py:901] (2/4) Epoch 17, batch 1450, loss[loss=0.2362, simple_loss=0.3028, pruned_loss=0.08485, over 8125.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2945, pruned_loss=0.06685, over 1622143.63 frames. ], batch size: 22, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:26:20,756 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 19:26:27,818 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:32,213 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130800.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:49,003 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0535, 1.5825, 1.3333, 1.5695, 1.4134, 1.1661, 1.2274, 1.3417], + device='cuda:2'), covar=tensor([0.1126, 0.0443, 0.1257, 0.0518, 0.0681, 0.1505, 0.0923, 0.0727], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0232, 0.0324, 0.0299, 0.0297, 0.0328, 0.0339, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:26:54,258 INFO [train.py:901] (2/4) Epoch 17, batch 1500, loss[loss=0.2521, simple_loss=0.3326, pruned_loss=0.08583, over 8448.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2956, pruned_loss=0.06745, over 1619566.67 frames. ], batch size: 27, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:27:17,137 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:27:22,935 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.370e+02 2.974e+02 3.638e+02 1.375e+03, threshold=5.949e+02, percent-clipped=1.0 +2023-02-06 19:27:29,127 INFO [train.py:901] (2/4) Epoch 17, batch 1550, loss[loss=0.2414, simple_loss=0.3293, pruned_loss=0.07671, over 8241.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2948, pruned_loss=0.06691, over 1620228.62 frames. ], batch size: 24, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:27:50,135 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:27:50,703 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130910.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:27:54,329 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130915.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:28:02,604 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:28:03,818 INFO [train.py:901] (2/4) Epoch 17, batch 1600, loss[loss=0.245, simple_loss=0.3224, pruned_loss=0.08375, over 8467.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2949, pruned_loss=0.06692, over 1620839.90 frames. ], batch size: 27, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:28:09,165 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 19:28:34,755 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.345e+02 2.992e+02 3.546e+02 8.486e+02, threshold=5.983e+02, percent-clipped=5.0 +2023-02-06 19:28:40,942 INFO [train.py:901] (2/4) Epoch 17, batch 1650, loss[loss=0.2169, simple_loss=0.2911, pruned_loss=0.07135, over 7807.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2944, pruned_loss=0.06681, over 1618689.91 frames. ], batch size: 20, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:29:10,874 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 19:29:13,496 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131025.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:29:16,141 INFO [train.py:901] (2/4) Epoch 17, batch 1700, loss[loss=0.2294, simple_loss=0.3159, pruned_loss=0.07143, over 8496.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2944, pruned_loss=0.06725, over 1616411.90 frames. ], batch size: 28, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:29:25,391 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:29:46,947 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.451e+02 3.155e+02 3.823e+02 7.811e+02, threshold=6.311e+02, percent-clipped=3.0 +2023-02-06 19:29:53,064 INFO [train.py:901] (2/4) Epoch 17, batch 1750, loss[loss=0.2346, simple_loss=0.3003, pruned_loss=0.08448, over 7433.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2934, pruned_loss=0.06712, over 1613159.90 frames. ], batch size: 17, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:30:19,622 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131117.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:30:23,362 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 +2023-02-06 19:30:27,911 INFO [train.py:901] (2/4) Epoch 17, batch 1800, loss[loss=0.2196, simple_loss=0.2966, pruned_loss=0.07135, over 7655.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2932, pruned_loss=0.06707, over 1612034.32 frames. ], batch size: 19, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:30:30,799 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5038, 2.0584, 3.3673, 1.3368, 2.4974, 2.0421, 1.6654, 2.5696], + device='cuda:2'), covar=tensor([0.2025, 0.2808, 0.0852, 0.4595, 0.1776, 0.3225, 0.2332, 0.2243], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0564, 0.0545, 0.0611, 0.0632, 0.0570, 0.0506, 0.0617], + 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') +2023-02-06 19:30:37,107 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:30:44,376 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5936, 1.8430, 2.8119, 1.4784, 1.8854, 2.0522, 1.6736, 1.9685], + device='cuda:2'), covar=tensor([0.1543, 0.2102, 0.0724, 0.3630, 0.1635, 0.2487, 0.1788, 0.1864], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0564, 0.0544, 0.0612, 0.0633, 0.0570, 0.0507, 0.0618], + 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') +2023-02-06 19:30:52,690 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:30:55,953 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.745e+02 3.356e+02 4.683e+02 1.105e+03, threshold=6.712e+02, percent-clipped=11.0 +2023-02-06 19:30:56,917 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131171.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:02,079 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.38 vs. limit=5.0 +2023-02-06 19:31:03,694 INFO [train.py:901] (2/4) Epoch 17, batch 1850, loss[loss=0.2336, simple_loss=0.3089, pruned_loss=0.07917, over 8286.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2929, pruned_loss=0.0676, over 1606444.19 frames. ], batch size: 23, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:31:12,488 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131190.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:13,835 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131192.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:16,712 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131196.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:39,965 INFO [train.py:901] (2/4) Epoch 17, batch 1900, loss[loss=0.2267, simple_loss=0.3005, pruned_loss=0.07642, over 8069.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.293, pruned_loss=0.0676, over 1603327.34 frames. ], batch size: 21, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:32:08,103 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.313e+02 2.955e+02 3.582e+02 5.685e+02, threshold=5.910e+02, percent-clipped=0.0 +2023-02-06 19:32:08,132 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 19:32:14,118 INFO [train.py:901] (2/4) Epoch 17, batch 1950, loss[loss=0.2282, simple_loss=0.3041, pruned_loss=0.07612, over 8589.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2946, pruned_loss=0.068, over 1610992.08 frames. ], batch size: 31, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:32:15,759 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131281.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:19,635 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 19:32:26,415 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 19:32:28,924 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131298.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:35,146 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131306.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:39,920 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 19:32:46,202 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6693, 2.1927, 4.0623, 1.4427, 2.8461, 2.2057, 1.6265, 2.7087], + device='cuda:2'), covar=tensor([0.1836, 0.2462, 0.0725, 0.4230, 0.1799, 0.3049, 0.2257, 0.2456], + device='cuda:2'), in_proj_covar=tensor([0.0510, 0.0568, 0.0547, 0.0617, 0.0638, 0.0574, 0.0511, 0.0623], + 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') +2023-02-06 19:32:47,522 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131323.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:51,198 INFO [train.py:901] (2/4) Epoch 17, batch 2000, loss[loss=0.2133, simple_loss=0.3105, pruned_loss=0.05803, over 8250.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2961, pruned_loss=0.06918, over 1614047.84 frames. ], batch size: 24, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:32:58,450 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.09 vs. limit=5.0 +2023-02-06 19:33:19,856 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.510e+02 3.128e+02 3.622e+02 6.098e+02, threshold=6.257e+02, percent-clipped=1.0 +2023-02-06 19:33:25,354 INFO [train.py:901] (2/4) Epoch 17, batch 2050, loss[loss=0.1959, simple_loss=0.2829, pruned_loss=0.05443, over 7931.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2964, pruned_loss=0.06868, over 1619109.65 frames. ], batch size: 20, lr: 4.51e-03, grad_scale: 8.0 +2023-02-06 19:33:41,634 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.19 vs. limit=5.0 +2023-02-06 19:34:00,670 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:34:01,969 INFO [train.py:901] (2/4) Epoch 17, batch 2100, loss[loss=0.2007, simple_loss=0.2782, pruned_loss=0.06159, over 8034.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2958, pruned_loss=0.06852, over 1615957.69 frames. ], batch size: 22, lr: 4.51e-03, grad_scale: 8.0 +2023-02-06 19:34:06,123 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:34:28,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5644, 1.5277, 1.8769, 1.4218, 1.1902, 1.8592, 0.7787, 1.4866], + device='cuda:2'), covar=tensor([0.1824, 0.1019, 0.0390, 0.1223, 0.2709, 0.0519, 0.2066, 0.1427], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0183, 0.0115, 0.0216, 0.0261, 0.0122, 0.0166, 0.0179], + 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') +2023-02-06 19:34:31,417 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.457e+02 2.884e+02 3.530e+02 8.686e+02, threshold=5.767e+02, percent-clipped=1.0 +2023-02-06 19:34:36,976 INFO [train.py:901] (2/4) Epoch 17, batch 2150, loss[loss=0.1994, simple_loss=0.2817, pruned_loss=0.0586, over 8145.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2965, pruned_loss=0.06909, over 1614583.63 frames. ], batch size: 22, lr: 4.51e-03, grad_scale: 8.0 +2023-02-06 19:34:58,678 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:35:12,358 INFO [train.py:901] (2/4) Epoch 17, batch 2200, loss[loss=0.2294, simple_loss=0.312, pruned_loss=0.07338, over 8759.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2968, pruned_loss=0.06852, over 1619781.64 frames. ], batch size: 40, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:35:17,212 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131536.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:35:43,520 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.550e+02 3.248e+02 4.465e+02 1.208e+03, threshold=6.496e+02, percent-clipped=6.0 +2023-02-06 19:35:45,151 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0254, 2.5085, 3.5838, 2.0761, 1.9516, 3.4237, 0.8253, 2.0793], + device='cuda:2'), covar=tensor([0.1564, 0.1476, 0.0336, 0.1958, 0.3309, 0.0703, 0.2670, 0.1792], + device='cuda:2'), in_proj_covar=tensor([0.0180, 0.0184, 0.0115, 0.0218, 0.0263, 0.0122, 0.0168, 0.0180], + 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') +2023-02-06 19:35:49,218 INFO [train.py:901] (2/4) Epoch 17, batch 2250, loss[loss=0.2002, simple_loss=0.2865, pruned_loss=0.05694, over 8681.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2963, pruned_loss=0.06799, over 1620768.58 frames. ], batch size: 39, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:36:23,879 INFO [train.py:901] (2/4) Epoch 17, batch 2300, loss[loss=0.2119, simple_loss=0.3037, pruned_loss=0.0601, over 8494.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2963, pruned_loss=0.06784, over 1620139.34 frames. ], batch size: 29, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:36:40,758 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:36:47,781 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7785, 4.7825, 4.2495, 2.2709, 4.1972, 4.4240, 4.2517, 4.1008], + device='cuda:2'), covar=tensor([0.0643, 0.0511, 0.1045, 0.4133, 0.0897, 0.0758, 0.1378, 0.0717], + device='cuda:2'), in_proj_covar=tensor([0.0500, 0.0411, 0.0415, 0.0511, 0.0403, 0.0411, 0.0398, 0.0358], + 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') +2023-02-06 19:36:55,870 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.553e+02 3.001e+02 3.824e+02 6.268e+02, threshold=6.003e+02, percent-clipped=0.0 +2023-02-06 19:37:01,525 INFO [train.py:901] (2/4) Epoch 17, batch 2350, loss[loss=0.2228, simple_loss=0.3016, pruned_loss=0.07205, over 8441.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2984, pruned_loss=0.06931, over 1621838.65 frames. ], batch size: 27, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:37:35,924 INFO [train.py:901] (2/4) Epoch 17, batch 2400, loss[loss=0.1932, simple_loss=0.2757, pruned_loss=0.05538, over 8137.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2981, pruned_loss=0.06929, over 1622072.84 frames. ], batch size: 22, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:38:06,374 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.467e+02 3.155e+02 3.892e+02 8.269e+02, threshold=6.310e+02, percent-clipped=4.0 +2023-02-06 19:38:06,494 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:38:12,216 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:38:12,839 INFO [train.py:901] (2/4) Epoch 17, batch 2450, loss[loss=0.2099, simple_loss=0.2838, pruned_loss=0.068, over 8196.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2971, pruned_loss=0.06896, over 1619483.76 frames. ], batch size: 23, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:38:47,885 INFO [train.py:901] (2/4) Epoch 17, batch 2500, loss[loss=0.2207, simple_loss=0.309, pruned_loss=0.0662, over 8292.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2971, pruned_loss=0.06867, over 1619994.59 frames. ], batch size: 23, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:39:05,481 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131854.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:17,124 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.481e+02 2.929e+02 3.320e+02 7.417e+02, threshold=5.858e+02, percent-clipped=2.0 +2023-02-06 19:39:20,114 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:22,845 INFO [train.py:901] (2/4) Epoch 17, batch 2550, loss[loss=0.1913, simple_loss=0.2751, pruned_loss=0.05375, over 8252.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2967, pruned_loss=0.06858, over 1619068.94 frames. ], batch size: 22, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:39:29,619 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:34,616 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131893.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:35,580 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.60 vs. limit=5.0 +2023-02-06 19:39:45,578 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:40:00,838 INFO [train.py:901] (2/4) Epoch 17, batch 2600, loss[loss=0.2253, simple_loss=0.2986, pruned_loss=0.07603, over 8084.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2962, pruned_loss=0.06834, over 1618344.59 frames. ], batch size: 21, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:40:03,026 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131932.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:40:05,735 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8223, 2.1424, 2.3221, 1.3352, 2.3957, 1.6695, 0.7403, 2.0326], + device='cuda:2'), covar=tensor([0.0510, 0.0268, 0.0180, 0.0481, 0.0277, 0.0683, 0.0721, 0.0259], + device='cuda:2'), in_proj_covar=tensor([0.0427, 0.0368, 0.0315, 0.0424, 0.0351, 0.0510, 0.0375, 0.0391], + device='cuda:2'), out_proj_covar=tensor([1.1662e-04, 9.8122e-05, 8.3504e-05, 1.1353e-04, 9.4333e-05, 1.4711e-04, + 1.0244e-04, 1.0512e-04], device='cuda:2') +2023-02-06 19:40:25,954 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1348, 1.5307, 1.2085, 2.3631, 1.1078, 1.1065, 1.6723, 1.6883], + device='cuda:2'), covar=tensor([0.1687, 0.1316, 0.2096, 0.0463, 0.1350, 0.2232, 0.0949, 0.1004], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0198, 0.0246, 0.0211, 0.0207, 0.0244, 0.0251, 0.0209], + 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') +2023-02-06 19:40:28,790 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:40:29,953 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.407e+02 2.887e+02 3.716e+02 6.826e+02, threshold=5.774e+02, percent-clipped=1.0 +2023-02-06 19:40:32,198 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8046, 1.6367, 1.8437, 1.6321, 0.8057, 1.5311, 2.0259, 1.9265], + device='cuda:2'), covar=tensor([0.0442, 0.1228, 0.1547, 0.1344, 0.0625, 0.1525, 0.0640, 0.0579], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0156, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 19:40:35,446 INFO [train.py:901] (2/4) Epoch 17, batch 2650, loss[loss=0.221, simple_loss=0.3047, pruned_loss=0.06871, over 8335.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2942, pruned_loss=0.06743, over 1614045.52 frames. ], batch size: 25, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:41:13,492 INFO [train.py:901] (2/4) Epoch 17, batch 2700, loss[loss=0.2253, simple_loss=0.2974, pruned_loss=0.07663, over 8321.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2945, pruned_loss=0.06769, over 1616215.93 frames. ], batch size: 26, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:41:17,451 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.82 vs. limit=5.0 +2023-02-06 19:41:27,379 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132049.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:41:42,494 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.450e+02 3.248e+02 4.102e+02 1.137e+03, threshold=6.496e+02, percent-clipped=12.0 +2023-02-06 19:41:43,387 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:41:48,254 INFO [train.py:901] (2/4) Epoch 17, batch 2750, loss[loss=0.2544, simple_loss=0.3146, pruned_loss=0.09715, over 7205.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2949, pruned_loss=0.06762, over 1616575.39 frames. ], batch size: 72, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:42:25,031 INFO [train.py:901] (2/4) Epoch 17, batch 2800, loss[loss=0.1877, simple_loss=0.2549, pruned_loss=0.0602, over 7542.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2952, pruned_loss=0.06853, over 1614065.83 frames. ], batch size: 18, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:42:35,386 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:42:40,211 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132149.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:42:50,606 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9741, 1.7122, 3.0973, 1.5167, 2.3228, 3.3746, 3.4604, 2.8838], + device='cuda:2'), covar=tensor([0.1082, 0.1532, 0.0363, 0.1935, 0.0924, 0.0255, 0.0574, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0309, 0.0275, 0.0303, 0.0294, 0.0253, 0.0388, 0.0299], + 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') +2023-02-06 19:42:52,683 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:42:55,265 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.217e+02 2.865e+02 3.623e+02 1.020e+03, threshold=5.730e+02, percent-clipped=3.0 +2023-02-06 19:42:57,334 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132174.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:43:00,626 INFO [train.py:901] (2/4) Epoch 17, batch 2850, loss[loss=0.2366, simple_loss=0.3168, pruned_loss=0.07821, over 8451.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2966, pruned_loss=0.06885, over 1618708.78 frames. ], batch size: 27, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:43:12,453 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9735, 1.6179, 1.3983, 1.5371, 1.3517, 1.2516, 1.2544, 1.3009], + device='cuda:2'), covar=tensor([0.1114, 0.0449, 0.1251, 0.0562, 0.0734, 0.1379, 0.0858, 0.0755], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0232, 0.0323, 0.0298, 0.0297, 0.0326, 0.0337, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:43:29,199 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:43:33,409 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132225.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:43:35,753 INFO [train.py:901] (2/4) Epoch 17, batch 2900, loss[loss=0.2317, simple_loss=0.2985, pruned_loss=0.08241, over 7781.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2981, pruned_loss=0.06973, over 1619520.64 frames. ], batch size: 19, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:43:37,428 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9093, 2.1167, 1.8287, 2.6257, 1.2398, 1.5898, 1.8506, 1.9605], + device='cuda:2'), covar=tensor([0.0755, 0.0787, 0.1025, 0.0408, 0.1150, 0.1374, 0.0860, 0.0892], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0198, 0.0246, 0.0210, 0.0209, 0.0244, 0.0252, 0.0209], + 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') +2023-02-06 19:43:52,933 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:44:08,372 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.407e+02 2.887e+02 3.454e+02 7.005e+02, threshold=5.774e+02, percent-clipped=2.0 +2023-02-06 19:44:09,829 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 19:44:13,738 INFO [train.py:901] (2/4) Epoch 17, batch 2950, loss[loss=0.1961, simple_loss=0.2669, pruned_loss=0.06265, over 7544.00 frames. ], tot_loss[loss=0.218, simple_loss=0.297, pruned_loss=0.06953, over 1615647.24 frames. ], batch size: 18, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:44:20,352 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.75 vs. limit=5.0 +2023-02-06 19:44:48,310 INFO [train.py:901] (2/4) Epoch 17, batch 3000, loss[loss=0.2445, simple_loss=0.3188, pruned_loss=0.08507, over 8189.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.298, pruned_loss=0.06941, over 1621541.23 frames. ], batch size: 23, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:44:48,310 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 19:45:00,594 INFO [train.py:935] (2/4) Epoch 17, validation: loss=0.1786, simple_loss=0.2786, pruned_loss=0.03928, over 944034.00 frames. +2023-02-06 19:45:00,595 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 19:45:04,454 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132334.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:45:31,446 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.492e+02 3.005e+02 3.786e+02 8.313e+02, threshold=6.010e+02, percent-clipped=11.0 +2023-02-06 19:45:37,096 INFO [train.py:901] (2/4) Epoch 17, batch 3050, loss[loss=0.2059, simple_loss=0.2884, pruned_loss=0.06167, over 8289.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2974, pruned_loss=0.06897, over 1619418.20 frames. ], batch size: 23, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:45:48,261 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:46:04,204 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132416.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:46:12,935 INFO [train.py:901] (2/4) Epoch 17, batch 3100, loss[loss=0.2172, simple_loss=0.2975, pruned_loss=0.06844, over 8239.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2984, pruned_loss=0.06912, over 1622608.29 frames. ], batch size: 22, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:46:13,064 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1849, 3.1029, 2.8447, 1.5846, 2.7547, 2.9655, 2.8661, 2.7694], + device='cuda:2'), covar=tensor([0.1364, 0.0930, 0.1496, 0.5592, 0.1288, 0.1278, 0.1824, 0.1223], + device='cuda:2'), in_proj_covar=tensor([0.0498, 0.0411, 0.0414, 0.0510, 0.0401, 0.0413, 0.0399, 0.0357], + 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') +2023-02-06 19:46:41,883 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.340e+02 2.843e+02 3.195e+02 7.960e+02, threshold=5.685e+02, percent-clipped=6.0 +2023-02-06 19:46:46,902 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1830, 2.1018, 1.6635, 1.9478, 1.7856, 1.4495, 1.6702, 1.7190], + device='cuda:2'), covar=tensor([0.1155, 0.0346, 0.1075, 0.0473, 0.0621, 0.1290, 0.0764, 0.0763], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0233, 0.0324, 0.0300, 0.0297, 0.0328, 0.0339, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:46:47,324 INFO [train.py:901] (2/4) Epoch 17, batch 3150, loss[loss=0.2265, simple_loss=0.2943, pruned_loss=0.07938, over 7974.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2975, pruned_loss=0.06904, over 1621022.04 frames. ], batch size: 21, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:46:59,436 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 19:47:09,744 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132508.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:47:24,999 INFO [train.py:901] (2/4) Epoch 17, batch 3200, loss[loss=0.26, simple_loss=0.3255, pruned_loss=0.09724, over 7210.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2976, pruned_loss=0.0689, over 1621298.99 frames. ], batch size: 73, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:47:26,593 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:47:28,227 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 19:47:54,176 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.495e+02 3.112e+02 3.824e+02 1.248e+03, threshold=6.223e+02, percent-clipped=6.0 +2023-02-06 19:47:59,505 INFO [train.py:901] (2/4) Epoch 17, batch 3250, loss[loss=0.1895, simple_loss=0.2708, pruned_loss=0.05414, over 7252.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2966, pruned_loss=0.06842, over 1620738.14 frames. ], batch size: 16, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:48:03,307 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-06 19:48:07,386 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132590.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:48:26,270 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132615.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:48:37,501 INFO [train.py:901] (2/4) Epoch 17, batch 3300, loss[loss=0.196, simple_loss=0.2939, pruned_loss=0.04907, over 8246.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2952, pruned_loss=0.06786, over 1618385.95 frames. ], batch size: 24, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:49:06,785 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 2.532e+02 2.971e+02 3.744e+02 7.972e+02, threshold=5.942e+02, percent-clipped=3.0 +2023-02-06 19:49:11,862 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132678.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:49:12,441 INFO [train.py:901] (2/4) Epoch 17, batch 3350, loss[loss=0.2142, simple_loss=0.2998, pruned_loss=0.06428, over 8342.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2956, pruned_loss=0.06829, over 1615365.07 frames. ], batch size: 49, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:49:49,261 INFO [train.py:901] (2/4) Epoch 17, batch 3400, loss[loss=0.2068, simple_loss=0.2786, pruned_loss=0.06753, over 7538.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2954, pruned_loss=0.06843, over 1605423.83 frames. ], batch size: 18, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:50:14,689 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:14,782 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:19,434 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.442e+02 2.969e+02 4.012e+02 9.663e+02, threshold=5.937e+02, percent-clipped=5.0 +2023-02-06 19:50:24,933 INFO [train.py:901] (2/4) Epoch 17, batch 3450, loss[loss=0.1622, simple_loss=0.2423, pruned_loss=0.04105, over 7549.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2952, pruned_loss=0.06882, over 1607035.86 frames. ], batch size: 18, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:50:30,886 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:32,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:47,675 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:59,800 INFO [train.py:901] (2/4) Epoch 17, batch 3500, loss[loss=0.1897, simple_loss=0.2704, pruned_loss=0.05449, over 7223.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2956, pruned_loss=0.06838, over 1608186.04 frames. ], batch size: 16, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:51:13,850 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 19:51:31,533 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.401e+02 3.009e+02 3.970e+02 8.620e+02, threshold=6.019e+02, percent-clipped=6.0 +2023-02-06 19:51:33,854 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.5049, 1.6404, 3.7619, 1.5950, 3.0687, 2.9050, 3.3979, 3.3150], + device='cuda:2'), covar=tensor([0.1382, 0.5926, 0.1333, 0.5085, 0.2380, 0.1898, 0.1114, 0.1297], + device='cuda:2'), in_proj_covar=tensor([0.0568, 0.0617, 0.0654, 0.0588, 0.0664, 0.0576, 0.0565, 0.0631], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:51:37,027 INFO [train.py:901] (2/4) Epoch 17, batch 3550, loss[loss=0.2687, simple_loss=0.3382, pruned_loss=0.09963, over 8645.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2959, pruned_loss=0.06844, over 1608920.18 frames. ], batch size: 34, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:52:11,138 INFO [train.py:901] (2/4) Epoch 17, batch 3600, loss[loss=0.2083, simple_loss=0.2825, pruned_loss=0.06706, over 7713.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2945, pruned_loss=0.06785, over 1604389.35 frames. ], batch size: 18, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:52:41,875 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.442e+02 2.775e+02 3.418e+02 6.006e+02, threshold=5.549e+02, percent-clipped=0.0 +2023-02-06 19:52:48,330 INFO [train.py:901] (2/4) Epoch 17, batch 3650, loss[loss=0.2038, simple_loss=0.2823, pruned_loss=0.06269, over 7658.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2954, pruned_loss=0.06815, over 1607533.45 frames. ], batch size: 19, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:52:55,313 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8325, 5.8950, 5.1253, 2.5865, 5.1243, 5.5481, 5.4653, 5.3726], + device='cuda:2'), covar=tensor([0.0470, 0.0410, 0.0936, 0.4161, 0.0749, 0.0740, 0.0982, 0.0535], + device='cuda:2'), in_proj_covar=tensor([0.0503, 0.0412, 0.0418, 0.0513, 0.0407, 0.0415, 0.0400, 0.0358], + 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') +2023-02-06 19:53:18,535 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:53:21,773 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 19:53:23,086 INFO [train.py:901] (2/4) Epoch 17, batch 3700, loss[loss=0.2213, simple_loss=0.3028, pruned_loss=0.06986, over 8109.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.295, pruned_loss=0.06811, over 1608211.33 frames. ], batch size: 23, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:53:53,576 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.419e+02 3.081e+02 4.194e+02 7.364e+02, threshold=6.162e+02, percent-clipped=6.0 +2023-02-06 19:53:59,127 INFO [train.py:901] (2/4) Epoch 17, batch 3750, loss[loss=0.2041, simple_loss=0.2885, pruned_loss=0.05982, over 8335.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2947, pruned_loss=0.06779, over 1612281.43 frames. ], batch size: 25, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:54:21,508 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133108.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:54:30,213 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1776, 2.1026, 1.5837, 1.8999, 1.7555, 1.3880, 1.6928, 1.6097], + device='cuda:2'), covar=tensor([0.1229, 0.0346, 0.1138, 0.0499, 0.0660, 0.1350, 0.0818, 0.0870], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0232, 0.0325, 0.0301, 0.0299, 0.0330, 0.0341, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 19:54:35,507 INFO [train.py:901] (2/4) Epoch 17, batch 3800, loss[loss=0.2041, simple_loss=0.2725, pruned_loss=0.06785, over 7802.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2956, pruned_loss=0.06822, over 1615319.94 frames. ], batch size: 20, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:54:41,302 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:55:04,560 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.594e+02 3.054e+02 3.718e+02 6.772e+02, threshold=6.108e+02, percent-clipped=5.0 +2023-02-06 19:55:09,947 INFO [train.py:901] (2/4) Epoch 17, batch 3850, loss[loss=0.2503, simple_loss=0.3137, pruned_loss=0.09346, over 8656.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.295, pruned_loss=0.06779, over 1612325.68 frames. ], batch size: 49, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:55:15,494 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7351, 1.5500, 3.9347, 1.4279, 3.4821, 3.2836, 3.5930, 3.4545], + device='cuda:2'), covar=tensor([0.0756, 0.4357, 0.0715, 0.4077, 0.1314, 0.1107, 0.0730, 0.0815], + device='cuda:2'), in_proj_covar=tensor([0.0577, 0.0626, 0.0666, 0.0596, 0.0674, 0.0586, 0.0572, 0.0640], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:55:31,132 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 19:55:42,999 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133223.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:55:46,893 INFO [train.py:901] (2/4) Epoch 17, batch 3900, loss[loss=0.2125, simple_loss=0.2867, pruned_loss=0.06918, over 8139.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2952, pruned_loss=0.06811, over 1611690.90 frames. ], batch size: 22, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:56:00,183 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5527, 1.4875, 4.7800, 1.8935, 4.2996, 4.0798, 4.4229, 4.2079], + device='cuda:2'), covar=tensor([0.0494, 0.4042, 0.0445, 0.3499, 0.0875, 0.0855, 0.0420, 0.0590], + device='cuda:2'), in_proj_covar=tensor([0.0570, 0.0617, 0.0659, 0.0589, 0.0665, 0.0579, 0.0566, 0.0634], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:56:14,633 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1652, 1.8885, 2.7123, 2.2384, 2.5872, 2.1090, 1.8589, 1.3366], + device='cuda:2'), covar=tensor([0.5144, 0.4790, 0.1609, 0.3375, 0.2368, 0.2965, 0.1873, 0.4987], + device='cuda:2'), in_proj_covar=tensor([0.0922, 0.0937, 0.0772, 0.0906, 0.0971, 0.0853, 0.0724, 0.0803], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 19:56:15,759 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.486e+02 2.968e+02 4.028e+02 1.073e+03, threshold=5.936e+02, percent-clipped=5.0 +2023-02-06 19:56:21,117 INFO [train.py:901] (2/4) Epoch 17, batch 3950, loss[loss=0.1896, simple_loss=0.2734, pruned_loss=0.05295, over 7965.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2955, pruned_loss=0.06847, over 1612065.06 frames. ], batch size: 21, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:56:56,943 INFO [train.py:901] (2/4) Epoch 17, batch 4000, loss[loss=0.2219, simple_loss=0.301, pruned_loss=0.07144, over 8497.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2945, pruned_loss=0.06796, over 1609291.56 frames. ], batch size: 26, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:57:27,415 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.588e+02 2.526e+02 3.333e+02 3.995e+02 7.649e+02, threshold=6.666e+02, percent-clipped=5.0 +2023-02-06 19:57:32,341 INFO [train.py:901] (2/4) Epoch 17, batch 4050, loss[loss=0.195, simple_loss=0.2844, pruned_loss=0.05281, over 8462.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2941, pruned_loss=0.06736, over 1612089.26 frames. ], batch size: 29, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:57:41,374 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133392.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:57:42,122 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:57:42,762 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133394.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:57:59,643 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:58:07,781 INFO [train.py:901] (2/4) Epoch 17, batch 4100, loss[loss=0.2218, simple_loss=0.3126, pruned_loss=0.06545, over 8634.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2942, pruned_loss=0.06726, over 1613333.69 frames. ], batch size: 39, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:58:40,085 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.513e+02 2.919e+02 3.658e+02 1.440e+03, threshold=5.839e+02, percent-clipped=2.0 +2023-02-06 19:58:45,039 INFO [train.py:901] (2/4) Epoch 17, batch 4150, loss[loss=0.2192, simple_loss=0.3054, pruned_loss=0.06653, over 8467.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2942, pruned_loss=0.06773, over 1611400.76 frames. ], batch size: 25, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:58:45,264 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:58:54,028 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133492.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:59:02,279 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:59:19,507 INFO [train.py:901] (2/4) Epoch 17, batch 4200, loss[loss=0.1968, simple_loss=0.2734, pruned_loss=0.06008, over 7929.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2943, pruned_loss=0.0678, over 1607110.07 frames. ], batch size: 20, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:59:32,479 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 19:59:37,348 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5982, 1.5616, 1.9216, 1.3751, 1.2270, 1.9432, 0.4343, 1.3414], + device='cuda:2'), covar=tensor([0.1924, 0.1246, 0.0449, 0.1183, 0.2897, 0.0440, 0.2276, 0.1323], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0184, 0.0116, 0.0218, 0.0264, 0.0123, 0.0168, 0.0179], + 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') +2023-02-06 19:59:51,075 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.565e+02 3.135e+02 3.827e+02 1.180e+03, threshold=6.269e+02, percent-clipped=6.0 +2023-02-06 19:59:56,768 INFO [train.py:901] (2/4) Epoch 17, batch 4250, loss[loss=0.2098, simple_loss=0.2943, pruned_loss=0.06268, over 7971.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2942, pruned_loss=0.06766, over 1604748.25 frames. ], batch size: 21, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:59:57,448 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 20:00:07,950 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 20:00:30,994 INFO [train.py:901] (2/4) Epoch 17, batch 4300, loss[loss=0.1723, simple_loss=0.2583, pruned_loss=0.04318, over 8361.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2938, pruned_loss=0.0676, over 1604499.73 frames. ], batch size: 24, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:00:53,053 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4080, 1.5285, 1.4501, 1.8450, 0.7156, 1.2906, 1.3139, 1.5313], + device='cuda:2'), covar=tensor([0.0854, 0.0802, 0.1004, 0.0494, 0.1133, 0.1484, 0.0784, 0.0739], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0199, 0.0248, 0.0211, 0.0208, 0.0246, 0.0254, 0.0212], + 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') +2023-02-06 20:00:55,152 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2561, 2.4652, 2.2465, 2.9178, 2.0686, 2.1352, 2.3123, 2.6610], + device='cuda:2'), covar=tensor([0.0675, 0.0735, 0.0703, 0.0509, 0.0869, 0.1030, 0.0716, 0.0598], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0199, 0.0248, 0.0211, 0.0208, 0.0246, 0.0254, 0.0212], + 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') +2023-02-06 20:01:00,693 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133670.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:01:01,924 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.551e+02 3.118e+02 3.976e+02 6.360e+02, threshold=6.236e+02, percent-clipped=1.0 +2023-02-06 20:01:06,898 INFO [train.py:901] (2/4) Epoch 17, batch 4350, loss[loss=0.1633, simple_loss=0.2459, pruned_loss=0.04033, over 7531.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2946, pruned_loss=0.06787, over 1605413.06 frames. ], batch size: 18, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:01:08,326 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5561, 4.4617, 4.0754, 2.2993, 4.0087, 4.0860, 4.2163, 3.9000], + device='cuda:2'), covar=tensor([0.0707, 0.0507, 0.0997, 0.4269, 0.0827, 0.1093, 0.1020, 0.0858], + device='cuda:2'), in_proj_covar=tensor([0.0502, 0.0411, 0.0421, 0.0515, 0.0407, 0.0413, 0.0400, 0.0359], + 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') +2023-02-06 20:01:17,795 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5443, 1.9198, 3.0164, 1.3715, 2.2425, 1.9746, 1.5859, 2.2814], + device='cuda:2'), covar=tensor([0.1855, 0.2543, 0.0733, 0.4495, 0.1794, 0.3015, 0.2266, 0.2133], + device='cuda:2'), in_proj_covar=tensor([0.0510, 0.0571, 0.0550, 0.0621, 0.0637, 0.0576, 0.0514, 0.0627], + 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') +2023-02-06 20:01:31,320 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 20:01:43,141 INFO [train.py:901] (2/4) Epoch 17, batch 4400, loss[loss=0.202, simple_loss=0.2627, pruned_loss=0.07069, over 7443.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2942, pruned_loss=0.06729, over 1611504.65 frames. ], batch size: 17, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:01:48,105 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133736.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 20:01:49,427 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:02:05,418 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5242, 1.9338, 3.1989, 1.3925, 2.4374, 1.9187, 1.5860, 2.3293], + device='cuda:2'), covar=tensor([0.1896, 0.2498, 0.0756, 0.4397, 0.1678, 0.3035, 0.2223, 0.2191], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0573, 0.0549, 0.0622, 0.0640, 0.0576, 0.0516, 0.0628], + 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') +2023-02-06 20:02:12,880 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.579e+02 3.148e+02 3.884e+02 8.584e+02, threshold=6.297e+02, percent-clipped=6.0 +2023-02-06 20:02:12,934 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 20:02:18,531 INFO [train.py:901] (2/4) Epoch 17, batch 4450, loss[loss=0.2323, simple_loss=0.3117, pruned_loss=0.07641, over 8290.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2949, pruned_loss=0.06708, over 1614700.21 frames. ], batch size: 23, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:02:55,021 INFO [train.py:901] (2/4) Epoch 17, batch 4500, loss[loss=0.2439, simple_loss=0.315, pruned_loss=0.0864, over 8618.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2958, pruned_loss=0.06794, over 1614735.77 frames. ], batch size: 34, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:03:00,079 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133836.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:03:10,422 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133851.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 20:03:10,900 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 20:03:11,735 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133853.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:03:24,313 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.330e+02 2.856e+02 3.592e+02 8.327e+02, threshold=5.711e+02, percent-clipped=1.0 +2023-02-06 20:03:29,175 INFO [train.py:901] (2/4) Epoch 17, batch 4550, loss[loss=0.2024, simple_loss=0.2868, pruned_loss=0.05901, over 8337.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2959, pruned_loss=0.06825, over 1613891.86 frames. ], batch size: 25, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:04:04,547 INFO [train.py:901] (2/4) Epoch 17, batch 4600, loss[loss=0.2514, simple_loss=0.3159, pruned_loss=0.09349, over 7811.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2955, pruned_loss=0.06815, over 1611268.08 frames. ], batch size: 20, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:04:07,611 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0637, 1.6266, 1.4199, 1.5922, 1.3664, 1.3023, 1.2807, 1.3398], + device='cuda:2'), covar=tensor([0.1146, 0.0478, 0.1208, 0.0523, 0.0737, 0.1427, 0.0907, 0.0827], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0233, 0.0325, 0.0301, 0.0297, 0.0332, 0.0341, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 20:04:18,981 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 20:04:21,359 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:04:35,429 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.386e+02 2.834e+02 3.772e+02 7.696e+02, threshold=5.668e+02, percent-clipped=3.0 +2023-02-06 20:04:40,243 INFO [train.py:901] (2/4) Epoch 17, batch 4650, loss[loss=0.2545, simple_loss=0.3326, pruned_loss=0.08819, over 8352.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2946, pruned_loss=0.06765, over 1613031.93 frames. ], batch size: 24, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:05:01,910 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 20:05:02,393 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0326, 1.7502, 3.3685, 1.4562, 2.3250, 3.6991, 3.7655, 3.1427], + device='cuda:2'), covar=tensor([0.1083, 0.1564, 0.0319, 0.2062, 0.0988, 0.0205, 0.0397, 0.0545], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0311, 0.0275, 0.0304, 0.0295, 0.0253, 0.0392, 0.0298], + 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') +2023-02-06 20:05:06,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134014.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:05:07,207 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:05:16,663 INFO [train.py:901] (2/4) Epoch 17, batch 4700, loss[loss=0.2727, simple_loss=0.3375, pruned_loss=0.104, over 8581.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2949, pruned_loss=0.06756, over 1612512.13 frames. ], batch size: 49, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:05:45,984 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 20:05:48,988 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.331e+02 2.674e+02 3.349e+02 6.559e+02, threshold=5.348e+02, percent-clipped=3.0 +2023-02-06 20:05:53,967 INFO [train.py:901] (2/4) Epoch 17, batch 4750, loss[loss=0.2242, simple_loss=0.3039, pruned_loss=0.07221, over 8660.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2958, pruned_loss=0.06776, over 1615126.01 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:06:13,291 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134107.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 20:06:14,638 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:06:17,894 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 20:06:20,569 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 20:06:28,174 INFO [train.py:901] (2/4) Epoch 17, batch 4800, loss[loss=0.1649, simple_loss=0.2466, pruned_loss=0.04164, over 7410.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2948, pruned_loss=0.06739, over 1613352.20 frames. ], batch size: 17, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:06:28,387 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:06:31,287 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134132.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 20:06:32,699 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:07:00,735 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.352e+02 2.869e+02 3.488e+02 8.440e+02, threshold=5.739e+02, percent-clipped=9.0 +2023-02-06 20:07:06,357 INFO [train.py:901] (2/4) Epoch 17, batch 4850, loss[loss=0.216, simple_loss=0.2938, pruned_loss=0.06916, over 7817.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2945, pruned_loss=0.06665, over 1619275.13 frames. ], batch size: 20, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:07:14,652 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 20:07:26,230 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134207.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:07:29,619 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5725, 1.5160, 1.8540, 1.4196, 1.2063, 1.8374, 0.2257, 1.2609], + device='cuda:2'), covar=tensor([0.1899, 0.1210, 0.0399, 0.0996, 0.2673, 0.0450, 0.2168, 0.1166], + device='cuda:2'), in_proj_covar=tensor([0.0179, 0.0185, 0.0117, 0.0219, 0.0264, 0.0123, 0.0167, 0.0182], + 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') +2023-02-06 20:07:41,126 INFO [train.py:901] (2/4) Epoch 17, batch 4900, loss[loss=0.252, simple_loss=0.3365, pruned_loss=0.0837, over 8469.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.295, pruned_loss=0.0676, over 1614739.54 frames. ], batch size: 25, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:07:43,500 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134232.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:08:13,115 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.657e+02 3.351e+02 4.707e+02 1.168e+03, threshold=6.701e+02, percent-clipped=12.0 +2023-02-06 20:08:17,820 INFO [train.py:901] (2/4) Epoch 17, batch 4950, loss[loss=0.1726, simple_loss=0.2593, pruned_loss=0.04292, over 7652.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2962, pruned_loss=0.0682, over 1607589.72 frames. ], batch size: 19, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:08:18,297 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 20:08:49,554 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:08:52,964 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:08:54,212 INFO [train.py:901] (2/4) Epoch 17, batch 5000, loss[loss=0.2181, simple_loss=0.28, pruned_loss=0.07813, over 7434.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2957, pruned_loss=0.06758, over 1609337.69 frames. ], batch size: 17, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:08:55,956 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 20:09:15,213 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:09:24,852 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.361e+02 2.656e+02 3.405e+02 6.362e+02, threshold=5.311e+02, percent-clipped=0.0 +2023-02-06 20:09:30,477 INFO [train.py:901] (2/4) Epoch 17, batch 5050, loss[loss=0.2461, simple_loss=0.3219, pruned_loss=0.08515, over 8509.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2966, pruned_loss=0.06848, over 1610136.74 frames. ], batch size: 48, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:09:35,097 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:09:37,963 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5437, 1.9862, 3.2543, 1.3983, 2.3566, 2.0295, 1.6903, 2.3470], + device='cuda:2'), covar=tensor([0.2024, 0.2602, 0.0884, 0.4384, 0.1866, 0.3012, 0.2133, 0.2336], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0566, 0.0543, 0.0614, 0.0634, 0.0572, 0.0507, 0.0620], + 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') +2023-02-06 20:09:54,070 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:09:58,687 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 20:10:04,495 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5835, 2.8019, 2.4685, 4.0907, 1.4862, 2.2174, 2.4381, 3.1748], + device='cuda:2'), covar=tensor([0.0700, 0.0797, 0.0781, 0.0208, 0.1230, 0.1232, 0.1037, 0.0639], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0200, 0.0249, 0.0211, 0.0211, 0.0247, 0.0255, 0.0211], + 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') +2023-02-06 20:10:07,172 INFO [train.py:901] (2/4) Epoch 17, batch 5100, loss[loss=0.2283, simple_loss=0.3059, pruned_loss=0.07534, over 8493.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2958, pruned_loss=0.06805, over 1607548.14 frames. ], batch size: 39, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:10:36,991 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.384e+02 2.769e+02 3.675e+02 1.185e+03, threshold=5.538e+02, percent-clipped=9.0 +2023-02-06 20:10:38,570 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:10:42,692 INFO [train.py:901] (2/4) Epoch 17, batch 5150, loss[loss=0.2002, simple_loss=0.2839, pruned_loss=0.05822, over 7959.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2961, pruned_loss=0.06811, over 1607465.91 frames. ], batch size: 21, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:10:44,804 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4682, 4.4104, 3.9725, 2.1210, 3.9463, 3.9334, 4.0334, 3.7930], + device='cuda:2'), covar=tensor([0.0682, 0.0509, 0.0995, 0.4295, 0.0775, 0.0789, 0.1159, 0.0739], + device='cuda:2'), in_proj_covar=tensor([0.0500, 0.0412, 0.0418, 0.0514, 0.0404, 0.0412, 0.0399, 0.0359], + 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') +2023-02-06 20:11:17,577 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4598, 4.3830, 3.9467, 2.3567, 3.8937, 4.0486, 4.0135, 3.7371], + device='cuda:2'), covar=tensor([0.0623, 0.0554, 0.1014, 0.4109, 0.0753, 0.0985, 0.1199, 0.0756], + device='cuda:2'), in_proj_covar=tensor([0.0498, 0.0410, 0.0416, 0.0513, 0.0402, 0.0409, 0.0398, 0.0358], + 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') +2023-02-06 20:11:20,277 INFO [train.py:901] (2/4) Epoch 17, batch 5200, loss[loss=0.2077, simple_loss=0.2713, pruned_loss=0.07202, over 7433.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2953, pruned_loss=0.06766, over 1606772.19 frames. ], batch size: 17, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:11:25,570 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 +2023-02-06 20:11:49,978 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.225e+02 2.783e+02 3.706e+02 1.482e+03, threshold=5.567e+02, percent-clipped=8.0 +2023-02-06 20:11:54,889 INFO [train.py:901] (2/4) Epoch 17, batch 5250, loss[loss=0.2107, simple_loss=0.2782, pruned_loss=0.07161, over 7663.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2953, pruned_loss=0.06777, over 1604785.23 frames. ], batch size: 19, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:11:57,602 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 20:12:17,327 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-06 20:12:21,443 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 20:12:31,032 INFO [train.py:901] (2/4) Epoch 17, batch 5300, loss[loss=0.2106, simple_loss=0.3051, pruned_loss=0.05803, over 8344.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2945, pruned_loss=0.0671, over 1602241.09 frames. ], batch size: 24, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:12:56,715 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 20:12:58,389 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:13:01,795 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134671.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:13:02,352 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.333e+02 2.884e+02 3.429e+02 1.143e+03, threshold=5.769e+02, percent-clipped=6.0 +2023-02-06 20:13:07,125 INFO [train.py:901] (2/4) Epoch 17, batch 5350, loss[loss=0.172, simple_loss=0.2589, pruned_loss=0.04253, over 7980.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2947, pruned_loss=0.06734, over 1606448.92 frames. ], batch size: 21, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:13:16,921 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8152, 2.0729, 1.8098, 2.6055, 1.1668, 1.4605, 1.8385, 2.0262], + device='cuda:2'), covar=tensor([0.0771, 0.0708, 0.0943, 0.0363, 0.1185, 0.1395, 0.0834, 0.0704], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0199, 0.0249, 0.0212, 0.0211, 0.0247, 0.0255, 0.0211], + 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') +2023-02-06 20:13:36,021 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-06 20:13:42,898 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6596, 2.2693, 4.1210, 1.4026, 2.7946, 2.2451, 1.6883, 2.7601], + device='cuda:2'), covar=tensor([0.1869, 0.2392, 0.0678, 0.4305, 0.1809, 0.2978, 0.2146, 0.2324], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0563, 0.0542, 0.0611, 0.0631, 0.0570, 0.0506, 0.0615], + 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') +2023-02-06 20:13:43,340 INFO [train.py:901] (2/4) Epoch 17, batch 5400, loss[loss=0.2059, simple_loss=0.2929, pruned_loss=0.05946, over 8530.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2944, pruned_loss=0.06756, over 1605370.60 frames. ], batch size: 28, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:13:44,323 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:01,973 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:10,291 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0514, 2.5517, 3.6022, 1.9621, 1.9326, 3.5853, 0.7371, 2.1503], + device='cuda:2'), covar=tensor([0.1513, 0.1451, 0.0240, 0.2087, 0.3109, 0.0503, 0.2671, 0.1670], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0185, 0.0116, 0.0219, 0.0263, 0.0123, 0.0167, 0.0181], + 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') +2023-02-06 20:14:14,300 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.337e+02 2.988e+02 3.635e+02 1.067e+03, threshold=5.976e+02, percent-clipped=7.0 +2023-02-06 20:14:17,748 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8932, 3.7775, 3.4384, 1.9282, 3.3906, 3.4360, 3.4885, 3.3468], + device='cuda:2'), covar=tensor([0.0910, 0.0686, 0.1248, 0.4841, 0.1029, 0.1140, 0.1440, 0.0981], + device='cuda:2'), in_proj_covar=tensor([0.0508, 0.0418, 0.0424, 0.0527, 0.0413, 0.0418, 0.0407, 0.0364], + 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') +2023-02-06 20:14:18,990 INFO [train.py:901] (2/4) Epoch 17, batch 5450, loss[loss=0.2154, simple_loss=0.2891, pruned_loss=0.07087, over 7509.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2935, pruned_loss=0.06693, over 1600528.60 frames. ], batch size: 18, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:14:20,487 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134781.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:23,985 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134786.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:54,763 INFO [train.py:901] (2/4) Epoch 17, batch 5500, loss[loss=0.1951, simple_loss=0.2777, pruned_loss=0.05622, over 8361.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.292, pruned_loss=0.06619, over 1601585.06 frames. ], batch size: 24, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:14:55,399 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 20:15:25,532 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.381e+02 2.895e+02 3.783e+02 8.489e+02, threshold=5.790e+02, percent-clipped=3.0 +2023-02-06 20:15:31,381 INFO [train.py:901] (2/4) Epoch 17, batch 5550, loss[loss=0.235, simple_loss=0.3099, pruned_loss=0.0801, over 8026.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2925, pruned_loss=0.06663, over 1603208.25 frames. ], batch size: 22, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:15:32,208 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134880.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:16:06,820 INFO [train.py:901] (2/4) Epoch 17, batch 5600, loss[loss=0.2108, simple_loss=0.2937, pruned_loss=0.06394, over 8520.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2926, pruned_loss=0.06648, over 1605183.39 frames. ], batch size: 26, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:16:34,848 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1969, 1.9345, 2.5724, 2.1437, 2.5664, 2.1876, 1.8485, 1.2275], + device='cuda:2'), covar=tensor([0.5072, 0.4325, 0.1761, 0.3193, 0.2010, 0.2637, 0.1839, 0.5043], + device='cuda:2'), in_proj_covar=tensor([0.0914, 0.0930, 0.0774, 0.0902, 0.0972, 0.0849, 0.0721, 0.0800], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 20:16:38,758 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.724e+02 3.292e+02 4.135e+02 9.276e+02, threshold=6.584e+02, percent-clipped=7.0 +2023-02-06 20:16:42,898 INFO [train.py:901] (2/4) Epoch 17, batch 5650, loss[loss=0.1607, simple_loss=0.2377, pruned_loss=0.04187, over 7202.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2947, pruned_loss=0.06712, over 1610770.74 frames. ], batch size: 16, lr: 4.45e-03, grad_scale: 4.0 +2023-02-06 20:17:04,353 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 20:17:18,725 INFO [train.py:901] (2/4) Epoch 17, batch 5700, loss[loss=0.226, simple_loss=0.3096, pruned_loss=0.07118, over 8644.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06726, over 1613810.66 frames. ], batch size: 34, lr: 4.45e-03, grad_scale: 4.0 +2023-02-06 20:17:24,533 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135037.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:28,048 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:40,716 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.90 vs. limit=5.0 +2023-02-06 20:17:42,583 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:45,954 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135067.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:49,729 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.516e+02 3.214e+02 3.973e+02 1.283e+03, threshold=6.427e+02, percent-clipped=6.0 +2023-02-06 20:17:53,700 INFO [train.py:901] (2/4) Epoch 17, batch 5750, loss[loss=0.2357, simple_loss=0.3047, pruned_loss=0.08337, over 8488.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2949, pruned_loss=0.06783, over 1609016.39 frames. ], batch size: 26, lr: 4.45e-03, grad_scale: 4.0 +2023-02-06 20:18:02,757 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2169, 1.2384, 1.5172, 1.1705, 0.7122, 1.2750, 1.1661, 0.9852], + device='cuda:2'), covar=tensor([0.0592, 0.1342, 0.1692, 0.1463, 0.0598, 0.1573, 0.0720, 0.0712], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0156, 0.0099, 0.0162, 0.0113, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:18:11,551 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 20:18:21,899 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8354, 3.8784, 2.3427, 2.8764, 2.7472, 2.2024, 2.8593, 3.1195], + device='cuda:2'), covar=tensor([0.1778, 0.0266, 0.1065, 0.0741, 0.0804, 0.1305, 0.1124, 0.1041], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0231, 0.0326, 0.0301, 0.0296, 0.0330, 0.0341, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 20:18:30,181 INFO [train.py:901] (2/4) Epoch 17, batch 5800, loss[loss=0.2058, simple_loss=0.2921, pruned_loss=0.05974, over 5115.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2939, pruned_loss=0.06726, over 1604470.82 frames. ], batch size: 11, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:18:36,169 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 20:19:00,545 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.313e+02 2.882e+02 3.681e+02 6.576e+02, threshold=5.764e+02, percent-clipped=1.0 +2023-02-06 20:19:04,573 INFO [train.py:901] (2/4) Epoch 17, batch 5850, loss[loss=0.2665, simple_loss=0.3323, pruned_loss=0.1004, over 8108.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.294, pruned_loss=0.06716, over 1607300.12 frames. ], batch size: 23, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:19:12,152 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135189.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:19:37,628 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135224.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:19:41,090 INFO [train.py:901] (2/4) Epoch 17, batch 5900, loss[loss=0.211, simple_loss=0.2947, pruned_loss=0.0637, over 8198.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2936, pruned_loss=0.06689, over 1607143.79 frames. ], batch size: 23, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:20:03,977 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.15 vs. limit=5.0 +2023-02-06 20:20:12,436 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.357e+02 3.084e+02 3.660e+02 6.807e+02, threshold=6.167e+02, percent-clipped=2.0 +2023-02-06 20:20:16,622 INFO [train.py:901] (2/4) Epoch 17, batch 5950, loss[loss=0.3018, simple_loss=0.3567, pruned_loss=0.1234, over 6963.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2938, pruned_loss=0.06724, over 1606832.55 frames. ], batch size: 71, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:20:25,207 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7070, 2.1178, 2.3932, 1.2569, 2.4279, 1.6006, 0.7073, 1.9292], + device='cuda:2'), covar=tensor([0.0488, 0.0267, 0.0192, 0.0519, 0.0271, 0.0700, 0.0673, 0.0278], + device='cuda:2'), in_proj_covar=tensor([0.0435, 0.0376, 0.0320, 0.0431, 0.0358, 0.0516, 0.0380, 0.0397], + device='cuda:2'), out_proj_covar=tensor([1.1862e-04, 9.9983e-05, 8.4629e-05, 1.1535e-04, 9.5846e-05, 1.4872e-04, + 1.0343e-04, 1.0629e-04], device='cuda:2') +2023-02-06 20:20:43,831 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.81 vs. limit=5.0 +2023-02-06 20:20:52,281 INFO [train.py:901] (2/4) Epoch 17, batch 6000, loss[loss=0.2272, simple_loss=0.3171, pruned_loss=0.06866, over 8025.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.293, pruned_loss=0.06675, over 1603882.12 frames. ], batch size: 22, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:20:52,281 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 20:21:05,422 INFO [train.py:935] (2/4) Epoch 17, validation: loss=0.1774, simple_loss=0.2777, pruned_loss=0.03857, over 944034.00 frames. +2023-02-06 20:21:05,423 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 20:21:12,597 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135339.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:21:30,590 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5808, 1.2722, 2.7852, 1.2373, 2.1559, 2.9683, 3.1010, 2.3508], + device='cuda:2'), covar=tensor([0.1313, 0.1885, 0.0538, 0.2381, 0.1003, 0.0431, 0.0745, 0.0963], + device='cuda:2'), in_proj_covar=tensor([0.0282, 0.0311, 0.0276, 0.0302, 0.0294, 0.0253, 0.0389, 0.0300], + 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') +2023-02-06 20:21:36,687 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.411e+02 3.026e+02 3.580e+02 8.983e+02, threshold=6.051e+02, percent-clipped=2.0 +2023-02-06 20:21:40,885 INFO [train.py:901] (2/4) Epoch 17, batch 6050, loss[loss=0.2579, simple_loss=0.3087, pruned_loss=0.1035, over 7698.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2949, pruned_loss=0.06785, over 1608435.37 frames. ], batch size: 18, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:22:16,316 INFO [train.py:901] (2/4) Epoch 17, batch 6100, loss[loss=0.2438, simple_loss=0.3167, pruned_loss=0.08546, over 8571.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2944, pruned_loss=0.0672, over 1614722.65 frames. ], batch size: 49, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:22:47,554 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.459e+02 2.890e+02 3.783e+02 6.848e+02, threshold=5.780e+02, percent-clipped=3.0 +2023-02-06 20:22:49,609 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 20:22:52,250 INFO [train.py:901] (2/4) Epoch 17, batch 6150, loss[loss=0.2107, simple_loss=0.2839, pruned_loss=0.06875, over 8496.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2933, pruned_loss=0.06643, over 1613416.05 frames. ], batch size: 26, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:23:26,579 INFO [train.py:901] (2/4) Epoch 17, batch 6200, loss[loss=0.2333, simple_loss=0.3112, pruned_loss=0.07769, over 8345.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2936, pruned_loss=0.06663, over 1610662.35 frames. ], batch size: 26, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:23:29,344 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:23:57,601 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.470e+02 3.035e+02 3.550e+02 6.137e+02, threshold=6.070e+02, percent-clipped=1.0 +2023-02-06 20:24:01,728 INFO [train.py:901] (2/4) Epoch 17, batch 6250, loss[loss=0.1897, simple_loss=0.276, pruned_loss=0.05163, over 7923.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2928, pruned_loss=0.06634, over 1607698.47 frames. ], batch size: 20, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:24:13,155 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:24:24,743 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9797, 2.0230, 1.8087, 2.3377, 1.3993, 1.6218, 1.8377, 2.0873], + device='cuda:2'), covar=tensor([0.0626, 0.0704, 0.0871, 0.0535, 0.1035, 0.1119, 0.0803, 0.0668], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0197, 0.0246, 0.0209, 0.0207, 0.0245, 0.0254, 0.0210], + 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') +2023-02-06 20:24:29,109 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 20:24:30,231 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:24:36,950 INFO [train.py:901] (2/4) Epoch 17, batch 6300, loss[loss=0.1833, simple_loss=0.2709, pruned_loss=0.04789, over 8109.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2939, pruned_loss=0.06646, over 1613276.86 frames. ], batch size: 23, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:24:49,744 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:24:49,781 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:25:07,391 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.756e+02 2.700e+02 3.426e+02 4.477e+02 8.691e+02, threshold=6.853e+02, percent-clipped=8.0 +2023-02-06 20:25:11,449 INFO [train.py:901] (2/4) Epoch 17, batch 6350, loss[loss=0.1803, simple_loss=0.2692, pruned_loss=0.04576, over 7964.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2932, pruned_loss=0.06608, over 1612008.53 frames. ], batch size: 21, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:25:46,610 INFO [train.py:901] (2/4) Epoch 17, batch 6400, loss[loss=0.2148, simple_loss=0.2986, pruned_loss=0.06548, over 8288.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2944, pruned_loss=0.06673, over 1611709.83 frames. ], batch size: 23, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:26:16,683 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.216e+02 2.648e+02 3.143e+02 6.334e+02, threshold=5.295e+02, percent-clipped=0.0 +2023-02-06 20:26:20,488 INFO [train.py:901] (2/4) Epoch 17, batch 6450, loss[loss=0.1711, simple_loss=0.2456, pruned_loss=0.04836, over 7260.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2955, pruned_loss=0.06759, over 1609777.62 frames. ], batch size: 16, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:26:56,417 INFO [train.py:901] (2/4) Epoch 17, batch 6500, loss[loss=0.2217, simple_loss=0.2965, pruned_loss=0.07349, over 8240.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2953, pruned_loss=0.06766, over 1612447.76 frames. ], batch size: 22, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:27:27,342 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.443e+02 3.095e+02 4.367e+02 8.897e+02, threshold=6.190e+02, percent-clipped=12.0 +2023-02-06 20:27:31,527 INFO [train.py:901] (2/4) Epoch 17, batch 6550, loss[loss=0.1923, simple_loss=0.2809, pruned_loss=0.05183, over 7939.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.295, pruned_loss=0.0672, over 1610441.66 frames. ], batch size: 20, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:27:45,159 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-06 20:27:48,291 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135904.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:27:56,443 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 20:28:06,531 INFO [train.py:901] (2/4) Epoch 17, batch 6600, loss[loss=0.298, simple_loss=0.3655, pruned_loss=0.1152, over 7330.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2943, pruned_loss=0.06728, over 1609512.54 frames. ], batch size: 71, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:28:06,731 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:28:16,335 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 20:28:27,298 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8690, 1.6403, 1.9836, 1.6682, 0.8645, 1.6504, 2.1568, 2.1791], + device='cuda:2'), covar=tensor([0.0428, 0.1232, 0.1594, 0.1382, 0.0598, 0.1387, 0.0626, 0.0535], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0153, 0.0191, 0.0158, 0.0101, 0.0163, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:28:36,536 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.554e+02 2.943e+02 3.634e+02 1.271e+03, threshold=5.887e+02, percent-clipped=2.0 +2023-02-06 20:28:40,553 INFO [train.py:901] (2/4) Epoch 17, batch 6650, loss[loss=0.2274, simple_loss=0.3208, pruned_loss=0.06703, over 8450.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2942, pruned_loss=0.06728, over 1613358.96 frames. ], batch size: 27, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:28:49,898 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135992.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:28:56,429 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136000.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:29:04,531 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136012.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:29:07,755 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7965, 5.9278, 5.1293, 2.4547, 5.2135, 5.5959, 5.5111, 5.2746], + device='cuda:2'), covar=tensor([0.0519, 0.0350, 0.0951, 0.4333, 0.0717, 0.0705, 0.0916, 0.0589], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0418, 0.0421, 0.0522, 0.0411, 0.0416, 0.0404, 0.0364], + 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') +2023-02-06 20:29:16,459 INFO [train.py:901] (2/4) Epoch 17, batch 6700, loss[loss=0.1883, simple_loss=0.2649, pruned_loss=0.05583, over 7527.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2942, pruned_loss=0.06762, over 1612326.49 frames. ], batch size: 18, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:29:47,790 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.437e+02 3.090e+02 3.837e+02 8.578e+02, threshold=6.181e+02, percent-clipped=4.0 +2023-02-06 20:29:51,765 INFO [train.py:901] (2/4) Epoch 17, batch 6750, loss[loss=0.1769, simple_loss=0.2557, pruned_loss=0.04902, over 7432.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2948, pruned_loss=0.06733, over 1618393.10 frames. ], batch size: 17, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:30:11,741 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:30:13,064 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9873, 2.6226, 3.1933, 1.3535, 3.2631, 1.7348, 1.5786, 1.9707], + device='cuda:2'), covar=tensor([0.0789, 0.0325, 0.0190, 0.0721, 0.0429, 0.0738, 0.0793, 0.0475], + device='cuda:2'), in_proj_covar=tensor([0.0429, 0.0371, 0.0320, 0.0423, 0.0353, 0.0514, 0.0375, 0.0394], + device='cuda:2'), out_proj_covar=tensor([1.1705e-04, 9.8736e-05, 8.4776e-05, 1.1300e-04, 9.4536e-05, 1.4809e-04, + 1.0197e-04, 1.0554e-04], device='cuda:2') +2023-02-06 20:30:26,347 INFO [train.py:901] (2/4) Epoch 17, batch 6800, loss[loss=0.2028, simple_loss=0.274, pruned_loss=0.06576, over 7535.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2959, pruned_loss=0.06769, over 1619453.33 frames. ], batch size: 18, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:30:35,092 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 20:30:45,877 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4385, 1.5955, 2.1892, 1.3063, 1.5068, 1.6528, 1.4925, 1.4478], + device='cuda:2'), covar=tensor([0.1812, 0.2391, 0.0819, 0.4310, 0.1772, 0.3242, 0.2224, 0.2076], + device='cuda:2'), in_proj_covar=tensor([0.0510, 0.0571, 0.0546, 0.0617, 0.0636, 0.0577, 0.0511, 0.0623], + 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') +2023-02-06 20:30:48,378 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136160.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:30:57,852 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.691e+02 3.204e+02 3.737e+02 8.793e+02, threshold=6.409e+02, percent-clipped=5.0 +2023-02-06 20:31:01,826 INFO [train.py:901] (2/4) Epoch 17, batch 6850, loss[loss=0.214, simple_loss=0.2942, pruned_loss=0.06691, over 8142.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2974, pruned_loss=0.06848, over 1621210.86 frames. ], batch size: 22, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:31:22,725 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 20:31:24,399 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0878, 2.3084, 3.5909, 1.9298, 1.7772, 3.5815, 0.8747, 2.1748], + device='cuda:2'), covar=tensor([0.1489, 0.1391, 0.0237, 0.1809, 0.2957, 0.0314, 0.2453, 0.1537], + device='cuda:2'), in_proj_covar=tensor([0.0176, 0.0184, 0.0115, 0.0216, 0.0259, 0.0123, 0.0165, 0.0181], + 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') +2023-02-06 20:31:37,319 INFO [train.py:901] (2/4) Epoch 17, batch 6900, loss[loss=0.2423, simple_loss=0.3182, pruned_loss=0.08321, over 8247.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2977, pruned_loss=0.06873, over 1616457.30 frames. ], batch size: 22, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:31:40,196 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7482, 1.2961, 3.9880, 1.4532, 3.4755, 3.3451, 3.5808, 3.4778], + device='cuda:2'), covar=tensor([0.0747, 0.4507, 0.0630, 0.3773, 0.1394, 0.0947, 0.0719, 0.0809], + device='cuda:2'), in_proj_covar=tensor([0.0574, 0.0617, 0.0654, 0.0590, 0.0670, 0.0569, 0.0569, 0.0631], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 20:32:08,491 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.541e+02 3.415e+02 4.318e+02 7.722e+02, threshold=6.831e+02, percent-clipped=4.0 +2023-02-06 20:32:12,488 INFO [train.py:901] (2/4) Epoch 17, batch 6950, loss[loss=0.2039, simple_loss=0.2779, pruned_loss=0.06488, over 7933.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2972, pruned_loss=0.06898, over 1607631.18 frames. ], batch size: 20, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:32:22,236 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2272, 1.3468, 1.5963, 1.3738, 0.6945, 1.4173, 1.2549, 1.2316], + device='cuda:2'), covar=tensor([0.0534, 0.1363, 0.1697, 0.1421, 0.0595, 0.1515, 0.0667, 0.0656], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0158, 0.0101, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:32:32,958 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 20:32:47,695 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9893, 1.2558, 1.1922, 0.5193, 1.2101, 0.9983, 0.0987, 1.1906], + device='cuda:2'), covar=tensor([0.0407, 0.0343, 0.0348, 0.0556, 0.0396, 0.0885, 0.0733, 0.0294], + device='cuda:2'), in_proj_covar=tensor([0.0425, 0.0366, 0.0317, 0.0419, 0.0348, 0.0509, 0.0371, 0.0390], + device='cuda:2'), out_proj_covar=tensor([1.1580e-04, 9.7235e-05, 8.4124e-05, 1.1157e-04, 9.2951e-05, 1.4670e-04, + 1.0103e-04, 1.0438e-04], device='cuda:2') +2023-02-06 20:32:48,136 INFO [train.py:901] (2/4) Epoch 17, batch 7000, loss[loss=0.2255, simple_loss=0.3063, pruned_loss=0.0723, over 8498.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.297, pruned_loss=0.06897, over 1611360.98 frames. ], batch size: 26, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:32:58,254 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:04,323 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:06,278 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136356.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:11,085 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:18,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.499e+02 2.956e+02 3.583e+02 7.307e+02, threshold=5.911e+02, percent-clipped=2.0 +2023-02-06 20:33:22,378 INFO [train.py:901] (2/4) Epoch 17, batch 7050, loss[loss=0.1802, simple_loss=0.2636, pruned_loss=0.04837, over 7680.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2967, pruned_loss=0.06893, over 1607360.69 frames. ], batch size: 18, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:33:29,402 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:46,653 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6490, 1.8336, 1.6380, 2.2621, 1.0470, 1.4191, 1.6626, 1.8383], + device='cuda:2'), covar=tensor([0.0798, 0.0763, 0.0999, 0.0474, 0.1158, 0.1359, 0.0833, 0.0768], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0198, 0.0249, 0.0212, 0.0209, 0.0247, 0.0254, 0.0210], + 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') +2023-02-06 20:33:57,954 INFO [train.py:901] (2/4) Epoch 17, batch 7100, loss[loss=0.2354, simple_loss=0.3226, pruned_loss=0.07406, over 8189.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2977, pruned_loss=0.06935, over 1609647.13 frames. ], batch size: 23, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:34:04,102 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2583, 3.1284, 2.9020, 1.6701, 2.8489, 2.8500, 2.8221, 2.7333], + device='cuda:2'), covar=tensor([0.1349, 0.0964, 0.1596, 0.4886, 0.1310, 0.1250, 0.1953, 0.1243], + device='cuda:2'), in_proj_covar=tensor([0.0507, 0.0417, 0.0422, 0.0520, 0.0411, 0.0417, 0.0404, 0.0364], + 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') +2023-02-06 20:34:18,670 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:34:26,520 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136471.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:34:27,626 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.503e+02 2.917e+02 3.905e+02 1.004e+03, threshold=5.834e+02, percent-clipped=4.0 +2023-02-06 20:34:31,743 INFO [train.py:901] (2/4) Epoch 17, batch 7150, loss[loss=0.2213, simple_loss=0.3061, pruned_loss=0.06819, over 8341.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2972, pruned_loss=0.06943, over 1608233.62 frames. ], batch size: 25, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:34:50,068 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:35:07,636 INFO [train.py:901] (2/4) Epoch 17, batch 7200, loss[loss=0.2179, simple_loss=0.2874, pruned_loss=0.07417, over 7198.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2963, pruned_loss=0.0692, over 1606144.57 frames. ], batch size: 16, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:35:37,959 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.498e+02 3.072e+02 3.698e+02 8.742e+02, threshold=6.145e+02, percent-clipped=2.0 +2023-02-06 20:35:42,151 INFO [train.py:901] (2/4) Epoch 17, batch 7250, loss[loss=0.2109, simple_loss=0.2936, pruned_loss=0.06409, over 8350.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2965, pruned_loss=0.06901, over 1612219.01 frames. ], batch size: 24, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:36:11,362 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136619.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:36:17,806 INFO [train.py:901] (2/4) Epoch 17, batch 7300, loss[loss=0.2477, simple_loss=0.3299, pruned_loss=0.08275, over 8480.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2955, pruned_loss=0.0681, over 1613063.17 frames. ], batch size: 29, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:36:40,630 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:36:48,514 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.506e+02 2.969e+02 3.762e+02 7.100e+02, threshold=5.939e+02, percent-clipped=2.0 +2023-02-06 20:36:52,584 INFO [train.py:901] (2/4) Epoch 17, batch 7350, loss[loss=0.1986, simple_loss=0.286, pruned_loss=0.05559, over 8357.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2954, pruned_loss=0.06824, over 1612447.23 frames. ], batch size: 24, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:36:59,052 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 20:37:05,620 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136697.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:14,745 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4233, 1.8239, 2.0238, 1.2252, 2.0789, 1.2129, 0.6522, 1.7799], + device='cuda:2'), covar=tensor([0.0643, 0.0364, 0.0245, 0.0604, 0.0414, 0.0981, 0.0807, 0.0311], + device='cuda:2'), in_proj_covar=tensor([0.0428, 0.0367, 0.0316, 0.0420, 0.0348, 0.0512, 0.0371, 0.0392], + device='cuda:2'), out_proj_covar=tensor([1.1670e-04, 9.7663e-05, 8.3805e-05, 1.1183e-04, 9.3130e-05, 1.4729e-04, + 1.0086e-04, 1.0478e-04], device='cuda:2') +2023-02-06 20:37:16,510 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 20:37:17,971 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:27,018 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136727.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:28,122 INFO [train.py:901] (2/4) Epoch 17, batch 7400, loss[loss=0.2208, simple_loss=0.2858, pruned_loss=0.07788, over 7797.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2957, pruned_loss=0.06822, over 1613966.84 frames. ], batch size: 19, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:37:35,034 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 20:37:36,604 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136740.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:45,398 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136752.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:55,327 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6530, 4.6579, 4.2433, 2.1770, 4.1226, 4.2342, 4.2749, 3.9716], + device='cuda:2'), covar=tensor([0.0721, 0.0522, 0.0922, 0.4662, 0.0875, 0.1010, 0.1236, 0.0882], + device='cuda:2'), in_proj_covar=tensor([0.0505, 0.0419, 0.0420, 0.0522, 0.0412, 0.0417, 0.0408, 0.0365], + 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') +2023-02-06 20:37:59,284 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.354e+02 2.898e+02 3.777e+02 7.037e+02, threshold=5.795e+02, percent-clipped=3.0 +2023-02-06 20:38:03,290 INFO [train.py:901] (2/4) Epoch 17, batch 7450, loss[loss=0.2185, simple_loss=0.2894, pruned_loss=0.07382, over 7658.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2952, pruned_loss=0.06796, over 1611876.37 frames. ], batch size: 19, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:38:16,571 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 20:38:26,054 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:38:33,499 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8957, 1.8954, 6.0248, 2.2068, 5.3777, 5.0131, 5.4857, 5.4107], + device='cuda:2'), covar=tensor([0.0432, 0.4159, 0.0385, 0.3737, 0.1083, 0.0829, 0.0504, 0.0488], + device='cuda:2'), in_proj_covar=tensor([0.0578, 0.0622, 0.0656, 0.0597, 0.0673, 0.0574, 0.0574, 0.0635], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 20:38:34,156 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:38:37,330 INFO [train.py:901] (2/4) Epoch 17, batch 7500, loss[loss=0.1824, simple_loss=0.2677, pruned_loss=0.04858, over 7773.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2955, pruned_loss=0.06804, over 1616105.97 frames. ], batch size: 19, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:38:38,168 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9826, 1.8464, 2.0350, 1.7374, 0.9872, 1.8005, 2.3726, 2.3179], + device='cuda:2'), covar=tensor([0.0422, 0.1245, 0.1632, 0.1370, 0.0553, 0.1421, 0.0606, 0.0550], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:39:09,457 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.462e+02 2.866e+02 3.948e+02 7.787e+02, threshold=5.732e+02, percent-clipped=6.0 +2023-02-06 20:39:11,049 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:39:13,444 INFO [train.py:901] (2/4) Epoch 17, batch 7550, loss[loss=0.2531, simple_loss=0.3407, pruned_loss=0.08276, over 8341.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2964, pruned_loss=0.06816, over 1619431.18 frames. ], batch size: 26, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:39:28,624 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136900.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:39:49,021 INFO [train.py:901] (2/4) Epoch 17, batch 7600, loss[loss=0.2499, simple_loss=0.3242, pruned_loss=0.08784, over 8516.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2959, pruned_loss=0.0687, over 1613183.26 frames. ], batch size: 28, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:40:21,096 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 2.516e+02 2.945e+02 3.717e+02 7.457e+02, threshold=5.891e+02, percent-clipped=6.0 +2023-02-06 20:40:25,198 INFO [train.py:901] (2/4) Epoch 17, batch 7650, loss[loss=0.2408, simple_loss=0.3257, pruned_loss=0.07793, over 8457.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2961, pruned_loss=0.06869, over 1614185.92 frames. ], batch size: 27, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:40:43,283 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:41:00,049 INFO [train.py:901] (2/4) Epoch 17, batch 7700, loss[loss=0.2205, simple_loss=0.3087, pruned_loss=0.06617, over 8247.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2971, pruned_loss=0.06914, over 1618249.28 frames. ], batch size: 24, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:41:26,488 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 20:41:26,651 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1092, 1.3859, 1.6593, 1.3590, 1.0399, 1.4323, 1.7731, 1.4380], + device='cuda:2'), covar=tensor([0.0520, 0.1389, 0.1716, 0.1486, 0.0597, 0.1561, 0.0684, 0.0673], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0157, 0.0100, 0.0161, 0.0113, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:41:26,691 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137068.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:41:30,556 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.598e+02 3.111e+02 3.900e+02 8.834e+02, threshold=6.222e+02, percent-clipped=1.0 +2023-02-06 20:41:34,727 INFO [train.py:901] (2/4) Epoch 17, batch 7750, loss[loss=0.176, simple_loss=0.2538, pruned_loss=0.04905, over 7548.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2956, pruned_loss=0.06825, over 1616404.16 frames. ], batch size: 18, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:41:44,962 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:42:03,587 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:42:09,490 INFO [train.py:901] (2/4) Epoch 17, batch 7800, loss[loss=0.2225, simple_loss=0.3014, pruned_loss=0.07185, over 8499.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2964, pruned_loss=0.06844, over 1618591.92 frames. ], batch size: 26, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:42:36,424 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137168.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:42:39,579 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.454e+02 2.768e+02 3.488e+02 7.043e+02, threshold=5.537e+02, percent-clipped=4.0 +2023-02-06 20:42:43,618 INFO [train.py:901] (2/4) Epoch 17, batch 7850, loss[loss=0.1961, simple_loss=0.2686, pruned_loss=0.06185, over 7652.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2958, pruned_loss=0.06788, over 1622283.95 frames. ], batch size: 19, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:42:59,279 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0676, 1.5215, 1.6879, 1.4180, 0.9325, 1.5356, 1.7700, 1.6371], + device='cuda:2'), covar=tensor([0.0519, 0.1261, 0.1736, 0.1428, 0.0609, 0.1491, 0.0666, 0.0623], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0157, 0.0100, 0.0160, 0.0113, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:43:11,946 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5212, 1.5025, 1.8049, 1.2397, 1.0848, 1.8361, 0.1557, 1.1538], + device='cuda:2'), covar=tensor([0.1853, 0.1323, 0.0409, 0.1230, 0.3338, 0.0507, 0.2405, 0.1450], + device='cuda:2'), in_proj_covar=tensor([0.0176, 0.0182, 0.0116, 0.0216, 0.0261, 0.0122, 0.0165, 0.0181], + 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') +2023-02-06 20:43:16,598 INFO [train.py:901] (2/4) Epoch 17, batch 7900, loss[loss=0.2469, simple_loss=0.3237, pruned_loss=0.0851, over 8372.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2957, pruned_loss=0.06821, over 1618934.29 frames. ], batch size: 24, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:43:32,169 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 20:43:45,781 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.478e+02 3.005e+02 3.961e+02 6.905e+02, threshold=6.010e+02, percent-clipped=7.0 +2023-02-06 20:43:49,863 INFO [train.py:901] (2/4) Epoch 17, batch 7950, loss[loss=0.1717, simple_loss=0.2462, pruned_loss=0.04858, over 7650.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2942, pruned_loss=0.06684, over 1616669.53 frames. ], batch size: 19, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:43:52,843 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:44:23,205 INFO [train.py:901] (2/4) Epoch 17, batch 8000, loss[loss=0.187, simple_loss=0.2672, pruned_loss=0.05345, over 8137.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2946, pruned_loss=0.06748, over 1615158.81 frames. ], batch size: 22, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:44:36,541 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1968, 1.9055, 2.6132, 2.1121, 2.4262, 2.2351, 1.9462, 1.3220], + device='cuda:2'), covar=tensor([0.5033, 0.4654, 0.1618, 0.3573, 0.2482, 0.2772, 0.1804, 0.4930], + device='cuda:2'), in_proj_covar=tensor([0.0916, 0.0936, 0.0768, 0.0903, 0.0969, 0.0849, 0.0721, 0.0798], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 20:44:52,977 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.583e+02 3.026e+02 3.684e+02 1.341e+03, threshold=6.053e+02, percent-clipped=4.0 +2023-02-06 20:44:55,373 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:44:57,245 INFO [train.py:901] (2/4) Epoch 17, batch 8050, loss[loss=0.1769, simple_loss=0.2521, pruned_loss=0.05082, over 7437.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2945, pruned_loss=0.0678, over 1604585.70 frames. ], batch size: 17, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:45:12,523 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137401.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:45:29,500 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 20:45:34,942 INFO [train.py:901] (2/4) Epoch 18, batch 0, loss[loss=0.2195, simple_loss=0.2904, pruned_loss=0.07435, over 7794.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2904, pruned_loss=0.07435, over 7794.00 frames. ], batch size: 19, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:45:34,943 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 20:45:46,129 INFO [train.py:935] (2/4) Epoch 18, validation: loss=0.1783, simple_loss=0.2784, pruned_loss=0.03907, over 944034.00 frames. +2023-02-06 20:45:46,129 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 20:46:00,873 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 20:46:20,803 INFO [train.py:901] (2/4) Epoch 18, batch 50, loss[loss=0.2131, simple_loss=0.2896, pruned_loss=0.06826, over 8190.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2996, pruned_loss=0.06864, over 369368.78 frames. ], batch size: 23, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:46:29,002 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.698e+02 3.585e+02 4.414e+02 8.769e+02, threshold=7.169e+02, percent-clipped=9.0 +2023-02-06 20:46:35,903 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 20:46:42,988 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5781, 1.9751, 3.3505, 1.3393, 2.6154, 1.9800, 1.6336, 2.5619], + device='cuda:2'), covar=tensor([0.1865, 0.2436, 0.0911, 0.4306, 0.1659, 0.3134, 0.2172, 0.2116], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0577, 0.0548, 0.0620, 0.0642, 0.0584, 0.0514, 0.0629], + 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') +2023-02-06 20:46:56,055 INFO [train.py:901] (2/4) Epoch 18, batch 100, loss[loss=0.1916, simple_loss=0.2745, pruned_loss=0.05434, over 8187.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2975, pruned_loss=0.06761, over 646260.52 frames. ], batch size: 23, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:46:58,818 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 20:47:16,442 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:47:31,279 INFO [train.py:901] (2/4) Epoch 18, batch 150, loss[loss=0.1803, simple_loss=0.2527, pruned_loss=0.05397, over 7427.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2939, pruned_loss=0.0651, over 862370.95 frames. ], batch size: 17, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:47:33,520 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:47:39,693 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.369e+02 2.797e+02 3.885e+02 6.122e+02, threshold=5.595e+02, percent-clipped=0.0 +2023-02-06 20:48:07,693 INFO [train.py:901] (2/4) Epoch 18, batch 200, loss[loss=0.2458, simple_loss=0.3235, pruned_loss=0.08409, over 8492.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2953, pruned_loss=0.06636, over 1031781.55 frames. ], batch size: 28, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:48:33,245 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 20:48:44,088 INFO [train.py:901] (2/4) Epoch 18, batch 250, loss[loss=0.2226, simple_loss=0.3013, pruned_loss=0.07193, over 8223.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2961, pruned_loss=0.06696, over 1160913.37 frames. ], batch size: 22, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:48:52,404 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.467e+02 3.008e+02 3.586e+02 6.135e+02, threshold=6.015e+02, percent-clipped=1.0 +2023-02-06 20:48:55,930 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 20:49:03,705 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 20:49:19,871 INFO [train.py:901] (2/4) Epoch 18, batch 300, loss[loss=0.1973, simple_loss=0.2798, pruned_loss=0.05741, over 8239.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.296, pruned_loss=0.06684, over 1259435.91 frames. ], batch size: 22, lr: 4.28e-03, grad_scale: 8.0 +2023-02-06 20:49:55,785 INFO [train.py:901] (2/4) Epoch 18, batch 350, loss[loss=0.2265, simple_loss=0.3064, pruned_loss=0.0733, over 8548.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2958, pruned_loss=0.06697, over 1335398.12 frames. ], batch size: 39, lr: 4.28e-03, grad_scale: 8.0 +2023-02-06 20:50:05,700 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.555e+02 3.034e+02 3.752e+02 7.695e+02, threshold=6.069e+02, percent-clipped=3.0 +2023-02-06 20:50:21,152 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2107, 2.2742, 1.9267, 2.8593, 1.4041, 1.6364, 1.9310, 2.3289], + device='cuda:2'), covar=tensor([0.0638, 0.0713, 0.0902, 0.0360, 0.1093, 0.1274, 0.0974, 0.0692], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0198, 0.0251, 0.0211, 0.0208, 0.0248, 0.0255, 0.0209], + 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') +2023-02-06 20:50:32,304 INFO [train.py:901] (2/4) Epoch 18, batch 400, loss[loss=0.2752, simple_loss=0.3467, pruned_loss=0.1018, over 8529.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.296, pruned_loss=0.06696, over 1403156.48 frames. ], batch size: 49, lr: 4.28e-03, grad_scale: 8.0 +2023-02-06 20:50:33,932 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8832, 2.3362, 3.4614, 1.9494, 1.6530, 3.5022, 0.4626, 2.1511], + device='cuda:2'), covar=tensor([0.1655, 0.1347, 0.0271, 0.1974, 0.3185, 0.0322, 0.2788, 0.1640], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0184, 0.0118, 0.0219, 0.0263, 0.0124, 0.0166, 0.0182], + 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') +2023-02-06 20:51:08,176 INFO [train.py:901] (2/4) Epoch 18, batch 450, loss[loss=0.2359, simple_loss=0.3124, pruned_loss=0.07972, over 8451.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2955, pruned_loss=0.06675, over 1450223.43 frames. ], batch size: 27, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:51:16,916 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.504e+02 3.016e+02 3.557e+02 6.367e+02, threshold=6.032e+02, percent-clipped=3.0 +2023-02-06 20:51:35,829 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8671, 1.7443, 5.9911, 2.4207, 5.4755, 5.0580, 5.5093, 5.4601], + device='cuda:2'), covar=tensor([0.0463, 0.4510, 0.0269, 0.3277, 0.0828, 0.0807, 0.0460, 0.0418], + device='cuda:2'), in_proj_covar=tensor([0.0576, 0.0617, 0.0656, 0.0591, 0.0670, 0.0574, 0.0570, 0.0634], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 20:51:43,047 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137910.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:51:43,593 INFO [train.py:901] (2/4) Epoch 18, batch 500, loss[loss=0.2126, simple_loss=0.3038, pruned_loss=0.06068, over 8187.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2961, pruned_loss=0.06758, over 1483054.79 frames. ], batch size: 23, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:51:45,192 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137913.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:51:50,424 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 20:52:20,557 INFO [train.py:901] (2/4) Epoch 18, batch 550, loss[loss=0.2831, simple_loss=0.3228, pruned_loss=0.1217, over 7411.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2953, pruned_loss=0.06781, over 1508482.60 frames. ], batch size: 17, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:52:29,487 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.606e+02 3.197e+02 3.974e+02 7.545e+02, threshold=6.394e+02, percent-clipped=3.0 +2023-02-06 20:52:35,965 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 20:52:56,991 INFO [train.py:901] (2/4) Epoch 18, batch 600, loss[loss=0.2237, simple_loss=0.3045, pruned_loss=0.07139, over 8537.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2956, pruned_loss=0.06789, over 1532133.77 frames. ], batch size: 28, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:53:11,899 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 20:53:32,793 INFO [train.py:901] (2/4) Epoch 18, batch 650, loss[loss=0.1881, simple_loss=0.2587, pruned_loss=0.05875, over 7645.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2951, pruned_loss=0.06763, over 1546379.46 frames. ], batch size: 19, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:53:43,406 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.461e+02 2.865e+02 3.365e+02 7.739e+02, threshold=5.729e+02, percent-clipped=1.0 +2023-02-06 20:54:00,543 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 20:54:09,457 INFO [train.py:901] (2/4) Epoch 18, batch 700, loss[loss=0.2299, simple_loss=0.3103, pruned_loss=0.07471, over 8761.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2944, pruned_loss=0.06745, over 1557310.55 frames. ], batch size: 49, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:54:29,540 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 20:54:30,790 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1008, 2.6567, 3.5468, 2.2481, 2.0891, 3.7118, 0.7116, 2.2171], + device='cuda:2'), covar=tensor([0.1657, 0.1417, 0.0269, 0.1887, 0.2930, 0.0318, 0.2686, 0.1640], + device='cuda:2'), in_proj_covar=tensor([0.0178, 0.0184, 0.0117, 0.0217, 0.0262, 0.0123, 0.0165, 0.0182], + 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') +2023-02-06 20:54:44,137 INFO [train.py:901] (2/4) Epoch 18, batch 750, loss[loss=0.1836, simple_loss=0.2574, pruned_loss=0.05492, over 7530.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.293, pruned_loss=0.06638, over 1570752.05 frames. ], batch size: 18, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:54:53,226 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.332e+02 3.041e+02 3.730e+02 6.216e+02, threshold=6.081e+02, percent-clipped=3.0 +2023-02-06 20:54:58,181 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 20:55:08,082 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 20:55:19,814 INFO [train.py:901] (2/4) Epoch 18, batch 800, loss[loss=0.1779, simple_loss=0.2695, pruned_loss=0.04317, over 8088.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2934, pruned_loss=0.06617, over 1586389.46 frames. ], batch size: 21, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:55:49,565 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:55:51,609 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138257.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:55:54,217 INFO [train.py:901] (2/4) Epoch 18, batch 850, loss[loss=0.2455, simple_loss=0.3233, pruned_loss=0.08389, over 8501.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.294, pruned_loss=0.06637, over 1597549.35 frames. ], batch size: 29, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:56:03,035 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.308e+02 2.906e+02 3.562e+02 8.427e+02, threshold=5.812e+02, percent-clipped=4.0 +2023-02-06 20:56:13,562 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:56:30,843 INFO [train.py:901] (2/4) Epoch 18, batch 900, loss[loss=0.2066, simple_loss=0.3007, pruned_loss=0.05624, over 8196.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06722, over 1603965.26 frames. ], batch size: 23, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:56:35,137 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2904, 1.7155, 1.9344, 1.6810, 1.1372, 1.7158, 1.9501, 1.9872], + device='cuda:2'), covar=tensor([0.0494, 0.1140, 0.1485, 0.1302, 0.0600, 0.1386, 0.0696, 0.0541], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0158, 0.0100, 0.0161, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 20:56:50,638 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138340.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:57:05,377 INFO [train.py:901] (2/4) Epoch 18, batch 950, loss[loss=0.1814, simple_loss=0.266, pruned_loss=0.04842, over 7807.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2947, pruned_loss=0.06722, over 1607544.72 frames. ], batch size: 20, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:57:10,956 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138369.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:57:13,033 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138372.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:57:14,179 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.531e+02 3.020e+02 3.937e+02 8.991e+02, threshold=6.039e+02, percent-clipped=7.0 +2023-02-06 20:57:14,411 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2494, 2.3551, 2.2072, 2.9693, 1.9779, 2.1383, 2.2962, 2.6604], + device='cuda:2'), covar=tensor([0.0693, 0.0734, 0.0720, 0.0456, 0.0846, 0.0912, 0.0715, 0.0559], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0200, 0.0251, 0.0212, 0.0207, 0.0249, 0.0253, 0.0208], + 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') +2023-02-06 20:57:29,251 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 20:57:40,369 INFO [train.py:901] (2/4) Epoch 18, batch 1000, loss[loss=0.2369, simple_loss=0.3159, pruned_loss=0.07894, over 8104.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2951, pruned_loss=0.06742, over 1609632.83 frames. ], batch size: 23, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:57:56,203 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3826, 1.7276, 2.6718, 1.2329, 1.8592, 1.7574, 1.4701, 1.8009], + device='cuda:2'), covar=tensor([0.1960, 0.2513, 0.0844, 0.4549, 0.1950, 0.3177, 0.2298, 0.2339], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0574, 0.0544, 0.0618, 0.0636, 0.0577, 0.0510, 0.0624], + 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') +2023-02-06 20:58:05,512 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 20:58:16,696 INFO [train.py:901] (2/4) Epoch 18, batch 1050, loss[loss=0.1801, simple_loss=0.2571, pruned_loss=0.05153, over 7260.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2944, pruned_loss=0.067, over 1609504.31 frames. ], batch size: 16, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:58:18,825 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 20:58:25,546 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.454e+02 3.228e+02 4.133e+02 8.765e+02, threshold=6.456e+02, percent-clipped=4.0 +2023-02-06 20:58:51,053 INFO [train.py:901] (2/4) Epoch 18, batch 1100, loss[loss=0.1734, simple_loss=0.2504, pruned_loss=0.04822, over 7705.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2945, pruned_loss=0.06714, over 1610063.76 frames. ], batch size: 18, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 20:58:56,213 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3129, 4.2554, 3.8820, 2.2147, 3.7883, 3.8314, 3.9121, 3.6571], + device='cuda:2'), covar=tensor([0.0656, 0.0548, 0.1030, 0.3918, 0.0783, 0.0996, 0.1199, 0.0834], + device='cuda:2'), in_proj_covar=tensor([0.0507, 0.0419, 0.0419, 0.0517, 0.0409, 0.0422, 0.0407, 0.0365], + 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') +2023-02-06 20:59:26,908 INFO [train.py:901] (2/4) Epoch 18, batch 1150, loss[loss=0.2112, simple_loss=0.3007, pruned_loss=0.06088, over 8187.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.294, pruned_loss=0.06682, over 1607824.67 frames. ], batch size: 23, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 20:59:29,613 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 20:59:35,879 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.366e+02 2.909e+02 3.553e+02 5.350e+02, threshold=5.817e+02, percent-clipped=0.0 +2023-02-06 20:59:44,957 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6499, 2.1302, 3.2036, 1.3304, 2.4370, 1.9738, 1.7472, 2.3156], + device='cuda:2'), covar=tensor([0.2094, 0.2597, 0.1023, 0.4908, 0.1933, 0.3495, 0.2292, 0.2545], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0577, 0.0550, 0.0625, 0.0642, 0.0583, 0.0515, 0.0630], + 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') +2023-02-06 21:00:02,026 INFO [train.py:901] (2/4) Epoch 18, batch 1200, loss[loss=0.2177, simple_loss=0.3048, pruned_loss=0.06534, over 8598.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2931, pruned_loss=0.06608, over 1610617.67 frames. ], batch size: 34, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:00:04,276 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2392, 1.1852, 3.3869, 1.0111, 2.9852, 2.8335, 3.0834, 3.0047], + device='cuda:2'), covar=tensor([0.0810, 0.3971, 0.0741, 0.3987, 0.1302, 0.1069, 0.0769, 0.0832], + device='cuda:2'), in_proj_covar=tensor([0.0582, 0.0618, 0.0662, 0.0591, 0.0674, 0.0577, 0.0569, 0.0642], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:00:11,879 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138625.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:13,945 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138628.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:16,607 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:29,710 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138650.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:31,808 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138653.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:37,163 INFO [train.py:901] (2/4) Epoch 18, batch 1250, loss[loss=0.2026, simple_loss=0.288, pruned_loss=0.05863, over 8105.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2934, pruned_loss=0.06589, over 1614510.45 frames. ], batch size: 23, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:00:46,117 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5739, 1.9320, 5.8880, 2.3813, 4.7879, 4.8043, 5.5386, 5.4211], + device='cuda:2'), covar=tensor([0.1012, 0.5625, 0.0758, 0.4443, 0.1953, 0.1380, 0.0771, 0.0730], + device='cuda:2'), in_proj_covar=tensor([0.0584, 0.0621, 0.0664, 0.0594, 0.0677, 0.0580, 0.0572, 0.0645], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:00:47,264 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.310e+02 2.834e+02 3.613e+02 5.274e+02, threshold=5.668e+02, percent-clipped=0.0 +2023-02-06 21:00:54,246 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138684.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:04,840 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138699.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:08,266 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:12,862 INFO [train.py:901] (2/4) Epoch 18, batch 1300, loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09773, over 8328.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2937, pruned_loss=0.06598, over 1615476.45 frames. ], batch size: 25, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:01:15,833 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:34,957 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3974, 1.7489, 3.3869, 1.2485, 2.4229, 2.0063, 1.4589, 2.4668], + device='cuda:2'), covar=tensor([0.2109, 0.2698, 0.0762, 0.4645, 0.1826, 0.3085, 0.2363, 0.2026], + device='cuda:2'), in_proj_covar=tensor([0.0514, 0.0577, 0.0549, 0.0624, 0.0643, 0.0583, 0.0513, 0.0630], + 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') +2023-02-06 21:01:37,541 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138747.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:38,945 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1702, 2.3341, 2.0668, 2.9256, 1.4564, 1.8221, 2.1374, 2.5414], + device='cuda:2'), covar=tensor([0.0685, 0.0765, 0.0790, 0.0348, 0.0995, 0.1181, 0.0784, 0.0620], + device='cuda:2'), in_proj_covar=tensor([0.0236, 0.0202, 0.0254, 0.0214, 0.0210, 0.0251, 0.0256, 0.0210], + 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') +2023-02-06 21:01:47,854 INFO [train.py:901] (2/4) Epoch 18, batch 1350, loss[loss=0.2419, simple_loss=0.3264, pruned_loss=0.07873, over 8523.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.293, pruned_loss=0.06571, over 1612643.41 frames. ], batch size: 48, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:01:56,586 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.404e+02 2.906e+02 3.545e+02 6.613e+02, threshold=5.812e+02, percent-clipped=4.0 +2023-02-06 21:02:15,436 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:02:23,336 INFO [train.py:901] (2/4) Epoch 18, batch 1400, loss[loss=0.2521, simple_loss=0.3084, pruned_loss=0.09785, over 7254.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2942, pruned_loss=0.06612, over 1616311.63 frames. ], batch size: 16, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:02:43,933 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.03 vs. limit=5.0 +2023-02-06 21:02:57,612 INFO [train.py:901] (2/4) Epoch 18, batch 1450, loss[loss=0.1529, simple_loss=0.2331, pruned_loss=0.03634, over 7202.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2934, pruned_loss=0.066, over 1612685.54 frames. ], batch size: 16, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:03:06,393 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 2.491e+02 3.050e+02 4.246e+02 7.467e+02, threshold=6.100e+02, percent-clipped=3.0 +2023-02-06 21:03:07,097 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 21:03:33,394 INFO [train.py:901] (2/4) Epoch 18, batch 1500, loss[loss=0.1929, simple_loss=0.274, pruned_loss=0.05587, over 7789.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2933, pruned_loss=0.06591, over 1614689.31 frames. ], batch size: 19, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:03:44,791 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:04:08,397 INFO [train.py:901] (2/4) Epoch 18, batch 1550, loss[loss=0.2079, simple_loss=0.2847, pruned_loss=0.06557, over 8230.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2943, pruned_loss=0.06668, over 1618869.51 frames. ], batch size: 22, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:04:17,340 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 2.366e+02 2.933e+02 3.736e+02 6.367e+02, threshold=5.865e+02, percent-clipped=3.0 +2023-02-06 21:04:38,201 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139003.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:04:43,428 INFO [train.py:901] (2/4) Epoch 18, batch 1600, loss[loss=0.1783, simple_loss=0.2515, pruned_loss=0.05257, over 7548.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2939, pruned_loss=0.06654, over 1618295.31 frames. ], batch size: 18, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:04:56,943 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:07,017 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:10,363 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:15,396 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139055.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:18,029 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:18,147 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8012, 2.8385, 2.6664, 4.2171, 1.6362, 2.2155, 2.3742, 3.3011], + device='cuda:2'), covar=tensor([0.0643, 0.0840, 0.0756, 0.0193, 0.1170, 0.1309, 0.1090, 0.0682], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0202, 0.0251, 0.0212, 0.0208, 0.0249, 0.0256, 0.0209], + 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') +2023-02-06 21:05:19,994 INFO [train.py:901] (2/4) Epoch 18, batch 1650, loss[loss=0.1673, simple_loss=0.2435, pruned_loss=0.04558, over 8040.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2934, pruned_loss=0.06612, over 1618961.85 frames. ], batch size: 20, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:05:28,835 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.375e+02 2.907e+02 3.508e+02 7.626e+02, threshold=5.813e+02, percent-clipped=3.0 +2023-02-06 21:05:33,157 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2101, 3.0751, 2.9164, 1.5447, 2.8553, 2.9281, 2.8539, 2.7220], + device='cuda:2'), covar=tensor([0.1283, 0.0890, 0.1377, 0.5227, 0.1277, 0.1360, 0.1695, 0.1204], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0417, 0.0418, 0.0523, 0.0412, 0.0424, 0.0410, 0.0366], + 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') +2023-02-06 21:05:33,259 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:54,417 INFO [train.py:901] (2/4) Epoch 18, batch 1700, loss[loss=0.1807, simple_loss=0.2774, pruned_loss=0.04198, over 8187.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2928, pruned_loss=0.06584, over 1616301.98 frames. ], batch size: 23, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:06:28,711 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139158.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:06:30,624 INFO [train.py:901] (2/4) Epoch 18, batch 1750, loss[loss=0.2188, simple_loss=0.3022, pruned_loss=0.06769, over 8509.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06714, over 1615992.49 frames. ], batch size: 26, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:06:32,200 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:06:39,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.310e+02 2.913e+02 3.912e+02 7.750e+02, threshold=5.826e+02, percent-clipped=6.0 +2023-02-06 21:06:39,806 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139174.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:06:49,072 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.86 vs. limit=5.0 +2023-02-06 21:07:05,719 INFO [train.py:901] (2/4) Epoch 18, batch 1800, loss[loss=0.2261, simple_loss=0.3077, pruned_loss=0.07227, over 8481.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2933, pruned_loss=0.0664, over 1614718.10 frames. ], batch size: 28, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:07:37,627 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:07:40,841 INFO [train.py:901] (2/4) Epoch 18, batch 1850, loss[loss=0.2232, simple_loss=0.3006, pruned_loss=0.07292, over 8597.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2931, pruned_loss=0.06627, over 1615191.86 frames. ], batch size: 31, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:07:49,483 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:07:51,403 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.274e+02 2.776e+02 3.369e+02 8.658e+02, threshold=5.552e+02, percent-clipped=2.0 +2023-02-06 21:07:55,859 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 21:08:01,848 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0868, 2.4244, 2.0047, 2.8780, 1.2155, 1.7058, 1.9422, 2.4797], + device='cuda:2'), covar=tensor([0.0686, 0.0726, 0.0856, 0.0380, 0.1203, 0.1299, 0.1007, 0.0663], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0199, 0.0249, 0.0211, 0.0206, 0.0247, 0.0253, 0.0208], + 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') +2023-02-06 21:08:05,261 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139294.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:08:14,572 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139307.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:08:14,591 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0224, 1.7525, 3.4564, 1.4899, 2.2557, 3.8392, 3.9267, 3.2866], + device='cuda:2'), covar=tensor([0.1113, 0.1740, 0.0418, 0.2174, 0.1264, 0.0230, 0.0480, 0.0586], + device='cuda:2'), in_proj_covar=tensor([0.0281, 0.0311, 0.0274, 0.0305, 0.0293, 0.0253, 0.0391, 0.0296], + 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') +2023-02-06 21:08:17,141 INFO [train.py:901] (2/4) Epoch 18, batch 1900, loss[loss=0.1947, simple_loss=0.2707, pruned_loss=0.05937, over 8247.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2934, pruned_loss=0.06654, over 1619310.05 frames. ], batch size: 22, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:08:32,784 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 21:08:52,412 INFO [train.py:901] (2/4) Epoch 18, batch 1950, loss[loss=0.2024, simple_loss=0.284, pruned_loss=0.06039, over 8340.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2928, pruned_loss=0.06588, over 1620059.79 frames. ], batch size: 25, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:08:55,262 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 21:09:01,254 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.421e+02 2.964e+02 3.877e+02 7.962e+02, threshold=5.927e+02, percent-clipped=5.0 +2023-02-06 21:09:08,112 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 21:09:11,155 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:28,235 INFO [train.py:901] (2/4) Epoch 18, batch 2000, loss[loss=0.2225, simple_loss=0.3046, pruned_loss=0.07021, over 7974.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.294, pruned_loss=0.06613, over 1624011.27 frames. ], batch size: 21, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:09:28,240 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 21:09:30,568 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:33,876 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139419.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:42,129 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139430.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:48,232 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:49,604 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1044, 2.3603, 1.9522, 2.9649, 1.3263, 1.6559, 2.0259, 2.3218], + device='cuda:2'), covar=tensor([0.0697, 0.0809, 0.0929, 0.0321, 0.1128, 0.1338, 0.0943, 0.0769], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0200, 0.0250, 0.0213, 0.0207, 0.0249, 0.0255, 0.0209], + 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') +2023-02-06 21:09:51,681 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139444.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:59,137 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139455.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:10:02,980 INFO [train.py:901] (2/4) Epoch 18, batch 2050, loss[loss=0.2026, simple_loss=0.2674, pruned_loss=0.0689, over 7287.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2933, pruned_loss=0.06606, over 1616969.55 frames. ], batch size: 16, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:10:12,675 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.515e+02 3.080e+02 3.592e+02 7.733e+02, threshold=6.160e+02, percent-clipped=3.0 +2023-02-06 21:10:39,816 INFO [train.py:901] (2/4) Epoch 18, batch 2100, loss[loss=0.2259, simple_loss=0.2972, pruned_loss=0.07734, over 7816.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2928, pruned_loss=0.06566, over 1613110.30 frames. ], batch size: 20, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:11:15,315 INFO [train.py:901] (2/4) Epoch 18, batch 2150, loss[loss=0.2272, simple_loss=0.3087, pruned_loss=0.07284, over 8339.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2941, pruned_loss=0.06653, over 1610411.87 frames. ], batch size: 26, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:11:24,960 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.487e+02 3.024e+02 3.808e+02 9.008e+02, threshold=6.048e+02, percent-clipped=4.0 +2023-02-06 21:11:34,711 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139589.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:11:43,086 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:11:43,157 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:11:50,324 INFO [train.py:901] (2/4) Epoch 18, batch 2200, loss[loss=0.2326, simple_loss=0.3195, pruned_loss=0.07286, over 8142.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2937, pruned_loss=0.067, over 1611301.11 frames. ], batch size: 22, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:12:10,498 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139638.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:13,210 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:19,363 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:19,531 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5615, 1.9554, 3.2536, 1.3018, 2.5302, 2.0087, 1.6058, 2.3300], + device='cuda:2'), covar=tensor([0.1854, 0.2597, 0.0854, 0.4643, 0.1715, 0.3050, 0.2244, 0.2256], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0577, 0.0554, 0.0622, 0.0640, 0.0577, 0.0513, 0.0627], + 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') +2023-02-06 21:12:26,632 INFO [train.py:901] (2/4) Epoch 18, batch 2250, loss[loss=0.204, simple_loss=0.2927, pruned_loss=0.0576, over 8041.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2943, pruned_loss=0.06759, over 1609803.87 frames. ], batch size: 22, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:12:31,130 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:36,176 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.519e+02 3.270e+02 4.475e+02 8.912e+02, threshold=6.540e+02, percent-clipped=11.0 +2023-02-06 21:13:01,635 INFO [train.py:901] (2/4) Epoch 18, batch 2300, loss[loss=0.196, simple_loss=0.2852, pruned_loss=0.05344, over 8569.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2933, pruned_loss=0.06718, over 1607779.47 frames. ], batch size: 31, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:13:04,655 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:13:32,011 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:13:37,206 INFO [train.py:901] (2/4) Epoch 18, batch 2350, loss[loss=0.2211, simple_loss=0.3041, pruned_loss=0.06905, over 8489.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2947, pruned_loss=0.06754, over 1613271.67 frames. ], batch size: 29, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:13:40,614 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139766.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:13:44,709 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7929, 1.5903, 4.0189, 1.4304, 3.5443, 3.2911, 3.6170, 3.5153], + device='cuda:2'), covar=tensor([0.0687, 0.4055, 0.0565, 0.4094, 0.1247, 0.1042, 0.0634, 0.0786], + device='cuda:2'), in_proj_covar=tensor([0.0585, 0.0618, 0.0668, 0.0594, 0.0673, 0.0578, 0.0574, 0.0646], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:13:47,226 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.439e+02 2.945e+02 3.859e+02 6.515e+02, threshold=5.891e+02, percent-clipped=0.0 +2023-02-06 21:13:49,997 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2714, 1.6678, 4.5562, 2.1081, 2.4946, 5.2215, 5.1808, 4.5204], + device='cuda:2'), covar=tensor([0.1056, 0.1752, 0.0260, 0.1701, 0.1114, 0.0134, 0.0411, 0.0505], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0314, 0.0274, 0.0306, 0.0295, 0.0254, 0.0394, 0.0297], + 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') +2023-02-06 21:13:51,087 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 21:13:55,504 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:09,010 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:11,605 INFO [train.py:901] (2/4) Epoch 18, batch 2400, loss[loss=0.1755, simple_loss=0.2747, pruned_loss=0.03809, over 8248.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2948, pruned_loss=0.06766, over 1616060.14 frames. ], batch size: 22, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:14:20,250 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:21,630 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:48,487 INFO [train.py:901] (2/4) Epoch 18, batch 2450, loss[loss=0.2067, simple_loss=0.2758, pruned_loss=0.06882, over 7702.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2936, pruned_loss=0.06735, over 1612011.33 frames. ], batch size: 18, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:14:53,489 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:58,180 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.359e+02 2.854e+02 3.442e+02 8.627e+02, threshold=5.708e+02, percent-clipped=1.0 +2023-02-06 21:15:23,627 INFO [train.py:901] (2/4) Epoch 18, batch 2500, loss[loss=0.2392, simple_loss=0.3133, pruned_loss=0.08256, over 7583.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2939, pruned_loss=0.06706, over 1615205.67 frames. ], batch size: 71, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:15:39,670 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139933.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:15:46,144 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.24 vs. limit=5.0 +2023-02-06 21:15:47,235 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:00,240 INFO [train.py:901] (2/4) Epoch 18, batch 2550, loss[loss=0.1901, simple_loss=0.2568, pruned_loss=0.06168, over 7209.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2925, pruned_loss=0.06652, over 1611663.36 frames. ], batch size: 16, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:16:07,337 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139971.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:09,810 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.329e+02 2.906e+02 3.594e+02 7.294e+02, threshold=5.811e+02, percent-clipped=3.0 +2023-02-06 21:16:25,053 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139996.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:32,758 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6977, 1.9270, 1.9724, 1.3277, 2.1065, 1.4136, 0.6158, 1.9707], + device='cuda:2'), covar=tensor([0.0402, 0.0279, 0.0234, 0.0449, 0.0341, 0.0677, 0.0744, 0.0188], + device='cuda:2'), in_proj_covar=tensor([0.0430, 0.0371, 0.0321, 0.0426, 0.0357, 0.0517, 0.0378, 0.0398], + device='cuda:2'), out_proj_covar=tensor([1.1712e-04, 9.8174e-05, 8.5056e-05, 1.1353e-04, 9.5178e-05, 1.4837e-04, + 1.0278e-04, 1.0658e-04], device='cuda:2') +2023-02-06 21:16:35,503 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140009.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:36,618 INFO [train.py:901] (2/4) Epoch 18, batch 2600, loss[loss=0.2398, simple_loss=0.3293, pruned_loss=0.07513, over 8242.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2928, pruned_loss=0.06667, over 1611017.68 frames. ], batch size: 24, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:16:40,978 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6606, 1.6099, 4.9242, 2.0669, 4.3634, 4.0551, 4.4212, 4.2352], + device='cuda:2'), covar=tensor([0.0556, 0.4332, 0.0404, 0.3425, 0.1046, 0.0815, 0.0484, 0.0648], + device='cuda:2'), in_proj_covar=tensor([0.0585, 0.0619, 0.0668, 0.0595, 0.0674, 0.0578, 0.0574, 0.0647], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:16:44,583 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:44,601 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4966, 1.8120, 2.6163, 1.3027, 1.8892, 1.8881, 1.5935, 1.7950], + device='cuda:2'), covar=tensor([0.1806, 0.2392, 0.1013, 0.4301, 0.1851, 0.3043, 0.2143, 0.2300], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0575, 0.0552, 0.0620, 0.0637, 0.0578, 0.0513, 0.0626], + 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') +2023-02-06 21:16:52,718 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140034.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:01,563 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:03,001 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:05,697 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7416, 1.9412, 1.6592, 2.2507, 1.3003, 1.4464, 1.7267, 1.8751], + device='cuda:2'), covar=tensor([0.0771, 0.0721, 0.0965, 0.0429, 0.0958, 0.1300, 0.0744, 0.0785], + device='cuda:2'), in_proj_covar=tensor([0.0238, 0.0202, 0.0254, 0.0215, 0.0209, 0.0253, 0.0258, 0.0214], + 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') +2023-02-06 21:17:10,538 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:11,784 INFO [train.py:901] (2/4) Epoch 18, batch 2650, loss[loss=0.2111, simple_loss=0.2998, pruned_loss=0.06119, over 8257.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2931, pruned_loss=0.06648, over 1614703.40 frames. ], batch size: 24, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:17:22,348 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.571e+02 2.973e+02 3.666e+02 6.732e+02, threshold=5.945e+02, percent-clipped=3.0 +2023-02-06 21:17:43,051 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 21:17:44,109 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.12 vs. limit=5.0 +2023-02-06 21:17:47,904 INFO [train.py:901] (2/4) Epoch 18, batch 2700, loss[loss=0.289, simple_loss=0.3412, pruned_loss=0.1184, over 6711.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2932, pruned_loss=0.06612, over 1615805.73 frames. ], batch size: 71, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:17:55,819 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140121.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:00,545 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:02,537 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140131.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:09,874 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 21:18:16,407 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140151.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:23,213 INFO [train.py:901] (2/4) Epoch 18, batch 2750, loss[loss=0.178, simple_loss=0.2585, pruned_loss=0.04872, over 7931.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2942, pruned_loss=0.06675, over 1616105.71 frames. ], batch size: 20, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:18:27,488 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140166.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:28,895 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140168.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:33,790 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.336e+02 2.919e+02 3.807e+02 8.313e+02, threshold=5.838e+02, percent-clipped=5.0 +2023-02-06 21:19:00,768 INFO [train.py:901] (2/4) Epoch 18, batch 2800, loss[loss=0.1924, simple_loss=0.2537, pruned_loss=0.06552, over 7243.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2943, pruned_loss=0.06657, over 1620372.64 frames. ], batch size: 16, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:19:01,489 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:02,875 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7447, 1.5740, 1.8192, 1.4716, 1.2179, 1.5509, 2.2256, 1.9581], + device='cuda:2'), covar=tensor([0.0474, 0.1297, 0.1733, 0.1511, 0.0572, 0.1562, 0.0636, 0.0682], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0191, 0.0158, 0.0100, 0.0161, 0.0114, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 21:19:25,737 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140246.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:35,885 INFO [train.py:901] (2/4) Epoch 18, batch 2850, loss[loss=0.2082, simple_loss=0.2934, pruned_loss=0.0615, over 7810.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2944, pruned_loss=0.06646, over 1622226.50 frames. ], batch size: 20, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:19:39,552 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140266.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:45,685 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.447e+02 2.919e+02 3.574e+02 5.806e+02, threshold=5.838e+02, percent-clipped=0.0 +2023-02-06 21:19:47,308 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140277.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:50,909 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140281.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:51,970 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 21:19:52,309 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:07,401 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140304.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:10,144 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7836, 2.3549, 3.2363, 1.7065, 1.5954, 3.3178, 0.7361, 2.0667], + device='cuda:2'), covar=tensor([0.1701, 0.1392, 0.0350, 0.2039, 0.3255, 0.0381, 0.2673, 0.1731], + device='cuda:2'), in_proj_covar=tensor([0.0180, 0.0189, 0.0119, 0.0219, 0.0263, 0.0126, 0.0166, 0.0184], + 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') +2023-02-06 21:20:11,951 INFO [train.py:901] (2/4) Epoch 18, batch 2900, loss[loss=0.2059, simple_loss=0.2889, pruned_loss=0.06146, over 8020.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2953, pruned_loss=0.06742, over 1619930.38 frames. ], batch size: 22, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:20:14,965 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140315.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:23,767 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:25,188 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140329.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:31,571 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6722, 1.6750, 2.1843, 1.4443, 1.2333, 2.2626, 0.4180, 1.3783], + device='cuda:2'), covar=tensor([0.1939, 0.1675, 0.0445, 0.1330, 0.3335, 0.0404, 0.2657, 0.1421], + device='cuda:2'), in_proj_covar=tensor([0.0181, 0.0190, 0.0120, 0.0220, 0.0265, 0.0127, 0.0166, 0.0185], + 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') +2023-02-06 21:20:33,501 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140340.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:44,337 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 21:20:47,991 INFO [train.py:901] (2/4) Epoch 18, batch 2950, loss[loss=0.2464, simple_loss=0.332, pruned_loss=0.08036, over 8474.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2952, pruned_loss=0.06732, over 1618049.82 frames. ], batch size: 29, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:20:57,357 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.671e+02 3.280e+02 4.327e+02 7.160e+02, threshold=6.561e+02, percent-clipped=5.0 +2023-02-06 21:21:02,935 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3402, 1.5333, 4.5299, 1.7625, 3.9935, 3.7625, 4.0463, 3.9261], + device='cuda:2'), covar=tensor([0.0570, 0.4388, 0.0460, 0.3833, 0.1123, 0.0939, 0.0616, 0.0664], + device='cuda:2'), in_proj_covar=tensor([0.0586, 0.0618, 0.0665, 0.0596, 0.0675, 0.0578, 0.0575, 0.0646], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:21:18,956 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140405.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:21:23,788 INFO [train.py:901] (2/4) Epoch 18, batch 3000, loss[loss=0.1846, simple_loss=0.2607, pruned_loss=0.05423, over 7703.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2959, pruned_loss=0.06785, over 1618790.19 frames. ], batch size: 18, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:21:23,788 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 21:21:37,684 INFO [train.py:935] (2/4) Epoch 18, validation: loss=0.1773, simple_loss=0.2774, pruned_loss=0.03861, over 944034.00 frames. +2023-02-06 21:21:37,685 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 21:22:14,087 INFO [train.py:901] (2/4) Epoch 18, batch 3050, loss[loss=0.2229, simple_loss=0.3157, pruned_loss=0.06504, over 8331.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2952, pruned_loss=0.06751, over 1617099.10 frames. ], batch size: 25, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:22:16,898 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:21,665 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:24,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.663e+02 3.172e+02 4.119e+02 9.916e+02, threshold=6.345e+02, percent-clipped=7.0 +2023-02-06 21:22:40,213 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1764, 2.0041, 4.0472, 1.7440, 2.5361, 4.5493, 4.5854, 3.9141], + device='cuda:2'), covar=tensor([0.1179, 0.1595, 0.0344, 0.2053, 0.1187, 0.0188, 0.0405, 0.0586], + device='cuda:2'), in_proj_covar=tensor([0.0286, 0.0316, 0.0280, 0.0310, 0.0300, 0.0258, 0.0402, 0.0301], + 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') +2023-02-06 21:22:42,883 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140502.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:46,268 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:48,874 INFO [train.py:901] (2/4) Epoch 18, batch 3100, loss[loss=0.207, simple_loss=0.2815, pruned_loss=0.0662, over 7790.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2943, pruned_loss=0.06713, over 1615190.27 frames. ], batch size: 19, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:22:56,950 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140522.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:00,962 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:01,021 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:01,872 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 21:23:07,875 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140537.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:09,291 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:15,572 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:25,516 INFO [train.py:901] (2/4) Epoch 18, batch 3150, loss[loss=0.1903, simple_loss=0.2651, pruned_loss=0.05769, over 7649.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2936, pruned_loss=0.06693, over 1611017.83 frames. ], batch size: 19, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:23:26,373 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140562.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:27,685 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:34,940 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.438e+02 2.948e+02 4.263e+02 1.019e+03, threshold=5.895e+02, percent-clipped=4.0 +2023-02-06 21:23:38,569 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:40,647 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:43,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:59,140 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:24:01,005 INFO [train.py:901] (2/4) Epoch 18, batch 3200, loss[loss=0.2719, simple_loss=0.3341, pruned_loss=0.1049, over 6770.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2939, pruned_loss=0.0672, over 1609599.38 frames. ], batch size: 71, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:24:06,658 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3461, 2.8650, 3.2448, 1.6509, 3.3367, 2.2749, 1.5972, 2.4336], + device='cuda:2'), covar=tensor([0.0704, 0.0265, 0.0196, 0.0700, 0.0397, 0.0593, 0.0781, 0.0414], + device='cuda:2'), in_proj_covar=tensor([0.0434, 0.0375, 0.0321, 0.0430, 0.0360, 0.0520, 0.0379, 0.0399], + device='cuda:2'), out_proj_covar=tensor([1.1818e-04, 9.9388e-05, 8.5057e-05, 1.1472e-04, 9.6150e-05, 1.4908e-04, + 1.0321e-04, 1.0654e-04], device='cuda:2') +2023-02-06 21:24:07,872 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:24:36,820 INFO [train.py:901] (2/4) Epoch 18, batch 3250, loss[loss=0.1816, simple_loss=0.2697, pruned_loss=0.04673, over 8300.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2945, pruned_loss=0.06728, over 1612310.86 frames. ], batch size: 23, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:24:46,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.256e+02 2.889e+02 3.448e+02 6.536e+02, threshold=5.777e+02, percent-clipped=1.0 +2023-02-06 21:25:13,075 INFO [train.py:901] (2/4) Epoch 18, batch 3300, loss[loss=0.2594, simple_loss=0.3243, pruned_loss=0.09719, over 7783.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06722, over 1613447.14 frames. ], batch size: 19, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:25:30,567 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:25:33,926 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140741.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:25:39,313 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140749.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:25:44,917 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8592, 1.3812, 4.2275, 1.7556, 3.3234, 3.4269, 3.7906, 3.7671], + device='cuda:2'), covar=tensor([0.1419, 0.6559, 0.1043, 0.4850, 0.2406, 0.1740, 0.1123, 0.1069], + device='cuda:2'), in_proj_covar=tensor([0.0586, 0.0617, 0.0664, 0.0593, 0.0672, 0.0575, 0.0572, 0.0640], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:25:47,443 INFO [train.py:901] (2/4) Epoch 18, batch 3350, loss[loss=0.1793, simple_loss=0.2685, pruned_loss=0.04505, over 8091.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2955, pruned_loss=0.06751, over 1614465.55 frames. ], batch size: 21, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:25:57,584 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.504e+02 2.969e+02 3.727e+02 7.020e+02, threshold=5.938e+02, percent-clipped=2.0 +2023-02-06 21:26:17,431 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4219, 2.1170, 2.9045, 2.4282, 2.8327, 2.3590, 2.0553, 1.5875], + device='cuda:2'), covar=tensor([0.4572, 0.4442, 0.1627, 0.3121, 0.2117, 0.2717, 0.1799, 0.4877], + device='cuda:2'), in_proj_covar=tensor([0.0925, 0.0943, 0.0781, 0.0909, 0.0979, 0.0864, 0.0729, 0.0808], + 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') +2023-02-06 21:26:23,947 INFO [train.py:901] (2/4) Epoch 18, batch 3400, loss[loss=0.2279, simple_loss=0.3052, pruned_loss=0.07528, over 8555.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.295, pruned_loss=0.06792, over 1612945.62 frames. ], batch size: 39, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:26:35,825 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:42,137 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140836.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:46,744 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140843.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:52,094 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:55,938 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 21:26:58,803 INFO [train.py:901] (2/4) Epoch 18, batch 3450, loss[loss=0.2291, simple_loss=0.3163, pruned_loss=0.07099, over 8502.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2958, pruned_loss=0.06825, over 1616000.70 frames. ], batch size: 26, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:26:59,023 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:01,086 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:03,751 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:05,723 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140871.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:08,275 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.419e+02 3.065e+02 3.703e+02 6.567e+02, threshold=6.131e+02, percent-clipped=3.0 +2023-02-06 21:27:34,152 INFO [train.py:901] (2/4) Epoch 18, batch 3500, loss[loss=0.2126, simple_loss=0.2986, pruned_loss=0.06329, over 8463.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2956, pruned_loss=0.06789, over 1616215.51 frames. ], batch size: 49, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:27:37,701 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8174, 3.8064, 3.4893, 2.0585, 3.3615, 3.5246, 3.3547, 3.2003], + device='cuda:2'), covar=tensor([0.1112, 0.0700, 0.1355, 0.4483, 0.1089, 0.1073, 0.1585, 0.0919], + device='cuda:2'), in_proj_covar=tensor([0.0508, 0.0419, 0.0418, 0.0520, 0.0412, 0.0420, 0.0405, 0.0367], + 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') +2023-02-06 21:27:51,057 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 21:27:51,204 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140935.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:09,210 INFO [train.py:901] (2/4) Epoch 18, batch 3550, loss[loss=0.3175, simple_loss=0.37, pruned_loss=0.1325, over 8504.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2972, pruned_loss=0.06899, over 1619699.26 frames. ], batch size: 49, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:28:12,006 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:12,767 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140966.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:18,756 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.456e+02 3.083e+02 3.681e+02 6.081e+02, threshold=6.167e+02, percent-clipped=0.0 +2023-02-06 21:28:26,488 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140986.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:30,650 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140992.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:44,277 INFO [train.py:901] (2/4) Epoch 18, batch 3600, loss[loss=0.212, simple_loss=0.2994, pruned_loss=0.06227, over 7808.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2961, pruned_loss=0.06812, over 1617783.67 frames. ], batch size: 20, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:28:49,270 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141017.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:29:20,362 INFO [train.py:901] (2/4) Epoch 18, batch 3650, loss[loss=0.2212, simple_loss=0.3061, pruned_loss=0.06811, over 8584.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2959, pruned_loss=0.06815, over 1619372.11 frames. ], batch size: 31, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:29:30,818 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.345e+02 2.956e+02 3.633e+02 6.454e+02, threshold=5.912e+02, percent-clipped=1.0 +2023-02-06 21:29:37,715 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141085.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:29:55,736 INFO [train.py:901] (2/4) Epoch 18, batch 3700, loss[loss=0.2033, simple_loss=0.2874, pruned_loss=0.0596, over 8131.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.295, pruned_loss=0.06759, over 1621385.87 frames. ], batch size: 22, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:29:57,135 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 21:30:02,953 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:20,680 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141145.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:32,121 INFO [train.py:901] (2/4) Epoch 18, batch 3750, loss[loss=0.2404, simple_loss=0.3115, pruned_loss=0.08464, over 7373.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2951, pruned_loss=0.0673, over 1620161.19 frames. ], batch size: 71, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:30:32,305 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:39,106 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141171.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:41,865 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.679e+02 3.309e+02 4.099e+02 7.455e+02, threshold=6.618e+02, percent-clipped=7.0 +2023-02-06 21:31:00,284 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141200.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:07,418 INFO [train.py:901] (2/4) Epoch 18, batch 3800, loss[loss=0.2361, simple_loss=0.3061, pruned_loss=0.08308, over 7798.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2949, pruned_loss=0.06785, over 1615579.59 frames. ], batch size: 19, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:31:15,006 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141222.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:15,221 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.69 vs. limit=5.0 +2023-02-06 21:31:29,267 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141242.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:32,638 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:42,607 INFO [train.py:901] (2/4) Epoch 18, batch 3850, loss[loss=0.2276, simple_loss=0.3074, pruned_loss=0.0739, over 8501.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2963, pruned_loss=0.06833, over 1620722.81 frames. ], batch size: 28, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:31:46,881 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141267.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:52,716 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.500e+02 3.018e+02 3.684e+02 7.912e+02, threshold=6.036e+02, percent-clipped=1.0 +2023-02-06 21:31:53,563 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8782, 1.5367, 3.2377, 1.3815, 2.2721, 3.5661, 3.6544, 3.0858], + device='cuda:2'), covar=tensor([0.1209, 0.1774, 0.0353, 0.2132, 0.1146, 0.0267, 0.0635, 0.0586], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0311, 0.0275, 0.0306, 0.0297, 0.0254, 0.0396, 0.0297], + 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') +2023-02-06 21:31:55,475 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141279.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:00,518 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:03,626 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 21:32:03,922 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 21:32:17,072 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141309.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:18,342 INFO [train.py:901] (2/4) Epoch 18, batch 3900, loss[loss=0.2244, simple_loss=0.306, pruned_loss=0.07142, over 8431.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2948, pruned_loss=0.06748, over 1613052.27 frames. ], batch size: 27, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:32:42,395 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:52,481 INFO [train.py:901] (2/4) Epoch 18, batch 3950, loss[loss=0.2173, simple_loss=0.2935, pruned_loss=0.07061, over 8091.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2964, pruned_loss=0.06836, over 1611656.91 frames. ], batch size: 21, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:33:02,713 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.421e+02 2.990e+02 3.795e+02 7.053e+02, threshold=5.979e+02, percent-clipped=3.0 +2023-02-06 21:33:15,875 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141394.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:33:27,610 INFO [train.py:901] (2/4) Epoch 18, batch 4000, loss[loss=0.1747, simple_loss=0.2585, pruned_loss=0.04541, over 7652.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2954, pruned_loss=0.06821, over 1613480.22 frames. ], batch size: 19, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:33:37,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:33:40,606 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2261, 2.0307, 2.6818, 2.1875, 2.6565, 2.2484, 1.9728, 1.4554], + device='cuda:2'), covar=tensor([0.4845, 0.4416, 0.1755, 0.3313, 0.2237, 0.2672, 0.1791, 0.4857], + device='cuda:2'), in_proj_covar=tensor([0.0921, 0.0942, 0.0779, 0.0906, 0.0979, 0.0858, 0.0725, 0.0804], + 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') +2023-02-06 21:33:44,509 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141435.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:33:45,254 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4592, 1.6299, 2.2218, 1.3281, 1.5904, 1.7565, 1.5048, 1.4616], + device='cuda:2'), covar=tensor([0.1795, 0.2533, 0.0918, 0.4250, 0.1729, 0.3098, 0.2106, 0.2167], + device='cuda:2'), in_proj_covar=tensor([0.0510, 0.0575, 0.0549, 0.0620, 0.0637, 0.0578, 0.0512, 0.0626], + 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') +2023-02-06 21:33:58,725 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141456.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:00,764 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:02,050 INFO [train.py:901] (2/4) Epoch 18, batch 4050, loss[loss=0.2202, simple_loss=0.2976, pruned_loss=0.07145, over 8243.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2954, pruned_loss=0.06767, over 1612940.17 frames. ], batch size: 24, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:34:12,707 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.514e+02 3.146e+02 4.229e+02 8.641e+02, threshold=6.293e+02, percent-clipped=9.0 +2023-02-06 21:34:16,937 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:34,631 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141505.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:38,486 INFO [train.py:901] (2/4) Epoch 18, batch 4100, loss[loss=0.1868, simple_loss=0.2767, pruned_loss=0.04838, over 8074.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2965, pruned_loss=0.06825, over 1614091.85 frames. ], batch size: 21, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:34:47,610 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3111, 1.6584, 1.6793, 0.9860, 1.7000, 1.3225, 0.2663, 1.6020], + device='cuda:2'), covar=tensor([0.0400, 0.0286, 0.0216, 0.0405, 0.0309, 0.0754, 0.0715, 0.0204], + device='cuda:2'), in_proj_covar=tensor([0.0431, 0.0370, 0.0317, 0.0428, 0.0359, 0.0516, 0.0374, 0.0395], + device='cuda:2'), out_proj_covar=tensor([1.1716e-04, 9.7988e-05, 8.3866e-05, 1.1409e-04, 9.5594e-05, 1.4816e-04, + 1.0182e-04, 1.0552e-04], device='cuda:2') +2023-02-06 21:34:50,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3122, 1.9806, 2.8062, 2.2647, 2.8316, 2.2715, 1.9112, 1.5040], + device='cuda:2'), covar=tensor([0.5012, 0.4974, 0.1624, 0.3247, 0.2089, 0.2825, 0.1940, 0.4932], + device='cuda:2'), in_proj_covar=tensor([0.0923, 0.0944, 0.0779, 0.0907, 0.0979, 0.0860, 0.0727, 0.0807], + 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') +2023-02-06 21:35:00,240 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141542.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:35:09,996 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6515, 2.0081, 2.0936, 1.3451, 2.2224, 1.6135, 0.5737, 1.9185], + device='cuda:2'), covar=tensor([0.0466, 0.0297, 0.0257, 0.0493, 0.0368, 0.0756, 0.0708, 0.0242], + device='cuda:2'), in_proj_covar=tensor([0.0432, 0.0372, 0.0318, 0.0430, 0.0359, 0.0518, 0.0376, 0.0397], + device='cuda:2'), out_proj_covar=tensor([1.1750e-04, 9.8399e-05, 8.4316e-05, 1.1459e-04, 9.5719e-05, 1.4871e-04, + 1.0227e-04, 1.0603e-04], device='cuda:2') +2023-02-06 21:35:13,072 INFO [train.py:901] (2/4) Epoch 18, batch 4150, loss[loss=0.2188, simple_loss=0.2903, pruned_loss=0.07363, over 7921.00 frames. ], tot_loss[loss=0.215, simple_loss=0.295, pruned_loss=0.06744, over 1615280.39 frames. ], batch size: 20, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:35:17,337 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141567.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:35:20,207 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8442, 1.9151, 2.2057, 2.0578, 1.0654, 1.8928, 2.4862, 2.4604], + device='cuda:2'), covar=tensor([0.0471, 0.1166, 0.1516, 0.1219, 0.0578, 0.1384, 0.0573, 0.0497], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0157, 0.0099, 0.0161, 0.0114, 0.0138], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 21:35:22,670 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.507e+02 2.964e+02 3.952e+02 7.900e+02, threshold=5.928e+02, percent-clipped=3.0 +2023-02-06 21:35:37,892 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.56 vs. limit=2.0 +2023-02-06 21:35:48,917 INFO [train.py:901] (2/4) Epoch 18, batch 4200, loss[loss=0.2195, simple_loss=0.3128, pruned_loss=0.06314, over 8108.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2952, pruned_loss=0.06731, over 1614637.91 frames. ], batch size: 23, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:35:55,112 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:02,347 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 21:36:15,990 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141650.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:23,889 INFO [train.py:901] (2/4) Epoch 18, batch 4250, loss[loss=0.1857, simple_loss=0.2719, pruned_loss=0.04977, over 8327.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2946, pruned_loss=0.0669, over 1614087.83 frames. ], batch size: 25, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:36:24,587 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 21:36:33,261 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.491e+02 2.994e+02 3.932e+02 8.485e+02, threshold=5.988e+02, percent-clipped=6.0 +2023-02-06 21:36:33,491 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141675.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:36,882 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:44,012 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:54,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141705.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:58,037 INFO [train.py:901] (2/4) Epoch 18, batch 4300, loss[loss=0.2053, simple_loss=0.2943, pruned_loss=0.05816, over 8477.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2959, pruned_loss=0.0676, over 1614862.31 frames. ], batch size: 29, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:37:32,932 INFO [train.py:901] (2/4) Epoch 18, batch 4350, loss[loss=0.2182, simple_loss=0.306, pruned_loss=0.06518, over 8493.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2959, pruned_loss=0.06718, over 1620231.05 frames. ], batch size: 26, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:37:38,875 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-06 21:37:43,208 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.620e+02 3.197e+02 4.150e+02 9.266e+02, threshold=6.393e+02, percent-clipped=5.0 +2023-02-06 21:37:46,100 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141779.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:37:54,221 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 21:38:02,426 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141803.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:38:04,521 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:38:07,895 INFO [train.py:901] (2/4) Epoch 18, batch 4400, loss[loss=0.1967, simple_loss=0.2939, pruned_loss=0.0497, over 8539.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2948, pruned_loss=0.06623, over 1619526.50 frames. ], batch size: 28, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:38:36,599 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 21:38:43,835 INFO [train.py:901] (2/4) Epoch 18, batch 4450, loss[loss=0.1959, simple_loss=0.285, pruned_loss=0.05338, over 8331.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2938, pruned_loss=0.06559, over 1620195.62 frames. ], batch size: 25, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:38:53,320 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.507e+02 2.868e+02 3.524e+02 7.777e+02, threshold=5.735e+02, percent-clipped=2.0 +2023-02-06 21:38:54,252 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:07,098 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141894.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:39:11,835 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141901.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:18,270 INFO [train.py:901] (2/4) Epoch 18, batch 4500, loss[loss=0.2304, simple_loss=0.3233, pruned_loss=0.06872, over 8614.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2946, pruned_loss=0.06668, over 1612332.32 frames. ], batch size: 31, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:39:23,247 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:27,783 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 21:39:34,728 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141934.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:42,674 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141946.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:53,383 INFO [train.py:901] (2/4) Epoch 18, batch 4550, loss[loss=0.1815, simple_loss=0.254, pruned_loss=0.05446, over 7414.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2946, pruned_loss=0.06671, over 1608678.76 frames. ], batch size: 17, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:40:03,496 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.488e+02 2.920e+02 3.454e+02 6.371e+02, threshold=5.840e+02, percent-clipped=2.0 +2023-02-06 21:40:29,728 INFO [train.py:901] (2/4) Epoch 18, batch 4600, loss[loss=0.2076, simple_loss=0.2691, pruned_loss=0.07309, over 7416.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2942, pruned_loss=0.06654, over 1610183.69 frames. ], batch size: 17, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:41:04,180 INFO [train.py:901] (2/4) Epoch 18, batch 4650, loss[loss=0.2353, simple_loss=0.3081, pruned_loss=0.08128, over 8501.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2947, pruned_loss=0.067, over 1612489.57 frames. ], batch size: 29, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:41:05,108 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:41:13,896 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 2.389e+02 2.901e+02 3.503e+02 7.256e+02, threshold=5.801e+02, percent-clipped=3.0 +2023-02-06 21:41:23,649 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:41:39,476 INFO [train.py:901] (2/4) Epoch 18, batch 4700, loss[loss=0.2069, simple_loss=0.2885, pruned_loss=0.06266, over 8469.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2941, pruned_loss=0.06651, over 1612590.39 frames. ], batch size: 25, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:41:42,996 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.8210, 4.7016, 4.3204, 2.9727, 4.2798, 4.3861, 4.4701, 4.0995], + device='cuda:2'), covar=tensor([0.0530, 0.0462, 0.0861, 0.3262, 0.0682, 0.0879, 0.1045, 0.0610], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0421, 0.0417, 0.0517, 0.0410, 0.0415, 0.0399, 0.0364], + 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') +2023-02-06 21:42:06,026 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142150.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:42:13,354 INFO [train.py:901] (2/4) Epoch 18, batch 4750, loss[loss=0.2241, simple_loss=0.3053, pruned_loss=0.07142, over 8143.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2944, pruned_loss=0.06687, over 1614298.60 frames. ], batch size: 22, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:42:15,763 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4864, 2.0269, 3.2225, 1.3762, 2.3519, 1.9287, 1.7037, 2.4747], + device='cuda:2'), covar=tensor([0.1874, 0.2372, 0.0941, 0.4181, 0.1798, 0.2976, 0.2073, 0.2050], + device='cuda:2'), in_proj_covar=tensor([0.0510, 0.0574, 0.0546, 0.0620, 0.0633, 0.0578, 0.0510, 0.0621], + 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') +2023-02-06 21:42:23,428 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142174.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:42:23,882 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.398e+02 2.792e+02 3.541e+02 9.190e+02, threshold=5.585e+02, percent-clipped=4.0 +2023-02-06 21:42:24,107 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142175.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 21:42:30,577 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 21:42:32,632 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 21:42:41,665 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142199.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:42:49,343 INFO [train.py:901] (2/4) Epoch 18, batch 4800, loss[loss=0.2299, simple_loss=0.314, pruned_loss=0.0729, over 8497.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2945, pruned_loss=0.06722, over 1613404.30 frames. ], batch size: 28, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:42:59,095 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8619, 1.4975, 3.9683, 1.4145, 3.5490, 3.2963, 3.5996, 3.5044], + device='cuda:2'), covar=tensor([0.0599, 0.4090, 0.0640, 0.3944, 0.1132, 0.0951, 0.0585, 0.0661], + device='cuda:2'), in_proj_covar=tensor([0.0595, 0.0626, 0.0670, 0.0602, 0.0682, 0.0582, 0.0575, 0.0643], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:43:23,840 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 21:43:24,503 INFO [train.py:901] (2/4) Epoch 18, batch 4850, loss[loss=0.2451, simple_loss=0.3138, pruned_loss=0.0882, over 7973.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2938, pruned_loss=0.06693, over 1613577.32 frames. ], batch size: 21, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:43:33,963 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.522e+02 3.053e+02 3.876e+02 6.315e+02, threshold=6.106e+02, percent-clipped=2.0 +2023-02-06 21:43:36,095 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:43:45,116 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142290.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:43:59,066 INFO [train.py:901] (2/4) Epoch 18, batch 4900, loss[loss=0.24, simple_loss=0.3283, pruned_loss=0.07589, over 8356.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2928, pruned_loss=0.06651, over 1612412.55 frames. ], batch size: 24, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:44:11,380 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1015, 1.4335, 4.3137, 1.6852, 3.7887, 3.5647, 3.9109, 3.7538], + device='cuda:2'), covar=tensor([0.0724, 0.4445, 0.0568, 0.3953, 0.1218, 0.1026, 0.0620, 0.0752], + device='cuda:2'), in_proj_covar=tensor([0.0589, 0.0620, 0.0664, 0.0596, 0.0675, 0.0578, 0.0571, 0.0638], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:44:34,497 INFO [train.py:901] (2/4) Epoch 18, batch 4950, loss[loss=0.2128, simple_loss=0.2888, pruned_loss=0.06843, over 7657.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2909, pruned_loss=0.06508, over 1611405.59 frames. ], batch size: 19, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:44:44,951 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.451e+02 2.943e+02 3.789e+02 7.945e+02, threshold=5.886e+02, percent-clipped=1.0 +2023-02-06 21:44:51,991 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2096, 2.2677, 2.0720, 2.5390, 1.9558, 2.0624, 2.1349, 2.4801], + device='cuda:2'), covar=tensor([0.0616, 0.0697, 0.0770, 0.0507, 0.0825, 0.0922, 0.0647, 0.0535], + device='cuda:2'), in_proj_covar=tensor([0.0235, 0.0201, 0.0253, 0.0214, 0.0207, 0.0250, 0.0256, 0.0213], + 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') +2023-02-06 21:44:57,236 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:45:05,907 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142405.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:45:09,747 INFO [train.py:901] (2/4) Epoch 18, batch 5000, loss[loss=0.2662, simple_loss=0.3432, pruned_loss=0.09462, over 8499.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2935, pruned_loss=0.0666, over 1617828.06 frames. ], batch size: 26, lr: 4.21e-03, grad_scale: 8.0 +2023-02-06 21:45:44,314 INFO [train.py:901] (2/4) Epoch 18, batch 5050, loss[loss=0.2039, simple_loss=0.2908, pruned_loss=0.05852, over 8329.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2935, pruned_loss=0.06645, over 1619481.64 frames. ], batch size: 26, lr: 4.21e-03, grad_scale: 8.0 +2023-02-06 21:45:48,971 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 21:45:54,478 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.384e+02 2.804e+02 3.417e+02 5.925e+02, threshold=5.609e+02, percent-clipped=1.0 +2023-02-06 21:46:04,022 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 21:46:19,170 INFO [train.py:901] (2/4) Epoch 18, batch 5100, loss[loss=0.2407, simple_loss=0.3194, pruned_loss=0.08102, over 8244.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2942, pruned_loss=0.06672, over 1618989.39 frames. ], batch size: 24, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:46:46,513 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1839, 2.0255, 2.4691, 1.6697, 1.6351, 2.3737, 1.2432, 2.0629], + device='cuda:2'), covar=tensor([0.1921, 0.1164, 0.0427, 0.1423, 0.2524, 0.0585, 0.2087, 0.1372], + device='cuda:2'), in_proj_covar=tensor([0.0182, 0.0190, 0.0122, 0.0217, 0.0265, 0.0129, 0.0166, 0.0184], + 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') +2023-02-06 21:46:54,257 INFO [train.py:901] (2/4) Epoch 18, batch 5150, loss[loss=0.209, simple_loss=0.2903, pruned_loss=0.06385, over 8319.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2941, pruned_loss=0.0671, over 1619084.49 frames. ], batch size: 25, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:47:04,404 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.481e+02 3.004e+02 4.323e+02 1.197e+03, threshold=6.009e+02, percent-clipped=7.0 +2023-02-06 21:47:14,692 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142591.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:47:29,009 INFO [train.py:901] (2/4) Epoch 18, batch 5200, loss[loss=0.2026, simple_loss=0.2745, pruned_loss=0.06537, over 7217.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2929, pruned_loss=0.06646, over 1612914.09 frames. ], batch size: 16, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:47:34,977 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2023-02-06 21:47:55,310 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:02,991 INFO [train.py:901] (2/4) Epoch 18, batch 5250, loss[loss=0.1981, simple_loss=0.2736, pruned_loss=0.06128, over 7552.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2923, pruned_loss=0.06634, over 1606737.17 frames. ], batch size: 18, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:48:03,209 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:03,672 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 21:48:13,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:14,359 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.428e+02 2.491e+02 3.102e+02 3.692e+02 6.533e+02, threshold=6.204e+02, percent-clipped=2.0 +2023-02-06 21:48:21,247 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:37,895 INFO [train.py:901] (2/4) Epoch 18, batch 5300, loss[loss=0.2592, simple_loss=0.3292, pruned_loss=0.09457, over 8757.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2949, pruned_loss=0.06767, over 1606866.42 frames. ], batch size: 30, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:48:52,020 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9890, 1.7103, 2.0096, 1.7126, 1.2766, 1.6732, 2.3593, 1.9964], + device='cuda:2'), covar=tensor([0.0411, 0.1149, 0.1493, 0.1290, 0.0522, 0.1327, 0.0540, 0.0586], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0151, 0.0189, 0.0157, 0.0099, 0.0161, 0.0113, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 21:48:59,519 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142742.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:49:12,900 INFO [train.py:901] (2/4) Epoch 18, batch 5350, loss[loss=0.199, simple_loss=0.2908, pruned_loss=0.05362, over 8259.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2928, pruned_loss=0.06662, over 1604081.09 frames. ], batch size: 24, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:49:22,762 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.581e+02 3.011e+02 3.651e+02 7.168e+02, threshold=6.023e+02, percent-clipped=3.0 +2023-02-06 21:49:48,114 INFO [train.py:901] (2/4) Epoch 18, batch 5400, loss[loss=0.2259, simple_loss=0.3091, pruned_loss=0.07136, over 8509.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.292, pruned_loss=0.06569, over 1610156.14 frames. ], batch size: 49, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:50:22,758 INFO [train.py:901] (2/4) Epoch 18, batch 5450, loss[loss=0.1982, simple_loss=0.2958, pruned_loss=0.05026, over 8114.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2906, pruned_loss=0.06517, over 1605662.95 frames. ], batch size: 23, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:50:33,557 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.381e+02 3.003e+02 4.378e+02 7.690e+02, threshold=6.006e+02, percent-clipped=4.0 +2023-02-06 21:50:50,022 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 21:50:58,893 INFO [train.py:901] (2/4) Epoch 18, batch 5500, loss[loss=0.2229, simple_loss=0.3034, pruned_loss=0.07117, over 8083.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2918, pruned_loss=0.06554, over 1607619.33 frames. ], batch size: 21, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:51:14,988 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142935.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:51:18,599 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3634, 2.2765, 3.1627, 2.4591, 2.8816, 2.4647, 2.1632, 1.7574], + device='cuda:2'), covar=tensor([0.5055, 0.4852, 0.1766, 0.3550, 0.2591, 0.2805, 0.1960, 0.5498], + device='cuda:2'), in_proj_covar=tensor([0.0926, 0.0946, 0.0778, 0.0912, 0.0984, 0.0865, 0.0732, 0.0811], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:51:33,201 INFO [train.py:901] (2/4) Epoch 18, batch 5550, loss[loss=0.1953, simple_loss=0.2778, pruned_loss=0.05637, over 8237.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2918, pruned_loss=0.06573, over 1604075.71 frames. ], batch size: 22, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:51:43,334 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.398e+02 2.938e+02 3.826e+02 1.126e+03, threshold=5.876e+02, percent-clipped=10.0 +2023-02-06 21:51:48,889 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5368, 1.9952, 3.1275, 1.3664, 2.3339, 1.9565, 1.6766, 2.3058], + device='cuda:2'), covar=tensor([0.1888, 0.2357, 0.0831, 0.4409, 0.1805, 0.3085, 0.2164, 0.2212], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0574, 0.0549, 0.0620, 0.0634, 0.0580, 0.0512, 0.0625], + 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') +2023-02-06 21:52:08,269 INFO [train.py:901] (2/4) Epoch 18, batch 5600, loss[loss=0.1819, simple_loss=0.262, pruned_loss=0.05085, over 7813.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2914, pruned_loss=0.06515, over 1604683.32 frames. ], batch size: 20, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:52:11,238 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0872, 1.7447, 6.1815, 2.2666, 5.5985, 5.2158, 5.7818, 5.7034], + device='cuda:2'), covar=tensor([0.0458, 0.4501, 0.0372, 0.3695, 0.1058, 0.0989, 0.0418, 0.0444], + device='cuda:2'), in_proj_covar=tensor([0.0595, 0.0623, 0.0670, 0.0602, 0.0680, 0.0584, 0.0576, 0.0647], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:52:18,470 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 21:52:36,271 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143050.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:52:43,531 INFO [train.py:901] (2/4) Epoch 18, batch 5650, loss[loss=0.1967, simple_loss=0.2671, pruned_loss=0.06312, over 7781.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2905, pruned_loss=0.06472, over 1601223.93 frames. ], batch size: 19, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:52:46,062 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.20 vs. limit=5.0 +2023-02-06 21:52:54,552 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.315e+02 3.071e+02 3.627e+02 7.364e+02, threshold=6.141e+02, percent-clipped=4.0 +2023-02-06 21:53:00,420 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 21:53:01,151 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143086.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:53:18,752 INFO [train.py:901] (2/4) Epoch 18, batch 5700, loss[loss=0.1985, simple_loss=0.2913, pruned_loss=0.05287, over 8027.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.291, pruned_loss=0.06494, over 1606806.32 frames. ], batch size: 22, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:53:19,575 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0290, 1.7733, 3.2541, 1.2053, 2.3196, 3.6057, 3.8277, 2.6251], + device='cuda:2'), covar=tensor([0.1336, 0.1908, 0.0508, 0.2908, 0.1325, 0.0383, 0.0638, 0.1089], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0312, 0.0276, 0.0307, 0.0297, 0.0254, 0.0397, 0.0297], + 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') +2023-02-06 21:53:53,726 INFO [train.py:901] (2/4) Epoch 18, batch 5750, loss[loss=0.2049, simple_loss=0.28, pruned_loss=0.06493, over 7534.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.291, pruned_loss=0.06489, over 1607538.82 frames. ], batch size: 18, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:54:04,013 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.423e+02 2.839e+02 3.621e+02 5.889e+02, threshold=5.677e+02, percent-clipped=0.0 +2023-02-06 21:54:04,723 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 21:54:21,856 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:54:28,668 INFO [train.py:901] (2/4) Epoch 18, batch 5800, loss[loss=0.1637, simple_loss=0.2395, pruned_loss=0.04394, over 7431.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2903, pruned_loss=0.06413, over 1610710.36 frames. ], batch size: 17, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:54:43,931 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5751, 5.5685, 4.9082, 2.6600, 4.9447, 5.1911, 5.2193, 4.9829], + device='cuda:2'), covar=tensor([0.0488, 0.0364, 0.0895, 0.3888, 0.0720, 0.0816, 0.0950, 0.0615], + device='cuda:2'), in_proj_covar=tensor([0.0506, 0.0421, 0.0420, 0.0517, 0.0412, 0.0419, 0.0402, 0.0368], + 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') +2023-02-06 21:55:04,395 INFO [train.py:901] (2/4) Epoch 18, batch 5850, loss[loss=0.1977, simple_loss=0.2883, pruned_loss=0.0536, over 8037.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2906, pruned_loss=0.06459, over 1610862.48 frames. ], batch size: 22, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:55:15,560 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.467e+02 2.892e+02 3.630e+02 6.628e+02, threshold=5.783e+02, percent-clipped=2.0 +2023-02-06 21:55:36,571 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143306.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:55:39,811 INFO [train.py:901] (2/4) Epoch 18, batch 5900, loss[loss=0.1859, simple_loss=0.2588, pruned_loss=0.05646, over 7693.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2912, pruned_loss=0.0648, over 1609165.51 frames. ], batch size: 18, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:55:53,316 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143331.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:56:03,174 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1903, 2.0252, 2.6838, 2.2099, 2.5993, 2.2423, 1.9546, 1.3801], + device='cuda:2'), covar=tensor([0.5285, 0.4799, 0.1831, 0.3393, 0.2260, 0.2811, 0.1835, 0.5066], + device='cuda:2'), in_proj_covar=tensor([0.0929, 0.0946, 0.0777, 0.0915, 0.0981, 0.0866, 0.0729, 0.0812], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 21:56:14,463 INFO [train.py:901] (2/4) Epoch 18, batch 5950, loss[loss=0.2188, simple_loss=0.2982, pruned_loss=0.06969, over 8532.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2926, pruned_loss=0.06552, over 1611317.05 frames. ], batch size: 31, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:56:25,158 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.388e+02 2.875e+02 3.741e+02 7.794e+02, threshold=5.749e+02, percent-clipped=3.0 +2023-02-06 21:56:49,476 INFO [train.py:901] (2/4) Epoch 18, batch 6000, loss[loss=0.242, simple_loss=0.3158, pruned_loss=0.08407, over 8452.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2905, pruned_loss=0.0643, over 1610727.54 frames. ], batch size: 27, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:56:49,476 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 21:57:03,430 INFO [train.py:935] (2/4) Epoch 18, validation: loss=0.1765, simple_loss=0.2767, pruned_loss=0.03814, over 944034.00 frames. +2023-02-06 21:57:03,431 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 21:57:07,208 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9626, 2.2037, 1.9479, 2.8238, 1.2612, 1.7254, 2.0898, 2.2499], + device='cuda:2'), covar=tensor([0.0734, 0.0830, 0.0856, 0.0352, 0.1211, 0.1254, 0.0851, 0.0843], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0199, 0.0251, 0.0213, 0.0208, 0.0247, 0.0254, 0.0214], + 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') +2023-02-06 21:57:08,523 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143418.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:57:33,582 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8328, 5.9689, 5.2323, 2.6779, 5.3235, 5.5432, 5.5477, 5.3397], + device='cuda:2'), covar=tensor([0.0602, 0.0403, 0.0888, 0.4084, 0.0792, 0.0760, 0.0976, 0.0655], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0428, 0.0427, 0.0527, 0.0419, 0.0425, 0.0410, 0.0375], + 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') +2023-02-06 21:57:35,795 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143457.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:57:38,433 INFO [train.py:901] (2/4) Epoch 18, batch 6050, loss[loss=0.2301, simple_loss=0.3048, pruned_loss=0.07768, over 8480.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.292, pruned_loss=0.06529, over 1612278.79 frames. ], batch size: 27, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:57:47,850 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 21:57:48,584 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.412e+02 3.060e+02 4.409e+02 1.030e+03, threshold=6.120e+02, percent-clipped=9.0 +2023-02-06 21:57:53,569 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143482.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:57:59,612 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143491.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 21:58:02,441 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5309, 1.3266, 1.6041, 1.2474, 0.8827, 1.3586, 1.5467, 1.3840], + device='cuda:2'), covar=tensor([0.0569, 0.1331, 0.1729, 0.1487, 0.0602, 0.1557, 0.0707, 0.0681], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0099, 0.0161, 0.0113, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 21:58:06,561 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7574, 1.8754, 1.6457, 2.2966, 0.9794, 1.4821, 1.6744, 1.8844], + device='cuda:2'), covar=tensor([0.0725, 0.0806, 0.0926, 0.0399, 0.1109, 0.1245, 0.0783, 0.0740], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0199, 0.0250, 0.0213, 0.0208, 0.0247, 0.0254, 0.0214], + 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') +2023-02-06 21:58:13,404 INFO [train.py:901] (2/4) Epoch 18, batch 6100, loss[loss=0.1747, simple_loss=0.2556, pruned_loss=0.04692, over 7425.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2926, pruned_loss=0.06565, over 1614201.00 frames. ], batch size: 17, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:58:22,898 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7848, 4.6963, 4.3373, 2.8914, 4.2951, 4.3329, 4.4286, 4.0654], + device='cuda:2'), covar=tensor([0.0649, 0.0468, 0.0841, 0.3436, 0.0762, 0.0966, 0.1074, 0.0746], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0426, 0.0425, 0.0524, 0.0418, 0.0423, 0.0408, 0.0374], + 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') +2023-02-06 21:58:39,427 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 21:58:40,456 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.25 vs. limit=5.0 +2023-02-06 21:58:49,838 INFO [train.py:901] (2/4) Epoch 18, batch 6150, loss[loss=0.1988, simple_loss=0.2848, pruned_loss=0.05635, over 7652.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2926, pruned_loss=0.06541, over 1616714.10 frames. ], batch size: 19, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:59:00,210 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.359e+02 3.030e+02 3.820e+02 7.737e+02, threshold=6.061e+02, percent-clipped=3.0 +2023-02-06 21:59:25,629 INFO [train.py:901] (2/4) Epoch 18, batch 6200, loss[loss=0.2497, simple_loss=0.3345, pruned_loss=0.08246, over 8583.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2923, pruned_loss=0.06516, over 1612960.29 frames. ], batch size: 39, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:59:56,260 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2206, 1.5805, 1.2018, 2.4054, 1.0676, 1.1109, 1.8290, 1.7221], + device='cuda:2'), covar=tensor([0.1800, 0.1403, 0.2283, 0.0478, 0.1500, 0.2398, 0.0914, 0.1124], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0200, 0.0252, 0.0213, 0.0209, 0.0249, 0.0255, 0.0214], + 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') +2023-02-06 22:00:01,323 INFO [train.py:901] (2/4) Epoch 18, batch 6250, loss[loss=0.2165, simple_loss=0.2905, pruned_loss=0.0713, over 8581.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.293, pruned_loss=0.06533, over 1614501.02 frames. ], batch size: 31, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:00:07,092 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2100, 2.0478, 1.4789, 1.8899, 1.7109, 1.2924, 1.6037, 1.6347], + device='cuda:2'), covar=tensor([0.1261, 0.0451, 0.1256, 0.0579, 0.0754, 0.1542, 0.0950, 0.0808], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0234, 0.0325, 0.0304, 0.0295, 0.0330, 0.0341, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:00:12,419 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.460e+02 3.089e+02 4.040e+02 1.017e+03, threshold=6.178e+02, percent-clipped=5.0 +2023-02-06 22:00:20,715 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6616, 1.9579, 2.1545, 1.4160, 2.2489, 1.5319, 0.6382, 1.8778], + device='cuda:2'), covar=tensor([0.0557, 0.0290, 0.0221, 0.0511, 0.0325, 0.0782, 0.0801, 0.0257], + device='cuda:2'), in_proj_covar=tensor([0.0436, 0.0372, 0.0322, 0.0433, 0.0362, 0.0523, 0.0380, 0.0398], + device='cuda:2'), out_proj_covar=tensor([1.1826e-04, 9.8424e-05, 8.5340e-05, 1.1531e-04, 9.6311e-05, 1.4985e-04, + 1.0310e-04, 1.0596e-04], device='cuda:2') +2023-02-06 22:00:37,041 INFO [train.py:901] (2/4) Epoch 18, batch 6300, loss[loss=0.1698, simple_loss=0.2477, pruned_loss=0.04593, over 7420.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2925, pruned_loss=0.06534, over 1612370.55 frames. ], batch size: 17, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:01:11,903 INFO [train.py:901] (2/4) Epoch 18, batch 6350, loss[loss=0.2364, simple_loss=0.2945, pruned_loss=0.08918, over 7196.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2923, pruned_loss=0.0651, over 1616718.15 frames. ], batch size: 16, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:01:13,345 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143762.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:01:22,533 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.204e+02 2.882e+02 3.589e+02 6.333e+02, threshold=5.763e+02, percent-clipped=1.0 +2023-02-06 22:01:47,363 INFO [train.py:901] (2/4) Epoch 18, batch 6400, loss[loss=0.2046, simple_loss=0.2801, pruned_loss=0.06459, over 8476.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2915, pruned_loss=0.06495, over 1615159.36 frames. ], batch size: 27, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:01:49,682 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 22:01:55,547 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7027, 1.9358, 2.1258, 1.3885, 2.2803, 1.5047, 0.7240, 1.9312], + device='cuda:2'), covar=tensor([0.0702, 0.0383, 0.0307, 0.0594, 0.0456, 0.0886, 0.0834, 0.0293], + device='cuda:2'), in_proj_covar=tensor([0.0437, 0.0373, 0.0322, 0.0435, 0.0362, 0.0524, 0.0380, 0.0399], + device='cuda:2'), out_proj_covar=tensor([1.1853e-04, 9.8686e-05, 8.5205e-05, 1.1575e-04, 9.6397e-05, 1.5007e-04, + 1.0315e-04, 1.0630e-04], device='cuda:2') +2023-02-06 22:02:04,438 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143835.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:02:20,746 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9643, 1.7956, 2.5971, 1.6453, 2.2411, 2.8824, 2.8447, 2.5764], + device='cuda:2'), covar=tensor([0.0893, 0.1337, 0.0712, 0.1689, 0.1480, 0.0265, 0.0698, 0.0460], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0313, 0.0277, 0.0306, 0.0295, 0.0254, 0.0399, 0.0297], + 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') +2023-02-06 22:02:21,898 INFO [train.py:901] (2/4) Epoch 18, batch 6450, loss[loss=0.1803, simple_loss=0.2595, pruned_loss=0.0506, over 7661.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2925, pruned_loss=0.06542, over 1610748.55 frames. ], batch size: 19, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:02:33,457 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.452e+02 2.973e+02 3.704e+02 1.405e+03, threshold=5.946e+02, percent-clipped=1.0 +2023-02-06 22:02:34,290 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143877.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:02:39,646 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 22:02:57,285 INFO [train.py:901] (2/4) Epoch 18, batch 6500, loss[loss=0.2034, simple_loss=0.2975, pruned_loss=0.05461, over 8361.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2939, pruned_loss=0.06598, over 1615168.24 frames. ], batch size: 24, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:03:24,066 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143950.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:03:31,366 INFO [train.py:901] (2/4) Epoch 18, batch 6550, loss[loss=0.1651, simple_loss=0.2471, pruned_loss=0.04157, over 7542.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2932, pruned_loss=0.06565, over 1613289.21 frames. ], batch size: 18, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:03:38,658 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2549, 2.0061, 2.7096, 2.2699, 2.6318, 2.2051, 2.0030, 1.4978], + device='cuda:2'), covar=tensor([0.4559, 0.4310, 0.1686, 0.3101, 0.2128, 0.2825, 0.1802, 0.4591], + device='cuda:2'), in_proj_covar=tensor([0.0928, 0.0945, 0.0779, 0.0913, 0.0982, 0.0866, 0.0731, 0.0809], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:03:41,856 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.526e+02 3.154e+02 3.765e+02 8.734e+02, threshold=6.308e+02, percent-clipped=5.0 +2023-02-06 22:03:48,802 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 22:04:08,861 INFO [train.py:901] (2/4) Epoch 18, batch 6600, loss[loss=0.1892, simple_loss=0.2675, pruned_loss=0.0554, over 7970.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2926, pruned_loss=0.06515, over 1614212.45 frames. ], batch size: 21, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:04:10,888 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 22:04:25,261 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 22:04:34,008 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 +2023-02-06 22:04:43,665 INFO [train.py:901] (2/4) Epoch 18, batch 6650, loss[loss=0.2016, simple_loss=0.2908, pruned_loss=0.05621, over 8504.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2925, pruned_loss=0.06493, over 1615004.02 frames. ], batch size: 26, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:04:54,721 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.298e+02 3.022e+02 3.555e+02 7.360e+02, threshold=6.043e+02, percent-clipped=4.0 +2023-02-06 22:05:17,161 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5565, 1.8007, 1.9362, 1.1325, 1.9891, 1.4313, 0.4516, 1.8195], + device='cuda:2'), covar=tensor([0.0427, 0.0280, 0.0204, 0.0475, 0.0313, 0.0705, 0.0669, 0.0203], + device='cuda:2'), in_proj_covar=tensor([0.0438, 0.0375, 0.0323, 0.0436, 0.0364, 0.0524, 0.0381, 0.0401], + device='cuda:2'), out_proj_covar=tensor([1.1901e-04, 9.9063e-05, 8.5492e-05, 1.1602e-04, 9.7010e-05, 1.5011e-04, + 1.0345e-04, 1.0684e-04], device='cuda:2') +2023-02-06 22:05:19,670 INFO [train.py:901] (2/4) Epoch 18, batch 6700, loss[loss=0.1789, simple_loss=0.2497, pruned_loss=0.05401, over 7707.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2904, pruned_loss=0.0637, over 1615851.33 frames. ], batch size: 18, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:05:34,398 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144133.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:05:51,948 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144158.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:05:53,841 INFO [train.py:901] (2/4) Epoch 18, batch 6750, loss[loss=0.2682, simple_loss=0.3335, pruned_loss=0.1014, over 6847.00 frames. ], tot_loss[loss=0.211, simple_loss=0.292, pruned_loss=0.06495, over 1615969.77 frames. ], batch size: 71, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:06:03,939 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.293e+02 3.003e+02 3.717e+02 7.578e+02, threshold=6.007e+02, percent-clipped=1.0 +2023-02-06 22:06:04,345 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 22:06:19,927 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1718, 4.1255, 3.7897, 1.9855, 3.6736, 3.8589, 3.6626, 3.5711], + device='cuda:2'), covar=tensor([0.0835, 0.0622, 0.1053, 0.4847, 0.0935, 0.1028, 0.1313, 0.0794], + device='cuda:2'), in_proj_covar=tensor([0.0517, 0.0428, 0.0429, 0.0529, 0.0420, 0.0426, 0.0413, 0.0376], + 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') +2023-02-06 22:06:25,432 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144206.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:06:27,290 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 22:06:28,560 INFO [train.py:901] (2/4) Epoch 18, batch 6800, loss[loss=0.3044, simple_loss=0.3552, pruned_loss=0.1268, over 7393.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2927, pruned_loss=0.06555, over 1615650.22 frames. ], batch size: 73, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:06:42,850 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144231.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:06:51,121 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 22:07:03,973 INFO [train.py:901] (2/4) Epoch 18, batch 6850, loss[loss=0.2012, simple_loss=0.2802, pruned_loss=0.06106, over 8090.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2929, pruned_loss=0.06604, over 1612455.26 frames. ], batch size: 21, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:07:13,993 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.379e+02 2.937e+02 3.634e+02 6.722e+02, threshold=5.873e+02, percent-clipped=2.0 +2023-02-06 22:07:17,341 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 22:07:22,930 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9289, 3.7843, 2.3687, 2.9499, 2.8142, 2.1815, 2.6969, 2.9231], + device='cuda:2'), covar=tensor([0.1958, 0.0401, 0.1226, 0.0765, 0.0839, 0.1476, 0.1223, 0.1131], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0237, 0.0328, 0.0306, 0.0300, 0.0332, 0.0345, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:07:38,060 INFO [train.py:901] (2/4) Epoch 18, batch 6900, loss[loss=0.238, simple_loss=0.308, pruned_loss=0.08402, over 7427.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2921, pruned_loss=0.06594, over 1607756.99 frames. ], batch size: 17, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:08:09,502 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0475, 1.6947, 6.0918, 2.1076, 5.5266, 5.1637, 5.7347, 5.6025], + device='cuda:2'), covar=tensor([0.0438, 0.4377, 0.0335, 0.3607, 0.0947, 0.0827, 0.0437, 0.0453], + device='cuda:2'), in_proj_covar=tensor([0.0593, 0.0628, 0.0670, 0.0600, 0.0679, 0.0581, 0.0578, 0.0647], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:08:13,481 INFO [train.py:901] (2/4) Epoch 18, batch 6950, loss[loss=0.1838, simple_loss=0.2614, pruned_loss=0.05307, over 7927.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2915, pruned_loss=0.06526, over 1611653.69 frames. ], batch size: 20, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:08:20,470 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 22:08:24,077 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.398e+02 2.919e+02 3.864e+02 7.610e+02, threshold=5.839e+02, percent-clipped=3.0 +2023-02-06 22:08:25,463 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 22:08:47,780 INFO [train.py:901] (2/4) Epoch 18, batch 7000, loss[loss=0.2158, simple_loss=0.2973, pruned_loss=0.06713, over 8086.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2918, pruned_loss=0.06522, over 1616321.36 frames. ], batch size: 21, lr: 4.18e-03, grad_scale: 16.0 +2023-02-06 22:08:49,927 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8080, 1.4702, 3.9718, 1.3947, 3.5234, 3.2395, 3.6039, 3.4815], + device='cuda:2'), covar=tensor([0.0656, 0.3879, 0.0585, 0.3755, 0.1102, 0.1004, 0.0633, 0.0703], + device='cuda:2'), in_proj_covar=tensor([0.0596, 0.0632, 0.0676, 0.0603, 0.0685, 0.0584, 0.0582, 0.0652], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:09:01,097 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144429.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:09:05,729 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0953, 1.7321, 3.4140, 1.5840, 2.3406, 3.7470, 3.8883, 3.2205], + device='cuda:2'), covar=tensor([0.1120, 0.1680, 0.0334, 0.2110, 0.1104, 0.0218, 0.0491, 0.0546], + device='cuda:2'), in_proj_covar=tensor([0.0283, 0.0314, 0.0278, 0.0307, 0.0295, 0.0255, 0.0400, 0.0298], + 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') +2023-02-06 22:09:22,578 INFO [train.py:901] (2/4) Epoch 18, batch 7050, loss[loss=0.1988, simple_loss=0.2757, pruned_loss=0.06092, over 7660.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2913, pruned_loss=0.06514, over 1613292.73 frames. ], batch size: 19, lr: 4.18e-03, grad_scale: 16.0 +2023-02-06 22:09:34,230 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.214e+02 2.534e+02 2.937e+02 3.689e+02 8.247e+02, threshold=5.874e+02, percent-clipped=3.0 +2023-02-06 22:09:58,438 INFO [train.py:901] (2/4) Epoch 18, batch 7100, loss[loss=0.2491, simple_loss=0.3349, pruned_loss=0.0817, over 8512.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2917, pruned_loss=0.06546, over 1607641.57 frames. ], batch size: 26, lr: 4.18e-03, grad_scale: 16.0 +2023-02-06 22:10:33,591 INFO [train.py:901] (2/4) Epoch 18, batch 7150, loss[loss=0.1618, simple_loss=0.2394, pruned_loss=0.04216, over 6352.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2917, pruned_loss=0.06493, over 1608215.68 frames. ], batch size: 14, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:10:43,980 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.496e+02 2.263e+02 2.906e+02 3.662e+02 1.305e+03, threshold=5.813e+02, percent-clipped=7.0 +2023-02-06 22:11:00,102 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.25 vs. limit=5.0 +2023-02-06 22:11:10,031 INFO [train.py:901] (2/4) Epoch 18, batch 7200, loss[loss=0.1962, simple_loss=0.2811, pruned_loss=0.05563, over 8044.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2916, pruned_loss=0.06506, over 1607247.18 frames. ], batch size: 22, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:11:12,572 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 22:11:20,762 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7859, 1.3959, 1.6371, 1.1427, 0.9610, 1.3605, 1.6775, 1.3608], + device='cuda:2'), covar=tensor([0.0568, 0.1325, 0.1712, 0.1580, 0.0603, 0.1586, 0.0707, 0.0714], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0151, 0.0190, 0.0157, 0.0099, 0.0161, 0.0113, 0.0139], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 22:11:44,457 INFO [train.py:901] (2/4) Epoch 18, batch 7250, loss[loss=0.1912, simple_loss=0.2694, pruned_loss=0.05647, over 7714.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2927, pruned_loss=0.0658, over 1607820.22 frames. ], batch size: 18, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:11:54,469 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.379e+02 2.816e+02 3.627e+02 9.857e+02, threshold=5.632e+02, percent-clipped=4.0 +2023-02-06 22:12:19,769 INFO [train.py:901] (2/4) Epoch 18, batch 7300, loss[loss=0.2178, simple_loss=0.2957, pruned_loss=0.06995, over 8707.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2927, pruned_loss=0.06563, over 1607150.05 frames. ], batch size: 34, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:12:50,206 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0358, 3.8722, 2.3082, 3.0713, 2.8679, 1.9540, 2.9448, 2.9963], + device='cuda:2'), covar=tensor([0.1614, 0.0256, 0.1181, 0.0687, 0.0763, 0.1502, 0.1053, 0.1012], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0233, 0.0324, 0.0301, 0.0294, 0.0328, 0.0340, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:12:53,982 INFO [train.py:901] (2/4) Epoch 18, batch 7350, loss[loss=0.1759, simple_loss=0.2554, pruned_loss=0.04818, over 8085.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2923, pruned_loss=0.06569, over 1607190.13 frames. ], batch size: 21, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:13:02,928 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=144773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:13:04,748 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.484e+02 2.992e+02 3.514e+02 8.978e+02, threshold=5.985e+02, percent-clipped=6.0 +2023-02-06 22:13:08,073 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 22:13:26,911 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 22:13:28,896 INFO [train.py:901] (2/4) Epoch 18, batch 7400, loss[loss=0.2055, simple_loss=0.2864, pruned_loss=0.06234, over 7815.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2932, pruned_loss=0.06652, over 1608718.73 frames. ], batch size: 20, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:14:04,302 INFO [train.py:901] (2/4) Epoch 18, batch 7450, loss[loss=0.2547, simple_loss=0.3401, pruned_loss=0.08462, over 8106.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2948, pruned_loss=0.06718, over 1610299.83 frames. ], batch size: 23, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:14:07,791 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 22:14:07,987 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9929, 3.8597, 2.2421, 2.9190, 2.9673, 2.0455, 3.0104, 2.9524], + device='cuda:2'), covar=tensor([0.1592, 0.0253, 0.0937, 0.0607, 0.0583, 0.1216, 0.0869, 0.0964], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0233, 0.0324, 0.0302, 0.0295, 0.0329, 0.0342, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:14:14,577 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.437e+02 2.433e+02 3.083e+02 4.140e+02 9.921e+02, threshold=6.167e+02, percent-clipped=3.0 +2023-02-06 22:14:23,398 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144888.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:14:38,529 INFO [train.py:901] (2/4) Epoch 18, batch 7500, loss[loss=0.2062, simple_loss=0.2905, pruned_loss=0.06096, over 8454.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2934, pruned_loss=0.06649, over 1603376.62 frames. ], batch size: 25, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:15:14,116 INFO [train.py:901] (2/4) Epoch 18, batch 7550, loss[loss=0.2021, simple_loss=0.2863, pruned_loss=0.05895, over 8109.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2923, pruned_loss=0.06576, over 1603882.52 frames. ], batch size: 23, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:15:24,766 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.361e+02 2.893e+02 3.293e+02 8.578e+02, threshold=5.785e+02, percent-clipped=2.0 +2023-02-06 22:15:48,827 INFO [train.py:901] (2/4) Epoch 18, batch 7600, loss[loss=0.2016, simple_loss=0.2803, pruned_loss=0.06149, over 7967.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2926, pruned_loss=0.06599, over 1607133.99 frames. ], batch size: 21, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:15:51,058 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145014.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:16:12,837 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:16:24,429 INFO [train.py:901] (2/4) Epoch 18, batch 7650, loss[loss=0.1943, simple_loss=0.2825, pruned_loss=0.05308, over 8466.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2922, pruned_loss=0.06541, over 1608558.15 frames. ], batch size: 25, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:16:35,693 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.409e+02 3.204e+02 3.806e+02 7.453e+02, threshold=6.408e+02, percent-clipped=5.0 +2023-02-06 22:16:40,599 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:16:57,521 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9305, 2.5816, 3.5415, 2.0637, 2.0824, 3.5542, 0.7398, 2.0961], + device='cuda:2'), covar=tensor([0.1428, 0.1591, 0.0404, 0.1894, 0.2717, 0.0310, 0.2543, 0.1579], + device='cuda:2'), in_proj_covar=tensor([0.0182, 0.0191, 0.0121, 0.0215, 0.0265, 0.0129, 0.0166, 0.0183], + 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') +2023-02-06 22:16:58,566 INFO [train.py:901] (2/4) Epoch 18, batch 7700, loss[loss=0.2315, simple_loss=0.3165, pruned_loss=0.07324, over 8102.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2927, pruned_loss=0.06584, over 1610484.63 frames. ], batch size: 23, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:17:16,300 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 22:17:17,759 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2392, 2.5372, 2.9193, 1.5480, 3.1826, 1.8196, 1.4815, 2.0313], + device='cuda:2'), covar=tensor([0.0678, 0.0347, 0.0269, 0.0736, 0.0398, 0.0773, 0.0846, 0.0501], + device='cuda:2'), in_proj_covar=tensor([0.0438, 0.0377, 0.0323, 0.0431, 0.0362, 0.0523, 0.0379, 0.0403], + device='cuda:2'), out_proj_covar=tensor([1.1892e-04, 9.9597e-05, 8.5574e-05, 1.1453e-04, 9.6450e-05, 1.4958e-04, + 1.0290e-04, 1.0766e-04], device='cuda:2') +2023-02-06 22:17:21,899 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145144.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:17:33,670 INFO [train.py:901] (2/4) Epoch 18, batch 7750, loss[loss=0.2518, simple_loss=0.3172, pruned_loss=0.09317, over 8704.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2925, pruned_loss=0.06609, over 1610093.90 frames. ], batch size: 40, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:17:40,047 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145169.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:17:45,202 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.617e+02 3.101e+02 3.765e+02 9.296e+02, threshold=6.202e+02, percent-clipped=3.0 +2023-02-06 22:18:08,798 INFO [train.py:901] (2/4) Epoch 18, batch 7800, loss[loss=0.1829, simple_loss=0.2742, pruned_loss=0.04577, over 7969.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2928, pruned_loss=0.06619, over 1609113.77 frames. ], batch size: 21, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:18:42,838 INFO [train.py:901] (2/4) Epoch 18, batch 7850, loss[loss=0.1798, simple_loss=0.2556, pruned_loss=0.05199, over 7447.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2927, pruned_loss=0.0661, over 1611292.48 frames. ], batch size: 17, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:18:53,262 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.506e+02 2.477e+02 2.948e+02 3.643e+02 1.044e+03, threshold=5.895e+02, percent-clipped=9.0 +2023-02-06 22:19:16,123 INFO [train.py:901] (2/4) Epoch 18, batch 7900, loss[loss=0.1824, simple_loss=0.2712, pruned_loss=0.04676, over 8473.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2923, pruned_loss=0.06543, over 1613485.11 frames. ], batch size: 25, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:19:47,151 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145358.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:19:49,081 INFO [train.py:901] (2/4) Epoch 18, batch 7950, loss[loss=0.1921, simple_loss=0.2813, pruned_loss=0.05147, over 8593.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2911, pruned_loss=0.06514, over 1611013.93 frames. ], batch size: 39, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:19:59,837 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.389e+02 3.012e+02 3.869e+02 1.111e+03, threshold=6.025e+02, percent-clipped=3.0 +2023-02-06 22:20:04,931 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 +2023-02-06 22:20:07,939 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:20:08,007 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:20:23,113 INFO [train.py:901] (2/4) Epoch 18, batch 8000, loss[loss=0.1646, simple_loss=0.2506, pruned_loss=0.03925, over 7922.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2903, pruned_loss=0.06448, over 1608857.16 frames. ], batch size: 20, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:20:25,591 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 22:20:31,455 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8700, 2.3414, 4.3476, 1.6325, 3.2322, 2.4145, 1.8774, 3.0634], + device='cuda:2'), covar=tensor([0.1797, 0.2551, 0.0787, 0.4213, 0.1618, 0.2890, 0.2158, 0.2190], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0578, 0.0548, 0.0625, 0.0638, 0.0583, 0.0515, 0.0628], + 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') +2023-02-06 22:20:34,670 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:20:57,108 INFO [train.py:901] (2/4) Epoch 18, batch 8050, loss[loss=0.2216, simple_loss=0.2957, pruned_loss=0.07371, over 7541.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2895, pruned_loss=0.06513, over 1587677.61 frames. ], batch size: 18, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:21:05,672 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145473.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:21:08,160 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.350e+02 2.866e+02 3.408e+02 5.747e+02, threshold=5.732e+02, percent-clipped=0.0 +2023-02-06 22:21:29,480 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 22:21:34,901 INFO [train.py:901] (2/4) Epoch 19, batch 0, loss[loss=0.1762, simple_loss=0.2647, pruned_loss=0.04381, over 8089.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2647, pruned_loss=0.04381, over 8089.00 frames. ], batch size: 21, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:21:34,901 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 22:21:46,549 INFO [train.py:935] (2/4) Epoch 19, validation: loss=0.1782, simple_loss=0.2779, pruned_loss=0.03928, over 944034.00 frames. +2023-02-06 22:21:46,550 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 22:21:54,197 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:22:03,067 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 22:22:22,457 INFO [train.py:901] (2/4) Epoch 19, batch 50, loss[loss=0.1894, simple_loss=0.2651, pruned_loss=0.05688, over 7554.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2861, pruned_loss=0.0623, over 364238.99 frames. ], batch size: 18, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:22:22,669 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:22:23,349 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9496, 1.5742, 3.3026, 1.3906, 2.4622, 3.6057, 3.7523, 3.0501], + device='cuda:2'), covar=tensor([0.1166, 0.1741, 0.0353, 0.2218, 0.0980, 0.0244, 0.0456, 0.0567], + device='cuda:2'), in_proj_covar=tensor([0.0281, 0.0312, 0.0275, 0.0303, 0.0293, 0.0253, 0.0396, 0.0296], + 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') +2023-02-06 22:22:25,796 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 22:22:40,524 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 22:22:42,317 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-06 22:22:45,196 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.340e+02 2.977e+02 3.641e+02 7.952e+02, threshold=5.953e+02, percent-clipped=6.0 +2023-02-06 22:22:56,254 INFO [train.py:901] (2/4) Epoch 19, batch 100, loss[loss=0.1994, simple_loss=0.2818, pruned_loss=0.05852, over 8246.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2907, pruned_loss=0.06546, over 642250.73 frames. ], batch size: 22, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:22:59,376 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6809, 2.0841, 3.3634, 1.5783, 2.5349, 2.0626, 1.7920, 2.4940], + device='cuda:2'), covar=tensor([0.1791, 0.2605, 0.0743, 0.4393, 0.1726, 0.3153, 0.2188, 0.2150], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0578, 0.0548, 0.0628, 0.0635, 0.0585, 0.0517, 0.0629], + 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') +2023-02-06 22:23:01,889 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 22:23:10,099 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145612.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:23:20,686 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7599, 4.6848, 4.1966, 2.0033, 4.1711, 4.3046, 4.3147, 4.0384], + device='cuda:2'), covar=tensor([0.0608, 0.0467, 0.0957, 0.4629, 0.0782, 0.0881, 0.1042, 0.0737], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0426, 0.0429, 0.0527, 0.0414, 0.0426, 0.0408, 0.0373], + 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') +2023-02-06 22:23:32,319 INFO [train.py:901] (2/4) Epoch 19, batch 150, loss[loss=0.1722, simple_loss=0.25, pruned_loss=0.04714, over 7429.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2912, pruned_loss=0.06427, over 860716.80 frames. ], batch size: 17, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:23:40,877 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0631, 1.7612, 2.4042, 1.9539, 2.2270, 2.0678, 1.8299, 1.1677], + device='cuda:2'), covar=tensor([0.5237, 0.4760, 0.1727, 0.3359, 0.2392, 0.2973, 0.1990, 0.4819], + device='cuda:2'), in_proj_covar=tensor([0.0925, 0.0945, 0.0773, 0.0911, 0.0976, 0.0864, 0.0728, 0.0808], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:23:46,186 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:23:57,040 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.454e+02 2.969e+02 3.777e+02 1.176e+03, threshold=5.938e+02, percent-clipped=4.0 +2023-02-06 22:24:07,965 INFO [train.py:901] (2/4) Epoch 19, batch 200, loss[loss=0.1813, simple_loss=0.2692, pruned_loss=0.04663, over 8087.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2921, pruned_loss=0.06378, over 1032203.45 frames. ], batch size: 21, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:24:33,094 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4402, 2.5029, 1.7938, 2.1509, 2.1339, 1.4192, 2.0260, 1.9669], + device='cuda:2'), covar=tensor([0.1598, 0.0397, 0.1158, 0.0624, 0.0681, 0.1601, 0.1021, 0.1010], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0234, 0.0323, 0.0301, 0.0296, 0.0328, 0.0339, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:24:33,110 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145729.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:35,723 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:43,201 INFO [train.py:901] (2/4) Epoch 19, batch 250, loss[loss=0.2447, simple_loss=0.322, pruned_loss=0.08372, over 8291.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2944, pruned_loss=0.06527, over 1163951.14 frames. ], batch size: 23, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:24:51,128 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:55,242 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:58,390 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 22:25:06,977 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 22:25:07,541 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.432e+02 3.022e+02 3.893e+02 7.688e+02, threshold=6.043e+02, percent-clipped=6.0 +2023-02-06 22:25:13,284 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:25:18,656 INFO [train.py:901] (2/4) Epoch 19, batch 300, loss[loss=0.1918, simple_loss=0.2716, pruned_loss=0.05604, over 8226.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.294, pruned_loss=0.06559, over 1266303.55 frames. ], batch size: 22, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:25:22,942 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:25:39,976 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:25:50,254 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145839.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:25:53,699 INFO [train.py:901] (2/4) Epoch 19, batch 350, loss[loss=0.2163, simple_loss=0.3035, pruned_loss=0.06453, over 8237.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2943, pruned_loss=0.06608, over 1343727.95 frames. ], batch size: 22, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:25:57,444 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:26:17,667 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.383e+02 2.952e+02 3.795e+02 9.100e+02, threshold=5.904e+02, percent-clipped=6.0 +2023-02-06 22:26:30,026 INFO [train.py:901] (2/4) Epoch 19, batch 400, loss[loss=0.2193, simple_loss=0.3043, pruned_loss=0.06712, over 8102.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2948, pruned_loss=0.06653, over 1405288.78 frames. ], batch size: 23, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:26:32,897 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.4245, 5.5156, 4.9408, 2.5376, 4.9571, 5.1564, 5.1168, 4.8917], + device='cuda:2'), covar=tensor([0.0688, 0.0450, 0.0910, 0.4491, 0.0704, 0.0868, 0.1075, 0.0680], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0422, 0.0425, 0.0524, 0.0411, 0.0424, 0.0402, 0.0370], + 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') +2023-02-06 22:26:44,768 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0829, 1.7574, 2.0331, 1.7669, 1.3531, 1.8067, 2.5077, 2.0546], + device='cuda:2'), covar=tensor([0.0416, 0.1132, 0.1551, 0.1278, 0.0524, 0.1339, 0.0547, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0157, 0.0099, 0.0161, 0.0112, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 22:27:04,031 INFO [train.py:901] (2/4) Epoch 19, batch 450, loss[loss=0.1984, simple_loss=0.2957, pruned_loss=0.05058, over 8475.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2962, pruned_loss=0.06683, over 1455804.32 frames. ], batch size: 25, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:27:12,889 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145956.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:27:23,353 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.09 vs. limit=5.0 +2023-02-06 22:27:28,514 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.474e+02 2.839e+02 3.457e+02 5.406e+02, threshold=5.677e+02, percent-clipped=0.0 +2023-02-06 22:27:40,173 INFO [train.py:901] (2/4) Epoch 19, batch 500, loss[loss=0.1959, simple_loss=0.2752, pruned_loss=0.05829, over 8240.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2948, pruned_loss=0.06572, over 1494681.99 frames. ], batch size: 22, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:27:50,093 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:28:03,430 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4279, 1.5097, 4.5809, 1.6743, 4.1090, 3.7570, 4.1663, 3.9965], + device='cuda:2'), covar=tensor([0.0549, 0.4464, 0.0457, 0.4053, 0.1002, 0.0946, 0.0553, 0.0634], + device='cuda:2'), in_proj_covar=tensor([0.0597, 0.0627, 0.0673, 0.0605, 0.0687, 0.0591, 0.0585, 0.0649], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 22:28:15,843 INFO [train.py:901] (2/4) Epoch 19, batch 550, loss[loss=0.1921, simple_loss=0.2844, pruned_loss=0.04989, over 8282.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2926, pruned_loss=0.06463, over 1517654.32 frames. ], batch size: 23, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:28:19,299 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4495, 4.4076, 3.9937, 1.9643, 3.9660, 4.0754, 4.0242, 3.7291], + device='cuda:2'), covar=tensor([0.0659, 0.0547, 0.1118, 0.4599, 0.0784, 0.0815, 0.1216, 0.0752], + device='cuda:2'), in_proj_covar=tensor([0.0519, 0.0428, 0.0430, 0.0530, 0.0416, 0.0430, 0.0411, 0.0375], + 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') +2023-02-06 22:28:35,110 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146071.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:28:38,925 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.557e+02 3.049e+02 4.000e+02 8.642e+02, threshold=6.099e+02, percent-clipped=4.0 +2023-02-06 22:28:50,769 INFO [train.py:901] (2/4) Epoch 19, batch 600, loss[loss=0.236, simple_loss=0.3124, pruned_loss=0.07983, over 8735.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2933, pruned_loss=0.06558, over 1540410.47 frames. ], batch size: 30, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:28:54,562 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7888, 1.5575, 3.9856, 1.4170, 3.5386, 3.3258, 3.6068, 3.4798], + device='cuda:2'), covar=tensor([0.0639, 0.4122, 0.0630, 0.3879, 0.1250, 0.1065, 0.0656, 0.0766], + device='cuda:2'), in_proj_covar=tensor([0.0594, 0.0624, 0.0670, 0.0600, 0.0684, 0.0588, 0.0582, 0.0646], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:28:59,454 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:03,469 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6236, 4.5942, 4.2057, 2.4450, 4.1176, 4.1790, 4.2963, 3.9827], + device='cuda:2'), covar=tensor([0.0714, 0.0484, 0.0867, 0.3865, 0.0788, 0.0975, 0.1065, 0.0753], + device='cuda:2'), in_proj_covar=tensor([0.0517, 0.0425, 0.0427, 0.0526, 0.0413, 0.0427, 0.0407, 0.0373], + 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') +2023-02-06 22:29:11,437 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 22:29:11,612 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:17,650 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:26,685 INFO [train.py:901] (2/4) Epoch 19, batch 650, loss[loss=0.1701, simple_loss=0.2606, pruned_loss=0.03976, over 7814.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2937, pruned_loss=0.06565, over 1559603.48 frames. ], batch size: 20, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:29:42,929 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:49,759 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.628e+02 2.995e+02 3.912e+02 8.872e+02, threshold=5.991e+02, percent-clipped=7.0 +2023-02-06 22:29:53,934 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146183.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:30:00,627 INFO [train.py:901] (2/4) Epoch 19, batch 700, loss[loss=0.2176, simple_loss=0.2983, pruned_loss=0.06843, over 8576.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.294, pruned_loss=0.06614, over 1568904.45 frames. ], batch size: 31, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:30:37,726 INFO [train.py:901] (2/4) Epoch 19, batch 750, loss[loss=0.248, simple_loss=0.3165, pruned_loss=0.08976, over 8607.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2927, pruned_loss=0.0655, over 1582737.29 frames. ], batch size: 39, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:30:58,055 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 22:31:00,739 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.187e+02 2.733e+02 3.387e+02 1.037e+03, threshold=5.466e+02, percent-clipped=4.0 +2023-02-06 22:31:06,860 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 22:31:11,497 INFO [train.py:901] (2/4) Epoch 19, batch 800, loss[loss=0.1978, simple_loss=0.2939, pruned_loss=0.05086, over 8035.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2938, pruned_loss=0.06658, over 1590161.72 frames. ], batch size: 22, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:31:14,849 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146298.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:31:15,523 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0693, 1.7646, 2.0273, 1.7231, 1.0895, 1.8121, 2.3063, 2.2282], + device='cuda:2'), covar=tensor([0.0401, 0.1200, 0.1529, 0.1327, 0.0535, 0.1384, 0.0574, 0.0584], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0099, 0.0162, 0.0113, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 22:31:17,171 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.61 vs. limit=5.0 +2023-02-06 22:31:35,749 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:31:47,235 INFO [train.py:901] (2/4) Epoch 19, batch 850, loss[loss=0.1824, simple_loss=0.2683, pruned_loss=0.04828, over 7517.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2919, pruned_loss=0.06534, over 1595100.24 frames. ], batch size: 18, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:31:54,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146352.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:32:10,865 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:32:11,294 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.470e+02 3.071e+02 3.941e+02 1.675e+03, threshold=6.141e+02, percent-clipped=6.0 +2023-02-06 22:32:22,255 INFO [train.py:901] (2/4) Epoch 19, batch 900, loss[loss=0.2364, simple_loss=0.3134, pruned_loss=0.07968, over 8110.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2933, pruned_loss=0.06579, over 1606684.21 frames. ], batch size: 23, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:32:27,822 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146401.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:32:41,230 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8052, 1.9935, 3.3727, 1.5127, 2.5288, 2.1117, 1.8703, 2.3178], + device='cuda:2'), covar=tensor([0.1630, 0.2287, 0.0733, 0.4085, 0.1638, 0.2905, 0.1906, 0.2182], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0580, 0.0550, 0.0630, 0.0639, 0.0584, 0.0518, 0.0629], + 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') +2023-02-06 22:32:56,375 INFO [train.py:901] (2/4) Epoch 19, batch 950, loss[loss=0.2381, simple_loss=0.2949, pruned_loss=0.0907, over 7981.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2926, pruned_loss=0.06586, over 1607039.45 frames. ], batch size: 21, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:33:09,723 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 +2023-02-06 22:33:20,833 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8952, 1.6551, 1.9726, 1.5869, 1.0220, 1.6977, 1.9889, 2.0191], + device='cuda:2'), covar=tensor([0.0430, 0.1228, 0.1555, 0.1369, 0.0617, 0.1449, 0.0672, 0.0603], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0157, 0.0099, 0.0162, 0.0112, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 22:33:21,310 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.324e+02 2.987e+02 4.077e+02 9.877e+02, threshold=5.974e+02, percent-clipped=4.0 +2023-02-06 22:33:22,690 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 22:33:32,153 INFO [train.py:901] (2/4) Epoch 19, batch 1000, loss[loss=0.2121, simple_loss=0.2879, pruned_loss=0.0682, over 7059.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2922, pruned_loss=0.06523, over 1613853.96 frames. ], batch size: 71, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:33:44,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146511.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:33:54,571 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 22:34:06,369 INFO [train.py:901] (2/4) Epoch 19, batch 1050, loss[loss=0.243, simple_loss=0.3219, pruned_loss=0.08202, over 8510.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2931, pruned_loss=0.0653, over 1620929.57 frames. ], batch size: 28, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:34:06,388 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 22:34:14,962 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146554.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:34:31,583 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.403e+02 2.837e+02 3.508e+02 6.242e+02, threshold=5.674e+02, percent-clipped=1.0 +2023-02-06 22:34:34,619 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146579.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:34:44,087 INFO [train.py:901] (2/4) Epoch 19, batch 1100, loss[loss=0.2235, simple_loss=0.3111, pruned_loss=0.068, over 8481.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2928, pruned_loss=0.06508, over 1621059.39 frames. ], batch size: 25, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:34:55,189 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:34:56,651 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4709, 2.2399, 3.2797, 2.4759, 2.9828, 2.4520, 2.2367, 1.7434], + device='cuda:2'), covar=tensor([0.5249, 0.4906, 0.1773, 0.3751, 0.2603, 0.2821, 0.1700, 0.5695], + device='cuda:2'), in_proj_covar=tensor([0.0931, 0.0948, 0.0782, 0.0913, 0.0978, 0.0866, 0.0726, 0.0809], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:35:06,981 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146626.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:35:18,380 INFO [train.py:901] (2/4) Epoch 19, batch 1150, loss[loss=0.1843, simple_loss=0.2563, pruned_loss=0.05619, over 7219.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2926, pruned_loss=0.0654, over 1618947.43 frames. ], batch size: 16, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:35:19,116 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 22:35:19,275 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:35:42,420 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.484e+02 2.879e+02 3.755e+02 5.922e+02, threshold=5.758e+02, percent-clipped=3.0 +2023-02-06 22:35:53,864 INFO [train.py:901] (2/4) Epoch 19, batch 1200, loss[loss=0.2295, simple_loss=0.3149, pruned_loss=0.07211, over 8496.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2926, pruned_loss=0.06581, over 1616702.95 frames. ], batch size: 26, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:36:03,787 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=5.29 vs. limit=5.0 +2023-02-06 22:36:28,998 INFO [train.py:901] (2/4) Epoch 19, batch 1250, loss[loss=0.1891, simple_loss=0.2618, pruned_loss=0.05823, over 7430.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2909, pruned_loss=0.06529, over 1612419.33 frames. ], batch size: 17, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:36:52,646 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.471e+02 2.976e+02 4.092e+02 7.603e+02, threshold=5.951e+02, percent-clipped=4.0 +2023-02-06 22:37:04,304 INFO [train.py:901] (2/4) Epoch 19, batch 1300, loss[loss=0.2124, simple_loss=0.3, pruned_loss=0.0624, over 8630.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2898, pruned_loss=0.06478, over 1608150.94 frames. ], batch size: 34, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:37:40,717 INFO [train.py:901] (2/4) Epoch 19, batch 1350, loss[loss=0.1792, simple_loss=0.2538, pruned_loss=0.05229, over 7242.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2901, pruned_loss=0.06456, over 1607545.03 frames. ], batch size: 16, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:37:53,751 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146862.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:38:03,884 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.302e+02 2.844e+02 3.659e+02 6.626e+02, threshold=5.688e+02, percent-clipped=1.0 +2023-02-06 22:38:07,775 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146882.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:38:15,247 INFO [train.py:901] (2/4) Epoch 19, batch 1400, loss[loss=0.2648, simple_loss=0.3269, pruned_loss=0.1013, over 8710.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2904, pruned_loss=0.06495, over 1608586.48 frames. ], batch size: 39, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:38:25,960 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:38:52,632 INFO [train.py:901] (2/4) Epoch 19, batch 1450, loss[loss=0.1921, simple_loss=0.2515, pruned_loss=0.06634, over 7708.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.291, pruned_loss=0.0654, over 1614923.68 frames. ], batch size: 18, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:38:56,664 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 22:38:59,402 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146953.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:39:16,185 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.362e+02 2.962e+02 3.993e+02 1.525e+03, threshold=5.923e+02, percent-clipped=6.0 +2023-02-06 22:39:17,114 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7300, 2.4726, 3.5439, 2.7400, 3.4107, 2.6903, 2.3675, 2.2123], + device='cuda:2'), covar=tensor([0.4596, 0.4569, 0.1443, 0.3585, 0.2211, 0.2666, 0.1732, 0.4970], + device='cuda:2'), in_proj_covar=tensor([0.0927, 0.0943, 0.0777, 0.0905, 0.0972, 0.0860, 0.0722, 0.0802], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:39:20,639 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4680, 1.5341, 2.0694, 1.2390, 1.3912, 1.6302, 1.4489, 1.2915], + device='cuda:2'), covar=tensor([0.1971, 0.2506, 0.1055, 0.4659, 0.2072, 0.3448, 0.2342, 0.2439], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0581, 0.0551, 0.0628, 0.0637, 0.0586, 0.0518, 0.0630], + 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') +2023-02-06 22:39:23,967 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146988.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:39:27,300 INFO [train.py:901] (2/4) Epoch 19, batch 1500, loss[loss=0.2027, simple_loss=0.2645, pruned_loss=0.07045, over 7251.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2904, pruned_loss=0.06503, over 1612911.13 frames. ], batch size: 16, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:39:50,166 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8808, 3.8606, 3.4789, 1.6354, 3.4177, 3.4238, 3.4639, 3.2747], + device='cuda:2'), covar=tensor([0.0907, 0.0642, 0.1233, 0.5027, 0.0997, 0.1093, 0.1369, 0.0897], + device='cuda:2'), in_proj_covar=tensor([0.0514, 0.0422, 0.0427, 0.0524, 0.0415, 0.0426, 0.0407, 0.0374], + 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') +2023-02-06 22:39:53,154 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 22:40:03,265 INFO [train.py:901] (2/4) Epoch 19, batch 1550, loss[loss=0.2395, simple_loss=0.3203, pruned_loss=0.07931, over 8526.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.292, pruned_loss=0.06584, over 1615913.80 frames. ], batch size: 31, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:40:22,638 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147068.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:40:28,398 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.434e+02 2.984e+02 3.600e+02 8.495e+02, threshold=5.968e+02, percent-clipped=1.0 +2023-02-06 22:40:39,456 INFO [train.py:901] (2/4) Epoch 19, batch 1600, loss[loss=0.2004, simple_loss=0.2823, pruned_loss=0.05926, over 8037.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2926, pruned_loss=0.06552, over 1617226.84 frames. ], batch size: 22, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:40:46,380 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:41:14,548 INFO [train.py:901] (2/4) Epoch 19, batch 1650, loss[loss=0.2483, simple_loss=0.3281, pruned_loss=0.08423, over 8482.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2928, pruned_loss=0.06583, over 1616250.79 frames. ], batch size: 28, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:41:40,975 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.588e+02 2.354e+02 2.709e+02 3.474e+02 7.081e+02, threshold=5.418e+02, percent-clipped=1.0 +2023-02-06 22:41:51,229 INFO [train.py:901] (2/4) Epoch 19, batch 1700, loss[loss=0.1946, simple_loss=0.2637, pruned_loss=0.06276, over 7423.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2919, pruned_loss=0.06522, over 1613281.97 frames. ], batch size: 17, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:42:00,577 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147206.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:42:25,936 INFO [train.py:901] (2/4) Epoch 19, batch 1750, loss[loss=0.2214, simple_loss=0.297, pruned_loss=0.07296, over 8506.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2921, pruned_loss=0.06533, over 1614899.63 frames. ], batch size: 26, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:42:37,955 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:42:40,819 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9378, 3.6435, 2.0838, 2.7817, 2.7605, 2.0194, 2.7202, 2.9199], + device='cuda:2'), covar=tensor([0.1621, 0.0366, 0.1156, 0.0719, 0.0798, 0.1369, 0.0921, 0.0989], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0236, 0.0326, 0.0306, 0.0301, 0.0332, 0.0343, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:42:51,054 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.541e+02 2.529e+02 3.043e+02 3.569e+02 7.736e+02, threshold=6.085e+02, percent-clipped=5.0 +2023-02-06 22:43:03,023 INFO [train.py:901] (2/4) Epoch 19, batch 1800, loss[loss=0.2194, simple_loss=0.2869, pruned_loss=0.07594, over 7798.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2914, pruned_loss=0.06488, over 1613555.84 frames. ], batch size: 19, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:43:22,551 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147321.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:24,598 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147324.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:27,345 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:37,467 INFO [train.py:901] (2/4) Epoch 19, batch 1850, loss[loss=0.1949, simple_loss=0.2669, pruned_loss=0.0614, over 7648.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2914, pruned_loss=0.0652, over 1612520.70 frames. ], batch size: 19, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:43:41,692 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:48,571 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:00,461 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1328, 1.6841, 4.3431, 1.5860, 3.8214, 3.6624, 3.9016, 3.7667], + device='cuda:2'), covar=tensor([0.0603, 0.4055, 0.0525, 0.3967, 0.1084, 0.0953, 0.0637, 0.0700], + device='cuda:2'), in_proj_covar=tensor([0.0600, 0.0634, 0.0674, 0.0612, 0.0691, 0.0592, 0.0589, 0.0651], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 22:44:02,407 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.300e+02 2.823e+02 3.606e+02 1.006e+03, threshold=5.645e+02, percent-clipped=2.0 +2023-02-06 22:44:06,242 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 22:44:06,694 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:12,515 INFO [train.py:901] (2/4) Epoch 19, batch 1900, loss[loss=0.2275, simple_loss=0.3163, pruned_loss=0.06931, over 8359.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2917, pruned_loss=0.06516, over 1617280.85 frames. ], batch size: 24, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:44:37,306 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147425.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:44,949 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 22:44:49,673 INFO [train.py:901] (2/4) Epoch 19, batch 1950, loss[loss=0.2281, simple_loss=0.303, pruned_loss=0.07656, over 7928.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2912, pruned_loss=0.06453, over 1619792.12 frames. ], batch size: 20, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:44:51,462 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 22:44:55,960 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147452.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:56,489 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 22:45:13,741 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.289e+02 2.862e+02 3.830e+02 8.439e+02, threshold=5.724e+02, percent-clipped=6.0 +2023-02-06 22:45:15,253 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 22:45:23,813 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.83 vs. limit=5.0 +2023-02-06 22:45:24,867 INFO [train.py:901] (2/4) Epoch 19, batch 2000, loss[loss=0.2174, simple_loss=0.3038, pruned_loss=0.06552, over 8477.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2919, pruned_loss=0.06464, over 1621970.69 frames. ], batch size: 27, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:46:01,793 INFO [train.py:901] (2/4) Epoch 19, batch 2050, loss[loss=0.184, simple_loss=0.2546, pruned_loss=0.05674, over 7676.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2914, pruned_loss=0.06436, over 1618551.66 frames. ], batch size: 18, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:46:25,336 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147577.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:46:25,778 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.500e+02 2.918e+02 3.445e+02 6.516e+02, threshold=5.836e+02, percent-clipped=2.0 +2023-02-06 22:46:35,051 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4243, 2.4373, 1.6362, 2.1735, 1.9733, 1.4786, 1.9722, 1.9180], + device='cuda:2'), covar=tensor([0.1705, 0.0370, 0.1340, 0.0642, 0.0796, 0.1559, 0.1042, 0.1050], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0233, 0.0325, 0.0302, 0.0300, 0.0330, 0.0340, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 22:46:36,262 INFO [train.py:901] (2/4) Epoch 19, batch 2100, loss[loss=0.2688, simple_loss=0.3324, pruned_loss=0.1027, over 8244.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2923, pruned_loss=0.06517, over 1619115.60 frames. ], batch size: 24, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:46:42,937 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147602.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:46:43,437 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:47:12,100 INFO [train.py:901] (2/4) Epoch 19, batch 2150, loss[loss=0.2246, simple_loss=0.3215, pruned_loss=0.06384, over 8327.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2917, pruned_loss=0.06481, over 1616879.67 frames. ], batch size: 25, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:47:33,572 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:47:36,909 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:47:37,414 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.398e+02 3.174e+02 3.852e+02 9.466e+02, threshold=6.348e+02, percent-clipped=6.0 +2023-02-06 22:47:46,173 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.90 vs. limit=5.0 +2023-02-06 22:47:47,715 INFO [train.py:901] (2/4) Epoch 19, batch 2200, loss[loss=0.1752, simple_loss=0.2566, pruned_loss=0.04689, over 8230.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2904, pruned_loss=0.06443, over 1611680.05 frames. ], batch size: 22, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:47:47,898 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0855, 1.6516, 3.1084, 1.4900, 2.1784, 3.3378, 3.4537, 2.8135], + device='cuda:2'), covar=tensor([0.0975, 0.1510, 0.0369, 0.1996, 0.1018, 0.0253, 0.0640, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0284, 0.0314, 0.0279, 0.0306, 0.0296, 0.0258, 0.0398, 0.0298], + 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') +2023-02-06 22:48:04,732 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:48:21,941 INFO [train.py:901] (2/4) Epoch 19, batch 2250, loss[loss=0.2123, simple_loss=0.3011, pruned_loss=0.06175, over 8194.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2917, pruned_loss=0.06499, over 1611743.87 frames. ], batch size: 23, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:48:41,110 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:48:46,060 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 22:48:47,097 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.392e+02 3.089e+02 3.849e+02 9.613e+02, threshold=6.179e+02, percent-clipped=2.0 +2023-02-06 22:48:53,337 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:48:56,991 INFO [train.py:901] (2/4) Epoch 19, batch 2300, loss[loss=0.2065, simple_loss=0.2935, pruned_loss=0.05976, over 8501.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2914, pruned_loss=0.06502, over 1611062.73 frames. ], batch size: 26, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:48:58,973 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147796.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:49:24,349 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:49:30,994 INFO [train.py:901] (2/4) Epoch 19, batch 2350, loss[loss=0.2206, simple_loss=0.2987, pruned_loss=0.07125, over 8463.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2911, pruned_loss=0.06439, over 1615741.02 frames. ], batch size: 29, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:49:53,255 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:49:55,895 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.451e+02 2.984e+02 3.607e+02 1.132e+03, threshold=5.968e+02, percent-clipped=4.0 +2023-02-06 22:50:01,542 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147884.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:50:07,522 INFO [train.py:901] (2/4) Epoch 19, batch 2400, loss[loss=0.2223, simple_loss=0.3125, pruned_loss=0.0661, over 8498.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2921, pruned_loss=0.06503, over 1618970.25 frames. ], batch size: 28, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:50:20,044 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147911.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:50:24,948 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6475, 1.5686, 2.2744, 1.6065, 1.2301, 2.2613, 0.3276, 1.3618], + device='cuda:2'), covar=tensor([0.1735, 0.1442, 0.0367, 0.1344, 0.3006, 0.0429, 0.2467, 0.1544], + device='cuda:2'), in_proj_covar=tensor([0.0185, 0.0194, 0.0123, 0.0220, 0.0269, 0.0131, 0.0169, 0.0186], + 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') +2023-02-06 22:50:32,261 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:50:41,465 INFO [train.py:901] (2/4) Epoch 19, batch 2450, loss[loss=0.1973, simple_loss=0.269, pruned_loss=0.06278, over 7411.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2922, pruned_loss=0.06556, over 1616006.53 frames. ], batch size: 17, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:50:56,377 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7051, 2.7700, 2.4300, 4.0089, 1.8877, 2.1625, 2.5455, 3.2001], + device='cuda:2'), covar=tensor([0.0578, 0.0782, 0.0778, 0.0203, 0.0954, 0.1073, 0.0914, 0.0687], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0197, 0.0249, 0.0212, 0.0207, 0.0246, 0.0253, 0.0212], + 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') +2023-02-06 22:51:03,067 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:51:04,979 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8379, 1.4593, 4.2371, 1.7450, 3.3911, 3.2712, 3.8056, 3.7428], + device='cuda:2'), covar=tensor([0.1281, 0.6040, 0.1134, 0.4811, 0.2111, 0.1972, 0.1025, 0.1009], + device='cuda:2'), in_proj_covar=tensor([0.0597, 0.0627, 0.0676, 0.0607, 0.0687, 0.0590, 0.0586, 0.0646], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 22:51:05,437 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.599e+02 2.990e+02 3.557e+02 6.406e+02, threshold=5.981e+02, percent-clipped=1.0 +2023-02-06 22:51:15,598 INFO [train.py:901] (2/4) Epoch 19, batch 2500, loss[loss=0.201, simple_loss=0.2887, pruned_loss=0.05663, over 8081.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2918, pruned_loss=0.06542, over 1612085.02 frames. ], batch size: 21, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:51:20,626 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:51:37,736 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:51:52,313 INFO [train.py:901] (2/4) Epoch 19, batch 2550, loss[loss=0.2142, simple_loss=0.2902, pruned_loss=0.06915, over 8575.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.291, pruned_loss=0.06473, over 1611953.14 frames. ], batch size: 31, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:51:52,574 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:52:09,290 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148068.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:52:15,629 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 2.379e+02 2.867e+02 3.516e+02 7.047e+02, threshold=5.734e+02, percent-clipped=3.0 +2023-02-06 22:52:23,195 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8821, 1.6261, 2.0514, 1.8131, 2.0577, 1.9392, 1.7542, 0.7631], + device='cuda:2'), covar=tensor([0.5627, 0.4682, 0.1903, 0.3235, 0.2238, 0.2937, 0.1860, 0.5024], + device='cuda:2'), in_proj_covar=tensor([0.0932, 0.0952, 0.0787, 0.0917, 0.0980, 0.0868, 0.0730, 0.0808], + 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') +2023-02-06 22:52:26,279 INFO [train.py:901] (2/4) Epoch 19, batch 2600, loss[loss=0.2262, simple_loss=0.2904, pruned_loss=0.08095, over 7521.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2905, pruned_loss=0.06521, over 1608211.38 frames. ], batch size: 18, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:52:28,762 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-06 22:52:57,246 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:00,040 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:01,854 INFO [train.py:901] (2/4) Epoch 19, batch 2650, loss[loss=0.2161, simple_loss=0.3039, pruned_loss=0.06413, over 8472.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2914, pruned_loss=0.0656, over 1606210.14 frames. ], batch size: 25, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:53:16,795 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:18,198 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:24,869 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148177.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:25,454 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.384e+02 2.853e+02 3.529e+02 7.126e+02, threshold=5.707e+02, percent-clipped=4.0 +2023-02-06 22:53:27,036 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7961, 2.3394, 3.4773, 2.0243, 1.7378, 3.4505, 0.7190, 2.1242], + device='cuda:2'), covar=tensor([0.1815, 0.1451, 0.0288, 0.1868, 0.3240, 0.0331, 0.2599, 0.1524], + device='cuda:2'), in_proj_covar=tensor([0.0185, 0.0193, 0.0123, 0.0219, 0.0270, 0.0131, 0.0169, 0.0186], + 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') +2023-02-06 22:53:35,180 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148192.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:35,669 INFO [train.py:901] (2/4) Epoch 19, batch 2700, loss[loss=0.1857, simple_loss=0.2774, pruned_loss=0.04702, over 8140.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.291, pruned_loss=0.06529, over 1610085.06 frames. ], batch size: 22, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:53:44,879 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3130, 2.0044, 2.7336, 2.2301, 2.6393, 2.1879, 2.0472, 1.8550], + device='cuda:2'), covar=tensor([0.3774, 0.4060, 0.1443, 0.2775, 0.1723, 0.2610, 0.1556, 0.3704], + device='cuda:2'), in_proj_covar=tensor([0.0936, 0.0958, 0.0790, 0.0922, 0.0986, 0.0874, 0.0734, 0.0814], + 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') +2023-02-06 22:53:54,174 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:11,924 INFO [train.py:901] (2/4) Epoch 19, batch 2750, loss[loss=0.2402, simple_loss=0.3169, pruned_loss=0.08172, over 8581.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2912, pruned_loss=0.06558, over 1610428.76 frames. ], batch size: 31, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:54:32,994 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148273.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:36,056 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.484e+02 2.895e+02 4.098e+02 9.310e+02, threshold=5.790e+02, percent-clipped=8.0 +2023-02-06 22:54:45,407 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148292.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:45,919 INFO [train.py:901] (2/4) Epoch 19, batch 2800, loss[loss=0.1979, simple_loss=0.2708, pruned_loss=0.06254, over 7432.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2921, pruned_loss=0.06602, over 1611597.24 frames. ], batch size: 17, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:54:54,041 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148305.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:55,513 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2170, 1.8656, 2.5305, 2.0856, 2.4259, 2.1803, 1.9377, 1.2891], + device='cuda:2'), covar=tensor([0.5213, 0.4743, 0.1658, 0.3486, 0.2320, 0.2957, 0.1928, 0.4924], + device='cuda:2'), in_proj_covar=tensor([0.0932, 0.0953, 0.0784, 0.0916, 0.0981, 0.0869, 0.0730, 0.0809], + 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') +2023-02-06 22:54:58,709 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8201, 1.7722, 2.5483, 1.5486, 1.2857, 2.5250, 0.4227, 1.5542], + device='cuda:2'), covar=tensor([0.2052, 0.1413, 0.0323, 0.1536, 0.3129, 0.0379, 0.2467, 0.1526], + device='cuda:2'), in_proj_covar=tensor([0.0184, 0.0193, 0.0123, 0.0219, 0.0270, 0.0131, 0.0169, 0.0186], + 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') +2023-02-06 22:55:13,965 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148334.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:19,425 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9640, 1.3355, 1.7273, 1.1791, 0.9282, 1.4820, 1.7234, 1.5242], + device='cuda:2'), covar=tensor([0.0501, 0.1272, 0.1637, 0.1538, 0.0613, 0.1524, 0.0692, 0.0693], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0188, 0.0157, 0.0100, 0.0161, 0.0113, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 22:55:19,842 INFO [train.py:901] (2/4) Epoch 19, batch 2850, loss[loss=0.2452, simple_loss=0.3137, pruned_loss=0.08833, over 8338.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.292, pruned_loss=0.06548, over 1614834.00 frames. ], batch size: 26, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:55:46,065 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.512e+02 2.931e+02 3.824e+02 7.566e+02, threshold=5.862e+02, percent-clipped=4.0 +2023-02-06 22:55:50,339 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7251, 1.6817, 4.8732, 1.8604, 4.3711, 3.9841, 4.4322, 4.2885], + device='cuda:2'), covar=tensor([0.0448, 0.4075, 0.0442, 0.3764, 0.0879, 0.1093, 0.0501, 0.0531], + device='cuda:2'), in_proj_covar=tensor([0.0591, 0.0624, 0.0669, 0.0597, 0.0676, 0.0585, 0.0580, 0.0642], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 22:55:51,897 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-06 22:55:52,987 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:55,660 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:56,117 INFO [train.py:901] (2/4) Epoch 19, batch 2900, loss[loss=0.1869, simple_loss=0.2735, pruned_loss=0.05013, over 7660.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.292, pruned_loss=0.06562, over 1609557.36 frames. ], batch size: 19, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:56:12,658 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:56:29,356 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 22:56:29,949 INFO [train.py:901] (2/4) Epoch 19, batch 2950, loss[loss=0.2199, simple_loss=0.2901, pruned_loss=0.07485, over 7807.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2927, pruned_loss=0.06613, over 1608870.17 frames. ], batch size: 19, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:56:32,822 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148447.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:56:41,585 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5768, 1.9324, 3.0919, 1.4134, 2.3549, 2.0118, 1.5953, 2.4623], + device='cuda:2'), covar=tensor([0.1800, 0.2522, 0.0893, 0.4274, 0.1747, 0.2944, 0.2141, 0.2133], + device='cuda:2'), in_proj_covar=tensor([0.0519, 0.0584, 0.0554, 0.0632, 0.0640, 0.0589, 0.0523, 0.0632], + 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') +2023-02-06 22:56:42,930 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9362, 2.4209, 4.0858, 1.7834, 3.0303, 2.3923, 2.0071, 2.9544], + device='cuda:2'), covar=tensor([0.1751, 0.2462, 0.1004, 0.4011, 0.1765, 0.2944, 0.2023, 0.2526], + device='cuda:2'), in_proj_covar=tensor([0.0518, 0.0583, 0.0554, 0.0632, 0.0639, 0.0589, 0.0522, 0.0631], + 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') +2023-02-06 22:56:54,933 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.514e+02 3.009e+02 3.973e+02 7.443e+02, threshold=6.017e+02, percent-clipped=3.0 +2023-02-06 22:57:06,344 INFO [train.py:901] (2/4) Epoch 19, batch 3000, loss[loss=0.2068, simple_loss=0.2913, pruned_loss=0.06114, over 8360.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2919, pruned_loss=0.06571, over 1610255.81 frames. ], batch size: 24, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:57:06,344 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 22:57:22,670 INFO [train.py:935] (2/4) Epoch 19, validation: loss=0.1752, simple_loss=0.2756, pruned_loss=0.03738, over 944034.00 frames. +2023-02-06 22:57:22,671 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 22:57:38,584 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148516.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:57:56,926 INFO [train.py:901] (2/4) Epoch 19, batch 3050, loss[loss=0.2112, simple_loss=0.2717, pruned_loss=0.07535, over 7529.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2922, pruned_loss=0.06601, over 1611467.03 frames. ], batch size: 18, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:58:00,731 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:58:17,727 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:58:21,053 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.398e+02 2.811e+02 3.727e+02 6.995e+02, threshold=5.622e+02, percent-clipped=3.0 +2023-02-06 22:58:30,233 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148590.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:58:32,157 INFO [train.py:901] (2/4) Epoch 19, batch 3100, loss[loss=0.1767, simple_loss=0.2644, pruned_loss=0.04453, over 8131.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2913, pruned_loss=0.06535, over 1606574.31 frames. ], batch size: 22, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:58:49,329 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148615.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:09,435 INFO [train.py:901] (2/4) Epoch 19, batch 3150, loss[loss=0.2008, simple_loss=0.2815, pruned_loss=0.06002, over 7919.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2915, pruned_loss=0.06503, over 1610154.16 frames. ], batch size: 20, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:59:10,323 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:13,424 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:26,360 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:26,421 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:32,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.358e+02 3.073e+02 3.824e+02 9.523e+02, threshold=6.146e+02, percent-clipped=8.0 +2023-02-06 22:59:42,404 INFO [train.py:901] (2/4) Epoch 19, batch 3200, loss[loss=0.2081, simple_loss=0.3, pruned_loss=0.05814, over 8591.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2929, pruned_loss=0.06572, over 1608510.32 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 22:59:50,082 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1937, 1.3609, 1.5993, 1.2709, 0.7329, 1.4319, 1.2848, 1.2129], + device='cuda:2'), covar=tensor([0.0559, 0.1250, 0.1633, 0.1415, 0.0543, 0.1415, 0.0646, 0.0618], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0100, 0.0162, 0.0113, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 22:59:53,525 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1045, 1.4844, 1.6425, 1.3845, 1.0894, 1.4638, 1.9111, 1.7158], + device='cuda:2'), covar=tensor([0.0491, 0.1216, 0.1622, 0.1391, 0.0597, 0.1444, 0.0627, 0.0585], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0100, 0.0162, 0.0113, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 23:00:12,965 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148734.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:00:19,659 INFO [train.py:901] (2/4) Epoch 19, batch 3250, loss[loss=0.2448, simple_loss=0.3149, pruned_loss=0.08737, over 8562.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2936, pruned_loss=0.06576, over 1615792.46 frames. ], batch size: 39, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:00:34,071 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:00:35,478 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2129, 1.8852, 2.5406, 2.0684, 2.4677, 2.1815, 1.9367, 1.1844], + device='cuda:2'), covar=tensor([0.5147, 0.4674, 0.1836, 0.3680, 0.2528, 0.3004, 0.1903, 0.5225], + device='cuda:2'), in_proj_covar=tensor([0.0932, 0.0953, 0.0783, 0.0916, 0.0982, 0.0871, 0.0729, 0.0809], + 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') +2023-02-06 23:00:40,075 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3639, 2.2240, 3.1456, 2.4720, 3.0647, 2.3560, 2.1637, 1.8348], + device='cuda:2'), covar=tensor([0.5709, 0.5494, 0.2046, 0.3932, 0.2608, 0.3097, 0.1979, 0.5919], + device='cuda:2'), in_proj_covar=tensor([0.0930, 0.0951, 0.0782, 0.0915, 0.0980, 0.0870, 0.0728, 0.0809], + 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') +2023-02-06 23:00:43,259 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.443e+02 3.073e+02 4.112e+02 8.183e+02, threshold=6.146e+02, percent-clipped=4.0 +2023-02-06 23:00:52,319 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:00:53,591 INFO [train.py:901] (2/4) Epoch 19, batch 3300, loss[loss=0.2619, simple_loss=0.3373, pruned_loss=0.09322, over 8506.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2923, pruned_loss=0.06481, over 1618969.54 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:01:28,246 INFO [train.py:901] (2/4) Epoch 19, batch 3350, loss[loss=0.2022, simple_loss=0.276, pruned_loss=0.06424, over 8086.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2931, pruned_loss=0.06466, over 1618160.28 frames. ], batch size: 21, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:01:41,953 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148860.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:01:53,952 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.291e+02 2.864e+02 3.449e+02 6.722e+02, threshold=5.728e+02, percent-clipped=1.0 +2023-02-06 23:02:02,990 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7730, 2.2428, 3.3610, 1.6285, 2.7046, 2.2198, 1.9372, 2.6226], + device='cuda:2'), covar=tensor([0.1619, 0.2176, 0.0710, 0.3891, 0.1507, 0.2683, 0.1894, 0.2009], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0581, 0.0552, 0.0630, 0.0638, 0.0588, 0.0521, 0.0630], + 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') +2023-02-06 23:02:04,168 INFO [train.py:901] (2/4) Epoch 19, batch 3400, loss[loss=0.2256, simple_loss=0.3077, pruned_loss=0.07181, over 8192.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2929, pruned_loss=0.06535, over 1619350.33 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:02:13,156 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148906.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:02:32,358 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1858, 1.8113, 2.3635, 2.0051, 2.3331, 2.1502, 1.9208, 1.1864], + device='cuda:2'), covar=tensor([0.4781, 0.4238, 0.1745, 0.3427, 0.2146, 0.2775, 0.1830, 0.4707], + device='cuda:2'), in_proj_covar=tensor([0.0930, 0.0951, 0.0785, 0.0917, 0.0979, 0.0870, 0.0729, 0.0810], + 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') +2023-02-06 23:02:38,071 INFO [train.py:901] (2/4) Epoch 19, batch 3450, loss[loss=0.2358, simple_loss=0.3229, pruned_loss=0.07441, over 8463.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2927, pruned_loss=0.0652, over 1618963.96 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:03:01,930 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148975.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:03:04,398 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.340e+02 2.956e+02 3.727e+02 1.104e+03, threshold=5.912e+02, percent-clipped=3.0 +2023-02-06 23:03:14,140 INFO [train.py:901] (2/4) Epoch 19, batch 3500, loss[loss=0.1949, simple_loss=0.264, pruned_loss=0.06291, over 7207.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2945, pruned_loss=0.06633, over 1617963.36 frames. ], batch size: 16, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:03:28,308 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149013.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:03:33,359 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:03:35,920 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 23:03:48,890 INFO [train.py:901] (2/4) Epoch 19, batch 3550, loss[loss=0.2448, simple_loss=0.327, pruned_loss=0.08129, over 8506.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.295, pruned_loss=0.06591, over 1624692.81 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:03:50,372 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:04:13,070 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:04:13,636 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.461e+02 3.087e+02 3.824e+02 7.251e+02, threshold=6.175e+02, percent-clipped=6.0 +2023-02-06 23:04:25,637 INFO [train.py:901] (2/4) Epoch 19, batch 3600, loss[loss=0.1997, simple_loss=0.2753, pruned_loss=0.06204, over 8023.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2941, pruned_loss=0.06575, over 1622502.55 frames. ], batch size: 22, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:04:49,822 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:04:59,721 INFO [train.py:901] (2/4) Epoch 19, batch 3650, loss[loss=0.2345, simple_loss=0.2953, pruned_loss=0.08679, over 8079.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2947, pruned_loss=0.06629, over 1621608.77 frames. ], batch size: 21, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:05:13,232 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:05:24,380 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.296e+02 2.731e+02 3.488e+02 6.725e+02, threshold=5.462e+02, percent-clipped=1.0 +2023-02-06 23:05:30,733 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149187.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:05:35,220 INFO [train.py:901] (2/4) Epoch 19, batch 3700, loss[loss=0.2015, simple_loss=0.2876, pruned_loss=0.05774, over 8246.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2938, pruned_loss=0.06584, over 1617606.43 frames. ], batch size: 22, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:05:35,409 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:05:38,051 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 23:06:02,788 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149231.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:06:10,460 INFO [train.py:901] (2/4) Epoch 19, batch 3750, loss[loss=0.1898, simple_loss=0.2636, pruned_loss=0.05804, over 7439.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.293, pruned_loss=0.06574, over 1609191.11 frames. ], batch size: 17, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:06:19,378 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:06:34,566 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.530e+02 3.028e+02 3.831e+02 7.632e+02, threshold=6.056e+02, percent-clipped=6.0 +2023-02-06 23:06:38,967 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9972, 3.6616, 2.2844, 2.8766, 2.8803, 1.9410, 2.7587, 3.0213], + device='cuda:2'), covar=tensor([0.1749, 0.0347, 0.1130, 0.0769, 0.0713, 0.1481, 0.1197, 0.1061], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0241, 0.0330, 0.0307, 0.0303, 0.0335, 0.0345, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:06:44,222 INFO [train.py:901] (2/4) Epoch 19, batch 3800, loss[loss=0.2027, simple_loss=0.295, pruned_loss=0.05514, over 8353.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.293, pruned_loss=0.06504, over 1615689.58 frames. ], batch size: 24, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:06:44,676 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 23:07:20,712 INFO [train.py:901] (2/4) Epoch 19, batch 3850, loss[loss=0.2276, simple_loss=0.3036, pruned_loss=0.07581, over 8580.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2912, pruned_loss=0.06418, over 1614248.93 frames. ], batch size: 34, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:07:42,416 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 23:07:45,095 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.409e+02 2.948e+02 3.728e+02 6.848e+02, threshold=5.896e+02, percent-clipped=3.0 +2023-02-06 23:07:48,713 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:07:54,635 INFO [train.py:901] (2/4) Epoch 19, batch 3900, loss[loss=0.2098, simple_loss=0.2878, pruned_loss=0.06586, over 8078.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2915, pruned_loss=0.06434, over 1614205.71 frames. ], batch size: 21, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:08:06,570 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:08:10,670 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1316, 1.8469, 1.9092, 1.7851, 1.2809, 1.8552, 2.0662, 1.8885], + device='cuda:2'), covar=tensor([0.0555, 0.0971, 0.1333, 0.1107, 0.0643, 0.1129, 0.0671, 0.0495], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0188, 0.0157, 0.0099, 0.0160, 0.0112, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 23:08:31,939 INFO [train.py:901] (2/4) Epoch 19, batch 3950, loss[loss=0.2054, simple_loss=0.289, pruned_loss=0.06093, over 7810.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2919, pruned_loss=0.06449, over 1616131.98 frames. ], batch size: 20, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:08:36,330 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149449.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:08:53,036 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:08:56,238 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.497e+02 2.881e+02 4.050e+02 6.266e+02, threshold=5.763e+02, percent-clipped=1.0 +2023-02-06 23:09:05,730 INFO [train.py:901] (2/4) Epoch 19, batch 4000, loss[loss=0.2241, simple_loss=0.3011, pruned_loss=0.07349, over 8698.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2916, pruned_loss=0.06475, over 1616280.80 frames. ], batch size: 34, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:09:11,133 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 23:09:13,966 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9124, 1.7574, 6.0799, 2.3578, 5.3972, 5.0521, 5.4881, 5.4073], + device='cuda:2'), covar=tensor([0.0454, 0.4913, 0.0346, 0.3779, 0.1018, 0.0839, 0.0548, 0.0505], + device='cuda:2'), in_proj_covar=tensor([0.0599, 0.0635, 0.0674, 0.0608, 0.0687, 0.0592, 0.0586, 0.0650], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 23:09:32,420 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:09:35,725 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1726, 1.0676, 1.2843, 1.1062, 0.9649, 1.3027, 0.0766, 0.8972], + device='cuda:2'), covar=tensor([0.1756, 0.1487, 0.0465, 0.0828, 0.3177, 0.0536, 0.2435, 0.1360], + device='cuda:2'), in_proj_covar=tensor([0.0184, 0.0194, 0.0124, 0.0222, 0.0271, 0.0132, 0.0171, 0.0188], + 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') +2023-02-06 23:09:40,142 INFO [train.py:901] (2/4) Epoch 19, batch 4050, loss[loss=0.2046, simple_loss=0.2956, pruned_loss=0.05682, over 8446.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2908, pruned_loss=0.06457, over 1618129.41 frames. ], batch size: 29, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:10:03,202 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 23:10:05,803 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.470e+02 3.003e+02 4.246e+02 8.728e+02, threshold=6.007e+02, percent-clipped=8.0 +2023-02-06 23:10:08,589 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:10:12,316 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 23:10:15,178 INFO [train.py:901] (2/4) Epoch 19, batch 4100, loss[loss=0.2223, simple_loss=0.3037, pruned_loss=0.07046, over 8334.00 frames. ], tot_loss[loss=0.209, simple_loss=0.29, pruned_loss=0.06406, over 1615990.84 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:10:39,673 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6935, 1.3361, 1.6339, 1.1545, 0.8522, 1.3932, 1.4332, 1.3644], + device='cuda:2'), covar=tensor([0.0526, 0.1324, 0.1717, 0.1535, 0.0601, 0.1540, 0.0721, 0.0659], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0157, 0.0099, 0.0161, 0.0113, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 23:10:49,871 INFO [train.py:901] (2/4) Epoch 19, batch 4150, loss[loss=0.2564, simple_loss=0.3242, pruned_loss=0.09425, over 7040.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2906, pruned_loss=0.06431, over 1615243.69 frames. ], batch size: 71, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:11:16,655 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.273e+02 2.791e+02 3.594e+02 5.057e+02, threshold=5.582e+02, percent-clipped=0.0 +2023-02-06 23:11:26,106 INFO [train.py:901] (2/4) Epoch 19, batch 4200, loss[loss=0.1841, simple_loss=0.2691, pruned_loss=0.04955, over 8240.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2898, pruned_loss=0.06371, over 1611726.02 frames. ], batch size: 22, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:11:36,335 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 23:11:36,601 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 23:11:59,497 INFO [train.py:901] (2/4) Epoch 19, batch 4250, loss[loss=0.1903, simple_loss=0.2657, pruned_loss=0.05743, over 7546.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2897, pruned_loss=0.06364, over 1614973.38 frames. ], batch size: 18, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:12:00,922 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 23:12:14,489 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149764.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 23:12:25,320 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.444e+02 3.025e+02 3.928e+02 1.033e+03, threshold=6.050e+02, percent-clipped=5.0 +2023-02-06 23:12:35,581 INFO [train.py:901] (2/4) Epoch 19, batch 4300, loss[loss=0.2339, simple_loss=0.3215, pruned_loss=0.07314, over 8104.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2907, pruned_loss=0.06403, over 1618510.90 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:12:42,699 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0957, 3.9912, 3.7276, 2.1090, 3.5944, 3.6319, 3.7182, 3.4384], + device='cuda:2'), covar=tensor([0.0797, 0.0635, 0.1065, 0.4335, 0.0872, 0.1084, 0.1286, 0.1008], + device='cuda:2'), in_proj_covar=tensor([0.0518, 0.0430, 0.0430, 0.0533, 0.0420, 0.0437, 0.0415, 0.0378], + 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') +2023-02-06 23:13:10,090 INFO [train.py:901] (2/4) Epoch 19, batch 4350, loss[loss=0.199, simple_loss=0.2812, pruned_loss=0.05846, over 7975.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2906, pruned_loss=0.06386, over 1620199.98 frames. ], batch size: 21, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:13:33,154 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 23:13:33,238 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:13:35,197 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.416e+02 2.972e+02 3.761e+02 1.184e+03, threshold=5.944e+02, percent-clipped=4.0 +2023-02-06 23:13:39,040 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 23:13:44,579 INFO [train.py:901] (2/4) Epoch 19, batch 4400, loss[loss=0.263, simple_loss=0.3299, pruned_loss=0.09803, over 8239.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2913, pruned_loss=0.0644, over 1616477.77 frames. ], batch size: 24, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:13:50,363 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1820, 2.0936, 1.6530, 1.9767, 1.7646, 1.3964, 1.5969, 1.5848], + device='cuda:2'), covar=tensor([0.1358, 0.0459, 0.1306, 0.0510, 0.0707, 0.1617, 0.0950, 0.0903], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0238, 0.0329, 0.0306, 0.0301, 0.0333, 0.0343, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:13:53,807 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6772, 2.2512, 4.0895, 1.4511, 3.0946, 2.2459, 1.8282, 2.9774], + device='cuda:2'), covar=tensor([0.1977, 0.2627, 0.0872, 0.4659, 0.1753, 0.3216, 0.2298, 0.2323], + device='cuda:2'), in_proj_covar=tensor([0.0522, 0.0590, 0.0560, 0.0638, 0.0645, 0.0595, 0.0531, 0.0637], + 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') +2023-02-06 23:14:09,977 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:14:14,611 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 23:14:20,891 INFO [train.py:901] (2/4) Epoch 19, batch 4450, loss[loss=0.2231, simple_loss=0.3191, pruned_loss=0.0636, over 8325.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.292, pruned_loss=0.06483, over 1615763.25 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:14:44,888 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.522e+02 2.925e+02 4.193e+02 1.036e+03, threshold=5.849e+02, percent-clipped=7.0 +2023-02-06 23:14:53,315 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:14:54,476 INFO [train.py:901] (2/4) Epoch 19, batch 4500, loss[loss=0.211, simple_loss=0.3044, pruned_loss=0.05881, over 8492.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2914, pruned_loss=0.06406, over 1617831.13 frames. ], batch size: 28, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:15:08,401 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 23:15:08,633 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6418, 1.9938, 3.2707, 1.4572, 2.4700, 2.0659, 1.6545, 2.3888], + device='cuda:2'), covar=tensor([0.1891, 0.2489, 0.0820, 0.4440, 0.1809, 0.3068, 0.2250, 0.2233], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0589, 0.0558, 0.0637, 0.0644, 0.0593, 0.0530, 0.0637], + 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') +2023-02-06 23:15:19,530 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9540, 1.6417, 3.2563, 1.4534, 2.1575, 3.5476, 3.6864, 2.9875], + device='cuda:2'), covar=tensor([0.1138, 0.1679, 0.0345, 0.2129, 0.1105, 0.0249, 0.0492, 0.0604], + device='cuda:2'), in_proj_covar=tensor([0.0290, 0.0316, 0.0286, 0.0312, 0.0302, 0.0265, 0.0406, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:15:31,625 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:15:32,128 INFO [train.py:901] (2/4) Epoch 19, batch 4550, loss[loss=0.1933, simple_loss=0.2856, pruned_loss=0.0505, over 8295.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2919, pruned_loss=0.06427, over 1617459.88 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:15:51,857 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4581, 2.7003, 1.6511, 2.3654, 2.0579, 1.3512, 1.9413, 2.1458], + device='cuda:2'), covar=tensor([0.1748, 0.0413, 0.1459, 0.0671, 0.0936, 0.1888, 0.1330, 0.1073], + device='cuda:2'), in_proj_covar=tensor([0.0348, 0.0236, 0.0326, 0.0305, 0.0300, 0.0330, 0.0341, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:15:56,361 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.399e+02 2.811e+02 3.428e+02 5.502e+02, threshold=5.622e+02, percent-clipped=0.0 +2023-02-06 23:16:05,764 INFO [train.py:901] (2/4) Epoch 19, batch 4600, loss[loss=0.1894, simple_loss=0.2724, pruned_loss=0.0532, over 8617.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2912, pruned_loss=0.06395, over 1618771.62 frames. ], batch size: 34, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:16:08,470 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:16:15,963 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150108.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 23:16:29,379 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3168, 2.8809, 2.2251, 3.9470, 1.7825, 2.0308, 2.4385, 2.9941], + device='cuda:2'), covar=tensor([0.0774, 0.0728, 0.0870, 0.0263, 0.1087, 0.1285, 0.1040, 0.0755], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0197, 0.0246, 0.0212, 0.0205, 0.0247, 0.0252, 0.0209], + 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') +2023-02-06 23:16:41,494 INFO [train.py:901] (2/4) Epoch 19, batch 4650, loss[loss=0.2215, simple_loss=0.3129, pruned_loss=0.065, over 8494.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2911, pruned_loss=0.06411, over 1617485.79 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:17:06,556 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.474e+02 2.856e+02 3.464e+02 8.049e+02, threshold=5.712e+02, percent-clipped=3.0 +2023-02-06 23:17:16,089 INFO [train.py:901] (2/4) Epoch 19, batch 4700, loss[loss=0.1797, simple_loss=0.2488, pruned_loss=0.05532, over 7429.00 frames. ], tot_loss[loss=0.21, simple_loss=0.291, pruned_loss=0.06453, over 1617156.30 frames. ], batch size: 17, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:17:30,094 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 23:17:36,661 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150223.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 23:17:50,830 INFO [train.py:901] (2/4) Epoch 19, batch 4750, loss[loss=0.2178, simple_loss=0.3024, pruned_loss=0.06661, over 8331.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2919, pruned_loss=0.06507, over 1614691.20 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:17:53,806 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:05,658 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8801, 1.9618, 2.5472, 1.6947, 1.4021, 2.5522, 0.5129, 1.5687], + device='cuda:2'), covar=tensor([0.1582, 0.1138, 0.0296, 0.1364, 0.2713, 0.0342, 0.2239, 0.1386], + device='cuda:2'), in_proj_covar=tensor([0.0187, 0.0194, 0.0124, 0.0223, 0.0272, 0.0133, 0.0170, 0.0189], + 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') +2023-02-06 23:18:12,327 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150272.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:13,462 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 23:18:15,517 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 23:18:16,856 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.315e+02 2.829e+02 3.523e+02 6.730e+02, threshold=5.657e+02, percent-clipped=3.0 +2023-02-06 23:18:25,824 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150292.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:26,369 INFO [train.py:901] (2/4) Epoch 19, batch 4800, loss[loss=0.2636, simple_loss=0.3255, pruned_loss=0.1009, over 6519.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2907, pruned_loss=0.06455, over 1611040.65 frames. ], batch size: 71, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:18:29,962 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150298.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:46,555 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150323.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:19:00,012 INFO [train.py:901] (2/4) Epoch 19, batch 4850, loss[loss=0.2478, simple_loss=0.3212, pruned_loss=0.08719, over 8324.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2907, pruned_loss=0.06472, over 1608152.08 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:19:05,339 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 23:19:10,226 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3975, 4.3336, 3.9182, 1.8649, 3.8426, 3.9606, 3.9739, 3.6498], + device='cuda:2'), covar=tensor([0.0810, 0.0564, 0.1153, 0.5085, 0.0885, 0.1036, 0.1277, 0.0924], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0425, 0.0427, 0.0528, 0.0414, 0.0432, 0.0411, 0.0374], + 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') +2023-02-06 23:19:27,011 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.476e+02 2.899e+02 3.621e+02 6.951e+02, threshold=5.799e+02, percent-clipped=6.0 +2023-02-06 23:19:32,496 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1616, 1.0669, 1.2914, 1.0516, 0.9918, 1.3258, 0.0659, 0.9057], + device='cuda:2'), covar=tensor([0.1728, 0.1310, 0.0476, 0.0844, 0.2479, 0.0557, 0.2249, 0.1295], + device='cuda:2'), in_proj_covar=tensor([0.0186, 0.0193, 0.0123, 0.0221, 0.0269, 0.0132, 0.0169, 0.0188], + 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') +2023-02-06 23:19:36,180 INFO [train.py:901] (2/4) Epoch 19, batch 4900, loss[loss=0.2398, simple_loss=0.3234, pruned_loss=0.0781, over 8588.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2913, pruned_loss=0.06497, over 1612258.95 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:20:07,720 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150441.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:20:09,003 INFO [train.py:901] (2/4) Epoch 19, batch 4950, loss[loss=0.2112, simple_loss=0.2901, pruned_loss=0.06615, over 8552.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2917, pruned_loss=0.06556, over 1614117.29 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:20:33,696 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.356e+02 2.775e+02 3.573e+02 1.033e+03, threshold=5.550e+02, percent-clipped=4.0 +2023-02-06 23:20:33,934 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150479.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 23:20:36,639 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9651, 2.4782, 3.7797, 1.9894, 1.9248, 3.6501, 0.8244, 2.1660], + device='cuda:2'), covar=tensor([0.1393, 0.1188, 0.0193, 0.1840, 0.2647, 0.0284, 0.2332, 0.1439], + device='cuda:2'), in_proj_covar=tensor([0.0187, 0.0194, 0.0124, 0.0222, 0.0269, 0.0132, 0.0169, 0.0188], + 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') +2023-02-06 23:20:43,979 INFO [train.py:901] (2/4) Epoch 19, batch 5000, loss[loss=0.1949, simple_loss=0.2829, pruned_loss=0.05342, over 8316.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2929, pruned_loss=0.06643, over 1614767.68 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:20:51,583 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.04 vs. limit=5.0 +2023-02-06 23:20:52,082 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150504.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 23:21:17,800 INFO [train.py:901] (2/4) Epoch 19, batch 5050, loss[loss=0.1886, simple_loss=0.2771, pruned_loss=0.04999, over 8139.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2934, pruned_loss=0.06612, over 1617961.64 frames. ], batch size: 22, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:21:25,940 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150555.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:21:26,620 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:21:40,934 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 23:21:41,605 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.501e+02 3.000e+02 3.972e+02 7.212e+02, threshold=5.999e+02, percent-clipped=3.0 +2023-02-06 23:21:51,778 INFO [train.py:901] (2/4) Epoch 19, batch 5100, loss[loss=0.2487, simple_loss=0.3248, pruned_loss=0.08635, over 8598.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2928, pruned_loss=0.06593, over 1616221.93 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:22:23,317 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:22:27,867 INFO [train.py:901] (2/4) Epoch 19, batch 5150, loss[loss=0.2009, simple_loss=0.2877, pruned_loss=0.05712, over 7969.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2937, pruned_loss=0.06612, over 1615732.03 frames. ], batch size: 21, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:22:49,043 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 23:22:49,427 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0330, 1.2618, 1.2170, 0.6891, 1.2317, 1.0264, 0.0638, 1.2086], + device='cuda:2'), covar=tensor([0.0409, 0.0361, 0.0321, 0.0528, 0.0390, 0.0936, 0.0766, 0.0335], + device='cuda:2'), in_proj_covar=tensor([0.0439, 0.0379, 0.0330, 0.0435, 0.0363, 0.0525, 0.0383, 0.0405], + device='cuda:2'), out_proj_covar=tensor([1.1883e-04, 1.0025e-04, 8.7162e-05, 1.1536e-04, 9.6351e-05, 1.4964e-04, + 1.0370e-04, 1.0816e-04], device='cuda:2') +2023-02-06 23:22:51,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.510e+02 3.215e+02 4.688e+02 9.098e+02, threshold=6.429e+02, percent-clipped=11.0 +2023-02-06 23:23:01,329 INFO [train.py:901] (2/4) Epoch 19, batch 5200, loss[loss=0.2215, simple_loss=0.2966, pruned_loss=0.07325, over 8421.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2933, pruned_loss=0.06616, over 1611355.97 frames. ], batch size: 49, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:23:29,363 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1042, 1.4783, 3.2785, 1.5075, 2.3658, 3.5933, 3.6909, 3.1131], + device='cuda:2'), covar=tensor([0.0982, 0.1649, 0.0354, 0.1939, 0.0965, 0.0225, 0.0496, 0.0511], + device='cuda:2'), in_proj_covar=tensor([0.0289, 0.0316, 0.0286, 0.0312, 0.0301, 0.0264, 0.0405, 0.0303], + 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') +2023-02-06 23:23:38,104 INFO [train.py:901] (2/4) Epoch 19, batch 5250, loss[loss=0.1708, simple_loss=0.2589, pruned_loss=0.04131, over 8321.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2933, pruned_loss=0.06621, over 1610738.81 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:23:40,636 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 23:23:42,649 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:23:43,385 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:24:01,539 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 2.565e+02 3.080e+02 4.191e+02 1.354e+03, threshold=6.160e+02, percent-clipped=9.0 +2023-02-06 23:24:03,321 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.78 vs. limit=5.0 +2023-02-06 23:24:10,893 INFO [train.py:901] (2/4) Epoch 19, batch 5300, loss[loss=0.2321, simple_loss=0.3146, pruned_loss=0.07478, over 8517.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2942, pruned_loss=0.06687, over 1613681.55 frames. ], batch size: 49, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:24:23,654 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:24:29,036 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0965, 1.6560, 1.7415, 1.4885, 0.9860, 1.5599, 1.7892, 1.6294], + device='cuda:2'), covar=tensor([0.0468, 0.1093, 0.1536, 0.1337, 0.0549, 0.1314, 0.0596, 0.0610], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0158, 0.0099, 0.0160, 0.0112, 0.0140], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 23:24:35,547 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 23:24:41,663 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150837.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:24:46,252 INFO [train.py:901] (2/4) Epoch 19, batch 5350, loss[loss=0.2243, simple_loss=0.3077, pruned_loss=0.07044, over 8463.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2945, pruned_loss=0.06737, over 1609148.09 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:25:10,968 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.499e+02 2.979e+02 3.723e+02 8.863e+02, threshold=5.959e+02, percent-clipped=1.0 +2023-02-06 23:25:20,521 INFO [train.py:901] (2/4) Epoch 19, batch 5400, loss[loss=0.1744, simple_loss=0.265, pruned_loss=0.04186, over 8081.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2941, pruned_loss=0.06697, over 1611071.16 frames. ], batch size: 21, lr: 3.98e-03, grad_scale: 16.0 +2023-02-06 23:25:24,749 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:25:37,931 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:25:55,451 INFO [train.py:901] (2/4) Epoch 19, batch 5450, loss[loss=0.1998, simple_loss=0.282, pruned_loss=0.05882, over 8140.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2936, pruned_loss=0.06625, over 1612908.06 frames. ], batch size: 22, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:26:22,622 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.573e+02 2.347e+02 2.658e+02 3.430e+02 7.604e+02, threshold=5.316e+02, percent-clipped=2.0 +2023-02-06 23:26:28,450 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 23:26:31,933 INFO [train.py:901] (2/4) Epoch 19, batch 5500, loss[loss=0.2429, simple_loss=0.3112, pruned_loss=0.08732, over 8666.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2934, pruned_loss=0.06596, over 1617126.23 frames. ], batch size: 39, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:26:33,443 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3893, 2.5293, 1.7188, 2.1825, 1.9351, 1.4459, 1.9598, 2.1517], + device='cuda:2'), covar=tensor([0.1899, 0.0437, 0.1404, 0.0811, 0.0935, 0.1821, 0.1156, 0.1087], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0237, 0.0327, 0.0305, 0.0301, 0.0331, 0.0341, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:26:41,794 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:26:46,651 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151014.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:26:58,972 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:27:06,213 INFO [train.py:901] (2/4) Epoch 19, batch 5550, loss[loss=0.2174, simple_loss=0.2977, pruned_loss=0.06849, over 8352.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2938, pruned_loss=0.06595, over 1617974.32 frames. ], batch size: 24, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:27:32,506 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.454e+02 3.027e+02 4.195e+02 6.901e+02, threshold=6.054e+02, percent-clipped=7.0 +2023-02-06 23:27:42,411 INFO [train.py:901] (2/4) Epoch 19, batch 5600, loss[loss=0.2173, simple_loss=0.3085, pruned_loss=0.06306, over 8537.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2934, pruned_loss=0.06599, over 1620192.59 frames. ], batch size: 31, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:27:43,136 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:27:47,922 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4352, 2.1354, 3.1324, 2.3470, 2.9858, 2.3998, 2.1004, 1.6951], + device='cuda:2'), covar=tensor([0.5373, 0.5611, 0.1852, 0.3788, 0.2471, 0.3011, 0.1995, 0.5707], + device='cuda:2'), in_proj_covar=tensor([0.0931, 0.0957, 0.0788, 0.0920, 0.0981, 0.0872, 0.0737, 0.0814], + 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') +2023-02-06 23:27:57,252 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7858, 1.9451, 4.6616, 2.4434, 2.8266, 5.2518, 5.3249, 4.5174], + device='cuda:2'), covar=tensor([0.1005, 0.1683, 0.0225, 0.1625, 0.0941, 0.0180, 0.0402, 0.0586], + device='cuda:2'), in_proj_covar=tensor([0.0292, 0.0320, 0.0288, 0.0313, 0.0302, 0.0265, 0.0407, 0.0305], + 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') +2023-02-06 23:27:58,841 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 23:28:15,302 INFO [train.py:901] (2/4) Epoch 19, batch 5650, loss[loss=0.2048, simple_loss=0.2842, pruned_loss=0.06274, over 7801.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2936, pruned_loss=0.0654, over 1618064.09 frames. ], batch size: 20, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:28:31,560 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 23:28:39,658 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.685e+02 3.149e+02 3.866e+02 8.044e+02, threshold=6.298e+02, percent-clipped=3.0 +2023-02-06 23:28:50,405 INFO [train.py:901] (2/4) Epoch 19, batch 5700, loss[loss=0.2218, simple_loss=0.3055, pruned_loss=0.06905, over 8345.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2934, pruned_loss=0.06565, over 1614368.98 frames. ], batch size: 26, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:28:57,636 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151202.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:02,317 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:20,368 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2848, 2.0515, 2.7794, 2.2785, 2.6891, 2.3157, 2.0383, 1.5166], + device='cuda:2'), covar=tensor([0.5068, 0.4953, 0.1801, 0.3411, 0.2477, 0.2719, 0.1862, 0.5238], + device='cuda:2'), in_proj_covar=tensor([0.0930, 0.0956, 0.0787, 0.0919, 0.0980, 0.0871, 0.0736, 0.0811], + 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') +2023-02-06 23:29:24,756 INFO [train.py:901] (2/4) Epoch 19, batch 5750, loss[loss=0.2054, simple_loss=0.2993, pruned_loss=0.0557, over 8473.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2928, pruned_loss=0.06532, over 1616509.74 frames. ], batch size: 28, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:29:36,100 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 23:29:37,493 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:43,022 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151270.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:48,610 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.389e+02 2.918e+02 3.727e+02 7.769e+02, threshold=5.836e+02, percent-clipped=3.0 +2023-02-06 23:29:58,855 INFO [train.py:901] (2/4) Epoch 19, batch 5800, loss[loss=0.2015, simple_loss=0.2981, pruned_loss=0.05248, over 8294.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2922, pruned_loss=0.06502, over 1615087.57 frames. ], batch size: 23, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:30:00,347 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151295.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:01,708 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8015, 2.0271, 2.2273, 1.4659, 2.3525, 1.5708, 0.7383, 2.0931], + device='cuda:2'), covar=tensor([0.0547, 0.0342, 0.0255, 0.0523, 0.0348, 0.0794, 0.0761, 0.0254], + device='cuda:2'), in_proj_covar=tensor([0.0443, 0.0381, 0.0331, 0.0438, 0.0367, 0.0526, 0.0384, 0.0407], + device='cuda:2'), out_proj_covar=tensor([1.1993e-04, 1.0078e-04, 8.7415e-05, 1.1621e-04, 9.7324e-05, 1.5027e-04, + 1.0398e-04, 1.0879e-04], device='cuda:2') +2023-02-06 23:30:04,360 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151300.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:16,598 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151316.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:34,587 INFO [train.py:901] (2/4) Epoch 19, batch 5850, loss[loss=0.1709, simple_loss=0.256, pruned_loss=0.04286, over 7664.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2919, pruned_loss=0.06488, over 1613014.17 frames. ], batch size: 19, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:30:44,232 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1951, 1.0854, 1.3194, 1.0693, 0.9320, 1.3293, 0.0512, 0.8616], + device='cuda:2'), covar=tensor([0.1894, 0.1407, 0.0542, 0.0911, 0.2884, 0.0609, 0.2255, 0.1246], + device='cuda:2'), in_proj_covar=tensor([0.0188, 0.0196, 0.0124, 0.0222, 0.0270, 0.0134, 0.0169, 0.0189], + 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') +2023-02-06 23:30:57,556 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151377.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:58,661 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.176e+02 2.714e+02 3.221e+02 1.387e+03, threshold=5.429e+02, percent-clipped=3.0 +2023-02-06 23:31:08,074 INFO [train.py:901] (2/4) Epoch 19, batch 5900, loss[loss=0.1869, simple_loss=0.2809, pruned_loss=0.04649, over 8560.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.291, pruned_loss=0.06469, over 1608039.14 frames. ], batch size: 34, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:31:44,895 INFO [train.py:901] (2/4) Epoch 19, batch 5950, loss[loss=0.2232, simple_loss=0.2917, pruned_loss=0.07732, over 7801.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2908, pruned_loss=0.06492, over 1608852.21 frames. ], batch size: 20, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:31:47,006 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5036, 2.2082, 3.2613, 2.6017, 3.0011, 2.4721, 2.2158, 1.7339], + device='cuda:2'), covar=tensor([0.5068, 0.4860, 0.1732, 0.3445, 0.2443, 0.2930, 0.1880, 0.5571], + device='cuda:2'), in_proj_covar=tensor([0.0932, 0.0961, 0.0789, 0.0922, 0.0982, 0.0876, 0.0737, 0.0817], + 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') +2023-02-06 23:31:59,861 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:32:09,188 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.424e+02 3.104e+02 3.851e+02 8.156e+02, threshold=6.208e+02, percent-clipped=3.0 +2023-02-06 23:32:16,816 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:32:18,557 INFO [train.py:901] (2/4) Epoch 19, batch 6000, loss[loss=0.217, simple_loss=0.2873, pruned_loss=0.07335, over 7529.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2915, pruned_loss=0.06576, over 1606616.03 frames. ], batch size: 18, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:32:18,558 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 23:32:32,010 INFO [train.py:935] (2/4) Epoch 19, validation: loss=0.1763, simple_loss=0.2764, pruned_loss=0.03805, over 944034.00 frames. +2023-02-06 23:32:32,011 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 23:33:06,934 INFO [train.py:901] (2/4) Epoch 19, batch 6050, loss[loss=0.1713, simple_loss=0.264, pruned_loss=0.03924, over 8462.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2924, pruned_loss=0.06581, over 1612948.17 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:33:09,100 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:33:20,065 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 23:33:32,578 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.535e+02 3.172e+02 3.888e+02 8.825e+02, threshold=6.343e+02, percent-clipped=4.0 +2023-02-06 23:33:42,766 INFO [train.py:901] (2/4) Epoch 19, batch 6100, loss[loss=0.2156, simple_loss=0.2958, pruned_loss=0.0677, over 7909.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.294, pruned_loss=0.06623, over 1621357.59 frames. ], batch size: 20, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:34:07,684 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 23:34:10,834 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151633.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:12,958 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2458, 2.5315, 2.9391, 1.5907, 3.1671, 1.8818, 1.5707, 2.2183], + device='cuda:2'), covar=tensor([0.0765, 0.0370, 0.0284, 0.0712, 0.0404, 0.0817, 0.0920, 0.0488], + device='cuda:2'), in_proj_covar=tensor([0.0442, 0.0380, 0.0332, 0.0438, 0.0367, 0.0526, 0.0384, 0.0405], + device='cuda:2'), out_proj_covar=tensor([1.1975e-04, 1.0018e-04, 8.7593e-05, 1.1622e-04, 9.7466e-05, 1.4998e-04, + 1.0391e-04, 1.0815e-04], device='cuda:2') +2023-02-06 23:34:17,586 INFO [train.py:901] (2/4) Epoch 19, batch 6150, loss[loss=0.2136, simple_loss=0.2997, pruned_loss=0.06373, over 8508.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2937, pruned_loss=0.06598, over 1622503.28 frames. ], batch size: 28, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:34:18,368 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:28,875 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:30,158 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151660.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:30,966 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:43,579 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.320e+02 2.846e+02 3.654e+02 5.745e+02, threshold=5.693e+02, percent-clipped=0.0 +2023-02-06 23:34:53,937 INFO [train.py:901] (2/4) Epoch 19, batch 6200, loss[loss=0.1926, simple_loss=0.2792, pruned_loss=0.053, over 7922.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2939, pruned_loss=0.06605, over 1622475.29 frames. ], batch size: 20, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:35:02,694 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151706.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:35:28,542 INFO [train.py:901] (2/4) Epoch 19, batch 6250, loss[loss=0.2145, simple_loss=0.2925, pruned_loss=0.06823, over 8754.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2932, pruned_loss=0.06568, over 1620519.32 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:35:37,322 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5908, 1.3590, 1.4996, 1.2720, 0.8644, 1.3074, 1.4585, 1.1127], + device='cuda:2'), covar=tensor([0.0571, 0.1285, 0.1777, 0.1537, 0.0615, 0.1541, 0.0730, 0.0750], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0159, 0.0100, 0.0161, 0.0112, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-06 23:35:39,361 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151759.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:35:50,938 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151775.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:35:53,500 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 2.555e+02 3.246e+02 4.070e+02 8.549e+02, threshold=6.492e+02, percent-clipped=6.0 +2023-02-06 23:36:03,716 INFO [train.py:901] (2/4) Epoch 19, batch 6300, loss[loss=0.227, simple_loss=0.3133, pruned_loss=0.07031, over 8284.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.294, pruned_loss=0.0661, over 1623026.24 frames. ], batch size: 23, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:36:22,214 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:36:39,103 INFO [train.py:901] (2/4) Epoch 19, batch 6350, loss[loss=0.2179, simple_loss=0.2897, pruned_loss=0.07301, over 8367.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.294, pruned_loss=0.06586, over 1625876.89 frames. ], batch size: 24, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:36:40,790 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 23:36:57,097 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8461, 1.9770, 1.7805, 2.4623, 1.3844, 1.5536, 1.9049, 2.1130], + device='cuda:2'), covar=tensor([0.0768, 0.0771, 0.0872, 0.0481, 0.0996, 0.1282, 0.0740, 0.0706], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0198, 0.0248, 0.0213, 0.0205, 0.0250, 0.0254, 0.0210], + 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') +2023-02-06 23:37:03,135 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.433e+02 2.376e+02 2.921e+02 3.593e+02 6.855e+02, threshold=5.841e+02, percent-clipped=1.0 +2023-02-06 23:37:13,200 INFO [train.py:901] (2/4) Epoch 19, batch 6400, loss[loss=0.2415, simple_loss=0.322, pruned_loss=0.08046, over 8632.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.292, pruned_loss=0.06521, over 1619752.82 frames. ], batch size: 34, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:37:30,637 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:37:48,045 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151942.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:37:48,525 INFO [train.py:901] (2/4) Epoch 19, batch 6450, loss[loss=0.2134, simple_loss=0.2964, pruned_loss=0.06522, over 8245.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2926, pruned_loss=0.0654, over 1621863.69 frames. ], batch size: 24, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:38:13,512 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.409e+02 2.943e+02 3.710e+02 6.232e+02, threshold=5.887e+02, percent-clipped=1.0 +2023-02-06 23:38:19,954 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5835, 1.9771, 3.1116, 1.4891, 2.2804, 2.0433, 1.7352, 2.2628], + device='cuda:2'), covar=tensor([0.1869, 0.2501, 0.0844, 0.4348, 0.1856, 0.3143, 0.2212, 0.2366], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0581, 0.0551, 0.0629, 0.0640, 0.0587, 0.0523, 0.0628], + 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') +2023-02-06 23:38:23,087 INFO [train.py:901] (2/4) Epoch 19, batch 6500, loss[loss=0.1855, simple_loss=0.273, pruned_loss=0.04895, over 7825.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2913, pruned_loss=0.06486, over 1616313.63 frames. ], batch size: 20, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:38:40,009 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:38:51,615 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:38:58,599 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152040.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:39:00,457 INFO [train.py:901] (2/4) Epoch 19, batch 6550, loss[loss=0.1709, simple_loss=0.2494, pruned_loss=0.04622, over 7697.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2907, pruned_loss=0.06462, over 1615164.41 frames. ], batch size: 18, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:39:04,936 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152050.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:39:09,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152056.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:39:21,605 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 23:39:24,910 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.379e+02 2.761e+02 3.695e+02 7.678e+02, threshold=5.522e+02, percent-clipped=3.0 +2023-02-06 23:39:34,322 INFO [train.py:901] (2/4) Epoch 19, batch 6600, loss[loss=0.246, simple_loss=0.3241, pruned_loss=0.0839, over 6838.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2922, pruned_loss=0.06567, over 1613360.64 frames. ], batch size: 72, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:39:36,561 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 23:39:39,626 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 23:39:59,802 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.80 vs. limit=5.0 +2023-02-06 23:40:09,007 INFO [train.py:901] (2/4) Epoch 19, batch 6650, loss[loss=0.1917, simple_loss=0.2805, pruned_loss=0.05148, over 8189.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.293, pruned_loss=0.0661, over 1614883.88 frames. ], batch size: 23, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:40:19,001 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 23:40:23,464 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:40:24,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:40:32,971 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2207, 1.6683, 4.3357, 2.0454, 2.4450, 4.9385, 4.9486, 4.2435], + device='cuda:2'), covar=tensor([0.1277, 0.1693, 0.0335, 0.1957, 0.1210, 0.0189, 0.0451, 0.0584], + device='cuda:2'), in_proj_covar=tensor([0.0290, 0.0317, 0.0283, 0.0309, 0.0301, 0.0260, 0.0404, 0.0304], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:40:34,183 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.686e+02 3.265e+02 3.895e+02 8.931e+02, threshold=6.531e+02, percent-clipped=7.0 +2023-02-06 23:40:44,518 INFO [train.py:901] (2/4) Epoch 19, batch 6700, loss[loss=0.1982, simple_loss=0.2881, pruned_loss=0.05411, over 8329.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2923, pruned_loss=0.0652, over 1616674.27 frames. ], batch size: 25, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:40:53,760 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 23:41:19,449 INFO [train.py:901] (2/4) Epoch 19, batch 6750, loss[loss=0.1656, simple_loss=0.2484, pruned_loss=0.0414, over 7442.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2916, pruned_loss=0.06473, over 1619251.63 frames. ], batch size: 17, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:41:19,592 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:41:41,477 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-06 23:41:44,675 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:41:45,118 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.378e+02 2.909e+02 3.491e+02 6.752e+02, threshold=5.817e+02, percent-clipped=2.0 +2023-02-06 23:41:53,507 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:41:54,062 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 23:41:54,741 INFO [train.py:901] (2/4) Epoch 19, batch 6800, loss[loss=0.1797, simple_loss=0.265, pruned_loss=0.04718, over 8236.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2914, pruned_loss=0.06422, over 1621229.37 frames. ], batch size: 22, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:42:29,085 INFO [train.py:901] (2/4) Epoch 19, batch 6850, loss[loss=0.1932, simple_loss=0.2706, pruned_loss=0.05791, over 7822.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2911, pruned_loss=0.06402, over 1616128.69 frames. ], batch size: 20, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:42:43,978 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 23:42:54,734 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.344e+02 3.012e+02 3.839e+02 8.073e+02, threshold=6.025e+02, percent-clipped=5.0 +2023-02-06 23:43:05,119 INFO [train.py:901] (2/4) Epoch 19, batch 6900, loss[loss=0.2062, simple_loss=0.2763, pruned_loss=0.06801, over 7653.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2905, pruned_loss=0.06366, over 1615559.76 frames. ], batch size: 19, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:43:17,188 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8134, 1.5839, 3.4516, 1.4869, 2.4218, 3.9168, 3.9649, 3.2756], + device='cuda:2'), covar=tensor([0.1403, 0.1747, 0.0353, 0.2127, 0.1060, 0.0197, 0.0501, 0.0605], + device='cuda:2'), in_proj_covar=tensor([0.0291, 0.0319, 0.0285, 0.0311, 0.0301, 0.0261, 0.0407, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:43:18,645 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3910, 1.3713, 2.3418, 1.1883, 2.1393, 2.5185, 2.6574, 2.1088], + device='cuda:2'), covar=tensor([0.1200, 0.1379, 0.0478, 0.2222, 0.0787, 0.0400, 0.0740, 0.0791], + device='cuda:2'), in_proj_covar=tensor([0.0291, 0.0319, 0.0285, 0.0311, 0.0301, 0.0261, 0.0406, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:43:25,516 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152421.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:43:27,852 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 +2023-02-06 23:43:40,394 INFO [train.py:901] (2/4) Epoch 19, batch 6950, loss[loss=0.1927, simple_loss=0.279, pruned_loss=0.05326, over 8355.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2902, pruned_loss=0.06379, over 1613484.41 frames. ], batch size: 24, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:43:42,607 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152446.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:43:53,798 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 23:43:53,936 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152463.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:43:58,611 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5237, 1.2566, 4.7246, 1.7499, 4.1523, 3.9040, 4.1983, 4.1145], + device='cuda:2'), covar=tensor([0.0618, 0.5003, 0.0461, 0.4317, 0.1172, 0.0975, 0.0656, 0.0671], + device='cuda:2'), in_proj_covar=tensor([0.0594, 0.0626, 0.0668, 0.0604, 0.0684, 0.0585, 0.0584, 0.0646], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 23:44:05,256 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.443e+02 3.132e+02 3.706e+02 6.613e+02, threshold=6.264e+02, percent-clipped=2.0 +2023-02-06 23:44:08,852 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8861, 2.3752, 3.6966, 1.8910, 1.9079, 3.7044, 0.7420, 2.1043], + device='cuda:2'), covar=tensor([0.1424, 0.1189, 0.0177, 0.1905, 0.2766, 0.0210, 0.2489, 0.1488], + device='cuda:2'), in_proj_covar=tensor([0.0186, 0.0193, 0.0123, 0.0220, 0.0266, 0.0132, 0.0168, 0.0185], + 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') +2023-02-06 23:44:14,626 INFO [train.py:901] (2/4) Epoch 19, batch 7000, loss[loss=0.2006, simple_loss=0.2866, pruned_loss=0.05731, over 8343.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2897, pruned_loss=0.06361, over 1608934.55 frames. ], batch size: 26, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:44:44,332 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152534.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:44:51,099 INFO [train.py:901] (2/4) Epoch 19, batch 7050, loss[loss=0.201, simple_loss=0.2885, pruned_loss=0.05676, over 8665.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2908, pruned_loss=0.06425, over 1612582.01 frames. ], batch size: 39, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:45:02,279 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152559.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:15,716 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.420e+02 2.800e+02 3.429e+02 5.549e+02, threshold=5.599e+02, percent-clipped=0.0 +2023-02-06 23:45:21,280 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:25,390 INFO [train.py:901] (2/4) Epoch 19, batch 7100, loss[loss=0.2199, simple_loss=0.2904, pruned_loss=0.07471, over 8047.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2908, pruned_loss=0.06413, over 1608904.56 frames. ], batch size: 22, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:45:30,673 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:35,440 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:56,367 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:46:01,876 INFO [train.py:901] (2/4) Epoch 19, batch 7150, loss[loss=0.1985, simple_loss=0.2736, pruned_loss=0.06172, over 7804.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2909, pruned_loss=0.06456, over 1608574.55 frames. ], batch size: 20, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:46:02,142 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7005, 2.3577, 4.2724, 1.5410, 3.1867, 2.1848, 1.8379, 2.8865], + device='cuda:2'), covar=tensor([0.1922, 0.2561, 0.0816, 0.4624, 0.1685, 0.3305, 0.2196, 0.2549], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0585, 0.0553, 0.0628, 0.0640, 0.0589, 0.0522, 0.0631], + 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') +2023-02-06 23:46:27,015 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 23:46:27,178 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.485e+02 2.441e+02 2.885e+02 3.630e+02 1.043e+03, threshold=5.770e+02, percent-clipped=5.0 +2023-02-06 23:46:36,619 INFO [train.py:901] (2/4) Epoch 19, batch 7200, loss[loss=0.227, simple_loss=0.3123, pruned_loss=0.07085, over 8534.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2917, pruned_loss=0.06471, over 1606786.75 frames. ], batch size: 28, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:46:42,742 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:46:56,239 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3993, 1.6339, 1.7201, 1.1059, 1.7403, 1.3804, 0.2900, 1.5911], + device='cuda:2'), covar=tensor([0.0428, 0.0347, 0.0295, 0.0443, 0.0381, 0.0848, 0.0788, 0.0233], + device='cuda:2'), in_proj_covar=tensor([0.0445, 0.0380, 0.0335, 0.0443, 0.0369, 0.0530, 0.0387, 0.0410], + device='cuda:2'), out_proj_covar=tensor([1.2024e-04, 1.0022e-04, 8.8461e-05, 1.1754e-04, 9.7831e-05, 1.5109e-04, + 1.0467e-04, 1.0941e-04], device='cuda:2') +2023-02-06 23:47:12,600 INFO [train.py:901] (2/4) Epoch 19, batch 7250, loss[loss=0.1913, simple_loss=0.263, pruned_loss=0.05975, over 7534.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2903, pruned_loss=0.06436, over 1605572.28 frames. ], batch size: 18, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:47:13,474 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152744.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:47:17,697 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:47:37,386 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 2.392e+02 2.877e+02 3.488e+02 7.359e+02, threshold=5.753e+02, percent-clipped=2.0 +2023-02-06 23:47:47,609 INFO [train.py:901] (2/4) Epoch 19, batch 7300, loss[loss=0.1631, simple_loss=0.2434, pruned_loss=0.04143, over 7434.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.06364, over 1608939.69 frames. ], batch size: 17, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:47:48,531 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4651, 1.6368, 2.1594, 1.3403, 1.4639, 1.6084, 1.5693, 1.4184], + device='cuda:2'), covar=tensor([0.2049, 0.2368, 0.1001, 0.4319, 0.1961, 0.3551, 0.2283, 0.2270], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0581, 0.0551, 0.0624, 0.0637, 0.0586, 0.0519, 0.0627], + 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') +2023-02-06 23:47:57,327 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:48:21,874 INFO [train.py:901] (2/4) Epoch 19, batch 7350, loss[loss=0.2632, simple_loss=0.3226, pruned_loss=0.1019, over 7663.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2903, pruned_loss=0.0643, over 1606721.73 frames. ], batch size: 71, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:48:22,993 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 23:48:45,481 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1946, 4.1995, 3.7951, 1.9118, 3.6772, 3.8139, 3.7765, 3.5244], + device='cuda:2'), covar=tensor([0.0856, 0.0601, 0.1114, 0.4816, 0.0922, 0.1038, 0.1318, 0.0902], + device='cuda:2'), in_proj_covar=tensor([0.0517, 0.0427, 0.0427, 0.0531, 0.0416, 0.0428, 0.0411, 0.0372], + 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') +2023-02-06 23:48:46,698 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 23:48:48,153 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.571e+02 3.070e+02 4.184e+02 8.940e+02, threshold=6.140e+02, percent-clipped=8.0 +2023-02-06 23:48:58,039 INFO [train.py:901] (2/4) Epoch 19, batch 7400, loss[loss=0.1849, simple_loss=0.2687, pruned_loss=0.05055, over 7655.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2896, pruned_loss=0.06413, over 1605458.10 frames. ], batch size: 19, lr: 3.95e-03, grad_scale: 32.0 +2023-02-06 23:49:07,685 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 23:49:17,393 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4392, 2.5982, 1.8142, 2.1399, 2.1784, 1.4719, 1.9524, 2.0580], + device='cuda:2'), covar=tensor([0.1568, 0.0402, 0.1259, 0.0731, 0.0778, 0.1579, 0.1166, 0.1199], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0233, 0.0328, 0.0304, 0.0299, 0.0330, 0.0342, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:49:18,786 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:25,744 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-06 23:49:32,918 INFO [train.py:901] (2/4) Epoch 19, batch 7450, loss[loss=0.2025, simple_loss=0.2919, pruned_loss=0.05658, over 8351.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2906, pruned_loss=0.06414, over 1608764.40 frames. ], batch size: 24, lr: 3.95e-03, grad_scale: 32.0 +2023-02-06 23:49:33,722 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:38,519 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:44,022 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152958.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:46,567 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 23:49:58,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.506e+02 3.079e+02 4.075e+02 8.166e+02, threshold=6.159e+02, percent-clipped=5.0 +2023-02-06 23:50:01,976 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:08,330 INFO [train.py:901] (2/4) Epoch 19, batch 7500, loss[loss=0.226, simple_loss=0.303, pruned_loss=0.07454, over 8192.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2916, pruned_loss=0.06471, over 1609685.36 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:50:17,498 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:34,844 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:42,932 INFO [train.py:901] (2/4) Epoch 19, batch 7550, loss[loss=0.2264, simple_loss=0.3021, pruned_loss=0.07531, over 6941.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2918, pruned_loss=0.06509, over 1610612.81 frames. ], batch size: 72, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:50:53,932 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:58,702 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:51:08,488 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.431e+02 2.980e+02 3.688e+02 7.634e+02, threshold=5.960e+02, percent-clipped=2.0 +2023-02-06 23:51:14,777 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153088.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:51:18,071 INFO [train.py:901] (2/4) Epoch 19, batch 7600, loss[loss=0.1976, simple_loss=0.2787, pruned_loss=0.05831, over 7439.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2915, pruned_loss=0.06495, over 1606814.61 frames. ], batch size: 17, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:51:37,054 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7022, 2.3814, 3.7107, 1.4969, 2.8585, 2.0608, 1.9600, 2.4001], + device='cuda:2'), covar=tensor([0.1962, 0.2331, 0.1006, 0.4578, 0.1836, 0.3598, 0.2113, 0.2871], + device='cuda:2'), in_proj_covar=tensor([0.0519, 0.0588, 0.0557, 0.0632, 0.0643, 0.0592, 0.0523, 0.0634], + 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') +2023-02-06 23:51:42,024 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 23:51:47,771 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:51:53,092 INFO [train.py:901] (2/4) Epoch 19, batch 7650, loss[loss=0.2186, simple_loss=0.2958, pruned_loss=0.07073, over 7972.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2922, pruned_loss=0.06568, over 1608321.48 frames. ], batch size: 21, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:51:59,413 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 23:52:00,688 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2023-02-06 23:52:17,643 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:52:18,733 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.290e+02 2.780e+02 3.362e+02 7.829e+02, threshold=5.561e+02, percent-clipped=2.0 +2023-02-06 23:52:28,393 INFO [train.py:901] (2/4) Epoch 19, batch 7700, loss[loss=0.2203, simple_loss=0.2985, pruned_loss=0.07102, over 8289.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2913, pruned_loss=0.06498, over 1609301.69 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:52:35,482 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:52:35,506 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:52:46,188 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 23:52:57,461 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 23:53:03,345 INFO [train.py:901] (2/4) Epoch 19, batch 7750, loss[loss=0.1907, simple_loss=0.2645, pruned_loss=0.05849, over 7228.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2919, pruned_loss=0.06547, over 1609271.58 frames. ], batch size: 16, lr: 3.94e-03, grad_scale: 16.0 +2023-02-06 23:53:18,703 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 23:53:28,921 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.456e+02 3.001e+02 3.725e+02 8.940e+02, threshold=6.003e+02, percent-clipped=11.0 +2023-02-06 23:53:37,740 INFO [train.py:901] (2/4) Epoch 19, batch 7800, loss[loss=0.2148, simple_loss=0.302, pruned_loss=0.06382, over 8188.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2917, pruned_loss=0.06522, over 1610504.01 frames. ], batch size: 23, lr: 3.94e-03, grad_scale: 16.0 +2023-02-06 23:53:39,909 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153296.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:53:53,359 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153315.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:53:58,000 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:54:09,674 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153340.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:54:11,468 INFO [train.py:901] (2/4) Epoch 19, batch 7850, loss[loss=0.2021, simple_loss=0.294, pruned_loss=0.05514, over 8359.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2902, pruned_loss=0.06446, over 1607866.49 frames. ], batch size: 24, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:54:14,365 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:54:36,618 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.457e+02 2.874e+02 3.581e+02 1.670e+03, threshold=5.749e+02, percent-clipped=9.0 +2023-02-06 23:54:44,306 INFO [train.py:901] (2/4) Epoch 19, batch 7900, loss[loss=0.1778, simple_loss=0.2566, pruned_loss=0.04944, over 7229.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2884, pruned_loss=0.06349, over 1602553.48 frames. ], batch size: 16, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:55:08,556 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6150, 1.7496, 2.3217, 1.3468, 1.1801, 2.3268, 0.3621, 1.3317], + device='cuda:2'), covar=tensor([0.2148, 0.1189, 0.0317, 0.1625, 0.3081, 0.0344, 0.2208, 0.1413], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0195, 0.0125, 0.0224, 0.0272, 0.0134, 0.0171, 0.0187], + 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') +2023-02-06 23:55:13,049 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6946, 1.5173, 4.9147, 1.7966, 4.3243, 4.0971, 4.4098, 4.2727], + device='cuda:2'), covar=tensor([0.0622, 0.4871, 0.0446, 0.4096, 0.1060, 0.0923, 0.0541, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0605, 0.0637, 0.0678, 0.0611, 0.0693, 0.0594, 0.0593, 0.0654], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 23:55:15,517 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:17,979 INFO [train.py:901] (2/4) Epoch 19, batch 7950, loss[loss=0.206, simple_loss=0.2966, pruned_loss=0.05771, over 8251.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2891, pruned_loss=0.06356, over 1607952.24 frames. ], batch size: 24, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:55:28,827 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:38,851 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2221, 1.9655, 2.7294, 2.2117, 2.6308, 2.2267, 2.0058, 1.4985], + device='cuda:2'), covar=tensor([0.5739, 0.5232, 0.1837, 0.3643, 0.2519, 0.3177, 0.2073, 0.5345], + device='cuda:2'), in_proj_covar=tensor([0.0926, 0.0955, 0.0783, 0.0922, 0.0981, 0.0867, 0.0736, 0.0809], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-06 23:55:41,873 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:43,176 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.434e+02 3.034e+02 3.983e+02 8.510e+02, threshold=6.068e+02, percent-clipped=6.0 +2023-02-06 23:55:43,590 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.09 vs. limit=5.0 +2023-02-06 23:55:45,342 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153484.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:49,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3200, 1.3015, 2.1426, 1.1673, 1.8730, 2.2677, 2.3603, 1.9529], + device='cuda:2'), covar=tensor([0.1031, 0.1259, 0.0464, 0.1898, 0.0898, 0.0384, 0.0720, 0.0640], + device='cuda:2'), in_proj_covar=tensor([0.0290, 0.0318, 0.0286, 0.0312, 0.0303, 0.0262, 0.0406, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:55:51,002 INFO [train.py:901] (2/4) Epoch 19, batch 8000, loss[loss=0.2225, simple_loss=0.3029, pruned_loss=0.07107, over 8251.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2899, pruned_loss=0.0639, over 1611409.32 frames. ], batch size: 24, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:56:25,144 INFO [train.py:901] (2/4) Epoch 19, batch 8050, loss[loss=0.1983, simple_loss=0.2678, pruned_loss=0.06445, over 7422.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2889, pruned_loss=0.064, over 1602472.44 frames. ], batch size: 17, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:56:27,180 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7539, 4.6643, 4.2671, 3.1522, 4.2026, 4.2417, 4.4392, 3.9136], + device='cuda:2'), covar=tensor([0.0566, 0.0453, 0.0898, 0.3192, 0.0718, 0.1023, 0.1057, 0.0895], + device='cuda:2'), in_proj_covar=tensor([0.0518, 0.0428, 0.0427, 0.0534, 0.0420, 0.0432, 0.0416, 0.0373], + 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') +2023-02-06 23:56:34,265 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 23:56:59,236 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 23:57:04,931 INFO [train.py:901] (2/4) Epoch 20, batch 0, loss[loss=0.1976, simple_loss=0.2768, pruned_loss=0.05918, over 7801.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2768, pruned_loss=0.05918, over 7801.00 frames. ], batch size: 19, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:57:04,931 INFO [train.py:926] (2/4) Computing validation loss +2023-02-06 23:57:12,510 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7191, 1.4997, 3.8866, 1.5686, 3.4355, 3.1864, 3.5413, 3.3766], + device='cuda:2'), covar=tensor([0.0744, 0.4588, 0.0523, 0.4158, 0.1139, 0.1044, 0.0703, 0.0788], + device='cuda:2'), in_proj_covar=tensor([0.0606, 0.0638, 0.0678, 0.0612, 0.0692, 0.0593, 0.0591, 0.0655], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-06 23:57:16,941 INFO [train.py:935] (2/4) Epoch 20, validation: loss=0.1757, simple_loss=0.276, pruned_loss=0.03766, over 944034.00 frames. +2023-02-06 23:57:16,942 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-06 23:57:20,454 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.577e+02 3.496e+02 4.495e+02 1.164e+03, threshold=6.992e+02, percent-clipped=12.0 +2023-02-06 23:57:29,442 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:57:31,330 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 23:57:51,322 INFO [train.py:901] (2/4) Epoch 20, batch 50, loss[loss=0.2204, simple_loss=0.304, pruned_loss=0.06845, over 8294.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2941, pruned_loss=0.06348, over 367979.95 frames. ], batch size: 23, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:58:01,109 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153640.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:58:06,571 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 23:58:26,060 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1196, 2.0976, 1.5196, 1.8537, 1.5779, 1.2142, 1.4303, 1.6452], + device='cuda:2'), covar=tensor([0.1616, 0.0529, 0.1510, 0.0648, 0.0938, 0.1978, 0.1243, 0.0987], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0236, 0.0332, 0.0305, 0.0300, 0.0334, 0.0344, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:58:27,837 INFO [train.py:901] (2/4) Epoch 20, batch 100, loss[loss=0.2534, simple_loss=0.3335, pruned_loss=0.08663, over 8290.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2941, pruned_loss=0.06485, over 648351.16 frames. ], batch size: 23, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:58:29,260 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 23:58:31,349 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.446e+02 2.844e+02 3.351e+02 7.473e+02, threshold=5.688e+02, percent-clipped=2.0 +2023-02-06 23:59:03,151 INFO [train.py:901] (2/4) Epoch 20, batch 150, loss[loss=0.2124, simple_loss=0.3016, pruned_loss=0.06164, over 8200.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2939, pruned_loss=0.06456, over 866937.96 frames. ], batch size: 23, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:59:06,099 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5031, 2.6234, 1.9276, 2.3146, 2.2093, 1.6485, 2.1461, 2.1641], + device='cuda:2'), covar=tensor([0.1318, 0.0373, 0.1140, 0.0545, 0.0620, 0.1452, 0.0849, 0.0881], + device='cuda:2'), in_proj_covar=tensor([0.0349, 0.0234, 0.0329, 0.0303, 0.0298, 0.0331, 0.0341, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-06 23:59:23,326 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:59:26,119 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5419, 1.5177, 1.8564, 1.3338, 1.1501, 1.8470, 0.2538, 1.1746], + device='cuda:2'), covar=tensor([0.1823, 0.1307, 0.0371, 0.0986, 0.3005, 0.0418, 0.2265, 0.1141], + device='cuda:2'), in_proj_covar=tensor([0.0187, 0.0194, 0.0124, 0.0221, 0.0270, 0.0133, 0.0169, 0.0186], + 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') +2023-02-06 23:59:39,296 INFO [train.py:901] (2/4) Epoch 20, batch 200, loss[loss=0.1752, simple_loss=0.2701, pruned_loss=0.0402, over 8108.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2923, pruned_loss=0.06357, over 1032415.76 frames. ], batch size: 23, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:59:42,414 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.84 vs. limit=5.0 +2023-02-06 23:59:42,528 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.177e+02 2.784e+02 3.416e+02 8.818e+02, threshold=5.569e+02, percent-clipped=1.0 +2023-02-06 23:59:43,921 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:00:15,026 INFO [train.py:901] (2/4) Epoch 20, batch 250, loss[loss=0.2065, simple_loss=0.2871, pruned_loss=0.06295, over 8464.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2912, pruned_loss=0.06374, over 1160675.15 frames. ], batch size: 25, lr: 3.84e-03, grad_scale: 8.0 +2023-02-07 00:00:26,528 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 00:00:31,635 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153850.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:00:34,732 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 00:00:48,282 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153875.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:00:48,774 INFO [train.py:901] (2/4) Epoch 20, batch 300, loss[loss=0.1855, simple_loss=0.2614, pruned_loss=0.05485, over 7663.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2913, pruned_loss=0.0639, over 1260729.13 frames. ], batch size: 19, lr: 3.84e-03, grad_scale: 8.0 +2023-02-07 00:00:52,001 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.425e+02 2.846e+02 3.739e+02 1.062e+03, threshold=5.691e+02, percent-clipped=2.0 +2023-02-07 00:01:05,174 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153898.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:01:24,550 INFO [train.py:901] (2/4) Epoch 20, batch 350, loss[loss=0.202, simple_loss=0.2956, pruned_loss=0.0542, over 8259.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2907, pruned_loss=0.06371, over 1340967.73 frames. ], batch size: 24, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:01:35,765 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153941.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:01:36,640 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 00:01:59,289 INFO [train.py:901] (2/4) Epoch 20, batch 400, loss[loss=0.222, simple_loss=0.3063, pruned_loss=0.06889, over 8596.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2929, pruned_loss=0.06501, over 1399792.87 frames. ], batch size: 31, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:02:02,805 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.483e+02 2.937e+02 3.652e+02 9.410e+02, threshold=5.874e+02, percent-clipped=4.0 +2023-02-07 00:02:22,963 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 00:02:25,626 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154011.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:02:36,308 INFO [train.py:901] (2/4) Epoch 20, batch 450, loss[loss=0.1901, simple_loss=0.2671, pruned_loss=0.05657, over 7808.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2915, pruned_loss=0.06424, over 1449397.54 frames. ], batch size: 20, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:02:44,056 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154036.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:03:05,787 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154067.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:03:11,784 INFO [train.py:901] (2/4) Epoch 20, batch 500, loss[loss=0.204, simple_loss=0.2871, pruned_loss=0.06041, over 8328.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2913, pruned_loss=0.06435, over 1489543.93 frames. ], batch size: 26, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:03:15,235 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 2.274e+02 2.685e+02 3.204e+02 7.760e+02, threshold=5.371e+02, percent-clipped=3.0 +2023-02-07 00:03:24,419 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:03:29,960 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154102.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:03:46,396 INFO [train.py:901] (2/4) Epoch 20, batch 550, loss[loss=0.22, simple_loss=0.2989, pruned_loss=0.07057, over 8488.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2901, pruned_loss=0.06354, over 1513931.08 frames. ], batch size: 29, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:03:49,712 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-07 00:03:59,236 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-07 00:04:07,821 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154154.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:04:23,276 INFO [train.py:901] (2/4) Epoch 20, batch 600, loss[loss=0.2218, simple_loss=0.2874, pruned_loss=0.07816, over 7701.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2909, pruned_loss=0.06462, over 1535015.87 frames. ], batch size: 18, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:04:25,542 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154179.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:04:26,652 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.477e+02 2.962e+02 3.836e+02 8.919e+02, threshold=5.925e+02, percent-clipped=6.0 +2023-02-07 00:04:45,283 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 00:04:57,544 INFO [train.py:901] (2/4) Epoch 20, batch 650, loss[loss=0.1915, simple_loss=0.2895, pruned_loss=0.04674, over 8456.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2915, pruned_loss=0.06505, over 1554966.41 frames. ], batch size: 25, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:04:58,453 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6336, 2.6557, 1.8109, 2.3762, 2.3003, 1.5438, 2.2293, 2.3238], + device='cuda:2'), covar=tensor([0.1563, 0.0404, 0.1277, 0.0678, 0.0760, 0.1605, 0.0960, 0.1006], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0237, 0.0332, 0.0307, 0.0302, 0.0335, 0.0346, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 00:05:06,564 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:05:14,624 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-07 00:05:34,064 INFO [train.py:901] (2/4) Epoch 20, batch 700, loss[loss=0.2193, simple_loss=0.3137, pruned_loss=0.06241, over 8129.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2913, pruned_loss=0.06461, over 1569102.91 frames. ], batch size: 22, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:05:37,467 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.357e+02 2.958e+02 3.586e+02 6.466e+02, threshold=5.915e+02, percent-clipped=2.0 +2023-02-07 00:05:40,241 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154285.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:05:54,042 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154304.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:05:57,544 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:06:08,843 INFO [train.py:901] (2/4) Epoch 20, batch 750, loss[loss=0.22, simple_loss=0.3104, pruned_loss=0.06482, over 8295.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2922, pruned_loss=0.0648, over 1583288.53 frames. ], batch size: 23, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:06:11,930 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 00:06:28,500 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154355.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:06:33,643 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 00:06:43,001 INFO [train.py:901] (2/4) Epoch 20, batch 800, loss[loss=0.1934, simple_loss=0.2818, pruned_loss=0.05253, over 8479.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2919, pruned_loss=0.06475, over 1591719.18 frames. ], batch size: 29, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:06:43,740 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 00:06:47,165 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.441e+02 3.052e+02 3.711e+02 8.675e+02, threshold=6.104e+02, percent-clipped=3.0 +2023-02-07 00:07:01,172 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:07:08,303 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154411.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:07:19,173 INFO [train.py:901] (2/4) Epoch 20, batch 850, loss[loss=0.2279, simple_loss=0.2994, pruned_loss=0.07819, over 8712.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2919, pruned_loss=0.06537, over 1597183.98 frames. ], batch size: 30, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:07:27,235 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154438.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:07:32,660 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154446.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:07:52,894 INFO [train.py:901] (2/4) Epoch 20, batch 900, loss[loss=0.2121, simple_loss=0.2944, pruned_loss=0.06489, over 8640.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2919, pruned_loss=0.06485, over 1605852.80 frames. ], batch size: 34, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:07:56,215 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.439e+02 2.923e+02 3.686e+02 1.072e+03, threshold=5.846e+02, percent-clipped=2.0 +2023-02-07 00:07:57,784 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1989, 2.5444, 2.0263, 2.9774, 1.5644, 1.8387, 2.0912, 2.4693], + device='cuda:2'), covar=tensor([0.0661, 0.0719, 0.0817, 0.0331, 0.1015, 0.1180, 0.0886, 0.0693], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0195, 0.0246, 0.0209, 0.0205, 0.0248, 0.0249, 0.0208], + 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') +2023-02-07 00:07:59,841 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4383, 2.1758, 3.2595, 2.6160, 2.9901, 2.2206, 2.1335, 2.2935], + device='cuda:2'), covar=tensor([0.4459, 0.4953, 0.1683, 0.2912, 0.2384, 0.3427, 0.2299, 0.3972], + device='cuda:2'), in_proj_covar=tensor([0.0930, 0.0960, 0.0786, 0.0924, 0.0978, 0.0871, 0.0734, 0.0813], + 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') +2023-02-07 00:08:04,403 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:08:29,162 INFO [train.py:901] (2/4) Epoch 20, batch 950, loss[loss=0.2148, simple_loss=0.2913, pruned_loss=0.06916, over 8286.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2929, pruned_loss=0.06523, over 1611469.28 frames. ], batch size: 23, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:08:29,372 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154526.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:08:48,739 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154553.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:08:54,164 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154561.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:00,292 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0726, 1.6629, 4.0500, 1.9264, 2.4027, 4.6178, 4.7444, 3.9714], + device='cuda:2'), covar=tensor([0.1281, 0.1918, 0.0351, 0.2060, 0.1366, 0.0225, 0.0454, 0.0621], + device='cuda:2'), in_proj_covar=tensor([0.0288, 0.0318, 0.0286, 0.0310, 0.0302, 0.0259, 0.0403, 0.0303], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 00:09:01,511 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 00:09:04,233 INFO [train.py:901] (2/4) Epoch 20, batch 1000, loss[loss=0.1994, simple_loss=0.2727, pruned_loss=0.06308, over 7648.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2911, pruned_loss=0.06431, over 1611033.69 frames. ], batch size: 19, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:09:07,495 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.511e+02 3.044e+02 3.807e+02 8.767e+02, threshold=6.087e+02, percent-clipped=2.0 +2023-02-07 00:09:08,976 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:35,150 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 00:09:38,960 INFO [train.py:901] (2/4) Epoch 20, batch 1050, loss[loss=0.2221, simple_loss=0.3011, pruned_loss=0.07158, over 8450.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2908, pruned_loss=0.06458, over 1608152.82 frames. ], batch size: 27, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:09:49,447 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 00:09:52,626 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2023-02-07 00:09:53,674 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154646.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:54,884 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154648.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:59,107 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:01,455 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154656.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:14,711 INFO [train.py:901] (2/4) Epoch 20, batch 1100, loss[loss=0.1894, simple_loss=0.2545, pruned_loss=0.06215, over 7433.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2909, pruned_loss=0.06476, over 1610691.90 frames. ], batch size: 17, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:10:18,092 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.486e+02 3.103e+02 3.988e+02 8.246e+02, threshold=6.206e+02, percent-clipped=6.0 +2023-02-07 00:10:18,331 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154681.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:29,758 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:30,330 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154699.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:10:45,302 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 00:10:48,873 INFO [train.py:901] (2/4) Epoch 20, batch 1150, loss[loss=0.2103, simple_loss=0.3027, pruned_loss=0.05892, over 8366.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2916, pruned_loss=0.06443, over 1616073.85 frames. ], batch size: 24, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:10:57,852 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154738.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:59,076 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 00:11:10,831 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4990, 1.8703, 3.1520, 1.4513, 2.4122, 1.8722, 1.6422, 2.4229], + device='cuda:2'), covar=tensor([0.1899, 0.2587, 0.0806, 0.4326, 0.1663, 0.3302, 0.2159, 0.2087], + device='cuda:2'), in_proj_covar=tensor([0.0522, 0.0588, 0.0554, 0.0632, 0.0643, 0.0590, 0.0525, 0.0631], + 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') +2023-02-07 00:11:16,270 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154763.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:19,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:25,036 INFO [train.py:901] (2/4) Epoch 20, batch 1200, loss[loss=0.2036, simple_loss=0.2885, pruned_loss=0.05936, over 8718.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2913, pruned_loss=0.06437, over 1614434.08 frames. ], batch size: 34, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:11:28,382 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.412e+02 2.746e+02 3.577e+02 9.067e+02, threshold=5.492e+02, percent-clipped=2.0 +2023-02-07 00:11:29,309 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154782.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:11:46,343 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154807.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:11:47,771 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154809.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:51,160 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154814.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:11:53,284 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154817.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:59,116 INFO [train.py:901] (2/4) Epoch 20, batch 1250, loss[loss=0.2083, simple_loss=0.3027, pruned_loss=0.05689, over 8190.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2918, pruned_loss=0.06451, over 1616263.81 frames. ], batch size: 23, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:12:05,338 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154834.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:12:06,466 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154836.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:12:11,372 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:12:34,985 INFO [train.py:901] (2/4) Epoch 20, batch 1300, loss[loss=0.2327, simple_loss=0.311, pruned_loss=0.07723, over 8135.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2909, pruned_loss=0.06431, over 1614192.37 frames. ], batch size: 22, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:12:38,325 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.433e+02 3.191e+02 3.995e+02 7.235e+02, threshold=6.381e+02, percent-clipped=6.0 +2023-02-07 00:12:56,079 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5876, 2.2198, 4.1559, 1.5157, 2.9814, 2.0670, 1.7289, 2.8887], + device='cuda:2'), covar=tensor([0.1851, 0.2487, 0.0674, 0.4440, 0.1731, 0.3236, 0.2294, 0.2297], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0588, 0.0553, 0.0633, 0.0644, 0.0590, 0.0527, 0.0631], + 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') +2023-02-07 00:13:01,362 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2123, 3.6785, 2.4407, 3.0756, 3.0682, 2.2537, 2.7450, 3.0873], + device='cuda:2'), covar=tensor([0.1518, 0.0365, 0.1109, 0.0644, 0.0608, 0.1361, 0.0972, 0.1024], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0238, 0.0332, 0.0306, 0.0300, 0.0336, 0.0346, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 00:13:09,378 INFO [train.py:901] (2/4) Epoch 20, batch 1350, loss[loss=0.1895, simple_loss=0.2622, pruned_loss=0.05839, over 7538.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2912, pruned_loss=0.06433, over 1615609.92 frames. ], batch size: 18, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:13:27,091 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154951.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:29,761 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154954.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:44,747 INFO [train.py:901] (2/4) Epoch 20, batch 1400, loss[loss=0.1908, simple_loss=0.2728, pruned_loss=0.05447, over 8033.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2902, pruned_loss=0.06374, over 1616390.89 frames. ], batch size: 22, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:13:47,808 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154979.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:48,964 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.419e+02 2.969e+02 3.620e+02 8.609e+02, threshold=5.938e+02, percent-clipped=3.0 +2023-02-07 00:13:55,292 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:58,165 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154994.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:16,376 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155019.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:19,578 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155024.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:20,735 INFO [train.py:901] (2/4) Epoch 20, batch 1450, loss[loss=0.169, simple_loss=0.2453, pruned_loss=0.04637, over 7706.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.291, pruned_loss=0.06423, over 1617933.03 frames. ], batch size: 18, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:14:29,172 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 00:14:33,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:36,494 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155049.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:48,932 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0208, 1.2461, 1.1907, 0.7940, 1.2641, 1.0526, 0.1157, 1.2257], + device='cuda:2'), covar=tensor([0.0379, 0.0322, 0.0303, 0.0458, 0.0376, 0.0861, 0.0758, 0.0279], + device='cuda:2'), in_proj_covar=tensor([0.0438, 0.0382, 0.0336, 0.0440, 0.0368, 0.0528, 0.0386, 0.0407], + device='cuda:2'), out_proj_covar=tensor([1.1838e-04, 1.0062e-04, 8.8634e-05, 1.1644e-04, 9.7577e-05, 1.5049e-04, + 1.0466e-04, 1.0849e-04], device='cuda:2') +2023-02-07 00:14:50,879 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155070.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:14:55,249 INFO [train.py:901] (2/4) Epoch 20, batch 1500, loss[loss=0.1973, simple_loss=0.2857, pruned_loss=0.05439, over 8029.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2912, pruned_loss=0.06454, over 1616925.64 frames. ], batch size: 22, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:14:58,577 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.504e+02 2.482e+02 3.072e+02 3.822e+02 6.990e+02, threshold=6.143e+02, percent-clipped=2.0 +2023-02-07 00:14:59,298 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155082.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:15:02,792 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155087.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:15:08,997 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155095.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:15:15,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155105.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:15:30,481 INFO [train.py:901] (2/4) Epoch 20, batch 1550, loss[loss=0.1793, simple_loss=0.2603, pruned_loss=0.04911, over 8134.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2892, pruned_loss=0.06327, over 1614705.58 frames. ], batch size: 22, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:16:04,700 INFO [train.py:901] (2/4) Epoch 20, batch 1600, loss[loss=0.1667, simple_loss=0.2516, pruned_loss=0.04095, over 7932.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.289, pruned_loss=0.06294, over 1616953.58 frames. ], batch size: 20, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:16:08,766 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.295e+02 2.863e+02 3.431e+02 6.352e+02, threshold=5.726e+02, percent-clipped=1.0 +2023-02-07 00:16:20,590 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155197.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:16:27,195 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155207.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:16:40,703 INFO [train.py:901] (2/4) Epoch 20, batch 1650, loss[loss=0.1957, simple_loss=0.2721, pruned_loss=0.05959, over 7656.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2891, pruned_loss=0.06317, over 1612507.78 frames. ], batch size: 19, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:16:45,142 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155232.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:17:15,938 INFO [train.py:901] (2/4) Epoch 20, batch 1700, loss[loss=0.2207, simple_loss=0.3055, pruned_loss=0.06799, over 8204.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2887, pruned_loss=0.06277, over 1613338.42 frames. ], batch size: 23, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:17:17,133 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 +2023-02-07 00:17:19,366 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.383e+02 2.759e+02 3.259e+02 7.427e+02, threshold=5.517e+02, percent-clipped=3.0 +2023-02-07 00:17:51,278 INFO [train.py:901] (2/4) Epoch 20, batch 1750, loss[loss=0.173, simple_loss=0.2577, pruned_loss=0.04416, over 7702.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2896, pruned_loss=0.06316, over 1616510.05 frames. ], batch size: 18, lr: 3.82e-03, grad_scale: 16.0 +2023-02-07 00:18:00,382 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155338.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:18:17,135 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155361.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:18:27,006 INFO [train.py:901] (2/4) Epoch 20, batch 1800, loss[loss=0.1711, simple_loss=0.2623, pruned_loss=0.03998, over 7535.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.0636, over 1616477.04 frames. ], batch size: 18, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:18:31,095 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.586e+02 2.965e+02 3.772e+02 7.314e+02, threshold=5.929e+02, percent-clipped=8.0 +2023-02-07 00:18:31,377 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5780, 2.3670, 3.3007, 2.6015, 3.1480, 2.5475, 2.3018, 1.8920], + device='cuda:2'), covar=tensor([0.4971, 0.4850, 0.1650, 0.3540, 0.2475, 0.2879, 0.1804, 0.5312], + device='cuda:2'), in_proj_covar=tensor([0.0941, 0.0971, 0.0795, 0.0935, 0.0994, 0.0884, 0.0743, 0.0823], + 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') +2023-02-07 00:18:34,019 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155386.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:18:36,205 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-07 00:18:51,553 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8704, 1.7701, 5.9703, 2.1807, 5.3388, 5.0219, 5.5175, 5.3702], + device='cuda:2'), covar=tensor([0.0570, 0.4799, 0.0486, 0.3991, 0.1188, 0.0971, 0.0562, 0.0626], + device='cuda:2'), in_proj_covar=tensor([0.0604, 0.0627, 0.0677, 0.0606, 0.0685, 0.0590, 0.0589, 0.0658], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 00:19:01,130 INFO [train.py:901] (2/4) Epoch 20, batch 1850, loss[loss=0.2293, simple_loss=0.2919, pruned_loss=0.08333, over 7531.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2894, pruned_loss=0.06396, over 1612332.35 frames. ], batch size: 18, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:19:04,531 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:20,142 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155453.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:20,185 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155453.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:36,708 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155475.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:37,247 INFO [train.py:901] (2/4) Epoch 20, batch 1900, loss[loss=0.2077, simple_loss=0.2798, pruned_loss=0.06785, over 8082.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2896, pruned_loss=0.06366, over 1614526.24 frames. ], batch size: 21, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:19:38,772 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155478.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:41,327 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.441e+02 2.899e+02 3.473e+02 6.405e+02, threshold=5.799e+02, percent-clipped=1.0 +2023-02-07 00:19:48,155 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 00:19:52,334 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-07 00:20:11,842 INFO [train.py:901] (2/4) Epoch 20, batch 1950, loss[loss=0.2127, simple_loss=0.2915, pruned_loss=0.06691, over 7937.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2895, pruned_loss=0.06359, over 1614792.84 frames. ], batch size: 20, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:20:13,301 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 00:20:20,905 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0889, 1.5927, 3.4416, 1.5240, 2.3933, 3.8708, 3.9313, 3.2629], + device='cuda:2'), covar=tensor([0.1055, 0.1725, 0.0325, 0.2050, 0.1040, 0.0209, 0.0435, 0.0552], + device='cuda:2'), in_proj_covar=tensor([0.0290, 0.0320, 0.0287, 0.0313, 0.0303, 0.0262, 0.0407, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 00:20:26,414 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155546.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:20:26,920 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 00:20:32,378 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9939, 2.1395, 1.7776, 2.3695, 1.4595, 1.7263, 1.8879, 2.0940], + device='cuda:2'), covar=tensor([0.0640, 0.0606, 0.0805, 0.0482, 0.0986, 0.1082, 0.0713, 0.0624], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0197, 0.0248, 0.0213, 0.0205, 0.0250, 0.0252, 0.0208], + 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') +2023-02-07 00:20:46,962 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 00:20:47,674 INFO [train.py:901] (2/4) Epoch 20, batch 2000, loss[loss=0.2014, simple_loss=0.283, pruned_loss=0.05984, over 8084.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2902, pruned_loss=0.06397, over 1614492.58 frames. ], batch size: 21, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:20:51,754 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.363e+02 2.911e+02 3.881e+02 1.027e+03, threshold=5.822e+02, percent-clipped=2.0 +2023-02-07 00:21:02,540 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.22 vs. limit=5.0 +2023-02-07 00:21:20,956 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155623.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:21:22,881 INFO [train.py:901] (2/4) Epoch 20, batch 2050, loss[loss=0.2235, simple_loss=0.3125, pruned_loss=0.06728, over 8447.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2902, pruned_loss=0.06395, over 1616852.56 frames. ], batch size: 27, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:21:57,546 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155675.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:21:58,082 INFO [train.py:901] (2/4) Epoch 20, batch 2100, loss[loss=0.1764, simple_loss=0.2533, pruned_loss=0.04976, over 7239.00 frames. ], tot_loss[loss=0.209, simple_loss=0.29, pruned_loss=0.06399, over 1618097.29 frames. ], batch size: 16, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:21:58,407 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 00:22:02,102 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.564e+02 2.968e+02 3.686e+02 8.256e+02, threshold=5.935e+02, percent-clipped=7.0 +2023-02-07 00:22:22,173 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155709.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:22:33,489 INFO [train.py:901] (2/4) Epoch 20, batch 2150, loss[loss=0.2089, simple_loss=0.2898, pruned_loss=0.06404, over 7808.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2907, pruned_loss=0.06454, over 1616412.43 frames. ], batch size: 20, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:22:39,018 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155734.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:07,323 INFO [train.py:901] (2/4) Epoch 20, batch 2200, loss[loss=0.2532, simple_loss=0.3419, pruned_loss=0.08224, over 8514.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2924, pruned_loss=0.06545, over 1618858.75 frames. ], batch size: 28, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:23:12,092 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.519e+02 2.939e+02 3.787e+02 7.175e+02, threshold=5.878e+02, percent-clipped=4.0 +2023-02-07 00:23:26,031 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155802.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:26,115 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155802.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:38,413 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155819.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:43,085 INFO [train.py:901] (2/4) Epoch 20, batch 2250, loss[loss=0.2028, simple_loss=0.2887, pruned_loss=0.05844, over 8827.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2931, pruned_loss=0.06536, over 1625340.84 frames. ], batch size: 49, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:23:44,831 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155827.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:24:17,860 INFO [train.py:901] (2/4) Epoch 20, batch 2300, loss[loss=0.2237, simple_loss=0.3089, pruned_loss=0.06927, over 8491.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2928, pruned_loss=0.0652, over 1624926.16 frames. ], batch size: 29, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:24:21,976 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.500e+02 2.966e+02 3.753e+02 6.656e+02, threshold=5.933e+02, percent-clipped=3.0 +2023-02-07 00:24:54,616 INFO [train.py:901] (2/4) Epoch 20, batch 2350, loss[loss=0.1727, simple_loss=0.2433, pruned_loss=0.05104, over 7710.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2926, pruned_loss=0.06521, over 1622721.77 frames. ], batch size: 18, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:25:00,027 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155934.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:25:00,857 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 00:25:04,969 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9940, 1.6538, 1.7902, 1.4778, 0.8716, 1.6176, 1.7142, 1.6271], + device='cuda:2'), covar=tensor([0.0553, 0.1200, 0.1584, 0.1388, 0.0634, 0.1417, 0.0710, 0.0614], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:25:23,233 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155967.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:25:28,796 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8234, 1.5224, 1.6309, 1.3357, 1.0457, 1.4607, 1.7074, 1.4241], + device='cuda:2'), covar=tensor([0.0568, 0.1221, 0.1704, 0.1496, 0.0610, 0.1464, 0.0691, 0.0681], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0099, 0.0161, 0.0112, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:25:29,309 INFO [train.py:901] (2/4) Epoch 20, batch 2400, loss[loss=0.2399, simple_loss=0.3257, pruned_loss=0.07701, over 8661.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2917, pruned_loss=0.06476, over 1617406.37 frames. ], batch size: 34, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:25:33,221 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.377e+02 2.729e+02 3.502e+02 6.388e+02, threshold=5.458e+02, percent-clipped=1.0 +2023-02-07 00:25:36,387 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-07 00:25:58,367 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9364, 1.6753, 2.0154, 1.7488, 1.9485, 2.0029, 1.7874, 0.7128], + device='cuda:2'), covar=tensor([0.5439, 0.4431, 0.1909, 0.3546, 0.2355, 0.3041, 0.1805, 0.4875], + device='cuda:2'), in_proj_covar=tensor([0.0939, 0.0968, 0.0793, 0.0932, 0.0991, 0.0881, 0.0743, 0.0821], + 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') +2023-02-07 00:26:00,994 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156019.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:26:05,601 INFO [train.py:901] (2/4) Epoch 20, batch 2450, loss[loss=0.2009, simple_loss=0.2906, pruned_loss=0.05557, over 8338.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2911, pruned_loss=0.06446, over 1616918.49 frames. ], batch size: 26, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:26:29,109 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 +2023-02-07 00:26:33,872 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 00:26:37,719 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8904, 1.6251, 1.7001, 1.4059, 0.9327, 1.5182, 1.5892, 1.4022], + device='cuda:2'), covar=tensor([0.0539, 0.1184, 0.1660, 0.1399, 0.0585, 0.1474, 0.0714, 0.0669], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0190, 0.0158, 0.0099, 0.0161, 0.0112, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:26:40,974 INFO [train.py:901] (2/4) Epoch 20, batch 2500, loss[loss=0.1814, simple_loss=0.2652, pruned_loss=0.04875, over 7792.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2905, pruned_loss=0.06413, over 1617869.27 frames. ], batch size: 19, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:26:45,021 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.463e+02 3.105e+02 3.826e+02 1.382e+03, threshold=6.210e+02, percent-clipped=11.0 +2023-02-07 00:26:45,218 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156082.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:26:49,176 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156088.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:27:15,809 INFO [train.py:901] (2/4) Epoch 20, batch 2550, loss[loss=0.233, simple_loss=0.3092, pruned_loss=0.0784, over 8516.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.06369, over 1616502.93 frames. ], batch size: 31, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:27:18,741 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1097, 1.1817, 1.2209, 0.7270, 1.2358, 1.0561, 0.1003, 1.1957], + device='cuda:2'), covar=tensor([0.0428, 0.0412, 0.0342, 0.0549, 0.0415, 0.0914, 0.0812, 0.0313], + device='cuda:2'), in_proj_covar=tensor([0.0439, 0.0380, 0.0332, 0.0434, 0.0364, 0.0524, 0.0385, 0.0406], + device='cuda:2'), out_proj_covar=tensor([1.1853e-04, 9.9870e-05, 8.7688e-05, 1.1495e-04, 9.6314e-05, 1.4922e-04, + 1.0419e-04, 1.0835e-04], device='cuda:2') +2023-02-07 00:27:21,365 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156134.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:27:29,846 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156146.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:27:50,757 INFO [train.py:901] (2/4) Epoch 20, batch 2600, loss[loss=0.2407, simple_loss=0.3099, pruned_loss=0.08579, over 8197.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2908, pruned_loss=0.06453, over 1616762.34 frames. ], batch size: 23, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:27:54,662 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.376e+02 3.118e+02 3.808e+02 9.704e+02, threshold=6.236e+02, percent-clipped=5.0 +2023-02-07 00:28:00,367 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156190.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:28:17,466 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:28:18,165 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8335, 2.2634, 3.9885, 1.5838, 2.9332, 2.3702, 1.7989, 2.8045], + device='cuda:2'), covar=tensor([0.1811, 0.2472, 0.0787, 0.4405, 0.1877, 0.3035, 0.2233, 0.2576], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0586, 0.0549, 0.0628, 0.0637, 0.0588, 0.0522, 0.0626], + 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') +2023-02-07 00:28:20,844 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6908, 2.9364, 2.4436, 4.0353, 1.8063, 2.2727, 2.4561, 3.1908], + device='cuda:2'), covar=tensor([0.0604, 0.0791, 0.0772, 0.0212, 0.1090, 0.1120, 0.1077, 0.0689], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0195, 0.0245, 0.0210, 0.0204, 0.0247, 0.0250, 0.0208], + 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') +2023-02-07 00:28:24,592 INFO [train.py:901] (2/4) Epoch 20, batch 2650, loss[loss=0.1849, simple_loss=0.2645, pruned_loss=0.05262, over 7831.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2888, pruned_loss=0.06332, over 1614804.32 frames. ], batch size: 20, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:28:30,654 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156234.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:28:49,181 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156261.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:29:00,117 INFO [train.py:901] (2/4) Epoch 20, batch 2700, loss[loss=0.2341, simple_loss=0.3143, pruned_loss=0.077, over 8497.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2903, pruned_loss=0.06473, over 1614620.03 frames. ], batch size: 26, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:29:04,086 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.401e+02 3.078e+02 3.829e+02 8.557e+02, threshold=6.156e+02, percent-clipped=4.0 +2023-02-07 00:29:23,258 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156308.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:29:35,120 INFO [train.py:901] (2/4) Epoch 20, batch 2750, loss[loss=0.2435, simple_loss=0.3199, pruned_loss=0.08355, over 8541.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2913, pruned_loss=0.06512, over 1613181.36 frames. ], batch size: 49, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:29:43,528 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156338.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:01,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156363.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:10,369 INFO [train.py:901] (2/4) Epoch 20, batch 2800, loss[loss=0.2389, simple_loss=0.3066, pruned_loss=0.0856, over 7302.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2912, pruned_loss=0.06523, over 1611475.01 frames. ], batch size: 72, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:30:15,856 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.534e+02 2.983e+02 3.648e+02 6.974e+02, threshold=5.966e+02, percent-clipped=1.0 +2023-02-07 00:30:20,852 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:38,823 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156415.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:46,249 INFO [train.py:901] (2/4) Epoch 20, batch 2850, loss[loss=0.215, simple_loss=0.2937, pruned_loss=0.06812, over 8334.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2902, pruned_loss=0.06401, over 1614822.79 frames. ], batch size: 26, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:30:50,405 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156432.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:52,534 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4183, 4.4129, 3.9796, 1.9640, 3.8876, 4.0457, 4.0029, 3.8400], + device='cuda:2'), covar=tensor([0.0731, 0.0490, 0.1032, 0.4928, 0.0822, 0.0895, 0.1251, 0.0786], + device='cuda:2'), in_proj_covar=tensor([0.0513, 0.0425, 0.0429, 0.0529, 0.0419, 0.0430, 0.0416, 0.0373], + 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') +2023-02-07 00:31:20,790 INFO [train.py:901] (2/4) Epoch 20, batch 2900, loss[loss=0.1646, simple_loss=0.2456, pruned_loss=0.04178, over 7542.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2913, pruned_loss=0.06449, over 1615866.71 frames. ], batch size: 18, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:31:26,316 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.692e+02 2.409e+02 2.783e+02 3.401e+02 8.568e+02, threshold=5.566e+02, percent-clipped=1.0 +2023-02-07 00:31:50,391 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:31:53,716 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 00:31:57,134 INFO [train.py:901] (2/4) Epoch 20, batch 2950, loss[loss=0.1862, simple_loss=0.2559, pruned_loss=0.05819, over 7529.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2915, pruned_loss=0.06519, over 1611921.94 frames. ], batch size: 18, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:32:08,279 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156542.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:32:11,760 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156547.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:32:31,023 INFO [train.py:901] (2/4) Epoch 20, batch 3000, loss[loss=0.2446, simple_loss=0.3136, pruned_loss=0.08777, over 8497.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2908, pruned_loss=0.06495, over 1610297.38 frames. ], batch size: 28, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:32:31,024 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 00:32:46,839 INFO [train.py:935] (2/4) Epoch 20, validation: loss=0.1756, simple_loss=0.2756, pruned_loss=0.03779, over 944034.00 frames. +2023-02-07 00:32:46,840 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 00:32:48,401 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156578.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:32:51,796 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.571e+02 2.420e+02 3.007e+02 3.801e+02 6.408e+02, threshold=6.014e+02, percent-clipped=4.0 +2023-02-07 00:33:22,163 INFO [train.py:901] (2/4) Epoch 20, batch 3050, loss[loss=0.206, simple_loss=0.2892, pruned_loss=0.06145, over 8071.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2897, pruned_loss=0.06404, over 1610296.55 frames. ], batch size: 21, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:33:33,203 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1736, 1.0826, 1.2919, 1.0785, 0.9227, 1.2893, 0.0961, 1.0187], + device='cuda:2'), covar=tensor([0.1751, 0.1439, 0.0471, 0.0791, 0.2837, 0.0581, 0.2295, 0.1161], + device='cuda:2'), in_proj_covar=tensor([0.0186, 0.0193, 0.0123, 0.0218, 0.0266, 0.0133, 0.0166, 0.0186], + 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') +2023-02-07 00:33:40,595 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156652.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:33:56,274 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1778, 1.0663, 1.3310, 1.1195, 0.9356, 1.3219, 0.0712, 0.9921], + device='cuda:2'), covar=tensor([0.1553, 0.1359, 0.0446, 0.0715, 0.2599, 0.0524, 0.2011, 0.1118], + device='cuda:2'), in_proj_covar=tensor([0.0187, 0.0194, 0.0124, 0.0220, 0.0268, 0.0134, 0.0167, 0.0188], + 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') +2023-02-07 00:33:57,463 INFO [train.py:901] (2/4) Epoch 20, batch 3100, loss[loss=0.1891, simple_loss=0.2619, pruned_loss=0.05818, over 7670.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2883, pruned_loss=0.06343, over 1605625.25 frames. ], batch size: 19, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:34:02,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.428e+02 2.992e+02 3.732e+02 8.006e+02, threshold=5.985e+02, percent-clipped=5.0 +2023-02-07 00:34:09,239 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156693.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:34:31,974 INFO [train.py:901] (2/4) Epoch 20, batch 3150, loss[loss=0.22, simple_loss=0.3016, pruned_loss=0.06914, over 8602.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.29, pruned_loss=0.06434, over 1606407.78 frames. ], batch size: 31, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:35:01,275 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156767.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:35:01,311 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156767.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:35:07,253 INFO [train.py:901] (2/4) Epoch 20, batch 3200, loss[loss=0.1779, simple_loss=0.2542, pruned_loss=0.05079, over 7969.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2889, pruned_loss=0.06365, over 1609907.55 frames. ], batch size: 21, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:35:11,888 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.338e+02 2.875e+02 3.612e+02 1.133e+03, threshold=5.749e+02, percent-clipped=4.0 +2023-02-07 00:35:26,503 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156803.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:35:41,818 INFO [train.py:901] (2/4) Epoch 20, batch 3250, loss[loss=0.1981, simple_loss=0.2892, pruned_loss=0.05353, over 8031.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2888, pruned_loss=0.06337, over 1608543.17 frames. ], batch size: 22, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:35:43,291 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156828.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:36:17,055 INFO [train.py:901] (2/4) Epoch 20, batch 3300, loss[loss=0.193, simple_loss=0.2761, pruned_loss=0.05499, over 8392.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2904, pruned_loss=0.06427, over 1611829.78 frames. ], batch size: 49, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:36:21,772 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.341e+02 2.967e+02 3.887e+02 7.432e+02, threshold=5.934e+02, percent-clipped=7.0 +2023-02-07 00:36:32,803 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9233, 1.4756, 1.7560, 1.3236, 1.0609, 1.4504, 1.8166, 1.4831], + device='cuda:2'), covar=tensor([0.0519, 0.1309, 0.1663, 0.1448, 0.0599, 0.1565, 0.0644, 0.0658], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:36:35,551 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156903.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:36:51,528 INFO [train.py:901] (2/4) Epoch 20, batch 3350, loss[loss=0.2079, simple_loss=0.2949, pruned_loss=0.06041, over 8463.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2887, pruned_loss=0.06298, over 1612898.45 frames. ], batch size: 29, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:37:05,904 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 00:37:07,167 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156949.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:37:25,819 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156974.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:37:27,098 INFO [train.py:901] (2/4) Epoch 20, batch 3400, loss[loss=0.2093, simple_loss=0.2822, pruned_loss=0.06818, over 8240.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2888, pruned_loss=0.06331, over 1612547.58 frames. ], batch size: 24, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:37:31,899 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.508e+02 3.011e+02 3.882e+02 8.239e+02, threshold=6.022e+02, percent-clipped=6.0 +2023-02-07 00:37:49,722 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6962, 1.6283, 1.9644, 1.5697, 1.0328, 1.6441, 2.2391, 1.9298], + device='cuda:2'), covar=tensor([0.0449, 0.1231, 0.1607, 0.1414, 0.0621, 0.1467, 0.0635, 0.0582], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0158, 0.0100, 0.0161, 0.0112, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:37:59,116 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1208, 1.5445, 1.7116, 1.3960, 1.1283, 1.5029, 1.9389, 1.4611], + device='cuda:2'), covar=tensor([0.0501, 0.1200, 0.1606, 0.1441, 0.0601, 0.1430, 0.0647, 0.0676], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0159, 0.0100, 0.0161, 0.0112, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:38:01,292 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:38:03,917 INFO [train.py:901] (2/4) Epoch 20, batch 3450, loss[loss=0.2284, simple_loss=0.3155, pruned_loss=0.07068, over 8322.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2892, pruned_loss=0.06316, over 1611300.38 frames. ], batch size: 25, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:38:05,349 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157028.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:38:11,405 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7045, 1.7349, 2.3149, 1.5958, 1.3587, 2.2358, 0.4073, 1.3687], + device='cuda:2'), covar=tensor([0.1759, 0.1327, 0.0371, 0.1147, 0.2844, 0.0615, 0.2362, 0.1323], + device='cuda:2'), in_proj_covar=tensor([0.0184, 0.0191, 0.0123, 0.0216, 0.0265, 0.0132, 0.0165, 0.0186], + 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') +2023-02-07 00:38:18,893 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157048.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:38:38,200 INFO [train.py:901] (2/4) Epoch 20, batch 3500, loss[loss=0.1927, simple_loss=0.2674, pruned_loss=0.05903, over 8098.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2897, pruned_loss=0.06396, over 1610837.78 frames. ], batch size: 21, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:38:43,584 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.548e+02 3.004e+02 3.939e+02 7.448e+02, threshold=6.007e+02, percent-clipped=9.0 +2023-02-07 00:39:02,212 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 00:39:03,069 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157111.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:39:03,905 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4443, 1.7683, 2.6832, 1.3556, 2.0274, 1.7966, 1.5245, 1.9121], + device='cuda:2'), covar=tensor([0.1897, 0.2563, 0.0802, 0.4350, 0.1723, 0.3096, 0.2205, 0.2234], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0593, 0.0556, 0.0633, 0.0643, 0.0592, 0.0529, 0.0633], + 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') +2023-02-07 00:39:13,059 INFO [train.py:901] (2/4) Epoch 20, batch 3550, loss[loss=0.2057, simple_loss=0.2918, pruned_loss=0.05973, over 8233.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2892, pruned_loss=0.06387, over 1610916.34 frames. ], batch size: 22, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:39:37,691 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157160.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:39:48,306 INFO [train.py:901] (2/4) Epoch 20, batch 3600, loss[loss=0.2322, simple_loss=0.3037, pruned_loss=0.08033, over 8187.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2894, pruned_loss=0.06368, over 1615410.81 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:39:53,028 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.446e+02 2.923e+02 3.668e+02 9.434e+02, threshold=5.847e+02, percent-clipped=4.0 +2023-02-07 00:40:19,603 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1900, 3.6250, 2.2690, 2.7380, 2.9251, 1.8251, 2.9893, 3.0466], + device='cuda:2'), covar=tensor([0.1493, 0.0328, 0.0996, 0.0727, 0.0632, 0.1432, 0.0965, 0.1023], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0236, 0.0330, 0.0309, 0.0300, 0.0336, 0.0344, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 00:40:24,177 INFO [train.py:901] (2/4) Epoch 20, batch 3650, loss[loss=0.1975, simple_loss=0.2719, pruned_loss=0.06158, over 7435.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2895, pruned_loss=0.06396, over 1615074.95 frames. ], batch size: 17, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:40:24,364 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157226.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:40:38,561 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157247.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:40:58,619 INFO [train.py:901] (2/4) Epoch 20, batch 3700, loss[loss=0.1862, simple_loss=0.2624, pruned_loss=0.05499, over 8246.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2908, pruned_loss=0.06465, over 1618525.13 frames. ], batch size: 22, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:41:03,188 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.545e+02 3.038e+02 3.849e+02 9.039e+02, threshold=6.076e+02, percent-clipped=6.0 +2023-02-07 00:41:05,241 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 00:41:33,567 INFO [train.py:901] (2/4) Epoch 20, batch 3750, loss[loss=0.2126, simple_loss=0.2875, pruned_loss=0.0689, over 6622.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2913, pruned_loss=0.06479, over 1618184.96 frames. ], batch size: 72, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:41:58,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157362.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:05,450 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157372.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:06,815 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157374.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:07,956 INFO [train.py:901] (2/4) Epoch 20, batch 3800, loss[loss=0.2998, simple_loss=0.362, pruned_loss=0.1188, over 8677.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2926, pruned_loss=0.06525, over 1619413.54 frames. ], batch size: 34, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:42:12,516 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.302e+02 2.981e+02 3.884e+02 7.104e+02, threshold=5.962e+02, percent-clipped=4.0 +2023-02-07 00:42:25,025 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:39,607 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157422.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:42:42,776 INFO [train.py:901] (2/4) Epoch 20, batch 3850, loss[loss=0.1872, simple_loss=0.2729, pruned_loss=0.0507, over 7968.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2931, pruned_loss=0.06583, over 1616352.76 frames. ], batch size: 21, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:43:05,839 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6172, 4.5644, 4.1613, 1.9833, 4.1348, 4.2686, 4.0810, 4.0276], + device='cuda:2'), covar=tensor([0.0624, 0.0480, 0.1039, 0.4292, 0.0770, 0.0798, 0.1218, 0.0604], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0421, 0.0426, 0.0526, 0.0415, 0.0428, 0.0413, 0.0371], + 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') +2023-02-07 00:43:09,739 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 00:43:17,658 INFO [train.py:901] (2/4) Epoch 20, batch 3900, loss[loss=0.2307, simple_loss=0.3142, pruned_loss=0.07356, over 8196.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2933, pruned_loss=0.06562, over 1617131.60 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:43:21,793 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157482.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:22,201 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.513e+02 3.153e+02 3.900e+02 7.255e+02, threshold=6.305e+02, percent-clipped=5.0 +2023-02-07 00:43:24,971 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157487.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:37,143 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157504.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:39,411 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157507.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:43,554 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1640, 1.3719, 4.3263, 1.3845, 3.7829, 3.6165, 3.9500, 3.7905], + device='cuda:2'), covar=tensor([0.0559, 0.4706, 0.0577, 0.4388, 0.1189, 0.1000, 0.0590, 0.0691], + device='cuda:2'), in_proj_covar=tensor([0.0612, 0.0635, 0.0687, 0.0616, 0.0697, 0.0603, 0.0599, 0.0665], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 00:43:51,269 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157524.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:43:52,528 INFO [train.py:901] (2/4) Epoch 20, batch 3950, loss[loss=0.2068, simple_loss=0.2889, pruned_loss=0.0624, over 8238.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2923, pruned_loss=0.0649, over 1614171.10 frames. ], batch size: 22, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:44:25,096 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5570, 4.5499, 4.1452, 2.0349, 4.0748, 4.2119, 4.1987, 4.0849], + device='cuda:2'), covar=tensor([0.0675, 0.0477, 0.0974, 0.4593, 0.0763, 0.0956, 0.1098, 0.0722], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0425, 0.0431, 0.0531, 0.0418, 0.0432, 0.0418, 0.0374], + 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') +2023-02-07 00:44:28,110 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.52 vs. limit=5.0 +2023-02-07 00:44:28,440 INFO [train.py:901] (2/4) Epoch 20, batch 4000, loss[loss=0.1842, simple_loss=0.2655, pruned_loss=0.05141, over 7661.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.291, pruned_loss=0.0641, over 1613624.25 frames. ], batch size: 19, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:44:33,882 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.441e+02 3.259e+02 3.960e+02 7.383e+02, threshold=6.518e+02, percent-clipped=3.0 +2023-02-07 00:44:34,064 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9162, 1.4337, 6.0152, 2.0411, 5.4275, 5.0987, 5.6199, 5.4974], + device='cuda:2'), covar=tensor([0.0419, 0.5187, 0.0385, 0.3957, 0.0979, 0.0896, 0.0453, 0.0445], + device='cuda:2'), in_proj_covar=tensor([0.0614, 0.0637, 0.0688, 0.0619, 0.0701, 0.0606, 0.0601, 0.0667], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 00:44:36,142 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:44:41,340 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0656, 3.8166, 2.3048, 2.8625, 2.8277, 2.1162, 3.0931, 3.1458], + device='cuda:2'), covar=tensor([0.1562, 0.0310, 0.1117, 0.0721, 0.0684, 0.1368, 0.0851, 0.1009], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0235, 0.0332, 0.0307, 0.0300, 0.0335, 0.0344, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 00:44:57,791 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157618.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:44:58,430 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157619.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:45:03,486 INFO [train.py:901] (2/4) Epoch 20, batch 4050, loss[loss=0.2009, simple_loss=0.3045, pruned_loss=0.0486, over 8143.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2904, pruned_loss=0.06396, over 1612697.70 frames. ], batch size: 22, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:45:15,101 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:45:29,209 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 00:45:33,138 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.94 vs. limit=5.0 +2023-02-07 00:45:38,025 INFO [train.py:901] (2/4) Epoch 20, batch 4100, loss[loss=0.2719, simple_loss=0.3514, pruned_loss=0.09619, over 8358.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.292, pruned_loss=0.06476, over 1613740.35 frames. ], batch size: 24, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:45:42,584 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.468e+02 3.178e+02 4.268e+02 8.149e+02, threshold=6.355e+02, percent-clipped=4.0 +2023-02-07 00:46:07,476 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157718.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:12,784 INFO [train.py:901] (2/4) Epoch 20, batch 4150, loss[loss=0.1857, simple_loss=0.2777, pruned_loss=0.04684, over 8108.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.291, pruned_loss=0.06446, over 1611440.66 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:46:25,111 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157743.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:25,658 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157744.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:40,586 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157766.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:46:41,942 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:45,978 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2795, 1.9848, 2.7114, 2.2062, 2.6783, 2.2802, 2.0456, 1.5494], + device='cuda:2'), covar=tensor([0.5247, 0.4926, 0.1744, 0.3544, 0.2410, 0.2852, 0.1839, 0.5106], + device='cuda:2'), in_proj_covar=tensor([0.0934, 0.0966, 0.0788, 0.0931, 0.0984, 0.0882, 0.0737, 0.0815], + 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') +2023-02-07 00:46:46,933 INFO [train.py:901] (2/4) Epoch 20, batch 4200, loss[loss=0.2508, simple_loss=0.3212, pruned_loss=0.09016, over 8095.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2904, pruned_loss=0.06419, over 1609385.82 frames. ], batch size: 21, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:46:52,349 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.385e+02 2.811e+02 3.577e+02 7.269e+02, threshold=5.621e+02, percent-clipped=2.0 +2023-02-07 00:47:08,775 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 00:47:10,321 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8344, 1.5981, 5.9817, 2.3583, 5.3447, 5.0618, 5.5422, 5.3847], + device='cuda:2'), covar=tensor([0.0502, 0.5005, 0.0371, 0.3538, 0.1004, 0.0844, 0.0510, 0.0516], + device='cuda:2'), in_proj_covar=tensor([0.0619, 0.0640, 0.0694, 0.0622, 0.0703, 0.0610, 0.0605, 0.0670], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 00:47:23,327 INFO [train.py:901] (2/4) Epoch 20, batch 4250, loss[loss=0.2117, simple_loss=0.302, pruned_loss=0.06066, over 8314.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2887, pruned_loss=0.06387, over 1606508.46 frames. ], batch size: 25, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:47:26,206 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8039, 1.7538, 1.9634, 1.6173, 1.2739, 1.7315, 2.3806, 2.0386], + device='cuda:2'), covar=tensor([0.0450, 0.1182, 0.1587, 0.1349, 0.0577, 0.1363, 0.0570, 0.0602], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0159, 0.0100, 0.0162, 0.0113, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 00:47:28,290 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:32,309 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 00:47:46,379 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157859.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:53,372 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157868.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:47:55,484 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:58,177 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157875.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:58,652 INFO [train.py:901] (2/4) Epoch 20, batch 4300, loss[loss=0.2008, simple_loss=0.2792, pruned_loss=0.06117, over 8132.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2893, pruned_loss=0.06385, over 1608488.11 frames. ], batch size: 22, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:48:02,059 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157881.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:48:03,195 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.270e+02 2.745e+02 3.400e+02 8.203e+02, threshold=5.491e+02, percent-clipped=7.0 +2023-02-07 00:48:15,327 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157900.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:48:17,999 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6537, 1.3450, 2.8958, 1.4184, 2.0654, 3.0476, 3.2123, 2.6280], + device='cuda:2'), covar=tensor([0.1147, 0.1656, 0.0402, 0.2030, 0.0939, 0.0296, 0.0635, 0.0589], + device='cuda:2'), in_proj_covar=tensor([0.0292, 0.0318, 0.0285, 0.0311, 0.0300, 0.0262, 0.0407, 0.0303], + 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') +2023-02-07 00:48:24,968 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 00:48:33,499 INFO [train.py:901] (2/4) Epoch 20, batch 4350, loss[loss=0.2208, simple_loss=0.3096, pruned_loss=0.06598, over 8185.00 frames. ], tot_loss[loss=0.208, simple_loss=0.289, pruned_loss=0.06356, over 1612469.77 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:48:36,288 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157930.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:48:48,937 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.01 vs. limit=5.0 +2023-02-07 00:49:04,106 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 00:49:08,264 INFO [train.py:901] (2/4) Epoch 20, batch 4400, loss[loss=0.2078, simple_loss=0.2889, pruned_loss=0.06335, over 8254.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2897, pruned_loss=0.06388, over 1611988.22 frames. ], batch size: 24, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:49:13,805 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.443e+02 2.894e+02 3.714e+02 1.238e+03, threshold=5.788e+02, percent-clipped=6.0 +2023-02-07 00:49:14,001 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157983.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:49:44,318 INFO [train.py:901] (2/4) Epoch 20, batch 4450, loss[loss=0.2223, simple_loss=0.3058, pruned_loss=0.06939, over 8473.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2908, pruned_loss=0.06397, over 1617761.06 frames. ], batch size: 25, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:49:45,687 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 00:49:57,297 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158045.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:18,894 INFO [train.py:901] (2/4) Epoch 20, batch 4500, loss[loss=0.1998, simple_loss=0.2886, pruned_loss=0.05549, over 8134.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2905, pruned_loss=0.06352, over 1619509.20 frames. ], batch size: 22, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:50:23,581 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.382e+02 2.908e+02 3.384e+02 7.082e+02, threshold=5.816e+02, percent-clipped=5.0 +2023-02-07 00:50:27,966 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158089.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:39,140 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 00:50:45,180 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158114.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:45,855 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158115.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:53,563 INFO [train.py:901] (2/4) Epoch 20, batch 4550, loss[loss=0.1889, simple_loss=0.273, pruned_loss=0.05241, over 8244.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.29, pruned_loss=0.06329, over 1618101.07 frames. ], batch size: 22, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:51:01,051 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158137.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:51:02,892 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158140.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:51:02,940 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158140.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:51:18,411 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158162.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:51:28,241 INFO [train.py:901] (2/4) Epoch 20, batch 4600, loss[loss=0.1978, simple_loss=0.2892, pruned_loss=0.05323, over 8037.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2898, pruned_loss=0.06291, over 1619237.72 frames. ], batch size: 22, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:51:32,831 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.506e+02 3.217e+02 3.763e+02 8.986e+02, threshold=6.435e+02, percent-clipped=3.0 +2023-02-07 00:51:54,923 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:52:01,859 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4524, 1.3911, 1.7429, 1.2584, 1.1176, 1.7186, 0.2564, 1.2109], + device='cuda:2'), covar=tensor([0.1757, 0.1314, 0.0441, 0.1097, 0.2813, 0.0536, 0.2253, 0.1369], + device='cuda:2'), in_proj_covar=tensor([0.0187, 0.0195, 0.0124, 0.0219, 0.0268, 0.0134, 0.0167, 0.0189], + 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') +2023-02-07 00:52:03,092 INFO [train.py:901] (2/4) Epoch 20, batch 4650, loss[loss=0.2072, simple_loss=0.2813, pruned_loss=0.06652, over 7522.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2907, pruned_loss=0.06395, over 1618592.53 frames. ], batch size: 18, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:52:12,064 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158239.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:52:30,113 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158264.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:52:37,867 INFO [train.py:901] (2/4) Epoch 20, batch 4700, loss[loss=0.1968, simple_loss=0.2811, pruned_loss=0.05627, over 8297.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2913, pruned_loss=0.06451, over 1616943.57 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:52:42,603 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.408e+02 3.012e+02 4.119e+02 1.091e+03, threshold=6.025e+02, percent-clipped=3.0 +2023-02-07 00:52:55,802 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:53:12,687 INFO [train.py:901] (2/4) Epoch 20, batch 4750, loss[loss=0.2124, simple_loss=0.3008, pruned_loss=0.06197, over 8503.00 frames. ], tot_loss[loss=0.209, simple_loss=0.29, pruned_loss=0.06399, over 1610311.09 frames. ], batch size: 28, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:53:12,915 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158326.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:53:15,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158330.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:53:40,929 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 00:53:43,689 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 00:53:48,277 INFO [train.py:901] (2/4) Epoch 20, batch 4800, loss[loss=0.209, simple_loss=0.2946, pruned_loss=0.06176, over 8674.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2903, pruned_loss=0.06429, over 1607387.80 frames. ], batch size: 39, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:53:52,907 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.386e+02 2.729e+02 3.445e+02 7.258e+02, threshold=5.458e+02, percent-clipped=2.0 +2023-02-07 00:54:06,936 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158402.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:54:22,901 INFO [train.py:901] (2/4) Epoch 20, batch 4850, loss[loss=0.1888, simple_loss=0.2836, pruned_loss=0.04704, over 8142.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2903, pruned_loss=0.06397, over 1608420.17 frames. ], batch size: 22, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:54:23,842 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1195, 1.8373, 2.3870, 1.9722, 2.2867, 2.1726, 1.8914, 1.0932], + device='cuda:2'), covar=tensor([0.5311, 0.4685, 0.1844, 0.3444, 0.2466, 0.2742, 0.1869, 0.5268], + device='cuda:2'), in_proj_covar=tensor([0.0934, 0.0966, 0.0788, 0.0933, 0.0990, 0.0882, 0.0740, 0.0820], + 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') +2023-02-07 00:54:33,530 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 00:54:35,079 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1846, 1.5334, 4.4214, 2.0350, 2.4686, 4.9638, 5.0344, 4.2714], + device='cuda:2'), covar=tensor([0.1292, 0.1848, 0.0278, 0.1934, 0.1202, 0.0208, 0.0502, 0.0610], + device='cuda:2'), in_proj_covar=tensor([0.0293, 0.0320, 0.0287, 0.0313, 0.0303, 0.0263, 0.0410, 0.0305], + 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') +2023-02-07 00:54:51,915 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0356, 1.8250, 3.1931, 1.7766, 2.4658, 3.4625, 3.5021, 2.9874], + device='cuda:2'), covar=tensor([0.1155, 0.1589, 0.0428, 0.1815, 0.1039, 0.0254, 0.0679, 0.0526], + device='cuda:2'), in_proj_covar=tensor([0.0293, 0.0319, 0.0286, 0.0312, 0.0303, 0.0262, 0.0409, 0.0305], + 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') +2023-02-07 00:54:57,239 INFO [train.py:901] (2/4) Epoch 20, batch 4900, loss[loss=0.2489, simple_loss=0.3209, pruned_loss=0.08847, over 8241.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2925, pruned_loss=0.06518, over 1608196.56 frames. ], batch size: 24, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:55:02,467 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.481e+02 3.123e+02 4.208e+02 8.958e+02, threshold=6.246e+02, percent-clipped=7.0 +2023-02-07 00:55:03,228 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158484.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:55:32,882 INFO [train.py:901] (2/4) Epoch 20, batch 4950, loss[loss=0.1889, simple_loss=0.2661, pruned_loss=0.05589, over 7709.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2911, pruned_loss=0.06424, over 1609984.34 frames. ], batch size: 18, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:56:07,759 INFO [train.py:901] (2/4) Epoch 20, batch 5000, loss[loss=0.2347, simple_loss=0.3122, pruned_loss=0.07858, over 8325.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2908, pruned_loss=0.06437, over 1609935.36 frames. ], batch size: 26, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:56:12,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.361e+02 2.881e+02 3.667e+02 7.563e+02, threshold=5.761e+02, percent-clipped=2.0 +2023-02-07 00:56:12,598 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.77 vs. limit=5.0 +2023-02-07 00:56:14,508 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:56:23,821 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158599.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:56:32,798 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158611.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:56:42,905 INFO [train.py:901] (2/4) Epoch 20, batch 5050, loss[loss=0.1984, simple_loss=0.2902, pruned_loss=0.05329, over 8349.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2925, pruned_loss=0.06563, over 1610631.23 frames. ], batch size: 24, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:56:54,127 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4687, 1.4400, 1.7976, 1.1603, 1.1007, 1.7888, 0.1642, 1.1528], + device='cuda:2'), covar=tensor([0.1977, 0.1477, 0.0452, 0.1132, 0.3152, 0.0510, 0.2283, 0.1370], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0197, 0.0127, 0.0222, 0.0273, 0.0135, 0.0171, 0.0192], + 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') +2023-02-07 00:57:10,205 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 00:57:17,761 INFO [train.py:901] (2/4) Epoch 20, batch 5100, loss[loss=0.1669, simple_loss=0.2612, pruned_loss=0.03625, over 8110.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2917, pruned_loss=0.06489, over 1611833.57 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:57:23,331 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.670e+02 3.233e+02 3.910e+02 8.185e+02, threshold=6.466e+02, percent-clipped=7.0 +2023-02-07 00:57:53,856 INFO [train.py:901] (2/4) Epoch 20, batch 5150, loss[loss=0.2076, simple_loss=0.2953, pruned_loss=0.0599, over 8292.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2921, pruned_loss=0.06486, over 1615567.89 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:58:08,320 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158746.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:58:28,393 INFO [train.py:901] (2/4) Epoch 20, batch 5200, loss[loss=0.2505, simple_loss=0.3273, pruned_loss=0.08692, over 8461.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2916, pruned_loss=0.06466, over 1612715.19 frames. ], batch size: 25, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:58:30,716 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3867, 1.8154, 1.3472, 2.9688, 1.4027, 1.2524, 2.0883, 1.9597], + device='cuda:2'), covar=tensor([0.1684, 0.1381, 0.2026, 0.0336, 0.1406, 0.2163, 0.0984, 0.1113], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0212, 0.0203, 0.0246, 0.0249, 0.0207], + 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') +2023-02-07 00:58:33,213 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.433e+02 2.837e+02 3.461e+02 7.505e+02, threshold=5.673e+02, percent-clipped=2.0 +2023-02-07 00:58:41,617 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158795.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:58:46,634 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158801.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:58:52,344 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 00:59:03,978 INFO [train.py:901] (2/4) Epoch 20, batch 5250, loss[loss=0.2647, simple_loss=0.3378, pruned_loss=0.09578, over 8540.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2914, pruned_loss=0.06496, over 1612488.30 frames. ], batch size: 49, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:59:11,285 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 00:59:22,853 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158853.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:59:24,328 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158855.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:28,231 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158861.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:38,588 INFO [train.py:901] (2/4) Epoch 20, batch 5300, loss[loss=0.1812, simple_loss=0.2527, pruned_loss=0.05485, over 7659.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2919, pruned_loss=0.06553, over 1614074.04 frames. ], batch size: 19, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 00:59:41,443 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158880.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:43,361 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.336e+02 2.792e+02 3.296e+02 7.091e+02, threshold=5.585e+02, percent-clipped=2.0 +2023-02-07 00:59:52,521 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.90 vs. limit=5.0 +2023-02-07 01:00:13,210 INFO [train.py:901] (2/4) Epoch 20, batch 5350, loss[loss=0.2053, simple_loss=0.2874, pruned_loss=0.06161, over 8665.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.291, pruned_loss=0.06501, over 1611816.05 frames. ], batch size: 49, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:00:15,070 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-07 01:00:27,706 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 01:00:48,000 INFO [train.py:901] (2/4) Epoch 20, batch 5400, loss[loss=0.2389, simple_loss=0.3128, pruned_loss=0.08245, over 8469.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2914, pruned_loss=0.06554, over 1609622.84 frames. ], batch size: 29, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:00:52,644 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.394e+02 2.966e+02 3.887e+02 6.953e+02, threshold=5.932e+02, percent-clipped=4.0 +2023-02-07 01:01:16,690 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 01:01:22,917 INFO [train.py:901] (2/4) Epoch 20, batch 5450, loss[loss=0.1898, simple_loss=0.2772, pruned_loss=0.05121, over 8028.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2914, pruned_loss=0.06538, over 1611226.74 frames. ], batch size: 22, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:01:42,893 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2616, 1.9942, 2.7589, 2.2079, 2.6565, 2.2896, 1.9844, 1.5760], + device='cuda:2'), covar=tensor([0.4920, 0.4790, 0.1703, 0.3381, 0.2254, 0.2830, 0.1916, 0.4881], + device='cuda:2'), in_proj_covar=tensor([0.0925, 0.0958, 0.0784, 0.0924, 0.0977, 0.0874, 0.0732, 0.0809], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-07 01:01:57,423 INFO [train.py:901] (2/4) Epoch 20, batch 5500, loss[loss=0.227, simple_loss=0.3056, pruned_loss=0.07415, over 8466.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2917, pruned_loss=0.06537, over 1614320.19 frames. ], batch size: 27, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:02:00,095 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 01:02:02,825 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.298e+02 2.656e+02 3.222e+02 6.486e+02, threshold=5.312e+02, percent-clipped=1.0 +2023-02-07 01:02:03,906 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 +2023-02-07 01:02:05,787 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159087.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:21,369 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 01:02:27,315 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159117.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:29,240 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5199, 1.7949, 2.7415, 1.4320, 2.0340, 1.8774, 1.6302, 1.9791], + device='cuda:2'), covar=tensor([0.1884, 0.2682, 0.0775, 0.4541, 0.1754, 0.3213, 0.2229, 0.2139], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0590, 0.0554, 0.0634, 0.0643, 0.0589, 0.0528, 0.0629], + 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') +2023-02-07 01:02:32,441 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2662, 1.6576, 4.3134, 2.0478, 2.5243, 4.9640, 4.9257, 4.2368], + device='cuda:2'), covar=tensor([0.1096, 0.1741, 0.0333, 0.1854, 0.1123, 0.0160, 0.0439, 0.0536], + device='cuda:2'), in_proj_covar=tensor([0.0294, 0.0319, 0.0286, 0.0314, 0.0304, 0.0262, 0.0410, 0.0305], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 01:02:32,989 INFO [train.py:901] (2/4) Epoch 20, batch 5550, loss[loss=0.2206, simple_loss=0.3015, pruned_loss=0.06987, over 8348.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.291, pruned_loss=0.06502, over 1604669.34 frames. ], batch size: 26, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:02:41,927 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159139.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:43,056 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 01:02:44,166 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159142.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:46,019 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159145.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:50,309 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-07 01:03:08,164 INFO [train.py:901] (2/4) Epoch 20, batch 5600, loss[loss=0.271, simple_loss=0.3359, pruned_loss=0.103, over 8543.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2902, pruned_loss=0.06489, over 1604701.26 frames. ], batch size: 31, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:03:09,019 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159177.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:03:12,911 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.638e+02 2.419e+02 2.780e+02 3.445e+02 7.739e+02, threshold=5.561e+02, percent-clipped=2.0 +2023-02-07 01:03:20,717 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5735, 2.0276, 3.0748, 1.4216, 2.3352, 1.9524, 1.7730, 2.2808], + device='cuda:2'), covar=tensor([0.1824, 0.2379, 0.0803, 0.4199, 0.1739, 0.3043, 0.2023, 0.2180], + device='cuda:2'), in_proj_covar=tensor([0.0519, 0.0589, 0.0555, 0.0634, 0.0643, 0.0589, 0.0528, 0.0630], + 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') +2023-02-07 01:03:23,295 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159197.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:03:43,998 INFO [train.py:901] (2/4) Epoch 20, batch 5650, loss[loss=0.1565, simple_loss=0.2456, pruned_loss=0.03376, over 7684.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2911, pruned_loss=0.06514, over 1606730.46 frames. ], batch size: 18, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:04:03,423 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159254.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:04:04,621 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 01:04:07,499 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159260.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:04:19,056 INFO [train.py:901] (2/4) Epoch 20, batch 5700, loss[loss=0.2026, simple_loss=0.2931, pruned_loss=0.05608, over 8464.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2925, pruned_loss=0.06581, over 1610446.25 frames. ], batch size: 25, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:04:25,332 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.576e+02 3.260e+02 4.013e+02 6.441e+02, threshold=6.520e+02, percent-clipped=4.0 +2023-02-07 01:04:42,299 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159308.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:04:45,040 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159312.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:04:54,516 INFO [train.py:901] (2/4) Epoch 20, batch 5750, loss[loss=0.1664, simple_loss=0.2514, pruned_loss=0.04073, over 7970.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2933, pruned_loss=0.06613, over 1614979.76 frames. ], batch size: 21, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:04:59,222 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 01:05:09,316 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 01:05:23,228 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8697, 1.6295, 3.2578, 1.5749, 2.3154, 3.5420, 3.6450, 3.0586], + device='cuda:2'), covar=tensor([0.1211, 0.1605, 0.0343, 0.1987, 0.1060, 0.0228, 0.0540, 0.0501], + device='cuda:2'), in_proj_covar=tensor([0.0291, 0.0316, 0.0284, 0.0311, 0.0302, 0.0260, 0.0407, 0.0301], + 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') +2023-02-07 01:05:29,355 INFO [train.py:901] (2/4) Epoch 20, batch 5800, loss[loss=0.2501, simple_loss=0.3397, pruned_loss=0.08023, over 8455.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2949, pruned_loss=0.06683, over 1617723.73 frames. ], batch size: 25, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:05:35,566 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.438e+02 2.992e+02 3.849e+02 1.447e+03, threshold=5.984e+02, percent-clipped=4.0 +2023-02-07 01:06:04,880 INFO [train.py:901] (2/4) Epoch 20, batch 5850, loss[loss=0.1779, simple_loss=0.2618, pruned_loss=0.04701, over 7805.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2934, pruned_loss=0.06631, over 1612950.78 frames. ], batch size: 19, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:06:08,469 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:06:30,805 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 01:06:39,994 INFO [train.py:901] (2/4) Epoch 20, batch 5900, loss[loss=0.1887, simple_loss=0.2798, pruned_loss=0.04879, over 8291.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2926, pruned_loss=0.06527, over 1617536.55 frames. ], batch size: 23, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:06:45,624 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.454e+02 2.951e+02 3.822e+02 7.063e+02, threshold=5.901e+02, percent-clipped=2.0 +2023-02-07 01:07:00,826 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6722, 2.2432, 4.1939, 1.4645, 2.9928, 2.2942, 1.8172, 2.8465], + device='cuda:2'), covar=tensor([0.1855, 0.2692, 0.0693, 0.4549, 0.1697, 0.3102, 0.2248, 0.2450], + device='cuda:2'), in_proj_covar=tensor([0.0522, 0.0593, 0.0556, 0.0637, 0.0645, 0.0592, 0.0530, 0.0632], + 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') +2023-02-07 01:07:04,117 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159510.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:08,825 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159516.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:12,107 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159521.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:15,459 INFO [train.py:901] (2/4) Epoch 20, batch 5950, loss[loss=0.1592, simple_loss=0.2419, pruned_loss=0.03822, over 7659.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2917, pruned_loss=0.06465, over 1615902.46 frames. ], batch size: 19, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:07:15,838 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 01:07:22,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159535.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:26,382 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159541.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:29,670 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159546.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:42,011 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5508, 1.8475, 1.9288, 1.2190, 1.9586, 1.4603, 0.4589, 1.8082], + device='cuda:2'), covar=tensor([0.0496, 0.0324, 0.0254, 0.0488, 0.0344, 0.0838, 0.0815, 0.0225], + device='cuda:2'), in_proj_covar=tensor([0.0455, 0.0386, 0.0337, 0.0444, 0.0369, 0.0533, 0.0392, 0.0415], + device='cuda:2'), out_proj_covar=tensor([1.2254e-04, 1.0147e-04, 8.8862e-05, 1.1734e-04, 9.7634e-05, 1.5151e-04, + 1.0602e-04, 1.1062e-04], device='cuda:2') +2023-02-07 01:07:45,567 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159568.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:07:50,925 INFO [train.py:901] (2/4) Epoch 20, batch 6000, loss[loss=0.2086, simple_loss=0.2805, pruned_loss=0.06836, over 7647.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2906, pruned_loss=0.06427, over 1610592.80 frames. ], batch size: 19, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:07:50,926 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 01:08:04,189 INFO [train.py:935] (2/4) Epoch 20, validation: loss=0.175, simple_loss=0.275, pruned_loss=0.03755, over 944034.00 frames. +2023-02-07 01:08:04,190 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 01:08:09,554 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.504e+02 2.869e+02 3.482e+02 8.370e+02, threshold=5.739e+02, percent-clipped=5.0 +2023-02-07 01:08:15,888 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159593.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:08:38,924 INFO [train.py:901] (2/4) Epoch 20, batch 6050, loss[loss=0.2062, simple_loss=0.2977, pruned_loss=0.05739, over 8335.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.06359, over 1610182.01 frames. ], batch size: 25, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:08:45,953 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159636.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:08:51,473 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3759, 2.8378, 2.3802, 3.8795, 1.5722, 2.1252, 2.2806, 2.6966], + device='cuda:2'), covar=tensor([0.0663, 0.0755, 0.0743, 0.0230, 0.1092, 0.1186, 0.0944, 0.0773], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0212, 0.0204, 0.0245, 0.0248, 0.0206], + 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') +2023-02-07 01:08:55,604 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159649.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:08:57,636 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159652.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:09:14,759 INFO [train.py:901] (2/4) Epoch 20, batch 6100, loss[loss=0.2149, simple_loss=0.3055, pruned_loss=0.06216, over 8103.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2913, pruned_loss=0.06461, over 1610601.10 frames. ], batch size: 23, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:09:21,004 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.453e+02 2.842e+02 3.745e+02 1.322e+03, threshold=5.684e+02, percent-clipped=4.0 +2023-02-07 01:09:30,876 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1397, 2.3174, 2.0008, 2.7860, 1.4553, 1.7524, 2.1176, 2.2324], + device='cuda:2'), covar=tensor([0.0613, 0.0711, 0.0792, 0.0406, 0.1009, 0.1161, 0.0783, 0.0716], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0197, 0.0245, 0.0213, 0.0204, 0.0245, 0.0248, 0.0207], + 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') +2023-02-07 01:09:41,580 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 01:09:49,993 INFO [train.py:901] (2/4) Epoch 20, batch 6150, loss[loss=0.1953, simple_loss=0.2957, pruned_loss=0.04746, over 8463.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2906, pruned_loss=0.06454, over 1611381.40 frames. ], batch size: 25, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:10:18,361 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159767.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:10:24,871 INFO [train.py:901] (2/4) Epoch 20, batch 6200, loss[loss=0.2032, simple_loss=0.2829, pruned_loss=0.06171, over 8262.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2903, pruned_loss=0.06443, over 1610297.58 frames. ], batch size: 24, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:10:30,207 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.429e+02 3.094e+02 3.753e+02 7.329e+02, threshold=6.188e+02, percent-clipped=3.0 +2023-02-07 01:10:33,806 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0644, 1.5889, 1.4320, 1.5018, 1.2951, 1.2450, 1.3205, 1.3282], + device='cuda:2'), covar=tensor([0.1105, 0.0505, 0.1248, 0.0539, 0.0769, 0.1536, 0.0809, 0.0780], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0237, 0.0331, 0.0309, 0.0301, 0.0337, 0.0344, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 01:10:43,480 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159802.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:11:00,336 INFO [train.py:901] (2/4) Epoch 20, batch 6250, loss[loss=0.2449, simple_loss=0.3242, pruned_loss=0.0828, over 8355.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2895, pruned_loss=0.06337, over 1611753.50 frames. ], batch size: 24, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:11:01,219 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159827.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:11:05,579 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 01:11:15,896 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6910, 2.4419, 4.0516, 1.6077, 3.0601, 2.2213, 1.8402, 2.8282], + device='cuda:2'), covar=tensor([0.2016, 0.2473, 0.0804, 0.4531, 0.1891, 0.3294, 0.2356, 0.2575], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0589, 0.0552, 0.0633, 0.0642, 0.0590, 0.0528, 0.0630], + 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') +2023-02-07 01:11:32,492 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159873.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:11:34,365 INFO [train.py:901] (2/4) Epoch 20, batch 6300, loss[loss=0.2256, simple_loss=0.3097, pruned_loss=0.07079, over 8831.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2894, pruned_loss=0.06355, over 1607891.69 frames. ], batch size: 32, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:11:40,337 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.354e+02 2.951e+02 3.644e+02 9.166e+02, threshold=5.902e+02, percent-clipped=5.0 +2023-02-07 01:11:45,861 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159892.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:12:03,493 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159917.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:12:09,078 INFO [train.py:901] (2/4) Epoch 20, batch 6350, loss[loss=0.2395, simple_loss=0.3227, pruned_loss=0.07814, over 8554.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2903, pruned_loss=0.06431, over 1611761.53 frames. ], batch size: 49, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:12:10,589 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159928.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:12:43,479 INFO [train.py:901] (2/4) Epoch 20, batch 6400, loss[loss=0.2315, simple_loss=0.3114, pruned_loss=0.07581, over 8731.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2909, pruned_loss=0.06458, over 1610884.45 frames. ], batch size: 34, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:12:48,765 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.436e+02 2.995e+02 3.881e+02 8.346e+02, threshold=5.989e+02, percent-clipped=6.0 +2023-02-07 01:12:52,384 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5750, 1.4725, 4.7852, 2.0011, 4.2753, 4.0408, 4.3846, 4.2304], + device='cuda:2'), covar=tensor([0.0563, 0.4884, 0.0508, 0.3848, 0.1046, 0.0944, 0.0542, 0.0621], + device='cuda:2'), in_proj_covar=tensor([0.0609, 0.0629, 0.0678, 0.0614, 0.0692, 0.0593, 0.0593, 0.0665], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 01:12:55,757 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159993.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:13:16,774 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:13:18,629 INFO [train.py:901] (2/4) Epoch 20, batch 6450, loss[loss=0.1858, simple_loss=0.2778, pruned_loss=0.04686, over 8254.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2911, pruned_loss=0.06475, over 1612244.16 frames. ], batch size: 22, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:13:31,724 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1318, 1.5183, 1.7454, 1.3226, 0.9470, 1.4952, 1.6942, 1.6805], + device='cuda:2'), covar=tensor([0.0503, 0.1269, 0.1622, 0.1457, 0.0624, 0.1465, 0.0698, 0.0620], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0158, 0.0099, 0.0161, 0.0112, 0.0141], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 01:13:34,503 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160048.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:13:53,994 INFO [train.py:901] (2/4) Epoch 20, batch 6500, loss[loss=0.2173, simple_loss=0.3018, pruned_loss=0.06644, over 8369.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2903, pruned_loss=0.06417, over 1612666.41 frames. ], batch size: 24, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:13:59,464 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.613e+02 3.061e+02 4.120e+02 1.100e+03, threshold=6.122e+02, percent-clipped=8.0 +2023-02-07 01:14:04,195 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.92 vs. limit=5.0 +2023-02-07 01:14:16,510 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:14:29,724 INFO [train.py:901] (2/4) Epoch 20, batch 6550, loss[loss=0.1819, simple_loss=0.2572, pruned_loss=0.0533, over 7173.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2888, pruned_loss=0.06329, over 1616048.78 frames. ], batch size: 16, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:14:52,109 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 01:14:53,115 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 01:15:05,564 INFO [train.py:901] (2/4) Epoch 20, batch 6600, loss[loss=0.1772, simple_loss=0.2667, pruned_loss=0.04391, over 8279.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2884, pruned_loss=0.063, over 1612413.15 frames. ], batch size: 23, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:15:10,796 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.488e+02 3.067e+02 3.982e+02 8.719e+02, threshold=6.134e+02, percent-clipped=3.0 +2023-02-07 01:15:12,142 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 01:15:33,421 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160217.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:15:39,330 INFO [train.py:901] (2/4) Epoch 20, batch 6650, loss[loss=0.2265, simple_loss=0.3048, pruned_loss=0.07413, over 8140.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2886, pruned_loss=0.06327, over 1613869.83 frames. ], batch size: 22, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:16:10,082 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 01:16:12,500 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160272.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:16:15,123 INFO [train.py:901] (2/4) Epoch 20, batch 6700, loss[loss=0.1988, simple_loss=0.2935, pruned_loss=0.05201, over 8195.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2888, pruned_loss=0.06368, over 1613752.36 frames. ], batch size: 23, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:16:20,502 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.299e+02 2.819e+02 3.357e+02 8.975e+02, threshold=5.638e+02, percent-clipped=4.0 +2023-02-07 01:16:44,747 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-07 01:16:48,910 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160325.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:16:49,344 INFO [train.py:901] (2/4) Epoch 20, batch 6750, loss[loss=0.3196, simple_loss=0.3658, pruned_loss=0.1367, over 8621.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.29, pruned_loss=0.0646, over 1612302.76 frames. ], batch size: 39, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:16:53,599 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160332.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:03,236 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8874, 1.5383, 6.0458, 2.0660, 5.4006, 5.0518, 5.5686, 5.4728], + device='cuda:2'), covar=tensor([0.0529, 0.5110, 0.0422, 0.4043, 0.1078, 0.0941, 0.0558, 0.0526], + device='cuda:2'), in_proj_covar=tensor([0.0612, 0.0629, 0.0681, 0.0614, 0.0694, 0.0594, 0.0596, 0.0665], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 01:17:16,338 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160364.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:18,899 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6749, 1.6239, 2.1440, 1.4984, 1.3091, 2.0687, 0.6929, 1.5097], + device='cuda:2'), covar=tensor([0.1649, 0.1242, 0.0363, 0.1062, 0.2607, 0.0439, 0.2247, 0.1185], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0196, 0.0126, 0.0223, 0.0272, 0.0134, 0.0170, 0.0190], + 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') +2023-02-07 01:17:23,988 INFO [train.py:901] (2/4) Epoch 20, batch 6800, loss[loss=0.2532, simple_loss=0.3253, pruned_loss=0.09058, over 6821.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2906, pruned_loss=0.0645, over 1616172.84 frames. ], batch size: 71, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:17:28,102 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 01:17:29,321 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.510e+02 3.096e+02 3.947e+02 9.727e+02, threshold=6.192e+02, percent-clipped=5.0 +2023-02-07 01:17:32,265 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160387.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:33,649 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160389.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:59,210 INFO [train.py:901] (2/4) Epoch 20, batch 6850, loss[loss=0.2309, simple_loss=0.3173, pruned_loss=0.07225, over 8351.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2904, pruned_loss=0.06405, over 1614997.08 frames. ], batch size: 26, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:18:19,449 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 01:18:34,196 INFO [train.py:901] (2/4) Epoch 20, batch 6900, loss[loss=0.1893, simple_loss=0.2688, pruned_loss=0.05494, over 7649.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2897, pruned_loss=0.06399, over 1614111.62 frames. ], batch size: 19, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:18:39,573 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 2.333e+02 2.912e+02 3.495e+02 9.213e+02, threshold=5.824e+02, percent-clipped=3.0 +2023-02-07 01:19:08,580 INFO [train.py:901] (2/4) Epoch 20, batch 6950, loss[loss=0.2454, simple_loss=0.3222, pruned_loss=0.08427, over 8320.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2903, pruned_loss=0.06387, over 1620984.46 frames. ], batch size: 25, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:19:15,598 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1651, 4.1607, 3.7809, 1.9294, 3.6826, 3.8747, 3.6899, 3.6710], + device='cuda:2'), covar=tensor([0.0786, 0.0597, 0.1101, 0.4374, 0.0890, 0.0902, 0.1382, 0.0814], + device='cuda:2'), in_proj_covar=tensor([0.0524, 0.0434, 0.0438, 0.0540, 0.0424, 0.0439, 0.0423, 0.0382], + 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') +2023-02-07 01:19:30,217 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 01:19:42,958 INFO [train.py:901] (2/4) Epoch 20, batch 7000, loss[loss=0.1701, simple_loss=0.2626, pruned_loss=0.03882, over 8099.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2908, pruned_loss=0.06351, over 1622460.04 frames. ], batch size: 23, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:19:43,788 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5997, 1.5141, 4.8100, 1.8139, 4.2044, 4.0297, 4.3815, 4.1927], + device='cuda:2'), covar=tensor([0.0595, 0.4741, 0.0494, 0.4227, 0.1138, 0.0969, 0.0562, 0.0689], + device='cuda:2'), in_proj_covar=tensor([0.0614, 0.0631, 0.0681, 0.0617, 0.0697, 0.0595, 0.0596, 0.0668], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 01:19:48,348 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.497e+02 2.987e+02 3.377e+02 5.985e+02, threshold=5.974e+02, percent-clipped=1.0 +2023-02-07 01:19:52,063 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160588.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:07,125 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 +2023-02-07 01:20:09,555 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160613.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:18,197 INFO [train.py:901] (2/4) Epoch 20, batch 7050, loss[loss=0.2173, simple_loss=0.2966, pruned_loss=0.06902, over 8125.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2908, pruned_loss=0.06344, over 1617595.36 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:20:30,576 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:48,791 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160668.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:49,340 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160669.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:20:53,994 INFO [train.py:901] (2/4) Epoch 20, batch 7100, loss[loss=0.2047, simple_loss=0.2857, pruned_loss=0.06187, over 8239.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2899, pruned_loss=0.06332, over 1618033.84 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:20:59,618 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.520e+02 2.814e+02 3.523e+02 7.232e+02, threshold=5.628e+02, percent-clipped=2.0 +2023-02-07 01:21:06,712 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7220, 4.7670, 4.2816, 1.9297, 4.2436, 4.3671, 4.3364, 4.2357], + device='cuda:2'), covar=tensor([0.0604, 0.0431, 0.0854, 0.4421, 0.0746, 0.0763, 0.1087, 0.0619], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0433, 0.0437, 0.0538, 0.0425, 0.0441, 0.0424, 0.0382], + 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') +2023-02-07 01:21:29,432 INFO [train.py:901] (2/4) Epoch 20, batch 7150, loss[loss=0.2418, simple_loss=0.322, pruned_loss=0.08078, over 8634.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2902, pruned_loss=0.0641, over 1614638.72 frames. ], batch size: 34, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:21:39,918 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 01:22:04,615 INFO [train.py:901] (2/4) Epoch 20, batch 7200, loss[loss=0.2769, simple_loss=0.3447, pruned_loss=0.1046, over 6948.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2906, pruned_loss=0.06426, over 1615434.88 frames. ], batch size: 73, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:22:09,775 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.432e+02 3.066e+02 3.972e+02 8.502e+02, threshold=6.132e+02, percent-clipped=3.0 +2023-02-07 01:22:09,991 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160784.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:22:39,217 INFO [train.py:901] (2/4) Epoch 20, batch 7250, loss[loss=0.2111, simple_loss=0.2922, pruned_loss=0.06496, over 8338.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2925, pruned_loss=0.06521, over 1617480.15 frames. ], batch size: 49, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:23:13,934 INFO [train.py:901] (2/4) Epoch 20, batch 7300, loss[loss=0.2415, simple_loss=0.3081, pruned_loss=0.08749, over 7965.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2927, pruned_loss=0.06536, over 1619752.56 frames. ], batch size: 21, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:23:19,309 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.519e+02 2.885e+02 3.982e+02 8.183e+02, threshold=5.771e+02, percent-clipped=5.0 +2023-02-07 01:23:44,251 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160919.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:23:48,785 INFO [train.py:901] (2/4) Epoch 20, batch 7350, loss[loss=0.1797, simple_loss=0.2618, pruned_loss=0.04877, over 8230.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.291, pruned_loss=0.06434, over 1617452.36 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:23:55,277 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 01:24:16,171 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 01:24:24,315 INFO [train.py:901] (2/4) Epoch 20, batch 7400, loss[loss=0.1824, simple_loss=0.2706, pruned_loss=0.04708, over 8147.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2917, pruned_loss=0.06405, over 1622067.69 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:24:29,920 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160983.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:24:30,422 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.344e+02 3.002e+02 3.673e+02 6.079e+02, threshold=6.004e+02, percent-clipped=1.0 +2023-02-07 01:24:37,301 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 01:24:59,969 INFO [train.py:901] (2/4) Epoch 20, batch 7450, loss[loss=0.1468, simple_loss=0.2298, pruned_loss=0.03187, over 7646.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.29, pruned_loss=0.06326, over 1620152.50 frames. ], batch size: 19, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:25:01,558 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3284, 2.1129, 3.5329, 2.1521, 2.8325, 3.9837, 3.9105, 3.5139], + device='cuda:2'), covar=tensor([0.1053, 0.1520, 0.0541, 0.1690, 0.1434, 0.0202, 0.0602, 0.0481], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0321, 0.0285, 0.0315, 0.0304, 0.0262, 0.0412, 0.0303], + 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') +2023-02-07 01:25:10,076 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161040.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:25:16,123 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 01:25:28,631 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161065.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:25:35,929 INFO [train.py:901] (2/4) Epoch 20, batch 7500, loss[loss=0.2253, simple_loss=0.303, pruned_loss=0.07382, over 8664.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2912, pruned_loss=0.06387, over 1619718.84 frames. ], batch size: 49, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:25:41,428 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.441e+02 3.010e+02 3.756e+02 8.900e+02, threshold=6.020e+02, percent-clipped=5.0 +2023-02-07 01:26:11,135 INFO [train.py:901] (2/4) Epoch 20, batch 7550, loss[loss=0.2082, simple_loss=0.2969, pruned_loss=0.05974, over 8464.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2918, pruned_loss=0.06472, over 1620779.56 frames. ], batch size: 25, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:26:46,334 INFO [train.py:901] (2/4) Epoch 20, batch 7600, loss[loss=0.2005, simple_loss=0.2759, pruned_loss=0.06252, over 8033.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2917, pruned_loss=0.06457, over 1621519.70 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:26:51,651 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.460e+02 3.037e+02 4.113e+02 9.859e+02, threshold=6.074e+02, percent-clipped=9.0 +2023-02-07 01:27:20,295 INFO [train.py:901] (2/4) Epoch 20, batch 7650, loss[loss=0.1866, simple_loss=0.2665, pruned_loss=0.05334, over 7928.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2921, pruned_loss=0.06495, over 1622985.47 frames. ], batch size: 20, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:27:25,750 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161234.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:27:36,289 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161249.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:27:45,607 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161263.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:27:54,278 INFO [train.py:901] (2/4) Epoch 20, batch 7700, loss[loss=0.2202, simple_loss=0.3023, pruned_loss=0.06906, over 8362.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2927, pruned_loss=0.065, over 1625949.94 frames. ], batch size: 24, lr: 3.75e-03, grad_scale: 16.0 +2023-02-07 01:27:59,457 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.411e+02 2.987e+02 3.572e+02 6.786e+02, threshold=5.975e+02, percent-clipped=3.0 +2023-02-07 01:28:22,560 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2882, 2.0102, 2.7066, 2.2042, 2.5153, 2.2724, 2.0499, 1.4251], + device='cuda:2'), covar=tensor([0.5188, 0.4734, 0.1714, 0.3674, 0.2505, 0.2923, 0.1878, 0.5144], + device='cuda:2'), in_proj_covar=tensor([0.0935, 0.0970, 0.0799, 0.0932, 0.0989, 0.0882, 0.0742, 0.0819], + 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') +2023-02-07 01:28:25,737 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 01:28:26,650 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5906, 2.7115, 1.9223, 2.2854, 2.3049, 1.5787, 2.1694, 2.4033], + device='cuda:2'), covar=tensor([0.1602, 0.0419, 0.1205, 0.0689, 0.0782, 0.1558, 0.1055, 0.1018], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0237, 0.0333, 0.0306, 0.0301, 0.0335, 0.0345, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 01:28:29,787 INFO [train.py:901] (2/4) Epoch 20, batch 7750, loss[loss=0.2039, simple_loss=0.2812, pruned_loss=0.06331, over 8242.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2909, pruned_loss=0.06434, over 1616595.34 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 16.0 +2023-02-07 01:28:30,570 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:05,270 INFO [train.py:901] (2/4) Epoch 20, batch 7800, loss[loss=0.2078, simple_loss=0.2882, pruned_loss=0.06366, over 7814.00 frames. ], tot_loss[loss=0.21, simple_loss=0.291, pruned_loss=0.06446, over 1616276.62 frames. ], batch size: 20, lr: 3.75e-03, grad_scale: 16.0 +2023-02-07 01:29:06,860 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:10,616 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.429e+02 2.909e+02 3.732e+02 6.331e+02, threshold=5.818e+02, percent-clipped=2.0 +2023-02-07 01:29:34,440 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161419.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:37,519 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-07 01:29:39,043 INFO [train.py:901] (2/4) Epoch 20, batch 7850, loss[loss=0.2405, simple_loss=0.318, pruned_loss=0.08151, over 8646.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2908, pruned_loss=0.0645, over 1615612.01 frames. ], batch size: 50, lr: 3.74e-03, grad_scale: 16.0 +2023-02-07 01:29:49,661 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161442.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:56,804 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5687, 2.7147, 1.9156, 2.2760, 2.3041, 1.6289, 2.0660, 2.3506], + device='cuda:2'), covar=tensor([0.1771, 0.0410, 0.1153, 0.0663, 0.0740, 0.1508, 0.1082, 0.0997], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0234, 0.0330, 0.0303, 0.0298, 0.0332, 0.0342, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 01:30:12,443 INFO [train.py:901] (2/4) Epoch 20, batch 7900, loss[loss=0.2055, simple_loss=0.2887, pruned_loss=0.06118, over 8352.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2903, pruned_loss=0.06401, over 1615099.06 frames. ], batch size: 24, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:30:13,741 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161478.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:30:15,380 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 01:30:18,875 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.320e+02 2.352e+02 2.923e+02 4.060e+02 8.940e+02, threshold=5.846e+02, percent-clipped=3.0 +2023-02-07 01:30:45,510 INFO [train.py:901] (2/4) Epoch 20, batch 7950, loss[loss=0.1696, simple_loss=0.26, pruned_loss=0.03962, over 7959.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2887, pruned_loss=0.0636, over 1610048.81 frames. ], batch size: 21, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:30:50,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8338, 3.7764, 3.4972, 1.8953, 3.4036, 3.4069, 3.4633, 3.2976], + device='cuda:2'), covar=tensor([0.0932, 0.0663, 0.1131, 0.4254, 0.0969, 0.1171, 0.1471, 0.0984], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0432, 0.0431, 0.0531, 0.0424, 0.0435, 0.0419, 0.0379], + 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') +2023-02-07 01:31:18,035 INFO [train.py:901] (2/4) Epoch 20, batch 8000, loss[loss=0.1734, simple_loss=0.2484, pruned_loss=0.04921, over 7533.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2884, pruned_loss=0.06363, over 1607748.41 frames. ], batch size: 18, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:31:19,436 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161578.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:31:23,852 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.449e+02 3.108e+02 3.740e+02 8.675e+02, threshold=6.215e+02, percent-clipped=6.0 +2023-02-07 01:31:29,406 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161593.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:31:36,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6607, 2.6881, 1.9200, 2.4009, 2.3543, 1.5969, 2.2255, 2.3836], + device='cuda:2'), covar=tensor([0.1605, 0.0427, 0.1234, 0.0640, 0.0740, 0.1622, 0.0947, 0.0926], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0234, 0.0330, 0.0303, 0.0298, 0.0332, 0.0341, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 01:31:51,325 INFO [train.py:901] (2/4) Epoch 20, batch 8050, loss[loss=0.1876, simple_loss=0.2748, pruned_loss=0.05015, over 6391.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2872, pruned_loss=0.06302, over 1599417.66 frames. ], batch size: 14, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:31:57,080 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161634.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:32:25,047 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 01:32:32,189 INFO [train.py:901] (2/4) Epoch 21, batch 0, loss[loss=0.1867, simple_loss=0.2611, pruned_loss=0.05613, over 8093.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2611, pruned_loss=0.05613, over 8093.00 frames. ], batch size: 21, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:32:32,190 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 01:32:44,209 INFO [train.py:935] (2/4) Epoch 21, validation: loss=0.1763, simple_loss=0.2762, pruned_loss=0.03818, over 944034.00 frames. +2023-02-07 01:32:44,210 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 01:32:44,430 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161659.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:32:59,349 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 01:33:02,224 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.415e+02 2.918e+02 3.924e+02 7.413e+02, threshold=5.835e+02, percent-clipped=4.0 +2023-02-07 01:33:07,970 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161693.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:11,616 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:18,551 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5540, 1.4710, 4.7469, 1.7547, 4.1237, 3.9858, 4.3382, 4.2036], + device='cuda:2'), covar=tensor([0.0585, 0.4756, 0.0466, 0.4029, 0.1220, 0.0936, 0.0548, 0.0652], + device='cuda:2'), in_proj_covar=tensor([0.0610, 0.0623, 0.0673, 0.0608, 0.0690, 0.0590, 0.0591, 0.0656], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 01:33:18,597 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161708.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:19,107 INFO [train.py:901] (2/4) Epoch 21, batch 50, loss[loss=0.2372, simple_loss=0.319, pruned_loss=0.07771, over 8190.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2934, pruned_loss=0.06612, over 368383.40 frames. ], batch size: 23, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:33:29,238 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:32,475 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 01:33:56,011 INFO [train.py:901] (2/4) Epoch 21, batch 100, loss[loss=0.1877, simple_loss=0.2689, pruned_loss=0.05326, over 7532.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.293, pruned_loss=0.06531, over 646436.88 frames. ], batch size: 18, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:33:57,250 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 01:33:58,658 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161763.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:34:14,164 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.511e+02 2.964e+02 4.065e+02 7.207e+02, threshold=5.927e+02, percent-clipped=4.0 +2023-02-07 01:34:18,656 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 01:34:30,766 INFO [train.py:901] (2/4) Epoch 21, batch 150, loss[loss=0.1546, simple_loss=0.245, pruned_loss=0.03213, over 7813.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2939, pruned_loss=0.06565, over 864878.42 frames. ], batch size: 20, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:34:33,279 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 01:34:39,725 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161822.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:34:47,260 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:35:06,331 INFO [train.py:901] (2/4) Epoch 21, batch 200, loss[loss=0.197, simple_loss=0.2654, pruned_loss=0.06425, over 7427.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2914, pruned_loss=0.06476, over 1029982.38 frames. ], batch size: 17, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:35:19,119 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161878.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:35:23,717 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.494e+02 2.791e+02 3.613e+02 7.338e+02, threshold=5.582e+02, percent-clipped=1.0 +2023-02-07 01:35:41,064 INFO [train.py:901] (2/4) Epoch 21, batch 250, loss[loss=0.2253, simple_loss=0.3095, pruned_loss=0.07053, over 8112.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2913, pruned_loss=0.06482, over 1157313.26 frames. ], batch size: 23, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:35:47,945 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 01:35:57,067 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 01:36:00,717 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161937.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:08,760 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161949.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:15,267 INFO [train.py:901] (2/4) Epoch 21, batch 300, loss[loss=0.2355, simple_loss=0.312, pruned_loss=0.07956, over 8488.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2922, pruned_loss=0.06493, over 1265340.80 frames. ], batch size: 48, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:36:19,021 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161964.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:20,639 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 01:36:26,617 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161974.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:33,754 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.407e+02 2.839e+02 3.558e+02 8.067e+02, threshold=5.678e+02, percent-clipped=5.0 +2023-02-07 01:36:36,660 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161989.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:51,874 INFO [train.py:901] (2/4) Epoch 21, batch 350, loss[loss=0.2948, simple_loss=0.3489, pruned_loss=0.1203, over 7093.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2899, pruned_loss=0.06355, over 1342194.00 frames. ], batch size: 71, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:36:57,889 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2023-02-07 01:37:25,806 INFO [train.py:901] (2/4) Epoch 21, batch 400, loss[loss=0.219, simple_loss=0.291, pruned_loss=0.07344, over 7813.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2904, pruned_loss=0.06383, over 1402486.18 frames. ], batch size: 20, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:37:44,476 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.323e+02 2.796e+02 3.394e+02 5.024e+02, threshold=5.592e+02, percent-clipped=0.0 +2023-02-07 01:37:52,929 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162095.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:37:58,454 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7314, 1.4881, 4.9302, 1.9156, 4.4189, 4.1634, 4.5072, 4.3877], + device='cuda:2'), covar=tensor([0.0480, 0.4603, 0.0400, 0.3578, 0.0891, 0.0732, 0.0512, 0.0539], + device='cuda:2'), in_proj_covar=tensor([0.0612, 0.0629, 0.0679, 0.0612, 0.0692, 0.0594, 0.0596, 0.0661], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 01:38:02,163 INFO [train.py:901] (2/4) Epoch 21, batch 450, loss[loss=0.2013, simple_loss=0.2893, pruned_loss=0.05668, over 8241.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2903, pruned_loss=0.06385, over 1452023.27 frames. ], batch size: 24, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:38:20,172 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162134.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:37,308 INFO [train.py:901] (2/4) Epoch 21, batch 500, loss[loss=0.2032, simple_loss=0.2979, pruned_loss=0.05419, over 8238.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2911, pruned_loss=0.064, over 1489739.11 frames. ], batch size: 24, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:38:37,582 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162159.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:50,065 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162177.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:55,586 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.501e+02 2.975e+02 3.750e+02 9.376e+02, threshold=5.950e+02, percent-clipped=8.0 +2023-02-07 01:39:01,456 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162193.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:39:09,311 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-07 01:39:13,736 INFO [train.py:901] (2/4) Epoch 21, batch 550, loss[loss=0.1775, simple_loss=0.2611, pruned_loss=0.0469, over 7796.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2905, pruned_loss=0.06332, over 1519346.68 frames. ], batch size: 19, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:39:20,156 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162218.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:39:42,304 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162249.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:39:45,165 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.84 vs. limit=5.0 +2023-02-07 01:39:48,780 INFO [train.py:901] (2/4) Epoch 21, batch 600, loss[loss=0.2273, simple_loss=0.2996, pruned_loss=0.07745, over 8028.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2895, pruned_loss=0.0635, over 1537853.73 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:40:02,428 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 01:40:06,587 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.365e+02 2.932e+02 3.412e+02 7.385e+02, threshold=5.863e+02, percent-clipped=2.0 +2023-02-07 01:40:11,434 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162292.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:40:16,973 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7180, 2.0735, 3.3524, 1.5349, 2.6268, 2.1966, 1.7836, 2.5794], + device='cuda:2'), covar=tensor([0.1848, 0.2467, 0.0761, 0.4431, 0.1619, 0.2968, 0.2210, 0.2151], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0592, 0.0549, 0.0631, 0.0640, 0.0591, 0.0529, 0.0631], + 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') +2023-02-07 01:40:22,966 INFO [train.py:901] (2/4) Epoch 21, batch 650, loss[loss=0.198, simple_loss=0.2827, pruned_loss=0.05671, over 8326.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2891, pruned_loss=0.06292, over 1557753.45 frames. ], batch size: 25, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:40:26,759 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 +2023-02-07 01:40:32,758 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2353, 2.5563, 2.9893, 1.7281, 3.1559, 1.9448, 1.5378, 2.0869], + device='cuda:2'), covar=tensor([0.0747, 0.0357, 0.0257, 0.0680, 0.0382, 0.0776, 0.0906, 0.0534], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0384, 0.0336, 0.0438, 0.0368, 0.0531, 0.0388, 0.0412], + device='cuda:2'), out_proj_covar=tensor([1.2043e-04, 1.0076e-04, 8.8479e-05, 1.1571e-04, 9.6954e-05, 1.5088e-04, + 1.0492e-04, 1.0957e-04], device='cuda:2') +2023-02-07 01:40:35,429 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2531, 3.1424, 2.9538, 1.6500, 2.8535, 2.9326, 2.8509, 2.8528], + device='cuda:2'), covar=tensor([0.1146, 0.0988, 0.1449, 0.4445, 0.1192, 0.1252, 0.1716, 0.0950], + device='cuda:2'), in_proj_covar=tensor([0.0515, 0.0432, 0.0430, 0.0531, 0.0419, 0.0435, 0.0413, 0.0378], + 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') +2023-02-07 01:40:59,236 INFO [train.py:901] (2/4) Epoch 21, batch 700, loss[loss=0.194, simple_loss=0.2878, pruned_loss=0.05012, over 8329.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2898, pruned_loss=0.06306, over 1568152.64 frames. ], batch size: 25, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:41:17,771 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.443e+02 3.111e+02 4.032e+02 8.821e+02, threshold=6.222e+02, percent-clipped=5.0 +2023-02-07 01:41:34,562 INFO [train.py:901] (2/4) Epoch 21, batch 750, loss[loss=0.2691, simple_loss=0.334, pruned_loss=0.1021, over 7338.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2902, pruned_loss=0.06348, over 1575547.23 frames. ], batch size: 72, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:41:45,486 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 01:41:49,073 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1908, 3.8538, 2.5360, 2.8223, 3.0757, 2.3371, 2.9746, 3.1294], + device='cuda:2'), covar=tensor([0.1613, 0.0400, 0.1160, 0.0773, 0.0702, 0.1303, 0.1094, 0.1031], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0236, 0.0334, 0.0308, 0.0300, 0.0336, 0.0347, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 01:41:54,320 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 01:41:56,380 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:42:11,055 INFO [train.py:901] (2/4) Epoch 21, batch 800, loss[loss=0.1989, simple_loss=0.2863, pruned_loss=0.05568, over 8237.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2902, pruned_loss=0.06316, over 1587921.30 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:42:29,948 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.455e+02 2.861e+02 3.570e+02 7.084e+02, threshold=5.721e+02, percent-clipped=3.0 +2023-02-07 01:42:47,169 INFO [train.py:901] (2/4) Epoch 21, batch 850, loss[loss=0.1959, simple_loss=0.2757, pruned_loss=0.05806, over 8297.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2898, pruned_loss=0.06327, over 1593784.91 frames. ], batch size: 23, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:43:01,484 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162529.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:16,266 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:18,607 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2023-02-07 01:43:21,103 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:24,449 INFO [train.py:901] (2/4) Epoch 21, batch 900, loss[loss=0.2159, simple_loss=0.2976, pruned_loss=0.06709, over 8335.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2896, pruned_loss=0.06343, over 1594483.34 frames. ], batch size: 25, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:43:34,390 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162573.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:38,964 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.53 vs. limit=5.0 +2023-02-07 01:43:42,630 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.319e+02 2.838e+02 3.637e+02 1.203e+03, threshold=5.677e+02, percent-clipped=5.0 +2023-02-07 01:43:49,237 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162593.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:44:00,580 INFO [train.py:901] (2/4) Epoch 21, batch 950, loss[loss=0.1974, simple_loss=0.2897, pruned_loss=0.05252, over 8862.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2897, pruned_loss=0.06311, over 1600157.17 frames. ], batch size: 40, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:44:14,210 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 01:44:35,848 INFO [train.py:901] (2/4) Epoch 21, batch 1000, loss[loss=0.1994, simple_loss=0.2816, pruned_loss=0.05861, over 7932.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2898, pruned_loss=0.06296, over 1599568.32 frames. ], batch size: 20, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:44:48,971 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 01:44:55,206 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.438e+02 2.954e+02 4.014e+02 9.557e+02, threshold=5.908e+02, percent-clipped=3.0 +2023-02-07 01:45:01,390 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 01:45:11,703 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162708.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:45:12,207 INFO [train.py:901] (2/4) Epoch 21, batch 1050, loss[loss=0.2096, simple_loss=0.2985, pruned_loss=0.06034, over 8436.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2894, pruned_loss=0.06251, over 1601781.49 frames. ], batch size: 27, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:45:17,835 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7046, 1.7426, 2.3332, 1.5118, 1.4244, 2.2920, 0.4466, 1.4433], + device='cuda:2'), covar=tensor([0.1779, 0.1347, 0.0315, 0.1317, 0.2419, 0.0418, 0.2173, 0.1388], + device='cuda:2'), in_proj_covar=tensor([0.0187, 0.0193, 0.0125, 0.0219, 0.0269, 0.0132, 0.0168, 0.0188], + 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') +2023-02-07 01:45:46,483 INFO [train.py:901] (2/4) Epoch 21, batch 1100, loss[loss=0.1832, simple_loss=0.261, pruned_loss=0.05268, over 7811.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2895, pruned_loss=0.06251, over 1606562.82 frames. ], batch size: 19, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:46:06,018 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.501e+02 3.059e+02 3.494e+02 1.150e+03, threshold=6.119e+02, percent-clipped=4.0 +2023-02-07 01:46:14,526 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 01:46:23,789 INFO [train.py:901] (2/4) Epoch 21, batch 1150, loss[loss=0.2074, simple_loss=0.2925, pruned_loss=0.06119, over 8466.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2893, pruned_loss=0.06233, over 1606998.51 frames. ], batch size: 25, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:46:24,671 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162810.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:46:42,980 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162835.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:46:50,072 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162845.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:46:54,517 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 01:46:59,716 INFO [train.py:901] (2/4) Epoch 21, batch 1200, loss[loss=0.1844, simple_loss=0.2661, pruned_loss=0.0514, over 8244.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2894, pruned_loss=0.0622, over 1607893.94 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:47:09,540 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162873.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:47:17,443 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.368e+02 3.051e+02 3.779e+02 6.869e+02, threshold=6.103e+02, percent-clipped=3.0 +2023-02-07 01:47:36,397 INFO [train.py:901] (2/4) Epoch 21, batch 1250, loss[loss=0.2517, simple_loss=0.3382, pruned_loss=0.08264, over 8478.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.289, pruned_loss=0.06212, over 1613687.80 frames. ], batch size: 29, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:47:58,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9528, 2.0700, 1.8134, 2.4529, 1.2674, 1.6582, 1.9706, 2.0891], + device='cuda:2'), covar=tensor([0.0660, 0.0721, 0.0905, 0.0499, 0.1059, 0.1156, 0.0738, 0.0706], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0195, 0.0243, 0.0211, 0.0203, 0.0244, 0.0247, 0.0206], + 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') +2023-02-07 01:48:11,274 INFO [train.py:901] (2/4) Epoch 21, batch 1300, loss[loss=0.1656, simple_loss=0.2358, pruned_loss=0.04766, over 7712.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2887, pruned_loss=0.06226, over 1612770.04 frames. ], batch size: 18, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:48:14,766 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162964.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:48:28,476 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.260e+02 2.727e+02 3.317e+02 5.773e+02, threshold=5.453e+02, percent-clipped=0.0 +2023-02-07 01:48:30,745 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162988.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:48:31,449 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162989.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:48:44,793 INFO [train.py:901] (2/4) Epoch 21, batch 1350, loss[loss=0.1732, simple_loss=0.2553, pruned_loss=0.04556, over 7932.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2896, pruned_loss=0.06315, over 1606014.84 frames. ], batch size: 20, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:49:21,731 INFO [train.py:901] (2/4) Epoch 21, batch 1400, loss[loss=0.2095, simple_loss=0.2916, pruned_loss=0.06375, over 8637.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2895, pruned_loss=0.06325, over 1606395.47 frames. ], batch size: 39, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:49:39,401 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.469e+02 3.010e+02 4.050e+02 1.060e+03, threshold=6.020e+02, percent-clipped=5.0 +2023-02-07 01:49:46,305 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 01:49:55,441 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:49:55,924 INFO [train.py:901] (2/4) Epoch 21, batch 1450, loss[loss=0.2893, simple_loss=0.3439, pruned_loss=0.1173, over 7664.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2908, pruned_loss=0.06409, over 1608546.34 frames. ], batch size: 71, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:50:01,717 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7613, 1.5269, 3.1779, 1.3690, 2.3155, 3.4112, 3.5257, 2.9432], + device='cuda:2'), covar=tensor([0.1201, 0.1584, 0.0341, 0.2032, 0.0900, 0.0251, 0.0531, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0294, 0.0319, 0.0289, 0.0313, 0.0307, 0.0263, 0.0412, 0.0304], + 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') +2023-02-07 01:50:32,151 INFO [train.py:901] (2/4) Epoch 21, batch 1500, loss[loss=0.1758, simple_loss=0.2588, pruned_loss=0.04642, over 7227.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2905, pruned_loss=0.06388, over 1608296.86 frames. ], batch size: 16, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:50:50,502 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.589e+02 2.250e+02 2.722e+02 3.392e+02 6.898e+02, threshold=5.444e+02, percent-clipped=4.0 +2023-02-07 01:50:51,635 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.96 vs. limit=5.0 +2023-02-07 01:50:53,290 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163189.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:02,320 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8649, 2.3832, 4.4404, 1.6251, 3.2514, 2.5496, 1.9679, 3.2287], + device='cuda:2'), covar=tensor([0.1916, 0.2833, 0.0822, 0.4801, 0.1688, 0.3106, 0.2457, 0.2185], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0592, 0.0554, 0.0636, 0.0641, 0.0589, 0.0531, 0.0630], + 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') +2023-02-07 01:51:06,797 INFO [train.py:901] (2/4) Epoch 21, batch 1550, loss[loss=0.2076, simple_loss=0.3035, pruned_loss=0.05584, over 8321.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2896, pruned_loss=0.06337, over 1608046.60 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:51:31,315 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163244.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:42,280 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163258.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:42,813 INFO [train.py:901] (2/4) Epoch 21, batch 1600, loss[loss=0.198, simple_loss=0.2746, pruned_loss=0.06076, over 7549.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2895, pruned_loss=0.0632, over 1610709.95 frames. ], batch size: 18, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:51:50,628 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163269.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:56,978 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 01:52:00,875 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.380e+02 3.009e+02 4.081e+02 9.131e+02, threshold=6.018e+02, percent-clipped=6.0 +2023-02-07 01:52:14,555 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163304.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:52:17,715 INFO [train.py:901] (2/4) Epoch 21, batch 1650, loss[loss=0.2215, simple_loss=0.2949, pruned_loss=0.07401, over 7711.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2897, pruned_loss=0.06304, over 1614972.61 frames. ], batch size: 18, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:52:50,771 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1880, 4.1840, 3.7949, 1.9027, 3.6849, 3.8011, 3.7789, 3.5890], + device='cuda:2'), covar=tensor([0.0753, 0.0585, 0.1066, 0.4579, 0.0891, 0.0978, 0.1294, 0.0839], + device='cuda:2'), in_proj_covar=tensor([0.0511, 0.0428, 0.0426, 0.0532, 0.0421, 0.0432, 0.0415, 0.0377], + 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') +2023-02-07 01:52:51,355 INFO [train.py:901] (2/4) Epoch 21, batch 1700, loss[loss=0.2004, simple_loss=0.2791, pruned_loss=0.06092, over 8290.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2906, pruned_loss=0.06393, over 1617130.18 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:53:09,954 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.344e+02 2.897e+02 3.678e+02 1.033e+03, threshold=5.793e+02, percent-clipped=5.0 +2023-02-07 01:53:27,413 INFO [train.py:901] (2/4) Epoch 21, batch 1750, loss[loss=0.1812, simple_loss=0.2677, pruned_loss=0.04733, over 8142.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2904, pruned_loss=0.06396, over 1615545.85 frames. ], batch size: 22, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:53:39,432 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3811, 2.6397, 3.0935, 1.8179, 3.3605, 2.0492, 1.6592, 2.2410], + device='cuda:2'), covar=tensor([0.0696, 0.0321, 0.0226, 0.0658, 0.0403, 0.0786, 0.0743, 0.0482], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0386, 0.0336, 0.0439, 0.0369, 0.0531, 0.0387, 0.0415], + device='cuda:2'), out_proj_covar=tensor([1.2041e-04, 1.0128e-04, 8.8523e-05, 1.1595e-04, 9.7193e-05, 1.5056e-04, + 1.0471e-04, 1.1038e-04], device='cuda:2') +2023-02-07 01:53:49,137 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.46 vs. limit=5.0 +2023-02-07 01:53:56,338 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:54:01,060 INFO [train.py:901] (2/4) Epoch 21, batch 1800, loss[loss=0.2156, simple_loss=0.2974, pruned_loss=0.06683, over 8105.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2896, pruned_loss=0.06317, over 1616067.54 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:54:18,713 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.661e+02 3.025e+02 4.067e+02 7.408e+02, threshold=6.049e+02, percent-clipped=6.0 +2023-02-07 01:54:37,342 INFO [train.py:901] (2/4) Epoch 21, batch 1850, loss[loss=0.2069, simple_loss=0.281, pruned_loss=0.06641, over 8112.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2889, pruned_loss=0.06308, over 1614287.01 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:54:53,428 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.72 vs. limit=5.0 +2023-02-07 01:55:07,798 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9369, 1.7083, 1.9219, 1.7944, 1.1654, 1.7448, 2.2904, 2.1927], + device='cuda:2'), covar=tensor([0.0416, 0.1237, 0.1670, 0.1371, 0.0625, 0.1440, 0.0613, 0.0581], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0158, 0.0099, 0.0162, 0.0113, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 01:55:11,340 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 01:55:11,696 INFO [train.py:901] (2/4) Epoch 21, batch 1900, loss[loss=0.1791, simple_loss=0.263, pruned_loss=0.04759, over 7972.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.288, pruned_loss=0.06236, over 1619236.91 frames. ], batch size: 21, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:55:12,601 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163560.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:17,242 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:26,433 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 01:55:28,236 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 01:55:29,011 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.410e+02 2.798e+02 3.588e+02 7.290e+02, threshold=5.595e+02, percent-clipped=1.0 +2023-02-07 01:55:29,212 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163585.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:37,705 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 01:55:40,618 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163602.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:45,159 INFO [train.py:901] (2/4) Epoch 21, batch 1950, loss[loss=0.2238, simple_loss=0.3138, pruned_loss=0.06687, over 8458.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2891, pruned_loss=0.06226, over 1627438.02 frames. ], batch size: 27, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:55:58,503 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 01:56:21,640 INFO [train.py:901] (2/4) Epoch 21, batch 2000, loss[loss=0.2048, simple_loss=0.2823, pruned_loss=0.06364, over 8086.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2889, pruned_loss=0.06192, over 1627262.25 frames. ], batch size: 21, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:56:39,058 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.546e+02 3.013e+02 3.975e+02 6.874e+02, threshold=6.025e+02, percent-clipped=4.0 +2023-02-07 01:56:44,558 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7143, 1.6323, 2.8228, 1.3182, 2.1454, 3.0349, 3.1614, 2.5963], + device='cuda:2'), covar=tensor([0.1200, 0.1493, 0.0400, 0.2102, 0.1008, 0.0291, 0.0598, 0.0593], + device='cuda:2'), in_proj_covar=tensor([0.0293, 0.0319, 0.0287, 0.0312, 0.0306, 0.0260, 0.0411, 0.0304], + 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') +2023-02-07 01:56:55,198 INFO [train.py:901] (2/4) Epoch 21, batch 2050, loss[loss=0.2151, simple_loss=0.2976, pruned_loss=0.0663, over 8111.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2892, pruned_loss=0.06246, over 1624910.81 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:57:00,609 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163717.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:57:30,300 INFO [train.py:901] (2/4) Epoch 21, batch 2100, loss[loss=0.2106, simple_loss=0.2879, pruned_loss=0.06664, over 8466.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2887, pruned_loss=0.06251, over 1619361.11 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:57:48,244 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1875, 2.0218, 2.7398, 2.2270, 2.6775, 2.2779, 2.0671, 1.5617], + device='cuda:2'), covar=tensor([0.5653, 0.5041, 0.1917, 0.3989, 0.2665, 0.3091, 0.1985, 0.5601], + device='cuda:2'), in_proj_covar=tensor([0.0942, 0.0974, 0.0798, 0.0935, 0.0994, 0.0887, 0.0743, 0.0824], + 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') +2023-02-07 01:57:48,642 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.575e+02 2.946e+02 3.630e+02 8.805e+02, threshold=5.893e+02, percent-clipped=3.0 +2023-02-07 01:58:04,870 INFO [train.py:901] (2/4) Epoch 21, batch 2150, loss[loss=0.2111, simple_loss=0.2898, pruned_loss=0.06625, over 7922.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2889, pruned_loss=0.06286, over 1619441.84 frames. ], batch size: 20, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:58:14,694 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163823.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:58:31,302 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163848.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:58:39,334 INFO [train.py:901] (2/4) Epoch 21, batch 2200, loss[loss=0.1819, simple_loss=0.2738, pruned_loss=0.04501, over 8324.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.289, pruned_loss=0.06297, over 1618532.70 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:58:49,741 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8828, 1.7432, 2.4862, 1.6555, 1.4227, 2.4142, 0.3614, 1.4883], + device='cuda:2'), covar=tensor([0.1432, 0.1426, 0.0301, 0.1122, 0.2883, 0.0389, 0.2315, 0.1355], + device='cuda:2'), in_proj_covar=tensor([0.0186, 0.0195, 0.0126, 0.0221, 0.0271, 0.0132, 0.0168, 0.0189], + 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') +2023-02-07 01:58:53,008 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0405, 2.4884, 3.7160, 2.1366, 2.0916, 3.6261, 0.8912, 2.2910], + device='cuda:2'), covar=tensor([0.1269, 0.1334, 0.0212, 0.1623, 0.2451, 0.0343, 0.2221, 0.1461], + device='cuda:2'), in_proj_covar=tensor([0.0186, 0.0195, 0.0126, 0.0221, 0.0270, 0.0133, 0.0168, 0.0189], + 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') +2023-02-07 01:58:58,248 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.475e+02 2.987e+02 3.670e+02 7.762e+02, threshold=5.973e+02, percent-clipped=3.0 +2023-02-07 01:59:15,124 INFO [train.py:901] (2/4) Epoch 21, batch 2250, loss[loss=0.2152, simple_loss=0.3031, pruned_loss=0.06358, over 8465.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2893, pruned_loss=0.06292, over 1622272.48 frames. ], batch size: 25, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 01:59:49,147 INFO [train.py:901] (2/4) Epoch 21, batch 2300, loss[loss=0.2082, simple_loss=0.2805, pruned_loss=0.06794, over 7820.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2894, pruned_loss=0.06319, over 1619603.03 frames. ], batch size: 20, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 01:59:58,910 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163973.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:00:08,073 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.361e+02 2.889e+02 3.736e+02 8.411e+02, threshold=5.778e+02, percent-clipped=4.0 +2023-02-07 02:00:17,857 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163998.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:00:26,190 INFO [train.py:901] (2/4) Epoch 21, batch 2350, loss[loss=0.2085, simple_loss=0.2861, pruned_loss=0.06544, over 8721.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2885, pruned_loss=0.06266, over 1619646.10 frames. ], batch size: 49, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 02:00:35,300 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6823, 4.7052, 4.1646, 2.3925, 4.1746, 4.3668, 4.1890, 4.2489], + device='cuda:2'), covar=tensor([0.0721, 0.0504, 0.1125, 0.4669, 0.0861, 0.1113, 0.1402, 0.0768], + device='cuda:2'), in_proj_covar=tensor([0.0516, 0.0433, 0.0434, 0.0538, 0.0426, 0.0440, 0.0421, 0.0382], + 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') +2023-02-07 02:01:01,226 INFO [train.py:901] (2/4) Epoch 21, batch 2400, loss[loss=0.2231, simple_loss=0.2971, pruned_loss=0.07456, over 8134.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2886, pruned_loss=0.06288, over 1618457.10 frames. ], batch size: 22, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 02:01:19,286 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.419e+02 2.926e+02 3.800e+02 6.132e+02, threshold=5.852e+02, percent-clipped=4.0 +2023-02-07 02:01:37,286 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:01:37,736 INFO [train.py:901] (2/4) Epoch 21, batch 2450, loss[loss=0.2298, simple_loss=0.3073, pruned_loss=0.07614, over 8558.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2888, pruned_loss=0.06292, over 1620744.59 frames. ], batch size: 31, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 02:02:12,735 INFO [train.py:901] (2/4) Epoch 21, batch 2500, loss[loss=0.1963, simple_loss=0.2825, pruned_loss=0.05507, over 7803.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.288, pruned_loss=0.06229, over 1621909.76 frames. ], batch size: 19, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:02:22,153 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164173.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:02:30,792 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.421e+02 3.174e+02 4.025e+02 1.090e+03, threshold=6.349e+02, percent-clipped=9.0 +2023-02-07 02:02:46,232 INFO [train.py:901] (2/4) Epoch 21, batch 2550, loss[loss=0.196, simple_loss=0.2813, pruned_loss=0.05538, over 7656.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.289, pruned_loss=0.06243, over 1626742.05 frames. ], batch size: 19, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:02:55,032 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1800, 2.3180, 1.8231, 2.8148, 1.2952, 1.6524, 1.9658, 2.2734], + device='cuda:2'), covar=tensor([0.0683, 0.0716, 0.0943, 0.0377, 0.1173, 0.1313, 0.0913, 0.0801], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0195, 0.0244, 0.0212, 0.0204, 0.0244, 0.0250, 0.0208], + 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') +2023-02-07 02:03:22,649 INFO [train.py:901] (2/4) Epoch 21, batch 2600, loss[loss=0.2238, simple_loss=0.304, pruned_loss=0.07185, over 8341.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.289, pruned_loss=0.06235, over 1625616.75 frames. ], batch size: 26, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:03:40,890 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.272e+02 2.670e+02 3.622e+02 6.852e+02, threshold=5.341e+02, percent-clipped=1.0 +2023-02-07 02:03:56,816 INFO [train.py:901] (2/4) Epoch 21, batch 2650, loss[loss=0.209, simple_loss=0.3014, pruned_loss=0.05825, over 8299.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2892, pruned_loss=0.0625, over 1625431.91 frames. ], batch size: 23, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:04:33,157 INFO [train.py:901] (2/4) Epoch 21, batch 2700, loss[loss=0.2147, simple_loss=0.2817, pruned_loss=0.07381, over 7794.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2881, pruned_loss=0.0621, over 1621441.63 frames. ], batch size: 19, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:04:46,950 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:04:52,078 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.228e+02 2.697e+02 3.361e+02 7.045e+02, threshold=5.394e+02, percent-clipped=4.0 +2023-02-07 02:05:07,796 INFO [train.py:901] (2/4) Epoch 21, batch 2750, loss[loss=0.2041, simple_loss=0.2954, pruned_loss=0.05637, over 8774.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2874, pruned_loss=0.06162, over 1619709.01 frames. ], batch size: 30, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:05:36,818 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:05:42,224 INFO [train.py:901] (2/4) Epoch 21, batch 2800, loss[loss=0.1841, simple_loss=0.272, pruned_loss=0.04815, over 7812.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2884, pruned_loss=0.06202, over 1621370.40 frames. ], batch size: 20, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:06:02,580 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.305e+02 2.813e+02 3.760e+02 7.507e+02, threshold=5.625e+02, percent-clipped=3.0 +2023-02-07 02:06:18,052 INFO [train.py:901] (2/4) Epoch 21, batch 2850, loss[loss=0.206, simple_loss=0.2918, pruned_loss=0.06008, over 8496.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2878, pruned_loss=0.06205, over 1619411.72 frames. ], batch size: 26, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:06:23,426 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:06:51,364 INFO [train.py:901] (2/4) Epoch 21, batch 2900, loss[loss=0.2209, simple_loss=0.3046, pruned_loss=0.06853, over 8475.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2881, pruned_loss=0.06186, over 1617655.30 frames. ], batch size: 25, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:06:56,997 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:07:09,755 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 02:07:11,685 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.599e+02 3.265e+02 4.069e+02 1.074e+03, threshold=6.531e+02, percent-clipped=8.0 +2023-02-07 02:07:11,862 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:07:19,712 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164596.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:07:28,727 INFO [train.py:901] (2/4) Epoch 21, batch 2950, loss[loss=0.2256, simple_loss=0.3162, pruned_loss=0.06749, over 8489.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2881, pruned_loss=0.06147, over 1617892.81 frames. ], batch size: 26, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:07:44,449 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:08:02,296 INFO [train.py:901] (2/4) Epoch 21, batch 3000, loss[loss=0.1933, simple_loss=0.2852, pruned_loss=0.05071, over 8474.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2883, pruned_loss=0.06196, over 1616139.48 frames. ], batch size: 25, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:08:02,296 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 02:08:15,066 INFO [train.py:935] (2/4) Epoch 21, validation: loss=0.1742, simple_loss=0.2744, pruned_loss=0.03706, over 944034.00 frames. +2023-02-07 02:08:15,067 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 02:08:26,763 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164676.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:08:33,568 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.380e+02 2.886e+02 3.399e+02 6.002e+02, threshold=5.772e+02, percent-clipped=0.0 +2023-02-07 02:08:46,930 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 02:08:49,843 INFO [train.py:901] (2/4) Epoch 21, batch 3050, loss[loss=0.211, simple_loss=0.2927, pruned_loss=0.06466, over 8257.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2886, pruned_loss=0.06249, over 1612808.52 frames. ], batch size: 24, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:08:59,357 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164722.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:09:08,316 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.4455, 5.5674, 4.8211, 2.3653, 4.8654, 5.2364, 5.2479, 5.0528], + device='cuda:2'), covar=tensor([0.0634, 0.0396, 0.0924, 0.4436, 0.0727, 0.0785, 0.0913, 0.0670], + device='cuda:2'), in_proj_covar=tensor([0.0512, 0.0428, 0.0428, 0.0529, 0.0420, 0.0432, 0.0411, 0.0377], + 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') +2023-02-07 02:09:25,484 INFO [train.py:901] (2/4) Epoch 21, batch 3100, loss[loss=0.1891, simple_loss=0.2688, pruned_loss=0.05469, over 7937.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2898, pruned_loss=0.0629, over 1615862.65 frames. ], batch size: 20, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:09:29,031 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164764.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:09:38,555 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4953, 2.5246, 1.7804, 2.2295, 2.1741, 1.4701, 2.0386, 2.0698], + device='cuda:2'), covar=tensor([0.1746, 0.0476, 0.1315, 0.0636, 0.0796, 0.1685, 0.1123, 0.1040], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0234, 0.0333, 0.0306, 0.0296, 0.0331, 0.0341, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:09:39,119 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9178, 6.1047, 5.3509, 2.5021, 5.4496, 5.7859, 5.5524, 5.4800], + device='cuda:2'), covar=tensor([0.0498, 0.0345, 0.0802, 0.4497, 0.0637, 0.0623, 0.1030, 0.0504], + device='cuda:2'), in_proj_covar=tensor([0.0509, 0.0426, 0.0426, 0.0526, 0.0417, 0.0430, 0.0409, 0.0376], + 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') +2023-02-07 02:09:43,637 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.375e+02 2.980e+02 3.572e+02 8.800e+02, threshold=5.960e+02, percent-clipped=5.0 +2023-02-07 02:09:59,117 INFO [train.py:901] (2/4) Epoch 21, batch 3150, loss[loss=0.2059, simple_loss=0.2754, pruned_loss=0.06816, over 7539.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2878, pruned_loss=0.06223, over 1610069.52 frames. ], batch size: 18, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:10:08,646 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164823.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:10:12,638 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8793, 1.6949, 3.0202, 1.4897, 2.3701, 3.2932, 3.3605, 2.8519], + device='cuda:2'), covar=tensor([0.1129, 0.1560, 0.0397, 0.2126, 0.0958, 0.0265, 0.0631, 0.0569], + device='cuda:2'), in_proj_covar=tensor([0.0290, 0.0320, 0.0288, 0.0314, 0.0304, 0.0261, 0.0410, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:10:19,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:10:26,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164848.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:10:34,948 INFO [train.py:901] (2/4) Epoch 21, batch 3200, loss[loss=0.2853, simple_loss=0.3359, pruned_loss=0.1173, over 6867.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2882, pruned_loss=0.06242, over 1608866.98 frames. ], batch size: 71, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:10:54,110 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.324e+02 2.650e+02 3.384e+02 7.808e+02, threshold=5.299e+02, percent-clipped=1.0 +2023-02-07 02:10:55,687 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164888.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:09,477 INFO [train.py:901] (2/4) Epoch 21, batch 3250, loss[loss=0.184, simple_loss=0.2791, pruned_loss=0.04443, over 8466.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2886, pruned_loss=0.06231, over 1608903.78 frames. ], batch size: 25, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:11:12,441 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164913.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:23,959 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164930.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:30,775 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164940.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:44,788 INFO [train.py:901] (2/4) Epoch 21, batch 3300, loss[loss=0.17, simple_loss=0.2648, pruned_loss=0.03759, over 8030.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2884, pruned_loss=0.06228, over 1611947.84 frames. ], batch size: 22, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:12:05,208 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.295e+02 2.742e+02 3.217e+02 7.829e+02, threshold=5.483e+02, percent-clipped=4.0 +2023-02-07 02:12:20,627 INFO [train.py:901] (2/4) Epoch 21, batch 3350, loss[loss=0.2237, simple_loss=0.3124, pruned_loss=0.06756, over 8245.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2888, pruned_loss=0.0624, over 1611849.41 frames. ], batch size: 24, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:12:28,063 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165020.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:12:45,361 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165045.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:12:52,348 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165055.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:12:54,954 INFO [train.py:901] (2/4) Epoch 21, batch 3400, loss[loss=0.2164, simple_loss=0.3017, pruned_loss=0.06549, over 8393.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2877, pruned_loss=0.06187, over 1613136.47 frames. ], batch size: 49, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:12:59,414 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.37 vs. limit=5.0 +2023-02-07 02:13:15,612 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.386e+02 2.300e+02 2.821e+02 3.884e+02 1.046e+03, threshold=5.643e+02, percent-clipped=8.0 +2023-02-07 02:13:20,665 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165093.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:31,317 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:31,892 INFO [train.py:901] (2/4) Epoch 21, batch 3450, loss[loss=0.2462, simple_loss=0.3333, pruned_loss=0.07953, over 8565.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2884, pruned_loss=0.06224, over 1617918.97 frames. ], batch size: 39, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:13:38,108 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:49,305 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165135.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:53,301 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:14:05,181 INFO [train.py:901] (2/4) Epoch 21, batch 3500, loss[loss=0.2149, simple_loss=0.2931, pruned_loss=0.06836, over 7805.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2901, pruned_loss=0.06354, over 1617669.23 frames. ], batch size: 20, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:14:07,492 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.22 vs. limit=5.0 +2023-02-07 02:14:10,626 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 02:14:19,675 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6142, 1.9552, 2.1163, 1.3217, 2.1066, 1.4208, 0.6992, 1.8431], + device='cuda:2'), covar=tensor([0.0868, 0.0410, 0.0326, 0.0725, 0.0580, 0.1051, 0.1073, 0.0395], + device='cuda:2'), in_proj_covar=tensor([0.0449, 0.0384, 0.0339, 0.0440, 0.0372, 0.0528, 0.0386, 0.0414], + device='cuda:2'), out_proj_covar=tensor([1.2068e-04, 1.0065e-04, 8.9298e-05, 1.1636e-04, 9.8080e-05, 1.4961e-04, + 1.0436e-04, 1.1000e-04], device='cuda:2') +2023-02-07 02:14:24,648 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.546e+02 2.436e+02 2.745e+02 3.695e+02 8.606e+02, threshold=5.490e+02, percent-clipped=3.0 +2023-02-07 02:14:31,313 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-07 02:14:41,284 INFO [train.py:901] (2/4) Epoch 21, batch 3550, loss[loss=0.2275, simple_loss=0.2866, pruned_loss=0.08414, over 7927.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2897, pruned_loss=0.06269, over 1620539.61 frames. ], batch size: 20, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:14:51,650 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165223.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:15:15,601 INFO [train.py:901] (2/4) Epoch 21, batch 3600, loss[loss=0.1928, simple_loss=0.2743, pruned_loss=0.05569, over 7968.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2896, pruned_loss=0.06293, over 1620866.52 frames. ], batch size: 21, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:15:34,163 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.347e+02 2.942e+02 3.699e+02 7.087e+02, threshold=5.884e+02, percent-clipped=2.0 +2023-02-07 02:15:36,933 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1451, 1.1972, 1.4462, 1.1088, 0.7146, 1.2257, 1.0599, 0.8531], + device='cuda:2'), covar=tensor([0.0628, 0.1191, 0.1494, 0.1373, 0.0583, 0.1386, 0.0717, 0.0714], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0158, 0.0099, 0.0161, 0.0112, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 02:15:44,469 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:15:51,140 INFO [train.py:901] (2/4) Epoch 21, batch 3650, loss[loss=0.2314, simple_loss=0.3116, pruned_loss=0.07557, over 7935.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2872, pruned_loss=0.06186, over 1617721.86 frames. ], batch size: 20, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:15:52,698 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:15:53,563 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 02:15:56,696 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7998, 2.2126, 1.6757, 2.7602, 1.3046, 1.5499, 1.9047, 2.1448], + device='cuda:2'), covar=tensor([0.0791, 0.0727, 0.1007, 0.0372, 0.1194, 0.1349, 0.1005, 0.0742], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0212, 0.0205, 0.0246, 0.0250, 0.0208], + 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') +2023-02-07 02:16:03,316 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165326.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:08,100 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8156, 2.4387, 4.3779, 1.6796, 3.3553, 2.3999, 2.0467, 3.1950], + device='cuda:2'), covar=tensor([0.1945, 0.2528, 0.0823, 0.4390, 0.1554, 0.3118, 0.2191, 0.2229], + device='cuda:2'), in_proj_covar=tensor([0.0523, 0.0596, 0.0557, 0.0639, 0.0645, 0.0592, 0.0535, 0.0634], + 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') +2023-02-07 02:16:10,826 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165336.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:16,761 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 02:16:25,986 INFO [train.py:901] (2/4) Epoch 21, batch 3700, loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.0882, over 8327.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2879, pruned_loss=0.06227, over 1614976.29 frames. ], batch size: 26, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:16:44,014 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.402e+02 2.885e+02 3.854e+02 8.848e+02, threshold=5.771e+02, percent-clipped=5.0 +2023-02-07 02:16:47,619 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165391.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:59,579 INFO [train.py:901] (2/4) Epoch 21, batch 3750, loss[loss=0.2002, simple_loss=0.278, pruned_loss=0.06117, over 8091.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2886, pruned_loss=0.06277, over 1609716.74 frames. ], batch size: 21, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:17:04,500 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165416.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:17:16,250 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3612, 2.8475, 2.3420, 3.9498, 1.7047, 2.2747, 2.3077, 2.8679], + device='cuda:2'), covar=tensor([0.0659, 0.0709, 0.0834, 0.0217, 0.1092, 0.1092, 0.1041, 0.0797], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0212, 0.0205, 0.0246, 0.0251, 0.0208], + 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') +2023-02-07 02:17:36,650 INFO [train.py:901] (2/4) Epoch 21, batch 3800, loss[loss=0.1875, simple_loss=0.2664, pruned_loss=0.0543, over 7552.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2886, pruned_loss=0.06313, over 1607042.61 frames. ], batch size: 18, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:17:49,673 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9530, 1.5577, 1.7311, 1.3794, 0.9129, 1.4752, 1.7942, 1.6294], + device='cuda:2'), covar=tensor([0.0544, 0.1242, 0.1612, 0.1477, 0.0612, 0.1525, 0.0688, 0.0637], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0160, 0.0099, 0.0162, 0.0113, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 02:17:50,402 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:17:54,319 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165485.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:17:54,906 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.401e+02 2.925e+02 3.673e+02 6.793e+02, threshold=5.851e+02, percent-clipped=2.0 +2023-02-07 02:17:59,010 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9704, 1.5263, 4.4564, 2.0112, 2.5793, 5.1125, 5.1629, 4.3895], + device='cuda:2'), covar=tensor([0.1285, 0.1911, 0.0271, 0.1922, 0.1094, 0.0158, 0.0429, 0.0578], + device='cuda:2'), in_proj_covar=tensor([0.0292, 0.0321, 0.0287, 0.0315, 0.0308, 0.0263, 0.0412, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:18:07,159 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165504.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:18:10,359 INFO [train.py:901] (2/4) Epoch 21, batch 3850, loss[loss=0.2028, simple_loss=0.2892, pruned_loss=0.0582, over 8458.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2884, pruned_loss=0.06268, over 1607881.26 frames. ], batch size: 25, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:18:18,565 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 02:18:31,966 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1524, 3.8856, 2.7277, 3.0208, 3.1919, 2.4157, 3.0900, 3.1118], + device='cuda:2'), covar=tensor([0.1695, 0.0332, 0.1051, 0.0814, 0.0678, 0.1351, 0.0985, 0.1085], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0235, 0.0334, 0.0310, 0.0298, 0.0336, 0.0345, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:18:46,107 INFO [train.py:901] (2/4) Epoch 21, batch 3900, loss[loss=0.1945, simple_loss=0.2791, pruned_loss=0.05492, over 8547.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2886, pruned_loss=0.06293, over 1608932.07 frames. ], batch size: 31, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:18:53,053 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165569.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:19:05,131 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.182e+02 2.809e+02 3.459e+02 6.713e+02, threshold=5.619e+02, percent-clipped=4.0 +2023-02-07 02:19:14,939 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165600.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:19:20,726 INFO [train.py:901] (2/4) Epoch 21, batch 3950, loss[loss=0.2504, simple_loss=0.3244, pruned_loss=0.08822, over 8133.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2891, pruned_loss=0.06303, over 1611957.61 frames. ], batch size: 22, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:19:55,186 INFO [train.py:901] (2/4) Epoch 21, batch 4000, loss[loss=0.2243, simple_loss=0.3206, pruned_loss=0.06397, over 8248.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2901, pruned_loss=0.06344, over 1611431.37 frames. ], batch size: 24, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:20:15,739 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.370e+02 2.936e+02 3.785e+02 6.204e+02, threshold=5.872e+02, percent-clipped=2.0 +2023-02-07 02:20:31,248 INFO [train.py:901] (2/4) Epoch 21, batch 4050, loss[loss=0.1608, simple_loss=0.2515, pruned_loss=0.03505, over 7646.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2897, pruned_loss=0.06297, over 1613653.73 frames. ], batch size: 19, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:20:47,659 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 02:21:03,807 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 02:21:04,761 INFO [train.py:901] (2/4) Epoch 21, batch 4100, loss[loss=0.171, simple_loss=0.2638, pruned_loss=0.03916, over 8291.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2881, pruned_loss=0.06221, over 1607704.90 frames. ], batch size: 23, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:21:11,130 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:21:24,843 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.510e+02 3.105e+02 3.860e+02 6.931e+02, threshold=6.209e+02, percent-clipped=6.0 +2023-02-07 02:21:41,898 INFO [train.py:901] (2/4) Epoch 21, batch 4150, loss[loss=0.2402, simple_loss=0.3145, pruned_loss=0.08299, over 8765.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2891, pruned_loss=0.06287, over 1612290.97 frames. ], batch size: 30, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:21:48,855 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6716, 1.5887, 2.2857, 1.5742, 1.2362, 2.1852, 0.4557, 1.3436], + device='cuda:2'), covar=tensor([0.1880, 0.1498, 0.0339, 0.1138, 0.2903, 0.0392, 0.2283, 0.1395], + device='cuda:2'), in_proj_covar=tensor([0.0188, 0.0195, 0.0128, 0.0221, 0.0272, 0.0135, 0.0171, 0.0191], + 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') +2023-02-07 02:21:50,124 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165821.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:22:13,881 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165856.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:22:15,094 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 02:22:15,767 INFO [train.py:901] (2/4) Epoch 21, batch 4200, loss[loss=0.1688, simple_loss=0.2483, pruned_loss=0.04466, over 7788.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2883, pruned_loss=0.06197, over 1615829.42 frames. ], batch size: 19, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:22:30,503 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165881.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:22:33,700 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.320e+02 2.907e+02 3.705e+02 7.802e+02, threshold=5.814e+02, percent-clipped=2.0 +2023-02-07 02:22:37,068 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 02:22:50,803 INFO [train.py:901] (2/4) Epoch 21, batch 4250, loss[loss=0.2027, simple_loss=0.2919, pruned_loss=0.05671, over 8466.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2882, pruned_loss=0.06187, over 1617822.44 frames. ], batch size: 25, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:22:53,703 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165913.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:23:14,527 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5363, 2.0140, 3.1892, 1.4163, 2.3595, 1.9050, 1.7349, 2.3655], + device='cuda:2'), covar=tensor([0.1927, 0.2453, 0.0864, 0.4409, 0.1906, 0.3295, 0.2255, 0.2343], + device='cuda:2'), in_proj_covar=tensor([0.0523, 0.0597, 0.0555, 0.0639, 0.0645, 0.0594, 0.0536, 0.0635], + 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') +2023-02-07 02:23:26,544 INFO [train.py:901] (2/4) Epoch 21, batch 4300, loss[loss=0.2391, simple_loss=0.329, pruned_loss=0.07463, over 8520.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2897, pruned_loss=0.06331, over 1617572.03 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:23:44,458 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.284e+02 2.728e+02 3.396e+02 7.954e+02, threshold=5.457e+02, percent-clipped=4.0 +2023-02-07 02:23:50,849 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5074, 1.9955, 3.1896, 1.3762, 2.3277, 1.9584, 1.7372, 2.3726], + device='cuda:2'), covar=tensor([0.1870, 0.2570, 0.0824, 0.4524, 0.1935, 0.3348, 0.2279, 0.2386], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0595, 0.0551, 0.0636, 0.0642, 0.0591, 0.0534, 0.0632], + 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') +2023-02-07 02:24:01,652 INFO [train.py:901] (2/4) Epoch 21, batch 4350, loss[loss=0.2126, simple_loss=0.2967, pruned_loss=0.06425, over 8328.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2897, pruned_loss=0.06364, over 1614560.40 frames. ], batch size: 25, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:24:11,727 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 02:24:15,244 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166028.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:24:16,542 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6916, 2.0714, 3.2315, 1.5139, 2.4615, 2.1491, 1.7986, 2.5012], + device='cuda:2'), covar=tensor([0.1810, 0.2465, 0.0779, 0.4313, 0.1777, 0.3015, 0.2245, 0.2197], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0593, 0.0551, 0.0635, 0.0641, 0.0590, 0.0533, 0.0631], + 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') +2023-02-07 02:24:36,828 INFO [train.py:901] (2/4) Epoch 21, batch 4400, loss[loss=0.2008, simple_loss=0.2856, pruned_loss=0.05796, over 7965.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2908, pruned_loss=0.06421, over 1614373.34 frames. ], batch size: 21, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:24:45,737 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166072.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:24:54,405 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 02:24:55,048 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.482e+02 3.095e+02 3.863e+02 7.424e+02, threshold=6.191e+02, percent-clipped=10.0 +2023-02-07 02:25:10,664 INFO [train.py:901] (2/4) Epoch 21, batch 4450, loss[loss=0.2024, simple_loss=0.2816, pruned_loss=0.06158, over 8036.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.0637, over 1615960.85 frames. ], batch size: 22, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:25:12,803 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166112.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:25:22,694 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5137, 2.7434, 3.2815, 1.8714, 3.4430, 2.2208, 1.6321, 2.2788], + device='cuda:2'), covar=tensor([0.0727, 0.0363, 0.0277, 0.0728, 0.0394, 0.0733, 0.0964, 0.0521], + device='cuda:2'), in_proj_covar=tensor([0.0454, 0.0387, 0.0341, 0.0441, 0.0374, 0.0533, 0.0391, 0.0415], + device='cuda:2'), out_proj_covar=tensor([1.2196e-04, 1.0144e-04, 8.9743e-05, 1.1657e-04, 9.8624e-05, 1.5080e-04, + 1.0559e-04, 1.1017e-04], device='cuda:2') +2023-02-07 02:25:27,717 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 02:25:45,423 INFO [train.py:901] (2/4) Epoch 21, batch 4500, loss[loss=0.2078, simple_loss=0.2969, pruned_loss=0.05935, over 8351.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2894, pruned_loss=0.06357, over 1611906.58 frames. ], batch size: 24, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:25:50,219 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 02:25:50,295 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:25:50,362 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:26:05,008 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.688e+02 3.191e+02 4.415e+02 1.086e+03, threshold=6.382e+02, percent-clipped=9.0 +2023-02-07 02:26:20,603 INFO [train.py:901] (2/4) Epoch 21, batch 4550, loss[loss=0.2042, simple_loss=0.297, pruned_loss=0.05572, over 8527.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2896, pruned_loss=0.06341, over 1615124.74 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:26:27,579 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9417, 1.6750, 2.0951, 1.8517, 2.0383, 1.9481, 1.7349, 0.8641], + device='cuda:2'), covar=tensor([0.5185, 0.4585, 0.1776, 0.3234, 0.2272, 0.2826, 0.1906, 0.4744], + device='cuda:2'), in_proj_covar=tensor([0.0935, 0.0972, 0.0794, 0.0932, 0.0989, 0.0885, 0.0741, 0.0820], + 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') +2023-02-07 02:26:33,046 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166227.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:26:54,696 INFO [train.py:901] (2/4) Epoch 21, batch 4600, loss[loss=0.2301, simple_loss=0.298, pruned_loss=0.08108, over 8250.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2898, pruned_loss=0.06365, over 1612895.11 frames. ], batch size: 22, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:27:10,560 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166280.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:27:14,133 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166284.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:27:15,243 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.551e+02 3.310e+02 4.080e+02 7.820e+02, threshold=6.621e+02, percent-clipped=4.0 +2023-02-07 02:27:30,339 INFO [train.py:901] (2/4) Epoch 21, batch 4650, loss[loss=0.1883, simple_loss=0.2629, pruned_loss=0.05687, over 7796.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2895, pruned_loss=0.06349, over 1614083.46 frames. ], batch size: 19, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:27:30,537 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:27:41,203 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166325.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:28:03,638 INFO [train.py:901] (2/4) Epoch 21, batch 4700, loss[loss=0.2176, simple_loss=0.3073, pruned_loss=0.06395, over 8574.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2897, pruned_loss=0.0633, over 1614024.10 frames. ], batch size: 39, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:28:23,890 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.373e+02 2.801e+02 3.877e+02 1.145e+03, threshold=5.601e+02, percent-clipped=4.0 +2023-02-07 02:28:40,127 INFO [train.py:901] (2/4) Epoch 21, batch 4750, loss[loss=0.1686, simple_loss=0.252, pruned_loss=0.04262, over 7534.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.291, pruned_loss=0.06385, over 1613820.00 frames. ], batch size: 18, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:28:45,011 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166416.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:28:52,945 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 02:28:55,070 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 02:29:14,172 INFO [train.py:901] (2/4) Epoch 21, batch 4800, loss[loss=0.2381, simple_loss=0.2965, pruned_loss=0.08987, over 7803.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2905, pruned_loss=0.0638, over 1610755.77 frames. ], batch size: 19, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:29:30,702 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166483.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:29:33,189 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.480e+02 2.407e+02 2.819e+02 3.849e+02 8.316e+02, threshold=5.639e+02, percent-clipped=5.0 +2023-02-07 02:29:39,662 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.82 vs. limit=5.0 +2023-02-07 02:29:42,736 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:29:43,963 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 02:29:48,838 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166508.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:29:49,313 INFO [train.py:901] (2/4) Epoch 21, batch 4850, loss[loss=0.1768, simple_loss=0.2499, pruned_loss=0.05186, over 7697.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.29, pruned_loss=0.0634, over 1605861.27 frames. ], batch size: 18, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:29:49,391 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166509.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:03,258 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166527.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:05,913 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166531.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:30:09,300 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166536.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:17,939 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166549.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:21,057 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.32 vs. limit=5.0 +2023-02-07 02:30:24,634 INFO [train.py:901] (2/4) Epoch 21, batch 4900, loss[loss=0.19, simple_loss=0.2784, pruned_loss=0.0508, over 8026.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2894, pruned_loss=0.06327, over 1604856.30 frames. ], batch size: 22, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:30:26,224 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166561.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:43,074 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.429e+02 3.059e+02 4.014e+02 7.599e+02, threshold=6.119e+02, percent-clipped=4.0 +2023-02-07 02:30:54,286 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1826, 1.0668, 1.2971, 1.0798, 1.0019, 1.3470, 0.0801, 0.8796], + device='cuda:2'), covar=tensor([0.1640, 0.1444, 0.0514, 0.0810, 0.2780, 0.0543, 0.2235, 0.1337], + device='cuda:2'), in_proj_covar=tensor([0.0190, 0.0196, 0.0128, 0.0223, 0.0272, 0.0136, 0.0172, 0.0192], + 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') +2023-02-07 02:30:58,617 INFO [train.py:901] (2/4) Epoch 21, batch 4950, loss[loss=0.1832, simple_loss=0.2645, pruned_loss=0.05094, over 7231.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2891, pruned_loss=0.06308, over 1607443.72 frames. ], batch size: 16, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:31:09,544 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166624.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:31:34,515 INFO [train.py:901] (2/4) Epoch 21, batch 5000, loss[loss=0.2276, simple_loss=0.2957, pruned_loss=0.07975, over 7807.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2896, pruned_loss=0.0632, over 1609091.76 frames. ], batch size: 20, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:31:41,052 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166669.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:31:52,855 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.262e+02 2.770e+02 3.475e+02 7.586e+02, threshold=5.540e+02, percent-clipped=2.0 +2023-02-07 02:32:07,630 INFO [train.py:901] (2/4) Epoch 21, batch 5050, loss[loss=0.2099, simple_loss=0.2947, pruned_loss=0.06255, over 8245.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2896, pruned_loss=0.0636, over 1606770.23 frames. ], batch size: 24, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:32:23,042 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 02:32:38,359 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166753.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:32:42,937 INFO [train.py:901] (2/4) Epoch 21, batch 5100, loss[loss=0.1937, simple_loss=0.2788, pruned_loss=0.05431, over 8480.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2895, pruned_loss=0.06364, over 1607041.83 frames. ], batch size: 29, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:32:59,530 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166782.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:33:00,920 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166784.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:33:02,720 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.499e+02 3.045e+02 3.729e+02 1.083e+03, threshold=6.090e+02, percent-clipped=5.0 +2023-02-07 02:33:02,983 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166787.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 02:33:17,643 INFO [train.py:901] (2/4) Epoch 21, batch 5150, loss[loss=0.2006, simple_loss=0.2802, pruned_loss=0.06047, over 8148.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2899, pruned_loss=0.06371, over 1614714.20 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:33:19,983 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166812.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 02:33:35,604 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0794, 2.3065, 1.8774, 2.7978, 1.3497, 1.6818, 1.8829, 2.3235], + device='cuda:2'), covar=tensor([0.0705, 0.0694, 0.0853, 0.0339, 0.1078, 0.1216, 0.0904, 0.0714], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0194, 0.0241, 0.0210, 0.0203, 0.0241, 0.0249, 0.0206], + 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') +2023-02-07 02:33:40,946 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:33:52,912 INFO [train.py:901] (2/4) Epoch 21, batch 5200, loss[loss=0.2063, simple_loss=0.2842, pruned_loss=0.0642, over 7926.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2887, pruned_loss=0.06322, over 1615710.29 frames. ], batch size: 20, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:33:54,692 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.33 vs. limit=5.0 +2023-02-07 02:34:01,085 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:08,928 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166880.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:13,412 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.573e+02 2.417e+02 2.893e+02 3.464e+02 9.071e+02, threshold=5.787e+02, percent-clipped=3.0 +2023-02-07 02:34:17,473 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166893.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:22,880 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 02:34:25,859 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166905.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:28,354 INFO [train.py:901] (2/4) Epoch 21, batch 5250, loss[loss=0.2085, simple_loss=0.2936, pruned_loss=0.06165, over 7929.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2875, pruned_loss=0.06262, over 1616390.79 frames. ], batch size: 20, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:34:53,833 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166947.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:00,990 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166958.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:01,468 INFO [train.py:901] (2/4) Epoch 21, batch 5300, loss[loss=0.1934, simple_loss=0.2735, pruned_loss=0.05663, over 8138.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2889, pruned_loss=0.06371, over 1614933.23 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:35:21,124 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166986.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:21,579 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 2.349e+02 2.996e+02 3.802e+02 6.845e+02, threshold=5.992e+02, percent-clipped=3.0 +2023-02-07 02:35:22,370 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2656, 3.6358, 2.4378, 2.9715, 2.8141, 2.2571, 2.9631, 3.1893], + device='cuda:2'), covar=tensor([0.1637, 0.0647, 0.1141, 0.0736, 0.0855, 0.1297, 0.1095, 0.1054], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0235, 0.0335, 0.0308, 0.0298, 0.0334, 0.0344, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:35:37,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167008.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:37,979 INFO [train.py:901] (2/4) Epoch 21, batch 5350, loss[loss=0.2085, simple_loss=0.3015, pruned_loss=0.05772, over 8484.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2891, pruned_loss=0.06392, over 1612103.71 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:35:58,871 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167040.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:12,081 INFO [train.py:901] (2/4) Epoch 21, batch 5400, loss[loss=0.184, simple_loss=0.2625, pruned_loss=0.0527, over 7641.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2886, pruned_loss=0.06366, over 1609037.17 frames. ], batch size: 19, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:36:16,473 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167065.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:32,172 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.459e+02 3.013e+02 3.547e+02 6.118e+02, threshold=6.026e+02, percent-clipped=1.0 +2023-02-07 02:36:39,934 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167097.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:47,942 INFO [train.py:901] (2/4) Epoch 21, batch 5450, loss[loss=0.2651, simple_loss=0.343, pruned_loss=0.09357, over 8244.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2907, pruned_loss=0.0644, over 1615547.70 frames. ], batch size: 24, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:36:55,797 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:58,552 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167122.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:37:01,213 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167126.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:37:12,854 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 02:37:19,304 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8747, 2.0167, 5.9898, 2.3388, 5.3949, 5.0756, 5.5589, 5.4530], + device='cuda:2'), covar=tensor([0.0465, 0.4365, 0.0330, 0.3876, 0.0981, 0.0845, 0.0511, 0.0498], + device='cuda:2'), in_proj_covar=tensor([0.0630, 0.0643, 0.0695, 0.0629, 0.0707, 0.0605, 0.0605, 0.0677], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 02:37:24,066 INFO [train.py:901] (2/4) Epoch 21, batch 5500, loss[loss=0.1917, simple_loss=0.2817, pruned_loss=0.05085, over 8196.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.29, pruned_loss=0.06385, over 1616310.93 frames. ], batch size: 23, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:37:43,662 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.538e+02 3.099e+02 3.967e+02 8.838e+02, threshold=6.197e+02, percent-clipped=3.0 +2023-02-07 02:37:58,454 INFO [train.py:901] (2/4) Epoch 21, batch 5550, loss[loss=0.2112, simple_loss=0.2847, pruned_loss=0.06887, over 8034.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2903, pruned_loss=0.06391, over 1616721.86 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:38:01,331 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167212.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:02,595 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:21,135 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:22,454 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167241.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:23,134 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167242.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:34,482 INFO [train.py:901] (2/4) Epoch 21, batch 5600, loss[loss=0.1836, simple_loss=0.2521, pruned_loss=0.05759, over 7661.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2898, pruned_loss=0.06382, over 1612998.74 frames. ], batch size: 19, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:38:34,957 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.61 vs. limit=5.0 +2023-02-07 02:38:36,019 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7033, 1.3005, 4.8691, 1.8519, 4.3328, 4.0786, 4.4105, 4.2757], + device='cuda:2'), covar=tensor([0.0500, 0.4833, 0.0442, 0.3928, 0.0948, 0.0821, 0.0542, 0.0573], + device='cuda:2'), in_proj_covar=tensor([0.0629, 0.0637, 0.0691, 0.0626, 0.0701, 0.0602, 0.0601, 0.0672], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 02:38:38,122 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167264.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:40,084 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167267.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:54,594 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.492e+02 3.097e+02 3.838e+02 7.086e+02, threshold=6.194e+02, percent-clipped=1.0 +2023-02-07 02:38:54,810 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167289.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:56,076 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167291.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:39:08,035 INFO [train.py:901] (2/4) Epoch 21, batch 5650, loss[loss=0.2166, simple_loss=0.2986, pruned_loss=0.06735, over 8504.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2902, pruned_loss=0.06382, over 1614365.08 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:39:18,773 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 02:39:21,791 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 +2023-02-07 02:39:44,497 INFO [train.py:901] (2/4) Epoch 21, batch 5700, loss[loss=0.2006, simple_loss=0.2827, pruned_loss=0.05921, over 7963.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2895, pruned_loss=0.06322, over 1615620.55 frames. ], batch size: 21, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:39:48,140 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5336, 1.9231, 2.1403, 1.2938, 2.1864, 1.4334, 0.5832, 1.7232], + device='cuda:2'), covar=tensor([0.0623, 0.0352, 0.0242, 0.0511, 0.0335, 0.0857, 0.0786, 0.0298], + device='cuda:2'), in_proj_covar=tensor([0.0448, 0.0385, 0.0337, 0.0439, 0.0372, 0.0528, 0.0385, 0.0411], + device='cuda:2'), out_proj_covar=tensor([1.2043e-04, 1.0098e-04, 8.8761e-05, 1.1606e-04, 9.8009e-05, 1.4936e-04, + 1.0401e-04, 1.0882e-04], device='cuda:2') +2023-02-07 02:40:04,669 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.585e+02 3.206e+02 3.925e+02 8.506e+02, threshold=6.412e+02, percent-clipped=6.0 +2023-02-07 02:40:16,551 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:40:18,411 INFO [train.py:901] (2/4) Epoch 21, batch 5750, loss[loss=0.2904, simple_loss=0.3477, pruned_loss=0.1166, over 7130.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2898, pruned_loss=0.06348, over 1614148.77 frames. ], batch size: 72, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:40:24,269 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 02:40:47,696 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167450.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:40:53,628 INFO [train.py:901] (2/4) Epoch 21, batch 5800, loss[loss=0.152, simple_loss=0.2396, pruned_loss=0.03223, over 7958.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2888, pruned_loss=0.06263, over 1616336.20 frames. ], batch size: 21, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:40:55,796 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:40:59,179 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167466.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:00,730 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167468.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:15,048 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.466e+02 2.953e+02 3.603e+02 7.254e+02, threshold=5.907e+02, percent-clipped=1.0 +2023-02-07 02:41:17,999 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167493.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:20,774 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167497.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:28,633 INFO [train.py:901] (2/4) Epoch 21, batch 5850, loss[loss=0.1935, simple_loss=0.2695, pruned_loss=0.05873, over 7662.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2891, pruned_loss=0.06248, over 1615945.97 frames. ], batch size: 19, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:41:36,514 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3671, 1.5538, 2.1760, 1.3283, 1.3801, 1.6366, 1.4504, 1.4422], + device='cuda:2'), covar=tensor([0.1941, 0.2474, 0.0851, 0.4216, 0.1989, 0.3336, 0.2311, 0.2260], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0593, 0.0551, 0.0634, 0.0637, 0.0587, 0.0528, 0.0626], + 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') +2023-02-07 02:41:37,829 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4972, 2.6528, 1.8999, 2.2901, 2.0902, 1.7211, 2.0220, 2.2244], + device='cuda:2'), covar=tensor([0.1612, 0.0388, 0.1262, 0.0711, 0.0789, 0.1482, 0.1086, 0.1001], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0234, 0.0333, 0.0306, 0.0298, 0.0333, 0.0343, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:41:37,835 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167522.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:42:03,758 INFO [train.py:901] (2/4) Epoch 21, batch 5900, loss[loss=0.2114, simple_loss=0.3014, pruned_loss=0.06064, over 8275.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2887, pruned_loss=0.06229, over 1618829.13 frames. ], batch size: 23, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:42:16,868 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167577.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:42:19,704 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167581.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:42:22,427 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7464, 2.0025, 2.1452, 1.5071, 2.1631, 1.4802, 0.7336, 1.8720], + device='cuda:2'), covar=tensor([0.0754, 0.0385, 0.0335, 0.0675, 0.0495, 0.1052, 0.0938, 0.0370], + device='cuda:2'), in_proj_covar=tensor([0.0451, 0.0387, 0.0339, 0.0441, 0.0374, 0.0531, 0.0390, 0.0415], + device='cuda:2'), out_proj_covar=tensor([1.2118e-04, 1.0146e-04, 8.9180e-05, 1.1651e-04, 9.8491e-05, 1.5011e-04, + 1.0522e-04, 1.0989e-04], device='cuda:2') +2023-02-07 02:42:25,698 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.257e+02 2.859e+02 3.440e+02 7.059e+02, threshold=5.718e+02, percent-clipped=2.0 +2023-02-07 02:42:40,372 INFO [train.py:901] (2/4) Epoch 21, batch 5950, loss[loss=0.2234, simple_loss=0.3065, pruned_loss=0.07018, over 8107.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2884, pruned_loss=0.0625, over 1617908.21 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 4.0 +2023-02-07 02:43:14,051 INFO [train.py:901] (2/4) Epoch 21, batch 6000, loss[loss=0.2063, simple_loss=0.2802, pruned_loss=0.06614, over 7526.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2882, pruned_loss=0.0621, over 1617297.91 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:43:14,052 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 02:43:26,396 INFO [train.py:935] (2/4) Epoch 21, validation: loss=0.174, simple_loss=0.2741, pruned_loss=0.03692, over 944034.00 frames. +2023-02-07 02:43:26,397 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 02:43:28,712 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:43:45,658 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167687.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:43:47,406 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.382e+02 2.918e+02 3.609e+02 5.587e+02, threshold=5.837e+02, percent-clipped=0.0 +2023-02-07 02:44:01,967 INFO [train.py:901] (2/4) Epoch 21, batch 6050, loss[loss=0.183, simple_loss=0.2745, pruned_loss=0.04578, over 8088.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2878, pruned_loss=0.06204, over 1615280.69 frames. ], batch size: 21, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:44:06,718 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-07 02:44:22,063 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167737.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:44:38,027 INFO [train.py:901] (2/4) Epoch 21, batch 6100, loss[loss=0.2363, simple_loss=0.3297, pruned_loss=0.0715, over 8345.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2882, pruned_loss=0.06273, over 1608358.68 frames. ], batch size: 26, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:44:56,065 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167785.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:44:57,193 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 02:44:58,541 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.390e+02 3.045e+02 3.849e+02 6.701e+02, threshold=6.089e+02, percent-clipped=2.0 +2023-02-07 02:45:02,109 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167794.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:13,115 INFO [train.py:901] (2/4) Epoch 21, batch 6150, loss[loss=0.2022, simple_loss=0.28, pruned_loss=0.06219, over 8090.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2884, pruned_loss=0.06309, over 1608188.17 frames. ], batch size: 21, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:45:30,426 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:33,105 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:44,766 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-07 02:45:48,361 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167858.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:48,817 INFO [train.py:901] (2/4) Epoch 21, batch 6200, loss[loss=0.242, simple_loss=0.3204, pruned_loss=0.08183, over 8338.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2888, pruned_loss=0.06324, over 1608513.77 frames. ], batch size: 26, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:45:51,065 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167862.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:46:09,469 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.266e+02 2.776e+02 3.727e+02 8.167e+02, threshold=5.552e+02, percent-clipped=4.0 +2023-02-07 02:46:23,394 INFO [train.py:901] (2/4) Epoch 21, batch 6250, loss[loss=0.1818, simple_loss=0.2797, pruned_loss=0.04199, over 8351.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2881, pruned_loss=0.06295, over 1609414.58 frames. ], batch size: 24, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:46:23,595 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:46:58,905 INFO [train.py:901] (2/4) Epoch 21, batch 6300, loss[loss=0.1664, simple_loss=0.2398, pruned_loss=0.04651, over 7541.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2875, pruned_loss=0.06253, over 1610009.08 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:47:06,584 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2023-02-07 02:47:20,718 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.475e+02 2.869e+02 3.545e+02 9.430e+02, threshold=5.737e+02, percent-clipped=7.0 +2023-02-07 02:47:22,361 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8669, 2.4124, 4.1086, 1.6089, 3.1210, 2.3233, 1.9038, 2.9446], + device='cuda:2'), covar=tensor([0.1877, 0.2552, 0.0789, 0.4542, 0.1808, 0.3125, 0.2288, 0.2407], + device='cuda:2'), in_proj_covar=tensor([0.0522, 0.0597, 0.0553, 0.0639, 0.0642, 0.0588, 0.0530, 0.0630], + 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') +2023-02-07 02:47:35,234 INFO [train.py:901] (2/4) Epoch 21, batch 6350, loss[loss=0.1848, simple_loss=0.2756, pruned_loss=0.047, over 8026.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2879, pruned_loss=0.06255, over 1608910.48 frames. ], batch size: 22, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:48:02,053 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9380, 2.0108, 6.0684, 2.3401, 5.4947, 5.1191, 5.5952, 5.4847], + device='cuda:2'), covar=tensor([0.0396, 0.4277, 0.0328, 0.3762, 0.0862, 0.0776, 0.0448, 0.0463], + device='cuda:2'), in_proj_covar=tensor([0.0627, 0.0639, 0.0693, 0.0628, 0.0705, 0.0603, 0.0606, 0.0677], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 02:48:06,024 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168053.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:48:09,816 INFO [train.py:901] (2/4) Epoch 21, batch 6400, loss[loss=0.184, simple_loss=0.2638, pruned_loss=0.05211, over 7926.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2875, pruned_loss=0.06217, over 1609923.02 frames. ], batch size: 20, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:48:25,123 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168081.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:48:30,484 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.236e+02 2.639e+02 3.603e+02 6.999e+02, threshold=5.279e+02, percent-clipped=2.0 +2023-02-07 02:48:45,445 INFO [train.py:901] (2/4) Epoch 21, batch 6450, loss[loss=0.1894, simple_loss=0.2819, pruned_loss=0.04838, over 8198.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2872, pruned_loss=0.06213, over 1610209.03 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:48:50,429 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1654, 3.7231, 2.3909, 2.8834, 3.0308, 2.0753, 2.9091, 3.0352], + device='cuda:2'), covar=tensor([0.1751, 0.0349, 0.1156, 0.0837, 0.0715, 0.1457, 0.1094, 0.1047], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0235, 0.0335, 0.0308, 0.0300, 0.0336, 0.0345, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:48:59,388 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168129.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:19,828 INFO [train.py:901] (2/4) Epoch 21, batch 6500, loss[loss=0.1798, simple_loss=0.2678, pruned_loss=0.04589, over 7806.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.287, pruned_loss=0.0621, over 1609212.75 frames. ], batch size: 20, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:49:24,797 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:26,330 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 02:49:26,879 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4415, 2.2860, 3.2010, 2.6081, 3.0161, 2.4303, 2.1900, 1.9306], + device='cuda:2'), covar=tensor([0.5136, 0.5025, 0.1896, 0.3434, 0.2504, 0.3101, 0.1980, 0.5049], + device='cuda:2'), in_proj_covar=tensor([0.0940, 0.0974, 0.0802, 0.0941, 0.0998, 0.0893, 0.0745, 0.0821], + 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') +2023-02-07 02:49:41,358 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.483e+02 3.129e+02 4.081e+02 1.148e+03, threshold=6.258e+02, percent-clipped=13.0 +2023-02-07 02:49:42,168 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168190.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:46,109 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168196.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:52,318 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0679, 1.8557, 2.3925, 2.0506, 2.2474, 2.1351, 1.8251, 1.1405], + device='cuda:2'), covar=tensor([0.5519, 0.4306, 0.1704, 0.2988, 0.2339, 0.2768, 0.1982, 0.4441], + device='cuda:2'), in_proj_covar=tensor([0.0939, 0.0974, 0.0801, 0.0941, 0.0998, 0.0892, 0.0745, 0.0821], + 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') +2023-02-07 02:49:54,742 INFO [train.py:901] (2/4) Epoch 21, batch 6550, loss[loss=0.1633, simple_loss=0.2427, pruned_loss=0.04194, over 7816.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2873, pruned_loss=0.06251, over 1611822.11 frames. ], batch size: 20, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:50:19,983 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168244.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:50:20,482 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 02:50:24,647 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168251.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:50:29,932 INFO [train.py:901] (2/4) Epoch 21, batch 6600, loss[loss=0.2056, simple_loss=0.2858, pruned_loss=0.06274, over 8079.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.287, pruned_loss=0.06207, over 1614167.24 frames. ], batch size: 21, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:50:38,734 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 02:50:50,801 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.414e+02 2.830e+02 3.481e+02 7.637e+02, threshold=5.659e+02, percent-clipped=3.0 +2023-02-07 02:51:05,095 INFO [train.py:901] (2/4) Epoch 21, batch 6650, loss[loss=0.2429, simple_loss=0.3282, pruned_loss=0.07883, over 8487.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2865, pruned_loss=0.06202, over 1610386.42 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:51:16,784 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9780, 1.9198, 3.2272, 2.4164, 2.7374, 1.9566, 1.7234, 1.8441], + device='cuda:2'), covar=tensor([0.6662, 0.5801, 0.1770, 0.3907, 0.3298, 0.4212, 0.2862, 0.5219], + device='cuda:2'), in_proj_covar=tensor([0.0932, 0.0967, 0.0797, 0.0935, 0.0992, 0.0887, 0.0741, 0.0817], + 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') +2023-02-07 02:51:29,486 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5071, 2.4711, 1.7989, 2.1874, 2.0417, 1.4908, 1.9819, 2.0219], + device='cuda:2'), covar=tensor([0.1554, 0.0418, 0.1298, 0.0654, 0.0791, 0.1607, 0.1031, 0.0978], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0235, 0.0334, 0.0307, 0.0300, 0.0335, 0.0344, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:51:40,100 INFO [train.py:901] (2/4) Epoch 21, batch 6700, loss[loss=0.1806, simple_loss=0.2625, pruned_loss=0.04929, over 8242.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2873, pruned_loss=0.06189, over 1613395.03 frames. ], batch size: 22, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:51:58,802 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-07 02:52:00,445 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.306e+02 2.933e+02 3.476e+02 6.537e+02, threshold=5.866e+02, percent-clipped=2.0 +2023-02-07 02:52:05,989 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168397.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:52:14,913 INFO [train.py:901] (2/4) Epoch 21, batch 6750, loss[loss=0.2004, simple_loss=0.2846, pruned_loss=0.0581, over 8193.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2873, pruned_loss=0.06138, over 1611756.62 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:52:16,706 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 02:52:32,330 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6827, 2.5938, 1.8343, 2.3246, 2.1722, 1.6037, 2.0323, 2.2170], + device='cuda:2'), covar=tensor([0.1383, 0.0349, 0.1231, 0.0583, 0.0746, 0.1470, 0.0956, 0.0867], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0236, 0.0336, 0.0307, 0.0300, 0.0336, 0.0344, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 02:52:39,060 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7896, 1.5820, 4.9311, 1.8835, 4.4117, 4.1597, 4.4743, 4.3784], + device='cuda:2'), covar=tensor([0.0452, 0.4426, 0.0405, 0.4088, 0.1008, 0.0925, 0.0513, 0.0635], + device='cuda:2'), in_proj_covar=tensor([0.0634, 0.0646, 0.0699, 0.0635, 0.0718, 0.0612, 0.0614, 0.0685], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 02:52:45,400 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:52:50,751 INFO [train.py:901] (2/4) Epoch 21, batch 6800, loss[loss=0.2079, simple_loss=0.2956, pruned_loss=0.06013, over 8484.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2875, pruned_loss=0.06146, over 1612681.69 frames. ], batch size: 27, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:52:58,515 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 02:53:04,338 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168477.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:09,208 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 02:53:12,378 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.333e+02 2.834e+02 3.373e+02 7.883e+02, threshold=5.669e+02, percent-clipped=5.0 +2023-02-07 02:53:15,413 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4530, 1.8875, 1.9779, 1.2202, 2.0062, 1.3733, 0.4100, 1.6761], + device='cuda:2'), covar=tensor([0.0593, 0.0348, 0.0253, 0.0547, 0.0389, 0.0924, 0.0915, 0.0282], + device='cuda:2'), in_proj_covar=tensor([0.0453, 0.0389, 0.0341, 0.0441, 0.0375, 0.0533, 0.0390, 0.0416], + device='cuda:2'), out_proj_covar=tensor([1.2177e-04, 1.0202e-04, 8.9875e-05, 1.1635e-04, 9.8628e-05, 1.5061e-04, + 1.0535e-04, 1.1017e-04], device='cuda:2') +2023-02-07 02:53:16,002 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5905, 4.6380, 4.1050, 2.2351, 4.0559, 4.1504, 4.1247, 3.9117], + device='cuda:2'), covar=tensor([0.0691, 0.0475, 0.1076, 0.4044, 0.0927, 0.1043, 0.1211, 0.0874], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0429, 0.0429, 0.0533, 0.0423, 0.0439, 0.0419, 0.0383], + 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') +2023-02-07 02:53:20,350 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168500.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:26,309 INFO [train.py:901] (2/4) Epoch 21, batch 6850, loss[loss=0.1611, simple_loss=0.2383, pruned_loss=0.04196, over 7802.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2878, pruned_loss=0.06149, over 1615173.58 frames. ], batch size: 19, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:53:28,535 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168512.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:31,835 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5097, 1.7546, 2.8888, 1.3988, 2.1742, 1.8942, 1.5495, 2.0879], + device='cuda:2'), covar=tensor([0.1918, 0.2623, 0.0932, 0.4430, 0.1808, 0.3049, 0.2345, 0.2223], + device='cuda:2'), in_proj_covar=tensor([0.0519, 0.0594, 0.0550, 0.0635, 0.0636, 0.0585, 0.0528, 0.0625], + 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') +2023-02-07 02:53:37,585 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168525.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:45,921 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 02:54:00,738 INFO [train.py:901] (2/4) Epoch 21, batch 6900, loss[loss=0.1664, simple_loss=0.2625, pruned_loss=0.03521, over 7977.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2898, pruned_loss=0.06253, over 1618388.41 frames. ], batch size: 21, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:54:17,245 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0226, 1.7778, 2.8154, 2.1231, 2.4283, 1.8775, 1.6971, 1.2631], + device='cuda:2'), covar=tensor([0.6723, 0.6378, 0.2088, 0.4268, 0.3299, 0.4600, 0.2916, 0.5695], + device='cuda:2'), in_proj_covar=tensor([0.0940, 0.0974, 0.0802, 0.0940, 0.0999, 0.0893, 0.0745, 0.0824], + 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') +2023-02-07 02:54:22,253 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.460e+02 2.867e+02 3.613e+02 6.820e+02, threshold=5.733e+02, percent-clipped=1.0 +2023-02-07 02:54:26,393 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168595.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:54:35,737 INFO [train.py:901] (2/4) Epoch 21, batch 6950, loss[loss=0.2464, simple_loss=0.3254, pruned_loss=0.08372, over 8617.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2908, pruned_loss=0.06327, over 1621073.45 frames. ], batch size: 39, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:54:53,399 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 02:55:10,659 INFO [train.py:901] (2/4) Epoch 21, batch 7000, loss[loss=0.1859, simple_loss=0.2879, pruned_loss=0.04193, over 8327.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2905, pruned_loss=0.06321, over 1617957.03 frames. ], batch size: 26, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:55:26,413 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-07 02:55:31,362 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.444e+02 3.041e+02 3.968e+02 8.528e+02, threshold=6.083e+02, percent-clipped=8.0 +2023-02-07 02:55:39,890 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2656, 1.9763, 2.6966, 2.2473, 2.6573, 2.2114, 2.0610, 1.6062], + device='cuda:2'), covar=tensor([0.4732, 0.4569, 0.1766, 0.3201, 0.2136, 0.2851, 0.1882, 0.4565], + device='cuda:2'), in_proj_covar=tensor([0.0938, 0.0972, 0.0801, 0.0940, 0.0995, 0.0891, 0.0744, 0.0822], + 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') +2023-02-07 02:55:45,700 INFO [train.py:901] (2/4) Epoch 21, batch 7050, loss[loss=0.2389, simple_loss=0.3131, pruned_loss=0.08235, over 8464.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2902, pruned_loss=0.06325, over 1614325.39 frames. ], batch size: 29, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:55:46,589 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168710.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:56:19,948 INFO [train.py:901] (2/4) Epoch 21, batch 7100, loss[loss=0.1953, simple_loss=0.287, pruned_loss=0.05175, over 8245.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2902, pruned_loss=0.06283, over 1616442.83 frames. ], batch size: 24, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:56:26,885 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:56:40,683 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.605e+02 3.011e+02 3.811e+02 1.077e+03, threshold=6.022e+02, percent-clipped=4.0 +2023-02-07 02:56:43,689 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168793.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:56:44,554 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2023-02-07 02:56:55,259 INFO [train.py:901] (2/4) Epoch 21, batch 7150, loss[loss=0.2121, simple_loss=0.2847, pruned_loss=0.06979, over 7653.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2905, pruned_loss=0.0629, over 1618895.98 frames. ], batch size: 19, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:57:20,050 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.02 vs. limit=5.0 +2023-02-07 02:57:29,818 INFO [train.py:901] (2/4) Epoch 21, batch 7200, loss[loss=0.2547, simple_loss=0.3219, pruned_loss=0.09378, over 7312.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2896, pruned_loss=0.06258, over 1617041.74 frames. ], batch size: 72, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:57:36,066 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1424, 2.5405, 2.9500, 1.5168, 3.1818, 1.9148, 1.4849, 2.0685], + device='cuda:2'), covar=tensor([0.0955, 0.0438, 0.0350, 0.0937, 0.0453, 0.0927, 0.0958, 0.0644], + device='cuda:2'), in_proj_covar=tensor([0.0451, 0.0389, 0.0341, 0.0442, 0.0372, 0.0532, 0.0388, 0.0415], + device='cuda:2'), out_proj_covar=tensor([1.2098e-04, 1.0219e-04, 8.9754e-05, 1.1650e-04, 9.7969e-05, 1.5034e-04, + 1.0475e-04, 1.1013e-04], device='cuda:2') +2023-02-07 02:57:38,238 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 02:57:51,146 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 2.343e+02 3.196e+02 4.097e+02 7.456e+02, threshold=6.392e+02, percent-clipped=6.0 +2023-02-07 02:58:02,866 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6066, 1.9652, 2.9496, 1.5270, 2.1528, 2.0637, 1.7442, 2.0341], + device='cuda:2'), covar=tensor([0.1840, 0.2560, 0.0905, 0.4454, 0.1806, 0.3040, 0.2223, 0.2162], + device='cuda:2'), in_proj_covar=tensor([0.0528, 0.0606, 0.0560, 0.0646, 0.0648, 0.0599, 0.0536, 0.0635], + 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') +2023-02-07 02:58:04,692 INFO [train.py:901] (2/4) Epoch 21, batch 7250, loss[loss=0.1812, simple_loss=0.2616, pruned_loss=0.05035, over 7970.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2894, pruned_loss=0.06276, over 1614306.12 frames. ], batch size: 21, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:58:40,072 INFO [train.py:901] (2/4) Epoch 21, batch 7300, loss[loss=0.1846, simple_loss=0.2815, pruned_loss=0.04381, over 8368.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2884, pruned_loss=0.06223, over 1613392.93 frames. ], batch size: 24, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:58:44,981 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168966.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:58:58,208 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168985.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:58:59,404 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8494, 3.7886, 3.4699, 1.8483, 3.4177, 3.4764, 3.3983, 3.2266], + device='cuda:2'), covar=tensor([0.0816, 0.0652, 0.1208, 0.4470, 0.0966, 0.0944, 0.1471, 0.0982], + device='cuda:2'), in_proj_covar=tensor([0.0523, 0.0431, 0.0432, 0.0533, 0.0423, 0.0442, 0.0421, 0.0384], + 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') +2023-02-07 02:58:59,448 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168987.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:59:00,602 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.375e+02 2.880e+02 4.111e+02 9.346e+02, threshold=5.760e+02, percent-clipped=6.0 +2023-02-07 02:59:02,086 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168991.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:59:14,682 INFO [train.py:901] (2/4) Epoch 21, batch 7350, loss[loss=0.1832, simple_loss=0.2682, pruned_loss=0.04914, over 7787.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2879, pruned_loss=0.06206, over 1614792.21 frames. ], batch size: 19, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:59:35,046 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 02:59:49,830 INFO [train.py:901] (2/4) Epoch 21, batch 7400, loss[loss=0.2241, simple_loss=0.3021, pruned_loss=0.07307, over 8434.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.288, pruned_loss=0.06221, over 1612171.94 frames. ], batch size: 27, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:59:53,389 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 03:00:01,565 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 +2023-02-07 03:00:10,727 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.483e+02 2.322e+02 3.020e+02 4.298e+02 1.187e+03, threshold=6.039e+02, percent-clipped=6.0 +2023-02-07 03:00:19,220 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169100.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:00:25,151 INFO [train.py:901] (2/4) Epoch 21, batch 7450, loss[loss=0.2383, simple_loss=0.317, pruned_loss=0.07977, over 8447.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2883, pruned_loss=0.06201, over 1613836.46 frames. ], batch size: 48, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:00:33,887 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 03:00:42,576 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169134.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:01:01,174 INFO [train.py:901] (2/4) Epoch 21, batch 7500, loss[loss=0.2047, simple_loss=0.295, pruned_loss=0.05718, over 8251.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2881, pruned_loss=0.062, over 1612958.12 frames. ], batch size: 24, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:01:13,527 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169177.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:01:21,459 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.287e+02 2.739e+02 3.438e+02 5.948e+02, threshold=5.478e+02, percent-clipped=0.0 +2023-02-07 03:01:35,755 INFO [train.py:901] (2/4) Epoch 21, batch 7550, loss[loss=0.2101, simple_loss=0.2893, pruned_loss=0.06542, over 7924.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2882, pruned_loss=0.0622, over 1611312.64 frames. ], batch size: 20, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:02:09,750 INFO [train.py:901] (2/4) Epoch 21, batch 7600, loss[loss=0.1773, simple_loss=0.257, pruned_loss=0.04874, over 7807.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2875, pruned_loss=0.06199, over 1608389.87 frames. ], batch size: 19, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:02:32,180 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.243e+02 2.742e+02 3.349e+02 1.012e+03, threshold=5.485e+02, percent-clipped=5.0 +2023-02-07 03:02:33,803 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3515, 2.9881, 2.1737, 3.8493, 1.9401, 2.0192, 2.4432, 2.8767], + device='cuda:2'), covar=tensor([0.0731, 0.0746, 0.0957, 0.0255, 0.1050, 0.1271, 0.0940, 0.0800], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0197, 0.0246, 0.0215, 0.0208, 0.0248, 0.0252, 0.0211], + 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') +2023-02-07 03:02:44,277 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 03:02:45,876 INFO [train.py:901] (2/4) Epoch 21, batch 7650, loss[loss=0.2452, simple_loss=0.3198, pruned_loss=0.08534, over 8502.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2876, pruned_loss=0.06202, over 1610243.62 frames. ], batch size: 26, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:03:00,457 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169329.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:03:01,834 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169331.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:03:21,300 INFO [train.py:901] (2/4) Epoch 21, batch 7700, loss[loss=0.242, simple_loss=0.3155, pruned_loss=0.08423, over 8328.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2868, pruned_loss=0.06168, over 1607340.94 frames. ], batch size: 26, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:03:24,165 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8843, 2.2908, 4.2429, 1.6145, 3.0407, 2.4585, 1.8023, 3.1910], + device='cuda:2'), covar=tensor([0.1896, 0.2745, 0.0736, 0.4434, 0.1721, 0.3051, 0.2437, 0.2014], + device='cuda:2'), in_proj_covar=tensor([0.0527, 0.0602, 0.0559, 0.0642, 0.0646, 0.0596, 0.0534, 0.0633], + 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') +2023-02-07 03:03:42,195 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.349e+02 2.901e+02 3.736e+02 6.675e+02, threshold=5.802e+02, percent-clipped=6.0 +2023-02-07 03:03:44,250 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 03:03:57,026 INFO [train.py:901] (2/4) Epoch 21, batch 7750, loss[loss=0.1965, simple_loss=0.2825, pruned_loss=0.05526, over 8347.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2875, pruned_loss=0.06196, over 1612450.90 frames. ], batch size: 26, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:04:21,910 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169444.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:04:22,019 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169444.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:04:23,340 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169446.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:04:32,641 INFO [train.py:901] (2/4) Epoch 21, batch 7800, loss[loss=0.1911, simple_loss=0.2727, pruned_loss=0.05476, over 7705.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2867, pruned_loss=0.06173, over 1604954.96 frames. ], batch size: 18, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:04:34,847 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5298, 1.8036, 1.9607, 1.4367, 2.0536, 1.3902, 0.5829, 1.7925], + device='cuda:2'), covar=tensor([0.0628, 0.0399, 0.0286, 0.0555, 0.0432, 0.0991, 0.0850, 0.0303], + device='cuda:2'), in_proj_covar=tensor([0.0453, 0.0391, 0.0342, 0.0442, 0.0374, 0.0533, 0.0389, 0.0419], + device='cuda:2'), out_proj_covar=tensor([1.2170e-04, 1.0256e-04, 9.0148e-05, 1.1650e-04, 9.8285e-05, 1.5050e-04, + 1.0519e-04, 1.1112e-04], device='cuda:2') +2023-02-07 03:04:45,393 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169478.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:04:52,675 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.145e+02 2.738e+02 3.428e+02 8.790e+02, threshold=5.476e+02, percent-clipped=3.0 +2023-02-07 03:04:52,926 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4001, 2.3407, 1.7148, 2.1289, 2.0045, 1.5215, 1.8423, 1.8619], + device='cuda:2'), covar=tensor([0.1520, 0.0469, 0.1281, 0.0613, 0.0742, 0.1520, 0.0993, 0.1062], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0236, 0.0335, 0.0307, 0.0301, 0.0336, 0.0343, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:04:55,617 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4135, 1.4426, 1.7967, 1.3635, 1.1448, 1.8161, 0.2364, 1.1909], + device='cuda:2'), covar=tensor([0.1639, 0.1187, 0.0437, 0.0845, 0.2417, 0.0420, 0.1964, 0.1103], + device='cuda:2'), in_proj_covar=tensor([0.0188, 0.0195, 0.0127, 0.0221, 0.0270, 0.0135, 0.0171, 0.0191], + 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') +2023-02-07 03:05:06,025 INFO [train.py:901] (2/4) Epoch 21, batch 7850, loss[loss=0.1827, simple_loss=0.2681, pruned_loss=0.0486, over 7811.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2861, pruned_loss=0.06078, over 1607233.00 frames. ], batch size: 20, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:05:06,872 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3598, 2.0368, 1.4935, 2.0318, 1.7742, 1.2818, 1.7002, 1.7982], + device='cuda:2'), covar=tensor([0.1160, 0.0424, 0.1368, 0.0441, 0.0716, 0.1677, 0.0860, 0.0724], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0236, 0.0335, 0.0308, 0.0300, 0.0336, 0.0344, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:05:10,168 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2013, 1.4707, 3.3163, 1.1583, 2.9333, 2.7499, 3.0289, 2.9190], + device='cuda:2'), covar=tensor([0.0801, 0.3801, 0.0818, 0.3958, 0.1398, 0.1179, 0.0819, 0.0891], + device='cuda:2'), in_proj_covar=tensor([0.0630, 0.0640, 0.0693, 0.0626, 0.0711, 0.0611, 0.0611, 0.0675], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 03:05:14,138 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169521.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:05:39,262 INFO [train.py:901] (2/4) Epoch 21, batch 7900, loss[loss=0.2746, simple_loss=0.3445, pruned_loss=0.1024, over 8681.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2882, pruned_loss=0.06192, over 1612540.57 frames. ], batch size: 34, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:05:39,446 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169559.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:05:41,601 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.77 vs. limit=5.0 +2023-02-07 03:05:59,284 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.405e+02 2.884e+02 3.520e+02 8.387e+02, threshold=5.767e+02, percent-clipped=5.0 +2023-02-07 03:06:02,050 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169593.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:06:08,079 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.08 vs. limit=5.0 +2023-02-07 03:06:12,852 INFO [train.py:901] (2/4) Epoch 21, batch 7950, loss[loss=0.217, simple_loss=0.2995, pruned_loss=0.06725, over 8256.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2883, pruned_loss=0.06209, over 1610717.56 frames. ], batch size: 24, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:06:31,342 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169636.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:06:33,343 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169639.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:06:46,581 INFO [train.py:901] (2/4) Epoch 21, batch 8000, loss[loss=0.1616, simple_loss=0.2393, pruned_loss=0.04196, over 7259.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2872, pruned_loss=0.06136, over 1611345.21 frames. ], batch size: 16, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:07:06,436 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.194e+02 2.844e+02 3.383e+02 6.688e+02, threshold=5.687e+02, percent-clipped=2.0 +2023-02-07 03:07:07,193 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9847, 6.2235, 5.3384, 2.6708, 5.4706, 5.8563, 5.6813, 5.6032], + device='cuda:2'), covar=tensor([0.0547, 0.0362, 0.0853, 0.3943, 0.0698, 0.0703, 0.1149, 0.0544], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0428, 0.0431, 0.0531, 0.0422, 0.0440, 0.0421, 0.0380], + 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') +2023-02-07 03:07:12,039 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9842, 1.6845, 2.0240, 1.8205, 1.9282, 2.0047, 1.8400, 0.8259], + device='cuda:2'), covar=tensor([0.5719, 0.5068, 0.2023, 0.3555, 0.2511, 0.3105, 0.1970, 0.5033], + device='cuda:2'), in_proj_covar=tensor([0.0945, 0.0977, 0.0803, 0.0946, 0.0998, 0.0892, 0.0746, 0.0823], + 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') +2023-02-07 03:07:12,145 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-07 03:07:14,027 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169700.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:07:15,418 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169702.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 03:07:19,700 INFO [train.py:901] (2/4) Epoch 21, batch 8050, loss[loss=0.2326, simple_loss=0.3121, pruned_loss=0.07651, over 7193.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2852, pruned_loss=0.06139, over 1586477.39 frames. ], batch size: 71, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:07:30,630 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169725.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:07:32,001 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169727.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:07:52,653 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 03:07:58,222 INFO [train.py:901] (2/4) Epoch 22, batch 0, loss[loss=0.1745, simple_loss=0.2549, pruned_loss=0.04706, over 7821.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2549, pruned_loss=0.04706, over 7821.00 frames. ], batch size: 20, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:07:58,222 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 03:08:05,038 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6955, 1.8440, 1.5897, 2.0750, 1.2268, 1.4524, 1.6733, 1.7797], + device='cuda:2'), covar=tensor([0.0736, 0.0631, 0.0922, 0.0537, 0.1046, 0.1256, 0.0741, 0.0751], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0196, 0.0245, 0.0213, 0.0207, 0.0246, 0.0250, 0.0209], + 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') +2023-02-07 03:08:09,347 INFO [train.py:935] (2/4) Epoch 22, validation: loss=0.1743, simple_loss=0.2746, pruned_loss=0.03702, over 944034.00 frames. +2023-02-07 03:08:09,348 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 03:08:12,912 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169747.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:08:17,071 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6966, 1.6528, 1.9986, 1.8593, 1.0779, 1.7082, 2.2375, 2.2141], + device='cuda:2'), covar=tensor([0.0432, 0.1220, 0.1597, 0.1294, 0.0559, 0.1393, 0.0583, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0160, 0.0100, 0.0164, 0.0113, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 03:08:24,252 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 03:08:25,073 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169765.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:08:36,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5692, 2.5386, 1.8092, 2.2599, 2.1150, 1.6367, 2.0404, 2.1955], + device='cuda:2'), covar=tensor([0.1501, 0.0473, 0.1225, 0.0628, 0.0749, 0.1467, 0.0972, 0.0997], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0234, 0.0331, 0.0305, 0.0298, 0.0332, 0.0340, 0.0315], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:08:42,193 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 2.482e+02 2.980e+02 3.558e+02 1.069e+03, threshold=5.959e+02, percent-clipped=8.0 +2023-02-07 03:08:44,166 INFO [train.py:901] (2/4) Epoch 22, batch 50, loss[loss=0.24, simple_loss=0.2982, pruned_loss=0.09092, over 7428.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2849, pruned_loss=0.06159, over 361730.18 frames. ], batch size: 17, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:08:54,109 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169804.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:01,057 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 03:09:02,044 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169815.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:19,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169840.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:20,336 INFO [train.py:901] (2/4) Epoch 22, batch 100, loss[loss=0.2066, simple_loss=0.2971, pruned_loss=0.05801, over 8359.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2869, pruned_loss=0.06269, over 639048.65 frames. ], batch size: 24, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:09:23,119 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 03:09:25,374 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169849.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:28,968 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 03:09:42,078 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169874.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:52,904 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.356e+02 3.069e+02 3.800e+02 7.981e+02, threshold=6.138e+02, percent-clipped=3.0 +2023-02-07 03:09:55,642 INFO [train.py:901] (2/4) Epoch 22, batch 150, loss[loss=0.2235, simple_loss=0.3038, pruned_loss=0.07157, over 8468.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2873, pruned_loss=0.06145, over 859170.34 frames. ], batch size: 25, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:09:55,877 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169892.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:10:10,890 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.08 vs. limit=5.0 +2023-02-07 03:10:12,774 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169917.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:10:23,673 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9657, 1.3088, 1.5736, 1.2552, 0.9158, 1.3810, 1.7578, 1.6070], + device='cuda:2'), covar=tensor([0.0515, 0.1349, 0.1876, 0.1584, 0.0662, 0.1586, 0.0694, 0.0669], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0190, 0.0159, 0.0099, 0.0163, 0.0112, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 03:10:30,755 INFO [train.py:901] (2/4) Epoch 22, batch 200, loss[loss=0.2061, simple_loss=0.3011, pruned_loss=0.05561, over 8495.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2884, pruned_loss=0.06127, over 1031482.82 frames. ], batch size: 49, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:10:58,689 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169983.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:11:02,622 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.362e+02 2.871e+02 3.395e+02 8.094e+02, threshold=5.742e+02, percent-clipped=2.0 +2023-02-07 03:11:04,627 INFO [train.py:901] (2/4) Epoch 22, batch 250, loss[loss=0.2707, simple_loss=0.3363, pruned_loss=0.1026, over 8513.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2872, pruned_loss=0.06134, over 1159818.56 frames. ], batch size: 28, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:11:17,880 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 03:11:26,113 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 03:11:41,665 INFO [train.py:901] (2/4) Epoch 22, batch 300, loss[loss=0.2038, simple_loss=0.2914, pruned_loss=0.05806, over 7940.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2861, pruned_loss=0.0602, over 1263827.46 frames. ], batch size: 20, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:11:50,133 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 03:11:56,574 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170063.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:13,696 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 2.486e+02 2.821e+02 3.492e+02 6.452e+02, threshold=5.641e+02, percent-clipped=3.0 +2023-02-07 03:12:15,179 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:15,769 INFO [train.py:901] (2/4) Epoch 22, batch 350, loss[loss=0.2177, simple_loss=0.3017, pruned_loss=0.06682, over 8507.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2868, pruned_loss=0.06064, over 1338116.23 frames. ], batch size: 26, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:12:19,958 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170098.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:27,046 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170109.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:49,714 INFO [train.py:901] (2/4) Epoch 22, batch 400, loss[loss=0.1995, simple_loss=0.2926, pruned_loss=0.05324, over 8257.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2859, pruned_loss=0.06021, over 1398861.83 frames. ], batch size: 24, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:12:53,699 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170148.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:13:22,569 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.277e+02 2.821e+02 3.460e+02 6.418e+02, threshold=5.643e+02, percent-clipped=3.0 +2023-02-07 03:13:24,663 INFO [train.py:901] (2/4) Epoch 22, batch 450, loss[loss=0.1989, simple_loss=0.2713, pruned_loss=0.0633, over 8086.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.286, pruned_loss=0.05986, over 1449873.14 frames. ], batch size: 21, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:13:34,380 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170206.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:13:46,333 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:13:58,268 INFO [train.py:901] (2/4) Epoch 22, batch 500, loss[loss=0.1816, simple_loss=0.2761, pruned_loss=0.04349, over 8034.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2859, pruned_loss=0.06042, over 1487203.41 frames. ], batch size: 22, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:14:13,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170263.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:14:31,687 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.263e+02 2.770e+02 3.716e+02 6.957e+02, threshold=5.540e+02, percent-clipped=5.0 +2023-02-07 03:14:34,516 INFO [train.py:901] (2/4) Epoch 22, batch 550, loss[loss=0.1947, simple_loss=0.2955, pruned_loss=0.04698, over 8536.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.287, pruned_loss=0.06114, over 1516891.84 frames. ], batch size: 31, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:15:00,844 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9104, 2.2664, 1.7213, 2.8947, 1.2527, 1.5441, 1.8616, 2.2332], + device='cuda:2'), covar=tensor([0.0823, 0.0789, 0.1053, 0.0405, 0.1256, 0.1428, 0.1065, 0.0835], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0196, 0.0243, 0.0213, 0.0206, 0.0245, 0.0248, 0.0208], + 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') +2023-02-07 03:15:08,213 INFO [train.py:901] (2/4) Epoch 22, batch 600, loss[loss=0.196, simple_loss=0.2804, pruned_loss=0.05581, over 7810.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2879, pruned_loss=0.06158, over 1536058.89 frames. ], batch size: 20, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:15:16,557 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170354.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:15:27,502 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 03:15:34,303 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170379.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:15:40,792 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.463e+02 3.010e+02 3.561e+02 9.437e+02, threshold=6.021e+02, percent-clipped=1.0 +2023-02-07 03:15:42,748 INFO [train.py:901] (2/4) Epoch 22, batch 650, loss[loss=0.2166, simple_loss=0.2944, pruned_loss=0.06944, over 8130.00 frames. ], tot_loss[loss=0.208, simple_loss=0.29, pruned_loss=0.06298, over 1558593.02 frames. ], batch size: 22, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:15:52,779 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5164, 2.4072, 2.2851, 1.3271, 2.2101, 2.2769, 2.2389, 2.1857], + device='cuda:2'), covar=tensor([0.1090, 0.0864, 0.1190, 0.3818, 0.1003, 0.1205, 0.1523, 0.1042], + device='cuda:2'), in_proj_covar=tensor([0.0524, 0.0430, 0.0428, 0.0530, 0.0422, 0.0441, 0.0423, 0.0381], + 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') +2023-02-07 03:15:53,431 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170407.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:17,636 INFO [train.py:901] (2/4) Epoch 22, batch 700, loss[loss=0.1883, simple_loss=0.2847, pruned_loss=0.04593, over 8469.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2887, pruned_loss=0.0626, over 1570550.02 frames. ], batch size: 25, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:16:31,460 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:42,963 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170479.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:43,708 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170480.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:49,120 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170487.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:50,916 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.347e+02 2.936e+02 3.672e+02 5.936e+02, threshold=5.871e+02, percent-clipped=0.0 +2023-02-07 03:16:52,904 INFO [train.py:901] (2/4) Epoch 22, batch 750, loss[loss=0.2622, simple_loss=0.3188, pruned_loss=0.1027, over 7806.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2889, pruned_loss=0.06272, over 1579810.90 frames. ], batch size: 20, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:17:01,713 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170505.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:11,626 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170519.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:13,543 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170522.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:14,716 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 03:17:23,976 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 03:17:27,394 INFO [train.py:901] (2/4) Epoch 22, batch 800, loss[loss=0.222, simple_loss=0.2987, pruned_loss=0.07263, over 8500.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2882, pruned_loss=0.06262, over 1587087.36 frames. ], batch size: 26, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:17:28,964 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170544.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:30,548 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 03:17:57,574 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170587.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:58,761 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.234e+02 2.598e+02 3.180e+02 6.753e+02, threshold=5.195e+02, percent-clipped=1.0 +2023-02-07 03:18:00,806 INFO [train.py:901] (2/4) Epoch 22, batch 850, loss[loss=0.1972, simple_loss=0.288, pruned_loss=0.05315, over 8518.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2873, pruned_loss=0.06174, over 1594771.55 frames. ], batch size: 28, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:18:14,727 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6144, 1.6707, 2.2506, 1.4339, 1.1844, 2.2663, 0.4571, 1.3637], + device='cuda:2'), covar=tensor([0.2023, 0.1305, 0.0388, 0.1347, 0.3113, 0.0376, 0.2174, 0.1387], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0196, 0.0127, 0.0222, 0.0270, 0.0135, 0.0171, 0.0193], + 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') +2023-02-07 03:18:31,338 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 03:18:36,994 INFO [train.py:901] (2/4) Epoch 22, batch 900, loss[loss=0.2293, simple_loss=0.3103, pruned_loss=0.07414, over 8457.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2875, pruned_loss=0.06194, over 1598287.66 frames. ], batch size: 49, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:18:53,611 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5662, 1.8495, 1.9311, 1.1942, 2.0450, 1.4049, 0.5990, 1.8218], + device='cuda:2'), covar=tensor([0.0622, 0.0354, 0.0279, 0.0570, 0.0389, 0.0880, 0.0859, 0.0287], + device='cuda:2'), in_proj_covar=tensor([0.0454, 0.0394, 0.0345, 0.0444, 0.0375, 0.0534, 0.0390, 0.0419], + device='cuda:2'), out_proj_covar=tensor([1.2187e-04, 1.0338e-04, 9.0653e-05, 1.1683e-04, 9.8477e-05, 1.5078e-04, + 1.0542e-04, 1.1114e-04], device='cuda:2') +2023-02-07 03:19:09,383 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.381e+02 2.827e+02 3.296e+02 7.509e+02, threshold=5.655e+02, percent-clipped=4.0 +2023-02-07 03:19:11,446 INFO [train.py:901] (2/4) Epoch 22, batch 950, loss[loss=0.1776, simple_loss=0.2548, pruned_loss=0.05019, over 7658.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2865, pruned_loss=0.06106, over 1599045.46 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:19:13,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2850, 1.4652, 4.2683, 2.0180, 2.5092, 4.9129, 4.9136, 4.2291], + device='cuda:2'), covar=tensor([0.1247, 0.2008, 0.0306, 0.1984, 0.1181, 0.0157, 0.0418, 0.0539], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0323, 0.0286, 0.0317, 0.0310, 0.0265, 0.0420, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:19:43,613 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 03:19:46,353 INFO [train.py:901] (2/4) Epoch 22, batch 1000, loss[loss=0.1683, simple_loss=0.2578, pruned_loss=0.03939, over 8254.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2871, pruned_loss=0.06159, over 1601561.42 frames. ], batch size: 22, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:19:49,225 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2496, 2.0838, 1.6516, 1.8865, 1.7540, 1.4463, 1.6586, 1.6608], + device='cuda:2'), covar=tensor([0.1364, 0.0435, 0.1181, 0.0563, 0.0722, 0.1465, 0.0939, 0.0899], + device='cuda:2'), in_proj_covar=tensor([0.0351, 0.0235, 0.0331, 0.0308, 0.0299, 0.0334, 0.0341, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:20:12,108 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170778.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:20:17,087 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 03:20:19,763 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.250e+02 2.326e+02 2.890e+02 3.504e+02 6.405e+02, threshold=5.779e+02, percent-clipped=4.0 +2023-02-07 03:20:21,012 INFO [train.py:901] (2/4) Epoch 22, batch 1050, loss[loss=0.2265, simple_loss=0.2975, pruned_loss=0.07774, over 7638.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2875, pruned_loss=0.06186, over 1603366.74 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:20:28,486 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 03:20:28,709 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170803.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:20:41,785 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170823.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:20:55,954 INFO [train.py:901] (2/4) Epoch 22, batch 1100, loss[loss=0.1998, simple_loss=0.2771, pruned_loss=0.06124, over 8589.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2884, pruned_loss=0.06224, over 1608092.98 frames. ], batch size: 34, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:21:27,511 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2938, 2.1197, 1.7538, 1.9541, 1.8290, 1.4851, 1.7275, 1.6844], + device='cuda:2'), covar=tensor([0.1231, 0.0455, 0.1120, 0.0511, 0.0728, 0.1468, 0.0943, 0.0831], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0237, 0.0334, 0.0310, 0.0300, 0.0336, 0.0344, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:21:29,315 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.576e+02 3.127e+02 3.706e+02 1.049e+03, threshold=6.255e+02, percent-clipped=5.0 +2023-02-07 03:21:30,680 INFO [train.py:901] (2/4) Epoch 22, batch 1150, loss[loss=0.1678, simple_loss=0.2594, pruned_loss=0.0381, over 7811.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2894, pruned_loss=0.06315, over 1609326.71 frames. ], batch size: 20, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:21:37,427 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 03:21:45,385 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8493, 1.4399, 3.2534, 1.2410, 2.1947, 3.5850, 3.8663, 2.6422], + device='cuda:2'), covar=tensor([0.1459, 0.2146, 0.0481, 0.2784, 0.1265, 0.0352, 0.0592, 0.1082], + device='cuda:2'), in_proj_covar=tensor([0.0293, 0.0320, 0.0283, 0.0314, 0.0306, 0.0262, 0.0415, 0.0304], + 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') +2023-02-07 03:21:52,887 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4214, 2.7099, 2.2009, 3.2705, 1.9467, 2.0606, 2.2906, 2.7397], + device='cuda:2'), covar=tensor([0.0638, 0.0715, 0.0800, 0.0460, 0.0959, 0.1138, 0.0841, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0198, 0.0247, 0.0215, 0.0209, 0.0249, 0.0251, 0.0210], + 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') +2023-02-07 03:21:56,817 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170931.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:22:01,706 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170938.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:22:04,213 INFO [train.py:901] (2/4) Epoch 22, batch 1200, loss[loss=0.2063, simple_loss=0.2864, pruned_loss=0.06308, over 8195.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2889, pruned_loss=0.0627, over 1608574.85 frames. ], batch size: 23, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:22:07,060 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170946.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:22:38,802 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.384e+02 2.807e+02 3.549e+02 5.873e+02, threshold=5.615e+02, percent-clipped=0.0 +2023-02-07 03:22:40,086 INFO [train.py:901] (2/4) Epoch 22, batch 1250, loss[loss=0.1794, simple_loss=0.2781, pruned_loss=0.04037, over 8366.00 frames. ], tot_loss[loss=0.207, simple_loss=0.289, pruned_loss=0.06246, over 1612735.29 frames. ], batch size: 24, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:22:57,672 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2086, 4.1700, 3.7681, 1.9819, 3.6646, 3.8081, 3.7020, 3.5828], + device='cuda:2'), covar=tensor([0.0835, 0.0580, 0.1264, 0.4681, 0.0984, 0.0810, 0.1422, 0.0731], + device='cuda:2'), in_proj_covar=tensor([0.0522, 0.0430, 0.0430, 0.0530, 0.0422, 0.0441, 0.0420, 0.0382], + 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') +2023-02-07 03:22:58,790 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 03:23:14,550 INFO [train.py:901] (2/4) Epoch 22, batch 1300, loss[loss=0.207, simple_loss=0.2881, pruned_loss=0.06296, over 8583.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2877, pruned_loss=0.0617, over 1612899.46 frames. ], batch size: 31, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:23:17,508 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171046.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:23:47,495 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.382e+02 2.988e+02 3.753e+02 7.309e+02, threshold=5.975e+02, percent-clipped=5.0 +2023-02-07 03:23:48,839 INFO [train.py:901] (2/4) Epoch 22, batch 1350, loss[loss=0.2062, simple_loss=0.2957, pruned_loss=0.05833, over 8638.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2878, pruned_loss=0.06162, over 1612674.71 frames. ], batch size: 31, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:24:01,669 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171110.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:24:02,359 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6442, 1.4850, 4.8870, 1.7520, 4.2901, 4.0117, 4.3466, 4.2100], + device='cuda:2'), covar=tensor([0.0639, 0.5129, 0.0439, 0.4368, 0.1193, 0.0993, 0.0632, 0.0678], + device='cuda:2'), in_proj_covar=tensor([0.0625, 0.0637, 0.0688, 0.0620, 0.0704, 0.0604, 0.0606, 0.0672], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 03:24:23,449 INFO [train.py:901] (2/4) Epoch 22, batch 1400, loss[loss=0.2092, simple_loss=0.2944, pruned_loss=0.06199, over 8027.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2882, pruned_loss=0.062, over 1613191.40 frames. ], batch size: 22, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:24:23,822 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.78 vs. limit=5.0 +2023-02-07 03:24:55,486 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.434e+02 3.047e+02 3.835e+02 9.203e+02, threshold=6.094e+02, percent-clipped=3.0 +2023-02-07 03:24:57,482 INFO [train.py:901] (2/4) Epoch 22, batch 1450, loss[loss=0.2702, simple_loss=0.3379, pruned_loss=0.1013, over 8536.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2874, pruned_loss=0.06169, over 1613904.35 frames. ], batch size: 49, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:24:58,894 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171194.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:25:06,244 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 03:25:12,522 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:25:16,690 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171219.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:25:32,453 INFO [train.py:901] (2/4) Epoch 22, batch 1500, loss[loss=0.2091, simple_loss=0.2804, pruned_loss=0.06893, over 7261.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2878, pruned_loss=0.06183, over 1616198.22 frames. ], batch size: 16, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:26:04,592 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.477e+02 2.962e+02 3.885e+02 1.079e+03, threshold=5.924e+02, percent-clipped=2.0 +2023-02-07 03:26:04,686 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171290.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:26:05,961 INFO [train.py:901] (2/4) Epoch 22, batch 1550, loss[loss=0.2025, simple_loss=0.2957, pruned_loss=0.05462, over 7978.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.289, pruned_loss=0.06273, over 1614916.27 frames. ], batch size: 21, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:26:12,939 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171302.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:26:25,492 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3800, 1.5271, 1.3834, 1.7823, 0.7532, 1.2605, 1.2549, 1.4901], + device='cuda:2'), covar=tensor([0.0862, 0.0766, 0.1073, 0.0540, 0.1210, 0.1435, 0.0848, 0.0774], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0214, 0.0207, 0.0247, 0.0250, 0.0211], + 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') +2023-02-07 03:26:30,106 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:26:37,452 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8512, 1.6389, 6.0313, 2.1266, 5.4167, 5.1198, 5.5382, 5.4356], + device='cuda:2'), covar=tensor([0.0451, 0.4809, 0.0311, 0.3933, 0.0921, 0.0836, 0.0487, 0.0500], + device='cuda:2'), in_proj_covar=tensor([0.0629, 0.0642, 0.0693, 0.0623, 0.0708, 0.0608, 0.0611, 0.0676], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 03:26:40,691 INFO [train.py:901] (2/4) Epoch 22, batch 1600, loss[loss=0.191, simple_loss=0.2746, pruned_loss=0.05368, over 8124.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.288, pruned_loss=0.06235, over 1611366.31 frames. ], batch size: 22, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:26:55,771 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171363.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:27:13,638 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.510e+02 3.045e+02 3.987e+02 6.104e+02, threshold=6.090e+02, percent-clipped=2.0 +2023-02-07 03:27:15,007 INFO [train.py:901] (2/4) Epoch 22, batch 1650, loss[loss=0.2066, simple_loss=0.2962, pruned_loss=0.05854, over 7800.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2879, pruned_loss=0.06169, over 1610928.80 frames. ], batch size: 20, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:27:24,117 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171405.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:27:51,036 INFO [train.py:901] (2/4) Epoch 22, batch 1700, loss[loss=0.1604, simple_loss=0.2424, pruned_loss=0.03924, over 7221.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.287, pruned_loss=0.06114, over 1611496.26 frames. ], batch size: 16, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:27:59,281 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171454.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:28:18,869 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-07 03:28:24,567 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.429e+02 3.050e+02 3.629e+02 7.357e+02, threshold=6.100e+02, percent-clipped=3.0 +2023-02-07 03:28:25,936 INFO [train.py:901] (2/4) Epoch 22, batch 1750, loss[loss=0.2093, simple_loss=0.2841, pruned_loss=0.06722, over 7804.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2867, pruned_loss=0.0611, over 1612242.03 frames. ], batch size: 20, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:28:42,135 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171516.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:29:00,472 INFO [train.py:901] (2/4) Epoch 22, batch 1800, loss[loss=0.1838, simple_loss=0.278, pruned_loss=0.04484, over 8485.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2871, pruned_loss=0.06129, over 1612737.56 frames. ], batch size: 27, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:29:11,509 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171558.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:29:19,681 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171569.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:29:34,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.408e+02 2.801e+02 3.784e+02 7.831e+02, threshold=5.602e+02, percent-clipped=2.0 +2023-02-07 03:29:35,951 INFO [train.py:901] (2/4) Epoch 22, batch 1850, loss[loss=0.2111, simple_loss=0.2935, pruned_loss=0.06437, over 8286.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2874, pruned_loss=0.06168, over 1610709.02 frames. ], batch size: 23, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:29:39,056 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 03:30:10,047 INFO [train.py:901] (2/4) Epoch 22, batch 1900, loss[loss=0.224, simple_loss=0.3088, pruned_loss=0.06956, over 8515.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2867, pruned_loss=0.06092, over 1613391.56 frames. ], batch size: 39, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:30:24,236 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171661.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:30:32,189 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171673.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:30:36,795 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 03:30:41,592 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171686.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:30:44,064 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.518e+02 3.035e+02 3.649e+02 9.576e+02, threshold=6.070e+02, percent-clipped=4.0 +2023-02-07 03:30:45,464 INFO [train.py:901] (2/4) Epoch 22, batch 1950, loss[loss=0.2202, simple_loss=0.3011, pruned_loss=0.06967, over 8461.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2866, pruned_loss=0.0608, over 1612006.30 frames. ], batch size: 25, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:30:48,017 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 03:30:56,293 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171707.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:31:07,790 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 03:31:20,038 INFO [train.py:901] (2/4) Epoch 22, batch 2000, loss[loss=0.2102, simple_loss=0.2979, pruned_loss=0.06127, over 8464.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2871, pruned_loss=0.06088, over 1617227.69 frames. ], batch size: 25, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:31:53,981 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.301e+02 2.928e+02 3.706e+02 6.798e+02, threshold=5.855e+02, percent-clipped=1.0 +2023-02-07 03:31:55,392 INFO [train.py:901] (2/4) Epoch 22, batch 2050, loss[loss=0.2293, simple_loss=0.3129, pruned_loss=0.07279, over 8190.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2872, pruned_loss=0.0607, over 1618581.05 frames. ], batch size: 23, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:32:17,603 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171822.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:19,740 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171825.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:31,138 INFO [train.py:901] (2/4) Epoch 22, batch 2100, loss[loss=0.2081, simple_loss=0.2962, pruned_loss=0.05998, over 8476.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.288, pruned_loss=0.06093, over 1619309.74 frames. ], batch size: 49, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:32:36,879 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171850.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:43,494 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171860.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:45,788 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4896, 1.5083, 1.9330, 1.3316, 1.0849, 1.9057, 0.4622, 1.2745], + device='cuda:2'), covar=tensor([0.1730, 0.1138, 0.0400, 0.1066, 0.2911, 0.0444, 0.2025, 0.1352], + device='cuda:2'), in_proj_covar=tensor([0.0190, 0.0198, 0.0127, 0.0223, 0.0272, 0.0137, 0.0171, 0.0194], + 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') +2023-02-07 03:33:05,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.505e+02 2.999e+02 3.749e+02 9.868e+02, threshold=5.998e+02, percent-clipped=7.0 +2023-02-07 03:33:06,894 INFO [train.py:901] (2/4) Epoch 22, batch 2150, loss[loss=0.1958, simple_loss=0.2829, pruned_loss=0.05436, over 8238.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2875, pruned_loss=0.06099, over 1617144.42 frames. ], batch size: 24, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:33:33,135 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:33:42,597 INFO [train.py:901] (2/4) Epoch 22, batch 2200, loss[loss=0.2213, simple_loss=0.2995, pruned_loss=0.0715, over 8426.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2884, pruned_loss=0.06161, over 1619113.60 frames. ], batch size: 29, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:33:51,043 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171954.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:33:51,692 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171955.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:34:04,373 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 +2023-02-07 03:34:05,584 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171975.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:34:15,551 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.362e+02 2.812e+02 3.623e+02 6.076e+02, threshold=5.624e+02, percent-clipped=1.0 +2023-02-07 03:34:16,933 INFO [train.py:901] (2/4) Epoch 22, batch 2250, loss[loss=0.2436, simple_loss=0.3095, pruned_loss=0.08884, over 8326.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2879, pruned_loss=0.06136, over 1621165.75 frames. ], batch size: 26, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:34:17,294 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-07 03:34:54,269 INFO [train.py:901] (2/4) Epoch 22, batch 2300, loss[loss=0.2066, simple_loss=0.2828, pruned_loss=0.06514, over 7980.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2893, pruned_loss=0.06229, over 1620870.47 frames. ], batch size: 21, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:35:19,366 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2000, 2.1162, 1.7638, 2.0169, 1.7792, 1.5346, 1.6750, 1.7040], + device='cuda:2'), covar=tensor([0.1343, 0.0413, 0.1170, 0.0481, 0.0721, 0.1445, 0.0952, 0.0877], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0233, 0.0331, 0.0307, 0.0299, 0.0336, 0.0343, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:35:20,117 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172078.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:35:28,306 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.401e+02 3.005e+02 3.667e+02 7.010e+02, threshold=6.010e+02, percent-clipped=1.0 +2023-02-07 03:35:29,624 INFO [train.py:901] (2/4) Epoch 22, batch 2350, loss[loss=0.2196, simple_loss=0.2957, pruned_loss=0.07179, over 8191.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2882, pruned_loss=0.0615, over 1619922.93 frames. ], batch size: 23, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:35:37,304 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172103.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:36:01,295 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172136.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:36:05,202 INFO [train.py:901] (2/4) Epoch 22, batch 2400, loss[loss=0.1888, simple_loss=0.267, pruned_loss=0.05531, over 7696.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2884, pruned_loss=0.06204, over 1618919.67 frames. ], batch size: 18, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:36:39,692 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.658e+02 3.455e+02 4.348e+02 7.809e+02, threshold=6.910e+02, percent-clipped=6.0 +2023-02-07 03:36:41,123 INFO [train.py:901] (2/4) Epoch 22, batch 2450, loss[loss=0.1882, simple_loss=0.2715, pruned_loss=0.05244, over 8472.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2882, pruned_loss=0.06206, over 1614098.34 frames. ], batch size: 25, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:36:45,727 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2835, 2.0309, 2.7069, 2.3271, 2.7035, 2.2833, 2.0822, 1.4778], + device='cuda:2'), covar=tensor([0.4970, 0.4819, 0.1888, 0.3402, 0.2333, 0.3039, 0.1868, 0.5184], + device='cuda:2'), in_proj_covar=tensor([0.0944, 0.0978, 0.0805, 0.0942, 0.0997, 0.0895, 0.0748, 0.0828], + 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') +2023-02-07 03:37:08,134 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172231.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:37:16,956 INFO [train.py:901] (2/4) Epoch 22, batch 2500, loss[loss=0.2203, simple_loss=0.2988, pruned_loss=0.07092, over 8499.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2868, pruned_loss=0.0614, over 1611793.11 frames. ], batch size: 26, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:37:26,699 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172256.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:37:36,998 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8529, 1.9946, 1.7809, 2.5600, 1.2001, 1.5751, 1.7590, 1.9471], + device='cuda:2'), covar=tensor([0.0741, 0.0777, 0.0856, 0.0389, 0.1097, 0.1232, 0.0858, 0.0856], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0197, 0.0243, 0.0215, 0.0206, 0.0248, 0.0251, 0.0210], + 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') +2023-02-07 03:37:50,849 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.288e+02 2.722e+02 3.540e+02 9.975e+02, threshold=5.443e+02, percent-clipped=1.0 +2023-02-07 03:37:52,251 INFO [train.py:901] (2/4) Epoch 22, batch 2550, loss[loss=0.204, simple_loss=0.2821, pruned_loss=0.06298, over 7935.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2868, pruned_loss=0.06127, over 1613426.88 frames. ], batch size: 20, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:37:56,720 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172299.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:38:25,418 INFO [train.py:901] (2/4) Epoch 22, batch 2600, loss[loss=0.2084, simple_loss=0.2741, pruned_loss=0.0713, over 7801.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2859, pruned_loss=0.06072, over 1613769.01 frames. ], batch size: 19, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:38:58,399 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.484e+02 3.096e+02 3.957e+02 1.134e+03, threshold=6.191e+02, percent-clipped=6.0 +2023-02-07 03:39:00,473 INFO [train.py:901] (2/4) Epoch 22, batch 2650, loss[loss=0.2734, simple_loss=0.3386, pruned_loss=0.1041, over 6752.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2862, pruned_loss=0.06122, over 1610698.86 frames. ], batch size: 72, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:39:10,806 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4452, 2.8656, 2.1340, 3.9654, 1.7484, 2.1647, 2.2655, 3.0009], + device='cuda:2'), covar=tensor([0.0697, 0.0736, 0.0853, 0.0278, 0.1036, 0.1189, 0.0973, 0.0750], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0197, 0.0244, 0.0216, 0.0206, 0.0248, 0.0251, 0.0210], + 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') +2023-02-07 03:39:16,288 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172414.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:39:35,342 INFO [train.py:901] (2/4) Epoch 22, batch 2700, loss[loss=0.1806, simple_loss=0.2653, pruned_loss=0.04793, over 8140.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2859, pruned_loss=0.06051, over 1613636.16 frames. ], batch size: 22, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:40:02,842 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172480.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:40:09,203 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 03:40:09,437 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.345e+02 2.798e+02 3.767e+02 1.133e+03, threshold=5.596e+02, percent-clipped=4.0 +2023-02-07 03:40:10,850 INFO [train.py:901] (2/4) Epoch 22, batch 2750, loss[loss=0.1958, simple_loss=0.2863, pruned_loss=0.05271, over 8610.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2866, pruned_loss=0.06095, over 1616936.60 frames. ], batch size: 31, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:40:45,666 INFO [train.py:901] (2/4) Epoch 22, batch 2800, loss[loss=0.1799, simple_loss=0.2549, pruned_loss=0.05248, over 7193.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2864, pruned_loss=0.06083, over 1615383.85 frames. ], batch size: 16, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:41:18,227 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.395e+02 2.840e+02 3.614e+02 7.820e+02, threshold=5.680e+02, percent-clipped=6.0 +2023-02-07 03:41:20,389 INFO [train.py:901] (2/4) Epoch 22, batch 2850, loss[loss=0.2315, simple_loss=0.3132, pruned_loss=0.07491, over 8754.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.287, pruned_loss=0.06084, over 1616310.49 frames. ], batch size: 30, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:41:23,208 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172595.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:41:25,863 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2501, 3.1636, 2.9217, 1.5378, 2.8256, 2.9490, 2.8651, 2.6968], + device='cuda:2'), covar=tensor([0.1223, 0.0864, 0.1442, 0.4954, 0.1286, 0.1366, 0.1795, 0.1135], + device='cuda:2'), in_proj_covar=tensor([0.0520, 0.0431, 0.0427, 0.0532, 0.0423, 0.0443, 0.0424, 0.0383], + 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') +2023-02-07 03:41:37,777 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:41:44,843 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 03:41:56,010 INFO [train.py:901] (2/4) Epoch 22, batch 2900, loss[loss=0.218, simple_loss=0.2979, pruned_loss=0.06906, over 7918.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2875, pruned_loss=0.06101, over 1617321.67 frames. ], batch size: 20, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:41:57,554 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:42:15,798 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172670.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:42:24,160 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 03:42:28,900 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.482e+02 2.975e+02 3.907e+02 6.756e+02, threshold=5.949e+02, percent-clipped=4.0 +2023-02-07 03:42:30,295 INFO [train.py:901] (2/4) Epoch 22, batch 2950, loss[loss=0.1659, simple_loss=0.2555, pruned_loss=0.03818, over 7930.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2885, pruned_loss=0.06143, over 1616604.47 frames. ], batch size: 20, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:42:32,546 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172695.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:42:56,537 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4341, 2.3536, 3.1626, 2.5707, 2.9509, 2.4143, 2.2899, 1.8199], + device='cuda:2'), covar=tensor([0.5341, 0.4962, 0.1946, 0.3461, 0.2608, 0.3140, 0.1819, 0.5349], + device='cuda:2'), in_proj_covar=tensor([0.0944, 0.0978, 0.0805, 0.0942, 0.0995, 0.0896, 0.0745, 0.0827], + 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') +2023-02-07 03:43:05,667 INFO [train.py:901] (2/4) Epoch 22, batch 3000, loss[loss=0.1741, simple_loss=0.262, pruned_loss=0.04317, over 7965.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2897, pruned_loss=0.06203, over 1619838.91 frames. ], batch size: 21, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:43:05,667 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 03:43:17,970 INFO [train.py:935] (2/4) Epoch 22, validation: loss=0.1735, simple_loss=0.2739, pruned_loss=0.03659, over 944034.00 frames. +2023-02-07 03:43:17,971 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 03:43:25,641 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172752.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:43:51,438 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.191e+02 2.765e+02 3.574e+02 6.067e+02, threshold=5.530e+02, percent-clipped=1.0 +2023-02-07 03:43:52,749 INFO [train.py:901] (2/4) Epoch 22, batch 3050, loss[loss=0.219, simple_loss=0.294, pruned_loss=0.07202, over 7805.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2895, pruned_loss=0.0626, over 1618588.86 frames. ], batch size: 20, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:44:23,922 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9607, 2.2404, 1.7948, 2.7387, 1.2216, 1.5869, 2.0288, 2.2228], + device='cuda:2'), covar=tensor([0.0737, 0.0800, 0.0884, 0.0369, 0.1153, 0.1312, 0.0811, 0.0729], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0198, 0.0246, 0.0217, 0.0208, 0.0249, 0.0252, 0.0211], + 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') +2023-02-07 03:44:26,386 INFO [train.py:901] (2/4) Epoch 22, batch 3100, loss[loss=0.1868, simple_loss=0.2688, pruned_loss=0.05243, over 8088.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2886, pruned_loss=0.06208, over 1619864.16 frames. ], batch size: 21, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:44:32,707 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172851.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:44:40,608 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172863.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:44:50,781 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:44:59,798 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.451e+02 3.163e+02 4.463e+02 7.617e+02, threshold=6.327e+02, percent-clipped=7.0 +2023-02-07 03:45:01,192 INFO [train.py:901] (2/4) Epoch 22, batch 3150, loss[loss=0.1988, simple_loss=0.2879, pruned_loss=0.05489, over 8318.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2887, pruned_loss=0.06272, over 1615370.39 frames. ], batch size: 25, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:45:33,032 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5883, 1.9463, 2.9155, 1.4852, 2.1558, 1.9509, 1.6291, 2.2153], + device='cuda:2'), covar=tensor([0.1917, 0.2525, 0.0920, 0.4414, 0.1840, 0.3188, 0.2322, 0.2110], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0605, 0.0558, 0.0642, 0.0645, 0.0591, 0.0534, 0.0630], + 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') +2023-02-07 03:45:35,470 INFO [train.py:901] (2/4) Epoch 22, batch 3200, loss[loss=0.1729, simple_loss=0.2673, pruned_loss=0.0393, over 8107.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2877, pruned_loss=0.06167, over 1617285.83 frames. ], batch size: 23, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:45:40,965 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3600, 1.4793, 1.4035, 1.7655, 0.6516, 1.2684, 1.2589, 1.4678], + device='cuda:2'), covar=tensor([0.0929, 0.0849, 0.1101, 0.0548, 0.1272, 0.1449, 0.0844, 0.0812], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0198, 0.0246, 0.0217, 0.0207, 0.0249, 0.0252, 0.0210], + 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') +2023-02-07 03:45:47,741 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172960.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:45:53,322 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1570, 1.4880, 1.6913, 1.4659, 1.0469, 1.4500, 1.9061, 1.6823], + device='cuda:2'), covar=tensor([0.0512, 0.1286, 0.1687, 0.1427, 0.0588, 0.1493, 0.0669, 0.0639], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0099, 0.0163, 0.0111, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 03:46:06,752 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172987.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:46:09,252 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.554e+02 2.964e+02 3.773e+02 6.891e+02, threshold=5.928e+02, percent-clipped=2.0 +2023-02-07 03:46:10,572 INFO [train.py:901] (2/4) Epoch 22, batch 3250, loss[loss=0.2349, simple_loss=0.3129, pruned_loss=0.0785, over 8248.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2876, pruned_loss=0.06188, over 1613009.45 frames. ], batch size: 24, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:46:28,195 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2779, 2.0673, 1.6320, 1.9357, 1.7041, 1.3959, 1.6687, 1.7431], + device='cuda:2'), covar=tensor([0.1346, 0.0425, 0.1226, 0.0530, 0.0728, 0.1572, 0.0990, 0.0861], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0235, 0.0336, 0.0312, 0.0302, 0.0341, 0.0348, 0.0321], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 03:46:45,372 INFO [train.py:901] (2/4) Epoch 22, batch 3300, loss[loss=0.2107, simple_loss=0.2914, pruned_loss=0.06502, over 8340.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2873, pruned_loss=0.06203, over 1611952.74 frames. ], batch size: 26, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:47:07,553 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173075.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:47:17,922 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.506e+02 2.831e+02 3.669e+02 6.075e+02, threshold=5.662e+02, percent-clipped=1.0 +2023-02-07 03:47:18,581 INFO [train.py:901] (2/4) Epoch 22, batch 3350, loss[loss=0.2251, simple_loss=0.3018, pruned_loss=0.07422, over 8043.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2871, pruned_loss=0.06155, over 1612359.96 frames. ], batch size: 22, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:47:22,039 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173096.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:47:26,075 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173102.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:47:54,974 INFO [train.py:901] (2/4) Epoch 22, batch 3400, loss[loss=0.1981, simple_loss=0.2692, pruned_loss=0.06346, over 7933.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.288, pruned_loss=0.06264, over 1612517.45 frames. ], batch size: 20, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:48:12,778 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173168.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:48:21,517 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3750, 1.6316, 4.5663, 1.6790, 4.0362, 3.7663, 4.1487, 3.9926], + device='cuda:2'), covar=tensor([0.0618, 0.4131, 0.0423, 0.3972, 0.1052, 0.1005, 0.0586, 0.0594], + device='cuda:2'), in_proj_covar=tensor([0.0625, 0.0636, 0.0685, 0.0618, 0.0700, 0.0604, 0.0603, 0.0669], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 03:48:28,293 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.505e+02 2.494e+02 3.128e+02 3.771e+02 6.972e+02, threshold=6.255e+02, percent-clipped=4.0 +2023-02-07 03:48:28,960 INFO [train.py:901] (2/4) Epoch 22, batch 3450, loss[loss=0.213, simple_loss=0.2945, pruned_loss=0.06581, over 8205.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2871, pruned_loss=0.06185, over 1615449.39 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:48:39,550 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173207.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:48:42,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173211.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:49:05,729 INFO [train.py:901] (2/4) Epoch 22, batch 3500, loss[loss=0.1926, simple_loss=0.272, pruned_loss=0.05665, over 7812.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2879, pruned_loss=0.06213, over 1611748.60 frames. ], batch size: 20, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:49:24,790 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 03:49:38,937 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.642e+02 3.082e+02 3.788e+02 9.506e+02, threshold=6.164e+02, percent-clipped=4.0 +2023-02-07 03:49:39,646 INFO [train.py:901] (2/4) Epoch 22, batch 3550, loss[loss=0.2091, simple_loss=0.287, pruned_loss=0.06554, over 8132.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2882, pruned_loss=0.06242, over 1609543.26 frames. ], batch size: 22, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:49:50,461 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6322, 4.6304, 4.1319, 2.0269, 4.0572, 4.1991, 4.2293, 4.0289], + device='cuda:2'), covar=tensor([0.0702, 0.0473, 0.1044, 0.4958, 0.0925, 0.0914, 0.1262, 0.0701], + device='cuda:2'), in_proj_covar=tensor([0.0523, 0.0434, 0.0432, 0.0536, 0.0427, 0.0447, 0.0426, 0.0386], + 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') +2023-02-07 03:50:00,040 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173322.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:03,373 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:06,896 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173331.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:10,303 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9630, 1.6451, 1.9663, 1.5526, 1.0077, 1.5940, 2.2711, 2.2757], + device='cuda:2'), covar=tensor([0.0430, 0.1248, 0.1629, 0.1460, 0.0588, 0.1480, 0.0620, 0.0569], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0158, 0.0099, 0.0163, 0.0111, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 03:50:14,764 INFO [train.py:901] (2/4) Epoch 22, batch 3600, loss[loss=0.1675, simple_loss=0.2399, pruned_loss=0.04755, over 7694.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2891, pruned_loss=0.06281, over 1614069.84 frames. ], batch size: 18, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:50:24,996 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173356.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:26,359 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:43,489 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173383.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:48,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.398e+02 3.034e+02 4.459e+02 8.281e+02, threshold=6.068e+02, percent-clipped=7.0 +2023-02-07 03:50:49,307 INFO [train.py:901] (2/4) Epoch 22, batch 3650, loss[loss=0.193, simple_loss=0.2711, pruned_loss=0.05742, over 7651.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2894, pruned_loss=0.06301, over 1612417.11 frames. ], batch size: 19, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:51:23,355 INFO [train.py:901] (2/4) Epoch 22, batch 3700, loss[loss=0.1791, simple_loss=0.259, pruned_loss=0.04954, over 7960.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2886, pruned_loss=0.06218, over 1614517.05 frames. ], batch size: 21, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:51:24,745 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 03:51:42,307 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173467.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:51:51,707 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-07 03:51:57,911 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.519e+02 2.931e+02 3.909e+02 7.363e+02, threshold=5.861e+02, percent-clipped=2.0 +2023-02-07 03:51:58,512 INFO [train.py:901] (2/4) Epoch 22, batch 3750, loss[loss=0.1783, simple_loss=0.2701, pruned_loss=0.04324, over 8268.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2893, pruned_loss=0.06266, over 1615231.68 frames. ], batch size: 24, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:51:58,725 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:52:11,077 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173509.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:52:12,777 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173512.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:52:18,141 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8067, 1.3291, 4.0058, 1.4603, 3.5143, 3.3472, 3.6507, 3.5161], + device='cuda:2'), covar=tensor([0.0667, 0.4642, 0.0596, 0.4094, 0.1259, 0.1012, 0.0653, 0.0724], + device='cuda:2'), in_proj_covar=tensor([0.0628, 0.0640, 0.0689, 0.0621, 0.0702, 0.0607, 0.0605, 0.0674], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 03:52:22,064 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9343, 1.7720, 2.0560, 1.7337, 0.8894, 1.7455, 2.1277, 2.2663], + device='cuda:2'), covar=tensor([0.0413, 0.1209, 0.1531, 0.1301, 0.0563, 0.1428, 0.0617, 0.0515], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0157, 0.0099, 0.0162, 0.0111, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 03:52:23,510 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 03:52:32,200 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6706, 1.5248, 1.8997, 1.5508, 0.9161, 1.6001, 2.0498, 1.8461], + device='cuda:2'), covar=tensor([0.0488, 0.1290, 0.1660, 0.1416, 0.0602, 0.1519, 0.0654, 0.0650], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0157, 0.0099, 0.0162, 0.0111, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 03:52:32,724 INFO [train.py:901] (2/4) Epoch 22, batch 3800, loss[loss=0.2018, simple_loss=0.2896, pruned_loss=0.057, over 8505.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2894, pruned_loss=0.0629, over 1614975.88 frames. ], batch size: 26, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:52:35,962 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.28 vs. limit=5.0 +2023-02-07 03:52:41,039 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8521, 1.6891, 2.4997, 1.6609, 1.3069, 2.5179, 0.5607, 1.5479], + device='cuda:2'), covar=tensor([0.1736, 0.1286, 0.0322, 0.1293, 0.2634, 0.0315, 0.2290, 0.1229], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0198, 0.0127, 0.0220, 0.0267, 0.0135, 0.0170, 0.0192], + 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') +2023-02-07 03:52:58,746 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173578.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:07,968 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.467e+02 3.140e+02 3.842e+02 8.904e+02, threshold=6.281e+02, percent-clipped=2.0 +2023-02-07 03:53:08,695 INFO [train.py:901] (2/4) Epoch 22, batch 3850, loss[loss=0.2099, simple_loss=0.2903, pruned_loss=0.06471, over 6972.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2882, pruned_loss=0.06226, over 1613344.88 frames. ], batch size: 71, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:53:16,529 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173603.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:30,745 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 03:53:33,549 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173627.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:36,894 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:43,567 INFO [train.py:901] (2/4) Epoch 22, batch 3900, loss[loss=0.2308, simple_loss=0.3165, pruned_loss=0.07252, over 8522.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2882, pruned_loss=0.0619, over 1615262.47 frames. ], batch size: 26, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:54:02,780 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8451, 2.2277, 3.5862, 1.9202, 2.0167, 3.5228, 0.9968, 2.0894], + device='cuda:2'), covar=tensor([0.1561, 0.1242, 0.0262, 0.1784, 0.2298, 0.0354, 0.2078, 0.1475], + device='cuda:2'), in_proj_covar=tensor([0.0190, 0.0198, 0.0128, 0.0221, 0.0269, 0.0136, 0.0171, 0.0193], + 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') +2023-02-07 03:54:03,323 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173671.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:54:17,229 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.507e+02 2.945e+02 3.654e+02 8.206e+02, threshold=5.890e+02, percent-clipped=3.0 +2023-02-07 03:54:17,888 INFO [train.py:901] (2/4) Epoch 22, batch 3950, loss[loss=0.1919, simple_loss=0.2822, pruned_loss=0.05074, over 8472.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2889, pruned_loss=0.06226, over 1619387.66 frames. ], batch size: 25, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:54:37,442 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8507, 2.2321, 3.5249, 1.8599, 1.7318, 3.4975, 0.6340, 2.0306], + device='cuda:2'), covar=tensor([0.1181, 0.1163, 0.0221, 0.1683, 0.2653, 0.0292, 0.2377, 0.1358], + device='cuda:2'), in_proj_covar=tensor([0.0188, 0.0196, 0.0127, 0.0219, 0.0266, 0.0135, 0.0170, 0.0192], + 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') +2023-02-07 03:54:52,908 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.12 vs. limit=5.0 +2023-02-07 03:54:53,285 INFO [train.py:901] (2/4) Epoch 22, batch 4000, loss[loss=0.1981, simple_loss=0.2829, pruned_loss=0.05661, over 8092.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2881, pruned_loss=0.06196, over 1618112.21 frames. ], batch size: 21, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:55:23,150 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173786.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:55:26,124 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.329e+02 2.821e+02 3.599e+02 1.045e+03, threshold=5.642e+02, percent-clipped=6.0 +2023-02-07 03:55:26,793 INFO [train.py:901] (2/4) Epoch 22, batch 4050, loss[loss=0.2931, simple_loss=0.3463, pruned_loss=0.1199, over 7144.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2893, pruned_loss=0.06241, over 1620675.45 frames. ], batch size: 71, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:56:02,760 INFO [train.py:901] (2/4) Epoch 22, batch 4100, loss[loss=0.2061, simple_loss=0.287, pruned_loss=0.06264, over 8564.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2892, pruned_loss=0.06244, over 1618438.54 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:56:10,365 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173853.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 03:56:31,255 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173883.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:56:36,362 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.329e+02 2.764e+02 3.605e+02 7.317e+02, threshold=5.528e+02, percent-clipped=2.0 +2023-02-07 03:56:37,019 INFO [train.py:901] (2/4) Epoch 22, batch 4150, loss[loss=0.2161, simple_loss=0.2842, pruned_loss=0.07406, over 7552.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2878, pruned_loss=0.06165, over 1616775.73 frames. ], batch size: 18, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:56:45,820 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5067, 1.5832, 2.0756, 1.2899, 1.2108, 2.0576, 0.3631, 1.2183], + device='cuda:2'), covar=tensor([0.1901, 0.1288, 0.0376, 0.1317, 0.2633, 0.0413, 0.2048, 0.1361], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0196, 0.0127, 0.0220, 0.0267, 0.0135, 0.0171, 0.0191], + 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') +2023-02-07 03:56:47,603 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173908.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:57:12,183 INFO [train.py:901] (2/4) Epoch 22, batch 4200, loss[loss=0.2069, simple_loss=0.2877, pruned_loss=0.06311, over 8525.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2896, pruned_loss=0.06261, over 1619802.01 frames. ], batch size: 28, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:57:28,908 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 03:57:29,769 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173968.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 03:57:35,170 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:57:46,762 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.391e+02 3.056e+02 3.931e+02 9.713e+02, threshold=6.111e+02, percent-clipped=5.0 +2023-02-07 03:57:46,782 INFO [train.py:901] (2/4) Epoch 22, batch 4250, loss[loss=0.1624, simple_loss=0.2335, pruned_loss=0.04568, over 7432.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2884, pruned_loss=0.06192, over 1613429.67 frames. ], batch size: 17, lr: 3.44e-03, grad_scale: 4.0 +2023-02-07 03:57:55,804 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 03:58:04,800 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1636, 4.1155, 3.7537, 1.8663, 3.7140, 3.8016, 3.7368, 3.6034], + device='cuda:2'), covar=tensor([0.0896, 0.0691, 0.1230, 0.5474, 0.0938, 0.1078, 0.1457, 0.0885], + device='cuda:2'), in_proj_covar=tensor([0.0527, 0.0436, 0.0433, 0.0538, 0.0425, 0.0448, 0.0428, 0.0388], + 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') +2023-02-07 03:58:15,727 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174033.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:17,065 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174035.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:21,636 INFO [train.py:901] (2/4) Epoch 22, batch 4300, loss[loss=0.1843, simple_loss=0.2781, pruned_loss=0.04523, over 8145.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2879, pruned_loss=0.06163, over 1607351.12 frames. ], batch size: 22, lr: 3.44e-03, grad_scale: 4.0 +2023-02-07 03:58:21,844 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174042.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:25,981 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 03:58:40,405 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174067.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:56,912 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:57,375 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.427e+02 2.775e+02 3.458e+02 5.995e+02, threshold=5.550e+02, percent-clipped=0.0 +2023-02-07 03:58:57,395 INFO [train.py:901] (2/4) Epoch 22, batch 4350, loss[loss=0.1977, simple_loss=0.2886, pruned_loss=0.0534, over 8483.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.288, pruned_loss=0.06182, over 1606579.11 frames. ], batch size: 29, lr: 3.43e-03, grad_scale: 4.0 +2023-02-07 03:59:25,061 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 03:59:32,547 INFO [train.py:901] (2/4) Epoch 22, batch 4400, loss[loss=0.2068, simple_loss=0.2875, pruned_loss=0.06305, over 8463.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2883, pruned_loss=0.06162, over 1607868.94 frames. ], batch size: 39, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 03:59:36,055 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1800, 2.3705, 1.9188, 2.9649, 1.5032, 1.7737, 2.3263, 2.4454], + device='cuda:2'), covar=tensor([0.0711, 0.0759, 0.0946, 0.0347, 0.0989, 0.1236, 0.0664, 0.0719], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0199, 0.0246, 0.0216, 0.0208, 0.0247, 0.0250, 0.0209], + 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') +2023-02-07 03:59:42,823 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:00:06,455 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 04:00:07,758 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.623e+02 3.065e+02 3.902e+02 1.119e+03, threshold=6.129e+02, percent-clipped=5.0 +2023-02-07 04:00:07,777 INFO [train.py:901] (2/4) Epoch 22, batch 4450, loss[loss=0.1876, simple_loss=0.2739, pruned_loss=0.05068, over 8290.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2885, pruned_loss=0.06229, over 1602088.44 frames. ], batch size: 23, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:00:24,478 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-07 04:00:26,069 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4538, 1.8563, 3.0207, 1.3517, 2.2916, 1.8611, 1.6242, 2.2655], + device='cuda:2'), covar=tensor([0.2014, 0.2635, 0.0769, 0.4741, 0.1764, 0.3350, 0.2337, 0.2158], + device='cuda:2'), in_proj_covar=tensor([0.0528, 0.0605, 0.0559, 0.0645, 0.0646, 0.0591, 0.0536, 0.0630], + 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') +2023-02-07 04:00:30,159 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174224.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:00:41,752 INFO [train.py:901] (2/4) Epoch 22, batch 4500, loss[loss=0.1901, simple_loss=0.2602, pruned_loss=0.05999, over 7673.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2882, pruned_loss=0.06239, over 1604030.64 frames. ], batch size: 19, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:00:46,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174249.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 04:00:56,983 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 04:01:17,041 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.524e+02 3.306e+02 4.354e+02 7.569e+02, threshold=6.612e+02, percent-clipped=6.0 +2023-02-07 04:01:17,062 INFO [train.py:901] (2/4) Epoch 22, batch 4550, loss[loss=0.233, simple_loss=0.3072, pruned_loss=0.07939, over 8085.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2884, pruned_loss=0.06237, over 1607368.28 frames. ], batch size: 21, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:01:23,306 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:01:28,571 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:01:44,047 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-07 04:01:51,053 INFO [train.py:901] (2/4) Epoch 22, batch 4600, loss[loss=0.2718, simple_loss=0.336, pruned_loss=0.1038, over 7258.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.289, pruned_loss=0.06279, over 1609647.87 frames. ], batch size: 72, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:01:54,751 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174347.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:12,035 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174372.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:15,444 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:16,892 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174379.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:25,978 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.375e+02 2.973e+02 3.873e+02 1.031e+03, threshold=5.946e+02, percent-clipped=3.0 +2023-02-07 04:02:25,999 INFO [train.py:901] (2/4) Epoch 22, batch 4650, loss[loss=0.2122, simple_loss=0.2797, pruned_loss=0.07233, over 7715.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2879, pruned_loss=0.06194, over 1612098.15 frames. ], batch size: 18, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:02:31,945 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 04:02:37,939 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:03:02,511 INFO [train.py:901] (2/4) Epoch 22, batch 4700, loss[loss=0.2274, simple_loss=0.3104, pruned_loss=0.07221, over 8509.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2876, pruned_loss=0.06173, over 1614865.39 frames. ], batch size: 49, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:03:21,327 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4608, 2.6551, 3.1384, 1.7709, 3.3068, 2.1522, 1.6700, 2.4090], + device='cuda:2'), covar=tensor([0.0763, 0.0367, 0.0335, 0.0753, 0.0380, 0.0768, 0.0924, 0.0528], + device='cuda:2'), in_proj_covar=tensor([0.0449, 0.0387, 0.0343, 0.0442, 0.0372, 0.0528, 0.0385, 0.0415], + device='cuda:2'), out_proj_covar=tensor([1.2022e-04, 1.0130e-04, 9.0373e-05, 1.1635e-04, 9.7755e-05, 1.4885e-04, + 1.0387e-04, 1.0982e-04], device='cuda:2') +2023-02-07 04:03:37,052 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.374e+02 2.923e+02 3.899e+02 9.329e+02, threshold=5.846e+02, percent-clipped=2.0 +2023-02-07 04:03:37,072 INFO [train.py:901] (2/4) Epoch 22, batch 4750, loss[loss=0.2375, simple_loss=0.3213, pruned_loss=0.07687, over 8457.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2865, pruned_loss=0.06117, over 1613960.11 frames. ], batch size: 25, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:03:37,251 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:03:38,612 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174494.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:03:43,329 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174501.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:04:04,458 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 04:04:06,510 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 04:04:12,369 INFO [train.py:901] (2/4) Epoch 22, batch 4800, loss[loss=0.2017, simple_loss=0.2911, pruned_loss=0.05616, over 8460.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2866, pruned_loss=0.06093, over 1612592.33 frames. ], batch size: 25, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:04:46,102 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.398e+02 2.995e+02 3.860e+02 8.125e+02, threshold=5.990e+02, percent-clipped=3.0 +2023-02-07 04:04:46,122 INFO [train.py:901] (2/4) Epoch 22, batch 4850, loss[loss=0.227, simple_loss=0.3111, pruned_loss=0.07148, over 8335.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2873, pruned_loss=0.0613, over 1614633.75 frames. ], batch size: 26, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:04:55,435 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 04:05:02,324 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:05:20,310 INFO [train.py:901] (2/4) Epoch 22, batch 4900, loss[loss=0.2146, simple_loss=0.3027, pruned_loss=0.0633, over 8133.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.288, pruned_loss=0.06187, over 1615630.62 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:05:23,131 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174645.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:05:29,310 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:05:56,489 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.582e+02 3.121e+02 3.821e+02 7.682e+02, threshold=6.242e+02, percent-clipped=2.0 +2023-02-07 04:05:56,510 INFO [train.py:901] (2/4) Epoch 22, batch 4950, loss[loss=0.1967, simple_loss=0.2815, pruned_loss=0.05596, over 8511.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2873, pruned_loss=0.0614, over 1613342.03 frames. ], batch size: 28, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:06:17,800 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:21,278 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3650, 2.0185, 1.6710, 2.0627, 1.7406, 1.2574, 1.7365, 1.8431], + device='cuda:2'), covar=tensor([0.1144, 0.0441, 0.1279, 0.0424, 0.0768, 0.1683, 0.0900, 0.0611], + device='cuda:2'), in_proj_covar=tensor([0.0350, 0.0231, 0.0331, 0.0305, 0.0297, 0.0336, 0.0339, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:06:27,422 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0723, 2.1023, 1.8469, 2.5290, 1.4454, 1.7466, 2.0214, 2.1497], + device='cuda:2'), covar=tensor([0.0625, 0.0732, 0.0807, 0.0444, 0.0979, 0.1047, 0.0715, 0.0668], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0198, 0.0245, 0.0216, 0.0208, 0.0248, 0.0250, 0.0210], + 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') +2023-02-07 04:06:30,654 INFO [train.py:901] (2/4) Epoch 22, batch 5000, loss[loss=0.2502, simple_loss=0.3252, pruned_loss=0.08756, over 7130.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2886, pruned_loss=0.06195, over 1614575.28 frames. ], batch size: 71, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:06:35,017 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174748.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:36,250 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:36,400 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:44,694 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174760.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:51,025 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:54,585 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174773.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:55,981 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174775.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:07:07,682 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.477e+02 2.928e+02 3.612e+02 7.754e+02, threshold=5.856e+02, percent-clipped=3.0 +2023-02-07 04:07:07,702 INFO [train.py:901] (2/4) Epoch 22, batch 5050, loss[loss=0.2168, simple_loss=0.307, pruned_loss=0.06334, over 8514.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2871, pruned_loss=0.06121, over 1616354.44 frames. ], batch size: 28, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:07:16,958 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1869, 1.3053, 1.5708, 1.2247, 0.7719, 1.3379, 1.2467, 1.0475], + device='cuda:2'), covar=tensor([0.0619, 0.1293, 0.1633, 0.1451, 0.0582, 0.1509, 0.0697, 0.0711], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0159, 0.0100, 0.0164, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 04:07:27,836 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4252, 2.3696, 3.2085, 2.5781, 3.0171, 2.5281, 2.2470, 1.7638], + device='cuda:2'), covar=tensor([0.5706, 0.4896, 0.1995, 0.3668, 0.2551, 0.2982, 0.1824, 0.5562], + device='cuda:2'), in_proj_covar=tensor([0.0935, 0.0975, 0.0800, 0.0941, 0.0989, 0.0888, 0.0745, 0.0821], + 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') +2023-02-07 04:07:34,469 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 04:07:42,673 INFO [train.py:901] (2/4) Epoch 22, batch 5100, loss[loss=0.1943, simple_loss=0.2775, pruned_loss=0.05556, over 8293.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2872, pruned_loss=0.06151, over 1616203.53 frames. ], batch size: 23, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:07:58,167 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:08:03,634 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:08:17,810 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.498e+02 3.130e+02 3.757e+02 7.363e+02, threshold=6.259e+02, percent-clipped=3.0 +2023-02-07 04:08:17,830 INFO [train.py:901] (2/4) Epoch 22, batch 5150, loss[loss=0.19, simple_loss=0.2638, pruned_loss=0.05811, over 7660.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2875, pruned_loss=0.06248, over 1610316.82 frames. ], batch size: 19, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:08:21,138 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174897.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:08:52,065 INFO [train.py:901] (2/4) Epoch 22, batch 5200, loss[loss=0.2162, simple_loss=0.2929, pruned_loss=0.06972, over 8507.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2884, pruned_loss=0.06257, over 1610843.38 frames. ], batch size: 26, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:08:52,925 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1250, 1.3867, 4.2866, 1.5795, 3.7814, 3.5479, 3.9165, 3.7509], + device='cuda:2'), covar=tensor([0.0683, 0.4953, 0.0550, 0.4426, 0.1284, 0.1044, 0.0650, 0.0785], + device='cuda:2'), in_proj_covar=tensor([0.0628, 0.0646, 0.0695, 0.0627, 0.0708, 0.0603, 0.0606, 0.0677], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 04:08:58,259 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2067, 4.1683, 3.7670, 1.8584, 3.6530, 3.7678, 3.8598, 3.5793], + device='cuda:2'), covar=tensor([0.0783, 0.0552, 0.1104, 0.4847, 0.0957, 0.1036, 0.1158, 0.0879], + device='cuda:2'), in_proj_covar=tensor([0.0522, 0.0433, 0.0432, 0.0536, 0.0424, 0.0444, 0.0424, 0.0387], + 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') +2023-02-07 04:09:26,992 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.432e+02 2.854e+02 3.739e+02 7.258e+02, threshold=5.708e+02, percent-clipped=1.0 +2023-02-07 04:09:27,012 INFO [train.py:901] (2/4) Epoch 22, batch 5250, loss[loss=0.2411, simple_loss=0.3145, pruned_loss=0.08384, over 8503.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2889, pruned_loss=0.06288, over 1605734.27 frames. ], batch size: 26, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:09:31,835 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 04:09:44,061 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175016.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:09:49,559 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175024.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:01,590 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175041.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:01,992 INFO [train.py:901] (2/4) Epoch 22, batch 5300, loss[loss=0.24, simple_loss=0.3086, pruned_loss=0.08566, over 8314.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2896, pruned_loss=0.06355, over 1611778.97 frames. ], batch size: 25, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:10:07,100 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175049.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:19,114 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175067.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:35,754 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.519e+02 3.149e+02 3.909e+02 1.075e+03, threshold=6.297e+02, percent-clipped=6.0 +2023-02-07 04:10:35,774 INFO [train.py:901] (2/4) Epoch 22, batch 5350, loss[loss=0.1874, simple_loss=0.2679, pruned_loss=0.05343, over 8025.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2888, pruned_loss=0.06286, over 1615406.97 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:10:57,896 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:11,826 INFO [train.py:901] (2/4) Epoch 22, batch 5400, loss[loss=0.1754, simple_loss=0.2532, pruned_loss=0.0488, over 7201.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2893, pruned_loss=0.06291, over 1617519.25 frames. ], batch size: 16, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:11:13,265 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2653, 3.1636, 2.9301, 1.5968, 2.8698, 2.8823, 2.8748, 2.8385], + device='cuda:2'), covar=tensor([0.1168, 0.0848, 0.1437, 0.4503, 0.1119, 0.1260, 0.1663, 0.1024], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0432, 0.0430, 0.0533, 0.0422, 0.0442, 0.0422, 0.0384], + 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') +2023-02-07 04:11:14,649 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175146.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:22,726 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:39,705 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175182.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:46,268 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.426e+02 2.833e+02 4.034e+02 1.686e+03, threshold=5.665e+02, percent-clipped=5.0 +2023-02-07 04:11:46,288 INFO [train.py:901] (2/4) Epoch 22, batch 5450, loss[loss=0.149, simple_loss=0.2283, pruned_loss=0.03484, over 7694.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2876, pruned_loss=0.06199, over 1613042.74 frames. ], batch size: 18, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:12:09,775 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175225.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:12:20,248 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 04:12:22,329 INFO [train.py:901] (2/4) Epoch 22, batch 5500, loss[loss=0.2011, simple_loss=0.2867, pruned_loss=0.05774, over 7239.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2854, pruned_loss=0.06077, over 1613037.27 frames. ], batch size: 16, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:12:40,057 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 04:12:56,619 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.376e+02 2.848e+02 3.508e+02 8.289e+02, threshold=5.697e+02, percent-clipped=6.0 +2023-02-07 04:12:56,639 INFO [train.py:901] (2/4) Epoch 22, batch 5550, loss[loss=0.2108, simple_loss=0.3004, pruned_loss=0.06062, over 8505.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2862, pruned_loss=0.06092, over 1615181.43 frames. ], batch size: 26, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:13:31,809 INFO [train.py:901] (2/4) Epoch 22, batch 5600, loss[loss=0.2097, simple_loss=0.2985, pruned_loss=0.06047, over 8559.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2856, pruned_loss=0.0605, over 1612541.88 frames. ], batch size: 34, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:13:46,081 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8241, 2.3045, 3.9167, 1.6599, 2.7860, 2.3698, 1.8079, 2.7687], + device='cuda:2'), covar=tensor([0.1798, 0.2480, 0.1018, 0.4376, 0.1919, 0.3049, 0.2387, 0.2601], + device='cuda:2'), in_proj_covar=tensor([0.0525, 0.0604, 0.0554, 0.0641, 0.0645, 0.0589, 0.0535, 0.0629], + 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') +2023-02-07 04:13:49,365 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9755, 1.8769, 2.5511, 1.5059, 1.5212, 2.5318, 0.5354, 1.5890], + device='cuda:2'), covar=tensor([0.1662, 0.1260, 0.0333, 0.1356, 0.2409, 0.0356, 0.2004, 0.1261], + device='cuda:2'), in_proj_covar=tensor([0.0191, 0.0197, 0.0129, 0.0220, 0.0267, 0.0137, 0.0168, 0.0194], + 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') +2023-02-07 04:13:50,743 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.30 vs. limit=5.0 +2023-02-07 04:14:03,188 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5612, 2.9099, 2.3279, 3.9589, 1.7090, 2.1611, 2.4850, 2.9671], + device='cuda:2'), covar=tensor([0.0656, 0.0775, 0.0826, 0.0217, 0.1108, 0.1134, 0.0902, 0.0783], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0198, 0.0247, 0.0216, 0.0208, 0.0247, 0.0252, 0.0209], + 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') +2023-02-07 04:14:03,829 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:06,408 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.532e+02 3.141e+02 4.135e+02 1.836e+03, threshold=6.283e+02, percent-clipped=10.0 +2023-02-07 04:14:06,428 INFO [train.py:901] (2/4) Epoch 22, batch 5650, loss[loss=0.1708, simple_loss=0.2523, pruned_loss=0.04467, over 8092.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2852, pruned_loss=0.06, over 1612197.36 frames. ], batch size: 21, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:14:09,955 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7511, 2.1491, 3.2733, 1.5229, 2.4357, 2.0008, 1.8586, 2.3541], + device='cuda:2'), covar=tensor([0.1880, 0.2272, 0.0800, 0.4442, 0.1871, 0.3263, 0.2220, 0.2323], + device='cuda:2'), in_proj_covar=tensor([0.0525, 0.0604, 0.0553, 0.0640, 0.0645, 0.0589, 0.0535, 0.0628], + 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') +2023-02-07 04:14:20,095 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175412.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:22,660 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 04:14:37,557 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175438.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:40,743 INFO [train.py:901] (2/4) Epoch 22, batch 5700, loss[loss=0.1839, simple_loss=0.2587, pruned_loss=0.05456, over 7815.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2869, pruned_loss=0.06091, over 1611391.79 frames. ], batch size: 19, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:14:56,174 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175463.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:15:15,527 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.452e+02 2.878e+02 3.661e+02 5.836e+02, threshold=5.755e+02, percent-clipped=0.0 +2023-02-07 04:15:15,548 INFO [train.py:901] (2/4) Epoch 22, batch 5750, loss[loss=0.2087, simple_loss=0.2966, pruned_loss=0.06038, over 8083.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2869, pruned_loss=0.06099, over 1613815.02 frames. ], batch size: 21, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:15:20,430 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:15:22,437 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175501.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:15:27,206 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 04:15:50,255 INFO [train.py:901] (2/4) Epoch 22, batch 5800, loss[loss=0.1957, simple_loss=0.2787, pruned_loss=0.0564, over 8465.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2867, pruned_loss=0.06113, over 1609898.91 frames. ], batch size: 25, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:16:09,022 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175569.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:16:25,902 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.280e+02 2.739e+02 3.457e+02 6.413e+02, threshold=5.479e+02, percent-clipped=3.0 +2023-02-07 04:16:25,922 INFO [train.py:901] (2/4) Epoch 22, batch 5850, loss[loss=0.2184, simple_loss=0.2946, pruned_loss=0.0711, over 7632.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2865, pruned_loss=0.06072, over 1610913.13 frames. ], batch size: 19, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:16:42,242 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:17:00,355 INFO [train.py:901] (2/4) Epoch 22, batch 5900, loss[loss=0.2153, simple_loss=0.2922, pruned_loss=0.06923, over 7973.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2866, pruned_loss=0.06075, over 1613926.18 frames. ], batch size: 21, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:17:29,299 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175684.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:17:35,264 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.520e+02 3.043e+02 3.699e+02 9.671e+02, threshold=6.086e+02, percent-clipped=7.0 +2023-02-07 04:17:35,284 INFO [train.py:901] (2/4) Epoch 22, batch 5950, loss[loss=0.1639, simple_loss=0.2572, pruned_loss=0.03526, over 7916.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.287, pruned_loss=0.06121, over 1609969.80 frames. ], batch size: 20, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:18:00,065 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175728.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:18:02,746 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175732.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:18:09,579 INFO [train.py:901] (2/4) Epoch 22, batch 6000, loss[loss=0.2631, simple_loss=0.3273, pruned_loss=0.09943, over 6491.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2868, pruned_loss=0.06139, over 1603829.89 frames. ], batch size: 71, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:18:09,579 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 04:18:21,635 INFO [train.py:935] (2/4) Epoch 22, validation: loss=0.1729, simple_loss=0.2732, pruned_loss=0.03632, over 944034.00 frames. +2023-02-07 04:18:21,636 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 04:18:24,899 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 04:18:31,486 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175756.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:18:34,523 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.94 vs. limit=5.0 +2023-02-07 04:18:41,928 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 +2023-02-07 04:18:56,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.478e+02 2.934e+02 3.623e+02 7.032e+02, threshold=5.869e+02, percent-clipped=2.0 +2023-02-07 04:18:56,237 INFO [train.py:901] (2/4) Epoch 22, batch 6050, loss[loss=0.2285, simple_loss=0.3115, pruned_loss=0.07273, over 8496.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2869, pruned_loss=0.06111, over 1607076.52 frames. ], batch size: 26, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:19:06,555 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1857, 2.3657, 1.8567, 2.8388, 1.4857, 1.7611, 2.1106, 2.4146], + device='cuda:2'), covar=tensor([0.0671, 0.0698, 0.0922, 0.0387, 0.1042, 0.1212, 0.0780, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0197, 0.0246, 0.0216, 0.0206, 0.0246, 0.0251, 0.0208], + 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') +2023-02-07 04:19:31,868 INFO [train.py:901] (2/4) Epoch 22, batch 6100, loss[loss=0.2111, simple_loss=0.2901, pruned_loss=0.06602, over 8195.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2872, pruned_loss=0.06128, over 1612281.92 frames. ], batch size: 23, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:19:32,679 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:35,610 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175847.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:51,976 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:52,698 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:56,544 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 04:20:07,208 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.365e+02 2.974e+02 3.880e+02 6.577e+02, threshold=5.949e+02, percent-clipped=2.0 +2023-02-07 04:20:07,228 INFO [train.py:901] (2/4) Epoch 22, batch 6150, loss[loss=0.2246, simple_loss=0.3139, pruned_loss=0.06763, over 8727.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2862, pruned_loss=0.06058, over 1612861.82 frames. ], batch size: 34, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:20:10,663 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175897.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:11,991 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175899.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:20,143 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9889, 1.6984, 2.0319, 1.8260, 1.9801, 2.0271, 1.8834, 0.7973], + device='cuda:2'), covar=tensor([0.5632, 0.4784, 0.1943, 0.3467, 0.2457, 0.3082, 0.1853, 0.5252], + device='cuda:2'), in_proj_covar=tensor([0.0943, 0.0981, 0.0806, 0.0946, 0.0997, 0.0896, 0.0748, 0.0826], + 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') +2023-02-07 04:20:40,360 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175940.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:41,476 INFO [train.py:901] (2/4) Epoch 22, batch 6200, loss[loss=0.2388, simple_loss=0.3247, pruned_loss=0.07641, over 8214.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2858, pruned_loss=0.06038, over 1613420.37 frames. ], batch size: 23, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:20:53,613 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175958.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:56,911 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175963.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:58,214 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175965.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:21:06,754 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0040, 1.5709, 1.4419, 1.5511, 1.3083, 1.2639, 1.2976, 1.2527], + device='cuda:2'), covar=tensor([0.1157, 0.0482, 0.1241, 0.0588, 0.0789, 0.1579, 0.0939, 0.0843], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0233, 0.0335, 0.0310, 0.0299, 0.0340, 0.0345, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:21:15,640 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.329e+02 2.882e+02 3.634e+02 1.217e+03, threshold=5.765e+02, percent-clipped=6.0 +2023-02-07 04:21:15,660 INFO [train.py:901] (2/4) Epoch 22, batch 6250, loss[loss=0.183, simple_loss=0.2504, pruned_loss=0.0578, over 6842.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2868, pruned_loss=0.06135, over 1613020.01 frames. ], batch size: 15, lr: 3.42e-03, grad_scale: 16.0 +2023-02-07 04:21:51,339 INFO [train.py:901] (2/4) Epoch 22, batch 6300, loss[loss=0.1896, simple_loss=0.2663, pruned_loss=0.05641, over 8133.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2868, pruned_loss=0.06208, over 1609781.13 frames. ], batch size: 22, lr: 3.42e-03, grad_scale: 16.0 +2023-02-07 04:22:12,729 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176072.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:22:26,686 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.297e+02 2.795e+02 3.577e+02 6.374e+02, threshold=5.590e+02, percent-clipped=1.0 +2023-02-07 04:22:26,706 INFO [train.py:901] (2/4) Epoch 22, batch 6350, loss[loss=0.1741, simple_loss=0.2654, pruned_loss=0.0414, over 7938.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2872, pruned_loss=0.06231, over 1610769.91 frames. ], batch size: 20, lr: 3.42e-03, grad_scale: 16.0 +2023-02-07 04:22:34,428 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176103.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:22:51,216 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176127.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:22:51,825 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176128.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:01,397 INFO [train.py:901] (2/4) Epoch 22, batch 6400, loss[loss=0.2073, simple_loss=0.2826, pruned_loss=0.06601, over 7933.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2869, pruned_loss=0.06235, over 1610845.96 frames. ], batch size: 20, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:23:08,437 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176152.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:33,466 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176187.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:36,645 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.375e+02 2.869e+02 3.334e+02 7.002e+02, threshold=5.738e+02, percent-clipped=1.0 +2023-02-07 04:23:36,665 INFO [train.py:901] (2/4) Epoch 22, batch 6450, loss[loss=0.157, simple_loss=0.2337, pruned_loss=0.04012, over 7408.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2858, pruned_loss=0.06163, over 1612201.37 frames. ], batch size: 17, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:23:52,614 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:54,751 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 +2023-02-07 04:24:05,443 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-07 04:24:09,914 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:11,679 INFO [train.py:901] (2/4) Epoch 22, batch 6500, loss[loss=0.2302, simple_loss=0.3094, pruned_loss=0.0755, over 8332.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2866, pruned_loss=0.06126, over 1615306.47 frames. ], batch size: 26, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:24:12,448 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:23,897 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176260.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:27,368 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2391, 3.6149, 2.3204, 2.8745, 2.6680, 2.0007, 2.7886, 3.0380], + device='cuda:2'), covar=tensor([0.1637, 0.0342, 0.1213, 0.0752, 0.0827, 0.1539, 0.1050, 0.1157], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0234, 0.0335, 0.0311, 0.0300, 0.0342, 0.0347, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:24:32,079 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6300, 2.5389, 1.7838, 2.2364, 2.1110, 1.5358, 1.9901, 2.1172], + device='cuda:2'), covar=tensor([0.1453, 0.0389, 0.1282, 0.0680, 0.0732, 0.1588, 0.1051, 0.1012], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0233, 0.0334, 0.0311, 0.0300, 0.0341, 0.0347, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:24:34,063 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176275.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:45,633 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.325e+02 2.725e+02 3.404e+02 5.159e+02, threshold=5.450e+02, percent-clipped=0.0 +2023-02-07 04:24:45,653 INFO [train.py:901] (2/4) Epoch 22, batch 6550, loss[loss=0.1841, simple_loss=0.2564, pruned_loss=0.05595, over 7543.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2855, pruned_loss=0.06085, over 1611348.00 frames. ], batch size: 18, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:24:55,361 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5792, 1.9661, 3.2891, 1.4628, 2.5122, 2.0744, 1.6334, 2.5062], + device='cuda:2'), covar=tensor([0.1816, 0.2503, 0.0789, 0.4345, 0.1706, 0.2886, 0.2321, 0.2033], + device='cuda:2'), in_proj_covar=tensor([0.0528, 0.0608, 0.0558, 0.0646, 0.0650, 0.0596, 0.0540, 0.0635], + 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') +2023-02-07 04:24:57,258 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176307.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:25:04,186 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3025, 2.0446, 2.6358, 2.2648, 2.5983, 2.2313, 2.1528, 1.8972], + device='cuda:2'), covar=tensor([0.3778, 0.4139, 0.1614, 0.2894, 0.1813, 0.2669, 0.1576, 0.3748], + device='cuda:2'), in_proj_covar=tensor([0.0941, 0.0979, 0.0806, 0.0945, 0.0997, 0.0896, 0.0749, 0.0828], + 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') +2023-02-07 04:25:09,377 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 04:25:21,086 INFO [train.py:901] (2/4) Epoch 22, batch 6600, loss[loss=0.1985, simple_loss=0.2828, pruned_loss=0.05711, over 8244.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2869, pruned_loss=0.06168, over 1612997.30 frames. ], batch size: 24, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:25:29,277 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 04:25:32,753 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:25:55,379 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.481e+02 2.510e+02 3.110e+02 4.060e+02 7.968e+02, threshold=6.221e+02, percent-clipped=4.0 +2023-02-07 04:25:55,400 INFO [train.py:901] (2/4) Epoch 22, batch 6650, loss[loss=0.2389, simple_loss=0.3164, pruned_loss=0.08069, over 8566.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2885, pruned_loss=0.06321, over 1608602.04 frames. ], batch size: 31, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:25:56,917 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3155, 2.8264, 2.4491, 4.0561, 1.8815, 1.9168, 2.6520, 3.0101], + device='cuda:2'), covar=tensor([0.0944, 0.0893, 0.1024, 0.0264, 0.1090, 0.1370, 0.0924, 0.0794], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0243, 0.0214, 0.0205, 0.0245, 0.0249, 0.0206], + 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') +2023-02-07 04:26:17,168 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176422.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:26:31,229 INFO [train.py:901] (2/4) Epoch 22, batch 6700, loss[loss=0.2237, simple_loss=0.3024, pruned_loss=0.07253, over 8646.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2875, pruned_loss=0.06254, over 1607202.47 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:26:32,108 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176443.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:26:49,608 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176468.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:27:00,366 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176484.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:27:05,583 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.672e+02 3.290e+02 4.002e+02 8.131e+02, threshold=6.579e+02, percent-clipped=6.0 +2023-02-07 04:27:05,603 INFO [train.py:901] (2/4) Epoch 22, batch 6750, loss[loss=0.2215, simple_loss=0.3027, pruned_loss=0.07021, over 8549.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2859, pruned_loss=0.06214, over 1607106.11 frames. ], batch size: 49, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:27:41,476 INFO [train.py:901] (2/4) Epoch 22, batch 6800, loss[loss=0.1766, simple_loss=0.2508, pruned_loss=0.0512, over 7720.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2864, pruned_loss=0.06225, over 1609980.82 frames. ], batch size: 18, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:27:45,668 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 04:28:10,347 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-07 04:28:16,786 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.317e+02 3.026e+02 3.783e+02 8.757e+02, threshold=6.052e+02, percent-clipped=1.0 +2023-02-07 04:28:16,806 INFO [train.py:901] (2/4) Epoch 22, batch 6850, loss[loss=0.2103, simple_loss=0.2955, pruned_loss=0.06256, over 8478.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2869, pruned_loss=0.06199, over 1610278.88 frames. ], batch size: 49, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:28:24,731 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176604.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:31,630 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176614.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:34,719 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 04:28:34,769 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176619.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:38,965 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8943, 2.2940, 4.2568, 1.6141, 3.1299, 2.4623, 1.8987, 3.0397], + device='cuda:2'), covar=tensor([0.1806, 0.2611, 0.0753, 0.4427, 0.1725, 0.2930, 0.2226, 0.2255], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0605, 0.0556, 0.0643, 0.0648, 0.0594, 0.0537, 0.0632], + 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') +2023-02-07 04:28:48,427 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176639.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:50,268 INFO [train.py:901] (2/4) Epoch 22, batch 6900, loss[loss=0.2151, simple_loss=0.2981, pruned_loss=0.06605, over 8252.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2879, pruned_loss=0.06261, over 1609754.25 frames. ], batch size: 24, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:28:51,036 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:17,272 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176678.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:20,580 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176683.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:26,746 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.293e+02 2.439e+02 3.078e+02 3.806e+02 5.995e+02, threshold=6.157e+02, percent-clipped=0.0 +2023-02-07 04:29:26,766 INFO [train.py:901] (2/4) Epoch 22, batch 6950, loss[loss=0.2114, simple_loss=0.2925, pruned_loss=0.06513, over 8454.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2877, pruned_loss=0.06202, over 1612434.42 frames. ], batch size: 27, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:29:35,481 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176703.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:44,284 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 04:29:46,554 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176719.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:48,146 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-07 04:29:56,844 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176734.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:30:02,046 INFO [train.py:901] (2/4) Epoch 22, batch 7000, loss[loss=0.1977, simple_loss=0.2866, pruned_loss=0.0544, over 7930.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.289, pruned_loss=0.06298, over 1611590.18 frames. ], batch size: 20, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:30:03,961 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-07 04:30:37,809 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.400e+02 2.923e+02 3.703e+02 8.900e+02, threshold=5.847e+02, percent-clipped=5.0 +2023-02-07 04:30:37,829 INFO [train.py:901] (2/4) Epoch 22, batch 7050, loss[loss=0.194, simple_loss=0.2812, pruned_loss=0.05339, over 8486.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2892, pruned_loss=0.06274, over 1613763.23 frames. ], batch size: 28, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:31:03,290 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176828.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:31:11,270 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0807, 3.5499, 2.3039, 2.8101, 2.7646, 2.0070, 2.6772, 2.9951], + device='cuda:2'), covar=tensor([0.1624, 0.0338, 0.1232, 0.0757, 0.0771, 0.1526, 0.1092, 0.1074], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0233, 0.0337, 0.0312, 0.0300, 0.0344, 0.0347, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:31:12,370 INFO [train.py:901] (2/4) Epoch 22, batch 7100, loss[loss=0.2189, simple_loss=0.3079, pruned_loss=0.06492, over 8608.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2896, pruned_loss=0.06301, over 1615322.33 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:31:31,794 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:31:46,089 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.316e+02 2.836e+02 3.633e+02 7.093e+02, threshold=5.673e+02, percent-clipped=3.0 +2023-02-07 04:31:46,109 INFO [train.py:901] (2/4) Epoch 22, batch 7150, loss[loss=0.2673, simple_loss=0.337, pruned_loss=0.09883, over 7654.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.29, pruned_loss=0.06325, over 1614867.05 frames. ], batch size: 19, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:31:51,640 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 04:32:22,266 INFO [train.py:901] (2/4) Epoch 22, batch 7200, loss[loss=0.1943, simple_loss=0.2785, pruned_loss=0.05501, over 8285.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2886, pruned_loss=0.06276, over 1617339.06 frames. ], batch size: 23, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:32:23,127 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176943.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:44,916 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176975.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:52,859 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176987.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:54,980 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176990.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:56,137 INFO [train.py:901] (2/4) Epoch 22, batch 7250, loss[loss=0.1744, simple_loss=0.2448, pruned_loss=0.05201, over 7445.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2886, pruned_loss=0.06247, over 1620654.07 frames. ], batch size: 17, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:32:56,793 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.387e+02 2.385e+02 2.852e+02 3.441e+02 7.839e+02, threshold=5.703e+02, percent-clipped=2.0 +2023-02-07 04:33:02,406 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177000.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:33:10,525 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6722, 4.6192, 4.1939, 2.0658, 4.1466, 4.2658, 4.1026, 4.0799], + device='cuda:2'), covar=tensor([0.0638, 0.0498, 0.0932, 0.4748, 0.0850, 0.0850, 0.1263, 0.0785], + device='cuda:2'), in_proj_covar=tensor([0.0528, 0.0438, 0.0432, 0.0542, 0.0427, 0.0448, 0.0428, 0.0388], + 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') +2023-02-07 04:33:14,105 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177015.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:33:15,421 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3004, 1.5175, 4.5126, 2.3136, 2.4151, 5.0563, 5.1245, 4.4045], + device='cuda:2'), covar=tensor([0.1131, 0.1850, 0.0273, 0.1643, 0.1225, 0.0182, 0.0507, 0.0538], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0319, 0.0284, 0.0313, 0.0309, 0.0265, 0.0418, 0.0303], + 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') +2023-02-07 04:33:21,898 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177027.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:33:31,972 INFO [train.py:901] (2/4) Epoch 22, batch 7300, loss[loss=0.1635, simple_loss=0.2352, pruned_loss=0.04591, over 7434.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06201, over 1614722.10 frames. ], batch size: 17, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:34:06,482 INFO [train.py:901] (2/4) Epoch 22, batch 7350, loss[loss=0.202, simple_loss=0.2907, pruned_loss=0.0567, over 8350.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2868, pruned_loss=0.0617, over 1611633.53 frames. ], batch size: 24, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:34:07,154 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.532e+02 3.310e+02 4.342e+02 9.656e+02, threshold=6.621e+02, percent-clipped=7.0 +2023-02-07 04:34:13,471 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177102.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:26,102 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 04:34:31,738 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177127.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:33,086 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177129.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:42,494 INFO [train.py:901] (2/4) Epoch 22, batch 7400, loss[loss=0.1817, simple_loss=0.2716, pruned_loss=0.04593, over 7814.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2853, pruned_loss=0.06081, over 1609399.89 frames. ], batch size: 20, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:34:42,675 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177142.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:42,687 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4090, 2.8185, 2.3209, 4.0480, 1.7237, 2.0914, 2.4673, 2.8758], + device='cuda:2'), covar=tensor([0.0763, 0.0801, 0.0867, 0.0212, 0.1168, 0.1299, 0.1046, 0.0776], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0243, 0.0212, 0.0206, 0.0245, 0.0248, 0.0206], + 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') +2023-02-07 04:34:47,987 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 04:35:09,031 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6990, 5.7123, 5.1220, 2.4716, 5.0860, 5.4656, 5.3356, 5.4150], + device='cuda:2'), covar=tensor([0.0612, 0.0481, 0.0924, 0.4701, 0.0728, 0.0777, 0.1067, 0.0530], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0440, 0.0433, 0.0542, 0.0428, 0.0449, 0.0429, 0.0389], + 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') +2023-02-07 04:35:16,507 INFO [train.py:901] (2/4) Epoch 22, batch 7450, loss[loss=0.1799, simple_loss=0.257, pruned_loss=0.05138, over 7930.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2863, pruned_loss=0.06099, over 1614993.78 frames. ], batch size: 20, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:35:17,190 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.327e+02 2.972e+02 3.761e+02 7.589e+02, threshold=5.944e+02, percent-clipped=3.0 +2023-02-07 04:35:21,630 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177199.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:35:27,656 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 04:35:32,257 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:35:38,466 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:35:51,507 INFO [train.py:901] (2/4) Epoch 22, batch 7500, loss[loss=0.1817, simple_loss=0.2538, pruned_loss=0.05484, over 7422.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2865, pruned_loss=0.06106, over 1612999.90 frames. ], batch size: 17, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:35:58,953 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8529, 1.6609, 1.8404, 1.7347, 0.9567, 1.5908, 2.1364, 2.2960], + device='cuda:2'), covar=tensor([0.0426, 0.1210, 0.1643, 0.1352, 0.0629, 0.1497, 0.0628, 0.0557], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0099, 0.0162, 0.0111, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 04:36:23,739 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.87 vs. limit=5.0 +2023-02-07 04:36:25,348 INFO [train.py:901] (2/4) Epoch 22, batch 7550, loss[loss=0.1594, simple_loss=0.2443, pruned_loss=0.03723, over 7458.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2861, pruned_loss=0.06082, over 1610542.34 frames. ], batch size: 17, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:36:26,041 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.508e+02 3.019e+02 3.781e+02 7.904e+02, threshold=6.039e+02, percent-clipped=4.0 +2023-02-07 04:36:43,691 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6845, 1.4264, 3.2010, 1.4355, 2.3893, 3.4617, 3.5530, 2.9711], + device='cuda:2'), covar=tensor([0.1271, 0.1760, 0.0286, 0.2010, 0.0854, 0.0226, 0.0475, 0.0518], + device='cuda:2'), in_proj_covar=tensor([0.0297, 0.0321, 0.0285, 0.0314, 0.0309, 0.0266, 0.0420, 0.0303], + 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') +2023-02-07 04:36:51,643 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177330.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:36:59,619 INFO [train.py:901] (2/4) Epoch 22, batch 7600, loss[loss=0.1808, simple_loss=0.2733, pruned_loss=0.04417, over 7926.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2853, pruned_loss=0.0602, over 1607300.32 frames. ], batch size: 20, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:37:11,460 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:37:29,851 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177383.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:37:35,783 INFO [train.py:901] (2/4) Epoch 22, batch 7650, loss[loss=0.2203, simple_loss=0.3042, pruned_loss=0.06819, over 8467.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2854, pruned_loss=0.06035, over 1606045.14 frames. ], batch size: 49, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:37:36,440 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.559e+02 3.074e+02 4.315e+02 1.263e+03, threshold=6.148e+02, percent-clipped=10.0 +2023-02-07 04:37:39,983 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177398.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:37:57,393 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177423.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:38:00,846 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0105, 2.6569, 2.1622, 2.3227, 2.3643, 2.0481, 2.2219, 2.4232], + device='cuda:2'), covar=tensor([0.1123, 0.0302, 0.0865, 0.0537, 0.0527, 0.1155, 0.0730, 0.0740], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0232, 0.0335, 0.0310, 0.0299, 0.0341, 0.0345, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:38:09,936 INFO [train.py:901] (2/4) Epoch 22, batch 7700, loss[loss=0.2241, simple_loss=0.3016, pruned_loss=0.07334, over 7153.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2848, pruned_loss=0.06038, over 1603215.37 frames. ], batch size: 72, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:38:30,444 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177471.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:38:31,726 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177473.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:38:38,598 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 04:38:45,982 INFO [train.py:901] (2/4) Epoch 22, batch 7750, loss[loss=0.2335, simple_loss=0.306, pruned_loss=0.08049, over 8101.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2859, pruned_loss=0.06107, over 1606144.83 frames. ], batch size: 23, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:38:46,657 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.486e+02 3.125e+02 4.090e+02 1.041e+03, threshold=6.251e+02, percent-clipped=8.0 +2023-02-07 04:39:20,412 INFO [train.py:901] (2/4) Epoch 22, batch 7800, loss[loss=0.1686, simple_loss=0.2468, pruned_loss=0.04524, over 7536.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2861, pruned_loss=0.06133, over 1605193.16 frames. ], batch size: 18, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:39:39,794 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177571.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:49,858 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:49,890 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:50,723 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2023-02-07 04:39:51,146 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177588.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:53,646 INFO [train.py:901] (2/4) Epoch 22, batch 7850, loss[loss=0.1963, simple_loss=0.2879, pruned_loss=0.05231, over 8355.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2871, pruned_loss=0.06164, over 1605791.50 frames. ], batch size: 24, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:39:54,300 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.387e+02 2.753e+02 3.373e+02 6.542e+02, threshold=5.505e+02, percent-clipped=2.0 +2023-02-07 04:40:06,551 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177611.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:40:26,686 INFO [train.py:901] (2/4) Epoch 22, batch 7900, loss[loss=0.2134, simple_loss=0.3005, pruned_loss=0.06316, over 8197.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2873, pruned_loss=0.06167, over 1605032.84 frames. ], batch size: 23, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:40:53,547 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:41:00,025 INFO [train.py:901] (2/4) Epoch 22, batch 7950, loss[loss=0.1731, simple_loss=0.2522, pruned_loss=0.04704, over 7765.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2865, pruned_loss=0.0611, over 1608906.88 frames. ], batch size: 19, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:41:00,683 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.424e+02 2.966e+02 3.766e+02 9.319e+02, threshold=5.931e+02, percent-clipped=7.0 +2023-02-07 04:41:00,853 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2378, 1.4968, 4.3660, 1.9953, 2.4422, 4.9613, 4.9993, 4.3531], + device='cuda:2'), covar=tensor([0.1186, 0.1918, 0.0264, 0.1953, 0.1251, 0.0158, 0.0384, 0.0541], + device='cuda:2'), in_proj_covar=tensor([0.0297, 0.0323, 0.0285, 0.0315, 0.0311, 0.0267, 0.0422, 0.0304], + 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') +2023-02-07 04:41:18,843 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.52 vs. limit=5.0 +2023-02-07 04:41:33,729 INFO [train.py:901] (2/4) Epoch 22, batch 8000, loss[loss=0.2034, simple_loss=0.2748, pruned_loss=0.066, over 7306.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06119, over 1606458.14 frames. ], batch size: 16, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:41:51,625 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 04:42:06,692 INFO [train.py:901] (2/4) Epoch 22, batch 8050, loss[loss=0.1941, simple_loss=0.2745, pruned_loss=0.05685, over 7528.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2853, pruned_loss=0.06142, over 1586136.95 frames. ], batch size: 18, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:42:07,272 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.308e+02 2.923e+02 3.618e+02 1.070e+03, threshold=5.846e+02, percent-clipped=4.0 +2023-02-07 04:42:10,809 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177798.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:42:22,386 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177815.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:42:39,547 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 04:42:44,817 INFO [train.py:901] (2/4) Epoch 23, batch 0, loss[loss=0.2402, simple_loss=0.3177, pruned_loss=0.08136, over 8455.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3177, pruned_loss=0.08136, over 8455.00 frames. ], batch size: 27, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:42:44,817 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 04:42:56,157 INFO [train.py:935] (2/4) Epoch 23, validation: loss=0.1743, simple_loss=0.274, pruned_loss=0.0373, over 944034.00 frames. +2023-02-07 04:42:56,158 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 04:43:08,349 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:10,538 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177844.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:12,384 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 04:43:26,738 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177867.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:28,089 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177869.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:32,010 INFO [train.py:901] (2/4) Epoch 23, batch 50, loss[loss=0.1893, simple_loss=0.2789, pruned_loss=0.04987, over 8511.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2928, pruned_loss=0.06369, over 367436.98 frames. ], batch size: 28, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:43:42,582 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9498, 2.1846, 3.1364, 1.8035, 2.7672, 2.2058, 2.0867, 2.5818], + device='cuda:2'), covar=tensor([0.1655, 0.2320, 0.0706, 0.3984, 0.1470, 0.2681, 0.2016, 0.1990], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0605, 0.0556, 0.0648, 0.0648, 0.0594, 0.0538, 0.0633], + 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') +2023-02-07 04:43:45,276 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.650e+02 3.149e+02 3.939e+02 1.519e+03, threshold=6.298e+02, percent-clipped=14.0 +2023-02-07 04:43:46,678 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 04:44:01,100 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177915.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:44:02,081 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 04:44:07,951 INFO [train.py:901] (2/4) Epoch 23, batch 100, loss[loss=0.2116, simple_loss=0.3022, pruned_loss=0.06054, over 8293.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2896, pruned_loss=0.0622, over 643482.62 frames. ], batch size: 23, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:44:09,372 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 04:44:15,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2270, 2.3698, 1.9144, 2.8975, 1.4186, 1.7578, 2.0592, 2.2561], + device='cuda:2'), covar=tensor([0.0608, 0.0634, 0.0845, 0.0355, 0.1072, 0.1147, 0.0881, 0.0771], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0214, 0.0207, 0.0246, 0.0250, 0.0207], + 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') +2023-02-07 04:44:42,228 INFO [train.py:901] (2/4) Epoch 23, batch 150, loss[loss=0.1896, simple_loss=0.2665, pruned_loss=0.05639, over 7969.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2918, pruned_loss=0.0638, over 861768.31 frames. ], batch size: 21, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:44:49,533 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1908, 1.5744, 4.0936, 1.8344, 2.4348, 4.5903, 4.6460, 3.9712], + device='cuda:2'), covar=tensor([0.1203, 0.1999, 0.0329, 0.2182, 0.1456, 0.0185, 0.0426, 0.0553], + device='cuda:2'), in_proj_covar=tensor([0.0300, 0.0323, 0.0287, 0.0318, 0.0313, 0.0269, 0.0424, 0.0307], + 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') +2023-02-07 04:44:54,927 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.352e+02 3.015e+02 3.767e+02 5.945e+02, threshold=6.031e+02, percent-clipped=0.0 +2023-02-07 04:44:55,344 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 04:45:18,302 INFO [train.py:901] (2/4) Epoch 23, batch 200, loss[loss=0.2349, simple_loss=0.325, pruned_loss=0.07244, over 8195.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2941, pruned_loss=0.06474, over 1028916.69 frames. ], batch size: 23, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:45:19,115 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178026.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:45:21,865 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178030.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:45:53,034 INFO [train.py:901] (2/4) Epoch 23, batch 250, loss[loss=0.1645, simple_loss=0.2529, pruned_loss=0.03804, over 8131.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2912, pruned_loss=0.06284, over 1160977.11 frames. ], batch size: 22, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:46:04,761 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 04:46:06,105 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.380e+02 2.804e+02 3.484e+02 6.736e+02, threshold=5.609e+02, percent-clipped=2.0 +2023-02-07 04:46:12,814 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 04:46:28,460 INFO [train.py:901] (2/4) Epoch 23, batch 300, loss[loss=0.2085, simple_loss=0.3019, pruned_loss=0.05755, over 8316.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2907, pruned_loss=0.06228, over 1267758.17 frames. ], batch size: 25, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:46:40,067 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:46:40,621 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178142.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:46:45,437 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0686, 1.5839, 1.4081, 1.4923, 1.3167, 1.2700, 1.2774, 1.3023], + device='cuda:2'), covar=tensor([0.1217, 0.0485, 0.1345, 0.0584, 0.0773, 0.1558, 0.0927, 0.0857], + device='cuda:2'), in_proj_covar=tensor([0.0352, 0.0231, 0.0332, 0.0308, 0.0297, 0.0337, 0.0342, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 04:46:52,877 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178159.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:46:53,827 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 04:46:54,585 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 04:46:59,077 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3107, 1.2206, 2.3559, 1.3109, 2.1568, 2.5425, 2.6908, 2.1596], + device='cuda:2'), covar=tensor([0.1284, 0.1480, 0.0430, 0.2010, 0.0716, 0.0380, 0.0671, 0.0677], + device='cuda:2'), in_proj_covar=tensor([0.0299, 0.0322, 0.0288, 0.0316, 0.0312, 0.0267, 0.0423, 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') +2023-02-07 04:47:03,746 INFO [train.py:901] (2/4) Epoch 23, batch 350, loss[loss=0.2085, simple_loss=0.2977, pruned_loss=0.05965, over 8500.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2896, pruned_loss=0.06196, over 1346784.96 frames. ], batch size: 29, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:47:16,037 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.418e+02 2.905e+02 3.451e+02 8.072e+02, threshold=5.809e+02, percent-clipped=5.0 +2023-02-07 04:47:23,326 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6783, 1.5208, 2.8067, 1.3575, 2.2235, 3.0416, 3.1703, 2.5723], + device='cuda:2'), covar=tensor([0.1229, 0.1603, 0.0370, 0.2122, 0.0895, 0.0292, 0.0557, 0.0600], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0320, 0.0286, 0.0315, 0.0311, 0.0266, 0.0421, 0.0304], + 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') +2023-02-07 04:47:38,678 INFO [train.py:901] (2/4) Epoch 23, batch 400, loss[loss=0.2, simple_loss=0.2919, pruned_loss=0.05406, over 8650.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2889, pruned_loss=0.06183, over 1401773.58 frames. ], batch size: 34, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:47:43,362 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 04:47:51,826 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3584, 1.5234, 2.2532, 1.2618, 1.8046, 1.5725, 1.4208, 1.7548], + device='cuda:2'), covar=tensor([0.1533, 0.2135, 0.0639, 0.3584, 0.1418, 0.2588, 0.1825, 0.2020], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0605, 0.0556, 0.0645, 0.0647, 0.0592, 0.0537, 0.0633], + 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') +2023-02-07 04:48:02,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178257.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:15,038 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178274.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:15,486 INFO [train.py:901] (2/4) Epoch 23, batch 450, loss[loss=0.2528, simple_loss=0.3118, pruned_loss=0.09688, over 7633.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.288, pruned_loss=0.06157, over 1448537.45 frames. ], batch size: 19, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:48:17,136 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3382, 2.7948, 2.3314, 3.9579, 1.7979, 2.0779, 2.5907, 2.8598], + device='cuda:2'), covar=tensor([0.0711, 0.0744, 0.0779, 0.0241, 0.1060, 0.1202, 0.0991, 0.0729], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0197, 0.0245, 0.0215, 0.0207, 0.0246, 0.0250, 0.0207], + 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') +2023-02-07 04:48:23,158 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178286.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:27,632 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.308e+02 2.812e+02 3.532e+02 1.107e+03, threshold=5.624e+02, percent-clipped=2.0 +2023-02-07 04:48:40,215 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:50,185 INFO [train.py:901] (2/4) Epoch 23, batch 500, loss[loss=0.2611, simple_loss=0.3187, pruned_loss=0.1018, over 7928.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2911, pruned_loss=0.06302, over 1492028.84 frames. ], batch size: 20, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:49:25,960 INFO [train.py:901] (2/4) Epoch 23, batch 550, loss[loss=0.1961, simple_loss=0.2852, pruned_loss=0.05353, over 8237.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2903, pruned_loss=0.06235, over 1515252.45 frames. ], batch size: 22, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:49:39,363 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.448e+02 3.105e+02 3.761e+02 9.562e+02, threshold=6.211e+02, percent-clipped=5.0 +2023-02-07 04:49:42,442 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178397.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:49:59,309 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178422.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:50:01,200 INFO [train.py:901] (2/4) Epoch 23, batch 600, loss[loss=0.2102, simple_loss=0.2819, pruned_loss=0.0693, over 8076.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2894, pruned_loss=0.0616, over 1540136.02 frames. ], batch size: 21, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:50:14,803 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 04:50:33,525 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178470.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:50:36,799 INFO [train.py:901] (2/4) Epoch 23, batch 650, loss[loss=0.1821, simple_loss=0.2668, pruned_loss=0.04869, over 8515.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2885, pruned_loss=0.06146, over 1559846.14 frames. ], batch size: 31, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:50:49,804 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.230e+02 2.701e+02 3.368e+02 8.641e+02, threshold=5.402e+02, percent-clipped=2.0 +2023-02-07 04:51:04,373 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178513.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:12,430 INFO [train.py:901] (2/4) Epoch 23, batch 700, loss[loss=0.1919, simple_loss=0.2692, pruned_loss=0.05728, over 7779.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2879, pruned_loss=0.0616, over 1569389.16 frames. ], batch size: 19, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:51:16,058 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178530.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:18,198 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1704, 2.0513, 2.6889, 2.2525, 2.6014, 2.2959, 2.0505, 1.5367], + device='cuda:2'), covar=tensor([0.6030, 0.5068, 0.2133, 0.3862, 0.2682, 0.2991, 0.1943, 0.5536], + device='cuda:2'), in_proj_covar=tensor([0.0941, 0.0986, 0.0813, 0.0952, 0.0997, 0.0899, 0.0754, 0.0831], + 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') +2023-02-07 04:51:21,532 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178538.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:33,985 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178555.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:47,511 INFO [train.py:901] (2/4) Epoch 23, batch 750, loss[loss=0.1894, simple_loss=0.2679, pruned_loss=0.05539, over 7428.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2872, pruned_loss=0.06132, over 1582382.28 frames. ], batch size: 17, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:51:49,831 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8753, 2.2371, 3.5668, 1.8429, 1.8588, 3.5226, 0.7215, 2.1568], + device='cuda:2'), covar=tensor([0.1322, 0.1270, 0.0230, 0.1668, 0.2448, 0.0324, 0.2067, 0.1376], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0200, 0.0130, 0.0221, 0.0270, 0.0137, 0.0171, 0.0196], + 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') +2023-02-07 04:51:59,472 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9860, 2.3744, 4.1835, 1.6793, 3.1539, 2.4419, 2.0293, 2.9363], + device='cuda:2'), covar=tensor([0.1798, 0.2694, 0.0910, 0.4521, 0.1783, 0.3138, 0.2172, 0.2584], + device='cuda:2'), in_proj_covar=tensor([0.0527, 0.0606, 0.0555, 0.0645, 0.0649, 0.0594, 0.0536, 0.0634], + 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') +2023-02-07 04:52:00,639 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.529e+02 2.988e+02 3.531e+02 9.866e+02, threshold=5.976e+02, percent-clipped=5.0 +2023-02-07 04:52:03,331 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 04:52:12,883 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 04:52:24,020 INFO [train.py:901] (2/4) Epoch 23, batch 800, loss[loss=0.1709, simple_loss=0.2479, pruned_loss=0.04699, over 7548.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.287, pruned_loss=0.06136, over 1586212.20 frames. ], batch size: 18, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:52:32,111 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178637.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:52:38,299 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0287, 2.2060, 1.8405, 2.8102, 1.3666, 1.6589, 2.0640, 2.1417], + device='cuda:2'), covar=tensor([0.0690, 0.0741, 0.0856, 0.0340, 0.1093, 0.1246, 0.0763, 0.0800], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0197, 0.0245, 0.0214, 0.0206, 0.0245, 0.0249, 0.0207], + 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') +2023-02-07 04:52:57,747 INFO [train.py:901] (2/4) Epoch 23, batch 850, loss[loss=0.2302, simple_loss=0.3084, pruned_loss=0.07603, over 8452.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2869, pruned_loss=0.06182, over 1591768.86 frames. ], batch size: 27, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:53:10,566 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.561e+02 2.992e+02 3.918e+02 1.040e+03, threshold=5.984e+02, percent-clipped=6.0 +2023-02-07 04:53:24,482 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178712.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:53:26,495 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178715.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:53:31,385 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1767, 4.1579, 3.7910, 1.9346, 3.6751, 3.8171, 3.6335, 3.6550], + device='cuda:2'), covar=tensor([0.0701, 0.0524, 0.0927, 0.4719, 0.0811, 0.0845, 0.1362, 0.0782], + device='cuda:2'), in_proj_covar=tensor([0.0535, 0.0440, 0.0435, 0.0544, 0.0426, 0.0448, 0.0431, 0.0389], + 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') +2023-02-07 04:53:34,048 INFO [train.py:901] (2/4) Epoch 23, batch 900, loss[loss=0.2104, simple_loss=0.2902, pruned_loss=0.06531, over 8498.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2873, pruned_loss=0.06155, over 1600026.84 frames. ], batch size: 28, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:53:55,223 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 +2023-02-07 04:54:03,482 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6402, 1.9305, 4.6702, 2.2721, 2.9836, 5.3287, 5.2976, 4.6739], + device='cuda:2'), covar=tensor([0.1066, 0.1773, 0.0239, 0.1804, 0.1053, 0.0138, 0.0307, 0.0452], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0321, 0.0286, 0.0315, 0.0310, 0.0267, 0.0422, 0.0304], + 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') +2023-02-07 04:54:09,414 INFO [train.py:901] (2/4) Epoch 23, batch 950, loss[loss=0.211, simple_loss=0.2971, pruned_loss=0.06241, over 8470.00 frames. ], tot_loss[loss=0.205, simple_loss=0.287, pruned_loss=0.06148, over 1602151.50 frames. ], batch size: 27, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:54:18,546 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178788.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:54:21,857 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.330e+02 2.907e+02 3.544e+02 9.473e+02, threshold=5.814e+02, percent-clipped=4.0 +2023-02-07 04:54:22,772 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6473, 1.9394, 2.0394, 1.2745, 2.3038, 1.3955, 0.8056, 1.7901], + device='cuda:2'), covar=tensor([0.0732, 0.0432, 0.0328, 0.0671, 0.0435, 0.1010, 0.0882, 0.0437], + device='cuda:2'), in_proj_covar=tensor([0.0453, 0.0395, 0.0347, 0.0448, 0.0380, 0.0536, 0.0393, 0.0425], + device='cuda:2'), out_proj_covar=tensor([1.2096e-04, 1.0363e-04, 9.1189e-05, 1.1773e-04, 9.9773e-05, 1.5107e-04, + 1.0597e-04, 1.1251e-04], device='cuda:2') +2023-02-07 04:54:35,810 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 04:54:37,115 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:54:45,364 INFO [train.py:901] (2/4) Epoch 23, batch 1000, loss[loss=0.1959, simple_loss=0.273, pruned_loss=0.05937, over 8032.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2854, pruned_loss=0.06053, over 1604649.92 frames. ], batch size: 22, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:55:12,384 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 04:55:21,361 INFO [train.py:901] (2/4) Epoch 23, batch 1050, loss[loss=0.1796, simple_loss=0.276, pruned_loss=0.04161, over 8454.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2848, pruned_loss=0.06008, over 1609453.11 frames. ], batch size: 27, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:55:25,409 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 04:55:33,401 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.332e+02 2.695e+02 3.454e+02 6.847e+02, threshold=5.390e+02, percent-clipped=5.0 +2023-02-07 04:55:46,651 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178912.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:55:56,207 INFO [train.py:901] (2/4) Epoch 23, batch 1100, loss[loss=0.2271, simple_loss=0.3071, pruned_loss=0.07356, over 8227.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2864, pruned_loss=0.06073, over 1610043.09 frames. ], batch size: 24, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:55:59,139 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:56:29,584 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0405, 2.4524, 3.8461, 2.2189, 2.0713, 3.8077, 0.8090, 2.2981], + device='cuda:2'), covar=tensor([0.1346, 0.1265, 0.0300, 0.1518, 0.2293, 0.0364, 0.2028, 0.1562], + device='cuda:2'), in_proj_covar=tensor([0.0192, 0.0200, 0.0129, 0.0220, 0.0268, 0.0136, 0.0170, 0.0194], + 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') +2023-02-07 04:56:32,124 INFO [train.py:901] (2/4) Epoch 23, batch 1150, loss[loss=0.1793, simple_loss=0.2656, pruned_loss=0.04648, over 7452.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2868, pruned_loss=0.06109, over 1608412.08 frames. ], batch size: 17, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:56:36,274 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 04:56:36,348 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178981.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:56:41,187 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2657, 3.1862, 2.9853, 1.5395, 2.9337, 2.9249, 2.9613, 2.8642], + device='cuda:2'), covar=tensor([0.1155, 0.0793, 0.1238, 0.4758, 0.1057, 0.1355, 0.1616, 0.1042], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0439, 0.0432, 0.0542, 0.0426, 0.0447, 0.0430, 0.0387], + 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') +2023-02-07 04:56:45,237 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.628e+02 3.162e+02 4.177e+02 1.087e+03, threshold=6.324e+02, percent-clipped=6.0 +2023-02-07 04:57:07,131 INFO [train.py:901] (2/4) Epoch 23, batch 1200, loss[loss=0.1933, simple_loss=0.288, pruned_loss=0.04934, over 8357.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2864, pruned_loss=0.06065, over 1612561.03 frames. ], batch size: 24, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:57:29,093 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:57:31,025 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179059.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 04:57:42,798 INFO [train.py:901] (2/4) Epoch 23, batch 1250, loss[loss=0.1953, simple_loss=0.2913, pruned_loss=0.04971, over 8331.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2869, pruned_loss=0.06108, over 1610041.13 frames. ], batch size: 25, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:57:55,982 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.427e+02 2.289e+02 2.896e+02 3.686e+02 5.954e+02, threshold=5.791e+02, percent-clipped=0.0 +2023-02-07 04:57:58,265 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179096.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:58:19,011 INFO [train.py:901] (2/4) Epoch 23, batch 1300, loss[loss=0.1806, simple_loss=0.2683, pruned_loss=0.04648, over 8126.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2864, pruned_loss=0.06026, over 1613044.45 frames. ], batch size: 22, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:58:24,104 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:58:51,907 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179171.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:58:53,991 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179174.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:58:54,481 INFO [train.py:901] (2/4) Epoch 23, batch 1350, loss[loss=0.1914, simple_loss=0.2815, pruned_loss=0.05065, over 7661.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2866, pruned_loss=0.06064, over 1609980.47 frames. ], batch size: 19, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:59:01,718 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179185.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:59:07,800 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.184e+02 2.635e+02 3.098e+02 5.270e+02, threshold=5.271e+02, percent-clipped=0.0 +2023-02-07 04:59:20,404 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179210.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:59:30,637 INFO [train.py:901] (2/4) Epoch 23, batch 1400, loss[loss=0.1626, simple_loss=0.2453, pruned_loss=0.03992, over 7429.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2864, pruned_loss=0.06087, over 1607924.76 frames. ], batch size: 17, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:59:47,155 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179247.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:59:53,436 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179256.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:00:06,546 INFO [train.py:901] (2/4) Epoch 23, batch 1450, loss[loss=0.2127, simple_loss=0.2954, pruned_loss=0.06496, over 8461.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2866, pruned_loss=0.06072, over 1611622.71 frames. ], batch size: 29, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:00:16,912 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 05:00:19,768 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.293e+02 2.971e+02 3.774e+02 8.745e+02, threshold=5.941e+02, percent-clipped=9.0 +2023-02-07 05:00:37,216 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-07 05:00:43,614 INFO [train.py:901] (2/4) Epoch 23, batch 1500, loss[loss=0.2259, simple_loss=0.3078, pruned_loss=0.07196, over 7924.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2863, pruned_loss=0.06079, over 1607482.86 frames. ], batch size: 20, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:01:03,400 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179352.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:16,373 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179371.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:01:18,849 INFO [train.py:901] (2/4) Epoch 23, batch 1550, loss[loss=0.1844, simple_loss=0.2723, pruned_loss=0.04827, over 8460.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2884, pruned_loss=0.06207, over 1610601.29 frames. ], batch size: 27, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:01:20,470 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:21,728 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179379.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:31,102 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.616e+02 2.349e+02 2.958e+02 3.969e+02 7.808e+02, threshold=5.916e+02, percent-clipped=1.0 +2023-02-07 05:01:37,883 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.04 vs. limit=5.0 +2023-02-07 05:01:40,056 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-07 05:01:54,021 INFO [train.py:901] (2/4) Epoch 23, batch 1600, loss[loss=0.2206, simple_loss=0.3027, pruned_loss=0.06919, over 8344.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2883, pruned_loss=0.0623, over 1613642.47 frames. ], batch size: 26, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:01:56,466 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179427.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:58,516 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179430.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:01:59,828 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4311, 1.4857, 1.3970, 1.7885, 0.7751, 1.2675, 1.3974, 1.5277], + device='cuda:2'), covar=tensor([0.0947, 0.0830, 0.1026, 0.0518, 0.1142, 0.1529, 0.0724, 0.0749], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0199, 0.0245, 0.0215, 0.0207, 0.0248, 0.0250, 0.0209], + 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') +2023-02-07 05:02:14,539 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:02:16,567 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179455.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:02:21,384 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179462.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:02:26,495 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6588, 1.6086, 2.4538, 1.3572, 1.0745, 2.3621, 0.5787, 1.4102], + device='cuda:2'), covar=tensor([0.1849, 0.1457, 0.0317, 0.1451, 0.2948, 0.0379, 0.1987, 0.1456], + device='cuda:2'), in_proj_covar=tensor([0.0190, 0.0198, 0.0129, 0.0219, 0.0267, 0.0136, 0.0169, 0.0193], + 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') +2023-02-07 05:02:31,190 INFO [train.py:901] (2/4) Epoch 23, batch 1650, loss[loss=0.2593, simple_loss=0.3335, pruned_loss=0.09256, over 8666.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2881, pruned_loss=0.06211, over 1612342.46 frames. ], batch size: 34, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:02:41,802 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3193, 1.8614, 1.3369, 2.7801, 1.2135, 1.1723, 2.1950, 2.0242], + device='cuda:2'), covar=tensor([0.1639, 0.1247, 0.2017, 0.0405, 0.1460, 0.2235, 0.0872, 0.0944], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0199, 0.0245, 0.0215, 0.0207, 0.0248, 0.0251, 0.0209], + 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') +2023-02-07 05:02:43,578 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.367e+02 2.783e+02 3.381e+02 8.055e+02, threshold=5.566e+02, percent-clipped=4.0 +2023-02-07 05:02:50,707 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179503.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:03:06,271 INFO [train.py:901] (2/4) Epoch 23, batch 1700, loss[loss=0.2371, simple_loss=0.3152, pruned_loss=0.07953, over 8523.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2888, pruned_loss=0.06238, over 1613494.14 frames. ], batch size: 39, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:03:08,727 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179528.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:03:42,331 INFO [train.py:901] (2/4) Epoch 23, batch 1750, loss[loss=0.2365, simple_loss=0.3182, pruned_loss=0.07736, over 8616.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2883, pruned_loss=0.06244, over 1612551.13 frames. ], batch size: 49, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:03:56,241 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.481e+02 2.857e+02 3.517e+02 8.396e+02, threshold=5.713e+02, percent-clipped=3.0 +2023-02-07 05:04:17,968 INFO [train.py:901] (2/4) Epoch 23, batch 1800, loss[loss=0.1869, simple_loss=0.2769, pruned_loss=0.04843, over 8477.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2881, pruned_loss=0.06201, over 1612150.19 frames. ], batch size: 25, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:04:19,578 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179627.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:04:33,217 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 05:04:37,264 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179652.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:04:38,536 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179654.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:04:54,476 INFO [train.py:901] (2/4) Epoch 23, batch 1850, loss[loss=0.2071, simple_loss=0.2951, pruned_loss=0.05956, over 8341.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2872, pruned_loss=0.0615, over 1612646.91 frames. ], batch size: 26, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:05:07,480 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.311e+02 2.831e+02 3.615e+02 8.108e+02, threshold=5.663e+02, percent-clipped=6.0 +2023-02-07 05:05:28,519 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:05:29,819 INFO [train.py:901] (2/4) Epoch 23, batch 1900, loss[loss=0.2431, simple_loss=0.3199, pruned_loss=0.08311, over 8235.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2868, pruned_loss=0.06162, over 1611840.32 frames. ], batch size: 22, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:05:59,876 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 05:06:05,566 INFO [train.py:901] (2/4) Epoch 23, batch 1950, loss[loss=0.2124, simple_loss=0.2812, pruned_loss=0.07186, over 7225.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2873, pruned_loss=0.06175, over 1610840.49 frames. ], batch size: 16, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:06:12,614 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 05:06:19,490 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.457e+02 2.986e+02 3.643e+02 8.972e+02, threshold=5.972e+02, percent-clipped=4.0 +2023-02-07 05:06:21,801 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3806, 1.4661, 1.5251, 1.1591, 1.6163, 1.2169, 0.6908, 1.4695], + device='cuda:2'), covar=tensor([0.0478, 0.0356, 0.0255, 0.0435, 0.0338, 0.0714, 0.0728, 0.0251], + device='cuda:2'), in_proj_covar=tensor([0.0455, 0.0393, 0.0346, 0.0449, 0.0380, 0.0534, 0.0393, 0.0422], + device='cuda:2'), out_proj_covar=tensor([1.2144e-04, 1.0307e-04, 9.0866e-05, 1.1813e-04, 1.0013e-04, 1.5032e-04, + 1.0593e-04, 1.1160e-04], device='cuda:2') +2023-02-07 05:06:28,060 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179806.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:06:31,241 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 05:06:34,176 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:06:37,042 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4255, 1.6150, 2.1663, 1.3164, 1.6116, 1.7048, 1.4705, 1.5393], + device='cuda:2'), covar=tensor([0.1852, 0.2379, 0.0951, 0.4324, 0.1823, 0.3217, 0.2269, 0.2041], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0608, 0.0556, 0.0647, 0.0650, 0.0594, 0.0538, 0.0633], + 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') +2023-02-07 05:06:41,620 INFO [train.py:901] (2/4) Epoch 23, batch 2000, loss[loss=0.1924, simple_loss=0.292, pruned_loss=0.04639, over 8578.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2871, pruned_loss=0.0611, over 1612837.11 frames. ], batch size: 39, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:06:50,597 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179838.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:07:16,428 INFO [train.py:901] (2/4) Epoch 23, batch 2050, loss[loss=0.1764, simple_loss=0.2489, pruned_loss=0.0519, over 7213.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2879, pruned_loss=0.06187, over 1611388.06 frames. ], batch size: 16, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:07:30,040 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.444e+02 2.856e+02 3.794e+02 1.051e+03, threshold=5.713e+02, percent-clipped=7.0 +2023-02-07 05:07:49,559 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179921.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:07:50,918 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1944, 4.1531, 3.7911, 1.9303, 3.6904, 3.7925, 3.7043, 3.6185], + device='cuda:2'), covar=tensor([0.0733, 0.0643, 0.1157, 0.4656, 0.0881, 0.1027, 0.1429, 0.0957], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0440, 0.0435, 0.0542, 0.0429, 0.0448, 0.0432, 0.0389], + 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') +2023-02-07 05:07:51,623 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179924.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:07:52,160 INFO [train.py:901] (2/4) Epoch 23, batch 2100, loss[loss=0.1811, simple_loss=0.2572, pruned_loss=0.05253, over 7788.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.288, pruned_loss=0.06234, over 1607018.44 frames. ], batch size: 19, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:08:04,845 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179942.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:08:24,974 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3636, 1.5017, 1.3507, 1.8191, 0.6643, 1.2408, 1.2731, 1.5078], + device='cuda:2'), covar=tensor([0.0882, 0.0795, 0.1011, 0.0515, 0.1249, 0.1436, 0.0800, 0.0750], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0198, 0.0243, 0.0214, 0.0206, 0.0246, 0.0249, 0.0207], + 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') +2023-02-07 05:08:27,554 INFO [train.py:901] (2/4) Epoch 23, batch 2150, loss[loss=0.2088, simple_loss=0.2985, pruned_loss=0.0596, over 8568.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2887, pruned_loss=0.0624, over 1610334.09 frames. ], batch size: 31, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:08:41,587 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.311e+02 2.940e+02 3.642e+02 8.826e+02, threshold=5.880e+02, percent-clipped=6.0 +2023-02-07 05:08:44,586 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179998.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:09:05,727 INFO [train.py:901] (2/4) Epoch 23, batch 2200, loss[loss=0.2226, simple_loss=0.3018, pruned_loss=0.07167, over 8239.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2875, pruned_loss=0.06169, over 1611055.06 frames. ], batch size: 22, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:09:18,804 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:09:23,719 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6291, 1.8083, 2.7693, 1.4637, 2.0312, 2.0748, 1.6283, 2.0062], + device='cuda:2'), covar=tensor([0.1779, 0.2667, 0.0826, 0.4553, 0.1787, 0.2898, 0.2324, 0.2025], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0610, 0.0557, 0.0648, 0.0650, 0.0595, 0.0539, 0.0633], + 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') +2023-02-07 05:09:40,654 INFO [train.py:901] (2/4) Epoch 23, batch 2250, loss[loss=0.1782, simple_loss=0.2727, pruned_loss=0.0419, over 8355.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.288, pruned_loss=0.06222, over 1612162.49 frames. ], batch size: 24, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:09:53,767 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.377e+02 2.815e+02 3.570e+02 6.536e+02, threshold=5.630e+02, percent-clipped=1.0 +2023-02-07 05:09:54,034 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:07,874 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180113.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:12,121 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180119.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:16,768 INFO [train.py:901] (2/4) Epoch 23, batch 2300, loss[loss=0.2109, simple_loss=0.2983, pruned_loss=0.06177, over 8498.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2886, pruned_loss=0.06213, over 1617730.81 frames. ], batch size: 26, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:10:40,104 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180158.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:52,677 INFO [train.py:901] (2/4) Epoch 23, batch 2350, loss[loss=0.3004, simple_loss=0.3539, pruned_loss=0.1235, over 7145.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2885, pruned_loss=0.06212, over 1613089.10 frames. ], batch size: 72, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:10:54,330 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180177.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:11:05,883 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.451e+02 2.928e+02 3.544e+02 9.883e+02, threshold=5.856e+02, percent-clipped=4.0 +2023-02-07 05:11:11,595 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180202.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:11:25,818 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180223.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:11:27,087 INFO [train.py:901] (2/4) Epoch 23, batch 2400, loss[loss=0.189, simple_loss=0.278, pruned_loss=0.05004, over 8286.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2877, pruned_loss=0.06155, over 1613758.97 frames. ], batch size: 23, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:11:59,473 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180268.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:12:02,922 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180273.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:12:04,127 INFO [train.py:901] (2/4) Epoch 23, batch 2450, loss[loss=0.2373, simple_loss=0.3072, pruned_loss=0.08366, over 8342.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2877, pruned_loss=0.06187, over 1615250.51 frames. ], batch size: 26, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:12:12,624 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180286.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:12:18,002 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.501e+02 2.918e+02 3.866e+02 1.157e+03, threshold=5.835e+02, percent-clipped=6.0 +2023-02-07 05:12:39,631 INFO [train.py:901] (2/4) Epoch 23, batch 2500, loss[loss=0.1733, simple_loss=0.2513, pruned_loss=0.04761, over 7791.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2879, pruned_loss=0.06223, over 1615538.87 frames. ], batch size: 19, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:13:00,484 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180354.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:12,934 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180369.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:15,763 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0745, 1.8545, 2.3502, 1.9848, 2.3345, 2.1344, 1.9484, 1.1794], + device='cuda:2'), covar=tensor([0.5616, 0.4874, 0.2008, 0.3578, 0.2333, 0.3056, 0.1856, 0.5065], + device='cuda:2'), in_proj_covar=tensor([0.0949, 0.0993, 0.0818, 0.0955, 0.1004, 0.0906, 0.0755, 0.0837], + 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') +2023-02-07 05:13:16,784 INFO [train.py:901] (2/4) Epoch 23, batch 2550, loss[loss=0.2321, simple_loss=0.2839, pruned_loss=0.09019, over 7807.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2884, pruned_loss=0.06261, over 1615676.67 frames. ], batch size: 19, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:13:22,394 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180383.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:13:25,751 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:27,629 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 05:13:29,883 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.435e+02 3.031e+02 3.942e+02 1.076e+03, threshold=6.063e+02, percent-clipped=1.0 +2023-02-07 05:13:30,128 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180394.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:32,962 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2389, 1.6139, 1.6652, 1.0004, 1.6798, 1.2759, 0.2511, 1.4965], + device='cuda:2'), covar=tensor([0.0556, 0.0374, 0.0290, 0.0544, 0.0457, 0.0960, 0.0895, 0.0297], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0396, 0.0349, 0.0449, 0.0382, 0.0537, 0.0395, 0.0424], + device='cuda:2'), out_proj_covar=tensor([1.2232e-04, 1.0370e-04, 9.1689e-05, 1.1798e-04, 1.0060e-04, 1.5134e-04, + 1.0646e-04, 1.1215e-04], device='cuda:2') +2023-02-07 05:13:35,750 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180401.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:51,957 INFO [train.py:901] (2/4) Epoch 23, batch 2600, loss[loss=0.2291, simple_loss=0.3087, pruned_loss=0.07475, over 8438.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2875, pruned_loss=0.06186, over 1612572.60 frames. ], batch size: 29, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:14:28,402 INFO [train.py:901] (2/4) Epoch 23, batch 2650, loss[loss=0.1817, simple_loss=0.265, pruned_loss=0.04921, over 7192.00 frames. ], tot_loss[loss=0.205, simple_loss=0.287, pruned_loss=0.0615, over 1614750.38 frames. ], batch size: 16, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:14:42,181 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.331e+02 2.876e+02 3.734e+02 9.435e+02, threshold=5.753e+02, percent-clipped=4.0 +2023-02-07 05:14:48,432 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180503.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:04,098 INFO [train.py:901] (2/4) Epoch 23, batch 2700, loss[loss=0.2124, simple_loss=0.3044, pruned_loss=0.06019, over 8440.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2867, pruned_loss=0.06123, over 1618412.68 frames. ], batch size: 27, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:15:07,062 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180529.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:24,088 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8070, 1.3532, 3.9877, 1.3842, 3.5247, 3.3234, 3.6021, 3.4963], + device='cuda:2'), covar=tensor([0.0698, 0.4889, 0.0657, 0.4415, 0.1364, 0.1038, 0.0685, 0.0857], + device='cuda:2'), in_proj_covar=tensor([0.0631, 0.0636, 0.0692, 0.0627, 0.0702, 0.0600, 0.0601, 0.0679], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 05:15:24,163 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:33,273 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:39,688 INFO [train.py:901] (2/4) Epoch 23, batch 2750, loss[loss=0.2146, simple_loss=0.3099, pruned_loss=0.05965, over 8449.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2862, pruned_loss=0.06132, over 1617941.15 frames. ], batch size: 27, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:15:48,799 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180588.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:53,491 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.355e+02 2.814e+02 3.432e+02 9.125e+02, threshold=5.629e+02, percent-clipped=4.0 +2023-02-07 05:16:15,674 INFO [train.py:901] (2/4) Epoch 23, batch 2800, loss[loss=0.1636, simple_loss=0.2525, pruned_loss=0.03735, over 8356.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2872, pruned_loss=0.06166, over 1619528.47 frames. ], batch size: 24, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:16:26,283 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180639.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:16:38,611 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180657.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:16:40,120 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 +2023-02-07 05:16:43,318 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180664.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:16:50,749 INFO [train.py:901] (2/4) Epoch 23, batch 2850, loss[loss=0.1922, simple_loss=0.2794, pruned_loss=0.05249, over 8478.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2865, pruned_loss=0.06123, over 1617188.34 frames. ], batch size: 25, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:16:55,715 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:16:55,739 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:04,507 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.469e+02 3.037e+02 3.866e+02 9.714e+02, threshold=6.075e+02, percent-clipped=7.0 +2023-02-07 05:17:07,453 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:27,370 INFO [train.py:901] (2/4) Epoch 23, batch 2900, loss[loss=0.1992, simple_loss=0.2795, pruned_loss=0.05946, over 8028.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06198, over 1619839.97 frames. ], batch size: 22, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:17:45,117 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:52,152 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180759.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:59,465 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 05:18:02,965 INFO [train.py:901] (2/4) Epoch 23, batch 2950, loss[loss=0.2701, simple_loss=0.3563, pruned_loss=0.09199, over 8453.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.287, pruned_loss=0.06178, over 1615160.50 frames. ], batch size: 29, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:18:05,925 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7904, 2.2813, 4.1420, 1.5482, 2.8862, 2.2629, 1.7746, 2.7305], + device='cuda:2'), covar=tensor([0.2012, 0.2727, 0.0867, 0.4708, 0.2061, 0.3142, 0.2415, 0.2603], + device='cuda:2'), in_proj_covar=tensor([0.0524, 0.0606, 0.0555, 0.0646, 0.0645, 0.0591, 0.0537, 0.0628], + 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') +2023-02-07 05:18:07,974 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1643, 2.5350, 2.8287, 1.6247, 3.0935, 1.8149, 1.5419, 2.0938], + device='cuda:2'), covar=tensor([0.0832, 0.0394, 0.0310, 0.0813, 0.0476, 0.0981, 0.1013, 0.0600], + device='cuda:2'), in_proj_covar=tensor([0.0456, 0.0394, 0.0348, 0.0447, 0.0380, 0.0535, 0.0393, 0.0422], + device='cuda:2'), out_proj_covar=tensor([1.2179e-04, 1.0310e-04, 9.1223e-05, 1.1749e-04, 1.0009e-04, 1.5077e-04, + 1.0586e-04, 1.1172e-04], device='cuda:2') +2023-02-07 05:18:09,324 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180784.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:18:16,017 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.356e+02 2.925e+02 3.942e+02 6.480e+02, threshold=5.850e+02, percent-clipped=1.0 +2023-02-07 05:18:30,322 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180813.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:18:38,125 INFO [train.py:901] (2/4) Epoch 23, batch 3000, loss[loss=0.1612, simple_loss=0.2452, pruned_loss=0.03863, over 8081.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2868, pruned_loss=0.06167, over 1617311.52 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:18:38,125 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 05:18:50,538 INFO [train.py:935] (2/4) Epoch 23, validation: loss=0.1735, simple_loss=0.2731, pruned_loss=0.03696, over 944034.00 frames. +2023-02-07 05:18:50,539 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 05:19:03,711 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:19:26,991 INFO [train.py:901] (2/4) Epoch 23, batch 3050, loss[loss=0.2175, simple_loss=0.2886, pruned_loss=0.07321, over 8193.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2869, pruned_loss=0.06169, over 1610575.78 frames. ], batch size: 23, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:19:40,681 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.515e+02 3.107e+02 3.968e+02 1.139e+03, threshold=6.214e+02, percent-clipped=7.0 +2023-02-07 05:20:02,335 INFO [train.py:901] (2/4) Epoch 23, batch 3100, loss[loss=0.1781, simple_loss=0.2644, pruned_loss=0.04595, over 7968.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2867, pruned_loss=0.06116, over 1615197.52 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:20:07,186 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180932.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:20:11,450 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180938.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:20:29,338 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180963.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:20:31,171 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-07 05:20:38,169 INFO [train.py:901] (2/4) Epoch 23, batch 3150, loss[loss=0.2157, simple_loss=0.2977, pruned_loss=0.06679, over 8460.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2862, pruned_loss=0.06095, over 1617574.49 frames. ], batch size: 29, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:20:51,966 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.467e+02 3.042e+02 3.660e+02 1.036e+03, threshold=6.084e+02, percent-clipped=2.0 +2023-02-07 05:21:14,468 INFO [train.py:901] (2/4) Epoch 23, batch 3200, loss[loss=0.1947, simple_loss=0.2752, pruned_loss=0.05709, over 8084.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2864, pruned_loss=0.06124, over 1620759.94 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:21:22,624 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 +2023-02-07 05:21:29,929 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181047.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:21:45,849 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181069.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:21:49,879 INFO [train.py:901] (2/4) Epoch 23, batch 3250, loss[loss=0.2206, simple_loss=0.311, pruned_loss=0.06505, over 8538.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2863, pruned_loss=0.06134, over 1616571.63 frames. ], batch size: 39, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:21:54,665 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 05:22:03,762 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.376e+02 2.917e+02 3.369e+02 6.745e+02, threshold=5.834e+02, percent-clipped=1.0 +2023-02-07 05:22:04,709 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:22:04,835 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:22:20,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7776, 1.6803, 2.4363, 1.5864, 1.2490, 2.3485, 0.4091, 1.4668], + device='cuda:2'), covar=tensor([0.1834, 0.1299, 0.0436, 0.1457, 0.3148, 0.0442, 0.2405, 0.1567], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0199, 0.0130, 0.0222, 0.0272, 0.0138, 0.0171, 0.0194], + 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') +2023-02-07 05:22:26,220 INFO [train.py:901] (2/4) Epoch 23, batch 3300, loss[loss=0.1563, simple_loss=0.2347, pruned_loss=0.03896, over 7433.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2864, pruned_loss=0.06112, over 1616046.64 frames. ], batch size: 17, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:23:01,365 INFO [train.py:901] (2/4) Epoch 23, batch 3350, loss[loss=0.1985, simple_loss=0.2802, pruned_loss=0.05836, over 8234.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2873, pruned_loss=0.06173, over 1614019.45 frames. ], batch size: 22, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:23:10,452 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181187.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:23:14,984 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.358e+02 3.053e+02 3.666e+02 9.674e+02, threshold=6.107e+02, percent-clipped=1.0 +2023-02-07 05:23:26,415 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181209.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:23:32,884 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8289, 3.8201, 3.4421, 1.8268, 3.3435, 3.4827, 3.4224, 3.3206], + device='cuda:2'), covar=tensor([0.0931, 0.0637, 0.1212, 0.4713, 0.1029, 0.1192, 0.1346, 0.0821], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0439, 0.0430, 0.0540, 0.0429, 0.0445, 0.0427, 0.0388], + 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') +2023-02-07 05:23:38,224 INFO [train.py:901] (2/4) Epoch 23, batch 3400, loss[loss=0.2042, simple_loss=0.2957, pruned_loss=0.05631, over 8093.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2876, pruned_loss=0.06148, over 1615550.81 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:23:50,593 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-07 05:24:13,228 INFO [train.py:901] (2/4) Epoch 23, batch 3450, loss[loss=0.2547, simple_loss=0.331, pruned_loss=0.08921, over 8693.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2887, pruned_loss=0.06225, over 1618159.22 frames. ], batch size: 49, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:24:22,686 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4707, 1.9953, 1.5496, 1.7691, 1.6974, 1.4731, 1.6853, 1.7418], + device='cuda:2'), covar=tensor([0.0982, 0.0341, 0.0993, 0.0502, 0.0642, 0.1189, 0.0733, 0.0688], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0233, 0.0336, 0.0309, 0.0300, 0.0338, 0.0344, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 05:24:27,416 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.466e+02 2.960e+02 3.783e+02 8.296e+02, threshold=5.920e+02, percent-clipped=4.0 +2023-02-07 05:24:32,991 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181302.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:24:33,725 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181303.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:24:42,789 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181315.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:24:49,555 INFO [train.py:901] (2/4) Epoch 23, batch 3500, loss[loss=0.1759, simple_loss=0.2656, pruned_loss=0.04314, over 8452.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2887, pruned_loss=0.06155, over 1622339.36 frames. ], batch size: 27, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:24:52,754 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181328.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:25:07,649 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 05:25:25,791 INFO [train.py:901] (2/4) Epoch 23, batch 3550, loss[loss=0.1781, simple_loss=0.2553, pruned_loss=0.05044, over 7799.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2878, pruned_loss=0.06153, over 1617207.49 frames. ], batch size: 19, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:25:39,006 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.351e+02 2.882e+02 3.469e+02 9.271e+02, threshold=5.765e+02, percent-clipped=2.0 +2023-02-07 05:25:42,664 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9009, 1.6218, 3.3426, 1.4151, 2.4500, 3.6652, 3.7481, 3.0924], + device='cuda:2'), covar=tensor([0.1229, 0.1818, 0.0313, 0.2171, 0.0963, 0.0215, 0.0463, 0.0541], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0319, 0.0285, 0.0314, 0.0310, 0.0267, 0.0422, 0.0304], + 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') +2023-02-07 05:26:01,200 INFO [train.py:901] (2/4) Epoch 23, batch 3600, loss[loss=0.2377, simple_loss=0.3136, pruned_loss=0.08092, over 8623.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2877, pruned_loss=0.06177, over 1617636.37 frames. ], batch size: 49, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:26:30,530 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181465.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:26:37,785 INFO [train.py:901] (2/4) Epoch 23, batch 3650, loss[loss=0.201, simple_loss=0.2731, pruned_loss=0.06445, over 7790.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2884, pruned_loss=0.06181, over 1617086.32 frames. ], batch size: 19, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:26:48,314 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181490.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:26:50,925 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.424e+02 2.919e+02 3.720e+02 6.119e+02, threshold=5.839e+02, percent-clipped=1.0 +2023-02-07 05:27:11,127 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 05:27:12,459 INFO [train.py:901] (2/4) Epoch 23, batch 3700, loss[loss=0.2549, simple_loss=0.3343, pruned_loss=0.08779, over 8602.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2878, pruned_loss=0.06161, over 1615387.62 frames. ], batch size: 34, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:27:30,293 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:27:37,397 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181558.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:27:49,549 INFO [train.py:901] (2/4) Epoch 23, batch 3750, loss[loss=0.2018, simple_loss=0.2877, pruned_loss=0.05796, over 7976.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2876, pruned_loss=0.06159, over 1616415.45 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:27:55,396 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:28:01,752 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3854, 2.0938, 2.8152, 2.3176, 2.7579, 2.3658, 2.1795, 1.6902], + device='cuda:2'), covar=tensor([0.5590, 0.4959, 0.1920, 0.3697, 0.2383, 0.2799, 0.1870, 0.5104], + device='cuda:2'), in_proj_covar=tensor([0.0949, 0.0992, 0.0815, 0.0957, 0.1003, 0.0905, 0.0757, 0.0834], + 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') +2023-02-07 05:28:02,823 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.354e+02 2.844e+02 3.677e+02 7.170e+02, threshold=5.688e+02, percent-clipped=4.0 +2023-02-07 05:28:19,922 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2023-02-07 05:28:24,873 INFO [train.py:901] (2/4) Epoch 23, batch 3800, loss[loss=0.1762, simple_loss=0.2599, pruned_loss=0.0462, over 7794.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2866, pruned_loss=0.0609, over 1614769.51 frames. ], batch size: 19, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:28:35,135 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-07 05:28:49,222 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181659.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:29:00,793 INFO [train.py:901] (2/4) Epoch 23, batch 3850, loss[loss=0.1637, simple_loss=0.2554, pruned_loss=0.03604, over 7970.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2875, pruned_loss=0.0614, over 1613599.31 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:29:14,859 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 2.361e+02 2.900e+02 3.650e+02 9.007e+02, threshold=5.800e+02, percent-clipped=7.0 +2023-02-07 05:29:22,388 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 05:29:36,633 INFO [train.py:901] (2/4) Epoch 23, batch 3900, loss[loss=0.1752, simple_loss=0.2594, pruned_loss=0.04549, over 7919.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2889, pruned_loss=0.06244, over 1612296.66 frames. ], batch size: 20, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:29:53,929 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9464, 1.4975, 1.6446, 1.4226, 0.9174, 1.4824, 1.7141, 1.6421], + device='cuda:2'), covar=tensor([0.0531, 0.1264, 0.1700, 0.1458, 0.0649, 0.1488, 0.0697, 0.0617], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0188, 0.0159, 0.0100, 0.0162, 0.0111, 0.0142], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 05:30:10,483 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181774.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:30:10,999 INFO [train.py:901] (2/4) Epoch 23, batch 3950, loss[loss=0.1744, simple_loss=0.2515, pruned_loss=0.04865, over 7531.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2874, pruned_loss=0.06179, over 1610333.97 frames. ], batch size: 18, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:30:26,249 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.359e+02 2.788e+02 3.393e+02 6.824e+02, threshold=5.575e+02, percent-clipped=4.0 +2023-02-07 05:30:29,315 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7547, 1.5919, 1.8260, 1.7027, 0.9267, 1.6551, 2.0860, 1.9209], + device='cuda:2'), covar=tensor([0.0454, 0.1266, 0.1685, 0.1377, 0.0599, 0.1429, 0.0643, 0.0626], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0100, 0.0163, 0.0112, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 05:30:47,710 INFO [train.py:901] (2/4) Epoch 23, batch 4000, loss[loss=0.2479, simple_loss=0.3131, pruned_loss=0.09133, over 6762.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.287, pruned_loss=0.06071, over 1611642.06 frames. ], batch size: 71, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:30:56,754 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3331, 1.2561, 4.5687, 1.6503, 4.0018, 3.7257, 4.0945, 4.0104], + device='cuda:2'), covar=tensor([0.0813, 0.5288, 0.0518, 0.4408, 0.1211, 0.1037, 0.0695, 0.0743], + device='cuda:2'), in_proj_covar=tensor([0.0640, 0.0649, 0.0705, 0.0640, 0.0717, 0.0615, 0.0613, 0.0686], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 05:31:22,578 INFO [train.py:901] (2/4) Epoch 23, batch 4050, loss[loss=0.1793, simple_loss=0.2589, pruned_loss=0.04982, over 7418.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2862, pruned_loss=0.06048, over 1609768.83 frames. ], batch size: 17, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:31:34,365 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181892.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:31:35,722 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.508e+02 2.885e+02 3.954e+02 8.020e+02, threshold=5.770e+02, percent-clipped=6.0 +2023-02-07 05:31:59,837 INFO [train.py:901] (2/4) Epoch 23, batch 4100, loss[loss=0.2458, simple_loss=0.3234, pruned_loss=0.08408, over 8590.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2868, pruned_loss=0.06088, over 1613851.26 frames. ], batch size: 34, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:32:34,913 INFO [train.py:901] (2/4) Epoch 23, batch 4150, loss[loss=0.166, simple_loss=0.2517, pruned_loss=0.04014, over 8241.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2862, pruned_loss=0.06053, over 1614304.69 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:32:48,433 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.356e+02 2.929e+02 3.956e+02 6.697e+02, threshold=5.858e+02, percent-clipped=3.0 +2023-02-07 05:32:49,334 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181996.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:32:58,751 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182007.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:33:11,759 INFO [train.py:901] (2/4) Epoch 23, batch 4200, loss[loss=0.2081, simple_loss=0.2881, pruned_loss=0.06402, over 8297.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2874, pruned_loss=0.06113, over 1618405.46 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:33:14,747 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6673, 1.6293, 2.3046, 1.6149, 1.3274, 2.2218, 0.4213, 1.4370], + device='cuda:2'), covar=tensor([0.1670, 0.1252, 0.0270, 0.1024, 0.2444, 0.0387, 0.1908, 0.1339], + device='cuda:2'), in_proj_covar=tensor([0.0191, 0.0199, 0.0128, 0.0219, 0.0267, 0.0136, 0.0168, 0.0191], + 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') +2023-02-07 05:33:16,200 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182030.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:33:25,750 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 05:33:33,429 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182055.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:33:47,600 INFO [train.py:901] (2/4) Epoch 23, batch 4250, loss[loss=0.1898, simple_loss=0.2865, pruned_loss=0.0466, over 8294.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2862, pruned_loss=0.06087, over 1616674.50 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:33:49,041 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 05:34:01,344 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.416e+02 2.989e+02 3.588e+02 6.339e+02, threshold=5.979e+02, percent-clipped=2.0 +2023-02-07 05:34:22,740 INFO [train.py:901] (2/4) Epoch 23, batch 4300, loss[loss=0.2056, simple_loss=0.2931, pruned_loss=0.05904, over 8355.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2851, pruned_loss=0.06029, over 1613739.64 frames. ], batch size: 24, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:34:58,708 INFO [train.py:901] (2/4) Epoch 23, batch 4350, loss[loss=0.2199, simple_loss=0.3014, pruned_loss=0.06916, over 8461.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2864, pruned_loss=0.06066, over 1616832.73 frames. ], batch size: 29, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:35:13,006 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0042, 2.2025, 1.8619, 2.7335, 1.2540, 1.6452, 1.9476, 2.2215], + device='cuda:2'), covar=tensor([0.0674, 0.0736, 0.0822, 0.0341, 0.1132, 0.1234, 0.0819, 0.0750], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0199, 0.0246, 0.0215, 0.0208, 0.0249, 0.0252, 0.0210], + 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') +2023-02-07 05:35:13,488 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.346e+02 2.960e+02 3.931e+02 9.702e+02, threshold=5.919e+02, percent-clipped=9.0 +2023-02-07 05:35:21,958 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 05:35:34,678 INFO [train.py:901] (2/4) Epoch 23, batch 4400, loss[loss=0.1867, simple_loss=0.2706, pruned_loss=0.05138, over 8135.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2861, pruned_loss=0.06027, over 1616494.81 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:35:46,164 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5005, 1.8960, 2.9049, 1.3522, 2.1360, 1.8639, 1.5819, 2.2019], + device='cuda:2'), covar=tensor([0.1992, 0.2664, 0.0867, 0.4694, 0.1981, 0.3368, 0.2428, 0.2257], + device='cuda:2'), in_proj_covar=tensor([0.0525, 0.0607, 0.0554, 0.0647, 0.0645, 0.0596, 0.0539, 0.0630], + 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') +2023-02-07 05:36:03,255 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182263.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:36:05,073 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 05:36:11,441 INFO [train.py:901] (2/4) Epoch 23, batch 4450, loss[loss=0.1944, simple_loss=0.279, pruned_loss=0.05488, over 8244.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2859, pruned_loss=0.06021, over 1615502.18 frames. ], batch size: 24, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:36:20,472 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182288.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:36:26,033 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.606e+02 3.225e+02 4.349e+02 9.132e+02, threshold=6.449e+02, percent-clipped=7.0 +2023-02-07 05:36:28,977 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182299.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:36:47,060 INFO [train.py:901] (2/4) Epoch 23, batch 4500, loss[loss=0.2457, simple_loss=0.3327, pruned_loss=0.07937, over 8470.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2853, pruned_loss=0.06016, over 1612022.61 frames. ], batch size: 25, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:36:56,852 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 05:36:57,584 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182340.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:37:14,933 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1067, 1.7893, 2.3487, 1.9955, 2.2919, 2.1318, 1.9845, 1.1542], + device='cuda:2'), covar=tensor([0.5898, 0.5036, 0.2032, 0.3707, 0.2379, 0.3033, 0.1887, 0.5070], + device='cuda:2'), in_proj_covar=tensor([0.0949, 0.0996, 0.0816, 0.0957, 0.1002, 0.0905, 0.0757, 0.0834], + 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') +2023-02-07 05:37:22,880 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 05:37:23,620 INFO [train.py:901] (2/4) Epoch 23, batch 4550, loss[loss=0.2495, simple_loss=0.3258, pruned_loss=0.0866, over 8610.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2862, pruned_loss=0.06118, over 1613806.48 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:37:34,953 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7578, 2.0286, 1.7886, 2.6380, 1.1896, 1.5296, 1.8843, 2.1049], + device='cuda:2'), covar=tensor([0.0822, 0.0766, 0.0860, 0.0364, 0.1107, 0.1313, 0.0813, 0.0750], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0198, 0.0246, 0.0215, 0.0208, 0.0249, 0.0252, 0.0209], + 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') +2023-02-07 05:37:37,489 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.347e+02 2.810e+02 3.651e+02 9.685e+02, threshold=5.619e+02, percent-clipped=2.0 +2023-02-07 05:37:46,727 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.44 vs. limit=5.0 +2023-02-07 05:37:59,258 INFO [train.py:901] (2/4) Epoch 23, batch 4600, loss[loss=0.1943, simple_loss=0.2811, pruned_loss=0.05377, over 8375.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2858, pruned_loss=0.061, over 1611464.97 frames. ], batch size: 49, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:38:07,959 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5909, 2.1129, 3.1600, 1.4727, 2.4723, 2.0433, 1.6701, 2.3645], + device='cuda:2'), covar=tensor([0.1847, 0.2400, 0.0897, 0.4411, 0.1696, 0.3047, 0.2340, 0.2210], + device='cuda:2'), in_proj_covar=tensor([0.0523, 0.0605, 0.0550, 0.0643, 0.0641, 0.0590, 0.0536, 0.0626], + 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') +2023-02-07 05:38:20,353 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182455.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:38:26,828 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.98 vs. limit=5.0 +2023-02-07 05:38:34,856 INFO [train.py:901] (2/4) Epoch 23, batch 4650, loss[loss=0.1996, simple_loss=0.2811, pruned_loss=0.05904, over 8195.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2853, pruned_loss=0.06079, over 1609911.49 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:38:50,626 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.204e+02 2.647e+02 3.638e+02 6.712e+02, threshold=5.294e+02, percent-clipped=7.0 +2023-02-07 05:39:12,439 INFO [train.py:901] (2/4) Epoch 23, batch 4700, loss[loss=0.2008, simple_loss=0.2919, pruned_loss=0.05487, over 8700.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2846, pruned_loss=0.06049, over 1604273.37 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:39:17,387 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182532.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:39:47,030 INFO [train.py:901] (2/4) Epoch 23, batch 4750, loss[loss=0.1702, simple_loss=0.2498, pruned_loss=0.04528, over 7190.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2849, pruned_loss=0.06054, over 1604168.41 frames. ], batch size: 16, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:40:01,608 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.297e+02 2.902e+02 3.418e+02 7.225e+02, threshold=5.805e+02, percent-clipped=3.0 +2023-02-07 05:40:05,960 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 05:40:08,826 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 05:40:11,282 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.41 vs. limit=5.0 +2023-02-07 05:40:12,598 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3987, 1.6760, 1.7137, 1.1487, 1.7413, 1.4169, 0.3131, 1.6214], + device='cuda:2'), covar=tensor([0.0471, 0.0347, 0.0294, 0.0484, 0.0452, 0.0882, 0.0923, 0.0280], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0394, 0.0348, 0.0449, 0.0383, 0.0536, 0.0394, 0.0423], + device='cuda:2'), out_proj_covar=tensor([1.2220e-04, 1.0303e-04, 9.1370e-05, 1.1794e-04, 1.0090e-04, 1.5095e-04, + 1.0622e-04, 1.1179e-04], device='cuda:2') +2023-02-07 05:40:24,102 INFO [train.py:901] (2/4) Epoch 23, batch 4800, loss[loss=0.196, simple_loss=0.2849, pruned_loss=0.05353, over 8031.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2842, pruned_loss=0.06019, over 1606400.19 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:40:36,438 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:40:59,186 INFO [train.py:901] (2/4) Epoch 23, batch 4850, loss[loss=0.1826, simple_loss=0.2749, pruned_loss=0.04508, over 8304.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2838, pruned_loss=0.05966, over 1605118.89 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:41:00,602 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 05:41:00,763 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0389, 1.6146, 3.2428, 1.4376, 2.3082, 3.5836, 3.6965, 2.9611], + device='cuda:2'), covar=tensor([0.1203, 0.1801, 0.0422, 0.2287, 0.1123, 0.0279, 0.0695, 0.0685], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0320, 0.0286, 0.0315, 0.0311, 0.0268, 0.0423, 0.0304], + 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') +2023-02-07 05:41:13,247 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.288e+02 2.781e+02 3.814e+02 7.165e+02, threshold=5.562e+02, percent-clipped=4.0 +2023-02-07 05:41:17,124 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3944, 1.2709, 1.6561, 1.1557, 1.1066, 1.6362, 0.3009, 1.1131], + device='cuda:2'), covar=tensor([0.1502, 0.1264, 0.0455, 0.0969, 0.2413, 0.0522, 0.2233, 0.1371], + device='cuda:2'), in_proj_covar=tensor([0.0191, 0.0197, 0.0129, 0.0219, 0.0268, 0.0136, 0.0168, 0.0192], + 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') +2023-02-07 05:41:25,677 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182711.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:41:36,258 INFO [train.py:901] (2/4) Epoch 23, batch 4900, loss[loss=0.2204, simple_loss=0.302, pruned_loss=0.06941, over 8421.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2842, pruned_loss=0.05994, over 1608811.21 frames. ], batch size: 27, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:41:44,975 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182736.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:41:50,003 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3283, 2.0023, 2.6140, 2.1552, 2.5279, 2.3198, 2.1344, 1.3444], + device='cuda:2'), covar=tensor([0.5065, 0.4581, 0.1816, 0.3779, 0.2511, 0.3040, 0.1980, 0.5328], + device='cuda:2'), in_proj_covar=tensor([0.0938, 0.0985, 0.0806, 0.0946, 0.0993, 0.0896, 0.0750, 0.0827], + 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') +2023-02-07 05:42:00,366 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182758.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:42:09,168 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-07 05:42:12,967 INFO [train.py:901] (2/4) Epoch 23, batch 4950, loss[loss=0.2144, simple_loss=0.3, pruned_loss=0.06436, over 8345.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2831, pruned_loss=0.05937, over 1608738.00 frames. ], batch size: 26, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:42:27,043 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.423e+02 2.989e+02 3.745e+02 1.524e+03, threshold=5.977e+02, percent-clipped=7.0 +2023-02-07 05:42:27,228 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3810, 1.3488, 2.3195, 1.2333, 2.1868, 2.5144, 2.7029, 2.0469], + device='cuda:2'), covar=tensor([0.1302, 0.1517, 0.0465, 0.2234, 0.0803, 0.0394, 0.0579, 0.0784], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0318, 0.0285, 0.0314, 0.0310, 0.0267, 0.0421, 0.0302], + 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') +2023-02-07 05:42:33,127 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6897, 2.4828, 3.2309, 2.6490, 3.0901, 2.6207, 2.5693, 2.3491], + device='cuda:2'), covar=tensor([0.3905, 0.4222, 0.1604, 0.3059, 0.1928, 0.2423, 0.1415, 0.3950], + device='cuda:2'), in_proj_covar=tensor([0.0939, 0.0985, 0.0806, 0.0947, 0.0993, 0.0895, 0.0750, 0.0827], + 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') +2023-02-07 05:42:37,324 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2863, 2.6082, 2.9455, 1.7303, 3.2032, 1.9254, 1.5677, 2.1072], + device='cuda:2'), covar=tensor([0.0783, 0.0419, 0.0327, 0.0794, 0.0427, 0.0869, 0.0868, 0.0599], + device='cuda:2'), in_proj_covar=tensor([0.0455, 0.0395, 0.0348, 0.0448, 0.0383, 0.0537, 0.0393, 0.0422], + device='cuda:2'), out_proj_covar=tensor([1.2140e-04, 1.0336e-04, 9.1409e-05, 1.1769e-04, 1.0070e-04, 1.5131e-04, + 1.0595e-04, 1.1145e-04], device='cuda:2') +2023-02-07 05:42:48,229 INFO [train.py:901] (2/4) Epoch 23, batch 5000, loss[loss=0.2006, simple_loss=0.2951, pruned_loss=0.05301, over 8245.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.283, pruned_loss=0.05907, over 1603250.85 frames. ], batch size: 24, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:43:25,219 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182874.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:43:25,722 INFO [train.py:901] (2/4) Epoch 23, batch 5050, loss[loss=0.2712, simple_loss=0.3367, pruned_loss=0.1029, over 8345.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2842, pruned_loss=0.06007, over 1610786.08 frames. ], batch size: 49, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:43:26,556 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:43:29,541 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182880.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:43:40,729 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.376e+02 2.932e+02 3.646e+02 6.966e+02, threshold=5.864e+02, percent-clipped=3.0 +2023-02-07 05:43:46,349 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 05:43:57,094 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7615, 1.8743, 1.6847, 2.3403, 1.1184, 1.4506, 1.7573, 1.8899], + device='cuda:2'), covar=tensor([0.0832, 0.0789, 0.0938, 0.0413, 0.1086, 0.1451, 0.0792, 0.0802], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0197, 0.0244, 0.0215, 0.0206, 0.0246, 0.0250, 0.0206], + 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') +2023-02-07 05:44:01,718 INFO [train.py:901] (2/4) Epoch 23, batch 5100, loss[loss=0.183, simple_loss=0.2723, pruned_loss=0.04689, over 8243.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2835, pruned_loss=0.0595, over 1612378.48 frames. ], batch size: 24, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:44:38,868 INFO [train.py:901] (2/4) Epoch 23, batch 5150, loss[loss=0.2169, simple_loss=0.2942, pruned_loss=0.06977, over 8032.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2845, pruned_loss=0.06013, over 1612875.10 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:44:50,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182991.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:44:53,603 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 2.409e+02 2.843e+02 3.449e+02 6.604e+02, threshold=5.686e+02, percent-clipped=1.0 +2023-02-07 05:44:55,928 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5534, 1.4309, 1.6327, 1.3453, 0.8778, 1.4212, 1.4699, 1.2151], + device='cuda:2'), covar=tensor([0.0591, 0.1242, 0.1615, 0.1423, 0.0645, 0.1442, 0.0736, 0.0690], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0160, 0.0101, 0.0162, 0.0112, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 05:45:07,264 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183014.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:45:14,590 INFO [train.py:901] (2/4) Epoch 23, batch 5200, loss[loss=0.2255, simple_loss=0.3035, pruned_loss=0.07381, over 8246.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2855, pruned_loss=0.06033, over 1615587.14 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:45:20,497 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 05:45:24,447 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183039.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:45:46,615 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 05:45:50,592 INFO [train.py:901] (2/4) Epoch 23, batch 5250, loss[loss=0.1875, simple_loss=0.2826, pruned_loss=0.04615, over 8242.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2856, pruned_loss=0.06027, over 1615584.88 frames. ], batch size: 24, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:45:55,889 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 05:46:05,165 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.492e+02 2.942e+02 3.798e+02 7.403e+02, threshold=5.885e+02, percent-clipped=3.0 +2023-02-07 05:46:27,065 INFO [train.py:901] (2/4) Epoch 23, batch 5300, loss[loss=0.2151, simple_loss=0.292, pruned_loss=0.06911, over 8696.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2868, pruned_loss=0.06099, over 1620350.43 frames. ], batch size: 34, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:46:38,061 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 05:46:48,959 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1050, 1.7761, 2.3051, 1.9479, 2.2633, 2.1089, 1.9155, 1.0723], + device='cuda:2'), covar=tensor([0.5409, 0.4755, 0.1992, 0.3782, 0.2397, 0.3079, 0.1993, 0.5116], + device='cuda:2'), in_proj_covar=tensor([0.0942, 0.0988, 0.0806, 0.0949, 0.0997, 0.0898, 0.0752, 0.0829], + 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') +2023-02-07 05:46:56,901 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6901, 2.1272, 3.4949, 1.8140, 1.5663, 3.4462, 0.5624, 2.0545], + device='cuda:2'), covar=tensor([0.1381, 0.1282, 0.0213, 0.1550, 0.2863, 0.0249, 0.2223, 0.1368], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0195, 0.0127, 0.0216, 0.0266, 0.0134, 0.0167, 0.0190], + 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') +2023-02-07 05:47:02,889 INFO [train.py:901] (2/4) Epoch 23, batch 5350, loss[loss=0.211, simple_loss=0.2907, pruned_loss=0.06566, over 8233.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2873, pruned_loss=0.06183, over 1620601.57 frames. ], batch size: 22, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:47:17,713 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.502e+02 3.193e+02 3.793e+02 7.809e+02, threshold=6.385e+02, percent-clipped=1.0 +2023-02-07 05:47:34,656 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183218.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:47:38,864 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:47:39,481 INFO [train.py:901] (2/4) Epoch 23, batch 5400, loss[loss=0.2217, simple_loss=0.3076, pruned_loss=0.06791, over 8031.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.287, pruned_loss=0.06092, over 1621505.07 frames. ], batch size: 22, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:47:55,899 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183247.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:48:10,988 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5217, 1.8938, 1.9308, 1.0617, 1.9382, 1.4667, 0.4418, 1.7638], + device='cuda:2'), covar=tensor([0.0615, 0.0351, 0.0319, 0.0618, 0.0444, 0.0924, 0.0938, 0.0307], + device='cuda:2'), in_proj_covar=tensor([0.0457, 0.0395, 0.0348, 0.0449, 0.0384, 0.0538, 0.0394, 0.0424], + device='cuda:2'), out_proj_covar=tensor([1.2197e-04, 1.0329e-04, 9.1275e-05, 1.1787e-04, 1.0107e-04, 1.5149e-04, + 1.0619e-04, 1.1176e-04], device='cuda:2') +2023-02-07 05:48:13,005 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183272.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:48:14,971 INFO [train.py:901] (2/4) Epoch 23, batch 5450, loss[loss=0.1783, simple_loss=0.2532, pruned_loss=0.05167, over 7440.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2861, pruned_loss=0.06055, over 1616143.44 frames. ], batch size: 17, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:48:15,404 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 05:48:30,404 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.334e+02 2.819e+02 3.622e+02 6.725e+02, threshold=5.637e+02, percent-clipped=1.0 +2023-02-07 05:48:41,152 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 05:48:52,623 INFO [train.py:901] (2/4) Epoch 23, batch 5500, loss[loss=0.1812, simple_loss=0.2628, pruned_loss=0.04984, over 8199.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2863, pruned_loss=0.06093, over 1615132.09 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:48:54,993 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6540, 2.4659, 3.2600, 2.6571, 3.3117, 2.6414, 2.4631, 2.0712], + device='cuda:2'), covar=tensor([0.5354, 0.4966, 0.1925, 0.3813, 0.2375, 0.2927, 0.1752, 0.5324], + device='cuda:2'), in_proj_covar=tensor([0.0941, 0.0985, 0.0805, 0.0946, 0.0994, 0.0896, 0.0750, 0.0828], + 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') +2023-02-07 05:48:55,604 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8971, 1.5579, 1.7149, 1.4207, 1.0297, 1.5258, 1.7056, 1.4764], + device='cuda:2'), covar=tensor([0.0558, 0.1171, 0.1629, 0.1438, 0.0613, 0.1368, 0.0684, 0.0669], + device='cuda:2'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0160, 0.0101, 0.0162, 0.0112, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 05:48:55,691 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2807, 1.9841, 2.6199, 2.1638, 2.6141, 2.3084, 2.1044, 1.4343], + device='cuda:2'), covar=tensor([0.5155, 0.4742, 0.1934, 0.4003, 0.2368, 0.2974, 0.1886, 0.5340], + device='cuda:2'), in_proj_covar=tensor([0.0941, 0.0985, 0.0805, 0.0946, 0.0994, 0.0896, 0.0750, 0.0827], + 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') +2023-02-07 05:48:58,326 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183333.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:49:02,352 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183339.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:49:27,073 INFO [train.py:901] (2/4) Epoch 23, batch 5550, loss[loss=0.1926, simple_loss=0.2909, pruned_loss=0.04721, over 8197.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2853, pruned_loss=0.0604, over 1613367.81 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:49:41,517 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.428e+02 3.119e+02 4.010e+02 1.058e+03, threshold=6.238e+02, percent-clipped=9.0 +2023-02-07 05:49:46,633 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5987, 1.8827, 1.9561, 1.2079, 2.0790, 1.4804, 0.6591, 1.8646], + device='cuda:2'), covar=tensor([0.0635, 0.0371, 0.0270, 0.0564, 0.0417, 0.0835, 0.0877, 0.0281], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0397, 0.0350, 0.0450, 0.0385, 0.0539, 0.0395, 0.0425], + device='cuda:2'), out_proj_covar=tensor([1.2231e-04, 1.0380e-04, 9.1876e-05, 1.1820e-04, 1.0126e-04, 1.5191e-04, + 1.0653e-04, 1.1211e-04], device='cuda:2') +2023-02-07 05:50:03,264 INFO [train.py:901] (2/4) Epoch 23, batch 5600, loss[loss=0.1676, simple_loss=0.2471, pruned_loss=0.04405, over 7527.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2853, pruned_loss=0.06062, over 1615564.68 frames. ], batch size: 18, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:50:04,084 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183426.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:50:39,035 INFO [train.py:901] (2/4) Epoch 23, batch 5650, loss[loss=0.1842, simple_loss=0.2631, pruned_loss=0.05269, over 7216.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2849, pruned_loss=0.0609, over 1611164.72 frames. ], batch size: 16, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:50:51,427 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 05:50:53,308 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.300e+02 3.084e+02 3.921e+02 7.530e+02, threshold=6.168e+02, percent-clipped=4.0 +2023-02-07 05:51:14,120 INFO [train.py:901] (2/4) Epoch 23, batch 5700, loss[loss=0.1743, simple_loss=0.2485, pruned_loss=0.05009, over 7690.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2865, pruned_loss=0.06093, over 1614978.65 frames. ], batch size: 18, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:51:50,329 INFO [train.py:901] (2/4) Epoch 23, batch 5750, loss[loss=0.1929, simple_loss=0.2751, pruned_loss=0.05531, over 8472.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2861, pruned_loss=0.06083, over 1614017.68 frames. ], batch size: 27, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:51:57,126 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 05:52:00,847 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183589.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:05,020 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183595.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:05,469 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.334e+02 3.030e+02 3.740e+02 1.347e+03, threshold=6.060e+02, percent-clipped=7.0 +2023-02-07 05:52:18,041 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183614.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:22,090 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183620.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:25,344 INFO [train.py:901] (2/4) Epoch 23, batch 5800, loss[loss=0.2103, simple_loss=0.295, pruned_loss=0.06287, over 8324.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2865, pruned_loss=0.06115, over 1619618.60 frames. ], batch size: 25, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:52:30,172 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:53:01,085 INFO [train.py:901] (2/4) Epoch 23, batch 5850, loss[loss=0.2477, simple_loss=0.3185, pruned_loss=0.08843, over 8665.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.287, pruned_loss=0.06101, over 1619957.62 frames. ], batch size: 50, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:53:16,206 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.232e+02 2.367e+02 2.798e+02 3.640e+02 5.951e+02, threshold=5.597e+02, percent-clipped=0.0 +2023-02-07 05:53:36,770 INFO [train.py:901] (2/4) Epoch 23, batch 5900, loss[loss=0.1735, simple_loss=0.2507, pruned_loss=0.04814, over 7696.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.286, pruned_loss=0.06053, over 1620568.86 frames. ], batch size: 18, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:53:58,119 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9769, 1.4950, 3.1171, 1.5515, 2.8192, 2.6354, 2.8440, 2.7755], + device='cuda:2'), covar=tensor([0.0747, 0.3291, 0.0794, 0.3531, 0.0899, 0.0891, 0.0599, 0.0660], + device='cuda:2'), in_proj_covar=tensor([0.0643, 0.0650, 0.0705, 0.0637, 0.0716, 0.0614, 0.0612, 0.0690], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 05:54:08,422 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183770.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:54:11,753 INFO [train.py:901] (2/4) Epoch 23, batch 5950, loss[loss=0.1995, simple_loss=0.281, pruned_loss=0.05903, over 7652.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2864, pruned_loss=0.06064, over 1616830.98 frames. ], batch size: 19, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:54:24,493 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5852, 1.4839, 1.7633, 1.3545, 0.8418, 1.5070, 1.4709, 1.4989], + device='cuda:2'), covar=tensor([0.0562, 0.1188, 0.1604, 0.1449, 0.0585, 0.1405, 0.0680, 0.0630], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0100, 0.0162, 0.0111, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:2') +2023-02-07 05:54:27,017 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.423e+02 2.784e+02 3.423e+02 5.836e+02, threshold=5.567e+02, percent-clipped=2.0 +2023-02-07 05:54:39,709 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 05:54:47,607 INFO [train.py:901] (2/4) Epoch 23, batch 6000, loss[loss=0.1723, simple_loss=0.2501, pruned_loss=0.04729, over 7930.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2872, pruned_loss=0.06139, over 1619570.59 frames. ], batch size: 20, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:54:47,607 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 05:55:00,697 INFO [train.py:935] (2/4) Epoch 23, validation: loss=0.1722, simple_loss=0.2724, pruned_loss=0.03597, over 944034.00 frames. +2023-02-07 05:55:00,698 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 05:55:25,818 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183860.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:55:29,284 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8667, 3.8200, 3.4687, 2.0289, 3.4191, 3.3898, 3.3623, 3.3228], + device='cuda:2'), covar=tensor([0.0957, 0.0695, 0.1241, 0.4487, 0.1099, 0.1268, 0.1587, 0.1060], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0444, 0.0431, 0.0540, 0.0431, 0.0447, 0.0430, 0.0387], + 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') +2023-02-07 05:55:36,130 INFO [train.py:901] (2/4) Epoch 23, batch 6050, loss[loss=0.1793, simple_loss=0.2517, pruned_loss=0.05345, over 7664.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2873, pruned_loss=0.06154, over 1616022.66 frames. ], batch size: 19, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:55:43,224 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183885.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:55:50,617 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.465e+02 3.097e+02 3.782e+02 8.398e+02, threshold=6.194e+02, percent-clipped=6.0 +2023-02-07 05:56:11,859 INFO [train.py:901] (2/4) Epoch 23, batch 6100, loss[loss=0.2262, simple_loss=0.3021, pruned_loss=0.0752, over 8607.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06199, over 1617158.85 frames. ], batch size: 31, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:56:32,460 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 05:56:47,365 INFO [train.py:901] (2/4) Epoch 23, batch 6150, loss[loss=0.2213, simple_loss=0.3062, pruned_loss=0.06821, over 8661.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.287, pruned_loss=0.0622, over 1614107.50 frames. ], batch size: 34, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:56:48,166 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:57:01,785 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.512e+02 2.876e+02 3.577e+02 6.799e+02, threshold=5.752e+02, percent-clipped=2.0 +2023-02-07 05:57:12,086 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0296, 2.2777, 3.7487, 1.9503, 1.8434, 3.6978, 0.8731, 2.2737], + device='cuda:2'), covar=tensor([0.1200, 0.1459, 0.0154, 0.1741, 0.2593, 0.0178, 0.1938, 0.1197], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0196, 0.0128, 0.0219, 0.0268, 0.0135, 0.0168, 0.0190], + 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') +2023-02-07 05:57:17,560 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8308, 3.8009, 3.4499, 1.8693, 3.4114, 3.4285, 3.4010, 3.2925], + device='cuda:2'), covar=tensor([0.0911, 0.0677, 0.1163, 0.4676, 0.0939, 0.1503, 0.1400, 0.1010], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0447, 0.0433, 0.0543, 0.0432, 0.0450, 0.0432, 0.0389], + 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') +2023-02-07 05:57:22,985 INFO [train.py:901] (2/4) Epoch 23, batch 6200, loss[loss=0.2291, simple_loss=0.3064, pruned_loss=0.07585, over 8643.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.287, pruned_loss=0.06167, over 1618239.23 frames. ], batch size: 34, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:57:54,993 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184068.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:57:59,614 INFO [train.py:901] (2/4) Epoch 23, batch 6250, loss[loss=0.1848, simple_loss=0.2721, pruned_loss=0.04876, over 8296.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2852, pruned_loss=0.06056, over 1612546.25 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:58:07,975 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0524, 1.7413, 6.2857, 2.2400, 5.6489, 5.2744, 5.7711, 5.6873], + device='cuda:2'), covar=tensor([0.0423, 0.4770, 0.0254, 0.3788, 0.0794, 0.0819, 0.0472, 0.0482], + device='cuda:2'), in_proj_covar=tensor([0.0644, 0.0651, 0.0707, 0.0641, 0.0719, 0.0616, 0.0617, 0.0693], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 05:58:11,448 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:58:14,638 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.476e+02 2.340e+02 2.866e+02 3.425e+02 5.984e+02, threshold=5.731e+02, percent-clipped=3.0 +2023-02-07 05:58:34,471 INFO [train.py:901] (2/4) Epoch 23, batch 6300, loss[loss=0.1679, simple_loss=0.2468, pruned_loss=0.04452, over 7432.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2852, pruned_loss=0.06026, over 1614005.97 frames. ], batch size: 17, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:58:45,745 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:59:04,550 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184166.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:59:10,598 INFO [train.py:901] (2/4) Epoch 23, batch 6350, loss[loss=0.1966, simple_loss=0.2742, pruned_loss=0.0595, over 7529.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2845, pruned_loss=0.05968, over 1616991.19 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 05:59:25,790 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.298e+02 2.703e+02 3.593e+02 9.198e+02, threshold=5.406e+02, percent-clipped=6.0 +2023-02-07 05:59:28,534 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 05:59:32,342 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184204.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:59:46,844 INFO [train.py:901] (2/4) Epoch 23, batch 6400, loss[loss=0.1614, simple_loss=0.2476, pruned_loss=0.03757, over 8235.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.285, pruned_loss=0.06029, over 1615628.58 frames. ], batch size: 22, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:00:22,071 INFO [train.py:901] (2/4) Epoch 23, batch 6450, loss[loss=0.2487, simple_loss=0.3279, pruned_loss=0.08473, over 8689.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2862, pruned_loss=0.06098, over 1615934.34 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:00:37,219 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.429e+02 3.055e+02 3.904e+02 7.071e+02, threshold=6.109e+02, percent-clipped=5.0 +2023-02-07 06:00:54,549 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184319.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:00:58,474 INFO [train.py:901] (2/4) Epoch 23, batch 6500, loss[loss=0.1782, simple_loss=0.2657, pruned_loss=0.04538, over 7801.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2865, pruned_loss=0.06106, over 1614467.17 frames. ], batch size: 20, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:01:13,631 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184347.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:01:30,746 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184372.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:01:32,660 INFO [train.py:901] (2/4) Epoch 23, batch 6550, loss[loss=0.1895, simple_loss=0.2674, pruned_loss=0.0558, over 7977.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2863, pruned_loss=0.06048, over 1615820.08 frames. ], batch size: 21, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:01:34,222 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:01:48,099 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.251e+02 2.720e+02 3.518e+02 7.175e+02, threshold=5.440e+02, percent-clipped=6.0 +2023-02-07 06:01:51,603 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 06:02:00,104 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184412.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:02:07,283 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2395, 2.5927, 2.8996, 1.6258, 3.2686, 2.0044, 1.5724, 2.1696], + device='cuda:2'), covar=tensor([0.0927, 0.0405, 0.0306, 0.0849, 0.0453, 0.0899, 0.1017, 0.0617], + device='cuda:2'), in_proj_covar=tensor([0.0455, 0.0396, 0.0349, 0.0448, 0.0381, 0.0536, 0.0392, 0.0425], + device='cuda:2'), out_proj_covar=tensor([1.2144e-04, 1.0353e-04, 9.1555e-05, 1.1783e-04, 1.0023e-04, 1.5082e-04, + 1.0563e-04, 1.1218e-04], device='cuda:2') +2023-02-07 06:02:09,827 INFO [train.py:901] (2/4) Epoch 23, batch 6600, loss[loss=0.1778, simple_loss=0.2596, pruned_loss=0.04804, over 7302.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2857, pruned_loss=0.06034, over 1614463.40 frames. ], batch size: 16, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:02:09,865 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 06:02:20,790 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-07 06:02:31,490 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7136, 1.6823, 2.1224, 1.7040, 1.0821, 1.8388, 2.1958, 2.1263], + device='cuda:2'), covar=tensor([0.0489, 0.1177, 0.1556, 0.1385, 0.0546, 0.1289, 0.0618, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0160, 0.0101, 0.0163, 0.0111, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 06:02:45,210 INFO [train.py:901] (2/4) Epoch 23, batch 6650, loss[loss=0.2248, simple_loss=0.3134, pruned_loss=0.06814, over 8459.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2865, pruned_loss=0.06049, over 1615842.78 frames. ], batch size: 27, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:03:00,392 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.187e+02 2.636e+02 3.150e+02 7.164e+02, threshold=5.273e+02, percent-clipped=1.0 +2023-02-07 06:03:06,026 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184504.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:03:21,187 INFO [train.py:901] (2/4) Epoch 23, batch 6700, loss[loss=0.2161, simple_loss=0.3029, pruned_loss=0.0646, over 8037.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2866, pruned_loss=0.06055, over 1616372.10 frames. ], batch size: 22, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:03:22,782 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184527.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:03:56,972 INFO [train.py:901] (2/4) Epoch 23, batch 6750, loss[loss=0.2358, simple_loss=0.3109, pruned_loss=0.08036, over 6791.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2857, pruned_loss=0.06019, over 1613056.60 frames. ], batch size: 72, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:03:57,212 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184575.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:04:11,515 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.299e+02 2.705e+02 3.689e+02 1.087e+03, threshold=5.410e+02, percent-clipped=6.0 +2023-02-07 06:04:14,467 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184600.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:04:30,946 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 06:04:32,277 INFO [train.py:901] (2/4) Epoch 23, batch 6800, loss[loss=0.2256, simple_loss=0.2996, pruned_loss=0.07579, over 7640.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2855, pruned_loss=0.05988, over 1615565.76 frames. ], batch size: 19, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:04:55,642 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 06:05:08,933 INFO [train.py:901] (2/4) Epoch 23, batch 6850, loss[loss=0.1682, simple_loss=0.2398, pruned_loss=0.04832, over 6819.00 frames. ], tot_loss[loss=0.203, simple_loss=0.286, pruned_loss=0.05998, over 1614776.62 frames. ], batch size: 15, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:05:19,332 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 06:05:23,524 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.645e+02 3.100e+02 4.179e+02 7.238e+02, threshold=6.201e+02, percent-clipped=8.0 +2023-02-07 06:05:40,732 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184721.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:05:43,468 INFO [train.py:901] (2/4) Epoch 23, batch 6900, loss[loss=0.1886, simple_loss=0.27, pruned_loss=0.05357, over 7917.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2859, pruned_loss=0.06016, over 1613818.52 frames. ], batch size: 20, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:06:21,332 INFO [train.py:901] (2/4) Epoch 23, batch 6950, loss[loss=0.1692, simple_loss=0.2423, pruned_loss=0.04804, over 7713.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2852, pruned_loss=0.05982, over 1615612.31 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:06:27,133 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:06:29,044 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 06:06:35,999 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.334e+02 2.863e+02 3.573e+02 6.345e+02, threshold=5.727e+02, percent-clipped=1.0 +2023-02-07 06:06:38,089 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4801, 4.4775, 4.0595, 2.0947, 4.0187, 4.0162, 3.9965, 3.9035], + device='cuda:2'), covar=tensor([0.0683, 0.0502, 0.0926, 0.3955, 0.0842, 0.0939, 0.1300, 0.0792], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0447, 0.0431, 0.0544, 0.0434, 0.0448, 0.0430, 0.0389], + 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') +2023-02-07 06:06:44,107 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.36 vs. limit=5.0 +2023-02-07 06:06:44,536 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184808.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:06:56,217 INFO [train.py:901] (2/4) Epoch 23, batch 7000, loss[loss=0.1937, simple_loss=0.287, pruned_loss=0.05024, over 8541.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2846, pruned_loss=0.05965, over 1608823.07 frames. ], batch size: 28, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:07:03,957 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184836.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:07:12,025 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184848.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:07:32,115 INFO [train.py:901] (2/4) Epoch 23, batch 7050, loss[loss=0.2098, simple_loss=0.2932, pruned_loss=0.06319, over 8456.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2847, pruned_loss=0.05951, over 1609332.60 frames. ], batch size: 27, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:07:38,586 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184884.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:07:48,061 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.247e+02 2.854e+02 3.580e+02 1.056e+03, threshold=5.709e+02, percent-clipped=4.0 +2023-02-07 06:07:53,398 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.47 vs. limit=5.0 +2023-02-07 06:08:08,157 INFO [train.py:901] (2/4) Epoch 23, batch 7100, loss[loss=0.1995, simple_loss=0.2867, pruned_loss=0.05615, over 8035.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2851, pruned_loss=0.06007, over 1608170.97 frames. ], batch size: 22, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:08:30,174 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184957.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:08:34,373 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184963.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:08:42,473 INFO [train.py:901] (2/4) Epoch 23, batch 7150, loss[loss=0.1926, simple_loss=0.2888, pruned_loss=0.04819, over 8333.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2843, pruned_loss=0.05954, over 1605696.14 frames. ], batch size: 25, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:08:53,467 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0924, 1.8615, 2.3540, 2.0355, 2.3080, 2.1573, 1.9436, 1.1697], + device='cuda:2'), covar=tensor([0.5486, 0.4703, 0.1891, 0.3330, 0.2369, 0.2896, 0.1840, 0.4910], + device='cuda:2'), in_proj_covar=tensor([0.0942, 0.0991, 0.0806, 0.0950, 0.0997, 0.0896, 0.0753, 0.0829], + 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') +2023-02-07 06:08:58,831 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.323e+02 2.664e+02 3.243e+02 7.163e+02, threshold=5.329e+02, percent-clipped=2.0 +2023-02-07 06:09:20,376 INFO [train.py:901] (2/4) Epoch 23, batch 7200, loss[loss=0.2521, simple_loss=0.3213, pruned_loss=0.09144, over 8625.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2837, pruned_loss=0.05909, over 1607931.81 frames. ], batch size: 49, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:09:29,466 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7513, 1.4901, 1.6486, 1.3912, 0.8567, 1.4341, 1.5164, 1.3472], + device='cuda:2'), covar=tensor([0.0564, 0.1304, 0.1689, 0.1474, 0.0627, 0.1511, 0.0763, 0.0700], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0161, 0.0102, 0.0164, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 06:09:35,037 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9701, 1.3880, 1.6884, 1.3201, 0.9280, 1.4803, 1.7966, 1.4931], + device='cuda:2'), covar=tensor([0.0577, 0.1613, 0.2235, 0.1792, 0.0689, 0.1866, 0.0732, 0.0762], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0102, 0.0164, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 06:09:35,780 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7634, 1.6612, 2.3517, 1.4459, 1.2484, 2.3578, 0.4156, 1.4187], + device='cuda:2'), covar=tensor([0.1661, 0.1215, 0.0298, 0.1324, 0.2652, 0.0302, 0.2275, 0.1448], + device='cuda:2'), in_proj_covar=tensor([0.0190, 0.0195, 0.0129, 0.0219, 0.0268, 0.0136, 0.0168, 0.0191], + 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') +2023-02-07 06:09:44,874 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5357, 1.3851, 1.6045, 1.2651, 0.9883, 1.4070, 1.5743, 1.2087], + device='cuda:2'), covar=tensor([0.0614, 0.1282, 0.1742, 0.1544, 0.0588, 0.1487, 0.0704, 0.0729], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0101, 0.0164, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 06:09:44,913 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4796, 1.4334, 1.8591, 1.2311, 1.0974, 1.7826, 0.2104, 1.1575], + device='cuda:2'), covar=tensor([0.1755, 0.1208, 0.0363, 0.1108, 0.2718, 0.0468, 0.2068, 0.1374], + device='cuda:2'), in_proj_covar=tensor([0.0190, 0.0196, 0.0129, 0.0220, 0.0268, 0.0136, 0.0169, 0.0191], + 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') +2023-02-07 06:09:54,644 INFO [train.py:901] (2/4) Epoch 23, batch 7250, loss[loss=0.2277, simple_loss=0.3021, pruned_loss=0.07669, over 7920.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2845, pruned_loss=0.0601, over 1605216.65 frames. ], batch size: 20, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:09:56,898 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185078.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:10:06,386 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185092.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:10:09,712 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.341e+02 2.701e+02 3.625e+02 6.528e+02, threshold=5.401e+02, percent-clipped=8.0 +2023-02-07 06:10:25,004 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185117.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:10:30,397 INFO [train.py:901] (2/4) Epoch 23, batch 7300, loss[loss=0.191, simple_loss=0.2707, pruned_loss=0.05563, over 7976.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2853, pruned_loss=0.06048, over 1605403.46 frames. ], batch size: 21, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:10:45,740 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1198, 1.9046, 2.4645, 2.0855, 2.3844, 2.2046, 1.9835, 1.2334], + device='cuda:2'), covar=tensor([0.5315, 0.4867, 0.1933, 0.3452, 0.2456, 0.2920, 0.1937, 0.5367], + device='cuda:2'), in_proj_covar=tensor([0.0939, 0.0986, 0.0805, 0.0948, 0.0995, 0.0896, 0.0751, 0.0826], + 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') +2023-02-07 06:10:46,969 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185147.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:11:04,728 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.51 vs. limit=5.0 +2023-02-07 06:11:06,505 INFO [train.py:901] (2/4) Epoch 23, batch 7350, loss[loss=0.2023, simple_loss=0.2718, pruned_loss=0.06644, over 8255.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2858, pruned_loss=0.06057, over 1609561.03 frames. ], batch size: 22, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:11:19,715 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 06:11:21,069 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.380e+02 2.863e+02 3.556e+02 7.708e+02, threshold=5.726e+02, percent-clipped=6.0 +2023-02-07 06:11:38,649 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5337, 1.5096, 1.8799, 1.3429, 1.2771, 1.8240, 0.1964, 1.2833], + device='cuda:2'), covar=tensor([0.1340, 0.1073, 0.0344, 0.0876, 0.2229, 0.0406, 0.1718, 0.1052], + device='cuda:2'), in_proj_covar=tensor([0.0189, 0.0195, 0.0128, 0.0219, 0.0268, 0.0136, 0.0168, 0.0191], + 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') +2023-02-07 06:11:38,653 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185219.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:11:41,244 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 06:11:42,632 INFO [train.py:901] (2/4) Epoch 23, batch 7400, loss[loss=0.1903, simple_loss=0.2867, pruned_loss=0.04695, over 8329.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2869, pruned_loss=0.06056, over 1616312.26 frames. ], batch size: 25, lr: 3.26e-03, grad_scale: 16.0 +2023-02-07 06:11:44,819 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185228.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:11:48,325 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7765, 1.6086, 2.8583, 1.3821, 2.0483, 3.0875, 3.2663, 2.5665], + device='cuda:2'), covar=tensor([0.1124, 0.1554, 0.0386, 0.2129, 0.0998, 0.0290, 0.0546, 0.0599], + device='cuda:2'), in_proj_covar=tensor([0.0297, 0.0322, 0.0287, 0.0315, 0.0314, 0.0269, 0.0424, 0.0305], + 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') +2023-02-07 06:11:56,745 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185244.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:12:04,426 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 06:12:18,684 INFO [train.py:901] (2/4) Epoch 23, batch 7450, loss[loss=0.203, simple_loss=0.2866, pruned_loss=0.05968, over 8490.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2868, pruned_loss=0.06091, over 1614526.05 frames. ], batch size: 39, lr: 3.26e-03, grad_scale: 16.0 +2023-02-07 06:12:18,947 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4040, 1.6277, 2.1279, 1.2814, 1.5338, 1.6714, 1.4783, 1.5446], + device='cuda:2'), covar=tensor([0.2010, 0.2634, 0.1007, 0.4843, 0.2063, 0.3491, 0.2475, 0.2294], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0611, 0.0554, 0.0646, 0.0647, 0.0595, 0.0541, 0.0632], + 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') +2023-02-07 06:12:21,577 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 06:12:33,471 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.310e+02 2.954e+02 3.827e+02 6.869e+02, threshold=5.908e+02, percent-clipped=4.0 +2023-02-07 06:12:37,081 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:12:53,822 INFO [train.py:901] (2/4) Epoch 23, batch 7500, loss[loss=0.1764, simple_loss=0.2458, pruned_loss=0.05345, over 7197.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2868, pruned_loss=0.06104, over 1613700.17 frames. ], batch size: 16, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:13:05,840 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-07 06:13:08,273 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185343.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:13:22,330 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3952, 1.5143, 1.2689, 1.7268, 1.0219, 1.1931, 1.4427, 1.5248], + device='cuda:2'), covar=tensor([0.0670, 0.0620, 0.0831, 0.0570, 0.0959, 0.1104, 0.0598, 0.0580], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0199, 0.0244, 0.0214, 0.0206, 0.0247, 0.0250, 0.0207], + 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') +2023-02-07 06:13:31,746 INFO [train.py:901] (2/4) Epoch 23, batch 7550, loss[loss=0.1692, simple_loss=0.2604, pruned_loss=0.03902, over 8104.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2857, pruned_loss=0.06056, over 1606816.08 frames. ], batch size: 23, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:13:39,223 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 06:13:46,072 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.461e+02 3.059e+02 3.860e+02 7.244e+02, threshold=6.118e+02, percent-clipped=3.0 +2023-02-07 06:14:00,491 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185416.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:04,573 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185422.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:06,593 INFO [train.py:901] (2/4) Epoch 23, batch 7600, loss[loss=0.1978, simple_loss=0.2842, pruned_loss=0.05564, over 8499.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.284, pruned_loss=0.0599, over 1604701.71 frames. ], batch size: 26, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:14:09,580 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.01 vs. limit=5.0 +2023-02-07 06:14:41,949 INFO [train.py:901] (2/4) Epoch 23, batch 7650, loss[loss=0.2329, simple_loss=0.3031, pruned_loss=0.08135, over 8036.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2862, pruned_loss=0.06084, over 1614338.39 frames. ], batch size: 22, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:14:50,192 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185486.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:54,420 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185491.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:57,809 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.478e+02 3.100e+02 3.999e+02 8.387e+02, threshold=6.200e+02, percent-clipped=6.0 +2023-02-07 06:15:17,431 INFO [train.py:901] (2/4) Epoch 23, batch 7700, loss[loss=0.2541, simple_loss=0.3279, pruned_loss=0.09018, over 8288.00 frames. ], tot_loss[loss=0.205, simple_loss=0.287, pruned_loss=0.06146, over 1614106.07 frames. ], batch size: 49, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:15:25,721 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185537.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:15:37,375 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 06:15:53,096 INFO [train.py:901] (2/4) Epoch 23, batch 7750, loss[loss=0.1985, simple_loss=0.2856, pruned_loss=0.05565, over 7782.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.287, pruned_loss=0.0614, over 1616566.45 frames. ], batch size: 19, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:16:08,172 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.336e+02 2.905e+02 3.607e+02 6.527e+02, threshold=5.810e+02, percent-clipped=2.0 +2023-02-07 06:16:10,502 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185599.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:16:15,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185606.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:16:28,336 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185624.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:16:28,832 INFO [train.py:901] (2/4) Epoch 23, batch 7800, loss[loss=0.1598, simple_loss=0.2443, pruned_loss=0.03762, over 7444.00 frames. ], tot_loss[loss=0.205, simple_loss=0.287, pruned_loss=0.06151, over 1613280.84 frames. ], batch size: 17, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:17:01,064 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185672.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:17:02,901 INFO [train.py:901] (2/4) Epoch 23, batch 7850, loss[loss=0.2218, simple_loss=0.3063, pruned_loss=0.0686, over 8628.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2864, pruned_loss=0.06111, over 1611019.39 frames. ], batch size: 34, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:17:17,286 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.434e+02 2.983e+02 3.607e+02 9.941e+02, threshold=5.966e+02, percent-clipped=5.0 +2023-02-07 06:17:18,198 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185697.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:17:37,217 INFO [train.py:901] (2/4) Epoch 23, batch 7900, loss[loss=0.1964, simple_loss=0.2729, pruned_loss=0.05995, over 8192.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2855, pruned_loss=0.06057, over 1612067.09 frames. ], batch size: 23, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:18:00,628 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-07 06:18:11,083 INFO [train.py:901] (2/4) Epoch 23, batch 7950, loss[loss=0.1761, simple_loss=0.2582, pruned_loss=0.04702, over 7808.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2841, pruned_loss=0.05995, over 1610198.81 frames. ], batch size: 20, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:18:12,605 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185777.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:18:12,818 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-07 06:18:23,328 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185793.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:18:25,078 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 2.254e+02 2.775e+02 3.427e+02 8.244e+02, threshold=5.550e+02, percent-clipped=2.0 +2023-02-07 06:18:39,405 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185817.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 06:18:40,117 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185818.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:18:44,661 INFO [train.py:901] (2/4) Epoch 23, batch 8000, loss[loss=0.2245, simple_loss=0.3181, pruned_loss=0.06544, over 8235.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2862, pruned_loss=0.06114, over 1612442.77 frames. ], batch size: 22, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:18:48,037 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185830.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:19:09,878 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185862.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:19:18,259 INFO [train.py:901] (2/4) Epoch 23, batch 8050, loss[loss=0.2507, simple_loss=0.3226, pruned_loss=0.08935, over 7038.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2856, pruned_loss=0.06183, over 1595740.69 frames. ], batch size: 71, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:19:26,749 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185887.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:19:32,783 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.662e+02 3.318e+02 4.159e+02 9.358e+02, threshold=6.635e+02, percent-clipped=7.0 +2023-02-07 06:19:51,747 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 06:19:57,763 INFO [train.py:901] (2/4) Epoch 24, batch 0, loss[loss=0.2107, simple_loss=0.2965, pruned_loss=0.06246, over 8493.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2965, pruned_loss=0.06246, over 8493.00 frames. ], batch size: 26, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:19:57,763 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 06:20:01,675 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6433, 1.4850, 1.7678, 1.4545, 0.9422, 1.5211, 1.6903, 1.3955], + device='cuda:2'), covar=tensor([0.0666, 0.1352, 0.1749, 0.1499, 0.0646, 0.1577, 0.0690, 0.0692], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0101, 0.0164, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 06:20:09,066 INFO [train.py:935] (2/4) Epoch 24, validation: loss=0.1731, simple_loss=0.2733, pruned_loss=0.03644, over 944034.00 frames. +2023-02-07 06:20:09,067 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 06:20:23,915 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 06:20:35,543 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185945.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:20:44,033 INFO [train.py:901] (2/4) Epoch 24, batch 50, loss[loss=0.1671, simple_loss=0.2434, pruned_loss=0.04544, over 7715.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2855, pruned_loss=0.06133, over 361757.31 frames. ], batch size: 18, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:20:57,559 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 06:21:11,398 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.437e+02 2.851e+02 3.663e+02 1.155e+03, threshold=5.702e+02, percent-clipped=3.0 +2023-02-07 06:21:20,557 INFO [train.py:901] (2/4) Epoch 24, batch 100, loss[loss=0.2487, simple_loss=0.3292, pruned_loss=0.0841, over 8327.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2902, pruned_loss=0.06372, over 641672.30 frames. ], batch size: 25, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:21:22,593 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 06:21:41,718 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5731, 1.8119, 5.7772, 2.2746, 5.2472, 4.8570, 5.3425, 5.2483], + device='cuda:2'), covar=tensor([0.0579, 0.4640, 0.0366, 0.3822, 0.0923, 0.0895, 0.0504, 0.0567], + device='cuda:2'), in_proj_covar=tensor([0.0645, 0.0655, 0.0708, 0.0641, 0.0720, 0.0620, 0.0615, 0.0690], + 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') +2023-02-07 06:21:56,096 INFO [train.py:901] (2/4) Epoch 24, batch 150, loss[loss=0.1656, simple_loss=0.2426, pruned_loss=0.04425, over 8088.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2882, pruned_loss=0.06199, over 855920.89 frames. ], batch size: 21, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:22:00,573 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.72 vs. limit=5.0 +2023-02-07 06:22:03,342 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 06:22:07,731 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0536, 1.2177, 1.2235, 0.7578, 1.2045, 1.0907, 0.1289, 1.2163], + device='cuda:2'), covar=tensor([0.0434, 0.0389, 0.0365, 0.0549, 0.0426, 0.0943, 0.0851, 0.0322], + device='cuda:2'), in_proj_covar=tensor([0.0462, 0.0402, 0.0354, 0.0455, 0.0385, 0.0545, 0.0398, 0.0428], + device='cuda:2'), out_proj_covar=tensor([1.2331e-04, 1.0515e-04, 9.2994e-05, 1.1942e-04, 1.0126e-04, 1.5330e-04, + 1.0710e-04, 1.1294e-04], device='cuda:2') +2023-02-07 06:22:21,896 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.384e+02 2.880e+02 3.401e+02 7.597e+02, threshold=5.761e+02, percent-clipped=1.0 +2023-02-07 06:22:30,261 INFO [train.py:901] (2/4) Epoch 24, batch 200, loss[loss=0.1782, simple_loss=0.2676, pruned_loss=0.0444, over 8357.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2866, pruned_loss=0.06106, over 1024272.48 frames. ], batch size: 24, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:22:35,413 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8509, 1.4354, 3.5045, 1.5315, 2.4010, 3.8034, 3.9024, 3.3002], + device='cuda:2'), covar=tensor([0.1268, 0.1914, 0.0280, 0.2050, 0.1060, 0.0220, 0.0430, 0.0526], + device='cuda:2'), in_proj_covar=tensor([0.0297, 0.0321, 0.0286, 0.0317, 0.0314, 0.0269, 0.0426, 0.0304], + 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') +2023-02-07 06:22:38,159 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:22:40,033 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:23:01,369 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-07 06:23:05,571 INFO [train.py:901] (2/4) Epoch 24, batch 250, loss[loss=0.2487, simple_loss=0.3252, pruned_loss=0.08609, over 6672.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2856, pruned_loss=0.061, over 1149499.44 frames. ], batch size: 72, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:23:07,697 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186161.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:23:16,544 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 06:23:18,802 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186176.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:23:25,610 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 06:23:32,345 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.402e+02 3.098e+02 3.972e+02 8.418e+02, threshold=6.197e+02, percent-clipped=5.0 +2023-02-07 06:23:36,037 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186201.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:23:40,506 INFO [train.py:901] (2/4) Epoch 24, batch 300, loss[loss=0.1975, simple_loss=0.2753, pruned_loss=0.05985, over 7811.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2871, pruned_loss=0.0615, over 1254186.77 frames. ], batch size: 19, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:23:53,002 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186226.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:24:00,553 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186236.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:24:00,861 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 06:24:13,913 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9836, 3.8039, 2.2423, 2.7787, 2.6553, 2.1931, 2.7851, 3.0198], + device='cuda:2'), covar=tensor([0.1528, 0.0296, 0.1178, 0.0754, 0.0814, 0.1351, 0.0977, 0.0933], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0235, 0.0337, 0.0311, 0.0303, 0.0342, 0.0350, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 06:24:15,084 INFO [train.py:901] (2/4) Epoch 24, batch 350, loss[loss=0.2768, simple_loss=0.3453, pruned_loss=0.1042, over 7382.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2873, pruned_loss=0.06129, over 1335754.39 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:24:24,625 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2036, 4.1639, 3.7892, 2.2002, 3.7543, 3.8464, 3.7040, 3.7269], + device='cuda:2'), covar=tensor([0.0763, 0.0577, 0.1098, 0.3984, 0.0892, 0.0777, 0.1253, 0.0764], + device='cuda:2'), in_proj_covar=tensor([0.0526, 0.0445, 0.0431, 0.0541, 0.0431, 0.0446, 0.0426, 0.0389], + 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') +2023-02-07 06:24:28,140 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186276.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 06:24:28,716 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186277.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:24:42,184 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.476e+02 2.428e+02 2.971e+02 3.348e+02 5.777e+02, threshold=5.941e+02, percent-clipped=0.0 +2023-02-07 06:24:47,709 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5924, 4.6469, 4.1602, 2.1649, 4.0607, 4.1984, 4.1402, 4.0573], + device='cuda:2'), covar=tensor([0.0701, 0.0501, 0.1062, 0.4241, 0.0881, 0.0774, 0.1203, 0.0688], + device='cuda:2'), in_proj_covar=tensor([0.0527, 0.0445, 0.0431, 0.0541, 0.0431, 0.0446, 0.0426, 0.0389], + 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') +2023-02-07 06:24:50,342 INFO [train.py:901] (2/4) Epoch 24, batch 400, loss[loss=0.2355, simple_loss=0.3229, pruned_loss=0.07408, over 8328.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2868, pruned_loss=0.06082, over 1398145.30 frames. ], batch size: 25, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:25:26,090 INFO [train.py:901] (2/4) Epoch 24, batch 450, loss[loss=0.1988, simple_loss=0.2984, pruned_loss=0.04955, over 8186.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2881, pruned_loss=0.0607, over 1452321.61 frames. ], batch size: 23, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:25:52,928 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.487e+02 2.919e+02 3.580e+02 7.824e+02, threshold=5.839e+02, percent-clipped=3.0 +2023-02-07 06:26:02,021 INFO [train.py:901] (2/4) Epoch 24, batch 500, loss[loss=0.2054, simple_loss=0.2916, pruned_loss=0.05966, over 8300.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2868, pruned_loss=0.05997, over 1490712.31 frames. ], batch size: 23, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:26:23,360 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:26:37,824 INFO [train.py:901] (2/4) Epoch 24, batch 550, loss[loss=0.1868, simple_loss=0.2677, pruned_loss=0.05299, over 8187.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2873, pruned_loss=0.06089, over 1513674.72 frames. ], batch size: 23, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:26:40,776 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:26:55,558 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7725, 1.9312, 2.0812, 1.3773, 2.1781, 1.5459, 0.6655, 1.9788], + device='cuda:2'), covar=tensor([0.0681, 0.0407, 0.0366, 0.0716, 0.0499, 0.1014, 0.0991, 0.0345], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0399, 0.0351, 0.0449, 0.0382, 0.0539, 0.0393, 0.0426], + device='cuda:2'), out_proj_covar=tensor([1.2231e-04, 1.0434e-04, 9.2115e-05, 1.1794e-04, 1.0032e-04, 1.5167e-04, + 1.0568e-04, 1.1243e-04], device='cuda:2') +2023-02-07 06:27:01,113 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:01,713 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186493.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:03,582 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.350e+02 3.005e+02 3.846e+02 7.955e+02, threshold=6.011e+02, percent-clipped=1.0 +2023-02-07 06:27:12,601 INFO [train.py:901] (2/4) Epoch 24, batch 600, loss[loss=0.1921, simple_loss=0.2725, pruned_loss=0.05584, over 8086.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2868, pruned_loss=0.0605, over 1536958.39 frames. ], batch size: 21, lr: 3.17e-03, grad_scale: 16.0 +2023-02-07 06:27:19,641 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:21,528 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186520.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:23,346 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.51 vs. limit=5.0 +2023-02-07 06:27:26,201 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 06:27:29,792 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186532.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:27:31,735 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186535.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:40,300 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7250, 1.4761, 3.1159, 1.4138, 2.2279, 3.3597, 3.4458, 2.8447], + device='cuda:2'), covar=tensor([0.1270, 0.1727, 0.0320, 0.2114, 0.0981, 0.0263, 0.0588, 0.0559], + device='cuda:2'), in_proj_covar=tensor([0.0297, 0.0321, 0.0286, 0.0314, 0.0312, 0.0268, 0.0424, 0.0303], + 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') +2023-02-07 06:27:46,469 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186557.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:46,539 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186557.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:27:47,019 INFO [train.py:901] (2/4) Epoch 24, batch 650, loss[loss=0.2481, simple_loss=0.3329, pruned_loss=0.08165, over 8518.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2876, pruned_loss=0.06119, over 1551600.91 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:27:52,098 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1436, 1.8969, 3.3329, 1.6777, 2.5484, 3.7436, 3.6957, 3.2601], + device='cuda:2'), covar=tensor([0.1115, 0.1649, 0.0390, 0.1985, 0.1197, 0.0212, 0.0618, 0.0517], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0321, 0.0285, 0.0314, 0.0311, 0.0268, 0.0423, 0.0302], + 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') +2023-02-07 06:28:01,233 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186577.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:07,510 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186585.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:13,143 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7673, 1.4464, 2.8190, 1.3443, 2.2068, 2.9815, 3.1868, 2.5805], + device='cuda:2'), covar=tensor([0.1133, 0.1704, 0.0368, 0.2206, 0.0930, 0.0316, 0.0560, 0.0582], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0320, 0.0284, 0.0312, 0.0310, 0.0267, 0.0422, 0.0301], + 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') +2023-02-07 06:28:15,659 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.377e+02 2.753e+02 3.513e+02 8.271e+02, threshold=5.505e+02, percent-clipped=2.0 +2023-02-07 06:28:20,763 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186604.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:23,403 INFO [train.py:901] (2/4) Epoch 24, batch 700, loss[loss=0.1691, simple_loss=0.2466, pruned_loss=0.04586, over 7562.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2859, pruned_loss=0.06013, over 1567711.49 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:28:24,187 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7993, 5.8855, 5.1640, 2.6435, 5.2338, 5.5312, 5.4108, 5.4255], + device='cuda:2'), covar=tensor([0.0483, 0.0385, 0.0874, 0.4005, 0.0669, 0.0791, 0.1017, 0.0568], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0446, 0.0432, 0.0543, 0.0430, 0.0448, 0.0427, 0.0389], + 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') +2023-02-07 06:28:33,216 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186621.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:43,748 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186635.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:59,668 INFO [train.py:901] (2/4) Epoch 24, batch 750, loss[loss=0.1603, simple_loss=0.2427, pruned_loss=0.03896, over 7542.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2864, pruned_loss=0.06007, over 1581433.09 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:29:00,494 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7736, 2.0970, 2.3065, 1.2564, 2.3053, 1.6248, 0.7719, 1.9077], + device='cuda:2'), covar=tensor([0.0802, 0.0416, 0.0326, 0.0751, 0.0514, 0.0936, 0.1069, 0.0417], + device='cuda:2'), in_proj_covar=tensor([0.0457, 0.0397, 0.0350, 0.0447, 0.0379, 0.0536, 0.0391, 0.0425], + device='cuda:2'), out_proj_covar=tensor([1.2187e-04, 1.0376e-04, 9.1745e-05, 1.1728e-04, 9.9550e-05, 1.5077e-04, + 1.0520e-04, 1.1211e-04], device='cuda:2') +2023-02-07 06:29:11,911 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 06:29:16,209 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2040, 4.1475, 3.7700, 1.9052, 3.7004, 3.7723, 3.7022, 3.5982], + device='cuda:2'), covar=tensor([0.0799, 0.0655, 0.1081, 0.4681, 0.0888, 0.1181, 0.1431, 0.0914], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0446, 0.0433, 0.0542, 0.0430, 0.0448, 0.0427, 0.0389], + 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') +2023-02-07 06:29:21,531 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 06:29:27,092 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.532e+02 3.077e+02 4.008e+02 9.294e+02, threshold=6.153e+02, percent-clipped=8.0 +2023-02-07 06:29:31,907 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 06:29:35,762 INFO [train.py:901] (2/4) Epoch 24, batch 800, loss[loss=0.239, simple_loss=0.308, pruned_loss=0.08498, over 8346.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.286, pruned_loss=0.06039, over 1589254.40 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:29:48,934 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186727.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:29:56,174 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186736.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:30:11,914 INFO [train.py:901] (2/4) Epoch 24, batch 850, loss[loss=0.1948, simple_loss=0.2706, pruned_loss=0.05955, over 6746.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2862, pruned_loss=0.05988, over 1595268.13 frames. ], batch size: 15, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:30:29,503 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:30:39,065 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.330e+02 2.764e+02 3.350e+02 7.186e+02, threshold=5.528e+02, percent-clipped=2.0 +2023-02-07 06:30:47,647 INFO [train.py:901] (2/4) Epoch 24, batch 900, loss[loss=0.1828, simple_loss=0.2774, pruned_loss=0.04413, over 8716.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2867, pruned_loss=0.0601, over 1603940.15 frames. ], batch size: 34, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:31:05,965 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:08,582 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:15,808 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3324, 2.1092, 2.6342, 2.2456, 2.6404, 2.3354, 2.1877, 1.6383], + device='cuda:2'), covar=tensor([0.4938, 0.4731, 0.1879, 0.3346, 0.2240, 0.2845, 0.1753, 0.4881], + device='cuda:2'), in_proj_covar=tensor([0.0948, 0.0992, 0.0812, 0.0958, 0.0998, 0.0901, 0.0756, 0.0830], + 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') +2023-02-07 06:31:24,267 INFO [train.py:901] (2/4) Epoch 24, batch 950, loss[loss=0.1924, simple_loss=0.2753, pruned_loss=0.05475, over 7969.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2864, pruned_loss=0.06001, over 1607327.68 frames. ], batch size: 21, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:31:24,450 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186858.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:24,512 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186858.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:30,134 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1127, 1.5024, 1.6899, 1.3708, 0.9700, 1.4929, 1.8077, 1.8116], + device='cuda:2'), covar=tensor([0.0548, 0.1261, 0.1741, 0.1511, 0.0592, 0.1503, 0.0670, 0.0608], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0100, 0.0162, 0.0111, 0.0143], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 06:31:39,988 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186879.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:43,466 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 06:31:48,340 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186891.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:50,408 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6229, 2.6277, 1.8994, 2.3297, 2.1727, 1.5234, 2.1465, 2.2650], + device='cuda:2'), covar=tensor([0.1551, 0.0393, 0.1202, 0.0705, 0.0773, 0.1659, 0.1071, 0.0968], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0234, 0.0336, 0.0310, 0.0301, 0.0341, 0.0346, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 06:31:52,263 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.324e+02 2.850e+02 3.567e+02 7.043e+02, threshold=5.700e+02, percent-clipped=2.0 +2023-02-07 06:31:53,149 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186898.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:55,093 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186901.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:59,853 INFO [train.py:901] (2/4) Epoch 24, batch 1000, loss[loss=0.2711, simple_loss=0.3556, pruned_loss=0.09328, over 8559.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2867, pruned_loss=0.06, over 1614436.93 frames. ], batch size: 31, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:32:05,644 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186916.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:15,421 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:15,491 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:19,256 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2843, 2.1044, 1.5912, 1.9255, 1.7001, 1.4032, 1.6920, 1.6797], + device='cuda:2'), covar=tensor([0.1320, 0.0476, 0.1283, 0.0583, 0.0753, 0.1554, 0.0984, 0.0966], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0235, 0.0336, 0.0310, 0.0301, 0.0341, 0.0347, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 06:32:20,497 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 06:32:29,434 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186948.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:32,167 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186952.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:33,359 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 06:32:36,112 INFO [train.py:901] (2/4) Epoch 24, batch 1050, loss[loss=0.1929, simple_loss=0.2673, pruned_loss=0.05921, over 7541.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2861, pruned_loss=0.05985, over 1610473.81 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:33:00,058 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2528, 1.2503, 3.3874, 1.0303, 3.0103, 2.8833, 3.0957, 3.0197], + device='cuda:2'), covar=tensor([0.0816, 0.4171, 0.0820, 0.4123, 0.1321, 0.1124, 0.0791, 0.0840], + device='cuda:2'), in_proj_covar=tensor([0.0642, 0.0649, 0.0703, 0.0634, 0.0718, 0.0618, 0.0610, 0.0686], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 06:33:01,571 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186992.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:02,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186994.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:04,738 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.475e+02 2.949e+02 3.829e+02 9.793e+02, threshold=5.897e+02, percent-clipped=8.0 +2023-02-07 06:33:12,427 INFO [train.py:901] (2/4) Epoch 24, batch 1100, loss[loss=0.1817, simple_loss=0.2595, pruned_loss=0.05191, over 7989.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.286, pruned_loss=0.05989, over 1610592.43 frames. ], batch size: 21, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:33:12,688 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6891, 2.1878, 4.1004, 1.4766, 2.8919, 2.1373, 1.8368, 2.9917], + device='cuda:2'), covar=tensor([0.2041, 0.3010, 0.0855, 0.4984, 0.2021, 0.3504, 0.2490, 0.2406], + device='cuda:2'), in_proj_covar=tensor([0.0527, 0.0615, 0.0556, 0.0650, 0.0653, 0.0600, 0.0544, 0.0634], + 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') +2023-02-07 06:33:18,320 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187016.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:19,042 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187017.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:37,517 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187043.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:38,301 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:45,706 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 06:33:47,140 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:48,355 INFO [train.py:901] (2/4) Epoch 24, batch 1150, loss[loss=0.2006, simple_loss=0.2834, pruned_loss=0.0589, over 8562.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2867, pruned_loss=0.06026, over 1609794.66 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:33:49,930 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0196, 1.5167, 3.5720, 1.5587, 2.5057, 3.9088, 4.0445, 3.3823], + device='cuda:2'), covar=tensor([0.1142, 0.1892, 0.0283, 0.1941, 0.1001, 0.0209, 0.0583, 0.0509], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0324, 0.0288, 0.0317, 0.0315, 0.0270, 0.0427, 0.0304], + 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') +2023-02-07 06:33:52,048 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187063.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:57,506 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187071.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:34:02,532 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0504, 1.6035, 1.3552, 1.5442, 1.3156, 1.2347, 1.3249, 1.2973], + device='cuda:2'), covar=tensor([0.1020, 0.0477, 0.1243, 0.0539, 0.0729, 0.1444, 0.0878, 0.0770], + device='cuda:2'), in_proj_covar=tensor([0.0355, 0.0234, 0.0336, 0.0309, 0.0301, 0.0341, 0.0346, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 06:34:16,153 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.205e+02 2.742e+02 3.279e+02 6.267e+02, threshold=5.485e+02, percent-clipped=2.0 +2023-02-07 06:34:24,623 INFO [train.py:901] (2/4) Epoch 24, batch 1200, loss[loss=0.2068, simple_loss=0.2985, pruned_loss=0.05757, over 8291.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.286, pruned_loss=0.06005, over 1607479.17 frames. ], batch size: 23, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:34:57,405 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187154.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:00,007 INFO [train.py:901] (2/4) Epoch 24, batch 1250, loss[loss=0.194, simple_loss=0.2797, pruned_loss=0.05413, over 8502.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2862, pruned_loss=0.06054, over 1610900.42 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:35:15,160 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187179.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:19,762 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187186.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:27,951 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.417e+02 2.916e+02 3.659e+02 9.833e+02, threshold=5.832e+02, percent-clipped=6.0 +2023-02-07 06:35:30,893 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187202.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:35,743 INFO [train.py:901] (2/4) Epoch 24, batch 1300, loss[loss=0.2152, simple_loss=0.2922, pruned_loss=0.06909, over 8830.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2864, pruned_loss=0.06013, over 1616386.05 frames. ], batch size: 40, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:35:35,952 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187208.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:51,427 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.89 vs. limit=5.0 +2023-02-07 06:35:53,890 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187233.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:05,804 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187250.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:10,944 INFO [train.py:901] (2/4) Epoch 24, batch 1350, loss[loss=0.1768, simple_loss=0.2506, pruned_loss=0.05152, over 7421.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2863, pruned_loss=0.05987, over 1619410.69 frames. ], batch size: 17, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:36:20,761 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187272.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:21,330 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187273.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:22,835 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187275.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:39,324 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187297.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:39,749 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.391e+02 3.088e+02 3.702e+02 1.176e+03, threshold=6.175e+02, percent-clipped=8.0 +2023-02-07 06:36:41,381 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187300.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:46,710 INFO [train.py:901] (2/4) Epoch 24, batch 1400, loss[loss=0.1647, simple_loss=0.2509, pruned_loss=0.03924, over 7922.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2869, pruned_loss=0.06049, over 1618584.45 frames. ], batch size: 20, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:36:52,967 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187317.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:54,421 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187319.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:59,275 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187325.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:03,363 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9407, 2.3617, 4.2871, 1.6721, 3.1490, 2.4234, 2.0737, 3.0454], + device='cuda:2'), covar=tensor([0.1799, 0.2550, 0.0701, 0.4456, 0.1633, 0.3000, 0.2224, 0.2303], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0617, 0.0558, 0.0652, 0.0653, 0.0600, 0.0545, 0.0636], + 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') +2023-02-07 06:37:05,352 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2608, 3.6471, 2.3822, 2.9519, 2.7933, 2.1092, 2.9791, 3.0780], + device='cuda:2'), covar=tensor([0.1669, 0.0391, 0.1197, 0.0760, 0.0767, 0.1459, 0.1024, 0.1151], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0233, 0.0335, 0.0308, 0.0300, 0.0338, 0.0346, 0.0316], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 06:37:12,877 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187344.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:16,689 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2023-02-07 06:37:21,749 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 06:37:22,421 INFO [train.py:901] (2/4) Epoch 24, batch 1450, loss[loss=0.1996, simple_loss=0.2941, pruned_loss=0.05258, over 8752.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.286, pruned_loss=0.05987, over 1616044.91 frames. ], batch size: 30, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:37:42,337 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187387.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:43,074 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:44,409 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:49,533 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.482e+02 2.870e+02 4.012e+02 8.494e+02, threshold=5.740e+02, percent-clipped=8.0 +2023-02-07 06:37:51,822 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:56,992 INFO [train.py:901] (2/4) Epoch 24, batch 1500, loss[loss=0.2154, simple_loss=0.3009, pruned_loss=0.06493, over 8496.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2848, pruned_loss=0.05925, over 1616008.31 frames. ], batch size: 27, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:38:22,162 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187442.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:38:33,247 INFO [train.py:901] (2/4) Epoch 24, batch 1550, loss[loss=0.2127, simple_loss=0.2949, pruned_loss=0.06527, over 8285.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2857, pruned_loss=0.0599, over 1618900.38 frames. ], batch size: 23, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:38:39,575 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187467.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:38:59,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.355e+02 2.764e+02 3.622e+02 7.454e+02, threshold=5.529e+02, percent-clipped=4.0 +2023-02-07 06:39:02,835 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187502.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:06,537 INFO [train.py:901] (2/4) Epoch 24, batch 1600, loss[loss=0.1984, simple_loss=0.2696, pruned_loss=0.06364, over 7430.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2856, pruned_loss=0.05963, over 1618559.84 frames. ], batch size: 17, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:39:11,516 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187515.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:22,052 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187530.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:27,626 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187537.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:30,380 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187541.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:41,433 INFO [train.py:901] (2/4) Epoch 24, batch 1650, loss[loss=0.206, simple_loss=0.2933, pruned_loss=0.05937, over 8091.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2854, pruned_loss=0.05941, over 1619668.79 frames. ], batch size: 21, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:39:52,523 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187573.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:40:09,686 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.451e+02 2.921e+02 3.516e+02 7.853e+02, threshold=5.842e+02, percent-clipped=7.0 +2023-02-07 06:40:09,873 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187598.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:40:10,451 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3385, 1.2338, 2.4122, 1.3753, 2.2145, 2.5421, 2.7104, 2.2116], + device='cuda:2'), covar=tensor([0.1176, 0.1391, 0.0398, 0.1986, 0.0734, 0.0381, 0.0668, 0.0638], + device='cuda:2'), in_proj_covar=tensor([0.0295, 0.0319, 0.0283, 0.0312, 0.0311, 0.0267, 0.0422, 0.0301], + 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') +2023-02-07 06:40:16,432 INFO [train.py:901] (2/4) Epoch 24, batch 1700, loss[loss=0.2287, simple_loss=0.3087, pruned_loss=0.07434, over 7168.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2863, pruned_loss=0.06022, over 1620087.23 frames. ], batch size: 16, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:40:40,839 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187644.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:40:50,891 INFO [train.py:901] (2/4) Epoch 24, batch 1750, loss[loss=0.196, simple_loss=0.2747, pruned_loss=0.05861, over 8081.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2865, pruned_loss=0.06057, over 1618995.91 frames. ], batch size: 21, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:40:58,580 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187669.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:41:18,565 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.502e+02 3.000e+02 3.757e+02 9.885e+02, threshold=5.999e+02, percent-clipped=2.0 +2023-02-07 06:41:25,258 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 06:41:26,170 INFO [train.py:901] (2/4) Epoch 24, batch 1800, loss[loss=0.195, simple_loss=0.2887, pruned_loss=0.05067, over 8483.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2854, pruned_loss=0.06038, over 1613668.61 frames. ], batch size: 25, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:41:43,298 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187734.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:41:59,215 INFO [train.py:901] (2/4) Epoch 24, batch 1850, loss[loss=0.2402, simple_loss=0.315, pruned_loss=0.08272, over 8617.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2859, pruned_loss=0.06058, over 1615687.36 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:41:59,464 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187758.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:09,322 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187771.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:18,458 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:27,406 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187796.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:28,601 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.405e+02 2.492e+02 2.912e+02 4.002e+02 8.326e+02, threshold=5.824e+02, percent-clipped=6.0 +2023-02-07 06:42:36,380 INFO [train.py:901] (2/4) Epoch 24, batch 1900, loss[loss=0.2149, simple_loss=0.2965, pruned_loss=0.06665, over 8456.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2851, pruned_loss=0.06018, over 1612064.01 frames. ], batch size: 49, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:42:42,657 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.2617, 5.2090, 4.7096, 2.7859, 4.7169, 4.9382, 4.8307, 4.7744], + device='cuda:2'), covar=tensor([0.0574, 0.0415, 0.0851, 0.3799, 0.0833, 0.1042, 0.1029, 0.0712], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0446, 0.0435, 0.0543, 0.0435, 0.0448, 0.0428, 0.0391], + 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') +2023-02-07 06:42:49,918 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 +2023-02-07 06:43:02,032 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 06:43:05,614 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 06:43:05,816 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187849.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:11,795 INFO [train.py:901] (2/4) Epoch 24, batch 1950, loss[loss=0.1777, simple_loss=0.2599, pruned_loss=0.04773, over 7419.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.285, pruned_loss=0.05996, over 1614363.39 frames. ], batch size: 17, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:43:18,404 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 06:43:22,599 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187874.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:27,989 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187881.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:30,733 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187885.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:39,205 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 2.375e+02 2.745e+02 3.412e+02 6.105e+02, threshold=5.491e+02, percent-clipped=1.0 +2023-02-07 06:43:39,241 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 06:43:42,659 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4408, 1.3483, 1.7863, 1.2127, 1.1050, 1.7437, 0.2391, 1.2147], + device='cuda:2'), covar=tensor([0.1615, 0.1219, 0.0395, 0.0958, 0.2541, 0.0439, 0.1886, 0.1134], + device='cuda:2'), in_proj_covar=tensor([0.0194, 0.0200, 0.0132, 0.0222, 0.0273, 0.0137, 0.0171, 0.0194], + 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') +2023-02-07 06:43:46,284 INFO [train.py:901] (2/4) Epoch 24, batch 2000, loss[loss=0.2026, simple_loss=0.2947, pruned_loss=0.05532, over 8454.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2866, pruned_loss=0.0606, over 1618952.45 frames. ], batch size: 27, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:44:21,541 INFO [train.py:901] (2/4) Epoch 24, batch 2050, loss[loss=0.1997, simple_loss=0.2893, pruned_loss=0.05506, over 8492.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2863, pruned_loss=0.06013, over 1619152.30 frames. ], batch size: 29, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:44:23,778 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0128, 1.2778, 1.2458, 0.6585, 1.2436, 1.0701, 0.1250, 1.1965], + device='cuda:2'), covar=tensor([0.0443, 0.0396, 0.0365, 0.0597, 0.0503, 0.1040, 0.0901, 0.0362], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0400, 0.0354, 0.0450, 0.0384, 0.0539, 0.0394, 0.0425], + device='cuda:2'), out_proj_covar=tensor([1.2219e-04, 1.0450e-04, 9.3072e-05, 1.1817e-04, 1.0087e-04, 1.5174e-04, + 1.0582e-04, 1.1198e-04], device='cuda:2') +2023-02-07 06:44:42,292 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187989.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:44:46,912 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187996.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:44:48,824 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.387e+02 2.958e+02 3.531e+02 6.524e+02, threshold=5.915e+02, percent-clipped=3.0 +2023-02-07 06:44:51,435 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188000.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:44:56,704 INFO [train.py:901] (2/4) Epoch 24, batch 2100, loss[loss=0.1847, simple_loss=0.281, pruned_loss=0.04425, over 8189.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2866, pruned_loss=0.06053, over 1617658.94 frames. ], batch size: 23, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:44:58,672 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 06:45:31,694 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6719, 1.5887, 2.2995, 1.5693, 1.2906, 2.2001, 0.3360, 1.4467], + device='cuda:2'), covar=tensor([0.1640, 0.1257, 0.0332, 0.1182, 0.2642, 0.0460, 0.2053, 0.1203], + device='cuda:2'), in_proj_covar=tensor([0.0194, 0.0200, 0.0132, 0.0222, 0.0273, 0.0137, 0.0172, 0.0195], + 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') +2023-02-07 06:45:32,170 INFO [train.py:901] (2/4) Epoch 24, batch 2150, loss[loss=0.2087, simple_loss=0.2943, pruned_loss=0.06152, over 8343.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2866, pruned_loss=0.06056, over 1617161.42 frames. ], batch size: 24, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:45:47,941 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6401, 2.3952, 3.2437, 2.6440, 3.0403, 2.7092, 2.5269, 1.8049], + device='cuda:2'), covar=tensor([0.5317, 0.5153, 0.1842, 0.3855, 0.2586, 0.2894, 0.1755, 0.5304], + device='cuda:2'), in_proj_covar=tensor([0.0951, 0.1000, 0.0818, 0.0965, 0.1002, 0.0908, 0.0761, 0.0835], + 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') +2023-02-07 06:45:58,763 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.464e+02 3.048e+02 3.692e+02 7.821e+02, threshold=6.095e+02, percent-clipped=5.0 +2023-02-07 06:46:03,900 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188105.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:46:05,701 INFO [train.py:901] (2/4) Epoch 24, batch 2200, loss[loss=0.2217, simple_loss=0.3007, pruned_loss=0.07131, over 8301.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2871, pruned_loss=0.06099, over 1613888.17 frames. ], batch size: 23, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:46:21,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188130.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:46:41,360 INFO [train.py:901] (2/4) Epoch 24, batch 2250, loss[loss=0.1791, simple_loss=0.2615, pruned_loss=0.04833, over 7928.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2858, pruned_loss=0.0603, over 1610313.73 frames. ], batch size: 20, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:47:09,411 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.349e+02 3.047e+02 3.898e+02 9.680e+02, threshold=6.095e+02, percent-clipped=4.0 +2023-02-07 06:47:16,302 INFO [train.py:901] (2/4) Epoch 24, batch 2300, loss[loss=0.2168, simple_loss=0.3017, pruned_loss=0.06597, over 8514.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2858, pruned_loss=0.05993, over 1612748.20 frames. ], batch size: 49, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:47:42,368 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188245.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:47:47,130 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188252.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:47:50,575 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188256.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:47:51,707 INFO [train.py:901] (2/4) Epoch 24, batch 2350, loss[loss=0.2292, simple_loss=0.3088, pruned_loss=0.07478, over 8470.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2847, pruned_loss=0.05967, over 1609316.62 frames. ], batch size: 25, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:48:00,162 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188270.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:05,509 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188277.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:08,186 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188281.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:20,112 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.514e+02 3.085e+02 3.939e+02 8.316e+02, threshold=6.171e+02, percent-clipped=4.0 +2023-02-07 06:48:22,399 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:27,099 INFO [train.py:901] (2/4) Epoch 24, batch 2400, loss[loss=0.1831, simple_loss=0.2546, pruned_loss=0.05574, over 7203.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2848, pruned_loss=0.06012, over 1606684.69 frames. ], batch size: 16, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:48:30,048 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4778, 1.2800, 2.3473, 1.2903, 2.1989, 2.5124, 2.7028, 2.1620], + device='cuda:2'), covar=tensor([0.1100, 0.1438, 0.0470, 0.2113, 0.0763, 0.0392, 0.0695, 0.0658], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0323, 0.0285, 0.0315, 0.0314, 0.0270, 0.0425, 0.0304], + 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') +2023-02-07 06:49:00,630 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.70 vs. limit=5.0 +2023-02-07 06:49:02,251 INFO [train.py:901] (2/4) Epoch 24, batch 2450, loss[loss=0.208, simple_loss=0.2999, pruned_loss=0.05806, over 8696.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2855, pruned_loss=0.06037, over 1607216.53 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:49:30,924 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.263e+02 2.982e+02 3.612e+02 7.179e+02, threshold=5.965e+02, percent-clipped=1.0 +2023-02-07 06:49:38,096 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-07 06:49:38,406 INFO [train.py:901] (2/4) Epoch 24, batch 2500, loss[loss=0.184, simple_loss=0.2632, pruned_loss=0.05239, over 8030.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2844, pruned_loss=0.05949, over 1612123.48 frames. ], batch size: 22, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:50:11,881 INFO [train.py:901] (2/4) Epoch 24, batch 2550, loss[loss=0.2257, simple_loss=0.3146, pruned_loss=0.0684, over 8260.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2843, pruned_loss=0.05954, over 1611872.57 frames. ], batch size: 49, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:50:40,459 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.347e+02 2.538e+02 2.905e+02 3.766e+02 9.788e+02, threshold=5.809e+02, percent-clipped=4.0 +2023-02-07 06:50:41,319 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:50:47,914 INFO [train.py:901] (2/4) Epoch 24, batch 2600, loss[loss=0.1722, simple_loss=0.2448, pruned_loss=0.04983, over 7931.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2851, pruned_loss=0.06002, over 1611153.27 frames. ], batch size: 20, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:51:21,923 INFO [train.py:901] (2/4) Epoch 24, batch 2650, loss[loss=0.1815, simple_loss=0.2628, pruned_loss=0.0501, over 7794.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2836, pruned_loss=0.05972, over 1610182.34 frames. ], batch size: 19, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:51:31,890 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-07 06:51:41,773 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 +2023-02-07 06:51:48,597 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.338e+02 2.924e+02 3.924e+02 7.774e+02, threshold=5.847e+02, percent-clipped=4.0 +2023-02-07 06:51:55,409 INFO [train.py:901] (2/4) Epoch 24, batch 2700, loss[loss=0.1896, simple_loss=0.2794, pruned_loss=0.04988, over 8566.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2845, pruned_loss=0.06013, over 1611252.01 frames. ], batch size: 39, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:52:02,963 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2350, 3.1371, 2.9544, 1.6801, 2.8913, 2.9299, 2.9068, 2.7823], + device='cuda:2'), covar=tensor([0.1132, 0.0795, 0.1217, 0.4477, 0.1138, 0.1277, 0.1545, 0.1023], + device='cuda:2'), in_proj_covar=tensor([0.0521, 0.0439, 0.0426, 0.0535, 0.0426, 0.0441, 0.0421, 0.0386], + 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') +2023-02-07 06:52:21,731 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=188645.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:52:31,276 INFO [train.py:901] (2/4) Epoch 24, batch 2750, loss[loss=0.2037, simple_loss=0.2886, pruned_loss=0.05942, over 8356.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2848, pruned_loss=0.06037, over 1611366.94 frames. ], batch size: 24, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:52:57,777 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 2.375e+02 3.087e+02 4.139e+02 1.460e+03, threshold=6.174e+02, percent-clipped=4.0 +2023-02-07 06:53:05,388 INFO [train.py:901] (2/4) Epoch 24, batch 2800, loss[loss=0.1879, simple_loss=0.2734, pruned_loss=0.05126, over 8136.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2849, pruned_loss=0.06067, over 1604621.37 frames. ], batch size: 22, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:53:23,062 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0291, 1.8710, 2.4322, 2.0317, 2.4257, 2.1113, 1.9572, 1.2742], + device='cuda:2'), covar=tensor([0.6001, 0.5154, 0.2124, 0.4038, 0.2571, 0.3275, 0.2044, 0.5617], + device='cuda:2'), in_proj_covar=tensor([0.0947, 0.0992, 0.0814, 0.0956, 0.0995, 0.0905, 0.0756, 0.0832], + 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') +2023-02-07 06:53:40,615 INFO [train.py:901] (2/4) Epoch 24, batch 2850, loss[loss=0.2556, simple_loss=0.3323, pruned_loss=0.08941, over 8706.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2847, pruned_loss=0.06015, over 1607311.12 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:53:42,192 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188760.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:54:07,932 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.346e+02 3.064e+02 3.754e+02 6.997e+02, threshold=6.129e+02, percent-clipped=3.0 +2023-02-07 06:54:14,861 INFO [train.py:901] (2/4) Epoch 24, batch 2900, loss[loss=0.1843, simple_loss=0.2719, pruned_loss=0.04835, over 8137.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2861, pruned_loss=0.06053, over 1612846.06 frames. ], batch size: 22, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:54:39,477 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=188843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:54:48,791 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8915, 2.1305, 1.6898, 2.5392, 1.1278, 1.5135, 1.8291, 2.0067], + device='cuda:2'), covar=tensor([0.0727, 0.0680, 0.0940, 0.0379, 0.1105, 0.1304, 0.0781, 0.0855], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0197, 0.0242, 0.0214, 0.0205, 0.0246, 0.0249, 0.0208], + 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') +2023-02-07 06:54:51,391 INFO [train.py:901] (2/4) Epoch 24, batch 2950, loss[loss=0.2118, simple_loss=0.2916, pruned_loss=0.06597, over 8238.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2856, pruned_loss=0.06041, over 1610853.37 frames. ], batch size: 22, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:54:51,400 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 06:55:19,106 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.217e+02 2.685e+02 3.700e+02 9.567e+02, threshold=5.370e+02, percent-clipped=4.0 +2023-02-07 06:55:25,889 INFO [train.py:901] (2/4) Epoch 24, batch 3000, loss[loss=0.2231, simple_loss=0.2944, pruned_loss=0.07588, over 7136.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2863, pruned_loss=0.06109, over 1606091.71 frames. ], batch size: 73, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:55:25,889 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 06:55:39,550 INFO [train.py:935] (2/4) Epoch 24, validation: loss=0.1724, simple_loss=0.2726, pruned_loss=0.03604, over 944034.00 frames. +2023-02-07 06:55:39,551 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 06:56:05,991 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188947.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:07,362 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188949.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:13,965 INFO [train.py:901] (2/4) Epoch 24, batch 3050, loss[loss=0.2089, simple_loss=0.2882, pruned_loss=0.06482, over 7982.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2871, pruned_loss=0.0614, over 1609225.05 frames. ], batch size: 21, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:56:14,161 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188958.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:41,554 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.422e+02 3.010e+02 3.817e+02 9.746e+02, threshold=6.020e+02, percent-clipped=4.0 +2023-02-07 06:56:49,112 INFO [train.py:901] (2/4) Epoch 24, batch 3100, loss[loss=0.2225, simple_loss=0.3207, pruned_loss=0.06219, over 8327.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2875, pruned_loss=0.06132, over 1606860.41 frames. ], batch size: 26, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:56:54,798 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189016.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:58,620 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6688, 4.7168, 4.1907, 1.9835, 4.1140, 4.3501, 4.2853, 4.1302], + device='cuda:2'), covar=tensor([0.0730, 0.0536, 0.1038, 0.5207, 0.0887, 0.0924, 0.1309, 0.0710], + device='cuda:2'), in_proj_covar=tensor([0.0527, 0.0442, 0.0430, 0.0542, 0.0429, 0.0444, 0.0424, 0.0389], + 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') +2023-02-07 06:57:12,120 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189041.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:16,985 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189048.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:19,144 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5590, 1.5058, 2.0811, 1.4463, 1.1765, 2.0204, 0.4018, 1.2628], + device='cuda:2'), covar=tensor([0.1466, 0.1156, 0.0327, 0.0879, 0.2321, 0.0355, 0.1844, 0.1070], + device='cuda:2'), in_proj_covar=tensor([0.0194, 0.0202, 0.0132, 0.0223, 0.0274, 0.0139, 0.0173, 0.0197], + 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') +2023-02-07 06:57:23,526 INFO [train.py:901] (2/4) Epoch 24, batch 3150, loss[loss=0.2308, simple_loss=0.3152, pruned_loss=0.07314, over 8560.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2872, pruned_loss=0.06139, over 1604378.77 frames. ], batch size: 31, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:57:26,637 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.98 vs. limit=5.0 +2023-02-07 06:57:40,666 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189082.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:50,855 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189097.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:51,360 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.394e+02 2.917e+02 3.565e+02 6.979e+02, threshold=5.834e+02, percent-clipped=3.0 +2023-02-07 06:57:59,738 INFO [train.py:901] (2/4) Epoch 24, batch 3200, loss[loss=0.1898, simple_loss=0.2699, pruned_loss=0.05487, over 7540.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2863, pruned_loss=0.06071, over 1604175.55 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:58:33,956 INFO [train.py:901] (2/4) Epoch 24, batch 3250, loss[loss=0.1955, simple_loss=0.284, pruned_loss=0.05351, over 8031.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2867, pruned_loss=0.06072, over 1609062.49 frames. ], batch size: 22, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 06:59:01,534 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.438e+02 3.003e+02 3.759e+02 6.490e+02, threshold=6.005e+02, percent-clipped=4.0 +2023-02-07 06:59:08,532 INFO [train.py:901] (2/4) Epoch 24, batch 3300, loss[loss=0.1713, simple_loss=0.2576, pruned_loss=0.04246, over 7923.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2863, pruned_loss=0.05998, over 1614420.16 frames. ], batch size: 20, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 06:59:12,858 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:59:30,912 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:59:39,008 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6220, 2.6806, 1.9724, 2.2945, 2.2679, 1.7286, 2.1352, 2.2141], + device='cuda:2'), covar=tensor([0.1649, 0.0398, 0.1181, 0.0709, 0.0691, 0.1557, 0.1077, 0.1126], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0235, 0.0334, 0.0310, 0.0299, 0.0341, 0.0345, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 06:59:44,170 INFO [train.py:901] (2/4) Epoch 24, batch 3350, loss[loss=0.2092, simple_loss=0.2789, pruned_loss=0.06975, over 7150.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.286, pruned_loss=0.06005, over 1609362.02 frames. ], batch size: 16, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 06:59:49,413 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189266.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:59:54,641 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6903, 1.5891, 2.5866, 1.9717, 2.2860, 1.7208, 1.5072, 1.2075], + device='cuda:2'), covar=tensor([0.8682, 0.7337, 0.2311, 0.4571, 0.3675, 0.5177, 0.3583, 0.6321], + device='cuda:2'), in_proj_covar=tensor([0.0945, 0.0991, 0.0811, 0.0955, 0.0994, 0.0902, 0.0754, 0.0828], + 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') +2023-02-07 07:00:05,776 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189291.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:00:07,043 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189293.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:00:10,342 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.367e+02 2.967e+02 3.575e+02 9.298e+02, threshold=5.934e+02, percent-clipped=5.0 +2023-02-07 07:00:17,750 INFO [train.py:901] (2/4) Epoch 24, batch 3400, loss[loss=0.2104, simple_loss=0.2945, pruned_loss=0.0631, over 8047.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2876, pruned_loss=0.06102, over 1615399.66 frames. ], batch size: 22, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:00:33,370 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6782, 1.3908, 2.9250, 1.1838, 2.2801, 3.1692, 3.4729, 2.3853], + device='cuda:2'), covar=tensor([0.1602, 0.2178, 0.0560, 0.2920, 0.1248, 0.0439, 0.0632, 0.1048], + device='cuda:2'), in_proj_covar=tensor([0.0300, 0.0324, 0.0286, 0.0317, 0.0316, 0.0271, 0.0429, 0.0304], + 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') +2023-02-07 07:00:34,013 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8494, 1.3626, 1.6083, 1.2278, 0.8644, 1.3909, 1.6033, 1.3751], + device='cuda:2'), covar=tensor([0.0533, 0.1340, 0.1698, 0.1570, 0.0648, 0.1554, 0.0754, 0.0677], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0160, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:00:52,446 INFO [train.py:901] (2/4) Epoch 24, batch 3450, loss[loss=0.2173, simple_loss=0.3095, pruned_loss=0.06254, over 8245.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2864, pruned_loss=0.06075, over 1615453.08 frames. ], batch size: 24, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:00:57,500 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189365.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:16,615 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189392.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:20,600 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.399e+02 2.884e+02 3.624e+02 7.571e+02, threshold=5.767e+02, percent-clipped=3.0 +2023-02-07 07:01:26,259 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:27,399 INFO [train.py:901] (2/4) Epoch 24, batch 3500, loss[loss=0.2523, simple_loss=0.3255, pruned_loss=0.08955, over 7178.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2863, pruned_loss=0.06082, over 1615127.60 frames. ], batch size: 71, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:01:27,623 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189408.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:27,663 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6989, 1.9824, 2.1136, 1.4704, 2.2275, 1.4623, 0.7167, 1.9093], + device='cuda:2'), covar=tensor([0.0741, 0.0411, 0.0320, 0.0666, 0.0461, 0.1069, 0.0957, 0.0383], + device='cuda:2'), in_proj_covar=tensor([0.0461, 0.0401, 0.0356, 0.0454, 0.0387, 0.0542, 0.0398, 0.0431], + device='cuda:2'), out_proj_covar=tensor([1.2292e-04, 1.0469e-04, 9.3477e-05, 1.1928e-04, 1.0174e-04, 1.5223e-04, + 1.0689e-04, 1.1361e-04], device='cuda:2') +2023-02-07 07:01:40,495 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 07:01:40,595 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189426.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:50,804 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189441.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:56,335 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 07:02:03,505 INFO [train.py:901] (2/4) Epoch 24, batch 3550, loss[loss=0.2083, simple_loss=0.2906, pruned_loss=0.06299, over 8329.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06117, over 1612088.41 frames. ], batch size: 25, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:02:26,873 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5603, 2.2826, 3.1231, 2.5192, 3.0714, 2.5146, 2.3265, 1.9292], + device='cuda:2'), covar=tensor([0.5475, 0.5040, 0.1972, 0.3835, 0.2426, 0.2961, 0.1840, 0.5456], + device='cuda:2'), in_proj_covar=tensor([0.0943, 0.0992, 0.0811, 0.0958, 0.0997, 0.0903, 0.0755, 0.0828], + 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') +2023-02-07 07:02:31,300 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.509e+02 2.981e+02 3.708e+02 7.370e+02, threshold=5.962e+02, percent-clipped=4.0 +2023-02-07 07:02:37,633 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189507.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:02:38,143 INFO [train.py:901] (2/4) Epoch 24, batch 3600, loss[loss=0.1984, simple_loss=0.2925, pruned_loss=0.05218, over 8355.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2861, pruned_loss=0.06091, over 1614672.71 frames. ], batch size: 26, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:02:42,682 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 07:02:45,848 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189519.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:03:01,847 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189541.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:03:12,157 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189556.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:03:13,407 INFO [train.py:901] (2/4) Epoch 24, batch 3650, loss[loss=0.229, simple_loss=0.2946, pruned_loss=0.08175, over 6755.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2862, pruned_loss=0.06096, over 1611573.09 frames. ], batch size: 71, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:03:41,105 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 07:03:41,749 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.496e+02 2.930e+02 3.600e+02 6.319e+02, threshold=5.860e+02, percent-clipped=2.0 +2023-02-07 07:03:48,377 INFO [train.py:901] (2/4) Epoch 24, batch 3700, loss[loss=0.181, simple_loss=0.27, pruned_loss=0.04595, over 7958.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2859, pruned_loss=0.06084, over 1608676.63 frames. ], batch size: 21, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:03:49,802 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189610.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:03:55,988 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7661, 1.9488, 2.0369, 1.4694, 2.2031, 1.4297, 0.7325, 2.0010], + device='cuda:2'), covar=tensor([0.0663, 0.0391, 0.0320, 0.0575, 0.0425, 0.0943, 0.0908, 0.0278], + device='cuda:2'), in_proj_covar=tensor([0.0459, 0.0399, 0.0354, 0.0451, 0.0385, 0.0540, 0.0394, 0.0427], + device='cuda:2'), out_proj_covar=tensor([1.2235e-04, 1.0415e-04, 9.2972e-05, 1.1844e-04, 1.0110e-04, 1.5183e-04, + 1.0590e-04, 1.1266e-04], device='cuda:2') +2023-02-07 07:04:00,074 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3911, 1.6015, 1.6510, 1.2360, 1.6191, 1.2931, 0.3412, 1.6052], + device='cuda:2'), covar=tensor([0.0531, 0.0398, 0.0339, 0.0502, 0.0482, 0.1054, 0.0922, 0.0282], + device='cuda:2'), in_proj_covar=tensor([0.0458, 0.0399, 0.0354, 0.0451, 0.0384, 0.0540, 0.0394, 0.0427], + device='cuda:2'), out_proj_covar=tensor([1.2229e-04, 1.0409e-04, 9.2928e-05, 1.1839e-04, 1.0099e-04, 1.5174e-04, + 1.0583e-04, 1.1259e-04], device='cuda:2') +2023-02-07 07:04:23,127 INFO [train.py:901] (2/4) Epoch 24, batch 3750, loss[loss=0.195, simple_loss=0.2908, pruned_loss=0.04962, over 8481.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2858, pruned_loss=0.06058, over 1609444.39 frames. ], batch size: 28, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:04:23,298 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189658.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:26,018 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1135, 1.2953, 1.2411, 0.8679, 1.2232, 1.0617, 0.1722, 1.2320], + device='cuda:2'), covar=tensor([0.0418, 0.0365, 0.0373, 0.0563, 0.0521, 0.1022, 0.0819, 0.0344], + device='cuda:2'), in_proj_covar=tensor([0.0456, 0.0397, 0.0352, 0.0449, 0.0382, 0.0537, 0.0393, 0.0425], + device='cuda:2'), out_proj_covar=tensor([1.2159e-04, 1.0350e-04, 9.2355e-05, 1.1792e-04, 1.0030e-04, 1.5080e-04, + 1.0553e-04, 1.1201e-04], device='cuda:2') +2023-02-07 07:04:26,023 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:27,227 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189664.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:42,766 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189687.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:43,975 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189689.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:51,127 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.609e+02 3.129e+02 4.249e+02 7.016e+02, threshold=6.258e+02, percent-clipped=8.0 +2023-02-07 07:04:57,800 INFO [train.py:901] (2/4) Epoch 24, batch 3800, loss[loss=0.1774, simple_loss=0.2687, pruned_loss=0.04299, over 8448.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2859, pruned_loss=0.06043, over 1610455.56 frames. ], batch size: 24, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:04:58,622 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189709.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:07,489 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189722.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:09,543 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189725.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:24,358 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189746.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:32,067 INFO [train.py:901] (2/4) Epoch 24, batch 3850, loss[loss=0.1776, simple_loss=0.2614, pruned_loss=0.04694, over 7701.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2864, pruned_loss=0.06093, over 1608762.84 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:05:35,653 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189763.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:47,661 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 07:05:53,007 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189788.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:53,958 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-07 07:05:59,103 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189797.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:59,539 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.407e+02 2.910e+02 3.432e+02 8.251e+02, threshold=5.819e+02, percent-clipped=1.0 +2023-02-07 07:06:06,335 INFO [train.py:901] (2/4) Epoch 24, batch 3900, loss[loss=0.1919, simple_loss=0.2842, pruned_loss=0.04983, over 8498.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2867, pruned_loss=0.06081, over 1608348.80 frames. ], batch size: 28, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:06:10,780 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189812.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:17,542 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189822.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:18,899 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189824.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:27,525 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:42,145 INFO [train.py:901] (2/4) Epoch 24, batch 3950, loss[loss=0.1922, simple_loss=0.2733, pruned_loss=0.05557, over 8020.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.287, pruned_loss=0.06102, over 1611867.92 frames. ], batch size: 22, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:06:45,564 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189863.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:07:09,594 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.387e+02 3.217e+02 3.997e+02 8.874e+02, threshold=6.434e+02, percent-clipped=5.0 +2023-02-07 07:07:16,321 INFO [train.py:901] (2/4) Epoch 24, batch 4000, loss[loss=0.1467, simple_loss=0.228, pruned_loss=0.03271, over 7436.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2864, pruned_loss=0.06046, over 1608713.86 frames. ], batch size: 17, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:07:51,143 INFO [train.py:901] (2/4) Epoch 24, batch 4050, loss[loss=0.2582, simple_loss=0.3339, pruned_loss=0.09122, over 8501.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2857, pruned_loss=0.0602, over 1609230.49 frames. ], batch size: 28, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:08:05,480 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189978.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:07,500 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189981.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:18,681 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.334e+02 2.770e+02 3.399e+02 1.124e+03, threshold=5.539e+02, percent-clipped=1.0 +2023-02-07 07:08:22,556 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190002.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:26,230 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190006.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:27,388 INFO [train.py:901] (2/4) Epoch 24, batch 4100, loss[loss=0.1898, simple_loss=0.273, pruned_loss=0.05326, over 7688.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2848, pruned_loss=0.0593, over 1611399.86 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:09:02,436 INFO [train.py:901] (2/4) Epoch 24, batch 4150, loss[loss=0.2254, simple_loss=0.3067, pruned_loss=0.07206, over 8321.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05927, over 1613239.65 frames. ], batch size: 26, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:09:08,094 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190066.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:17,893 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190080.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:25,160 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190090.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:30,579 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.686e+02 2.377e+02 2.724e+02 3.400e+02 7.023e+02, threshold=5.448e+02, percent-clipped=3.0 +2023-02-07 07:09:35,508 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190105.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:37,392 INFO [train.py:901] (2/4) Epoch 24, batch 4200, loss[loss=0.2153, simple_loss=0.3045, pruned_loss=0.06304, over 8100.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2847, pruned_loss=0.05896, over 1615704.84 frames. ], batch size: 23, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:09:43,678 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190117.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:45,031 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 07:09:48,185 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 07:09:50,482 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.09 vs. limit=5.0 +2023-02-07 07:10:10,654 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 07:10:11,425 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190157.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:10:11,873 INFO [train.py:901] (2/4) Epoch 24, batch 4250, loss[loss=0.237, simple_loss=0.3136, pruned_loss=0.0802, over 8446.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2849, pruned_loss=0.05938, over 1611589.46 frames. ], batch size: 24, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:10:19,343 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190169.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:10:28,695 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190181.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:10:40,134 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.309e+02 2.865e+02 3.517e+02 8.092e+02, threshold=5.730e+02, percent-clipped=6.0 +2023-02-07 07:10:45,780 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190205.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:10:47,629 INFO [train.py:901] (2/4) Epoch 24, batch 4300, loss[loss=0.207, simple_loss=0.2817, pruned_loss=0.06615, over 7161.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2853, pruned_loss=0.05952, over 1612856.88 frames. ], batch size: 72, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:10:56,464 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8555, 1.4685, 4.2075, 1.7095, 3.3969, 3.3308, 3.8011, 3.7241], + device='cuda:2'), covar=tensor([0.1404, 0.6531, 0.1124, 0.5419, 0.2360, 0.1636, 0.1069, 0.1177], + device='cuda:2'), in_proj_covar=tensor([0.0651, 0.0657, 0.0718, 0.0645, 0.0723, 0.0622, 0.0622, 0.0697], + 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') +2023-02-07 07:11:05,299 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190234.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:11:21,901 INFO [train.py:901] (2/4) Epoch 24, batch 4350, loss[loss=0.2185, simple_loss=0.2972, pruned_loss=0.06986, over 8585.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2854, pruned_loss=0.05958, over 1612014.63 frames. ], batch size: 31, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:11:22,814 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190259.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:11:40,172 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 07:11:50,405 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.430e+02 2.823e+02 3.493e+02 1.012e+03, threshold=5.646e+02, percent-clipped=3.0 +2023-02-07 07:11:57,334 INFO [train.py:901] (2/4) Epoch 24, batch 4400, loss[loss=0.2174, simple_loss=0.2931, pruned_loss=0.07085, over 8036.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2853, pruned_loss=0.05927, over 1612850.28 frames. ], batch size: 22, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:12:15,560 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190334.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:12:23,140 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 07:12:32,981 INFO [train.py:901] (2/4) Epoch 24, batch 4450, loss[loss=0.2296, simple_loss=0.3058, pruned_loss=0.0767, over 8358.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2848, pruned_loss=0.059, over 1615021.58 frames. ], batch size: 24, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:12:43,423 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:00,241 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.201e+02 2.691e+02 3.403e+02 6.534e+02, threshold=5.381e+02, percent-clipped=2.0 +2023-02-07 07:13:00,480 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190398.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:07,778 INFO [train.py:901] (2/4) Epoch 24, batch 4500, loss[loss=0.1625, simple_loss=0.25, pruned_loss=0.03752, over 7922.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.285, pruned_loss=0.05902, over 1616486.04 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:13:14,930 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190417.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:17,414 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 07:13:29,015 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190437.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:42,696 INFO [train.py:901] (2/4) Epoch 24, batch 4550, loss[loss=0.2135, simple_loss=0.296, pruned_loss=0.06544, over 8470.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.285, pruned_loss=0.05919, over 1615023.57 frames. ], batch size: 27, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:13:44,874 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190461.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:45,487 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:47,442 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.1688, 2.4789, 3.9178, 2.0430, 2.2191, 3.9416, 1.0962, 2.3975], + device='cuda:2'), covar=tensor([0.1134, 0.1292, 0.0228, 0.1571, 0.2125, 0.0189, 0.1856, 0.1221], + device='cuda:2'), in_proj_covar=tensor([0.0194, 0.0201, 0.0130, 0.0222, 0.0273, 0.0138, 0.0171, 0.0196], + 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') +2023-02-07 07:14:02,626 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190486.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:14:10,719 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.430e+02 2.973e+02 3.981e+02 9.647e+02, threshold=5.946e+02, percent-clipped=9.0 +2023-02-07 07:14:12,831 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190501.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:14:17,566 INFO [train.py:901] (2/4) Epoch 24, batch 4600, loss[loss=0.1837, simple_loss=0.2587, pruned_loss=0.05439, over 7445.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2843, pruned_loss=0.05911, over 1613202.83 frames. ], batch size: 17, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:14:21,246 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190513.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:14:42,210 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 +2023-02-07 07:14:54,258 INFO [train.py:901] (2/4) Epoch 24, batch 4650, loss[loss=0.1826, simple_loss=0.2703, pruned_loss=0.0475, over 8250.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2843, pruned_loss=0.05888, over 1611283.47 frames. ], batch size: 24, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:15:22,378 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.616e+02 2.469e+02 2.987e+02 3.787e+02 1.231e+03, threshold=5.974e+02, percent-clipped=5.0 +2023-02-07 07:15:29,195 INFO [train.py:901] (2/4) Epoch 24, batch 4700, loss[loss=0.2225, simple_loss=0.2979, pruned_loss=0.07355, over 7939.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2835, pruned_loss=0.0588, over 1610581.01 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:15:29,957 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190609.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:15:34,417 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190616.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:15:41,743 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190627.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:15:42,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190628.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:16:04,007 INFO [train.py:901] (2/4) Epoch 24, batch 4750, loss[loss=0.2298, simple_loss=0.3084, pruned_loss=0.07559, over 8076.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.283, pruned_loss=0.05844, over 1610760.77 frames. ], batch size: 21, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:16:18,717 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190678.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:16:20,750 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 07:16:22,886 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 07:16:29,418 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190693.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:16:32,691 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.279e+02 2.789e+02 3.393e+02 7.815e+02, threshold=5.578e+02, percent-clipped=3.0 +2023-02-07 07:16:40,387 INFO [train.py:901] (2/4) Epoch 24, batch 4800, loss[loss=0.1865, simple_loss=0.2798, pruned_loss=0.04665, over 8508.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2838, pruned_loss=0.05839, over 1613279.61 frames. ], batch size: 26, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:17:13,014 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 07:17:15,035 INFO [train.py:901] (2/4) Epoch 24, batch 4850, loss[loss=0.1723, simple_loss=0.2487, pruned_loss=0.048, over 7180.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2846, pruned_loss=0.05916, over 1612057.19 frames. ], batch size: 16, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:17:17,934 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:17:25,450 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0423, 1.7884, 2.3086, 1.9665, 2.2533, 2.1211, 1.9407, 1.1555], + device='cuda:2'), covar=tensor([0.5900, 0.5022, 0.2069, 0.3780, 0.2704, 0.3237, 0.2099, 0.5372], + device='cuda:2'), in_proj_covar=tensor([0.0947, 0.0996, 0.0813, 0.0966, 0.1002, 0.0907, 0.0756, 0.0831], + 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') +2023-02-07 07:17:40,140 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190793.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:17:43,500 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.278e+02 2.775e+02 3.178e+02 7.824e+02, threshold=5.550e+02, percent-clipped=3.0 +2023-02-07 07:17:50,723 INFO [train.py:901] (2/4) Epoch 24, batch 4900, loss[loss=0.186, simple_loss=0.2829, pruned_loss=0.04451, over 8595.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2842, pruned_loss=0.05911, over 1610956.85 frames. ], batch size: 31, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:18:06,938 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5897, 1.4341, 2.8955, 1.4094, 2.1879, 3.0534, 3.2197, 2.6206], + device='cuda:2'), covar=tensor([0.1316, 0.1770, 0.0340, 0.2142, 0.0859, 0.0313, 0.0563, 0.0577], + device='cuda:2'), in_proj_covar=tensor([0.0297, 0.0324, 0.0285, 0.0316, 0.0315, 0.0272, 0.0429, 0.0302], + 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') +2023-02-07 07:18:16,604 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.38 vs. limit=5.0 +2023-02-07 07:18:25,852 INFO [train.py:901] (2/4) Epoch 24, batch 4950, loss[loss=0.1858, simple_loss=0.2663, pruned_loss=0.05265, over 7924.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2854, pruned_loss=0.05996, over 1609320.39 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:18:34,367 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 07:18:36,138 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190872.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:18:39,587 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:18:45,197 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190884.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:18:53,852 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190897.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:18:54,299 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.393e+02 2.890e+02 3.701e+02 7.772e+02, threshold=5.780e+02, percent-clipped=4.0 +2023-02-07 07:19:01,823 INFO [train.py:901] (2/4) Epoch 24, batch 5000, loss[loss=0.2236, simple_loss=0.3084, pruned_loss=0.06941, over 8186.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2869, pruned_loss=0.06093, over 1614575.57 frames. ], batch size: 23, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:19:02,713 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:19:23,201 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.3046, 3.2037, 3.0326, 1.6072, 2.9402, 2.9086, 2.9268, 2.7932], + device='cuda:2'), covar=tensor([0.0999, 0.0779, 0.1108, 0.4372, 0.1090, 0.1335, 0.1513, 0.0993], + device='cuda:2'), in_proj_covar=tensor([0.0532, 0.0446, 0.0432, 0.0543, 0.0427, 0.0448, 0.0425, 0.0391], + 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') +2023-02-07 07:19:33,489 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190953.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:19:36,599 INFO [train.py:901] (2/4) Epoch 24, batch 5050, loss[loss=0.1776, simple_loss=0.254, pruned_loss=0.05058, over 7655.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2861, pruned_loss=0.06066, over 1613962.30 frames. ], batch size: 19, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:19:43,537 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-07 07:19:45,289 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190971.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:19:51,185 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 07:20:04,894 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.504e+02 2.521e+02 3.221e+02 4.359e+02 8.705e+02, threshold=6.442e+02, percent-clipped=11.0 +2023-02-07 07:20:11,606 INFO [train.py:901] (2/4) Epoch 24, batch 5100, loss[loss=0.1895, simple_loss=0.2613, pruned_loss=0.05886, over 7701.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2857, pruned_loss=0.06056, over 1611868.93 frames. ], batch size: 18, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:20:18,613 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0976, 1.7861, 6.2556, 2.3390, 5.6759, 5.3192, 5.8052, 5.7350], + device='cuda:2'), covar=tensor([0.0463, 0.4301, 0.0347, 0.3499, 0.0832, 0.0746, 0.0397, 0.0446], + device='cuda:2'), in_proj_covar=tensor([0.0647, 0.0653, 0.0711, 0.0641, 0.0722, 0.0619, 0.0618, 0.0690], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 07:20:31,917 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191037.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:20:36,679 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8480, 1.6182, 4.0048, 1.4757, 3.5658, 3.3325, 3.6522, 3.5276], + device='cuda:2'), covar=tensor([0.0665, 0.4069, 0.0592, 0.4201, 0.1089, 0.0949, 0.0598, 0.0722], + device='cuda:2'), in_proj_covar=tensor([0.0645, 0.0651, 0.0710, 0.0639, 0.0719, 0.0617, 0.0616, 0.0688], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 07:20:40,076 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191049.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:20:46,716 INFO [train.py:901] (2/4) Epoch 24, batch 5150, loss[loss=0.1868, simple_loss=0.2696, pruned_loss=0.05202, over 7912.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.286, pruned_loss=0.06089, over 1609395.66 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:20:53,686 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191068.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:20:57,803 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191074.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:05,746 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191086.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:05,781 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191086.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:13,676 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.286e+02 2.691e+02 3.617e+02 7.196e+02, threshold=5.383e+02, percent-clipped=2.0 +2023-02-07 07:21:20,836 INFO [train.py:901] (2/4) Epoch 24, batch 5200, loss[loss=0.1921, simple_loss=0.2731, pruned_loss=0.05555, over 8472.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2849, pruned_loss=0.06002, over 1608887.31 frames. ], batch size: 25, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:21:38,122 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:41,676 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-07 07:21:44,218 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7142, 2.3404, 4.0174, 1.6494, 2.9349, 2.2851, 1.8292, 2.7221], + device='cuda:2'), covar=tensor([0.2026, 0.2760, 0.0803, 0.4669, 0.1855, 0.3162, 0.2550, 0.2609], + device='cuda:2'), in_proj_covar=tensor([0.0531, 0.0617, 0.0556, 0.0655, 0.0651, 0.0598, 0.0549, 0.0637], + 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') +2023-02-07 07:21:49,618 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8115, 1.9430, 2.0883, 1.3914, 2.2478, 1.5574, 0.6891, 1.9012], + device='cuda:2'), covar=tensor([0.0609, 0.0398, 0.0346, 0.0618, 0.0469, 0.0954, 0.0992, 0.0360], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0406, 0.0360, 0.0457, 0.0388, 0.0544, 0.0401, 0.0434], + device='cuda:2'), out_proj_covar=tensor([1.2363e-04, 1.0604e-04, 9.4292e-05, 1.1997e-04, 1.0206e-04, 1.5281e-04, + 1.0776e-04, 1.1453e-04], device='cuda:2') +2023-02-07 07:21:50,033 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 07:21:50,822 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:52,267 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191152.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:55,601 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:56,101 INFO [train.py:901] (2/4) Epoch 24, batch 5250, loss[loss=0.176, simple_loss=0.2518, pruned_loss=0.05013, over 7451.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2848, pruned_loss=0.06032, over 1609243.94 frames. ], batch size: 17, lr: 3.14e-03, grad_scale: 8.0 +2023-02-07 07:22:11,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3883, 1.4554, 1.3957, 1.7671, 0.7099, 1.2656, 1.3030, 1.4992], + device='cuda:2'), covar=tensor([0.0861, 0.0800, 0.0982, 0.0508, 0.1099, 0.1341, 0.0744, 0.0702], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0199, 0.0244, 0.0215, 0.0205, 0.0247, 0.0252, 0.0207], + 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') +2023-02-07 07:22:14,956 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7557, 1.6956, 2.4592, 1.5970, 1.3315, 2.4381, 0.4175, 1.5378], + device='cuda:2'), covar=tensor([0.1660, 0.1292, 0.0327, 0.1295, 0.2531, 0.0397, 0.2074, 0.1301], + device='cuda:2'), in_proj_covar=tensor([0.0196, 0.0202, 0.0132, 0.0223, 0.0275, 0.0140, 0.0172, 0.0197], + 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') +2023-02-07 07:22:25,865 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.418e+02 2.847e+02 3.981e+02 6.971e+02, threshold=5.694e+02, percent-clipped=11.0 +2023-02-07 07:22:28,317 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191203.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:22:31,598 INFO [train.py:901] (2/4) Epoch 24, batch 5300, loss[loss=0.1692, simple_loss=0.26, pruned_loss=0.03917, over 8288.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2845, pruned_loss=0.06016, over 1608956.34 frames. ], batch size: 23, lr: 3.14e-03, grad_scale: 8.0 +2023-02-07 07:23:07,067 INFO [train.py:901] (2/4) Epoch 24, batch 5350, loss[loss=0.1894, simple_loss=0.2713, pruned_loss=0.05379, over 7799.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2851, pruned_loss=0.06029, over 1611421.49 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 8.0 +2023-02-07 07:23:36,917 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.294e+02 2.754e+02 3.342e+02 1.056e+03, threshold=5.508e+02, percent-clipped=2.0 +2023-02-07 07:23:42,385 INFO [train.py:901] (2/4) Epoch 24, batch 5400, loss[loss=0.1866, simple_loss=0.2735, pruned_loss=0.04983, over 8088.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2853, pruned_loss=0.0601, over 1611545.06 frames. ], batch size: 21, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:23:53,142 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191324.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:23:56,006 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 07:24:05,744 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191342.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:24:11,125 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191349.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:24:17,037 INFO [train.py:901] (2/4) Epoch 24, batch 5450, loss[loss=0.2009, simple_loss=0.2784, pruned_loss=0.06173, over 7806.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2848, pruned_loss=0.0597, over 1611923.77 frames. ], batch size: 20, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:24:21,964 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4277, 2.1149, 2.2540, 2.1284, 1.4805, 2.1550, 2.4030, 2.3032], + device='cuda:2'), covar=tensor([0.0531, 0.0914, 0.1205, 0.1040, 0.0572, 0.1067, 0.0591, 0.0442], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:24:23,296 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191367.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:24:34,967 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191383.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:24:39,460 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 07:24:46,090 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.294e+02 2.951e+02 3.676e+02 7.135e+02, threshold=5.902e+02, percent-clipped=5.0 +2023-02-07 07:24:52,420 INFO [train.py:901] (2/4) Epoch 24, batch 5500, loss[loss=0.2273, simple_loss=0.3098, pruned_loss=0.0724, over 7818.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2861, pruned_loss=0.06049, over 1612088.22 frames. ], batch size: 20, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:24:52,647 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191408.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:07,818 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191430.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:10,089 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191433.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:12,913 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3010, 2.1180, 1.6625, 1.9589, 1.8272, 1.4633, 1.7390, 1.6367], + device='cuda:2'), covar=tensor([0.1256, 0.0423, 0.1218, 0.0516, 0.0710, 0.1613, 0.0918, 0.0904], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0234, 0.0336, 0.0309, 0.0298, 0.0342, 0.0346, 0.0317], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 07:25:28,138 INFO [train.py:901] (2/4) Epoch 24, batch 5550, loss[loss=0.1721, simple_loss=0.2561, pruned_loss=0.04404, over 7976.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2853, pruned_loss=0.05971, over 1613601.31 frames. ], batch size: 21, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:25:39,666 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6545, 2.3961, 3.1442, 2.5367, 3.1033, 2.6623, 2.4196, 1.9167], + device='cuda:2'), covar=tensor([0.5359, 0.5501, 0.2202, 0.4022, 0.2676, 0.2999, 0.1852, 0.5700], + device='cuda:2'), in_proj_covar=tensor([0.0943, 0.0992, 0.0812, 0.0962, 0.0999, 0.0904, 0.0754, 0.0830], + 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') +2023-02-07 07:25:52,797 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1090, 1.2971, 1.6832, 1.3078, 0.7330, 1.3908, 1.1665, 1.0507], + device='cuda:2'), covar=tensor([0.0633, 0.1198, 0.1604, 0.1423, 0.0558, 0.1384, 0.0696, 0.0720], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:25:53,334 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191494.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:57,931 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.420e+02 3.039e+02 3.989e+02 7.925e+02, threshold=6.078e+02, percent-clipped=5.0 +2023-02-07 07:26:03,389 INFO [train.py:901] (2/4) Epoch 24, batch 5600, loss[loss=0.2072, simple_loss=0.2781, pruned_loss=0.06809, over 7235.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2853, pruned_loss=0.06013, over 1613467.42 frames. ], batch size: 16, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:26:29,660 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191545.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:26:30,969 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191547.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:26:38,448 INFO [train.py:901] (2/4) Epoch 24, batch 5650, loss[loss=0.1842, simple_loss=0.2655, pruned_loss=0.05151, over 7712.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2854, pruned_loss=0.06024, over 1615381.33 frames. ], batch size: 18, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:26:45,397 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 07:27:08,570 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.600e+02 2.481e+02 2.901e+02 3.753e+02 9.237e+02, threshold=5.802e+02, percent-clipped=5.0 +2023-02-07 07:27:13,001 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7696, 1.4919, 3.0885, 1.4490, 2.2241, 3.2756, 3.4541, 2.8187], + device='cuda:2'), covar=tensor([0.1215, 0.1813, 0.0360, 0.2164, 0.1084, 0.0322, 0.0572, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0296, 0.0324, 0.0286, 0.0315, 0.0315, 0.0273, 0.0430, 0.0303], + 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') +2023-02-07 07:27:14,232 INFO [train.py:901] (2/4) Epoch 24, batch 5700, loss[loss=0.1682, simple_loss=0.2504, pruned_loss=0.04298, over 7924.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2842, pruned_loss=0.05969, over 1610505.93 frames. ], batch size: 20, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:27:15,043 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191609.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:27:49,745 INFO [train.py:901] (2/4) Epoch 24, batch 5750, loss[loss=0.2176, simple_loss=0.2931, pruned_loss=0.07109, over 8474.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2835, pruned_loss=0.05915, over 1612232.12 frames. ], batch size: 28, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:27:52,729 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:27:53,259 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 07:28:19,025 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.394e+02 2.726e+02 3.524e+02 6.240e+02, threshold=5.452e+02, percent-clipped=4.0 +2023-02-07 07:28:25,087 INFO [train.py:901] (2/4) Epoch 24, batch 5800, loss[loss=0.2483, simple_loss=0.3174, pruned_loss=0.08953, over 8158.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.285, pruned_loss=0.0598, over 1615809.31 frames. ], batch size: 22, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:28:38,093 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191727.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:28:39,806 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 07:28:40,459 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2023-02-07 07:28:59,639 INFO [train.py:901] (2/4) Epoch 24, batch 5850, loss[loss=0.1914, simple_loss=0.2854, pruned_loss=0.04865, over 8361.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2842, pruned_loss=0.05927, over 1612106.60 frames. ], batch size: 26, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:29:29,158 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.223e+02 2.821e+02 3.422e+02 9.012e+02, threshold=5.641e+02, percent-clipped=8.0 +2023-02-07 07:29:30,134 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191801.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:29:34,717 INFO [train.py:901] (2/4) Epoch 24, batch 5900, loss[loss=0.215, simple_loss=0.3086, pruned_loss=0.06068, over 8333.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2832, pruned_loss=0.05878, over 1606509.74 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:29:48,202 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191826.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:29:59,325 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:30:10,950 INFO [train.py:901] (2/4) Epoch 24, batch 5950, loss[loss=0.2061, simple_loss=0.2942, pruned_loss=0.05905, over 8617.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2841, pruned_loss=0.0592, over 1610982.92 frames. ], batch size: 34, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:30:16,034 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:30:33,604 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191890.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:30:40,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.338e+02 2.991e+02 3.628e+02 7.270e+02, threshold=5.982e+02, percent-clipped=3.0 +2023-02-07 07:30:45,681 INFO [train.py:901] (2/4) Epoch 24, batch 6000, loss[loss=0.2201, simple_loss=0.3172, pruned_loss=0.06153, over 8465.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2852, pruned_loss=0.05979, over 1615370.74 frames. ], batch size: 29, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:30:45,681 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 07:30:59,033 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7141, 1.7122, 3.8428, 1.6687, 3.4615, 3.1975, 3.5350, 3.3572], + device='cuda:2'), covar=tensor([0.0618, 0.4198, 0.0536, 0.4245, 0.1005, 0.1034, 0.0609, 0.0687], + device='cuda:2'), in_proj_covar=tensor([0.0648, 0.0653, 0.0710, 0.0643, 0.0720, 0.0617, 0.0618, 0.0690], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 07:31:01,028 INFO [train.py:935] (2/4) Epoch 24, validation: loss=0.1718, simple_loss=0.2718, pruned_loss=0.0359, over 944034.00 frames. +2023-02-07 07:31:01,030 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 07:31:08,194 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191918.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:31:12,832 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1763, 1.4057, 1.6709, 1.3380, 0.7284, 1.4272, 1.1964, 1.0995], + device='cuda:2'), covar=tensor([0.0601, 0.1202, 0.1598, 0.1383, 0.0553, 0.1440, 0.0700, 0.0712], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0101, 0.0163, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:31:24,848 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191943.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:31:26,283 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8734, 2.3201, 4.1486, 1.5830, 3.1707, 2.3483, 1.9279, 2.9737], + device='cuda:2'), covar=tensor([0.2059, 0.2984, 0.1017, 0.5109, 0.1895, 0.3498, 0.2626, 0.2685], + device='cuda:2'), in_proj_covar=tensor([0.0531, 0.0615, 0.0555, 0.0652, 0.0652, 0.0600, 0.0547, 0.0637], + 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') +2023-02-07 07:31:35,315 INFO [train.py:901] (2/4) Epoch 24, batch 6050, loss[loss=0.1952, simple_loss=0.289, pruned_loss=0.0507, over 8327.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2852, pruned_loss=0.05997, over 1615249.16 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:31:41,183 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5047, 1.9400, 2.8837, 1.4137, 2.0910, 1.9206, 1.6314, 2.1110], + device='cuda:2'), covar=tensor([0.2000, 0.2599, 0.0936, 0.4690, 0.2100, 0.3297, 0.2409, 0.2525], + device='cuda:2'), in_proj_covar=tensor([0.0531, 0.0615, 0.0555, 0.0652, 0.0652, 0.0600, 0.0548, 0.0636], + 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') +2023-02-07 07:31:57,329 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0402, 1.5685, 1.7955, 1.3691, 0.9564, 1.5767, 1.7456, 1.6230], + device='cuda:2'), covar=tensor([0.0525, 0.1212, 0.1601, 0.1457, 0.0612, 0.1441, 0.0694, 0.0662], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0159, 0.0100, 0.0162, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:31:58,021 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7707, 2.0254, 1.7319, 2.5520, 1.1057, 1.5114, 1.8247, 2.0065], + device='cuda:2'), covar=tensor([0.0825, 0.0759, 0.0855, 0.0384, 0.1137, 0.1334, 0.0836, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0199, 0.0245, 0.0216, 0.0205, 0.0247, 0.0253, 0.0208], + 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') +2023-02-07 07:32:04,605 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.415e+02 2.742e+02 3.441e+02 8.508e+02, threshold=5.485e+02, percent-clipped=2.0 +2023-02-07 07:32:10,731 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9575, 2.4675, 4.1307, 1.6655, 3.1303, 2.3223, 2.0631, 2.6156], + device='cuda:2'), covar=tensor([0.1673, 0.2349, 0.0789, 0.4373, 0.1672, 0.3096, 0.2133, 0.2646], + device='cuda:2'), in_proj_covar=tensor([0.0528, 0.0613, 0.0552, 0.0649, 0.0649, 0.0598, 0.0545, 0.0634], + 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') +2023-02-07 07:32:11,925 INFO [train.py:901] (2/4) Epoch 24, batch 6100, loss[loss=0.2134, simple_loss=0.2965, pruned_loss=0.06518, over 8452.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2845, pruned_loss=0.0592, over 1616629.85 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:32:32,939 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1023, 1.5927, 1.4369, 1.5849, 1.4030, 1.3133, 1.2861, 1.2990], + device='cuda:2'), covar=tensor([0.1178, 0.0480, 0.1367, 0.0569, 0.0746, 0.1589, 0.1031, 0.0842], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0234, 0.0338, 0.0311, 0.0302, 0.0343, 0.0348, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 07:32:34,056 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 07:32:37,030 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2023-02-07 07:32:46,752 INFO [train.py:901] (2/4) Epoch 24, batch 6150, loss[loss=0.1883, simple_loss=0.2756, pruned_loss=0.0505, over 8112.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2839, pruned_loss=0.0592, over 1615295.39 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:33:15,574 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192098.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:33:16,713 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.367e+02 2.762e+02 3.348e+02 6.106e+02, threshold=5.524e+02, percent-clipped=2.0 +2023-02-07 07:33:22,017 INFO [train.py:901] (2/4) Epoch 24, batch 6200, loss[loss=0.1758, simple_loss=0.26, pruned_loss=0.04582, over 7211.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2837, pruned_loss=0.05924, over 1610897.06 frames. ], batch size: 16, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:33:30,652 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192120.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:33:32,773 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192123.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:33:56,608 INFO [train.py:901] (2/4) Epoch 24, batch 6250, loss[loss=0.1918, simple_loss=0.2678, pruned_loss=0.05791, over 7262.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2824, pruned_loss=0.05805, over 1611239.75 frames. ], batch size: 16, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:34:07,838 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192173.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:34:26,659 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.364e+02 2.949e+02 3.646e+02 8.976e+02, threshold=5.898e+02, percent-clipped=7.0 +2023-02-07 07:34:33,021 INFO [train.py:901] (2/4) Epoch 24, batch 6300, loss[loss=0.1985, simple_loss=0.2865, pruned_loss=0.05529, over 8252.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2832, pruned_loss=0.05848, over 1609406.05 frames. ], batch size: 22, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:34:42,831 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-07 07:34:52,694 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192237.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:34:54,074 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:35:02,889 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9971, 1.5462, 1.7381, 1.4549, 1.0910, 1.5469, 1.9308, 1.5404], + device='cuda:2'), covar=tensor([0.0512, 0.1176, 0.1587, 0.1391, 0.0572, 0.1420, 0.0631, 0.0646], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0153, 0.0188, 0.0159, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:35:07,359 INFO [train.py:901] (2/4) Epoch 24, batch 6350, loss[loss=0.1981, simple_loss=0.2827, pruned_loss=0.05673, over 8336.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05932, over 1612977.05 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:35:36,835 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.329e+02 2.896e+02 3.640e+02 5.459e+02, threshold=5.791e+02, percent-clipped=0.0 +2023-02-07 07:35:43,003 INFO [train.py:901] (2/4) Epoch 24, batch 6400, loss[loss=0.2073, simple_loss=0.3027, pruned_loss=0.05596, over 8336.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2843, pruned_loss=0.05876, over 1614160.71 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:36:19,118 INFO [train.py:901] (2/4) Epoch 24, batch 6450, loss[loss=0.2004, simple_loss=0.2795, pruned_loss=0.06065, over 8341.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05924, over 1611054.86 frames. ], batch size: 26, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:36:49,115 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.491e+02 2.965e+02 3.858e+02 7.678e+02, threshold=5.930e+02, percent-clipped=7.0 +2023-02-07 07:36:54,691 INFO [train.py:901] (2/4) Epoch 24, batch 6500, loss[loss=0.1943, simple_loss=0.2943, pruned_loss=0.04713, over 8331.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2846, pruned_loss=0.05895, over 1609545.10 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:36:56,155 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192410.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:00,677 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192417.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:25,023 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192451.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:29,466 INFO [train.py:901] (2/4) Epoch 24, batch 6550, loss[loss=0.1998, simple_loss=0.2896, pruned_loss=0.05494, over 8619.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2845, pruned_loss=0.05885, over 1614317.76 frames. ], batch size: 31, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:37:33,654 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192464.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:48,320 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 07:37:58,367 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.233e+02 2.734e+02 3.455e+02 6.558e+02, threshold=5.467e+02, percent-clipped=2.0 +2023-02-07 07:38:03,872 INFO [train.py:901] (2/4) Epoch 24, batch 6600, loss[loss=0.2062, simple_loss=0.2904, pruned_loss=0.06103, over 8139.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2837, pruned_loss=0.05902, over 1611611.23 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:38:08,110 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 07:38:10,834 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:38:39,632 INFO [train.py:901] (2/4) Epoch 24, batch 6650, loss[loss=0.2009, simple_loss=0.2916, pruned_loss=0.05512, over 8663.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.285, pruned_loss=0.0596, over 1617668.22 frames. ], batch size: 39, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:38:53,937 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192579.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:38:55,134 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192581.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:38:56,512 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:39:08,651 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.365e+02 2.876e+02 3.746e+02 9.522e+02, threshold=5.752e+02, percent-clipped=3.0 +2023-02-07 07:39:11,895 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-07 07:39:14,187 INFO [train.py:901] (2/4) Epoch 24, batch 6700, loss[loss=0.2224, simple_loss=0.3073, pruned_loss=0.06879, over 8322.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2843, pruned_loss=0.05879, over 1615528.60 frames. ], batch size: 26, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:39:31,147 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1637, 4.1005, 3.7338, 2.8838, 3.6377, 3.7132, 3.8073, 3.5866], + device='cuda:2'), covar=tensor([0.0722, 0.0565, 0.0870, 0.3147, 0.0851, 0.1335, 0.1073, 0.0898], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0446, 0.0434, 0.0546, 0.0434, 0.0450, 0.0429, 0.0392], + 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') +2023-02-07 07:39:31,226 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:39:32,518 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192634.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:39:36,667 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-07 07:39:49,033 INFO [train.py:901] (2/4) Epoch 24, batch 6750, loss[loss=0.1877, simple_loss=0.2775, pruned_loss=0.04895, over 8232.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2849, pruned_loss=0.05925, over 1614105.22 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:40:12,033 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192691.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:15,248 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192696.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:16,608 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:17,823 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.601e+02 3.163e+02 3.841e+02 9.507e+02, threshold=6.325e+02, percent-clipped=3.0 +2023-02-07 07:40:23,402 INFO [train.py:901] (2/4) Epoch 24, batch 6800, loss[loss=0.164, simple_loss=0.2426, pruned_loss=0.04275, over 7809.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2854, pruned_loss=0.05978, over 1616289.45 frames. ], batch size: 19, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:40:24,829 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 07:40:30,217 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 07:40:44,401 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 07:40:54,020 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4492, 4.4636, 4.0228, 2.0786, 3.9311, 3.9369, 3.9565, 3.8506], + device='cuda:2'), covar=tensor([0.0663, 0.0489, 0.1001, 0.3959, 0.0888, 0.0916, 0.1252, 0.0705], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0446, 0.0434, 0.0547, 0.0434, 0.0449, 0.0429, 0.0392], + 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') +2023-02-07 07:40:56,188 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192753.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:56,811 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192754.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:58,551 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.71 vs. limit=5.0 +2023-02-07 07:40:59,525 INFO [train.py:901] (2/4) Epoch 24, batch 6850, loss[loss=0.21, simple_loss=0.294, pruned_loss=0.06302, over 8254.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2856, pruned_loss=0.0599, over 1619236.79 frames. ], batch size: 24, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:41:01,649 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:41:15,299 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 07:41:26,175 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192795.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:41:29,430 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.230e+02 2.864e+02 3.611e+02 9.090e+02, threshold=5.729e+02, percent-clipped=1.0 +2023-02-07 07:41:35,200 INFO [train.py:901] (2/4) Epoch 24, batch 6900, loss[loss=0.222, simple_loss=0.3028, pruned_loss=0.07062, over 8028.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2849, pruned_loss=0.05916, over 1621154.45 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:41:51,988 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4879, 1.2762, 2.2123, 1.1694, 2.1549, 2.3771, 2.5633, 2.0514], + device='cuda:2'), covar=tensor([0.1039, 0.1489, 0.0505, 0.2168, 0.0819, 0.0430, 0.0747, 0.0631], + device='cuda:2'), in_proj_covar=tensor([0.0299, 0.0326, 0.0288, 0.0317, 0.0316, 0.0274, 0.0433, 0.0305], + 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') +2023-02-07 07:41:54,061 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192835.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:03,100 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192849.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:42:08,875 INFO [train.py:901] (2/4) Epoch 24, batch 6950, loss[loss=0.2565, simple_loss=0.3457, pruned_loss=0.08366, over 8494.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2861, pruned_loss=0.05988, over 1625057.99 frames. ], batch size: 26, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:42:10,320 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192860.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:17,129 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192869.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:21,212 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2386, 3.1461, 2.9432, 1.6409, 2.8658, 2.9049, 2.8365, 2.8320], + device='cuda:2'), covar=tensor([0.1193, 0.0861, 0.1269, 0.4626, 0.1190, 0.1419, 0.1601, 0.1084], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0444, 0.0432, 0.0544, 0.0432, 0.0446, 0.0427, 0.0390], + 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') +2023-02-07 07:42:21,972 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:23,173 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 07:42:24,001 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8225, 1.3967, 3.9857, 1.4263, 3.5139, 3.2905, 3.5613, 3.4696], + device='cuda:2'), covar=tensor([0.0727, 0.4323, 0.0611, 0.4224, 0.1220, 0.0996, 0.0773, 0.0803], + device='cuda:2'), in_proj_covar=tensor([0.0644, 0.0654, 0.0710, 0.0639, 0.0715, 0.0614, 0.0617, 0.0686], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 07:42:30,054 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192888.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:38,585 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.384e+02 2.955e+02 3.597e+02 9.319e+02, threshold=5.910e+02, percent-clipped=1.0 +2023-02-07 07:42:44,706 INFO [train.py:901] (2/4) Epoch 24, batch 7000, loss[loss=0.1683, simple_loss=0.247, pruned_loss=0.04482, over 7241.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2864, pruned_loss=0.06045, over 1623390.33 frames. ], batch size: 16, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:42:46,238 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192910.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:48,256 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192913.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:55,821 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-07 07:43:15,536 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192952.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:16,862 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192954.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:19,408 INFO [train.py:901] (2/4) Epoch 24, batch 7050, loss[loss=0.1667, simple_loss=0.2594, pruned_loss=0.03698, over 7928.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2865, pruned_loss=0.06072, over 1622713.28 frames. ], batch size: 20, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:43:32,655 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192977.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:33,234 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192978.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:34,079 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192979.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:48,986 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.407e+02 3.090e+02 3.925e+02 9.689e+02, threshold=6.179e+02, percent-clipped=7.0 +2023-02-07 07:43:54,449 INFO [train.py:901] (2/4) Epoch 24, batch 7100, loss[loss=0.1906, simple_loss=0.2547, pruned_loss=0.0633, over 7540.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2854, pruned_loss=0.0603, over 1618457.17 frames. ], batch size: 18, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:43:57,556 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.38 vs. limit=5.0 +2023-02-07 07:44:14,217 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193035.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:29,517 INFO [train.py:901] (2/4) Epoch 24, batch 7150, loss[loss=0.1914, simple_loss=0.2702, pruned_loss=0.05626, over 7650.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2845, pruned_loss=0.05955, over 1615330.25 frames. ], batch size: 19, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:44:30,374 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193059.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:54,294 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193093.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:56,929 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193097.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:58,917 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.272e+02 2.945e+02 3.915e+02 7.728e+02, threshold=5.890e+02, percent-clipped=4.0 +2023-02-07 07:45:05,030 INFO [train.py:901] (2/4) Epoch 24, batch 7200, loss[loss=0.2151, simple_loss=0.2989, pruned_loss=0.06563, over 8506.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2847, pruned_loss=0.05958, over 1617688.98 frames. ], batch size: 26, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:45:16,876 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193125.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:21,521 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:34,266 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:34,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:39,616 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:40,079 INFO [train.py:901] (2/4) Epoch 24, batch 7250, loss[loss=0.152, simple_loss=0.2341, pruned_loss=0.03493, over 7536.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2837, pruned_loss=0.05935, over 1611869.33 frames. ], batch size: 18, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:45:45,719 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193166.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:03,182 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193191.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:04,339 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193193.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:46:08,901 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.389e+02 2.780e+02 3.377e+02 1.311e+03, threshold=5.561e+02, percent-clipped=2.0 +2023-02-07 07:46:14,403 INFO [train.py:901] (2/4) Epoch 24, batch 7300, loss[loss=0.2077, simple_loss=0.2902, pruned_loss=0.06258, over 8106.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2828, pruned_loss=0.0588, over 1611310.13 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:46:17,193 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193212.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:29,133 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193229.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:44,639 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 07:46:48,081 INFO [train.py:901] (2/4) Epoch 24, batch 7350, loss[loss=0.1877, simple_loss=0.2786, pruned_loss=0.0484, over 8250.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2838, pruned_loss=0.05924, over 1613037.48 frames. ], batch size: 24, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:46:54,914 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5933, 1.8340, 1.5349, 2.2799, 1.0550, 1.3662, 1.6953, 1.8075], + device='cuda:2'), covar=tensor([0.0859, 0.0694, 0.0966, 0.0407, 0.1098, 0.1404, 0.0796, 0.0776], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0197, 0.0245, 0.0214, 0.0205, 0.0247, 0.0251, 0.0207], + 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') +2023-02-07 07:47:00,372 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2776, 1.9681, 2.5269, 1.5827, 1.7277, 2.4822, 1.2395, 2.0995], + device='cuda:2'), covar=tensor([0.1323, 0.1061, 0.0291, 0.1110, 0.1904, 0.0394, 0.1648, 0.0987], + device='cuda:2'), in_proj_covar=tensor([0.0195, 0.0201, 0.0130, 0.0221, 0.0273, 0.0140, 0.0171, 0.0197], + 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') +2023-02-07 07:47:05,436 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 07:47:11,624 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 07:47:17,798 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.477e+02 2.971e+02 3.853e+02 6.522e+02, threshold=5.942e+02, percent-clipped=4.0 +2023-02-07 07:47:19,306 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193302.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:47:23,212 INFO [train.py:901] (2/4) Epoch 24, batch 7400, loss[loss=0.2018, simple_loss=0.2922, pruned_loss=0.05566, over 8434.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2841, pruned_loss=0.05944, over 1613576.60 frames. ], batch size: 49, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:47:23,401 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193308.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:47:26,298 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 07:47:31,906 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 07:47:52,394 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193349.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:47:58,366 INFO [train.py:901] (2/4) Epoch 24, batch 7450, loss[loss=0.1908, simple_loss=0.2793, pruned_loss=0.05116, over 8476.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2856, pruned_loss=0.06035, over 1616427.67 frames. ], batch size: 25, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:48:09,526 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 07:48:10,330 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193374.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:28,444 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.340e+02 2.930e+02 4.048e+02 8.147e+02, threshold=5.861e+02, percent-clipped=5.0 +2023-02-07 07:48:30,642 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193403.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:32,906 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:34,037 INFO [train.py:901] (2/4) Epoch 24, batch 7500, loss[loss=0.1768, simple_loss=0.2497, pruned_loss=0.05196, over 6821.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.0591, over 1613740.76 frames. ], batch size: 15, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:48:50,719 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:56,214 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:49:07,186 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5287, 2.9948, 2.3840, 4.1040, 1.8512, 1.8984, 2.5752, 2.8761], + device='cuda:2'), covar=tensor([0.0709, 0.0734, 0.0778, 0.0208, 0.1010, 0.1310, 0.0864, 0.0747], + device='cuda:2'), in_proj_covar=tensor([0.0234, 0.0198, 0.0245, 0.0214, 0.0205, 0.0248, 0.0252, 0.0207], + 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') +2023-02-07 07:49:08,070 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.69 vs. limit=5.0 +2023-02-07 07:49:09,712 INFO [train.py:901] (2/4) Epoch 24, batch 7550, loss[loss=0.2016, simple_loss=0.3039, pruned_loss=0.04965, over 8642.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2851, pruned_loss=0.0598, over 1610763.42 frames. ], batch size: 49, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:49:16,844 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193468.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:49:19,413 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193472.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:49:34,387 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193493.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:49:39,066 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.474e+02 3.046e+02 3.751e+02 6.843e+02, threshold=6.092e+02, percent-clipped=3.0 +2023-02-07 07:49:40,770 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2350, 2.0766, 2.7205, 2.2163, 2.7434, 2.3367, 2.0883, 1.5549], + device='cuda:2'), covar=tensor([0.5619, 0.4995, 0.2053, 0.4068, 0.2662, 0.3168, 0.2004, 0.5506], + device='cuda:2'), in_proj_covar=tensor([0.0949, 0.0998, 0.0820, 0.0969, 0.1010, 0.0911, 0.0760, 0.0834], + 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') +2023-02-07 07:49:45,273 INFO [train.py:901] (2/4) Epoch 24, batch 7600, loss[loss=0.1908, simple_loss=0.2797, pruned_loss=0.051, over 8106.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2853, pruned_loss=0.05997, over 1611582.96 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:49:51,992 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193518.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:50:12,034 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9643, 1.4462, 1.6874, 1.3905, 1.0268, 1.4576, 1.9277, 1.6513], + device='cuda:2'), covar=tensor([0.0550, 0.1316, 0.1704, 0.1490, 0.0622, 0.1551, 0.0684, 0.0640], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0188, 0.0159, 0.0100, 0.0162, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 07:50:19,356 INFO [train.py:901] (2/4) Epoch 24, batch 7650, loss[loss=0.1671, simple_loss=0.2524, pruned_loss=0.04094, over 7912.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2849, pruned_loss=0.05956, over 1611043.31 frames. ], batch size: 20, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:50:24,425 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193564.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:50:30,317 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193573.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:50:41,160 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193589.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:50:48,585 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.626e+02 3.196e+02 4.372e+02 7.437e+02, threshold=6.392e+02, percent-clipped=4.0 +2023-02-07 07:50:53,961 INFO [train.py:901] (2/4) Epoch 24, batch 7700, loss[loss=0.2444, simple_loss=0.3305, pruned_loss=0.07912, over 8294.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2863, pruned_loss=0.06056, over 1613108.48 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:51:11,800 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193633.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:51:13,831 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4886, 1.6440, 2.1151, 1.3885, 1.5104, 1.7456, 1.5127, 1.5804], + device='cuda:2'), covar=tensor([0.1932, 0.2467, 0.0921, 0.4518, 0.2005, 0.3251, 0.2406, 0.2079], + device='cuda:2'), in_proj_covar=tensor([0.0531, 0.0618, 0.0557, 0.0653, 0.0652, 0.0600, 0.0548, 0.0637], + 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') +2023-02-07 07:51:14,903 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 07:51:17,909 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 07:51:20,928 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193646.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:51:29,069 INFO [train.py:901] (2/4) Epoch 24, batch 7750, loss[loss=0.1773, simple_loss=0.2512, pruned_loss=0.05166, over 7818.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2855, pruned_loss=0.06028, over 1614733.40 frames. ], batch size: 20, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:51:50,190 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193688.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:51:58,096 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.447e+02 2.679e+02 3.147e+02 3.999e+02 8.742e+02, threshold=6.294e+02, percent-clipped=3.0 +2023-02-07 07:52:03,342 INFO [train.py:901] (2/4) Epoch 24, batch 7800, loss[loss=0.2163, simple_loss=0.3046, pruned_loss=0.064, over 8451.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2852, pruned_loss=0.05961, over 1610030.52 frames. ], batch size: 27, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:52:26,012 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6076, 1.9961, 3.0780, 1.4787, 2.1894, 2.0716, 1.6411, 2.3222], + device='cuda:2'), covar=tensor([0.1934, 0.2764, 0.0796, 0.4585, 0.2182, 0.3225, 0.2521, 0.2414], + device='cuda:2'), in_proj_covar=tensor([0.0530, 0.0617, 0.0557, 0.0651, 0.0651, 0.0599, 0.0548, 0.0637], + 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') +2023-02-07 07:52:37,261 INFO [train.py:901] (2/4) Epoch 24, batch 7850, loss[loss=0.1786, simple_loss=0.2649, pruned_loss=0.04617, over 8195.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2854, pruned_loss=0.05976, over 1608121.54 frames. ], batch size: 23, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:52:39,508 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:52:48,224 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193774.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:52:54,182 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:53:05,054 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193799.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:53:05,510 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.471e+02 2.801e+02 3.652e+02 8.352e+02, threshold=5.603e+02, percent-clipped=2.0 +2023-02-07 07:53:08,051 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.38 vs. limit=5.0 +2023-02-07 07:53:10,834 INFO [train.py:901] (2/4) Epoch 24, batch 7900, loss[loss=0.165, simple_loss=0.2398, pruned_loss=0.04513, over 7429.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2845, pruned_loss=0.05917, over 1609853.78 frames. ], batch size: 17, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:53:11,731 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1845, 1.0593, 1.2791, 1.0344, 0.9078, 1.2898, 0.0867, 0.9626], + device='cuda:2'), covar=tensor([0.1501, 0.1188, 0.0464, 0.0728, 0.2504, 0.0587, 0.1976, 0.1165], + device='cuda:2'), in_proj_covar=tensor([0.0196, 0.0202, 0.0130, 0.0221, 0.0274, 0.0139, 0.0172, 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') +2023-02-07 07:53:16,269 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193816.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:53:28,542 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 07:53:43,829 INFO [train.py:901] (2/4) Epoch 24, batch 7950, loss[loss=0.2078, simple_loss=0.2973, pruned_loss=0.05921, over 8758.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2851, pruned_loss=0.05965, over 1611230.38 frames. ], batch size: 30, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:54:02,330 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8118, 1.4981, 2.9063, 1.3788, 2.1979, 3.0917, 3.2500, 2.6658], + device='cuda:2'), covar=tensor([0.1104, 0.1545, 0.0335, 0.2130, 0.0847, 0.0285, 0.0633, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0323, 0.0288, 0.0316, 0.0316, 0.0272, 0.0430, 0.0304], + 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') +2023-02-07 07:54:11,266 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193898.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:54:12,502 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.494e+02 3.061e+02 3.521e+02 6.741e+02, threshold=6.122e+02, percent-clipped=2.0 +2023-02-07 07:54:13,606 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.22 vs. limit=5.0 +2023-02-07 07:54:13,965 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193902.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:54:17,817 INFO [train.py:901] (2/4) Epoch 24, batch 8000, loss[loss=0.222, simple_loss=0.3071, pruned_loss=0.06846, over 8578.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2859, pruned_loss=0.06, over 1613430.55 frames. ], batch size: 31, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:54:25,815 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 07:54:33,388 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193931.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:54:42,133 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193944.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:54:51,049 INFO [train.py:901] (2/4) Epoch 24, batch 8050, loss[loss=0.218, simple_loss=0.302, pruned_loss=0.06703, over 7445.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2843, pruned_loss=0.0602, over 1593128.89 frames. ], batch size: 17, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:54:58,154 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193969.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:55:03,439 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193977.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:55:23,351 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 07:55:28,455 INFO [train.py:901] (2/4) Epoch 25, batch 0, loss[loss=0.2383, simple_loss=0.2949, pruned_loss=0.09081, over 7541.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.2949, pruned_loss=0.09081, over 7541.00 frames. ], batch size: 18, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:55:28,455 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 07:55:39,670 INFO [train.py:935] (2/4) Epoch 25, validation: loss=0.1722, simple_loss=0.2724, pruned_loss=0.03604, over 944034.00 frames. +2023-02-07 07:55:39,671 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 07:55:46,475 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.577e+02 3.086e+02 3.975e+02 9.885e+02, threshold=6.172e+02, percent-clipped=3.0 +2023-02-07 07:55:57,074 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 07:56:00,133 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194017.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:15,966 INFO [train.py:901] (2/4) Epoch 25, batch 50, loss[loss=0.2081, simple_loss=0.2947, pruned_loss=0.06076, over 8182.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2857, pruned_loss=0.05744, over 366118.39 frames. ], batch size: 23, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:56:17,563 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194042.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:32,519 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 07:56:51,133 INFO [train.py:901] (2/4) Epoch 25, batch 100, loss[loss=0.2381, simple_loss=0.3325, pruned_loss=0.07188, over 8486.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2869, pruned_loss=0.05929, over 644985.71 frames. ], batch size: 28, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:56:51,282 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194090.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:52,672 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194092.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:55,684 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 07:56:57,734 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.655e+02 3.251e+02 4.247e+02 7.218e+02, threshold=6.502e+02, percent-clipped=2.0 +2023-02-07 07:57:19,066 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 07:57:22,502 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 07:57:25,371 INFO [train.py:901] (2/4) Epoch 25, batch 150, loss[loss=0.2087, simple_loss=0.2797, pruned_loss=0.06881, over 7531.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2888, pruned_loss=0.06135, over 858675.81 frames. ], batch size: 18, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:57:35,092 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194154.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:57:52,138 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194179.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:57:58,199 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194187.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:58:00,648 INFO [train.py:901] (2/4) Epoch 25, batch 200, loss[loss=0.2034, simple_loss=0.2901, pruned_loss=0.05836, over 8145.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2866, pruned_loss=0.05989, over 1027842.56 frames. ], batch size: 22, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:58:07,390 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.343e+02 2.842e+02 3.543e+02 5.999e+02, threshold=5.685e+02, percent-clipped=0.0 +2023-02-07 07:58:16,690 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194212.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:58:35,360 INFO [train.py:901] (2/4) Epoch 25, batch 250, loss[loss=0.179, simple_loss=0.2582, pruned_loss=0.04984, over 8069.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2878, pruned_loss=0.06074, over 1157472.48 frames. ], batch size: 21, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:58:39,446 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194246.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:58:49,462 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 07:58:58,129 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 07:59:08,685 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-07 07:59:09,575 INFO [train.py:901] (2/4) Epoch 25, batch 300, loss[loss=0.1908, simple_loss=0.27, pruned_loss=0.0558, over 7970.00 frames. ], tot_loss[loss=0.205, simple_loss=0.288, pruned_loss=0.06105, over 1260693.54 frames. ], batch size: 21, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:59:17,105 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.353e+02 2.857e+02 3.504e+02 7.851e+02, threshold=5.715e+02, percent-clipped=2.0 +2023-02-07 07:59:19,403 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0172, 1.7326, 3.2990, 1.4545, 2.3084, 3.5333, 3.6584, 3.0206], + device='cuda:2'), covar=tensor([0.1204, 0.1652, 0.0343, 0.2233, 0.1126, 0.0287, 0.0721, 0.0532], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0322, 0.0287, 0.0316, 0.0316, 0.0273, 0.0429, 0.0303], + 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') +2023-02-07 07:59:43,441 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194336.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:59:45,911 INFO [train.py:901] (2/4) Epoch 25, batch 350, loss[loss=0.1991, simple_loss=0.2851, pruned_loss=0.05656, over 7973.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.287, pruned_loss=0.06069, over 1337865.93 frames. ], batch size: 21, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:59:51,532 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194348.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:00,425 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194361.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:08,052 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194371.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:09,431 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:14,181 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194380.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:00:20,650 INFO [train.py:901] (2/4) Epoch 25, batch 400, loss[loss=0.1862, simple_loss=0.2747, pruned_loss=0.04889, over 8457.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2872, pruned_loss=0.06079, over 1403366.54 frames. ], batch size: 25, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 08:00:27,610 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.445e+02 3.013e+02 3.982e+02 8.525e+02, threshold=6.027e+02, percent-clipped=7.0 +2023-02-07 08:00:29,487 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 +2023-02-07 08:00:52,193 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194434.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:56,958 INFO [train.py:901] (2/4) Epoch 25, batch 450, loss[loss=0.2053, simple_loss=0.274, pruned_loss=0.06834, over 7647.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2864, pruned_loss=0.05986, over 1450576.66 frames. ], batch size: 19, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 08:01:30,923 INFO [train.py:901] (2/4) Epoch 25, batch 500, loss[loss=0.2434, simple_loss=0.3298, pruned_loss=0.07849, over 8412.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2879, pruned_loss=0.0616, over 1487125.42 frames. ], batch size: 49, lr: 3.04e-03, grad_scale: 16.0 +2023-02-07 08:01:37,840 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.459e+02 3.156e+02 4.025e+02 7.800e+02, threshold=6.312e+02, percent-clipped=3.0 +2023-02-07 08:02:06,155 INFO [train.py:901] (2/4) Epoch 25, batch 550, loss[loss=0.1965, simple_loss=0.284, pruned_loss=0.05447, over 8105.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2874, pruned_loss=0.06116, over 1515142.75 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 16.0 +2023-02-07 08:02:08,553 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8426, 1.4805, 1.7443, 1.4762, 1.0303, 1.5344, 1.7442, 1.6082], + device='cuda:2'), covar=tensor([0.0572, 0.1239, 0.1593, 0.1421, 0.0621, 0.1463, 0.0737, 0.0627], + device='cuda:2'), in_proj_covar=tensor([0.0097, 0.0152, 0.0188, 0.0159, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 08:02:13,479 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194549.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:02:24,462 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2205, 3.1418, 2.9035, 1.6178, 2.8273, 2.8847, 2.8924, 2.7349], + device='cuda:2'), covar=tensor([0.1250, 0.0931, 0.1433, 0.5053, 0.1174, 0.1346, 0.1638, 0.1177], + device='cuda:2'), in_proj_covar=tensor([0.0535, 0.0454, 0.0435, 0.0552, 0.0436, 0.0457, 0.0433, 0.0398], + 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') +2023-02-07 08:02:38,877 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4818, 2.3623, 3.0432, 2.5307, 3.1051, 2.5979, 2.3983, 1.8739], + device='cuda:2'), covar=tensor([0.5742, 0.5033, 0.2204, 0.3941, 0.2574, 0.2943, 0.1789, 0.5606], + device='cuda:2'), in_proj_covar=tensor([0.0946, 0.0999, 0.0818, 0.0964, 0.1005, 0.0908, 0.0758, 0.0834], + 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') +2023-02-07 08:02:40,191 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9118, 1.4797, 1.7364, 1.3986, 1.0721, 1.4479, 1.7966, 1.4521], + device='cuda:2'), covar=tensor([0.0577, 0.1268, 0.1660, 0.1466, 0.0612, 0.1518, 0.0693, 0.0699], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0189, 0.0159, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 08:02:42,130 INFO [train.py:901] (2/4) Epoch 25, batch 600, loss[loss=0.202, simple_loss=0.2947, pruned_loss=0.05463, over 8500.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2869, pruned_loss=0.06079, over 1539519.35 frames. ], batch size: 26, lr: 3.04e-03, grad_scale: 16.0 +2023-02-07 08:02:48,737 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.589e+02 2.361e+02 2.970e+02 3.663e+02 1.001e+03, threshold=5.941e+02, percent-clipped=3.0 +2023-02-07 08:03:01,141 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 08:03:01,339 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194617.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:16,776 INFO [train.py:901] (2/4) Epoch 25, batch 650, loss[loss=0.2038, simple_loss=0.2875, pruned_loss=0.06003, over 8592.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2864, pruned_loss=0.06034, over 1559668.14 frames. ], batch size: 39, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:03:18,074 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194642.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:25,643 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194652.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:35,792 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7134, 1.6886, 2.2886, 1.5071, 1.3146, 2.2694, 0.3796, 1.4664], + device='cuda:2'), covar=tensor([0.1444, 0.1118, 0.0336, 0.0970, 0.2324, 0.0346, 0.1800, 0.1030], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0198, 0.0129, 0.0218, 0.0270, 0.0138, 0.0169, 0.0195], + 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') +2023-02-07 08:03:45,808 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194680.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:52,562 INFO [train.py:901] (2/4) Epoch 25, batch 700, loss[loss=0.2382, simple_loss=0.316, pruned_loss=0.0802, over 8580.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2867, pruned_loss=0.06033, over 1570609.37 frames. ], batch size: 31, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:04:00,039 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.448e+02 2.849e+02 3.638e+02 5.412e+02, threshold=5.698e+02, percent-clipped=0.0 +2023-02-07 08:04:09,641 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194715.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:04:09,782 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8338, 2.1679, 3.6178, 1.7993, 1.7937, 3.5028, 0.5569, 2.1852], + device='cuda:2'), covar=tensor([0.1343, 0.1154, 0.0220, 0.1590, 0.2311, 0.0254, 0.2024, 0.1118], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0198, 0.0129, 0.0218, 0.0270, 0.0138, 0.0169, 0.0195], + 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') +2023-02-07 08:04:16,433 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194724.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:04:27,401 INFO [train.py:901] (2/4) Epoch 25, batch 750, loss[loss=0.1578, simple_loss=0.2501, pruned_loss=0.03276, over 7414.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2857, pruned_loss=0.06025, over 1579739.19 frames. ], batch size: 17, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:04:42,121 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6417, 2.5973, 1.7920, 2.2621, 2.1894, 1.5279, 2.1009, 2.1565], + device='cuda:2'), covar=tensor([0.1734, 0.0462, 0.1443, 0.0778, 0.0829, 0.1755, 0.1183, 0.1122], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0235, 0.0337, 0.0310, 0.0300, 0.0342, 0.0346, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 08:04:49,489 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 08:04:58,548 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 08:05:03,379 INFO [train.py:901] (2/4) Epoch 25, batch 800, loss[loss=0.2168, simple_loss=0.2963, pruned_loss=0.0686, over 8358.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2844, pruned_loss=0.0601, over 1580661.03 frames. ], batch size: 26, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:05:07,600 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194795.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:11,591 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.325e+02 2.417e+02 2.991e+02 3.771e+02 6.788e+02, threshold=5.982e+02, percent-clipped=2.0 +2023-02-07 08:05:14,552 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194805.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:19,316 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9134, 6.0075, 5.2139, 2.4933, 5.3297, 5.6606, 5.4419, 5.5015], + device='cuda:2'), covar=tensor([0.0493, 0.0355, 0.0892, 0.4348, 0.0751, 0.0617, 0.1049, 0.0480], + device='cuda:2'), in_proj_covar=tensor([0.0533, 0.0452, 0.0436, 0.0551, 0.0436, 0.0456, 0.0432, 0.0398], + 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') +2023-02-07 08:05:31,802 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194830.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:31,827 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194830.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:34,489 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1049, 1.6458, 1.4161, 1.5827, 1.4078, 1.2706, 1.3594, 1.3017], + device='cuda:2'), covar=tensor([0.1081, 0.0496, 0.1295, 0.0568, 0.0770, 0.1588, 0.0903, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0354, 0.0234, 0.0335, 0.0308, 0.0298, 0.0339, 0.0344, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 08:05:37,856 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194839.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:05:38,297 INFO [train.py:901] (2/4) Epoch 25, batch 850, loss[loss=0.2407, simple_loss=0.3247, pruned_loss=0.07829, over 8605.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2859, pruned_loss=0.06077, over 1594146.87 frames. ], batch size: 31, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:05:56,607 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:06:14,011 INFO [train.py:901] (2/4) Epoch 25, batch 900, loss[loss=0.1885, simple_loss=0.2681, pruned_loss=0.05448, over 7423.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2856, pruned_loss=0.06057, over 1595484.94 frames. ], batch size: 17, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:06:17,739 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7800, 1.3861, 1.6194, 1.3487, 0.9919, 1.4479, 1.6313, 1.3195], + device='cuda:2'), covar=tensor([0.0567, 0.1318, 0.1651, 0.1474, 0.0616, 0.1496, 0.0726, 0.0751], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0101, 0.0164, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 08:06:22,179 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.491e+02 2.923e+02 3.701e+02 8.623e+02, threshold=5.846e+02, percent-clipped=3.0 +2023-02-07 08:06:34,964 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6716, 1.7170, 2.5159, 1.5307, 1.2939, 2.4718, 0.4189, 1.5288], + device='cuda:2'), covar=tensor([0.1948, 0.1243, 0.0309, 0.1258, 0.2560, 0.0315, 0.2021, 0.1246], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0198, 0.0128, 0.0218, 0.0269, 0.0137, 0.0169, 0.0194], + 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') +2023-02-07 08:06:49,722 INFO [train.py:901] (2/4) Epoch 25, batch 950, loss[loss=0.1958, simple_loss=0.278, pruned_loss=0.05682, over 8089.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2861, pruned_loss=0.06023, over 1600665.94 frames. ], batch size: 21, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:07:19,513 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 08:07:24,290 INFO [train.py:901] (2/4) Epoch 25, batch 1000, loss[loss=0.1687, simple_loss=0.2494, pruned_loss=0.04399, over 7214.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2858, pruned_loss=0.06024, over 1602606.59 frames. ], batch size: 16, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:07:25,388 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 08:07:29,064 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194996.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:07:32,158 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.541e+02 2.651e+02 3.101e+02 3.894e+02 6.477e+02, threshold=6.202e+02, percent-clipped=4.0 +2023-02-07 08:07:51,778 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195029.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:07:54,450 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 08:07:59,734 INFO [train.py:901] (2/4) Epoch 25, batch 1050, loss[loss=0.2245, simple_loss=0.2983, pruned_loss=0.07529, over 8024.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2864, pruned_loss=0.06072, over 1602598.27 frames. ], batch size: 22, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:08:06,387 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 08:08:07,157 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195051.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:14,489 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195062.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:08:23,831 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195076.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:31,299 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195086.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:31,922 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3928, 1.3942, 4.5645, 1.7468, 4.0501, 3.7662, 4.1254, 4.0146], + device='cuda:2'), covar=tensor([0.0559, 0.5006, 0.0485, 0.4210, 0.1014, 0.0943, 0.0559, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0640, 0.0646, 0.0702, 0.0637, 0.0712, 0.0605, 0.0610, 0.0684], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 08:08:33,316 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195089.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:33,849 INFO [train.py:901] (2/4) Epoch 25, batch 1100, loss[loss=0.2004, simple_loss=0.292, pruned_loss=0.05444, over 8241.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.288, pruned_loss=0.0615, over 1609433.73 frames. ], batch size: 24, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:08:37,351 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195095.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:08:41,228 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.550e+02 3.152e+02 4.111e+02 6.650e+02, threshold=6.304e+02, percent-clipped=3.0 +2023-02-07 08:08:48,159 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195111.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:48,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195111.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:53,475 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5746, 4.5886, 4.0850, 2.1959, 4.0560, 4.2352, 4.1339, 4.1027], + device='cuda:2'), covar=tensor([0.0648, 0.0455, 0.0907, 0.4518, 0.0796, 0.0814, 0.1078, 0.0682], + device='cuda:2'), in_proj_covar=tensor([0.0531, 0.0450, 0.0432, 0.0545, 0.0434, 0.0452, 0.0426, 0.0396], + 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') +2023-02-07 08:08:54,917 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195120.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:09:09,221 INFO [train.py:901] (2/4) Epoch 25, batch 1150, loss[loss=0.2548, simple_loss=0.3367, pruned_loss=0.08647, over 8337.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2868, pruned_loss=0.06094, over 1611183.26 frames. ], batch size: 26, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:09:16,871 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 08:09:43,696 INFO [train.py:901] (2/4) Epoch 25, batch 1200, loss[loss=0.1659, simple_loss=0.2516, pruned_loss=0.04009, over 7524.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2857, pruned_loss=0.06, over 1611086.15 frames. ], batch size: 18, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:09:44,707 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 08:09:51,821 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.229e+02 2.843e+02 3.492e+02 1.399e+03, threshold=5.685e+02, percent-clipped=2.0 +2023-02-07 08:09:57,122 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195209.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:10:18,379 INFO [train.py:901] (2/4) Epoch 25, batch 1250, loss[loss=0.1659, simple_loss=0.2496, pruned_loss=0.04104, over 7240.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2857, pruned_loss=0.06045, over 1610247.68 frames. ], batch size: 16, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:10:34,134 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.69 vs. limit=5.0 +2023-02-07 08:10:53,103 INFO [train.py:901] (2/4) Epoch 25, batch 1300, loss[loss=0.2042, simple_loss=0.3028, pruned_loss=0.05281, over 8024.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2857, pruned_loss=0.06024, over 1608985.10 frames. ], batch size: 22, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:11:00,278 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.452e+02 2.850e+02 4.025e+02 1.071e+03, threshold=5.700e+02, percent-clipped=7.0 +2023-02-07 08:11:16,291 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:11:26,831 INFO [train.py:901] (2/4) Epoch 25, batch 1350, loss[loss=0.2251, simple_loss=0.3084, pruned_loss=0.07087, over 7793.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.286, pruned_loss=0.06051, over 1609436.79 frames. ], batch size: 20, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:11:45,671 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195367.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:11:50,114 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:11:57,385 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-07 08:12:02,392 INFO [train.py:901] (2/4) Epoch 25, batch 1400, loss[loss=0.1998, simple_loss=0.2919, pruned_loss=0.05379, over 8317.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2843, pruned_loss=0.05951, over 1611044.10 frames. ], batch size: 25, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:12:04,001 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195392.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:12:10,480 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.570e+02 2.915e+02 3.833e+02 8.465e+02, threshold=5.831e+02, percent-clipped=6.0 +2023-02-07 08:12:13,227 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195406.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:12:15,773 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195410.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:12:31,396 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195433.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:12:36,125 INFO [train.py:901] (2/4) Epoch 25, batch 1450, loss[loss=0.2327, simple_loss=0.3229, pruned_loss=0.07124, over 8327.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2843, pruned_loss=0.05987, over 1610802.98 frames. ], batch size: 25, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:12:36,339 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6389, 2.1912, 3.4062, 1.6993, 1.7314, 3.4008, 0.7326, 2.1086], + device='cuda:2'), covar=tensor([0.1661, 0.1288, 0.0262, 0.1698, 0.2318, 0.0290, 0.2046, 0.1517], + device='cuda:2'), in_proj_covar=tensor([0.0195, 0.0201, 0.0130, 0.0221, 0.0273, 0.0139, 0.0171, 0.0196], + 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') +2023-02-07 08:12:44,148 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 08:13:10,177 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195488.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:13:11,402 INFO [train.py:901] (2/4) Epoch 25, batch 1500, loss[loss=0.1913, simple_loss=0.2669, pruned_loss=0.05783, over 8081.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2849, pruned_loss=0.05983, over 1614309.59 frames. ], batch size: 21, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:13:19,806 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.401e+02 3.375e+02 4.255e+02 1.024e+03, threshold=6.749e+02, percent-clipped=12.0 +2023-02-07 08:13:33,926 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195521.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:13:46,247 INFO [train.py:901] (2/4) Epoch 25, batch 1550, loss[loss=0.2121, simple_loss=0.2878, pruned_loss=0.06818, over 8102.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.285, pruned_loss=0.05974, over 1616977.15 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:13:51,917 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:14:14,118 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195580.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:14:21,359 INFO [train.py:901] (2/4) Epoch 25, batch 1600, loss[loss=0.2228, simple_loss=0.3049, pruned_loss=0.07036, over 8435.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2845, pruned_loss=0.05939, over 1611222.65 frames. ], batch size: 29, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:14:29,485 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.317e+02 3.105e+02 3.813e+02 7.132e+02, threshold=6.211e+02, percent-clipped=3.0 +2023-02-07 08:14:32,325 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195605.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:14:37,017 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2058, 2.0178, 2.6869, 2.1951, 2.6957, 2.2857, 2.1177, 1.5906], + device='cuda:2'), covar=tensor([0.5670, 0.5427, 0.2169, 0.4102, 0.2639, 0.3364, 0.1998, 0.5745], + device='cuda:2'), in_proj_covar=tensor([0.0954, 0.1009, 0.0826, 0.0977, 0.1019, 0.0919, 0.0765, 0.0842], + 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') +2023-02-07 08:14:55,981 INFO [train.py:901] (2/4) Epoch 25, batch 1650, loss[loss=0.1692, simple_loss=0.2585, pruned_loss=0.03994, over 7989.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2846, pruned_loss=0.05944, over 1609559.18 frames. ], batch size: 21, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:15:29,744 INFO [train.py:901] (2/4) Epoch 25, batch 1700, loss[loss=0.1791, simple_loss=0.256, pruned_loss=0.0511, over 7698.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2843, pruned_loss=0.05915, over 1609914.69 frames. ], batch size: 18, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:15:38,027 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.590e+02 3.116e+02 3.996e+02 7.880e+02, threshold=6.232e+02, percent-clipped=2.0 +2023-02-07 08:16:05,347 INFO [train.py:901] (2/4) Epoch 25, batch 1750, loss[loss=0.1737, simple_loss=0.2587, pruned_loss=0.04438, over 7650.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2848, pruned_loss=0.05971, over 1608660.26 frames. ], batch size: 19, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:16:06,321 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5956, 2.9907, 2.5196, 4.0726, 1.7527, 2.1696, 2.5246, 2.9733], + device='cuda:2'), covar=tensor([0.0643, 0.0710, 0.0720, 0.0198, 0.1046, 0.1127, 0.0859, 0.0720], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0245, 0.0213, 0.0205, 0.0245, 0.0249, 0.0206], + 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') +2023-02-07 08:16:09,112 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195744.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:16:15,705 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195754.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:16:17,273 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4839, 2.4142, 3.1828, 2.4816, 3.1501, 2.5357, 2.3016, 1.9786], + device='cuda:2'), covar=tensor([0.5430, 0.4946, 0.1868, 0.3766, 0.2447, 0.2955, 0.1846, 0.5415], + device='cuda:2'), in_proj_covar=tensor([0.0953, 0.1008, 0.0825, 0.0976, 0.1016, 0.0917, 0.0764, 0.0841], + 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') +2023-02-07 08:16:26,010 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195769.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:16:31,518 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195777.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:16:40,098 INFO [train.py:901] (2/4) Epoch 25, batch 1800, loss[loss=0.1643, simple_loss=0.2457, pruned_loss=0.04149, over 7810.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2851, pruned_loss=0.06015, over 1609112.61 frames. ], batch size: 19, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:16:48,976 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.456e+02 2.857e+02 3.484e+02 7.816e+02, threshold=5.715e+02, percent-clipped=1.0 +2023-02-07 08:16:49,215 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195802.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:16:50,549 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195804.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:17:07,905 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195829.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:17:15,239 INFO [train.py:901] (2/4) Epoch 25, batch 1850, loss[loss=0.2038, simple_loss=0.2945, pruned_loss=0.05653, over 8444.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2862, pruned_loss=0.0602, over 1610647.85 frames. ], batch size: 25, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:17:36,394 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195869.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:17:50,144 INFO [train.py:901] (2/4) Epoch 25, batch 1900, loss[loss=0.1869, simple_loss=0.2704, pruned_loss=0.05173, over 8248.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2858, pruned_loss=0.06016, over 1607852.63 frames. ], batch size: 24, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:17:58,360 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.686e+02 3.045e+02 3.689e+02 8.196e+02, threshold=6.090e+02, percent-clipped=3.0 +2023-02-07 08:18:24,461 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 08:18:25,126 INFO [train.py:901] (2/4) Epoch 25, batch 1950, loss[loss=0.1583, simple_loss=0.2452, pruned_loss=0.03565, over 7801.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2857, pruned_loss=0.05979, over 1613403.19 frames. ], batch size: 19, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:18:37,850 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 08:18:56,451 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2023-02-07 08:18:57,285 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 08:18:58,124 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5209, 2.4578, 1.7908, 2.2694, 2.0978, 1.5065, 1.9967, 2.1343], + device='cuda:2'), covar=tensor([0.1464, 0.0383, 0.1199, 0.0623, 0.0754, 0.1551, 0.1040, 0.0941], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0237, 0.0341, 0.0312, 0.0302, 0.0345, 0.0349, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 08:19:00,594 INFO [train.py:901] (2/4) Epoch 25, batch 2000, loss[loss=0.2102, simple_loss=0.2994, pruned_loss=0.0605, over 8619.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2853, pruned_loss=0.05941, over 1615660.74 frames. ], batch size: 50, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:19:09,749 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.344e+02 2.823e+02 3.287e+02 7.423e+02, threshold=5.646e+02, percent-clipped=4.0 +2023-02-07 08:19:36,116 INFO [train.py:901] (2/4) Epoch 25, batch 2050, loss[loss=0.2024, simple_loss=0.2895, pruned_loss=0.05769, over 8344.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2846, pruned_loss=0.05899, over 1617073.43 frames. ], batch size: 26, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:20:11,105 INFO [train.py:901] (2/4) Epoch 25, batch 2100, loss[loss=0.2109, simple_loss=0.2909, pruned_loss=0.06543, over 8079.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2848, pruned_loss=0.05913, over 1617799.82 frames. ], batch size: 21, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:20:20,386 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.403e+02 2.946e+02 3.659e+02 8.101e+02, threshold=5.892e+02, percent-clipped=3.0 +2023-02-07 08:20:35,889 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196125.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:20:46,046 INFO [train.py:901] (2/4) Epoch 25, batch 2150, loss[loss=0.1866, simple_loss=0.2597, pruned_loss=0.0567, over 7794.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2858, pruned_loss=0.0596, over 1619397.46 frames. ], batch size: 19, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:20:54,019 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:21:22,072 INFO [train.py:901] (2/4) Epoch 25, batch 2200, loss[loss=0.2032, simple_loss=0.2898, pruned_loss=0.05832, over 8460.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2867, pruned_loss=0.06038, over 1621811.33 frames. ], batch size: 27, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:21:30,652 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.555e+02 3.213e+02 4.289e+02 6.887e+02, threshold=6.426e+02, percent-clipped=5.0 +2023-02-07 08:21:37,544 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2424, 3.1655, 2.9092, 1.9065, 2.8679, 2.8703, 2.8404, 2.7760], + device='cuda:2'), covar=tensor([0.0898, 0.0803, 0.1188, 0.3718, 0.0974, 0.1294, 0.1492, 0.0942], + device='cuda:2'), in_proj_covar=tensor([0.0532, 0.0449, 0.0435, 0.0546, 0.0433, 0.0452, 0.0427, 0.0396], + 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') +2023-02-07 08:21:56,963 INFO [train.py:901] (2/4) Epoch 25, batch 2250, loss[loss=0.1889, simple_loss=0.2647, pruned_loss=0.05656, over 8289.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2867, pruned_loss=0.06047, over 1624637.10 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:22:07,576 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0478, 1.5750, 3.4326, 1.5708, 2.4024, 3.8070, 3.9019, 3.2755], + device='cuda:2'), covar=tensor([0.1167, 0.1852, 0.0377, 0.2112, 0.1202, 0.0232, 0.0495, 0.0542], + device='cuda:2'), in_proj_covar=tensor([0.0300, 0.0325, 0.0289, 0.0317, 0.0317, 0.0275, 0.0432, 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') +2023-02-07 08:22:12,519 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5970, 2.4572, 1.8252, 2.3190, 2.1413, 1.5948, 2.0582, 2.1202], + device='cuda:2'), covar=tensor([0.1459, 0.0441, 0.1278, 0.0632, 0.0774, 0.1599, 0.1063, 0.0990], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0238, 0.0341, 0.0313, 0.0303, 0.0345, 0.0350, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 08:22:32,057 INFO [train.py:901] (2/4) Epoch 25, batch 2300, loss[loss=0.185, simple_loss=0.2813, pruned_loss=0.0443, over 8324.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2871, pruned_loss=0.06026, over 1624737.42 frames. ], batch size: 25, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:22:40,954 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.420e+02 2.794e+02 3.530e+02 9.865e+02, threshold=5.587e+02, percent-clipped=2.0 +2023-02-07 08:23:04,009 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8567, 1.9659, 1.7165, 2.6611, 1.1382, 1.5345, 1.8525, 1.9714], + device='cuda:2'), covar=tensor([0.0715, 0.0784, 0.0928, 0.0384, 0.1152, 0.1364, 0.0821, 0.0809], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0194, 0.0242, 0.0210, 0.0203, 0.0243, 0.0247, 0.0205], + 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') +2023-02-07 08:23:07,217 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196339.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:23:07,751 INFO [train.py:901] (2/4) Epoch 25, batch 2350, loss[loss=0.188, simple_loss=0.2631, pruned_loss=0.05645, over 7698.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2862, pruned_loss=0.05971, over 1624641.04 frames. ], batch size: 18, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:23:34,294 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:23:42,090 INFO [train.py:901] (2/4) Epoch 25, batch 2400, loss[loss=0.194, simple_loss=0.2731, pruned_loss=0.05743, over 7911.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.285, pruned_loss=0.05918, over 1620446.32 frames. ], batch size: 20, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:23:50,271 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.420e+02 2.902e+02 3.432e+02 7.434e+02, threshold=5.805e+02, percent-clipped=2.0 +2023-02-07 08:24:17,357 INFO [train.py:901] (2/4) Epoch 25, batch 2450, loss[loss=0.1561, simple_loss=0.2328, pruned_loss=0.03974, over 7436.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2844, pruned_loss=0.05862, over 1623574.70 frames. ], batch size: 17, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:24:51,933 INFO [train.py:901] (2/4) Epoch 25, batch 2500, loss[loss=0.2539, simple_loss=0.3303, pruned_loss=0.08874, over 6941.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2842, pruned_loss=0.05854, over 1623080.50 frames. ], batch size: 71, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:25:00,795 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.398e+02 2.858e+02 3.242e+02 5.404e+02, threshold=5.717e+02, percent-clipped=0.0 +2023-02-07 08:25:26,994 INFO [train.py:901] (2/4) Epoch 25, batch 2550, loss[loss=0.1888, simple_loss=0.2742, pruned_loss=0.05167, over 8130.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2849, pruned_loss=0.05898, over 1625063.17 frames. ], batch size: 22, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:26:02,156 INFO [train.py:901] (2/4) Epoch 25, batch 2600, loss[loss=0.2085, simple_loss=0.2878, pruned_loss=0.06463, over 8488.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05932, over 1619581.10 frames. ], batch size: 29, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:26:06,422 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196596.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:26:10,247 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.366e+02 2.911e+02 3.287e+02 8.101e+02, threshold=5.822e+02, percent-clipped=1.0 +2023-02-07 08:26:22,284 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7305, 5.8314, 4.9746, 2.5946, 5.0738, 5.5830, 5.2859, 5.3639], + device='cuda:2'), covar=tensor([0.0502, 0.0393, 0.0962, 0.4294, 0.0817, 0.0761, 0.1069, 0.0487], + device='cuda:2'), in_proj_covar=tensor([0.0531, 0.0447, 0.0434, 0.0543, 0.0435, 0.0451, 0.0425, 0.0397], + 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') +2023-02-07 08:26:36,595 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5472, 1.5304, 2.1236, 1.2463, 1.2134, 2.0840, 0.2591, 1.2156], + device='cuda:2'), covar=tensor([0.1629, 0.1147, 0.0345, 0.1094, 0.2457, 0.0437, 0.2017, 0.1270], + device='cuda:2'), in_proj_covar=tensor([0.0195, 0.0201, 0.0131, 0.0220, 0.0272, 0.0140, 0.0171, 0.0197], + 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') +2023-02-07 08:26:37,068 INFO [train.py:901] (2/4) Epoch 25, batch 2650, loss[loss=0.2264, simple_loss=0.318, pruned_loss=0.06735, over 8115.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2848, pruned_loss=0.0593, over 1616478.39 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:27:08,211 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196683.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:27:12,709 INFO [train.py:901] (2/4) Epoch 25, batch 2700, loss[loss=0.2156, simple_loss=0.3029, pruned_loss=0.06415, over 8188.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2837, pruned_loss=0.05861, over 1613358.25 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:27:20,585 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.452e+02 2.909e+02 3.648e+02 8.771e+02, threshold=5.818e+02, percent-clipped=3.0 +2023-02-07 08:27:33,982 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196722.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:27:46,997 INFO [train.py:901] (2/4) Epoch 25, batch 2750, loss[loss=0.1964, simple_loss=0.2908, pruned_loss=0.051, over 8495.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2826, pruned_loss=0.05831, over 1610921.70 frames. ], batch size: 26, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:27:55,334 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3042, 2.4548, 2.9010, 1.6486, 3.1609, 1.9109, 1.5521, 2.2418], + device='cuda:2'), covar=tensor([0.1047, 0.0559, 0.0379, 0.0993, 0.0485, 0.0959, 0.1119, 0.0636], + device='cuda:2'), in_proj_covar=tensor([0.0465, 0.0404, 0.0360, 0.0456, 0.0388, 0.0543, 0.0402, 0.0432], + device='cuda:2'), out_proj_covar=tensor([1.2356e-04, 1.0535e-04, 9.4351e-05, 1.1950e-04, 1.0162e-04, 1.5227e-04, + 1.0762e-04, 1.1362e-04], device='cuda:2') +2023-02-07 08:28:11,176 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.89 vs. limit=5.0 +2023-02-07 08:28:22,165 INFO [train.py:901] (2/4) Epoch 25, batch 2800, loss[loss=0.1802, simple_loss=0.2694, pruned_loss=0.04545, over 8507.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2818, pruned_loss=0.05795, over 1608452.25 frames. ], batch size: 28, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:28:27,859 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196797.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:28:28,544 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196798.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:28:31,148 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.445e+02 2.946e+02 3.604e+02 6.151e+02, threshold=5.892e+02, percent-clipped=2.0 +2023-02-07 08:28:52,420 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-07 08:28:54,961 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:28:56,675 INFO [train.py:901] (2/4) Epoch 25, batch 2850, loss[loss=0.2362, simple_loss=0.3263, pruned_loss=0.07302, over 8467.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2822, pruned_loss=0.05747, over 1613590.18 frames. ], batch size: 25, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:29:19,400 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:29:32,392 INFO [train.py:901] (2/4) Epoch 25, batch 2900, loss[loss=0.1924, simple_loss=0.2886, pruned_loss=0.04814, over 8028.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2812, pruned_loss=0.05676, over 1613387.51 frames. ], batch size: 22, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:29:39,439 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196899.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:29:41,309 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.504e+02 3.053e+02 3.742e+02 6.617e+02, threshold=6.106e+02, percent-clipped=2.0 +2023-02-07 08:30:08,123 INFO [train.py:901] (2/4) Epoch 25, batch 2950, loss[loss=0.2274, simple_loss=0.3125, pruned_loss=0.07109, over 8595.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2831, pruned_loss=0.05779, over 1617399.86 frames. ], batch size: 49, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:30:08,203 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196940.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:30:08,823 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 08:30:19,842 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-07 08:30:41,610 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 08:30:42,478 INFO [train.py:901] (2/4) Epoch 25, batch 3000, loss[loss=0.1939, simple_loss=0.2804, pruned_loss=0.05366, over 8114.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2842, pruned_loss=0.05887, over 1617004.76 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:30:42,478 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 08:30:55,640 INFO [train.py:935] (2/4) Epoch 25, validation: loss=0.1722, simple_loss=0.2721, pruned_loss=0.03618, over 944034.00 frames. +2023-02-07 08:30:55,641 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 08:31:03,958 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.269e+02 2.477e+02 2.955e+02 3.925e+02 7.788e+02, threshold=5.910e+02, percent-clipped=1.0 +2023-02-07 08:31:30,699 INFO [train.py:901] (2/4) Epoch 25, batch 3050, loss[loss=0.1961, simple_loss=0.286, pruned_loss=0.05305, over 8337.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2859, pruned_loss=0.05965, over 1620318.88 frames. ], batch size: 26, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:31:40,531 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197054.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:31:41,168 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197055.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:31:57,855 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197079.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:05,206 INFO [train.py:901] (2/4) Epoch 25, batch 3100, loss[loss=0.234, simple_loss=0.3143, pruned_loss=0.07687, over 8287.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2863, pruned_loss=0.06001, over 1620121.60 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:32:07,488 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197093.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:13,235 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.425e+02 3.089e+02 3.818e+02 7.102e+02, threshold=6.178e+02, percent-clipped=3.0 +2023-02-07 08:32:24,977 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:40,196 INFO [train.py:901] (2/4) Epoch 25, batch 3150, loss[loss=0.1685, simple_loss=0.2636, pruned_loss=0.03674, over 8203.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2865, pruned_loss=0.06007, over 1619544.22 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:32:40,984 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:33:15,302 INFO [train.py:901] (2/4) Epoch 25, batch 3200, loss[loss=0.177, simple_loss=0.2579, pruned_loss=0.04806, over 7937.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2859, pruned_loss=0.06032, over 1617053.19 frames. ], batch size: 20, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:33:23,539 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.400e+02 2.739e+02 3.315e+02 1.024e+03, threshold=5.479e+02, percent-clipped=5.0 +2023-02-07 08:33:33,159 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197216.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:33:50,294 INFO [train.py:901] (2/4) Epoch 25, batch 3250, loss[loss=0.2129, simple_loss=0.2978, pruned_loss=0.064, over 8037.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2855, pruned_loss=0.05979, over 1617471.13 frames. ], batch size: 22, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:33:52,455 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:02,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197256.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:25,382 INFO [train.py:901] (2/4) Epoch 25, batch 3300, loss[loss=0.1594, simple_loss=0.2403, pruned_loss=0.03932, over 7701.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2845, pruned_loss=0.05913, over 1615700.52 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:34:34,250 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.560e+02 3.230e+02 4.212e+02 8.703e+02, threshold=6.460e+02, percent-clipped=10.0 +2023-02-07 08:34:40,503 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:46,284 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-07 08:34:54,230 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197331.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:57,734 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197336.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:35:00,243 INFO [train.py:901] (2/4) Epoch 25, batch 3350, loss[loss=0.1695, simple_loss=0.2552, pruned_loss=0.04193, over 7815.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.0593, over 1617767.22 frames. ], batch size: 20, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:35:13,338 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:35:17,450 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2496, 1.1262, 1.2996, 1.0313, 0.9911, 1.3270, 0.1317, 0.9882], + device='cuda:2'), covar=tensor([0.1388, 0.1212, 0.0494, 0.0660, 0.2549, 0.0503, 0.1906, 0.1116], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0200, 0.0130, 0.0220, 0.0271, 0.0139, 0.0170, 0.0197], + 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') +2023-02-07 08:35:36,171 INFO [train.py:901] (2/4) Epoch 25, batch 3400, loss[loss=0.2039, simple_loss=0.2794, pruned_loss=0.0642, over 7917.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2848, pruned_loss=0.05905, over 1619655.71 frames. ], batch size: 20, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:35:39,769 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4724, 1.4483, 1.8457, 1.1421, 1.1024, 1.8616, 0.1636, 1.1093], + device='cuda:2'), covar=tensor([0.1567, 0.1296, 0.0388, 0.1090, 0.2640, 0.0380, 0.1981, 0.1317], + device='cuda:2'), in_proj_covar=tensor([0.0193, 0.0199, 0.0130, 0.0220, 0.0271, 0.0138, 0.0170, 0.0197], + 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') +2023-02-07 08:35:44,271 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.489e+02 3.044e+02 3.734e+02 7.163e+02, threshold=6.087e+02, percent-clipped=2.0 +2023-02-07 08:36:06,304 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197433.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:36:11,058 INFO [train.py:901] (2/4) Epoch 25, batch 3450, loss[loss=0.2259, simple_loss=0.3011, pruned_loss=0.07537, over 7805.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.285, pruned_loss=0.05942, over 1620789.94 frames. ], batch size: 20, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:36:16,731 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3131, 2.1381, 1.6264, 1.9451, 1.7927, 1.3937, 1.7247, 1.7397], + device='cuda:2'), covar=tensor([0.1274, 0.0406, 0.1225, 0.0516, 0.0656, 0.1547, 0.0929, 0.0877], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0236, 0.0340, 0.0311, 0.0300, 0.0345, 0.0350, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 08:36:46,262 INFO [train.py:901] (2/4) Epoch 25, batch 3500, loss[loss=0.1979, simple_loss=0.2697, pruned_loss=0.06309, over 8204.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2848, pruned_loss=0.05923, over 1621178.63 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:36:54,912 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.387e+02 2.953e+02 3.537e+02 5.869e+02, threshold=5.907e+02, percent-clipped=0.0 +2023-02-07 08:37:02,036 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197512.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:07,262 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 08:37:17,736 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197534.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:19,770 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197537.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:21,600 INFO [train.py:901] (2/4) Epoch 25, batch 3550, loss[loss=0.2027, simple_loss=0.286, pruned_loss=0.05965, over 7236.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.05914, over 1613545.72 frames. ], batch size: 71, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:37:22,461 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8911, 1.8423, 2.4390, 1.5751, 1.4736, 2.4773, 0.4807, 1.4890], + device='cuda:2'), covar=tensor([0.1456, 0.1136, 0.0314, 0.1055, 0.2121, 0.0302, 0.1708, 0.1149], + device='cuda:2'), in_proj_covar=tensor([0.0194, 0.0201, 0.0131, 0.0221, 0.0273, 0.0139, 0.0172, 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') +2023-02-07 08:37:24,413 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9501, 1.6936, 2.0293, 1.8071, 2.0013, 2.0206, 1.8635, 0.8609], + device='cuda:2'), covar=tensor([0.5872, 0.4684, 0.2201, 0.3694, 0.2527, 0.3189, 0.2003, 0.4994], + device='cuda:2'), in_proj_covar=tensor([0.0951, 0.1006, 0.0822, 0.0974, 0.1014, 0.0914, 0.0763, 0.0840], + 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') +2023-02-07 08:37:37,368 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197563.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:52,449 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8013, 1.9026, 1.6977, 2.3214, 1.0348, 1.5818, 1.6941, 1.8523], + device='cuda:2'), covar=tensor([0.0722, 0.0781, 0.0894, 0.0381, 0.1133, 0.1271, 0.0790, 0.0739], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0194, 0.0245, 0.0212, 0.0204, 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') +2023-02-07 08:37:54,506 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197587.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:56,269 INFO [train.py:901] (2/4) Epoch 25, batch 3600, loss[loss=0.2028, simple_loss=0.2849, pruned_loss=0.06041, over 8346.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2838, pruned_loss=0.05892, over 1611404.17 frames. ], batch size: 26, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:38:05,202 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.311e+02 2.881e+02 3.803e+02 6.346e+02, threshold=5.762e+02, percent-clipped=1.0 +2023-02-07 08:38:12,109 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197612.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:38:13,509 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197614.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:38:31,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197639.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:38:31,671 INFO [train.py:901] (2/4) Epoch 25, batch 3650, loss[loss=0.2551, simple_loss=0.3255, pruned_loss=0.0924, over 8583.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2828, pruned_loss=0.05871, over 1612006.55 frames. ], batch size: 31, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:38:40,965 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7426, 1.6564, 2.4220, 1.4951, 1.3139, 2.3869, 0.3916, 1.4523], + device='cuda:2'), covar=tensor([0.1599, 0.1245, 0.0325, 0.1238, 0.2580, 0.0348, 0.2037, 0.1365], + device='cuda:2'), in_proj_covar=tensor([0.0194, 0.0201, 0.0131, 0.0221, 0.0273, 0.0139, 0.0171, 0.0197], + 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') +2023-02-07 08:38:54,159 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7931, 2.3330, 4.0892, 1.6408, 3.0077, 2.3929, 1.7867, 3.1550], + device='cuda:2'), covar=tensor([0.1853, 0.2599, 0.0761, 0.4447, 0.1850, 0.2997, 0.2377, 0.2109], + device='cuda:2'), in_proj_covar=tensor([0.0532, 0.0621, 0.0557, 0.0659, 0.0654, 0.0603, 0.0549, 0.0637], + 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') +2023-02-07 08:39:06,718 INFO [train.py:901] (2/4) Epoch 25, batch 3700, loss[loss=0.1674, simple_loss=0.259, pruned_loss=0.03791, over 8091.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2831, pruned_loss=0.05894, over 1615354.06 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:39:09,551 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 08:39:15,746 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.484e+02 2.942e+02 3.783e+02 7.174e+02, threshold=5.884e+02, percent-clipped=5.0 +2023-02-07 08:39:35,834 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.5442, 1.7206, 3.6878, 1.9910, 3.3470, 3.1511, 3.4235, 3.3422], + device='cuda:2'), covar=tensor([0.0707, 0.3657, 0.0848, 0.3716, 0.1059, 0.0939, 0.0601, 0.0659], + device='cuda:2'), in_proj_covar=tensor([0.0655, 0.0660, 0.0721, 0.0650, 0.0730, 0.0624, 0.0626, 0.0701], + 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') +2023-02-07 08:39:40,720 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 08:39:43,102 INFO [train.py:901] (2/4) Epoch 25, batch 3750, loss[loss=0.1765, simple_loss=0.2562, pruned_loss=0.04844, over 7972.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2837, pruned_loss=0.05933, over 1611305.56 frames. ], batch size: 21, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:40:09,409 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197777.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:40:18,191 INFO [train.py:901] (2/4) Epoch 25, batch 3800, loss[loss=0.2014, simple_loss=0.2858, pruned_loss=0.05853, over 8109.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2854, pruned_loss=0.05991, over 1617169.24 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:40:26,484 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.549e+02 3.044e+02 3.681e+02 9.424e+02, threshold=6.087e+02, percent-clipped=5.0 +2023-02-07 08:40:34,310 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 08:40:43,133 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-07 08:40:53,458 INFO [train.py:901] (2/4) Epoch 25, batch 3850, loss[loss=0.2099, simple_loss=0.2962, pruned_loss=0.06185, over 8632.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2866, pruned_loss=0.06081, over 1616150.81 frames. ], batch size: 49, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:40:57,609 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2707, 2.8397, 2.2799, 3.9113, 1.6145, 1.9330, 2.4206, 2.8496], + device='cuda:2'), covar=tensor([0.0773, 0.0811, 0.0810, 0.0267, 0.1200, 0.1346, 0.0976, 0.0766], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0211, 0.0205, 0.0247, 0.0247, 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') +2023-02-07 08:41:12,902 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 08:41:19,585 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197878.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:41:28,508 INFO [train.py:901] (2/4) Epoch 25, batch 3900, loss[loss=0.1695, simple_loss=0.2546, pruned_loss=0.0422, over 7545.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2869, pruned_loss=0.06055, over 1617252.10 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:41:29,963 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197892.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:41:36,374 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.445e+02 2.982e+02 3.609e+02 8.629e+02, threshold=5.963e+02, percent-clipped=3.0 +2023-02-07 08:41:39,828 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197907.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:41:41,928 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197910.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:42:02,716 INFO [train.py:901] (2/4) Epoch 25, batch 3950, loss[loss=0.2171, simple_loss=0.2855, pruned_loss=0.07438, over 7272.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2871, pruned_loss=0.0607, over 1616127.86 frames. ], batch size: 16, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:42:24,235 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1301, 1.2824, 4.3136, 1.6508, 3.8031, 3.5927, 3.9011, 3.8058], + device='cuda:2'), covar=tensor([0.0635, 0.5040, 0.0602, 0.4104, 0.1127, 0.0936, 0.0631, 0.0707], + device='cuda:2'), in_proj_covar=tensor([0.0648, 0.0653, 0.0714, 0.0642, 0.0723, 0.0618, 0.0621, 0.0692], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 08:42:37,828 INFO [train.py:901] (2/4) Epoch 25, batch 4000, loss[loss=0.2498, simple_loss=0.3167, pruned_loss=0.09145, over 8442.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2871, pruned_loss=0.06071, over 1621655.54 frames. ], batch size: 27, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:42:40,144 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197993.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:42:47,780 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.312e+02 2.768e+02 3.562e+02 7.475e+02, threshold=5.536e+02, percent-clipped=2.0 +2023-02-07 08:43:01,546 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198022.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:43:14,006 INFO [train.py:901] (2/4) Epoch 25, batch 4050, loss[loss=0.1909, simple_loss=0.2875, pruned_loss=0.04718, over 8402.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2865, pruned_loss=0.06023, over 1618261.77 frames. ], batch size: 49, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:43:48,799 INFO [train.py:901] (2/4) Epoch 25, batch 4100, loss[loss=0.2125, simple_loss=0.2941, pruned_loss=0.06546, over 8343.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2869, pruned_loss=0.0605, over 1615817.14 frames. ], batch size: 24, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:43:55,129 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198099.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:43:57,000 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.413e+02 2.876e+02 3.434e+02 5.292e+02, threshold=5.752e+02, percent-clipped=1.0 +2023-02-07 08:44:24,272 INFO [train.py:901] (2/4) Epoch 25, batch 4150, loss[loss=0.2759, simple_loss=0.3423, pruned_loss=0.1047, over 8496.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2871, pruned_loss=0.0605, over 1620051.78 frames. ], batch size: 26, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:44:29,943 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198148.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:44:47,411 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198173.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:44:58,964 INFO [train.py:901] (2/4) Epoch 25, batch 4200, loss[loss=0.2043, simple_loss=0.2915, pruned_loss=0.05853, over 8623.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2857, pruned_loss=0.05985, over 1618028.76 frames. ], batch size: 39, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:45:08,033 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.351e+02 3.091e+02 3.845e+02 7.201e+02, threshold=6.182e+02, percent-clipped=4.0 +2023-02-07 08:45:09,395 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 08:45:33,127 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 08:45:35,178 INFO [train.py:901] (2/4) Epoch 25, batch 4250, loss[loss=0.1837, simple_loss=0.2763, pruned_loss=0.04549, over 8196.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2858, pruned_loss=0.05995, over 1624189.22 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:45:35,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8429, 1.7204, 2.9755, 1.4147, 2.4432, 3.3371, 3.4900, 2.5219], + device='cuda:2'), covar=tensor([0.1489, 0.1900, 0.0542, 0.2541, 0.1316, 0.0373, 0.0667, 0.0922], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0325, 0.0290, 0.0317, 0.0318, 0.0275, 0.0434, 0.0307], + 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') +2023-02-07 08:45:41,587 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198249.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:45:44,710 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198254.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:45:59,339 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198274.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:46:02,109 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198278.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:46:09,935 INFO [train.py:901] (2/4) Epoch 25, batch 4300, loss[loss=0.2028, simple_loss=0.2803, pruned_loss=0.06262, over 7006.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2854, pruned_loss=0.05999, over 1620092.39 frames. ], batch size: 71, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:46:18,872 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.314e+02 2.735e+02 3.533e+02 6.805e+02, threshold=5.471e+02, percent-clipped=1.0 +2023-02-07 08:46:19,824 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198303.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:46:45,627 INFO [train.py:901] (2/4) Epoch 25, batch 4350, loss[loss=0.1901, simple_loss=0.2741, pruned_loss=0.05305, over 8360.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2846, pruned_loss=0.05935, over 1619119.66 frames. ], batch size: 24, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:47:04,271 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 08:47:06,469 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198369.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:47:21,566 INFO [train.py:901] (2/4) Epoch 25, batch 4400, loss[loss=0.2275, simple_loss=0.3079, pruned_loss=0.0736, over 8243.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2848, pruned_loss=0.05931, over 1620877.74 frames. ], batch size: 24, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:47:29,511 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.496e+02 2.496e+02 2.935e+02 3.768e+02 7.665e+02, threshold=5.870e+02, percent-clipped=6.0 +2023-02-07 08:47:45,263 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 08:47:50,270 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 +2023-02-07 08:47:56,500 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 08:47:56,744 INFO [train.py:901] (2/4) Epoch 25, batch 4450, loss[loss=0.2401, simple_loss=0.3222, pruned_loss=0.07898, over 8353.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2846, pruned_loss=0.05876, over 1624052.12 frames. ], batch size: 26, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:47:58,932 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198443.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:48:31,962 INFO [train.py:901] (2/4) Epoch 25, batch 4500, loss[loss=0.1934, simple_loss=0.281, pruned_loss=0.05296, over 8722.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2854, pruned_loss=0.05904, over 1624278.67 frames. ], batch size: 34, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:48:36,294 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6886, 4.7470, 4.1378, 2.0528, 4.1154, 4.3001, 4.2203, 4.0655], + device='cuda:2'), covar=tensor([0.0656, 0.0481, 0.1061, 0.4703, 0.0802, 0.0913, 0.1238, 0.0812], + device='cuda:2'), in_proj_covar=tensor([0.0535, 0.0451, 0.0439, 0.0550, 0.0437, 0.0456, 0.0428, 0.0400], + 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') +2023-02-07 08:48:40,440 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.274e+02 2.771e+02 3.541e+02 5.802e+02, threshold=5.543e+02, percent-clipped=0.0 +2023-02-07 08:48:40,476 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 08:48:51,908 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:49:08,831 INFO [train.py:901] (2/4) Epoch 25, batch 4550, loss[loss=0.2095, simple_loss=0.2814, pruned_loss=0.06882, over 7544.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05889, over 1621163.80 frames. ], batch size: 18, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:49:22,069 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198558.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:49:39,132 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-07 08:49:44,786 INFO [train.py:901] (2/4) Epoch 25, batch 4600, loss[loss=0.1444, simple_loss=0.2228, pruned_loss=0.033, over 7933.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2841, pruned_loss=0.05895, over 1614662.19 frames. ], batch size: 20, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:49:52,976 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.449e+02 2.940e+02 3.432e+02 8.422e+02, threshold=5.881e+02, percent-clipped=6.0 +2023-02-07 08:50:09,328 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198625.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:50:14,653 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198633.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:50:19,157 INFO [train.py:901] (2/4) Epoch 25, batch 4650, loss[loss=0.1767, simple_loss=0.2578, pruned_loss=0.04783, over 7240.00 frames. ], tot_loss[loss=0.201, simple_loss=0.284, pruned_loss=0.05901, over 1612104.51 frames. ], batch size: 16, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:50:26,823 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198650.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:50:54,512 INFO [train.py:901] (2/4) Epoch 25, batch 4700, loss[loss=0.2011, simple_loss=0.2884, pruned_loss=0.05686, over 8444.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2839, pruned_loss=0.05924, over 1607032.17 frames. ], batch size: 27, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:51:03,370 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.164e+02 2.735e+02 3.323e+02 7.623e+02, threshold=5.470e+02, percent-clipped=2.0 +2023-02-07 08:51:29,738 INFO [train.py:901] (2/4) Epoch 25, batch 4750, loss[loss=0.2667, simple_loss=0.3409, pruned_loss=0.09627, over 8498.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.284, pruned_loss=0.05921, over 1609089.82 frames. ], batch size: 28, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:51:38,030 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3883, 1.4796, 4.6009, 1.7748, 4.0786, 3.7873, 4.1550, 4.0195], + device='cuda:2'), covar=tensor([0.0564, 0.4709, 0.0489, 0.3992, 0.1110, 0.0915, 0.0543, 0.0675], + device='cuda:2'), in_proj_covar=tensor([0.0646, 0.0650, 0.0709, 0.0640, 0.0723, 0.0616, 0.0617, 0.0690], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 08:51:39,045 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 08:51:42,013 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 08:51:45,373 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 08:52:05,184 INFO [train.py:901] (2/4) Epoch 25, batch 4800, loss[loss=0.2541, simple_loss=0.3304, pruned_loss=0.08893, over 8297.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2837, pruned_loss=0.05889, over 1610371.90 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:52:13,388 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.326e+02 2.392e+02 2.917e+02 3.409e+02 6.169e+02, threshold=5.835e+02, percent-clipped=3.0 +2023-02-07 08:52:22,063 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:52:36,112 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 08:52:39,635 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198839.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:52:40,102 INFO [train.py:901] (2/4) Epoch 25, batch 4850, loss[loss=0.1626, simple_loss=0.2497, pruned_loss=0.03774, over 7539.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2848, pruned_loss=0.05946, over 1612521.58 frames. ], batch size: 18, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:52:55,414 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198861.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:53:01,725 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198870.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:53:15,869 INFO [train.py:901] (2/4) Epoch 25, batch 4900, loss[loss=0.2158, simple_loss=0.3008, pruned_loss=0.06541, over 8129.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.285, pruned_loss=0.05967, over 1616326.91 frames. ], batch size: 22, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:53:24,155 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198901.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:53:24,666 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.376e+02 2.954e+02 3.660e+02 6.336e+02, threshold=5.908e+02, percent-clipped=3.0 +2023-02-07 08:53:41,107 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 08:53:43,155 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 08:53:50,028 INFO [train.py:901] (2/4) Epoch 25, batch 4950, loss[loss=0.248, simple_loss=0.3295, pruned_loss=0.08328, over 8353.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2845, pruned_loss=0.0594, over 1616228.45 frames. ], batch size: 24, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:53:54,444 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198945.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:54:15,961 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:54:16,532 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198977.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:54:24,177 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1884, 2.0107, 2.7058, 2.2330, 2.6890, 2.3089, 2.0734, 1.6260], + device='cuda:2'), covar=tensor([0.5920, 0.5249, 0.2081, 0.3997, 0.2694, 0.3311, 0.2147, 0.5387], + device='cuda:2'), in_proj_covar=tensor([0.0949, 0.1002, 0.0817, 0.0969, 0.1011, 0.0914, 0.0758, 0.0835], + 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') +2023-02-07 08:54:25,247 INFO [train.py:901] (2/4) Epoch 25, batch 5000, loss[loss=0.2019, simple_loss=0.2929, pruned_loss=0.05544, over 8352.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2845, pruned_loss=0.05917, over 1615122.59 frames. ], batch size: 24, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:54:33,925 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.471e+02 2.360e+02 2.883e+02 3.509e+02 6.136e+02, threshold=5.766e+02, percent-clipped=1.0 +2023-02-07 08:54:38,389 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.71 vs. limit=5.0 +2023-02-07 08:54:59,851 INFO [train.py:901] (2/4) Epoch 25, batch 5050, loss[loss=0.2738, simple_loss=0.3361, pruned_loss=0.1057, over 8530.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2852, pruned_loss=0.0595, over 1617051.42 frames. ], batch size: 28, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:55:14,361 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 08:55:35,763 INFO [train.py:901] (2/4) Epoch 25, batch 5100, loss[loss=0.2236, simple_loss=0.3151, pruned_loss=0.06607, over 8026.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2849, pruned_loss=0.05927, over 1619352.06 frames. ], batch size: 22, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:55:37,406 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199092.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:55:44,130 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.470e+02 3.005e+02 3.768e+02 7.063e+02, threshold=6.010e+02, percent-clipped=5.0 +2023-02-07 08:56:11,856 INFO [train.py:901] (2/4) Epoch 25, batch 5150, loss[loss=0.1791, simple_loss=0.2689, pruned_loss=0.04459, over 8189.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2852, pruned_loss=0.05935, over 1627150.03 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:56:12,464 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 08:56:47,046 INFO [train.py:901] (2/4) Epoch 25, batch 5200, loss[loss=0.1867, simple_loss=0.2562, pruned_loss=0.05863, over 7223.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.285, pruned_loss=0.05939, over 1621563.47 frames. ], batch size: 16, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:56:49,913 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199194.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:56:55,026 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.381e+02 2.894e+02 3.514e+02 1.206e+03, threshold=5.788e+02, percent-clipped=6.0 +2023-02-07 08:57:03,628 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 08:57:04,083 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:12,756 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 08:57:17,162 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199232.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:19,825 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6633, 1.4879, 1.6669, 1.3349, 0.8850, 1.4317, 1.4703, 1.4059], + device='cuda:2'), covar=tensor([0.0597, 0.1236, 0.1675, 0.1501, 0.0593, 0.1491, 0.0719, 0.0687], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0100, 0.0164, 0.0113, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 08:57:22,313 INFO [train.py:901] (2/4) Epoch 25, batch 5250, loss[loss=0.1905, simple_loss=0.276, pruned_loss=0.05254, over 6403.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2856, pruned_loss=0.05907, over 1623615.12 frames. ], batch size: 14, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:57:25,798 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199245.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:34,838 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199257.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:41,279 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1817, 4.0963, 3.7054, 2.0544, 3.7216, 3.8132, 3.6825, 3.6965], + device='cuda:2'), covar=tensor([0.0779, 0.0587, 0.1058, 0.4289, 0.0908, 0.0962, 0.1327, 0.0752], + device='cuda:2'), in_proj_covar=tensor([0.0542, 0.0456, 0.0442, 0.0555, 0.0439, 0.0460, 0.0433, 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') +2023-02-07 08:57:44,762 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7325, 1.6564, 2.2581, 1.4877, 1.3126, 2.2170, 0.4373, 1.3824], + device='cuda:2'), covar=tensor([0.1563, 0.1088, 0.0350, 0.1037, 0.2310, 0.0405, 0.1823, 0.1341], + device='cuda:2'), in_proj_covar=tensor([0.0196, 0.0202, 0.0132, 0.0222, 0.0277, 0.0142, 0.0173, 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') +2023-02-07 08:57:56,836 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199289.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:57,413 INFO [train.py:901] (2/4) Epoch 25, batch 5300, loss[loss=0.1842, simple_loss=0.2779, pruned_loss=0.04521, over 8014.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2855, pruned_loss=0.05889, over 1621253.52 frames. ], batch size: 22, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:58:05,707 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.313e+02 2.718e+02 3.488e+02 6.386e+02, threshold=5.437e+02, percent-clipped=3.0 +2023-02-07 08:58:25,230 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199329.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:58:32,735 INFO [train.py:901] (2/4) Epoch 25, batch 5350, loss[loss=0.1926, simple_loss=0.291, pruned_loss=0.04707, over 8290.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2854, pruned_loss=0.05911, over 1614560.89 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:58:38,543 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199348.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:58:47,625 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199360.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:58:57,357 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:59:08,730 INFO [train.py:901] (2/4) Epoch 25, batch 5400, loss[loss=0.1958, simple_loss=0.2734, pruned_loss=0.05906, over 8091.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.05872, over 1608154.46 frames. ], batch size: 21, lr: 3.01e-03, grad_scale: 32.0 +2023-02-07 08:59:18,149 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.292e+02 2.858e+02 3.757e+02 5.815e+02, threshold=5.716e+02, percent-clipped=3.0 +2023-02-07 08:59:18,353 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199404.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:59:22,498 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8596, 1.8722, 2.9271, 2.1920, 2.6935, 1.9669, 1.6935, 1.4124], + device='cuda:2'), covar=tensor([0.7374, 0.6152, 0.2163, 0.4363, 0.3219, 0.4454, 0.3041, 0.5954], + device='cuda:2'), in_proj_covar=tensor([0.0951, 0.1002, 0.0818, 0.0972, 0.1014, 0.0916, 0.0760, 0.0838], + 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') +2023-02-07 08:59:28,367 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199418.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:59:43,202 INFO [train.py:901] (2/4) Epoch 25, batch 5450, loss[loss=0.1701, simple_loss=0.2471, pruned_loss=0.04658, over 8078.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2834, pruned_loss=0.05862, over 1602524.95 frames. ], batch size: 21, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:00:08,085 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 09:00:08,226 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199476.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:00:17,983 INFO [train.py:901] (2/4) Epoch 25, batch 5500, loss[loss=0.1985, simple_loss=0.2763, pruned_loss=0.06035, over 7645.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2828, pruned_loss=0.05852, over 1602625.01 frames. ], batch size: 19, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:00:28,263 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.278e+02 2.767e+02 3.622e+02 8.817e+02, threshold=5.534e+02, percent-clipped=3.0 +2023-02-07 09:00:33,827 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199512.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:00:52,118 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199538.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:00:53,325 INFO [train.py:901] (2/4) Epoch 25, batch 5550, loss[loss=0.1991, simple_loss=0.2766, pruned_loss=0.0608, over 7770.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2829, pruned_loss=0.05861, over 1605295.76 frames. ], batch size: 19, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:01:02,093 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199553.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:01:13,072 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.83 vs. limit=5.0 +2023-02-07 09:01:24,423 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199585.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:01:27,624 INFO [train.py:901] (2/4) Epoch 25, batch 5600, loss[loss=0.1999, simple_loss=0.2931, pruned_loss=0.05334, over 8504.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2819, pruned_loss=0.0583, over 1602826.22 frames. ], batch size: 28, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:01:38,057 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.538e+02 3.116e+02 4.016e+02 1.228e+03, threshold=6.232e+02, percent-clipped=11.0 +2023-02-07 09:01:43,186 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199610.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:01:47,370 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:03,338 INFO [train.py:901] (2/4) Epoch 25, batch 5650, loss[loss=0.209, simple_loss=0.2902, pruned_loss=0.06393, over 8253.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2826, pruned_loss=0.05845, over 1606331.70 frames. ], batch size: 22, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:02:04,213 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199641.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:13,533 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:13,999 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 09:02:18,327 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199660.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:36,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199685.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:39,918 INFO [train.py:901] (2/4) Epoch 25, batch 5700, loss[loss=0.211, simple_loss=0.2968, pruned_loss=0.06262, over 8496.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2837, pruned_loss=0.05915, over 1605152.82 frames. ], batch size: 31, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:02:49,766 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.200e+02 2.647e+02 3.419e+02 7.306e+02, threshold=5.294e+02, percent-clipped=3.0 +2023-02-07 09:03:12,405 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 09:03:16,059 INFO [train.py:901] (2/4) Epoch 25, batch 5750, loss[loss=0.2159, simple_loss=0.301, pruned_loss=0.06536, over 8327.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2824, pruned_loss=0.05879, over 1597616.80 frames. ], batch size: 25, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:03:16,267 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1739, 1.9960, 1.4843, 1.9104, 1.5918, 1.2677, 1.5158, 1.6860], + device='cuda:2'), covar=tensor([0.1495, 0.0526, 0.1472, 0.0628, 0.0980, 0.1945, 0.1246, 0.0945], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0235, 0.0339, 0.0310, 0.0301, 0.0341, 0.0346, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 09:03:21,550 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 09:03:30,824 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199762.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:03:39,208 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6974, 1.3979, 4.9167, 1.8176, 4.3277, 3.9880, 4.4051, 4.2470], + device='cuda:2'), covar=tensor([0.0581, 0.4969, 0.0431, 0.4218, 0.1075, 0.0891, 0.0567, 0.0687], + device='cuda:2'), in_proj_covar=tensor([0.0656, 0.0657, 0.0724, 0.0647, 0.0731, 0.0621, 0.0622, 0.0699], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 09:03:50,501 INFO [train.py:901] (2/4) Epoch 25, batch 5800, loss[loss=0.2356, simple_loss=0.3176, pruned_loss=0.07674, over 8458.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2832, pruned_loss=0.05908, over 1601684.39 frames. ], batch size: 29, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:04:00,801 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.348e+02 2.869e+02 3.742e+02 6.332e+02, threshold=5.738e+02, percent-clipped=6.0 +2023-02-07 09:04:11,740 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199820.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:04:12,724 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 09:04:26,575 INFO [train.py:901] (2/4) Epoch 25, batch 5850, loss[loss=0.2547, simple_loss=0.3182, pruned_loss=0.09557, over 7092.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2837, pruned_loss=0.05891, over 1611329.90 frames. ], batch size: 71, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:04:37,359 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199856.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:04:51,702 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199877.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:01,006 INFO [train.py:901] (2/4) Epoch 25, batch 5900, loss[loss=0.2336, simple_loss=0.3094, pruned_loss=0.07893, over 8252.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2838, pruned_loss=0.05903, over 1610069.77 frames. ], batch size: 24, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:05:05,824 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199897.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:05:10,363 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.352e+02 2.828e+02 3.481e+02 7.421e+02, threshold=5.657e+02, percent-clipped=3.0 +2023-02-07 09:05:13,998 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:26,462 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199927.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:31,352 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199934.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:32,025 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199935.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:35,262 INFO [train.py:901] (2/4) Epoch 25, batch 5950, loss[loss=0.2137, simple_loss=0.2955, pruned_loss=0.06593, over 8630.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2843, pruned_loss=0.05892, over 1613909.11 frames. ], batch size: 34, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:05:54,665 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5278, 1.4438, 4.7856, 1.8949, 4.2309, 3.9580, 4.3271, 4.1768], + device='cuda:2'), covar=tensor([0.0648, 0.4885, 0.0460, 0.4157, 0.1146, 0.0921, 0.0577, 0.0696], + device='cuda:2'), in_proj_covar=tensor([0.0656, 0.0656, 0.0723, 0.0647, 0.0731, 0.0620, 0.0622, 0.0698], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 09:05:58,231 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199971.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:06:08,486 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4682, 2.2216, 2.7616, 2.4152, 2.7980, 2.4302, 2.3650, 2.0691], + device='cuda:2'), covar=tensor([0.4218, 0.4162, 0.1924, 0.3221, 0.1990, 0.2854, 0.1646, 0.4134], + device='cuda:2'), in_proj_covar=tensor([0.0955, 0.1004, 0.0822, 0.0976, 0.1017, 0.0917, 0.0763, 0.0839], + 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') +2023-02-07 09:06:11,076 INFO [train.py:901] (2/4) Epoch 25, batch 6000, loss[loss=0.2225, simple_loss=0.3035, pruned_loss=0.07073, over 8331.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.284, pruned_loss=0.05923, over 1608394.72 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:06:11,077 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 09:06:23,701 INFO [train.py:935] (2/4) Epoch 25, validation: loss=0.1725, simple_loss=0.2721, pruned_loss=0.03643, over 944034.00 frames. +2023-02-07 09:06:23,702 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 09:06:34,577 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.373e+02 2.952e+02 3.581e+02 7.260e+02, threshold=5.903e+02, percent-clipped=4.0 +2023-02-07 09:06:40,184 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200012.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:06:59,672 INFO [train.py:901] (2/4) Epoch 25, batch 6050, loss[loss=0.1681, simple_loss=0.2439, pruned_loss=0.04614, over 7551.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2843, pruned_loss=0.0591, over 1613985.16 frames. ], batch size: 18, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:07:35,027 INFO [train.py:901] (2/4) Epoch 25, batch 6100, loss[loss=0.1923, simple_loss=0.2844, pruned_loss=0.05006, over 8329.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2841, pruned_loss=0.05901, over 1612890.84 frames. ], batch size: 25, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:07:36,616 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7942, 1.3291, 3.9907, 1.5503, 3.4713, 3.3332, 3.5917, 3.4662], + device='cuda:2'), covar=tensor([0.0730, 0.4773, 0.0642, 0.4108, 0.1341, 0.0970, 0.0661, 0.0826], + device='cuda:2'), in_proj_covar=tensor([0.0657, 0.0656, 0.0724, 0.0647, 0.0732, 0.0621, 0.0624, 0.0699], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 09:07:38,692 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4034, 1.4483, 1.4037, 1.8294, 0.8236, 1.2840, 1.3790, 1.4994], + device='cuda:2'), covar=tensor([0.0860, 0.0766, 0.1006, 0.0483, 0.1076, 0.1444, 0.0727, 0.0812], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0194, 0.0245, 0.0211, 0.0205, 0.0247, 0.0249, 0.0207], + 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') +2023-02-07 09:07:45,358 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.345e+02 2.959e+02 3.596e+02 7.197e+02, threshold=5.919e+02, percent-clipped=3.0 +2023-02-07 09:07:54,345 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 09:08:05,859 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200133.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:08:11,249 INFO [train.py:901] (2/4) Epoch 25, batch 6150, loss[loss=0.195, simple_loss=0.2768, pruned_loss=0.05664, over 8490.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2826, pruned_loss=0.05818, over 1610328.98 frames. ], batch size: 28, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:08:22,234 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 09:08:23,404 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200158.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:08:27,288 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-07 09:08:46,113 INFO [train.py:901] (2/4) Epoch 25, batch 6200, loss[loss=0.1588, simple_loss=0.2528, pruned_loss=0.03243, over 8235.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.283, pruned_loss=0.05862, over 1610175.69 frames. ], batch size: 22, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:08:46,772 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-07 09:08:47,061 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200191.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:08:49,547 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200195.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:08:55,699 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.314e+02 2.821e+02 3.535e+02 6.331e+02, threshold=5.643e+02, percent-clipped=2.0 +2023-02-07 09:09:05,007 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200216.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:09:08,358 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5883, 1.5689, 1.9440, 1.2863, 1.2676, 1.9380, 0.4643, 1.3580], + device='cuda:2'), covar=tensor([0.1577, 0.1008, 0.0402, 0.0937, 0.2219, 0.0487, 0.1899, 0.1160], + device='cuda:2'), in_proj_covar=tensor([0.0195, 0.0200, 0.0131, 0.0219, 0.0274, 0.0141, 0.0172, 0.0197], + 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') +2023-02-07 09:09:13,103 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200227.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:09:21,663 INFO [train.py:901] (2/4) Epoch 25, batch 6250, loss[loss=0.1741, simple_loss=0.2706, pruned_loss=0.03874, over 8314.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2836, pruned_loss=0.05919, over 1610162.26 frames. ], batch size: 25, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:09:29,786 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200252.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:09:41,365 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200268.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:09:43,283 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200271.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:09:56,028 INFO [train.py:901] (2/4) Epoch 25, batch 6300, loss[loss=0.1558, simple_loss=0.2356, pruned_loss=0.03799, over 7656.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2824, pruned_loss=0.05848, over 1610813.00 frames. ], batch size: 19, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:09:58,153 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200293.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:10:00,758 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200297.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:10:06,125 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.537e+02 3.046e+02 4.211e+02 7.306e+02, threshold=6.092e+02, percent-clipped=6.0 +2023-02-07 09:10:31,233 INFO [train.py:901] (2/4) Epoch 25, batch 6350, loss[loss=0.1879, simple_loss=0.2735, pruned_loss=0.05113, over 8609.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2831, pruned_loss=0.059, over 1613724.82 frames. ], batch size: 34, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:11:03,766 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200386.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:11:06,376 INFO [train.py:901] (2/4) Epoch 25, batch 6400, loss[loss=0.2177, simple_loss=0.2964, pruned_loss=0.06949, over 7814.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2839, pruned_loss=0.05972, over 1614132.57 frames. ], batch size: 20, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:11:15,859 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.247e+02 2.600e+02 3.696e+02 8.014e+02, threshold=5.200e+02, percent-clipped=2.0 +2023-02-07 09:11:40,859 INFO [train.py:901] (2/4) Epoch 25, batch 6450, loss[loss=0.2083, simple_loss=0.2957, pruned_loss=0.06044, over 8194.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2836, pruned_loss=0.05911, over 1617131.37 frames. ], batch size: 23, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:12:16,064 INFO [train.py:901] (2/4) Epoch 25, batch 6500, loss[loss=0.2057, simple_loss=0.2931, pruned_loss=0.05915, over 8514.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2837, pruned_loss=0.05899, over 1617346.30 frames. ], batch size: 28, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:12:26,035 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.181e+02 2.613e+02 3.190e+02 4.719e+02, threshold=5.226e+02, percent-clipped=0.0 +2023-02-07 09:12:49,388 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200539.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:12:49,933 INFO [train.py:901] (2/4) Epoch 25, batch 6550, loss[loss=0.2289, simple_loss=0.3018, pruned_loss=0.07796, over 8236.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2839, pruned_loss=0.05912, over 1619889.37 frames. ], batch size: 22, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:13:09,849 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 09:13:16,316 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-07 09:13:26,026 INFO [train.py:901] (2/4) Epoch 25, batch 6600, loss[loss=0.198, simple_loss=0.2711, pruned_loss=0.0625, over 7805.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2831, pruned_loss=0.05901, over 1616979.04 frames. ], batch size: 20, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:13:30,833 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 09:13:35,559 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.321e+02 2.722e+02 3.541e+02 8.507e+02, threshold=5.445e+02, percent-clipped=6.0 +2023-02-07 09:13:43,069 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5336, 1.8360, 4.7197, 2.2284, 4.2546, 3.9442, 4.3416, 4.2280], + device='cuda:2'), covar=tensor([0.0592, 0.4131, 0.0517, 0.3830, 0.1036, 0.0868, 0.0537, 0.0575], + device='cuda:2'), in_proj_covar=tensor([0.0655, 0.0650, 0.0718, 0.0644, 0.0730, 0.0619, 0.0621, 0.0694], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 09:14:00,777 INFO [train.py:901] (2/4) Epoch 25, batch 6650, loss[loss=0.1674, simple_loss=0.2494, pruned_loss=0.04268, over 7798.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2826, pruned_loss=0.0586, over 1616528.37 frames. ], batch size: 19, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:14:01,607 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200641.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:02,401 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200642.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:10,432 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200654.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:14:16,369 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200663.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:19,883 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200667.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:35,339 INFO [train.py:901] (2/4) Epoch 25, batch 6700, loss[loss=0.2295, simple_loss=0.3071, pruned_loss=0.0759, over 8336.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2829, pruned_loss=0.05856, over 1616004.31 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:14:45,630 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.443e+02 2.859e+02 3.397e+02 5.440e+02, threshold=5.717e+02, percent-clipped=0.0 +2023-02-07 09:14:48,885 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 09:15:10,909 INFO [train.py:901] (2/4) Epoch 25, batch 6750, loss[loss=0.1919, simple_loss=0.2581, pruned_loss=0.06288, over 7453.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2818, pruned_loss=0.0579, over 1608682.82 frames. ], batch size: 17, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:15:22,782 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200756.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:15:47,008 INFO [train.py:901] (2/4) Epoch 25, batch 6800, loss[loss=0.2257, simple_loss=0.3066, pruned_loss=0.07234, over 8621.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2815, pruned_loss=0.05784, over 1607519.00 frames. ], batch size: 34, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:15:51,859 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 09:15:56,788 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.322e+02 2.853e+02 3.502e+02 6.162e+02, threshold=5.706e+02, percent-clipped=1.0 +2023-02-07 09:16:21,871 INFO [train.py:901] (2/4) Epoch 25, batch 6850, loss[loss=0.2026, simple_loss=0.2957, pruned_loss=0.05479, over 8455.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2814, pruned_loss=0.05777, over 1602483.73 frames. ], batch size: 49, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:16:29,578 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7847, 1.5806, 3.9737, 1.4083, 3.4744, 3.2592, 3.5900, 3.4951], + device='cuda:2'), covar=tensor([0.0756, 0.4193, 0.0700, 0.4581, 0.1372, 0.1114, 0.0694, 0.0802], + device='cuda:2'), in_proj_covar=tensor([0.0663, 0.0659, 0.0729, 0.0655, 0.0739, 0.0629, 0.0629, 0.0703], + 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') +2023-02-07 09:16:39,682 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9138, 3.8157, 3.5316, 1.7804, 3.5098, 3.5575, 3.3775, 3.4326], + device='cuda:2'), covar=tensor([0.0859, 0.0723, 0.1217, 0.4284, 0.0936, 0.0993, 0.1604, 0.0693], + device='cuda:2'), in_proj_covar=tensor([0.0535, 0.0453, 0.0440, 0.0552, 0.0438, 0.0457, 0.0433, 0.0401], + 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') +2023-02-07 09:16:40,950 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 09:16:56,596 INFO [train.py:901] (2/4) Epoch 25, batch 6900, loss[loss=0.1907, simple_loss=0.2541, pruned_loss=0.06362, over 7430.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.05809, over 1605300.51 frames. ], batch size: 17, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:17:06,815 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.244e+02 2.770e+02 3.533e+02 6.127e+02, threshold=5.541e+02, percent-clipped=2.0 +2023-02-07 09:17:11,230 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200910.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:17:28,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200935.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:17:31,363 INFO [train.py:901] (2/4) Epoch 25, batch 6950, loss[loss=0.208, simple_loss=0.3004, pruned_loss=0.05777, over 7816.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2834, pruned_loss=0.05836, over 1608974.23 frames. ], batch size: 20, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:17:50,964 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 09:18:01,551 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8437, 1.4410, 4.2008, 1.8039, 3.3255, 3.2916, 3.7928, 3.7496], + device='cuda:2'), covar=tensor([0.1356, 0.6567, 0.1165, 0.5137, 0.2521, 0.1774, 0.1144, 0.1081], + device='cuda:2'), in_proj_covar=tensor([0.0659, 0.0655, 0.0726, 0.0650, 0.0736, 0.0624, 0.0626, 0.0699], + 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') +2023-02-07 09:18:07,787 INFO [train.py:901] (2/4) Epoch 25, batch 7000, loss[loss=0.1987, simple_loss=0.29, pruned_loss=0.05367, over 8549.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2841, pruned_loss=0.05866, over 1610177.68 frames. ], batch size: 49, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:18:17,572 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.593e+02 3.026e+02 3.851e+02 8.547e+02, threshold=6.052e+02, percent-clipped=7.0 +2023-02-07 09:18:19,773 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201007.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:18:23,197 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201012.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:18:37,955 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 09:18:40,487 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201037.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:18:42,298 INFO [train.py:901] (2/4) Epoch 25, batch 7050, loss[loss=0.167, simple_loss=0.2473, pruned_loss=0.04333, over 8232.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2833, pruned_loss=0.05766, over 1612009.72 frames. ], batch size: 22, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:19:16,807 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8448, 3.7106, 3.4855, 1.8541, 3.3726, 3.4708, 3.3304, 3.3154], + device='cuda:2'), covar=tensor([0.0940, 0.0735, 0.1146, 0.4236, 0.1041, 0.1178, 0.1586, 0.0918], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0455, 0.0440, 0.0554, 0.0441, 0.0461, 0.0434, 0.0401], + 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') +2023-02-07 09:19:16,852 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201089.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:19:17,388 INFO [train.py:901] (2/4) Epoch 25, batch 7100, loss[loss=0.2202, simple_loss=0.2939, pruned_loss=0.07326, over 8332.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.282, pruned_loss=0.05743, over 1608150.21 frames. ], batch size: 25, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:19:26,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.246e+02 2.728e+02 3.277e+02 5.322e+02, threshold=5.456e+02, percent-clipped=0.0 +2023-02-07 09:19:39,432 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2705, 1.9995, 2.6209, 2.1996, 2.6170, 2.3468, 2.1470, 1.4833], + device='cuda:2'), covar=tensor([0.5425, 0.5109, 0.2108, 0.3956, 0.2611, 0.3128, 0.1991, 0.5577], + device='cuda:2'), in_proj_covar=tensor([0.0953, 0.1005, 0.0824, 0.0978, 0.1017, 0.0915, 0.0764, 0.0841], + 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') +2023-02-07 09:19:40,023 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201122.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:19:52,241 INFO [train.py:901] (2/4) Epoch 25, batch 7150, loss[loss=0.1976, simple_loss=0.2891, pruned_loss=0.05304, over 8355.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2815, pruned_loss=0.05717, over 1610225.94 frames. ], batch size: 24, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:20:14,505 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1663, 3.7759, 2.4360, 2.9883, 2.8966, 2.1124, 2.7393, 3.1314], + device='cuda:2'), covar=tensor([0.1683, 0.0361, 0.1144, 0.0715, 0.0724, 0.1458, 0.1194, 0.1181], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0239, 0.0346, 0.0316, 0.0304, 0.0349, 0.0352, 0.0324], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 09:20:28,435 INFO [train.py:901] (2/4) Epoch 25, batch 7200, loss[loss=0.1806, simple_loss=0.2563, pruned_loss=0.05239, over 7444.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2815, pruned_loss=0.05762, over 1608279.24 frames. ], batch size: 17, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:20:34,964 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 09:20:38,233 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.475e+02 3.123e+02 4.294e+02 9.608e+02, threshold=6.246e+02, percent-clipped=8.0 +2023-02-07 09:21:03,474 INFO [train.py:901] (2/4) Epoch 25, batch 7250, loss[loss=0.2211, simple_loss=0.3049, pruned_loss=0.06866, over 8196.00 frames. ], tot_loss[loss=0.199, simple_loss=0.282, pruned_loss=0.05794, over 1606543.03 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:21:20,244 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2023-02-07 09:21:25,204 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201271.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:21:31,945 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8733, 3.8008, 3.4926, 1.9192, 3.4182, 3.5904, 3.4583, 3.3096], + device='cuda:2'), covar=tensor([0.0863, 0.0630, 0.1194, 0.4508, 0.1036, 0.1044, 0.1417, 0.0899], + device='cuda:2'), in_proj_covar=tensor([0.0534, 0.0451, 0.0439, 0.0549, 0.0437, 0.0456, 0.0432, 0.0398], + 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') +2023-02-07 09:21:37,952 INFO [train.py:901] (2/4) Epoch 25, batch 7300, loss[loss=0.1579, simple_loss=0.2403, pruned_loss=0.03779, over 7718.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2822, pruned_loss=0.05833, over 1608965.59 frames. ], batch size: 18, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:21:39,387 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201292.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:21:48,706 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.341e+02 2.809e+02 3.464e+02 9.506e+02, threshold=5.617e+02, percent-clipped=4.0 +2023-02-07 09:22:13,162 INFO [train.py:901] (2/4) Epoch 25, batch 7350, loss[loss=0.2005, simple_loss=0.2924, pruned_loss=0.05432, over 8497.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2822, pruned_loss=0.05826, over 1614662.69 frames. ], batch size: 28, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:22:39,015 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 09:22:40,537 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:22:48,513 INFO [train.py:901] (2/4) Epoch 25, batch 7400, loss[loss=0.1987, simple_loss=0.2872, pruned_loss=0.05504, over 8328.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2839, pruned_loss=0.05871, over 1618239.44 frames. ], batch size: 25, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:22:57,503 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201403.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:22:57,969 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.231e+02 2.880e+02 3.857e+02 7.685e+02, threshold=5.759e+02, percent-clipped=5.0 +2023-02-07 09:22:58,716 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 09:23:19,305 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201433.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:23:23,998 INFO [train.py:901] (2/4) Epoch 25, batch 7450, loss[loss=0.1666, simple_loss=0.2487, pruned_loss=0.04221, over 7542.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2834, pruned_loss=0.05855, over 1615899.03 frames. ], batch size: 18, lr: 2.99e-03, grad_scale: 16.0 +2023-02-07 09:23:37,753 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 09:23:59,788 INFO [train.py:901] (2/4) Epoch 25, batch 7500, loss[loss=0.2085, simple_loss=0.2865, pruned_loss=0.06528, over 8130.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2835, pruned_loss=0.05839, over 1615504.01 frames. ], batch size: 22, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:24:09,797 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.281e+02 2.758e+02 3.564e+02 6.593e+02, threshold=5.515e+02, percent-clipped=6.0 +2023-02-07 09:24:34,699 INFO [train.py:901] (2/4) Epoch 25, batch 7550, loss[loss=0.2147, simple_loss=0.2979, pruned_loss=0.06577, over 8493.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2847, pruned_loss=0.05932, over 1613121.59 frames. ], batch size: 49, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:24:40,271 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:25:09,407 INFO [train.py:901] (2/4) Epoch 25, batch 7600, loss[loss=0.2139, simple_loss=0.2937, pruned_loss=0.06711, over 8191.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2857, pruned_loss=0.06004, over 1616295.38 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:25:20,530 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.467e+02 2.939e+02 3.909e+02 7.265e+02, threshold=5.878e+02, percent-clipped=5.0 +2023-02-07 09:25:27,306 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201615.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:25:41,304 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201636.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:25:43,931 INFO [train.py:901] (2/4) Epoch 25, batch 7650, loss[loss=0.2204, simple_loss=0.3027, pruned_loss=0.06904, over 8102.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2842, pruned_loss=0.05935, over 1613560.41 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:26:00,389 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:12,136 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201679.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:19,459 INFO [train.py:901] (2/4) Epoch 25, batch 7700, loss[loss=0.1643, simple_loss=0.2428, pruned_loss=0.04297, over 7541.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.284, pruned_loss=0.05914, over 1613575.97 frames. ], batch size: 18, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:26:30,382 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.306e+02 2.805e+02 3.732e+02 7.115e+02, threshold=5.609e+02, percent-clipped=1.0 +2023-02-07 09:26:43,956 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201724.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:48,035 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201730.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:49,269 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 09:26:54,623 INFO [train.py:901] (2/4) Epoch 25, batch 7750, loss[loss=0.1948, simple_loss=0.2634, pruned_loss=0.06305, over 7438.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2847, pruned_loss=0.05971, over 1613712.81 frames. ], batch size: 17, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:27:02,275 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201751.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:27:21,319 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.8030, 5.8700, 5.1521, 2.8460, 5.2294, 5.6473, 5.4755, 5.4964], + device='cuda:2'), covar=tensor([0.0568, 0.0377, 0.0876, 0.3760, 0.0704, 0.0684, 0.1027, 0.0528], + device='cuda:2'), in_proj_covar=tensor([0.0533, 0.0452, 0.0439, 0.0550, 0.0436, 0.0457, 0.0431, 0.0398], + 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') +2023-02-07 09:27:29,979 INFO [train.py:901] (2/4) Epoch 25, batch 7800, loss[loss=0.1719, simple_loss=0.2462, pruned_loss=0.04884, over 7668.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2849, pruned_loss=0.05927, over 1617098.19 frames. ], batch size: 19, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:27:40,028 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201804.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:27:40,470 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.301e+02 2.955e+02 3.831e+02 1.047e+03, threshold=5.910e+02, percent-clipped=5.0 +2023-02-07 09:27:56,326 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201829.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:28:03,383 INFO [train.py:901] (2/4) Epoch 25, batch 7850, loss[loss=0.1819, simple_loss=0.2697, pruned_loss=0.04701, over 7927.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.284, pruned_loss=0.05892, over 1614843.79 frames. ], batch size: 20, lr: 2.99e-03, grad_scale: 4.0 +2023-02-07 09:28:27,652 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3633, 1.5445, 2.0532, 1.2912, 1.4716, 1.6268, 1.3783, 1.5981], + device='cuda:2'), covar=tensor([0.1831, 0.2508, 0.0964, 0.4221, 0.1948, 0.3089, 0.2284, 0.2132], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0628, 0.0559, 0.0663, 0.0662, 0.0607, 0.0556, 0.0644], + 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') +2023-02-07 09:28:36,527 INFO [train.py:901] (2/4) Epoch 25, batch 7900, loss[loss=0.1813, simple_loss=0.2686, pruned_loss=0.04703, over 8470.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.284, pruned_loss=0.05889, over 1617302.52 frames. ], batch size: 25, lr: 2.99e-03, grad_scale: 4.0 +2023-02-07 09:28:47,146 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.505e+02 3.187e+02 3.787e+02 7.491e+02, threshold=6.375e+02, percent-clipped=2.0 +2023-02-07 09:29:09,582 INFO [train.py:901] (2/4) Epoch 25, batch 7950, loss[loss=0.2148, simple_loss=0.2876, pruned_loss=0.07104, over 7979.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2835, pruned_loss=0.05856, over 1615446.85 frames. ], batch size: 21, lr: 2.99e-03, grad_scale: 4.0 +2023-02-07 09:29:40,350 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201986.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:42,799 INFO [train.py:901] (2/4) Epoch 25, batch 8000, loss[loss=0.1698, simple_loss=0.2694, pruned_loss=0.03513, over 8182.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2827, pruned_loss=0.05809, over 1612154.22 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:29:46,048 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9380, 1.4732, 3.2853, 1.4403, 2.4905, 3.6191, 3.7172, 3.0906], + device='cuda:2'), covar=tensor([0.1207, 0.1876, 0.0342, 0.2187, 0.0975, 0.0234, 0.0631, 0.0575], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0323, 0.0290, 0.0318, 0.0318, 0.0274, 0.0435, 0.0307], + 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') +2023-02-07 09:29:46,070 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9479, 1.4235, 1.6346, 1.3281, 0.8840, 1.4490, 1.7345, 1.5216], + device='cuda:2'), covar=tensor([0.0578, 0.1252, 0.1779, 0.1554, 0.0644, 0.1470, 0.0701, 0.0681], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 09:29:54,391 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 2.298e+02 3.131e+02 3.789e+02 6.155e+02, threshold=6.263e+02, percent-clipped=0.0 +2023-02-07 09:29:54,489 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202006.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:55,330 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202007.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:55,933 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202008.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:58,056 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202011.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:06,542 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:12,417 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202032.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:17,709 INFO [train.py:901] (2/4) Epoch 25, batch 8050, loss[loss=0.201, simple_loss=0.2826, pruned_loss=0.05965, over 7930.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2822, pruned_loss=0.05856, over 1602840.51 frames. ], batch size: 20, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:30:36,924 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202068.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:50,320 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 09:30:55,054 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 09:30:55,321 INFO [train.py:901] (2/4) Epoch 26, batch 0, loss[loss=0.1926, simple_loss=0.2648, pruned_loss=0.06018, over 7281.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2648, pruned_loss=0.06018, over 7281.00 frames. ], batch size: 16, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:30:55,321 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 09:31:04,873 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3323, 2.1111, 1.6051, 1.8532, 1.7381, 1.5305, 1.6685, 1.7044], + device='cuda:2'), covar=tensor([0.1358, 0.0404, 0.1257, 0.0602, 0.0736, 0.1497, 0.0944, 0.0924], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0239, 0.0341, 0.0314, 0.0303, 0.0346, 0.0350, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 09:31:06,905 INFO [train.py:935] (2/4) Epoch 26, validation: loss=0.1717, simple_loss=0.2716, pruned_loss=0.03591, over 944034.00 frames. +2023-02-07 09:31:06,905 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 09:31:21,613 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 09:31:29,810 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.411e+02 2.993e+02 3.956e+02 9.314e+02, threshold=5.987e+02, percent-clipped=4.0 +2023-02-07 09:31:40,837 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:31:41,325 INFO [train.py:901] (2/4) Epoch 26, batch 50, loss[loss=0.1817, simple_loss=0.2673, pruned_loss=0.04809, over 8136.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2868, pruned_loss=0.05916, over 368321.88 frames. ], batch size: 22, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:31:52,558 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202138.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:31:55,759 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 09:32:15,981 INFO [train.py:901] (2/4) Epoch 26, batch 100, loss[loss=0.1907, simple_loss=0.2829, pruned_loss=0.0493, over 8453.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2893, pruned_loss=0.06139, over 650163.24 frames. ], batch size: 27, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:32:18,600 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 09:32:23,606 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202183.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:32:40,585 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.435e+02 2.962e+02 3.649e+02 8.375e+02, threshold=5.925e+02, percent-clipped=4.0 +2023-02-07 09:32:51,116 INFO [train.py:901] (2/4) Epoch 26, batch 150, loss[loss=0.2034, simple_loss=0.2838, pruned_loss=0.06148, over 8081.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.29, pruned_loss=0.06244, over 865320.49 frames. ], batch size: 21, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:33:22,414 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2368, 1.4304, 3.3841, 1.0802, 2.9843, 2.8646, 3.1049, 3.0514], + device='cuda:2'), covar=tensor([0.0984, 0.4073, 0.0844, 0.4619, 0.1456, 0.1172, 0.0807, 0.0911], + device='cuda:2'), in_proj_covar=tensor([0.0660, 0.0653, 0.0721, 0.0646, 0.0730, 0.0626, 0.0625, 0.0699], + 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') +2023-02-07 09:33:26,391 INFO [train.py:901] (2/4) Epoch 26, batch 200, loss[loss=0.1931, simple_loss=0.2775, pruned_loss=0.05436, over 8286.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.29, pruned_loss=0.06212, over 1035163.20 frames. ], batch size: 23, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:33:49,948 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 2.408e+02 2.928e+02 3.669e+02 9.390e+02, threshold=5.857e+02, percent-clipped=3.0 +2023-02-07 09:34:01,571 INFO [train.py:901] (2/4) Epoch 26, batch 250, loss[loss=0.2433, simple_loss=0.326, pruned_loss=0.0803, over 8250.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2899, pruned_loss=0.0618, over 1167401.14 frames. ], batch size: 24, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:34:09,725 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 09:34:19,979 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 09:34:22,771 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202352.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:35,408 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 09:34:36,400 INFO [train.py:901] (2/4) Epoch 26, batch 300, loss[loss=0.1907, simple_loss=0.2841, pruned_loss=0.04861, over 8456.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2892, pruned_loss=0.06194, over 1265856.13 frames. ], batch size: 25, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:34:40,015 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:52,235 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202394.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:57,267 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202402.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:59,616 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.510e+02 3.033e+02 3.572e+02 1.183e+03, threshold=6.066e+02, percent-clipped=2.0 +2023-02-07 09:35:08,710 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202419.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:10,555 INFO [train.py:901] (2/4) Epoch 26, batch 350, loss[loss=0.1617, simple_loss=0.2364, pruned_loss=0.04356, over 7447.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2876, pruned_loss=0.06094, over 1344594.88 frames. ], batch size: 17, lr: 2.93e-03, grad_scale: 4.0 +2023-02-07 09:35:22,053 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3354, 2.1086, 2.6079, 2.2594, 2.6272, 2.2831, 2.2324, 1.8458], + device='cuda:2'), covar=tensor([0.4017, 0.4234, 0.1659, 0.2944, 0.1935, 0.2730, 0.1556, 0.4085], + device='cuda:2'), in_proj_covar=tensor([0.0954, 0.1006, 0.0820, 0.0977, 0.1015, 0.0916, 0.0764, 0.0840], + 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') +2023-02-07 09:35:23,407 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:40,988 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202464.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:43,101 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202467.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:46,423 INFO [train.py:901] (2/4) Epoch 26, batch 400, loss[loss=0.2118, simple_loss=0.2876, pruned_loss=0.06798, over 8513.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2868, pruned_loss=0.05993, over 1408376.80 frames. ], batch size: 26, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:35:46,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5942, 2.4259, 3.2259, 2.5758, 3.2511, 2.5917, 2.4458, 1.9018], + device='cuda:2'), covar=tensor([0.5654, 0.5131, 0.2129, 0.4438, 0.2949, 0.3131, 0.1896, 0.6090], + device='cuda:2'), in_proj_covar=tensor([0.0953, 0.1005, 0.0820, 0.0977, 0.1015, 0.0915, 0.0763, 0.0840], + 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') +2023-02-07 09:36:09,159 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8645, 2.0139, 1.6993, 2.4617, 1.2746, 1.5331, 1.9203, 1.9943], + device='cuda:2'), covar=tensor([0.0724, 0.0745, 0.0909, 0.0474, 0.1050, 0.1288, 0.0736, 0.0739], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0212, 0.0206, 0.0246, 0.0249, 0.0207], + 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') +2023-02-07 09:36:11,055 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.432e+02 2.504e+02 3.071e+02 3.633e+02 8.131e+02, threshold=6.142e+02, percent-clipped=3.0 +2023-02-07 09:36:21,079 INFO [train.py:901] (2/4) Epoch 26, batch 450, loss[loss=0.2231, simple_loss=0.3194, pruned_loss=0.06342, over 8312.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2848, pruned_loss=0.05923, over 1453218.59 frames. ], batch size: 49, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:36:40,725 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 +2023-02-07 09:36:55,498 INFO [train.py:901] (2/4) Epoch 26, batch 500, loss[loss=0.1756, simple_loss=0.2571, pruned_loss=0.04706, over 7546.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2847, pruned_loss=0.05928, over 1490003.49 frames. ], batch size: 18, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:37:19,245 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.396e+02 2.962e+02 4.085e+02 8.069e+02, threshold=5.924e+02, percent-clipped=6.0 +2023-02-07 09:37:29,372 INFO [train.py:901] (2/4) Epoch 26, batch 550, loss[loss=0.2719, simple_loss=0.3418, pruned_loss=0.101, over 6794.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2857, pruned_loss=0.05995, over 1520752.55 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:37:41,179 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 09:37:52,232 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 09:38:01,512 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1343, 1.6407, 4.4379, 2.0684, 2.5502, 5.1002, 5.1966, 4.3982], + device='cuda:2'), covar=tensor([0.1316, 0.1929, 0.0292, 0.2009, 0.1216, 0.0181, 0.0580, 0.0568], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0326, 0.0289, 0.0319, 0.0318, 0.0274, 0.0433, 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') +2023-02-07 09:38:05,142 INFO [train.py:901] (2/4) Epoch 26, batch 600, loss[loss=0.1743, simple_loss=0.2491, pruned_loss=0.04976, over 7537.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2847, pruned_loss=0.05931, over 1541009.00 frames. ], batch size: 18, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:38:21,711 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 09:38:29,023 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.445e+02 2.916e+02 3.512e+02 6.749e+02, threshold=5.833e+02, percent-clipped=3.0 +2023-02-07 09:38:38,962 INFO [train.py:901] (2/4) Epoch 26, batch 650, loss[loss=0.1759, simple_loss=0.2731, pruned_loss=0.03941, over 7974.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.285, pruned_loss=0.05961, over 1553901.20 frames. ], batch size: 21, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:38:39,871 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:38:57,381 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202748.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:39:14,890 INFO [train.py:901] (2/4) Epoch 26, batch 700, loss[loss=0.2079, simple_loss=0.294, pruned_loss=0.06094, over 8284.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2846, pruned_loss=0.05945, over 1561596.07 frames. ], batch size: 23, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:39:38,616 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.498e+02 3.029e+02 3.750e+02 8.351e+02, threshold=6.058e+02, percent-clipped=3.0 +2023-02-07 09:39:45,991 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 09:39:49,880 INFO [train.py:901] (2/4) Epoch 26, batch 750, loss[loss=0.1743, simple_loss=0.2633, pruned_loss=0.0426, over 7804.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2851, pruned_loss=0.05945, over 1575323.35 frames. ], batch size: 20, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:40:05,054 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 09:40:07,994 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202848.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:40:13,834 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 09:40:24,779 INFO [train.py:901] (2/4) Epoch 26, batch 800, loss[loss=0.1996, simple_loss=0.2926, pruned_loss=0.05333, over 8667.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2865, pruned_loss=0.06014, over 1589666.18 frames. ], batch size: 34, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:40:49,974 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.489e+02 2.818e+02 3.827e+02 7.280e+02, threshold=5.635e+02, percent-clipped=3.0 +2023-02-07 09:40:59,899 INFO [train.py:901] (2/4) Epoch 26, batch 850, loss[loss=0.1548, simple_loss=0.2357, pruned_loss=0.03697, over 7442.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2848, pruned_loss=0.05895, over 1594090.95 frames. ], batch size: 17, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:41:33,349 INFO [train.py:901] (2/4) Epoch 26, batch 900, loss[loss=0.1894, simple_loss=0.2616, pruned_loss=0.05857, over 7655.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2856, pruned_loss=0.05923, over 1602892.15 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:41:58,978 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.742e+02 3.265e+02 4.005e+02 6.934e+02, threshold=6.531e+02, percent-clipped=5.0 +2023-02-07 09:42:08,858 INFO [train.py:901] (2/4) Epoch 26, batch 950, loss[loss=0.2053, simple_loss=0.2932, pruned_loss=0.05866, over 8466.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2859, pruned_loss=0.05941, over 1610065.27 frames. ], batch size: 25, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:42:12,078 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 09:42:31,729 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 09:42:32,525 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:42:43,151 INFO [train.py:901] (2/4) Epoch 26, batch 1000, loss[loss=0.1879, simple_loss=0.2689, pruned_loss=0.05345, over 7983.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2856, pruned_loss=0.05931, over 1612585.94 frames. ], batch size: 21, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:42:53,364 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0393, 1.7193, 3.5390, 1.4982, 2.3630, 3.9273, 3.9703, 3.3416], + device='cuda:2'), covar=tensor([0.1133, 0.1643, 0.0319, 0.2110, 0.1061, 0.0220, 0.0505, 0.0574], + device='cuda:2'), in_proj_covar=tensor([0.0301, 0.0323, 0.0288, 0.0316, 0.0316, 0.0273, 0.0432, 0.0304], + 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') +2023-02-07 09:43:04,711 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203103.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:43:05,256 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 09:43:07,153 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.540e+02 2.348e+02 2.857e+02 3.355e+02 6.976e+02, threshold=5.714e+02, percent-clipped=1.0 +2023-02-07 09:43:17,239 INFO [train.py:901] (2/4) Epoch 26, batch 1050, loss[loss=0.172, simple_loss=0.2492, pruned_loss=0.04742, over 7804.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05873, over 1610111.21 frames. ], batch size: 20, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:43:18,670 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 09:43:53,220 INFO [train.py:901] (2/4) Epoch 26, batch 1100, loss[loss=0.1785, simple_loss=0.2552, pruned_loss=0.05089, over 7425.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.0587, over 1612451.10 frames. ], batch size: 17, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:44:06,848 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203192.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:44:16,729 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.511e+02 2.912e+02 3.711e+02 8.666e+02, threshold=5.824e+02, percent-clipped=4.0 +2023-02-07 09:44:27,591 INFO [train.py:901] (2/4) Epoch 26, batch 1150, loss[loss=0.2393, simple_loss=0.3209, pruned_loss=0.07884, over 8455.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2833, pruned_loss=0.0588, over 1608938.33 frames. ], batch size: 27, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:44:27,597 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 09:45:02,742 INFO [train.py:901] (2/4) Epoch 26, batch 1200, loss[loss=0.2255, simple_loss=0.3177, pruned_loss=0.06663, over 8187.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2837, pruned_loss=0.05871, over 1606524.57 frames. ], batch size: 23, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:45:27,283 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203307.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:45:27,743 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.402e+02 2.806e+02 3.306e+02 6.331e+02, threshold=5.612e+02, percent-clipped=2.0 +2023-02-07 09:45:37,087 INFO [train.py:901] (2/4) Epoch 26, batch 1250, loss[loss=0.1776, simple_loss=0.2585, pruned_loss=0.04837, over 7425.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2828, pruned_loss=0.05836, over 1608331.26 frames. ], batch size: 17, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:45:49,554 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7602, 1.9207, 2.0612, 1.3198, 2.1257, 1.5353, 0.5538, 1.9492], + device='cuda:2'), covar=tensor([0.0579, 0.0400, 0.0314, 0.0619, 0.0421, 0.0978, 0.0928, 0.0302], + device='cuda:2'), in_proj_covar=tensor([0.0461, 0.0400, 0.0356, 0.0453, 0.0387, 0.0539, 0.0395, 0.0428], + device='cuda:2'), out_proj_covar=tensor([1.2258e-04, 1.0409e-04, 9.2922e-05, 1.1871e-04, 1.0138e-04, 1.5074e-04, + 1.0581e-04, 1.1255e-04], device='cuda:2') +2023-02-07 09:46:02,956 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9979, 2.2019, 1.7640, 2.8514, 1.3360, 1.5955, 2.0701, 2.2636], + device='cuda:2'), covar=tensor([0.0674, 0.0754, 0.0910, 0.0301, 0.1048, 0.1221, 0.0771, 0.0688], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0194, 0.0246, 0.0212, 0.0205, 0.0247, 0.0249, 0.0207], + 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') +2023-02-07 09:46:11,111 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.67 vs. limit=5.0 +2023-02-07 09:46:12,721 INFO [train.py:901] (2/4) Epoch 26, batch 1300, loss[loss=0.1647, simple_loss=0.2426, pruned_loss=0.04346, over 7928.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2838, pruned_loss=0.05885, over 1612722.58 frames. ], batch size: 20, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:46:29,987 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5998, 1.5406, 2.0702, 1.4353, 1.1535, 2.0676, 0.3860, 1.2970], + device='cuda:2'), covar=tensor([0.1364, 0.1306, 0.0361, 0.0902, 0.2438, 0.0357, 0.1763, 0.1195], + device='cuda:2'), in_proj_covar=tensor([0.0198, 0.0204, 0.0132, 0.0224, 0.0278, 0.0143, 0.0173, 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') +2023-02-07 09:46:31,892 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:46:37,070 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.437e+02 2.919e+02 3.429e+02 9.499e+02, threshold=5.838e+02, percent-clipped=5.0 +2023-02-07 09:46:46,433 INFO [train.py:901] (2/4) Epoch 26, batch 1350, loss[loss=0.1713, simple_loss=0.2453, pruned_loss=0.04863, over 7239.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.05868, over 1614328.12 frames. ], batch size: 16, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:47:04,177 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203447.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:47:22,494 INFO [train.py:901] (2/4) Epoch 26, batch 1400, loss[loss=0.1841, simple_loss=0.2735, pruned_loss=0.0474, over 8364.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2831, pruned_loss=0.05848, over 1614652.06 frames. ], batch size: 24, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:47:24,885 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 09:47:31,505 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203485.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:47:47,758 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.419e+02 2.906e+02 3.589e+02 5.599e+02, threshold=5.812e+02, percent-clipped=0.0 +2023-02-07 09:47:52,596 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203515.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:47:54,386 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 09:47:57,074 INFO [train.py:901] (2/4) Epoch 26, batch 1450, loss[loss=0.211, simple_loss=0.2921, pruned_loss=0.06493, over 8464.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2827, pruned_loss=0.05802, over 1613483.52 frames. ], batch size: 29, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:48:01,588 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 09:48:10,650 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5887, 2.0943, 3.3466, 1.4623, 2.5243, 2.0556, 1.7009, 2.7517], + device='cuda:2'), covar=tensor([0.1967, 0.2720, 0.0744, 0.4783, 0.1804, 0.3288, 0.2447, 0.1909], + device='cuda:2'), in_proj_covar=tensor([0.0533, 0.0623, 0.0554, 0.0659, 0.0654, 0.0602, 0.0552, 0.0637], + 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') +2023-02-07 09:48:18,154 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4596, 2.1540, 2.8297, 2.3493, 2.8157, 2.4752, 2.2913, 1.7260], + device='cuda:2'), covar=tensor([0.5576, 0.5134, 0.2043, 0.3938, 0.2657, 0.3180, 0.1809, 0.5613], + device='cuda:2'), in_proj_covar=tensor([0.0962, 0.1018, 0.0830, 0.0988, 0.1025, 0.0926, 0.0772, 0.0849], + 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') +2023-02-07 09:48:24,037 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203562.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:48:24,758 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203563.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:48:31,362 INFO [train.py:901] (2/4) Epoch 26, batch 1500, loss[loss=0.2012, simple_loss=0.2857, pruned_loss=0.05831, over 8082.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2829, pruned_loss=0.05869, over 1612877.55 frames. ], batch size: 21, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:48:33,571 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 09:48:42,963 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203588.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:48:56,866 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.262e+02 2.668e+02 3.517e+02 8.500e+02, threshold=5.335e+02, percent-clipped=2.0 +2023-02-07 09:49:06,819 INFO [train.py:901] (2/4) Epoch 26, batch 1550, loss[loss=0.2054, simple_loss=0.2739, pruned_loss=0.06851, over 7639.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2828, pruned_loss=0.05842, over 1616282.60 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:49:15,918 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6846, 1.9108, 1.9859, 1.2574, 2.0749, 1.5175, 0.5144, 1.9273], + device='cuda:2'), covar=tensor([0.0601, 0.0449, 0.0342, 0.0728, 0.0488, 0.0934, 0.0998, 0.0393], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0405, 0.0359, 0.0457, 0.0392, 0.0545, 0.0399, 0.0432], + device='cuda:2'), out_proj_covar=tensor([1.2384e-04, 1.0548e-04, 9.3703e-05, 1.1969e-04, 1.0256e-04, 1.5250e-04, + 1.0683e-04, 1.1347e-04], device='cuda:2') +2023-02-07 09:49:40,409 INFO [train.py:901] (2/4) Epoch 26, batch 1600, loss[loss=0.1868, simple_loss=0.2662, pruned_loss=0.05369, over 7649.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2817, pruned_loss=0.05787, over 1614225.37 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:50:05,089 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.410e+02 3.021e+02 3.901e+02 1.362e+03, threshold=6.042e+02, percent-clipped=8.0 +2023-02-07 09:50:15,006 INFO [train.py:901] (2/4) Epoch 26, batch 1650, loss[loss=0.2286, simple_loss=0.3116, pruned_loss=0.07283, over 8515.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2837, pruned_loss=0.05914, over 1616875.21 frames. ], batch size: 26, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:50:35,769 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 09:50:44,790 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203765.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:50:48,717 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203771.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:50:49,154 INFO [train.py:901] (2/4) Epoch 26, batch 1700, loss[loss=0.2223, simple_loss=0.3081, pruned_loss=0.06829, over 8477.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2843, pruned_loss=0.05943, over 1620347.46 frames. ], batch size: 28, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:50:54,930 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 09:50:57,401 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203784.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:05,469 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203796.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:14,292 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.478e+02 3.017e+02 3.791e+02 8.735e+02, threshold=6.035e+02, percent-clipped=4.0 +2023-02-07 09:51:20,898 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203818.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:23,317 INFO [train.py:901] (2/4) Epoch 26, batch 1750, loss[loss=0.1748, simple_loss=0.2621, pruned_loss=0.04378, over 7785.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2845, pruned_loss=0.05958, over 1621246.42 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:51:28,107 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203829.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:38,336 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:54,079 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:58,819 INFO [train.py:901] (2/4) Epoch 26, batch 1800, loss[loss=0.1887, simple_loss=0.2757, pruned_loss=0.05078, over 7821.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2845, pruned_loss=0.05965, over 1619966.25 frames. ], batch size: 20, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:52:23,002 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.441e+02 2.799e+02 3.336e+02 4.977e+02, threshold=5.598e+02, percent-clipped=0.0 +2023-02-07 09:52:23,386 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 09:52:29,710 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203918.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:52:32,265 INFO [train.py:901] (2/4) Epoch 26, batch 1850, loss[loss=0.2027, simple_loss=0.2748, pruned_loss=0.06527, over 7713.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2846, pruned_loss=0.05945, over 1618345.25 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:52:47,701 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203944.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:53:07,703 INFO [train.py:901] (2/4) Epoch 26, batch 1900, loss[loss=0.218, simple_loss=0.3054, pruned_loss=0.06529, over 8331.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2851, pruned_loss=0.05937, over 1624018.65 frames. ], batch size: 26, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:53:09,159 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9921, 1.7133, 3.5166, 1.6595, 2.4517, 3.8997, 3.9808, 3.3642], + device='cuda:2'), covar=tensor([0.1150, 0.1649, 0.0334, 0.2005, 0.1015, 0.0215, 0.0523, 0.0496], + device='cuda:2'), in_proj_covar=tensor([0.0299, 0.0319, 0.0286, 0.0314, 0.0314, 0.0271, 0.0428, 0.0300], + 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') +2023-02-07 09:53:33,457 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.507e+02 3.073e+02 4.108e+02 9.647e+02, threshold=6.146e+02, percent-clipped=9.0 +2023-02-07 09:53:35,030 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5302, 1.4735, 1.8697, 1.2982, 1.1885, 1.8618, 0.2523, 1.2316], + device='cuda:2'), covar=tensor([0.1465, 0.1156, 0.0350, 0.0839, 0.2360, 0.0377, 0.1811, 0.1143], + device='cuda:2'), in_proj_covar=tensor([0.0198, 0.0204, 0.0133, 0.0225, 0.0278, 0.0143, 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') +2023-02-07 09:53:36,901 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 09:53:42,909 INFO [train.py:901] (2/4) Epoch 26, batch 1950, loss[loss=0.1988, simple_loss=0.2794, pruned_loss=0.05913, over 7780.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2842, pruned_loss=0.05871, over 1623453.79 frames. ], batch size: 19, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:53:49,444 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 09:54:07,148 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 09:54:17,198 INFO [train.py:901] (2/4) Epoch 26, batch 2000, loss[loss=0.1734, simple_loss=0.2621, pruned_loss=0.0424, over 8080.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2848, pruned_loss=0.05917, over 1623820.19 frames. ], batch size: 21, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:54:34,348 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204095.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:54:43,741 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.388e+02 3.050e+02 3.690e+02 7.171e+02, threshold=6.101e+02, percent-clipped=4.0 +2023-02-07 09:54:44,462 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204109.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:54:48,534 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8538, 1.6632, 2.0451, 1.5837, 1.0408, 1.8259, 2.3674, 2.3078], + device='cuda:2'), covar=tensor([0.0459, 0.1241, 0.1560, 0.1460, 0.0609, 0.1452, 0.0590, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 09:54:53,081 INFO [train.py:901] (2/4) Epoch 26, batch 2050, loss[loss=0.1841, simple_loss=0.2753, pruned_loss=0.04648, over 8087.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2841, pruned_loss=0.05936, over 1618634.66 frames. ], batch size: 21, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:54:57,140 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204128.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:26,603 INFO [train.py:901] (2/4) Epoch 26, batch 2100, loss[loss=0.2061, simple_loss=0.2858, pruned_loss=0.06317, over 8249.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2844, pruned_loss=0.05965, over 1616009.10 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:55:33,648 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204181.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:47,785 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204200.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:52,931 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.312e+02 2.797e+02 3.552e+02 6.063e+02, threshold=5.595e+02, percent-clipped=0.0 +2023-02-07 09:55:53,721 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204209.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:02,439 INFO [train.py:901] (2/4) Epoch 26, batch 2150, loss[loss=0.1701, simple_loss=0.2397, pruned_loss=0.05027, over 7694.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2836, pruned_loss=0.05908, over 1617317.57 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:56:03,969 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:04,624 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204225.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:17,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:29,891 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204262.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:56:36,274 INFO [train.py:901] (2/4) Epoch 26, batch 2200, loss[loss=0.2235, simple_loss=0.3056, pruned_loss=0.07071, over 8448.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2843, pruned_loss=0.05965, over 1618005.34 frames. ], batch size: 27, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:56:48,681 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4191, 1.6681, 1.6980, 1.1606, 1.7092, 1.3569, 0.2995, 1.6729], + device='cuda:2'), covar=tensor([0.0499, 0.0397, 0.0318, 0.0560, 0.0403, 0.1007, 0.0979, 0.0300], + device='cuda:2'), in_proj_covar=tensor([0.0464, 0.0402, 0.0357, 0.0453, 0.0389, 0.0544, 0.0396, 0.0430], + device='cuda:2'), out_proj_covar=tensor([1.2326e-04, 1.0480e-04, 9.3285e-05, 1.1880e-04, 1.0173e-04, 1.5213e-04, + 1.0605e-04, 1.1306e-04], device='cuda:2') +2023-02-07 09:57:01,384 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.362e+02 3.099e+02 4.074e+02 1.599e+03, threshold=6.197e+02, percent-clipped=8.0 +2023-02-07 09:57:11,815 INFO [train.py:901] (2/4) Epoch 26, batch 2250, loss[loss=0.1973, simple_loss=0.2863, pruned_loss=0.05414, over 8363.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2843, pruned_loss=0.05938, over 1614136.75 frames. ], batch size: 24, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:57:13,367 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:57:30,422 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5495, 4.5741, 4.1593, 2.0242, 4.0097, 4.1970, 4.1177, 4.0436], + device='cuda:2'), covar=tensor([0.0669, 0.0496, 0.0942, 0.4878, 0.0829, 0.1019, 0.1247, 0.0678], + device='cuda:2'), in_proj_covar=tensor([0.0533, 0.0451, 0.0433, 0.0548, 0.0434, 0.0454, 0.0431, 0.0394], + 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') +2023-02-07 09:57:46,482 INFO [train.py:901] (2/4) Epoch 26, batch 2300, loss[loss=0.223, simple_loss=0.3053, pruned_loss=0.07037, over 8491.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2849, pruned_loss=0.05951, over 1617369.68 frames. ], batch size: 39, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:57:50,092 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204377.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:58:10,713 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.304e+02 2.813e+02 3.713e+02 7.684e+02, threshold=5.626e+02, percent-clipped=3.0 +2023-02-07 09:58:21,043 INFO [train.py:901] (2/4) Epoch 26, batch 2350, loss[loss=0.1899, simple_loss=0.2735, pruned_loss=0.05311, over 8025.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2853, pruned_loss=0.05961, over 1619554.49 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:58:33,860 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:58:34,647 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0286, 1.8608, 2.0937, 1.8451, 0.9548, 1.8334, 2.4596, 3.0270], + device='cuda:2'), covar=tensor([0.0438, 0.1173, 0.1572, 0.1349, 0.0578, 0.1418, 0.0542, 0.0417], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0191, 0.0161, 0.0100, 0.0164, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 09:58:57,197 INFO [train.py:901] (2/4) Epoch 26, batch 2400, loss[loss=0.28, simple_loss=0.3459, pruned_loss=0.1071, over 6867.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2851, pruned_loss=0.05949, over 1618647.73 frames. ], batch size: 72, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:58:58,110 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7922, 2.1568, 3.6440, 2.0580, 1.7676, 3.5504, 0.7599, 2.1646], + device='cuda:2'), covar=tensor([0.1266, 0.1183, 0.0189, 0.1380, 0.2314, 0.0329, 0.1778, 0.1110], + device='cuda:2'), in_proj_covar=tensor([0.0198, 0.0204, 0.0132, 0.0224, 0.0278, 0.0144, 0.0173, 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') +2023-02-07 09:59:02,876 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204480.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:16,094 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:20,398 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204505.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:22,312 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.551e+02 2.904e+02 3.805e+02 7.023e+02, threshold=5.807e+02, percent-clipped=3.0 +2023-02-07 09:59:32,133 INFO [train.py:901] (2/4) Epoch 26, batch 2450, loss[loss=0.2534, simple_loss=0.326, pruned_loss=0.09036, over 8528.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2841, pruned_loss=0.05927, over 1619321.75 frames. ], batch size: 28, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:59:33,748 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204524.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:34,344 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204525.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:43,679 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2464, 2.0368, 2.6279, 2.2249, 2.7228, 2.3340, 2.1224, 1.5742], + device='cuda:2'), covar=tensor([0.5738, 0.5240, 0.2355, 0.4193, 0.2586, 0.3394, 0.1981, 0.5911], + device='cuda:2'), in_proj_covar=tensor([0.0956, 0.1011, 0.0824, 0.0983, 0.1020, 0.0919, 0.0766, 0.0845], + 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') +2023-02-07 09:59:56,603 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:00:08,715 INFO [train.py:901] (2/4) Epoch 26, batch 2500, loss[loss=0.2185, simple_loss=0.3005, pruned_loss=0.06825, over 8241.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2842, pruned_loss=0.05926, over 1617917.93 frames. ], batch size: 24, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:00:15,138 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204580.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:00:32,007 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204605.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:00:33,807 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.483e+02 3.074e+02 3.585e+02 8.993e+02, threshold=6.148e+02, percent-clipped=7.0 +2023-02-07 10:00:42,666 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9179, 1.5317, 1.8152, 1.3866, 1.0272, 1.6245, 1.7151, 1.5310], + device='cuda:2'), covar=tensor([0.0517, 0.1187, 0.1607, 0.1434, 0.0588, 0.1359, 0.0676, 0.0670], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 10:00:43,170 INFO [train.py:901] (2/4) Epoch 26, batch 2550, loss[loss=0.1593, simple_loss=0.2369, pruned_loss=0.04079, over 7536.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2843, pruned_loss=0.05931, over 1618972.73 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:00:50,556 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204633.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 10:00:55,148 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204640.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:01:02,546 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0190, 1.6225, 1.8943, 1.5427, 1.0073, 1.6602, 1.7714, 1.7093], + device='cuda:2'), covar=tensor([0.0541, 0.1185, 0.1561, 0.1339, 0.0607, 0.1361, 0.0689, 0.0598], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 10:01:07,864 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204658.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 10:01:18,433 INFO [train.py:901] (2/4) Epoch 26, batch 2600, loss[loss=0.2138, simple_loss=0.3026, pruned_loss=0.06254, over 8141.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2839, pruned_loss=0.0593, over 1617317.81 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:01:43,342 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.465e+02 3.094e+02 3.874e+02 9.576e+02, threshold=6.187e+02, percent-clipped=4.0 +2023-02-07 10:01:52,895 INFO [train.py:901] (2/4) Epoch 26, batch 2650, loss[loss=0.199, simple_loss=0.2889, pruned_loss=0.05458, over 8487.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2833, pruned_loss=0.05892, over 1614425.99 frames. ], batch size: 26, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:02:28,028 INFO [train.py:901] (2/4) Epoch 26, batch 2700, loss[loss=0.1778, simple_loss=0.254, pruned_loss=0.05076, over 7416.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2827, pruned_loss=0.05868, over 1608903.95 frames. ], batch size: 17, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:02:53,803 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.359e+02 2.865e+02 3.674e+02 6.992e+02, threshold=5.730e+02, percent-clipped=1.0 +2023-02-07 10:02:55,424 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204810.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:03:04,000 INFO [train.py:901] (2/4) Epoch 26, batch 2750, loss[loss=0.2066, simple_loss=0.2832, pruned_loss=0.06496, over 7687.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2825, pruned_loss=0.05885, over 1602759.96 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:03:13,268 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204835.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:03:16,026 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204839.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:03:38,127 INFO [train.py:901] (2/4) Epoch 26, batch 2800, loss[loss=0.1838, simple_loss=0.2654, pruned_loss=0.05109, over 7428.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2824, pruned_loss=0.05883, over 1599094.61 frames. ], batch size: 17, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:03:55,277 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204896.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:04:04,627 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.285e+02 3.108e+02 3.828e+02 9.944e+02, threshold=6.216e+02, percent-clipped=6.0 +2023-02-07 10:04:13,926 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204921.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:04:14,419 INFO [train.py:901] (2/4) Epoch 26, batch 2850, loss[loss=0.1986, simple_loss=0.2647, pruned_loss=0.06626, over 7652.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2826, pruned_loss=0.05885, over 1599221.43 frames. ], batch size: 19, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:04:23,885 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3878, 1.1622, 2.2782, 1.3721, 2.0819, 2.4583, 2.6171, 2.0667], + device='cuda:2'), covar=tensor([0.1235, 0.1669, 0.0457, 0.2030, 0.0877, 0.0426, 0.0802, 0.0691], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0327, 0.0290, 0.0319, 0.0321, 0.0276, 0.0435, 0.0307], + 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') +2023-02-07 10:04:48,524 INFO [train.py:901] (2/4) Epoch 26, batch 2900, loss[loss=0.1939, simple_loss=0.2845, pruned_loss=0.05164, over 8247.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2821, pruned_loss=0.05829, over 1600810.56 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:04:59,571 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.16 vs. limit=5.0 +2023-02-07 10:05:00,983 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 10:05:13,383 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.634e+02 2.393e+02 3.052e+02 3.991e+02 9.487e+02, threshold=6.105e+02, percent-clipped=5.0 +2023-02-07 10:05:20,486 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 10:05:23,875 INFO [train.py:901] (2/4) Epoch 26, batch 2950, loss[loss=0.1982, simple_loss=0.2818, pruned_loss=0.05723, over 8040.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2817, pruned_loss=0.05836, over 1599074.66 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:05:58,487 INFO [train.py:901] (2/4) Epoch 26, batch 3000, loss[loss=0.2039, simple_loss=0.2883, pruned_loss=0.05977, over 8470.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2821, pruned_loss=0.05836, over 1602878.53 frames. ], batch size: 25, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:05:58,488 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 10:06:07,785 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.7745, 1.4910, 3.9225, 1.6138, 3.5183, 3.2459, 3.6190, 3.5082], + device='cuda:2'), covar=tensor([0.0702, 0.4520, 0.0513, 0.4127, 0.1008, 0.1045, 0.0630, 0.0697], + device='cuda:2'), in_proj_covar=tensor([0.0665, 0.0662, 0.0728, 0.0650, 0.0742, 0.0629, 0.0629, 0.0706], + 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') +2023-02-07 10:06:11,419 INFO [train.py:935] (2/4) Epoch 26, validation: loss=0.1716, simple_loss=0.2713, pruned_loss=0.03593, over 944034.00 frames. +2023-02-07 10:06:11,419 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 10:06:31,073 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-07 10:06:36,703 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.246e+02 2.785e+02 3.735e+02 7.523e+02, threshold=5.571e+02, percent-clipped=3.0 +2023-02-07 10:06:46,003 INFO [train.py:901] (2/4) Epoch 26, batch 3050, loss[loss=0.1993, simple_loss=0.2837, pruned_loss=0.05749, over 8233.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2833, pruned_loss=0.05855, over 1607871.38 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:06:49,436 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205127.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:07:22,789 INFO [train.py:901] (2/4) Epoch 26, batch 3100, loss[loss=0.198, simple_loss=0.285, pruned_loss=0.05552, over 8760.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2843, pruned_loss=0.05889, over 1611111.64 frames. ], batch size: 30, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:07:30,227 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=205183.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:07:48,156 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.327e+02 2.997e+02 4.038e+02 1.256e+03, threshold=5.993e+02, percent-clipped=7.0 +2023-02-07 10:07:53,693 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205216.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:07:57,548 INFO [train.py:901] (2/4) Epoch 26, batch 3150, loss[loss=0.2151, simple_loss=0.2939, pruned_loss=0.06811, over 8498.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2846, pruned_loss=0.05944, over 1610541.22 frames. ], batch size: 26, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:08:33,476 INFO [train.py:901] (2/4) Epoch 26, batch 3200, loss[loss=0.1457, simple_loss=0.2303, pruned_loss=0.03049, over 7459.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2836, pruned_loss=0.05918, over 1603992.11 frames. ], batch size: 17, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:08:52,247 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205298.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:08:57,569 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4959, 4.5055, 4.0924, 2.2058, 4.0150, 4.0851, 4.0131, 3.9135], + device='cuda:2'), covar=tensor([0.0688, 0.0477, 0.1010, 0.4352, 0.0826, 0.0964, 0.1265, 0.0776], + device='cuda:2'), in_proj_covar=tensor([0.0534, 0.0449, 0.0437, 0.0549, 0.0432, 0.0455, 0.0431, 0.0394], + 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') +2023-02-07 10:08:58,816 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.527e+02 3.010e+02 3.735e+02 6.895e+02, threshold=6.021e+02, percent-clipped=2.0 +2023-02-07 10:09:09,091 INFO [train.py:901] (2/4) Epoch 26, batch 3250, loss[loss=0.2258, simple_loss=0.3123, pruned_loss=0.06964, over 8567.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2836, pruned_loss=0.05914, over 1603604.99 frames. ], batch size: 31, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:09:43,189 INFO [train.py:901] (2/4) Epoch 26, batch 3300, loss[loss=0.1961, simple_loss=0.2874, pruned_loss=0.05239, over 8751.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.284, pruned_loss=0.05939, over 1604943.00 frames. ], batch size: 40, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:09:53,668 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9405, 1.9732, 1.8489, 2.5903, 1.0053, 1.5969, 1.8696, 2.0142], + device='cuda:2'), covar=tensor([0.0754, 0.0819, 0.0874, 0.0372, 0.1192, 0.1370, 0.0874, 0.0803], + device='cuda:2'), in_proj_covar=tensor([0.0228, 0.0192, 0.0243, 0.0210, 0.0203, 0.0245, 0.0248, 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') +2023-02-07 10:10:10,309 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.313e+02 2.653e+02 3.358e+02 9.214e+02, threshold=5.305e+02, percent-clipped=4.0 +2023-02-07 10:10:16,830 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5811, 1.9008, 2.6439, 1.4765, 1.9982, 2.0301, 1.5459, 2.1126], + device='cuda:2'), covar=tensor([0.1893, 0.2515, 0.0995, 0.4527, 0.1869, 0.3023, 0.2553, 0.2006], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0626, 0.0559, 0.0663, 0.0658, 0.0607, 0.0557, 0.0640], + 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') +2023-02-07 10:10:20,066 INFO [train.py:901] (2/4) Epoch 26, batch 3350, loss[loss=0.2476, simple_loss=0.3163, pruned_loss=0.08944, over 6919.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2832, pruned_loss=0.0588, over 1606023.58 frames. ], batch size: 72, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:10:54,101 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=205471.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:10:54,721 INFO [train.py:901] (2/4) Epoch 26, batch 3400, loss[loss=0.2141, simple_loss=0.2964, pruned_loss=0.06588, over 8507.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.0581, over 1609404.69 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:11:02,812 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-07 10:11:03,624 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2604, 1.7978, 4.4248, 2.0625, 2.4586, 5.0920, 5.1722, 4.3691], + device='cuda:2'), covar=tensor([0.1293, 0.1790, 0.0253, 0.1950, 0.1106, 0.0176, 0.0381, 0.0612], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0327, 0.0290, 0.0319, 0.0321, 0.0276, 0.0435, 0.0307], + 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') +2023-02-07 10:11:20,313 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.409e+02 2.883e+02 3.635e+02 7.106e+02, threshold=5.767e+02, percent-clipped=3.0 +2023-02-07 10:11:30,463 INFO [train.py:901] (2/4) Epoch 26, batch 3450, loss[loss=0.1821, simple_loss=0.2674, pruned_loss=0.04842, over 8087.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2834, pruned_loss=0.05838, over 1610625.64 frames. ], batch size: 21, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:11:53,166 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:11:57,170 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=205560.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:12:05,234 INFO [train.py:901] (2/4) Epoch 26, batch 3500, loss[loss=0.1932, simple_loss=0.2902, pruned_loss=0.04814, over 8623.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2829, pruned_loss=0.05779, over 1610240.74 frames. ], batch size: 39, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:12:10,358 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205579.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:12:15,152 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:12:21,539 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6268, 2.0151, 3.0617, 1.4704, 2.2313, 2.0715, 1.6080, 2.3049], + device='cuda:2'), covar=tensor([0.1902, 0.2791, 0.0872, 0.4794, 0.2083, 0.3239, 0.2669, 0.2447], + device='cuda:2'), in_proj_covar=tensor([0.0536, 0.0626, 0.0558, 0.0662, 0.0658, 0.0605, 0.0557, 0.0640], + 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') +2023-02-07 10:12:24,736 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 10:12:30,113 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.208e+02 2.714e+02 3.358e+02 5.744e+02, threshold=5.428e+02, percent-clipped=0.0 +2023-02-07 10:12:35,785 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 10:12:39,511 INFO [train.py:901] (2/4) Epoch 26, batch 3550, loss[loss=0.2034, simple_loss=0.2896, pruned_loss=0.05863, over 8503.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2824, pruned_loss=0.05746, over 1613774.18 frames. ], batch size: 28, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:13:15,378 INFO [train.py:901] (2/4) Epoch 26, batch 3600, loss[loss=0.1702, simple_loss=0.2581, pruned_loss=0.04114, over 8136.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2814, pruned_loss=0.05678, over 1611407.49 frames. ], batch size: 22, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:13:16,200 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205673.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:13:17,475 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205675.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:13:39,661 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.295e+02 2.882e+02 3.730e+02 8.207e+02, threshold=5.763e+02, percent-clipped=6.0 +2023-02-07 10:13:49,108 INFO [train.py:901] (2/4) Epoch 26, batch 3650, loss[loss=0.2152, simple_loss=0.3048, pruned_loss=0.06277, over 8627.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2816, pruned_loss=0.0566, over 1613964.09 frames. ], batch size: 39, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:14:18,432 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205762.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:14:24,803 INFO [train.py:901] (2/4) Epoch 26, batch 3700, loss[loss=0.2104, simple_loss=0.2967, pruned_loss=0.06201, over 8643.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2821, pruned_loss=0.05688, over 1610537.71 frames. ], batch size: 34, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:14:27,592 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 10:14:49,669 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.377e+02 2.968e+02 3.727e+02 1.221e+03, threshold=5.937e+02, percent-clipped=5.0 +2023-02-07 10:14:59,197 INFO [train.py:901] (2/4) Epoch 26, batch 3750, loss[loss=0.1826, simple_loss=0.2507, pruned_loss=0.05721, over 7922.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2814, pruned_loss=0.05693, over 1611530.79 frames. ], batch size: 20, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:15:12,916 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:15:31,310 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205867.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:15:34,014 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2021, 1.0764, 1.2933, 1.0409, 0.9181, 1.2943, 0.0977, 0.8944], + device='cuda:2'), covar=tensor([0.1385, 0.1304, 0.0537, 0.0707, 0.2534, 0.0544, 0.1976, 0.1172], + device='cuda:2'), in_proj_covar=tensor([0.0196, 0.0202, 0.0132, 0.0221, 0.0274, 0.0144, 0.0171, 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') +2023-02-07 10:15:34,438 INFO [train.py:901] (2/4) Epoch 26, batch 3800, loss[loss=0.2558, simple_loss=0.3477, pruned_loss=0.08193, over 8489.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2821, pruned_loss=0.0572, over 1614482.47 frames. ], batch size: 28, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:15:59,218 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.381e+02 2.847e+02 3.364e+02 6.986e+02, threshold=5.694e+02, percent-clipped=1.0 +2023-02-07 10:15:59,376 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205908.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:16:08,812 INFO [train.py:901] (2/4) Epoch 26, batch 3850, loss[loss=0.1891, simple_loss=0.2775, pruned_loss=0.05033, over 8453.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2828, pruned_loss=0.05761, over 1617807.25 frames. ], batch size: 24, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:16:15,126 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205931.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:16:20,580 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1609, 1.7855, 3.4826, 1.6325, 2.5208, 3.9042, 3.9962, 3.3309], + device='cuda:2'), covar=tensor([0.1149, 0.1682, 0.0340, 0.2069, 0.1034, 0.0205, 0.0537, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0328, 0.0291, 0.0321, 0.0322, 0.0279, 0.0438, 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') +2023-02-07 10:16:23,272 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7303, 2.1236, 3.0945, 1.5365, 2.5144, 2.1295, 1.8167, 2.4466], + device='cuda:2'), covar=tensor([0.1798, 0.2504, 0.0833, 0.4457, 0.1695, 0.2974, 0.2259, 0.2146], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0626, 0.0558, 0.0662, 0.0657, 0.0605, 0.0557, 0.0641], + 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') +2023-02-07 10:16:29,178 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 10:16:32,102 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205956.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:16:42,796 INFO [train.py:901] (2/4) Epoch 26, batch 3900, loss[loss=0.1831, simple_loss=0.2738, pruned_loss=0.04614, over 8106.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2828, pruned_loss=0.05753, over 1618633.72 frames. ], batch size: 23, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:17:09,995 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.441e+02 2.892e+02 3.706e+02 7.796e+02, threshold=5.785e+02, percent-clipped=3.0 +2023-02-07 10:17:16,699 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206017.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:17:18,464 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-07 10:17:20,037 INFO [train.py:901] (2/4) Epoch 26, batch 3950, loss[loss=0.1778, simple_loss=0.2702, pruned_loss=0.04265, over 8199.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.05767, over 1616866.24 frames. ], batch size: 23, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:17:53,922 INFO [train.py:901] (2/4) Epoch 26, batch 4000, loss[loss=0.1931, simple_loss=0.2718, pruned_loss=0.05721, over 7975.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2835, pruned_loss=0.05824, over 1617784.58 frames. ], batch size: 21, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:17:55,451 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206074.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:18:18,674 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206106.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:18:19,947 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.407e+02 2.986e+02 3.556e+02 8.558e+02, threshold=5.971e+02, percent-clipped=6.0 +2023-02-07 10:18:23,569 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1376, 1.8770, 3.5457, 1.5460, 2.5654, 3.8979, 4.0415, 3.2945], + device='cuda:2'), covar=tensor([0.1187, 0.1727, 0.0299, 0.2189, 0.0969, 0.0225, 0.0548, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0303, 0.0326, 0.0291, 0.0320, 0.0321, 0.0278, 0.0437, 0.0308], + 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') +2023-02-07 10:18:26,939 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.1113, 1.5054, 6.2129, 2.2127, 5.6939, 5.2859, 5.7740, 5.6112], + device='cuda:2'), covar=tensor([0.0423, 0.5062, 0.0263, 0.3864, 0.0777, 0.0763, 0.0390, 0.0459], + device='cuda:2'), in_proj_covar=tensor([0.0659, 0.0655, 0.0721, 0.0644, 0.0734, 0.0620, 0.0621, 0.0696], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:2') +2023-02-07 10:18:29,517 INFO [train.py:901] (2/4) Epoch 26, batch 4050, loss[loss=0.2367, simple_loss=0.3198, pruned_loss=0.07676, over 8327.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.05883, over 1615415.99 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:18:37,176 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:19:03,800 INFO [train.py:901] (2/4) Epoch 26, batch 4100, loss[loss=0.1679, simple_loss=0.2532, pruned_loss=0.04129, over 7823.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2837, pruned_loss=0.05868, over 1612509.80 frames. ], batch size: 20, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:19:27,817 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4859, 1.8860, 2.9452, 1.3390, 2.1677, 1.8570, 1.5418, 2.2482], + device='cuda:2'), covar=tensor([0.2094, 0.2771, 0.1096, 0.4904, 0.2235, 0.3465, 0.2601, 0.2512], + device='cuda:2'), in_proj_covar=tensor([0.0536, 0.0626, 0.0558, 0.0662, 0.0658, 0.0605, 0.0558, 0.0641], + 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') +2023-02-07 10:19:28,869 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.376e+02 2.755e+02 3.418e+02 9.873e+02, threshold=5.510e+02, percent-clipped=4.0 +2023-02-07 10:19:34,501 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:19:39,255 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206221.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:19:39,709 INFO [train.py:901] (2/4) Epoch 26, batch 4150, loss[loss=0.2229, simple_loss=0.3101, pruned_loss=0.06782, over 8592.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2843, pruned_loss=0.05834, over 1614515.02 frames. ], batch size: 34, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:19:58,907 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1090, 1.1485, 1.4886, 1.1605, 0.7698, 1.2585, 1.1708, 0.8794], + device='cuda:2'), covar=tensor([0.0680, 0.1415, 0.1785, 0.1574, 0.0609, 0.1636, 0.0768, 0.0835], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0100, 0.0162, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 10:20:00,780 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206252.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:20:14,200 INFO [train.py:901] (2/4) Epoch 26, batch 4200, loss[loss=0.2384, simple_loss=0.3328, pruned_loss=0.07199, over 8251.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2848, pruned_loss=0.0585, over 1616166.91 frames. ], batch size: 24, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:20:22,976 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 10:20:35,120 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0009, 1.0197, 0.9915, 1.2379, 0.5889, 0.8808, 0.9519, 1.0422], + device='cuda:2'), covar=tensor([0.0656, 0.0623, 0.0736, 0.0549, 0.0894, 0.1045, 0.0596, 0.0561], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0194, 0.0244, 0.0212, 0.0203, 0.0246, 0.0248, 0.0205], + 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') +2023-02-07 10:20:38,381 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 2.335e+02 2.968e+02 3.755e+02 9.805e+02, threshold=5.936e+02, percent-clipped=3.0 +2023-02-07 10:20:44,912 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 10:20:49,135 INFO [train.py:901] (2/4) Epoch 26, batch 4250, loss[loss=0.2146, simple_loss=0.3012, pruned_loss=0.06397, over 8030.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2846, pruned_loss=0.05874, over 1616329.18 frames. ], batch size: 22, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:21:21,328 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206367.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:24,643 INFO [train.py:901] (2/4) Epoch 26, batch 4300, loss[loss=0.2149, simple_loss=0.2925, pruned_loss=0.06865, over 8658.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2834, pruned_loss=0.05817, over 1615330.29 frames. ], batch size: 39, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:21:35,883 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:50,295 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.331e+02 2.890e+02 3.800e+02 6.492e+02, threshold=5.781e+02, percent-clipped=2.0 +2023-02-07 10:21:53,277 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206413.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:56,607 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206418.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:59,344 INFO [train.py:901] (2/4) Epoch 26, batch 4350, loss[loss=0.1934, simple_loss=0.2901, pruned_loss=0.04839, over 8317.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2818, pruned_loss=0.05764, over 1616198.11 frames. ], batch size: 25, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:22:18,984 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 10:22:20,070 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 +2023-02-07 10:22:20,820 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.92 vs. limit=5.0 +2023-02-07 10:22:32,191 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4004, 4.4304, 3.9741, 2.3107, 3.8589, 4.0161, 3.9296, 3.8116], + device='cuda:2'), covar=tensor([0.0714, 0.0502, 0.1103, 0.3983, 0.0912, 0.0866, 0.1320, 0.0771], + device='cuda:2'), in_proj_covar=tensor([0.0536, 0.0454, 0.0439, 0.0552, 0.0435, 0.0457, 0.0434, 0.0399], + 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') +2023-02-07 10:22:34,782 INFO [train.py:901] (2/4) Epoch 26, batch 4400, loss[loss=0.1677, simple_loss=0.2536, pruned_loss=0.04091, over 7660.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2815, pruned_loss=0.05759, over 1613282.63 frames. ], batch size: 19, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:22:36,353 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3637, 1.5598, 1.5829, 1.1175, 1.6532, 1.2440, 0.3311, 1.5683], + device='cuda:2'), covar=tensor([0.0530, 0.0413, 0.0355, 0.0587, 0.0446, 0.1049, 0.0964, 0.0321], + device='cuda:2'), in_proj_covar=tensor([0.0468, 0.0405, 0.0361, 0.0457, 0.0392, 0.0550, 0.0401, 0.0436], + device='cuda:2'), out_proj_covar=tensor([1.2424e-04, 1.0562e-04, 9.4419e-05, 1.1984e-04, 1.0259e-04, 1.5377e-04, + 1.0723e-04, 1.1472e-04], device='cuda:2') +2023-02-07 10:22:38,332 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206477.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:22:55,674 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206502.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:23:00,161 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.634e+02 2.619e+02 3.000e+02 3.925e+02 8.429e+02, threshold=6.000e+02, percent-clipped=7.0 +2023-02-07 10:23:00,193 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 10:23:04,480 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 +2023-02-07 10:23:08,811 INFO [train.py:901] (2/4) Epoch 26, batch 4450, loss[loss=0.1756, simple_loss=0.2664, pruned_loss=0.04245, over 8244.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2829, pruned_loss=0.05818, over 1612070.60 frames. ], batch size: 24, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:23:16,257 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206533.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:23:34,156 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206559.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:23:44,254 INFO [train.py:901] (2/4) Epoch 26, batch 4500, loss[loss=0.2005, simple_loss=0.2911, pruned_loss=0.05493, over 8473.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2821, pruned_loss=0.05793, over 1612173.20 frames. ], batch size: 25, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:23:55,214 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 10:24:07,620 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7301, 2.3927, 4.8665, 2.9025, 4.4902, 4.2091, 4.5540, 4.4326], + device='cuda:2'), covar=tensor([0.0586, 0.3644, 0.0564, 0.3355, 0.0889, 0.0826, 0.0508, 0.0575], + device='cuda:2'), in_proj_covar=tensor([0.0662, 0.0655, 0.0723, 0.0648, 0.0735, 0.0623, 0.0623, 0.0697], + 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') +2023-02-07 10:24:10,074 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.363e+02 2.961e+02 3.499e+02 6.135e+02, threshold=5.921e+02, percent-clipped=1.0 +2023-02-07 10:24:18,689 INFO [train.py:901] (2/4) Epoch 26, batch 4550, loss[loss=0.206, simple_loss=0.2736, pruned_loss=0.06926, over 7975.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2823, pruned_loss=0.05853, over 1609791.15 frames. ], batch size: 21, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:24:19,497 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206623.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:24:35,787 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206648.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:24:51,524 INFO [train.py:901] (2/4) Epoch 26, batch 4600, loss[loss=0.1515, simple_loss=0.2224, pruned_loss=0.04029, over 7188.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2829, pruned_loss=0.05871, over 1608833.38 frames. ], batch size: 16, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:24:53,053 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206674.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:25:01,034 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5678, 4.5670, 4.1308, 2.1483, 4.0416, 4.1370, 4.0580, 3.9226], + device='cuda:2'), covar=tensor([0.0687, 0.0474, 0.0965, 0.4556, 0.0902, 0.0780, 0.1253, 0.0702], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0452, 0.0438, 0.0550, 0.0434, 0.0456, 0.0434, 0.0399], + 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') +2023-02-07 10:25:18,499 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.342e+02 2.811e+02 3.625e+02 9.770e+02, threshold=5.622e+02, percent-clipped=5.0 +2023-02-07 10:25:21,726 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2023-02-07 10:25:23,271 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 10:25:28,309 INFO [train.py:901] (2/4) Epoch 26, batch 4650, loss[loss=0.2042, simple_loss=0.2937, pruned_loss=0.05734, over 8255.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2829, pruned_loss=0.05829, over 1614274.83 frames. ], batch size: 24, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:25:40,978 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 10:26:02,059 INFO [train.py:901] (2/4) Epoch 26, batch 4700, loss[loss=0.2104, simple_loss=0.2894, pruned_loss=0.06571, over 7926.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2818, pruned_loss=0.05762, over 1616181.83 frames. ], batch size: 20, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:26:08,410 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6100, 1.6890, 1.6832, 1.3806, 1.8008, 1.3956, 0.9312, 1.6478], + device='cuda:2'), covar=tensor([0.0569, 0.0409, 0.0322, 0.0500, 0.0403, 0.0694, 0.0817, 0.0299], + device='cuda:2'), in_proj_covar=tensor([0.0466, 0.0402, 0.0360, 0.0455, 0.0389, 0.0547, 0.0399, 0.0434], + device='cuda:2'), out_proj_covar=tensor([1.2370e-04, 1.0477e-04, 9.4011e-05, 1.1921e-04, 1.0177e-04, 1.5321e-04, + 1.0686e-04, 1.1408e-04], device='cuda:2') +2023-02-07 10:26:13,621 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206789.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:26:28,921 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.506e+02 2.890e+02 3.298e+02 6.611e+02, threshold=5.779e+02, percent-clipped=3.0 +2023-02-07 10:26:32,506 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:26:34,593 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5014, 2.3318, 2.9182, 2.5077, 2.8257, 2.5269, 2.3774, 1.8311], + device='cuda:2'), covar=tensor([0.4698, 0.4505, 0.1769, 0.3553, 0.2462, 0.2973, 0.1750, 0.4910], + device='cuda:2'), in_proj_covar=tensor([0.0952, 0.1003, 0.0824, 0.0979, 0.1014, 0.0917, 0.0764, 0.0838], + 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') +2023-02-07 10:26:37,698 INFO [train.py:901] (2/4) Epoch 26, batch 4750, loss[loss=0.1961, simple_loss=0.2801, pruned_loss=0.05606, over 7967.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2828, pruned_loss=0.05831, over 1605640.91 frames. ], batch size: 21, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:26:53,337 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 10:26:55,380 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 10:26:58,853 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206852.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:27:00,992 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2023-02-07 10:27:12,096 INFO [train.py:901] (2/4) Epoch 26, batch 4800, loss[loss=0.1729, simple_loss=0.2554, pruned_loss=0.04519, over 7689.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2825, pruned_loss=0.05774, over 1607336.47 frames. ], batch size: 18, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:27:21,155 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 10:27:37,298 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.341e+02 2.410e+02 2.886e+02 3.541e+02 7.542e+02, threshold=5.772e+02, percent-clipped=6.0 +2023-02-07 10:27:46,700 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 10:27:47,377 INFO [train.py:901] (2/4) Epoch 26, batch 4850, loss[loss=0.2028, simple_loss=0.2924, pruned_loss=0.05658, over 8195.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2836, pruned_loss=0.05865, over 1608429.06 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:27:52,971 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206930.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:28:10,403 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206955.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:28:21,661 INFO [train.py:901] (2/4) Epoch 26, batch 4900, loss[loss=0.2372, simple_loss=0.3195, pruned_loss=0.07749, over 8237.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2842, pruned_loss=0.05902, over 1612257.86 frames. ], batch size: 24, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:28:25,971 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.38 vs. limit=5.0 +2023-02-07 10:28:46,038 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.535e+02 3.142e+02 3.836e+02 8.051e+02, threshold=6.285e+02, percent-clipped=2.0 +2023-02-07 10:28:55,274 INFO [train.py:901] (2/4) Epoch 26, batch 4950, loss[loss=0.2055, simple_loss=0.2938, pruned_loss=0.05857, over 8188.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05893, over 1612746.85 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:29:04,275 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0907, 1.3405, 1.5166, 1.2417, 0.7668, 1.3232, 1.1703, 0.9117], + device='cuda:2'), covar=tensor([0.0644, 0.1197, 0.1602, 0.1440, 0.0582, 0.1430, 0.0718, 0.0720], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0189, 0.0160, 0.0101, 0.0162, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 10:29:32,338 INFO [train.py:901] (2/4) Epoch 26, batch 5000, loss[loss=0.2017, simple_loss=0.2885, pruned_loss=0.0574, over 8766.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2843, pruned_loss=0.05912, over 1613967.62 frames. ], batch size: 30, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:29:35,480 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 +2023-02-07 10:29:57,376 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.413e+02 2.985e+02 3.933e+02 1.062e+03, threshold=5.970e+02, percent-clipped=3.0 +2023-02-07 10:30:06,448 INFO [train.py:901] (2/4) Epoch 26, batch 5050, loss[loss=0.2151, simple_loss=0.3006, pruned_loss=0.06485, over 8351.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2834, pruned_loss=0.05813, over 1617770.43 frames. ], batch size: 24, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:30:13,946 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1689, 3.5811, 2.2994, 2.8857, 2.8533, 2.0425, 2.8801, 3.0939], + device='cuda:2'), covar=tensor([0.1627, 0.0357, 0.1130, 0.0761, 0.0758, 0.1495, 0.1007, 0.0975], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0240, 0.0341, 0.0313, 0.0302, 0.0348, 0.0349, 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') +2023-02-07 10:30:24,760 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 10:30:42,638 INFO [train.py:901] (2/4) Epoch 26, batch 5100, loss[loss=0.1798, simple_loss=0.2869, pruned_loss=0.03637, over 8261.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2834, pruned_loss=0.058, over 1618518.21 frames. ], batch size: 24, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:30:58,157 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207194.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:30:59,444 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=207196.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:31:08,233 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.087e+02 2.633e+02 3.622e+02 6.552e+02, threshold=5.265e+02, percent-clipped=1.0 +2023-02-07 10:31:16,943 INFO [train.py:901] (2/4) Epoch 26, batch 5150, loss[loss=0.2002, simple_loss=0.2922, pruned_loss=0.05412, over 8463.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.283, pruned_loss=0.05796, over 1615741.69 frames. ], batch size: 27, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:31:48,270 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6106, 2.0603, 3.2679, 1.4975, 2.3830, 2.0747, 1.6378, 2.4903], + device='cuda:2'), covar=tensor([0.2007, 0.2670, 0.0831, 0.4826, 0.2109, 0.3326, 0.2634, 0.2296], + device='cuda:2'), in_proj_covar=tensor([0.0535, 0.0626, 0.0556, 0.0661, 0.0660, 0.0604, 0.0557, 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') +2023-02-07 10:31:52,802 INFO [train.py:901] (2/4) Epoch 26, batch 5200, loss[loss=0.2118, simple_loss=0.2978, pruned_loss=0.06294, over 8333.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2836, pruned_loss=0.0581, over 1618237.74 frames. ], batch size: 26, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:32:17,988 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.585e+02 3.464e+02 4.468e+02 1.375e+03, threshold=6.928e+02, percent-clipped=16.0 +2023-02-07 10:32:19,443 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:32:19,956 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 10:32:26,696 INFO [train.py:901] (2/4) Epoch 26, batch 5250, loss[loss=0.2221, simple_loss=0.303, pruned_loss=0.0706, over 8519.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2835, pruned_loss=0.05844, over 1614608.86 frames. ], batch size: 28, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:32:41,207 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0526, 1.6932, 3.4713, 1.5659, 2.6894, 3.8893, 3.9499, 3.2137], + device='cuda:2'), covar=tensor([0.1186, 0.1895, 0.0364, 0.2147, 0.0963, 0.0252, 0.0527, 0.0633], + device='cuda:2'), in_proj_covar=tensor([0.0298, 0.0322, 0.0286, 0.0314, 0.0314, 0.0273, 0.0429, 0.0301], + 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') +2023-02-07 10:32:52,557 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7333, 1.9039, 1.6396, 2.3291, 0.9699, 1.4992, 1.6934, 1.8492], + device='cuda:2'), covar=tensor([0.0745, 0.0672, 0.0859, 0.0413, 0.1108, 0.1302, 0.0774, 0.0680], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0193, 0.0244, 0.0212, 0.0202, 0.0245, 0.0248, 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') +2023-02-07 10:33:00,346 INFO [train.py:901] (2/4) Epoch 26, batch 5300, loss[loss=0.2134, simple_loss=0.2964, pruned_loss=0.06522, over 8317.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2844, pruned_loss=0.05873, over 1616267.89 frames. ], batch size: 25, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:33:27,782 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.392e+02 2.913e+02 3.782e+02 6.658e+02, threshold=5.826e+02, percent-clipped=0.0 +2023-02-07 10:33:36,850 INFO [train.py:901] (2/4) Epoch 26, batch 5350, loss[loss=0.1682, simple_loss=0.2464, pruned_loss=0.04498, over 7937.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2828, pruned_loss=0.05796, over 1612214.98 frames. ], batch size: 20, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:34:10,287 INFO [train.py:901] (2/4) Epoch 26, batch 5400, loss[loss=0.2257, simple_loss=0.3103, pruned_loss=0.07054, over 8497.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2838, pruned_loss=0.05859, over 1613662.79 frames. ], batch size: 39, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:34:37,323 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.344e+02 3.061e+02 4.157e+02 9.885e+02, threshold=6.122e+02, percent-clipped=8.0 +2023-02-07 10:34:46,156 INFO [train.py:901] (2/4) Epoch 26, batch 5450, loss[loss=0.2114, simple_loss=0.2996, pruned_loss=0.06165, over 8521.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2835, pruned_loss=0.05843, over 1614667.34 frames. ], batch size: 26, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:34:57,844 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=207538.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:35:06,589 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 10:35:17,351 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:35:20,579 INFO [train.py:901] (2/4) Epoch 26, batch 5500, loss[loss=0.2402, simple_loss=0.3192, pruned_loss=0.08063, over 8443.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2847, pruned_loss=0.0594, over 1614527.25 frames. ], batch size: 27, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:35:34,396 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207592.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:35:47,181 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.395e+02 2.975e+02 3.460e+02 7.775e+02, threshold=5.949e+02, percent-clipped=2.0 +2023-02-07 10:35:56,712 INFO [train.py:901] (2/4) Epoch 26, batch 5550, loss[loss=0.21, simple_loss=0.2967, pruned_loss=0.06162, over 7974.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.284, pruned_loss=0.05932, over 1612919.96 frames. ], batch size: 21, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:36:17,865 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:36:30,421 INFO [train.py:901] (2/4) Epoch 26, batch 5600, loss[loss=0.1852, simple_loss=0.2533, pruned_loss=0.05857, over 7785.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2842, pruned_loss=0.05939, over 1619325.77 frames. ], batch size: 19, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:36:55,073 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.412e+02 3.079e+02 3.750e+02 8.490e+02, threshold=6.158e+02, percent-clipped=5.0 +2023-02-07 10:37:04,607 INFO [train.py:901] (2/4) Epoch 26, batch 5650, loss[loss=0.2096, simple_loss=0.2947, pruned_loss=0.06228, over 8123.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2836, pruned_loss=0.0591, over 1613145.09 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:37:12,996 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 10:37:40,860 INFO [train.py:901] (2/4) Epoch 26, batch 5700, loss[loss=0.214, simple_loss=0.2957, pruned_loss=0.06613, over 8506.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.283, pruned_loss=0.05878, over 1607650.91 frames. ], batch size: 26, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:38:05,994 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.232e+02 2.912e+02 3.330e+02 6.698e+02, threshold=5.824e+02, percent-clipped=1.0 +2023-02-07 10:38:14,805 INFO [train.py:901] (2/4) Epoch 26, batch 5750, loss[loss=0.2119, simple_loss=0.295, pruned_loss=0.06439, over 8037.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2839, pruned_loss=0.05907, over 1611909.81 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:38:16,882 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 10:38:50,530 INFO [train.py:901] (2/4) Epoch 26, batch 5800, loss[loss=0.2178, simple_loss=0.3005, pruned_loss=0.06758, over 8332.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.285, pruned_loss=0.05976, over 1611910.87 frames. ], batch size: 25, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:39:15,995 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.614e+02 3.148e+02 4.020e+02 8.026e+02, threshold=6.297e+02, percent-clipped=4.0 +2023-02-07 10:39:16,245 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:39:24,804 INFO [train.py:901] (2/4) Epoch 26, batch 5850, loss[loss=0.1623, simple_loss=0.2591, pruned_loss=0.03277, over 8130.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.05885, over 1616907.42 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:39:32,990 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207934.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:39:59,799 INFO [train.py:901] (2/4) Epoch 26, batch 5900, loss[loss=0.2051, simple_loss=0.2872, pruned_loss=0.06147, over 8097.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2827, pruned_loss=0.0581, over 1617054.09 frames. ], batch size: 21, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:40:15,809 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6965, 1.3185, 2.8248, 1.4997, 2.1750, 3.0183, 3.2061, 2.5528], + device='cuda:2'), covar=tensor([0.1201, 0.1812, 0.0382, 0.1983, 0.0902, 0.0309, 0.0636, 0.0639], + device='cuda:2'), in_proj_covar=tensor([0.0300, 0.0324, 0.0290, 0.0316, 0.0317, 0.0275, 0.0435, 0.0305], + 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') +2023-02-07 10:40:26,842 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.451e+02 2.961e+02 3.656e+02 5.483e+02, threshold=5.923e+02, percent-clipped=0.0 +2023-02-07 10:40:35,640 INFO [train.py:901] (2/4) Epoch 26, batch 5950, loss[loss=0.1756, simple_loss=0.2654, pruned_loss=0.04292, over 8236.00 frames. ], tot_loss[loss=0.2, simple_loss=0.283, pruned_loss=0.05849, over 1618989.42 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:40:54,935 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208050.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:41:09,815 INFO [train.py:901] (2/4) Epoch 26, batch 6000, loss[loss=0.1495, simple_loss=0.2291, pruned_loss=0.03496, over 5549.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2827, pruned_loss=0.05835, over 1609089.27 frames. ], batch size: 12, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:41:09,815 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 10:41:24,450 INFO [train.py:935] (2/4) Epoch 26, validation: loss=0.1721, simple_loss=0.2717, pruned_loss=0.03627, over 944034.00 frames. +2023-02-07 10:41:24,451 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 10:41:32,222 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7469, 1.5718, 2.2838, 1.4797, 1.2155, 2.2324, 0.3921, 1.3838], + device='cuda:2'), covar=tensor([0.1546, 0.1390, 0.0354, 0.1035, 0.2376, 0.0371, 0.1776, 0.1201], + device='cuda:2'), in_proj_covar=tensor([0.0197, 0.0203, 0.0133, 0.0222, 0.0276, 0.0144, 0.0170, 0.0197], + 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') +2023-02-07 10:41:51,019 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.308e+02 2.837e+02 3.630e+02 6.769e+02, threshold=5.675e+02, percent-clipped=2.0 +2023-02-07 10:42:00,859 INFO [train.py:901] (2/4) Epoch 26, batch 6050, loss[loss=0.1583, simple_loss=0.2621, pruned_loss=0.02722, over 8367.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2821, pruned_loss=0.05764, over 1618427.69 frames. ], batch size: 24, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:42:25,847 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2213, 4.2027, 3.7709, 2.0873, 3.6849, 3.7798, 3.7095, 3.5909], + device='cuda:2'), covar=tensor([0.0762, 0.0607, 0.1064, 0.4294, 0.0931, 0.1149, 0.1322, 0.0942], + device='cuda:2'), in_proj_covar=tensor([0.0529, 0.0446, 0.0435, 0.0541, 0.0427, 0.0451, 0.0427, 0.0395], + 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') +2023-02-07 10:42:36,027 INFO [train.py:901] (2/4) Epoch 26, batch 6100, loss[loss=0.2414, simple_loss=0.3173, pruned_loss=0.08271, over 7005.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2806, pruned_loss=0.05712, over 1611616.30 frames. ], batch size: 71, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:42:48,563 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 10:43:01,329 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.422e+02 2.947e+02 3.994e+02 1.088e+03, threshold=5.894e+02, percent-clipped=8.0 +2023-02-07 10:43:06,114 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5691, 4.5165, 4.1055, 1.8967, 3.9896, 4.2190, 4.1190, 3.9988], + device='cuda:2'), covar=tensor([0.0682, 0.0574, 0.1066, 0.5188, 0.0902, 0.0911, 0.1237, 0.0721], + device='cuda:2'), in_proj_covar=tensor([0.0532, 0.0450, 0.0438, 0.0545, 0.0429, 0.0454, 0.0429, 0.0397], + 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') +2023-02-07 10:43:10,773 INFO [train.py:901] (2/4) Epoch 26, batch 6150, loss[loss=0.2211, simple_loss=0.2861, pruned_loss=0.07806, over 7658.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2809, pruned_loss=0.0575, over 1609914.23 frames. ], batch size: 19, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:43:45,586 INFO [train.py:901] (2/4) Epoch 26, batch 6200, loss[loss=0.2491, simple_loss=0.3463, pruned_loss=0.07597, over 8495.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2819, pruned_loss=0.05744, over 1611094.91 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:43:53,058 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208283.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:44:10,220 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 2.196e+02 2.837e+02 3.308e+02 7.178e+02, threshold=5.674e+02, percent-clipped=2.0 +2023-02-07 10:44:19,005 INFO [train.py:901] (2/4) Epoch 26, batch 6250, loss[loss=0.2222, simple_loss=0.3148, pruned_loss=0.06482, over 8319.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.282, pruned_loss=0.05743, over 1613519.10 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:44:55,635 INFO [train.py:901] (2/4) Epoch 26, batch 6300, loss[loss=0.1982, simple_loss=0.2853, pruned_loss=0.05554, over 8241.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2812, pruned_loss=0.05718, over 1610854.41 frames. ], batch size: 22, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:45:10,475 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=208394.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:45:20,542 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.443e+02 2.919e+02 3.618e+02 1.192e+03, threshold=5.838e+02, percent-clipped=3.0 +2023-02-07 10:45:29,138 INFO [train.py:901] (2/4) Epoch 26, batch 6350, loss[loss=0.1603, simple_loss=0.2364, pruned_loss=0.04207, over 7552.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2823, pruned_loss=0.05812, over 1609903.64 frames. ], batch size: 18, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:46:04,930 INFO [train.py:901] (2/4) Epoch 26, batch 6400, loss[loss=0.1939, simple_loss=0.2621, pruned_loss=0.06282, over 7535.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.05799, over 1608565.61 frames. ], batch size: 18, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:46:30,392 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4884, 2.2788, 3.1138, 2.5195, 3.0600, 2.5190, 2.4354, 1.9649], + device='cuda:2'), covar=tensor([0.5616, 0.5598, 0.2140, 0.4126, 0.2873, 0.3433, 0.1958, 0.6297], + device='cuda:2'), in_proj_covar=tensor([0.0962, 0.1013, 0.0831, 0.0985, 0.1022, 0.0923, 0.0770, 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') +2023-02-07 10:46:30,798 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.580e+02 3.188e+02 3.813e+02 6.849e+02, threshold=6.376e+02, percent-clipped=3.0 +2023-02-07 10:46:30,996 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208509.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:46:39,661 INFO [train.py:901] (2/4) Epoch 26, batch 6450, loss[loss=0.1646, simple_loss=0.2487, pruned_loss=0.04029, over 8241.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2825, pruned_loss=0.05831, over 1609309.76 frames. ], batch size: 22, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:47:13,749 INFO [train.py:901] (2/4) Epoch 26, batch 6500, loss[loss=0.2004, simple_loss=0.2994, pruned_loss=0.05071, over 8466.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2835, pruned_loss=0.05837, over 1609905.85 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:47:29,481 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208594.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:47:39,234 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.371e+02 2.869e+02 3.528e+02 8.936e+02, threshold=5.738e+02, percent-clipped=3.0 +2023-02-07 10:47:44,298 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0524, 1.8696, 2.3170, 2.0310, 2.3486, 2.1438, 1.9856, 1.1821], + device='cuda:2'), covar=tensor([0.5849, 0.4933, 0.2061, 0.3754, 0.2434, 0.3109, 0.1980, 0.5477], + device='cuda:2'), in_proj_covar=tensor([0.0968, 0.1019, 0.0835, 0.0991, 0.1025, 0.0928, 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') +2023-02-07 10:47:48,770 INFO [train.py:901] (2/4) Epoch 26, batch 6550, loss[loss=0.206, simple_loss=0.2852, pruned_loss=0.06344, over 8363.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2833, pruned_loss=0.05837, over 1610806.93 frames. ], batch size: 24, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:47:52,197 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=208627.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:47:56,150 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 10:48:12,882 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 10:48:22,496 INFO [train.py:901] (2/4) Epoch 26, batch 6600, loss[loss=0.2177, simple_loss=0.2966, pruned_loss=0.06937, over 8283.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2829, pruned_loss=0.05786, over 1615989.91 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:48:49,205 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.581e+02 2.930e+02 3.571e+02 6.165e+02, threshold=5.859e+02, percent-clipped=2.0 +2023-02-07 10:48:58,727 INFO [train.py:901] (2/4) Epoch 26, batch 6650, loss[loss=0.1995, simple_loss=0.2763, pruned_loss=0.06134, over 7928.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.284, pruned_loss=0.05752, over 1622003.97 frames. ], batch size: 20, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:49:12,762 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208742.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:49:28,861 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208765.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:49:29,598 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2109, 1.9402, 2.4882, 2.1090, 2.5488, 2.2420, 2.1126, 1.4728], + device='cuda:2'), covar=tensor([0.5751, 0.4671, 0.1982, 0.3667, 0.2391, 0.3297, 0.1962, 0.4937], + device='cuda:2'), in_proj_covar=tensor([0.0967, 0.1017, 0.0834, 0.0989, 0.1024, 0.0927, 0.0771, 0.0849], + 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') +2023-02-07 10:49:33,534 INFO [train.py:901] (2/4) Epoch 26, batch 6700, loss[loss=0.1686, simple_loss=0.2584, pruned_loss=0.03937, over 7963.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2837, pruned_loss=0.05793, over 1614471.06 frames. ], batch size: 21, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:49:46,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208790.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:49:59,642 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.530e+02 3.053e+02 4.076e+02 9.744e+02, threshold=6.106e+02, percent-clipped=7.0 +2023-02-07 10:50:03,312 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7518, 1.7933, 1.6606, 2.2586, 1.0634, 1.4756, 1.7699, 1.8124], + device='cuda:2'), covar=tensor([0.0776, 0.0727, 0.0908, 0.0459, 0.1047, 0.1245, 0.0668, 0.0753], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0195, 0.0245, 0.0212, 0.0203, 0.0246, 0.0250, 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') +2023-02-07 10:50:09,370 INFO [train.py:901] (2/4) Epoch 26, batch 6750, loss[loss=0.2214, simple_loss=0.3047, pruned_loss=0.06907, over 7308.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2832, pruned_loss=0.05794, over 1612501.18 frames. ], batch size: 72, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:50:30,915 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 10:50:43,657 INFO [train.py:901] (2/4) Epoch 26, batch 6800, loss[loss=0.1532, simple_loss=0.2347, pruned_loss=0.03581, over 7700.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2828, pruned_loss=0.05795, over 1609092.69 frames. ], batch size: 18, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:50:57,050 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 10:51:08,807 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.376e+02 2.847e+02 3.449e+02 1.016e+03, threshold=5.694e+02, percent-clipped=2.0 +2023-02-07 10:51:18,814 INFO [train.py:901] (2/4) Epoch 26, batch 6850, loss[loss=0.2104, simple_loss=0.3019, pruned_loss=0.05947, over 8283.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2837, pruned_loss=0.05803, over 1614328.54 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:51:19,503 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 10:51:30,530 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=208938.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:51:33,401 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9909, 1.9940, 1.7828, 2.2391, 1.6281, 1.7338, 2.0005, 2.1021], + device='cuda:2'), covar=tensor([0.0613, 0.0716, 0.0778, 0.0613, 0.0860, 0.1038, 0.0642, 0.0631], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0212, 0.0202, 0.0245, 0.0249, 0.0205], + 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') +2023-02-07 10:51:37,687 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1848, 2.0295, 2.5892, 2.1293, 2.6395, 2.2552, 2.1122, 1.5553], + device='cuda:2'), covar=tensor([0.5763, 0.5284, 0.2168, 0.3873, 0.2548, 0.3197, 0.2048, 0.5522], + device='cuda:2'), in_proj_covar=tensor([0.0961, 0.1014, 0.0830, 0.0986, 0.1020, 0.0924, 0.0769, 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') +2023-02-07 10:51:42,427 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3099, 2.1025, 2.6675, 2.1960, 2.6761, 2.4174, 2.2062, 1.6336], + device='cuda:2'), covar=tensor([0.5666, 0.5106, 0.2131, 0.4344, 0.2861, 0.3152, 0.2131, 0.5672], + device='cuda:2'), in_proj_covar=tensor([0.0961, 0.1013, 0.0829, 0.0986, 0.1019, 0.0924, 0.0769, 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') +2023-02-07 10:51:45,311 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 10:51:54,549 INFO [train.py:901] (2/4) Epoch 26, batch 6900, loss[loss=0.2252, simple_loss=0.3058, pruned_loss=0.0723, over 8343.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.285, pruned_loss=0.05877, over 1615999.27 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:52:10,631 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4187, 2.4327, 1.8115, 2.1881, 1.9473, 1.6450, 1.9304, 1.9527], + device='cuda:2'), covar=tensor([0.1470, 0.0410, 0.1372, 0.0594, 0.0784, 0.1606, 0.1010, 0.0900], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0239, 0.0339, 0.0310, 0.0302, 0.0345, 0.0347, 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') +2023-02-07 10:52:12,015 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208998.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:52:12,733 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4888, 2.2827, 2.9938, 2.3972, 2.9756, 2.5026, 2.3736, 1.7625], + device='cuda:2'), covar=tensor([0.5727, 0.5585, 0.2172, 0.4370, 0.2931, 0.3427, 0.1944, 0.6338], + device='cuda:2'), in_proj_covar=tensor([0.0961, 0.1013, 0.0829, 0.0986, 0.1018, 0.0923, 0.0769, 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') +2023-02-07 10:52:20,224 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.430e+02 2.933e+02 3.890e+02 9.541e+02, threshold=5.866e+02, percent-clipped=7.0 +2023-02-07 10:52:23,022 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 10:52:23,726 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2470, 3.1524, 2.9757, 1.5822, 2.8598, 2.8842, 2.8956, 2.8070], + device='cuda:2'), covar=tensor([0.1140, 0.0822, 0.1235, 0.4523, 0.1175, 0.1402, 0.1538, 0.1123], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0456, 0.0442, 0.0552, 0.0436, 0.0460, 0.0436, 0.0402], + 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') +2023-02-07 10:52:28,989 INFO [train.py:901] (2/4) Epoch 26, batch 6950, loss[loss=0.2382, simple_loss=0.3235, pruned_loss=0.07648, over 8105.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.285, pruned_loss=0.05902, over 1617887.30 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:52:29,847 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:52:51,124 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209053.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:52:51,885 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7487, 1.6313, 2.3730, 1.9456, 2.2259, 1.7649, 1.5547, 1.0543], + device='cuda:2'), covar=tensor([0.7499, 0.6374, 0.2241, 0.4555, 0.3203, 0.4820, 0.3220, 0.5855], + device='cuda:2'), in_proj_covar=tensor([0.0961, 0.1013, 0.0829, 0.0986, 0.1017, 0.0922, 0.0770, 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') +2023-02-07 10:53:04,210 INFO [train.py:901] (2/4) Epoch 26, batch 7000, loss[loss=0.153, simple_loss=0.237, pruned_loss=0.03452, over 7429.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05876, over 1615448.06 frames. ], batch size: 17, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:53:30,446 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.543e+02 3.075e+02 4.223e+02 1.225e+03, threshold=6.150e+02, percent-clipped=4.0 +2023-02-07 10:53:38,406 INFO [train.py:901] (2/4) Epoch 26, batch 7050, loss[loss=0.2395, simple_loss=0.3229, pruned_loss=0.07808, over 8624.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2851, pruned_loss=0.05922, over 1617087.70 frames. ], batch size: 39, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:53:44,275 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3603, 1.5922, 2.1385, 1.2645, 1.7315, 1.5652, 1.4005, 1.6726], + device='cuda:2'), covar=tensor([0.1430, 0.2013, 0.0720, 0.3513, 0.1452, 0.2553, 0.1887, 0.2241], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0633, 0.0563, 0.0668, 0.0660, 0.0610, 0.0561, 0.0648], + 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') +2023-02-07 10:54:15,155 INFO [train.py:901] (2/4) Epoch 26, batch 7100, loss[loss=0.1775, simple_loss=0.2684, pruned_loss=0.04329, over 7940.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2834, pruned_loss=0.05896, over 1611312.92 frames. ], batch size: 20, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:54:42,203 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.538e+02 3.057e+02 3.964e+02 1.199e+03, threshold=6.114e+02, percent-clipped=9.0 +2023-02-07 10:54:49,204 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209220.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:54:50,428 INFO [train.py:901] (2/4) Epoch 26, batch 7150, loss[loss=0.2141, simple_loss=0.2977, pruned_loss=0.06522, over 8091.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2832, pruned_loss=0.05802, over 1617030.91 frames. ], batch size: 21, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:55:25,268 INFO [train.py:901] (2/4) Epoch 26, batch 7200, loss[loss=0.1447, simple_loss=0.2265, pruned_loss=0.03145, over 7791.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05874, over 1621679.84 frames. ], batch size: 19, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:55:26,104 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209273.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:55:51,825 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:55:52,275 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.329e+02 2.733e+02 3.562e+02 6.414e+02, threshold=5.467e+02, percent-clipped=2.0 +2023-02-07 10:55:56,507 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209316.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:56:00,325 INFO [train.py:901] (2/4) Epoch 26, batch 7250, loss[loss=0.1839, simple_loss=0.2797, pruned_loss=0.04401, over 8188.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2831, pruned_loss=0.05809, over 1620958.28 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:56:08,449 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209334.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:56:33,552 INFO [train.py:901] (2/4) Epoch 26, batch 7300, loss[loss=0.1844, simple_loss=0.2635, pruned_loss=0.05263, over 7540.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.284, pruned_loss=0.05821, over 1622812.12 frames. ], batch size: 18, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:56:55,996 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1894, 4.1719, 3.8439, 2.1424, 3.7077, 3.8078, 3.7692, 3.6566], + device='cuda:2'), covar=tensor([0.0801, 0.0597, 0.1049, 0.4315, 0.0964, 0.1115, 0.1408, 0.0829], + device='cuda:2'), in_proj_covar=tensor([0.0541, 0.0458, 0.0445, 0.0556, 0.0439, 0.0464, 0.0439, 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') +2023-02-07 10:56:59,843 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.463e+02 3.055e+02 3.934e+02 7.151e+02, threshold=6.111e+02, percent-clipped=5.0 +2023-02-07 10:57:06,739 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 10:57:09,434 INFO [train.py:901] (2/4) Epoch 26, batch 7350, loss[loss=0.1847, simple_loss=0.2805, pruned_loss=0.04444, over 8192.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2847, pruned_loss=0.05923, over 1621034.10 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:57:26,301 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 10:57:42,852 INFO [train.py:901] (2/4) Epoch 26, batch 7400, loss[loss=0.1418, simple_loss=0.2254, pruned_loss=0.02915, over 6313.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2845, pruned_loss=0.05884, over 1623083.56 frames. ], batch size: 14, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:57:51,018 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209484.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:58:04,811 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 10:58:09,479 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.534e+02 3.042e+02 3.812e+02 9.347e+02, threshold=6.084e+02, percent-clipped=5.0 +2023-02-07 10:58:17,134 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8073, 1.7340, 2.5371, 1.4771, 1.3223, 2.5054, 0.4638, 1.5120], + device='cuda:2'), covar=tensor([0.1675, 0.1411, 0.0349, 0.1756, 0.2556, 0.0338, 0.1926, 0.1624], + device='cuda:2'), in_proj_covar=tensor([0.0197, 0.0205, 0.0133, 0.0224, 0.0276, 0.0144, 0.0172, 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') +2023-02-07 10:58:17,541 INFO [train.py:901] (2/4) Epoch 26, batch 7450, loss[loss=0.1877, simple_loss=0.2744, pruned_loss=0.05046, over 8072.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2848, pruned_loss=0.05907, over 1619867.51 frames. ], batch size: 21, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:58:22,477 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209528.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:58:47,409 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209564.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:58:52,744 INFO [train.py:901] (2/4) Epoch 26, batch 7500, loss[loss=0.2026, simple_loss=0.2904, pruned_loss=0.05743, over 8496.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2841, pruned_loss=0.05888, over 1615337.53 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:59:18,692 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.436e+02 2.983e+02 3.503e+02 8.056e+02, threshold=5.967e+02, percent-clipped=5.0 +2023-02-07 10:59:24,167 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209617.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:59:27,341 INFO [train.py:901] (2/4) Epoch 26, batch 7550, loss[loss=0.1986, simple_loss=0.2865, pruned_loss=0.05534, over 8244.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2839, pruned_loss=0.05827, over 1617888.24 frames. ], batch size: 24, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 10:59:28,768 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209624.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:59:54,107 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209660.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:59:55,496 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2746, 3.1708, 2.9436, 1.6649, 2.8606, 2.9535, 2.8165, 2.8063], + device='cuda:2'), covar=tensor([0.1108, 0.0777, 0.1310, 0.4466, 0.1112, 0.1264, 0.1502, 0.1121], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0456, 0.0443, 0.0554, 0.0436, 0.0462, 0.0436, 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') +2023-02-07 11:00:01,941 INFO [train.py:901] (2/4) Epoch 26, batch 7600, loss[loss=0.2186, simple_loss=0.2905, pruned_loss=0.07336, over 7784.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2841, pruned_loss=0.05889, over 1614989.87 frames. ], batch size: 19, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:00:07,010 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209679.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:00:25,111 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209706.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:00:27,671 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.404e+02 2.880e+02 3.478e+02 6.437e+02, threshold=5.761e+02, percent-clipped=3.0 +2023-02-07 11:00:35,725 INFO [train.py:901] (2/4) Epoch 26, batch 7650, loss[loss=0.2233, simple_loss=0.2997, pruned_loss=0.07344, over 8523.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2831, pruned_loss=0.05829, over 1611622.82 frames. ], batch size: 28, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:00:43,090 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209732.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:01:00,269 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.24 vs. limit=5.0 +2023-02-07 11:01:10,730 INFO [train.py:901] (2/4) Epoch 26, batch 7700, loss[loss=0.2135, simple_loss=0.3014, pruned_loss=0.06282, over 8583.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2851, pruned_loss=0.05941, over 1614657.24 frames. ], batch size: 31, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:01:12,912 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209775.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:01:14,129 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 11:01:23,013 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 +2023-02-07 11:01:36,854 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.688e+02 3.083e+02 3.850e+02 9.382e+02, threshold=6.167e+02, percent-clipped=8.0 +2023-02-07 11:01:39,195 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5155, 1.8296, 2.9072, 1.4203, 2.0489, 1.9056, 1.5662, 2.1466], + device='cuda:2'), covar=tensor([0.2068, 0.2580, 0.0886, 0.4777, 0.2116, 0.3370, 0.2530, 0.2347], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0631, 0.0560, 0.0666, 0.0659, 0.0610, 0.0559, 0.0644], + 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') +2023-02-07 11:01:44,890 INFO [train.py:901] (2/4) Epoch 26, batch 7750, loss[loss=0.2276, simple_loss=0.3104, pruned_loss=0.07246, over 8453.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2867, pruned_loss=0.06019, over 1621133.71 frames. ], batch size: 49, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:01:48,919 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209828.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:02:19,758 INFO [train.py:901] (2/4) Epoch 26, batch 7800, loss[loss=0.1408, simple_loss=0.2255, pruned_loss=0.02811, over 7799.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2854, pruned_loss=0.05959, over 1614385.78 frames. ], batch size: 19, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:02:19,818 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:02:44,106 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([6.0770, 1.8418, 6.1625, 2.3051, 5.6404, 5.3197, 5.7438, 5.6807], + device='cuda:2'), covar=tensor([0.0488, 0.5075, 0.0393, 0.3870, 0.0992, 0.0888, 0.0493, 0.0492], + device='cuda:2'), in_proj_covar=tensor([0.0672, 0.0666, 0.0734, 0.0658, 0.0742, 0.0633, 0.0633, 0.0709], + 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') +2023-02-07 11:02:45,320 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.465e+02 2.962e+02 3.430e+02 5.705e+02, threshold=5.924e+02, percent-clipped=0.0 +2023-02-07 11:02:53,248 INFO [train.py:901] (2/4) Epoch 26, batch 7850, loss[loss=0.2022, simple_loss=0.2881, pruned_loss=0.05815, over 8338.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2867, pruned_loss=0.06003, over 1616179.40 frames. ], batch size: 26, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:03:01,886 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209935.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:07,203 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209943.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:18,180 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209960.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:23,279 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209968.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:25,891 INFO [train.py:901] (2/4) Epoch 26, batch 7900, loss[loss=0.2169, simple_loss=0.3066, pruned_loss=0.06358, over 8482.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2857, pruned_loss=0.05962, over 1617006.93 frames. ], batch size: 27, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:03:35,880 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209987.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:36,569 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209988.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:51,903 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.305e+02 2.795e+02 3.387e+02 5.942e+02, threshold=5.591e+02, percent-clipped=1.0 +2023-02-07 11:03:54,070 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210013.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:59,739 INFO [train.py:901] (2/4) Epoch 26, batch 7950, loss[loss=0.1777, simple_loss=0.2606, pruned_loss=0.04737, over 7814.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.05913, over 1615086.38 frames. ], batch size: 20, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:04:06,008 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210031.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:18,574 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=210050.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:22,764 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:33,086 INFO [train.py:901] (2/4) Epoch 26, batch 8000, loss[loss=0.2085, simple_loss=0.2909, pruned_loss=0.06304, over 8456.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2839, pruned_loss=0.05907, over 1614326.14 frames. ], batch size: 27, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:04:35,219 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210075.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:40,668 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210083.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:57,993 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.316e+02 2.819e+02 3.710e+02 9.270e+02, threshold=5.638e+02, percent-clipped=7.0 +2023-02-07 11:05:05,823 INFO [train.py:901] (2/4) Epoch 26, batch 8050, loss[loss=0.196, simple_loss=0.2709, pruned_loss=0.06055, over 7551.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2812, pruned_loss=0.05862, over 1596298.97 frames. ], batch size: 18, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:05:13,568 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5897, 1.7791, 4.7672, 1.8428, 4.3041, 3.9656, 4.2861, 4.2367], + device='cuda:2'), covar=tensor([0.0535, 0.4667, 0.0494, 0.4231, 0.0943, 0.0898, 0.0550, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0672, 0.0666, 0.0734, 0.0657, 0.0742, 0.0632, 0.0634, 0.0709], + 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') +2023-02-07 11:05:15,074 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.94 vs. limit=5.0 +2023-02-07 11:05:37,888 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 11:05:43,120 INFO [train.py:901] (2/4) Epoch 27, batch 0, loss[loss=0.2029, simple_loss=0.2969, pruned_loss=0.05445, over 8499.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2969, pruned_loss=0.05445, over 8499.00 frames. ], batch size: 28, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:05:43,121 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 11:05:54,192 INFO [train.py:935] (2/4) Epoch 27, validation: loss=0.172, simple_loss=0.2713, pruned_loss=0.03628, over 944034.00 frames. +2023-02-07 11:05:54,193 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 11:06:01,185 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:06:08,369 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 11:06:24,781 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210199.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:06:28,658 INFO [train.py:901] (2/4) Epoch 27, batch 50, loss[loss=0.2023, simple_loss=0.2758, pruned_loss=0.06441, over 8236.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2882, pruned_loss=0.059, over 368502.70 frames. ], batch size: 22, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:06:33,573 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.417e+02 2.930e+02 3.516e+02 7.088e+02, threshold=5.860e+02, percent-clipped=5.0 +2023-02-07 11:06:41,911 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 11:06:43,307 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:06:56,876 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:07:04,541 INFO [train.py:901] (2/4) Epoch 27, batch 100, loss[loss=0.1955, simple_loss=0.2731, pruned_loss=0.05892, over 8251.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2875, pruned_loss=0.05945, over 650061.38 frames. ], batch size: 22, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:07:05,004 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 11:07:05,182 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 11:07:13,376 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210268.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:07:38,001 INFO [train.py:901] (2/4) Epoch 27, batch 150, loss[loss=0.1907, simple_loss=0.2816, pruned_loss=0.04983, over 8252.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2848, pruned_loss=0.05838, over 865500.51 frames. ], batch size: 24, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:07:41,156 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.352e+02 2.905e+02 3.661e+02 1.089e+03, threshold=5.811e+02, percent-clipped=3.0 +2023-02-07 11:07:41,641 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 11:08:02,854 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210339.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:08:14,089 INFO [train.py:901] (2/4) Epoch 27, batch 200, loss[loss=0.2333, simple_loss=0.3165, pruned_loss=0.07503, over 8351.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.286, pruned_loss=0.05911, over 1035954.31 frames. ], batch size: 24, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:08:20,481 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210364.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:08:48,346 INFO [train.py:901] (2/4) Epoch 27, batch 250, loss[loss=0.2198, simple_loss=0.3046, pruned_loss=0.06754, over 8107.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.05906, over 1161028.33 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:08:49,371 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 11:08:51,589 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.304e+02 2.819e+02 3.559e+02 6.263e+02, threshold=5.638e+02, percent-clipped=1.0 +2023-02-07 11:08:57,578 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 11:08:57,629 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=210419.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:08:59,063 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210421.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:09:06,468 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 11:09:10,764 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1624, 1.9918, 2.5417, 2.1441, 2.5281, 2.2743, 2.1093, 1.4099], + device='cuda:2'), covar=tensor([0.6128, 0.5176, 0.2186, 0.3906, 0.2747, 0.2988, 0.1982, 0.5703], + device='cuda:2'), in_proj_covar=tensor([0.0965, 0.1017, 0.0828, 0.0985, 0.1021, 0.0926, 0.0772, 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') +2023-02-07 11:09:16,567 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210446.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:09:23,015 INFO [train.py:901] (2/4) Epoch 27, batch 300, loss[loss=0.2118, simple_loss=0.2983, pruned_loss=0.06262, over 8285.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2845, pruned_loss=0.05862, over 1263195.89 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:09:57,338 INFO [train.py:901] (2/4) Epoch 27, batch 350, loss[loss=0.2063, simple_loss=0.2914, pruned_loss=0.06059, over 8520.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2847, pruned_loss=0.05917, over 1342998.80 frames. ], batch size: 28, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:10:00,682 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.321e+02 2.740e+02 3.479e+02 7.751e+02, threshold=5.481e+02, percent-clipped=4.0 +2023-02-07 11:10:06,930 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3549, 1.6905, 1.2909, 2.7675, 1.1900, 1.2518, 1.9702, 1.8591], + device='cuda:2'), covar=tensor([0.1661, 0.1458, 0.2024, 0.0413, 0.1456, 0.2120, 0.1038, 0.1105], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0194, 0.0246, 0.0211, 0.0202, 0.0245, 0.0249, 0.0205], + 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') +2023-02-07 11:10:16,905 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210534.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:10:29,882 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 11:10:30,766 INFO [train.py:901] (2/4) Epoch 27, batch 400, loss[loss=0.2196, simple_loss=0.3099, pruned_loss=0.06465, over 8327.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2848, pruned_loss=0.05875, over 1404963.08 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:11:06,891 INFO [train.py:901] (2/4) Epoch 27, batch 450, loss[loss=0.2056, simple_loss=0.2826, pruned_loss=0.06427, over 8752.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2843, pruned_loss=0.05798, over 1455614.14 frames. ], batch size: 34, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:11:10,232 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.445e+02 3.096e+02 3.744e+02 6.670e+02, threshold=6.192e+02, percent-clipped=5.0 +2023-02-07 11:11:37,199 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210650.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:11:40,168 INFO [train.py:901] (2/4) Epoch 27, batch 500, loss[loss=0.2149, simple_loss=0.3085, pruned_loss=0.06066, over 8515.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2837, pruned_loss=0.05798, over 1488927.82 frames. ], batch size: 28, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:12:01,200 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210684.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:12:12,813 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210700.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:12:15,964 INFO [train.py:901] (2/4) Epoch 27, batch 550, loss[loss=0.2061, simple_loss=0.2803, pruned_loss=0.06591, over 7649.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2842, pruned_loss=0.05792, over 1521863.51 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:12:19,367 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.336e+02 2.792e+02 3.793e+02 8.487e+02, threshold=5.584e+02, percent-clipped=3.0 +2023-02-07 11:12:50,305 INFO [train.py:901] (2/4) Epoch 27, batch 600, loss[loss=0.2052, simple_loss=0.2884, pruned_loss=0.06099, over 8504.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2842, pruned_loss=0.05806, over 1547919.36 frames. ], batch size: 28, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:13:08,465 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210782.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:13:11,592 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 11:13:13,655 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210790.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:13:23,582 INFO [train.py:901] (2/4) Epoch 27, batch 650, loss[loss=0.1636, simple_loss=0.2463, pruned_loss=0.04051, over 7971.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2831, pruned_loss=0.05719, over 1562270.44 frames. ], batch size: 21, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:13:28,267 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.351e+02 2.894e+02 3.474e+02 6.032e+02, threshold=5.788e+02, percent-clipped=3.0 +2023-02-07 11:13:32,523 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210815.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:13:54,801 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2462, 2.0690, 2.6135, 2.1600, 2.5711, 2.2914, 2.1744, 1.4694], + device='cuda:2'), covar=tensor([0.5796, 0.5102, 0.2198, 0.3767, 0.2733, 0.2898, 0.1837, 0.5536], + device='cuda:2'), in_proj_covar=tensor([0.0960, 0.1014, 0.0826, 0.0983, 0.1017, 0.0925, 0.0768, 0.0845], + 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') +2023-02-07 11:13:54,955 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-07 11:13:59,783 INFO [train.py:901] (2/4) Epoch 27, batch 700, loss[loss=0.1954, simple_loss=0.2848, pruned_loss=0.05296, over 8445.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2838, pruned_loss=0.05771, over 1575496.58 frames. ], batch size: 27, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:14:32,723 INFO [train.py:901] (2/4) Epoch 27, batch 750, loss[loss=0.1937, simple_loss=0.2729, pruned_loss=0.05725, over 8083.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2832, pruned_loss=0.05778, over 1583045.51 frames. ], batch size: 21, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:14:34,140 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8278, 1.5207, 1.7218, 1.4199, 1.0882, 1.5239, 1.7237, 1.3324], + device='cuda:2'), covar=tensor([0.0536, 0.1208, 0.1575, 0.1402, 0.0575, 0.1441, 0.0675, 0.0673], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0154, 0.0190, 0.0161, 0.0101, 0.0164, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 11:14:35,973 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.536e+02 2.996e+02 3.960e+02 1.304e+03, threshold=5.993e+02, percent-clipped=7.0 +2023-02-07 11:14:39,933 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 11:14:54,964 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 11:15:04,237 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 11:15:08,843 INFO [train.py:901] (2/4) Epoch 27, batch 800, loss[loss=0.1586, simple_loss=0.2407, pruned_loss=0.03821, over 7789.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2831, pruned_loss=0.0579, over 1592106.59 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:15:35,001 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=210994.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:15:40,378 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211002.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:15:42,219 INFO [train.py:901] (2/4) Epoch 27, batch 850, loss[loss=0.2132, simple_loss=0.3065, pruned_loss=0.05994, over 8291.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2829, pruned_loss=0.0578, over 1597868.95 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:15:45,626 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.265e+02 2.725e+02 3.482e+02 8.151e+02, threshold=5.450e+02, percent-clipped=2.0 +2023-02-07 11:15:57,721 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211028.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:16:05,943 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.18 vs. limit=5.0 +2023-02-07 11:16:10,467 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:16:13,259 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0965, 1.4136, 3.5246, 1.7304, 2.4846, 3.9186, 3.9852, 3.3690], + device='cuda:2'), covar=tensor([0.1126, 0.2015, 0.0343, 0.1897, 0.1029, 0.0249, 0.0562, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0304, 0.0327, 0.0291, 0.0319, 0.0318, 0.0277, 0.0434, 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') +2023-02-07 11:16:17,666 INFO [train.py:901] (2/4) Epoch 27, batch 900, loss[loss=0.1537, simple_loss=0.2381, pruned_loss=0.03468, over 7791.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2837, pruned_loss=0.05842, over 1602436.05 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:16:51,925 INFO [train.py:901] (2/4) Epoch 27, batch 950, loss[loss=0.2134, simple_loss=0.2979, pruned_loss=0.06441, over 8473.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2835, pruned_loss=0.05852, over 1603422.77 frames. ], batch size: 48, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:16:54,832 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211109.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:16:55,242 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.483e+02 2.981e+02 4.008e+02 9.530e+02, threshold=5.961e+02, percent-clipped=10.0 +2023-02-07 11:17:06,152 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211126.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:17:17,582 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211143.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:17:18,072 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 11:17:26,813 INFO [train.py:901] (2/4) Epoch 27, batch 1000, loss[loss=0.1964, simple_loss=0.2735, pruned_loss=0.05968, over 7652.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2837, pruned_loss=0.05875, over 1606182.17 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:17:27,718 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8703, 1.6689, 3.1717, 1.4383, 2.3714, 3.4264, 3.5602, 2.9372], + device='cuda:2'), covar=tensor([0.1202, 0.1645, 0.0373, 0.2172, 0.0882, 0.0296, 0.0652, 0.0590], + device='cuda:2'), in_proj_covar=tensor([0.0306, 0.0327, 0.0293, 0.0321, 0.0320, 0.0278, 0.0436, 0.0308], + 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') +2023-02-07 11:17:30,477 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211159.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:17:53,311 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 11:18:03,302 INFO [train.py:901] (2/4) Epoch 27, batch 1050, loss[loss=0.2064, simple_loss=0.2874, pruned_loss=0.06275, over 8407.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2851, pruned_loss=0.05922, over 1610295.10 frames. ], batch size: 49, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:18:05,253 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 11:18:07,260 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.494e+02 3.070e+02 3.818e+02 8.233e+02, threshold=6.140e+02, percent-clipped=4.0 +2023-02-07 11:18:27,179 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211241.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:18:27,212 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6075, 1.5404, 2.1246, 1.3253, 1.1892, 2.0711, 0.3364, 1.2833], + device='cuda:2'), covar=tensor([0.1447, 0.1124, 0.0341, 0.1013, 0.2316, 0.0445, 0.1801, 0.1142], + device='cuda:2'), in_proj_covar=tensor([0.0197, 0.0203, 0.0133, 0.0221, 0.0275, 0.0145, 0.0171, 0.0197], + 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') +2023-02-07 11:18:36,463 INFO [train.py:901] (2/4) Epoch 27, batch 1100, loss[loss=0.1741, simple_loss=0.2574, pruned_loss=0.04536, over 8240.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2861, pruned_loss=0.05982, over 1606145.45 frames. ], batch size: 22, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:18:48,268 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9521, 3.5979, 2.1896, 2.8157, 2.7239, 2.0049, 2.8807, 3.0185], + device='cuda:2'), covar=tensor([0.1780, 0.0416, 0.1292, 0.0784, 0.0888, 0.1566, 0.1140, 0.1257], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0240, 0.0341, 0.0312, 0.0303, 0.0345, 0.0346, 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') +2023-02-07 11:19:01,144 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0790, 2.2529, 1.7857, 2.8027, 1.4690, 1.5572, 2.1327, 2.2738], + device='cuda:2'), covar=tensor([0.0731, 0.0737, 0.0920, 0.0360, 0.1081, 0.1367, 0.0793, 0.0728], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0195, 0.0246, 0.0212, 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, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-07 11:19:13,342 INFO [train.py:901] (2/4) Epoch 27, batch 1150, loss[loss=0.2431, simple_loss=0.3096, pruned_loss=0.08825, over 6980.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2842, pruned_loss=0.05867, over 1610497.83 frames. ], batch size: 71, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:19:15,964 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 11:19:17,279 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.371e+02 2.782e+02 3.549e+02 6.262e+02, threshold=5.564e+02, percent-clipped=1.0 +2023-02-07 11:19:30,075 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1962, 1.4618, 1.7286, 1.3515, 0.9506, 1.5031, 1.7393, 1.6440], + device='cuda:2'), covar=tensor([0.0483, 0.1290, 0.1715, 0.1529, 0.0613, 0.1478, 0.0720, 0.0649], + 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([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 11:19:40,861 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211346.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:19:46,763 INFO [train.py:901] (2/4) Epoch 27, batch 1200, loss[loss=0.2153, simple_loss=0.3041, pruned_loss=0.06326, over 7975.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2848, pruned_loss=0.05856, over 1616117.20 frames. ], batch size: 21, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:19:53,751 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211365.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:11,163 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:18,555 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211399.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:21,778 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211404.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:22,315 INFO [train.py:901] (2/4) Epoch 27, batch 1250, loss[loss=0.1782, simple_loss=0.271, pruned_loss=0.04272, over 8277.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2838, pruned_loss=0.0583, over 1612791.35 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:20:26,160 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.344e+02 2.922e+02 3.484e+02 6.390e+02, threshold=5.843e+02, percent-clipped=2.0 +2023-02-07 11:20:29,700 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211415.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:35,529 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211424.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:39,478 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211430.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:46,233 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211440.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:56,234 INFO [train.py:901] (2/4) Epoch 27, batch 1300, loss[loss=0.2206, simple_loss=0.3029, pruned_loss=0.0691, over 8105.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2828, pruned_loss=0.05764, over 1612148.41 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:21:00,472 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211461.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:21:24,569 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211497.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:21:30,515 INFO [train.py:901] (2/4) Epoch 27, batch 1350, loss[loss=0.1755, simple_loss=0.2667, pruned_loss=0.0421, over 8513.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2821, pruned_loss=0.05733, over 1609608.55 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:21:34,485 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.433e+02 2.859e+02 3.519e+02 6.900e+02, threshold=5.717e+02, percent-clipped=5.0 +2023-02-07 11:21:43,605 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211522.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:22:05,944 INFO [train.py:901] (2/4) Epoch 27, batch 1400, loss[loss=0.2075, simple_loss=0.2867, pruned_loss=0.06409, over 7930.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2827, pruned_loss=0.05772, over 1614520.03 frames. ], batch size: 20, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:22:39,557 INFO [train.py:901] (2/4) Epoch 27, batch 1450, loss[loss=0.1877, simple_loss=0.2712, pruned_loss=0.05205, over 8137.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2825, pruned_loss=0.0578, over 1612243.93 frames. ], batch size: 22, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:22:43,647 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.739e+02 3.417e+02 5.363e+02 1.739e+03, threshold=6.835e+02, percent-clipped=22.0 +2023-02-07 11:22:45,696 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 11:23:15,887 INFO [train.py:901] (2/4) Epoch 27, batch 1500, loss[loss=0.2142, simple_loss=0.3025, pruned_loss=0.06291, over 8328.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2818, pruned_loss=0.0575, over 1611591.41 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:23:49,651 INFO [train.py:901] (2/4) Epoch 27, batch 1550, loss[loss=0.1795, simple_loss=0.2786, pruned_loss=0.04014, over 8192.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.282, pruned_loss=0.05781, over 1610015.61 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:23:53,681 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.372e+02 3.027e+02 3.476e+02 5.786e+02, threshold=6.054e+02, percent-clipped=0.0 +2023-02-07 11:23:57,833 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211717.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:15,280 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211742.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:19,426 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211748.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:24,773 INFO [train.py:901] (2/4) Epoch 27, batch 1600, loss[loss=0.1974, simple_loss=0.2822, pruned_loss=0.05627, over 8135.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2822, pruned_loss=0.05824, over 1613632.08 frames. ], batch size: 22, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:24:28,338 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0278, 2.3224, 3.7952, 2.1602, 2.0104, 3.8109, 0.7402, 2.2621], + device='cuda:2'), covar=tensor([0.1011, 0.1084, 0.0173, 0.1360, 0.2059, 0.0172, 0.1922, 0.1269], + device='cuda:2'), in_proj_covar=tensor([0.0197, 0.0203, 0.0133, 0.0221, 0.0275, 0.0144, 0.0172, 0.0197], + 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') +2023-02-07 11:24:38,873 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211774.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:59,721 INFO [train.py:901] (2/4) Epoch 27, batch 1650, loss[loss=0.1763, simple_loss=0.2676, pruned_loss=0.04243, over 8481.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2811, pruned_loss=0.05808, over 1610530.54 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:25:03,546 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 11:25:03,802 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.547e+02 3.089e+02 3.889e+02 1.356e+03, threshold=6.177e+02, percent-clipped=3.0 +2023-02-07 11:25:34,306 INFO [train.py:901] (2/4) Epoch 27, batch 1700, loss[loss=0.1825, simple_loss=0.2656, pruned_loss=0.0497, over 7531.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2809, pruned_loss=0.0578, over 1609288.13 frames. ], batch size: 18, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:25:39,902 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211863.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:25:57,119 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8755, 1.4342, 3.5724, 1.5493, 2.4756, 3.9447, 4.0374, 3.3704], + device='cuda:2'), covar=tensor([0.1251, 0.1956, 0.0289, 0.2070, 0.0980, 0.0222, 0.0526, 0.0536], + device='cuda:2'), in_proj_covar=tensor([0.0307, 0.0329, 0.0295, 0.0322, 0.0323, 0.0279, 0.0440, 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') +2023-02-07 11:25:59,112 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211889.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:26:09,561 INFO [train.py:901] (2/4) Epoch 27, batch 1750, loss[loss=0.1802, simple_loss=0.259, pruned_loss=0.05072, over 7957.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.28, pruned_loss=0.05737, over 1603824.02 frames. ], batch size: 21, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:26:13,477 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 2.460e+02 2.972e+02 3.773e+02 5.726e+02, threshold=5.944e+02, percent-clipped=0.0 +2023-02-07 11:26:24,666 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.77 vs. limit=5.0 +2023-02-07 11:26:43,426 INFO [train.py:901] (2/4) Epoch 27, batch 1800, loss[loss=0.1734, simple_loss=0.2558, pruned_loss=0.04551, over 7931.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2806, pruned_loss=0.05736, over 1606051.21 frames. ], batch size: 20, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:27:03,804 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-07 11:27:20,496 INFO [train.py:901] (2/4) Epoch 27, batch 1850, loss[loss=0.1971, simple_loss=0.2698, pruned_loss=0.06216, over 7182.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2803, pruned_loss=0.05715, over 1606832.65 frames. ], batch size: 16, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:27:24,575 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.252e+02 2.767e+02 3.484e+02 5.487e+02, threshold=5.534e+02, percent-clipped=0.0 +2023-02-07 11:27:54,169 INFO [train.py:901] (2/4) Epoch 27, batch 1900, loss[loss=0.2143, simple_loss=0.296, pruned_loss=0.06633, over 8592.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2807, pruned_loss=0.05715, over 1609794.59 frames. ], batch size: 31, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:28:23,913 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8978, 3.8304, 3.5082, 1.8938, 3.4307, 3.4815, 3.3766, 3.3869], + device='cuda:2'), covar=tensor([0.0911, 0.0681, 0.1145, 0.4251, 0.0990, 0.1178, 0.1433, 0.0859], + device='cuda:2'), in_proj_covar=tensor([0.0540, 0.0456, 0.0442, 0.0552, 0.0438, 0.0460, 0.0437, 0.0402], + 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') +2023-02-07 11:28:25,181 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 11:28:28,382 INFO [train.py:901] (2/4) Epoch 27, batch 1950, loss[loss=0.2339, simple_loss=0.3248, pruned_loss=0.07155, over 8430.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2818, pruned_loss=0.05743, over 1610826.63 frames. ], batch size: 49, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:28:33,125 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.526e+02 2.484e+02 3.059e+02 3.727e+02 7.478e+02, threshold=6.119e+02, percent-clipped=3.0 +2023-02-07 11:28:38,414 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 11:28:39,363 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212119.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:28:41,242 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212122.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 11:28:56,776 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212144.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:28:57,275 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 11:28:57,487 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212145.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:29:03,979 INFO [train.py:901] (2/4) Epoch 27, batch 2000, loss[loss=0.2168, simple_loss=0.2995, pruned_loss=0.06703, over 8572.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.282, pruned_loss=0.05736, over 1615712.20 frames. ], batch size: 31, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:29:14,160 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212170.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:29:37,644 INFO [train.py:901] (2/4) Epoch 27, batch 2050, loss[loss=0.2179, simple_loss=0.3045, pruned_loss=0.06567, over 8116.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2816, pruned_loss=0.05749, over 1607476.97 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:29:41,746 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.364e+02 2.966e+02 3.655e+02 9.314e+02, threshold=5.932e+02, percent-clipped=4.0 +2023-02-07 11:29:53,466 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6724, 1.3776, 1.6966, 1.3474, 0.9906, 1.4609, 1.5091, 1.5036], + device='cuda:2'), covar=tensor([0.0589, 0.1284, 0.1647, 0.1522, 0.0569, 0.1436, 0.0683, 0.0650], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0154, 0.0190, 0.0163, 0.0102, 0.0164, 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') +2023-02-07 11:30:13,811 INFO [train.py:901] (2/4) Epoch 27, batch 2100, loss[loss=0.195, simple_loss=0.2769, pruned_loss=0.05651, over 8448.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2819, pruned_loss=0.0577, over 1607434.71 frames. ], batch size: 27, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:30:24,389 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1650, 4.1174, 3.7745, 2.1365, 3.6280, 3.8693, 3.7479, 3.6539], + device='cuda:2'), covar=tensor([0.0871, 0.0607, 0.1118, 0.4289, 0.0974, 0.0915, 0.1286, 0.0761], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0456, 0.0442, 0.0554, 0.0438, 0.0459, 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') +2023-02-07 11:30:25,127 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212271.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:30:42,129 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 +2023-02-07 11:30:47,851 INFO [train.py:901] (2/4) Epoch 27, batch 2150, loss[loss=0.2232, simple_loss=0.3125, pruned_loss=0.06692, over 8584.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2824, pruned_loss=0.05792, over 1608780.77 frames. ], batch size: 34, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:30:48,029 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3256, 2.2670, 1.6507, 2.0497, 1.8752, 1.3885, 1.7281, 1.9148], + device='cuda:2'), covar=tensor([0.1569, 0.0495, 0.1350, 0.0675, 0.0899, 0.1776, 0.1169, 0.1014], + device='cuda:2'), in_proj_covar=tensor([0.0365, 0.0240, 0.0341, 0.0316, 0.0304, 0.0349, 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') +2023-02-07 11:30:51,762 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.395e+02 2.764e+02 3.582e+02 6.444e+02, threshold=5.527e+02, percent-clipped=1.0 +2023-02-07 11:31:22,776 INFO [train.py:901] (2/4) Epoch 27, batch 2200, loss[loss=0.1875, simple_loss=0.2766, pruned_loss=0.04919, over 8506.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2831, pruned_loss=0.05831, over 1611821.48 frames. ], batch size: 28, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:31:57,346 INFO [train.py:901] (2/4) Epoch 27, batch 2250, loss[loss=0.1834, simple_loss=0.2652, pruned_loss=0.0508, over 7226.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2821, pruned_loss=0.05788, over 1612610.48 frames. ], batch size: 16, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:32:01,575 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.301e+02 2.815e+02 3.457e+02 5.141e+02, threshold=5.631e+02, percent-clipped=0.0 +2023-02-07 11:32:31,457 INFO [train.py:901] (2/4) Epoch 27, batch 2300, loss[loss=0.2107, simple_loss=0.2893, pruned_loss=0.06607, over 8552.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2816, pruned_loss=0.05784, over 1614878.09 frames. ], batch size: 34, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:32:33,511 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212457.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:32:39,373 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=212466.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 11:33:07,931 INFO [train.py:901] (2/4) Epoch 27, batch 2350, loss[loss=0.2051, simple_loss=0.3035, pruned_loss=0.05331, over 8456.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2818, pruned_loss=0.05795, over 1616825.20 frames. ], batch size: 27, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:33:12,143 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.366e+02 2.801e+02 3.492e+02 6.818e+02, threshold=5.601e+02, percent-clipped=4.0 +2023-02-07 11:33:42,851 INFO [train.py:901] (2/4) Epoch 27, batch 2400, loss[loss=0.2212, simple_loss=0.2984, pruned_loss=0.07203, over 8296.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2819, pruned_loss=0.0578, over 1616437.76 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:34:01,664 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212581.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 11:34:04,906 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7284, 4.7189, 4.2137, 2.0998, 4.1709, 4.3331, 4.2121, 4.1409], + device='cuda:2'), covar=tensor([0.0676, 0.0516, 0.1074, 0.4410, 0.0892, 0.0848, 0.1181, 0.0754], + device='cuda:2'), in_proj_covar=tensor([0.0542, 0.0460, 0.0444, 0.0558, 0.0440, 0.0464, 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') +2023-02-07 11:34:19,817 INFO [train.py:901] (2/4) Epoch 27, batch 2450, loss[loss=0.1605, simple_loss=0.2397, pruned_loss=0.04067, over 8084.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2809, pruned_loss=0.05748, over 1614097.91 frames. ], batch size: 21, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:34:23,906 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.406e+02 2.879e+02 3.948e+02 9.646e+02, threshold=5.757e+02, percent-clipped=9.0 +2023-02-07 11:34:26,763 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=212615.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:34:46,276 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-07 11:34:54,050 INFO [train.py:901] (2/4) Epoch 27, batch 2500, loss[loss=0.2213, simple_loss=0.3087, pruned_loss=0.06695, over 8565.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2803, pruned_loss=0.05691, over 1613188.39 frames. ], batch size: 31, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:34:59,685 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6558, 1.4962, 3.1701, 1.4199, 2.3780, 3.3827, 3.5626, 2.9078], + device='cuda:2'), covar=tensor([0.1371, 0.1803, 0.0340, 0.2238, 0.0902, 0.0251, 0.0508, 0.0551], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0327, 0.0292, 0.0321, 0.0320, 0.0277, 0.0437, 0.0308], + 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') +2023-02-07 11:35:28,280 INFO [train.py:901] (2/4) Epoch 27, batch 2550, loss[loss=0.1836, simple_loss=0.2735, pruned_loss=0.04685, over 8253.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2802, pruned_loss=0.05662, over 1613776.04 frames. ], batch size: 24, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:35:33,072 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.333e+02 2.985e+02 3.926e+02 7.498e+02, threshold=5.971e+02, percent-clipped=4.0 +2023-02-07 11:35:37,331 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212716.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:35:46,877 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212729.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:35:47,588 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212730.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:36:04,555 INFO [train.py:901] (2/4) Epoch 27, batch 2600, loss[loss=0.1679, simple_loss=0.2462, pruned_loss=0.04484, over 7444.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2798, pruned_loss=0.05654, over 1614075.95 frames. ], batch size: 17, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:36:09,599 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 11:36:18,126 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8686, 1.6982, 3.1288, 1.4403, 2.3582, 3.3664, 3.5325, 2.8503], + device='cuda:2'), covar=tensor([0.1294, 0.1719, 0.0364, 0.2314, 0.0913, 0.0268, 0.0529, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0327, 0.0292, 0.0320, 0.0320, 0.0277, 0.0438, 0.0307], + 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') +2023-02-07 11:36:21,581 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2688, 2.0838, 1.7455, 1.9008, 1.7792, 1.4890, 1.6719, 1.6751], + device='cuda:2'), covar=tensor([0.1486, 0.0503, 0.1250, 0.0593, 0.0773, 0.1587, 0.1071, 0.1051], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0238, 0.0337, 0.0312, 0.0301, 0.0345, 0.0346, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-07 11:36:29,684 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212792.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:36:33,834 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5578, 1.4502, 1.8453, 1.2307, 1.1893, 1.8188, 0.1874, 1.1530], + device='cuda:2'), covar=tensor([0.1410, 0.1134, 0.0384, 0.0854, 0.2255, 0.0438, 0.1868, 0.1242], + device='cuda:2'), in_proj_covar=tensor([0.0198, 0.0205, 0.0134, 0.0223, 0.0276, 0.0144, 0.0171, 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') +2023-02-07 11:36:35,855 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=212801.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:36:38,487 INFO [train.py:901] (2/4) Epoch 27, batch 2650, loss[loss=0.2578, simple_loss=0.3311, pruned_loss=0.09223, over 7227.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2801, pruned_loss=0.05668, over 1612207.34 frames. ], batch size: 71, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:36:43,315 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.517e+02 2.957e+02 3.589e+02 7.428e+02, threshold=5.913e+02, percent-clipped=3.0 +2023-02-07 11:37:02,116 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212837.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 11:37:04,095 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6465, 1.5192, 1.8975, 1.2671, 1.3290, 1.8621, 0.8476, 1.5324], + device='cuda:2'), covar=tensor([0.1360, 0.0913, 0.0373, 0.0837, 0.1674, 0.0418, 0.1580, 0.1214], + device='cuda:2'), in_proj_covar=tensor([0.0198, 0.0205, 0.0134, 0.0223, 0.0276, 0.0145, 0.0172, 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') +2023-02-07 11:37:14,947 INFO [train.py:901] (2/4) Epoch 27, batch 2700, loss[loss=0.2023, simple_loss=0.2838, pruned_loss=0.06041, over 7238.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2809, pruned_loss=0.05733, over 1612549.86 frames. ], batch size: 16, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:37:19,850 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212862.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 11:37:49,369 INFO [train.py:901] (2/4) Epoch 27, batch 2750, loss[loss=0.2161, simple_loss=0.3014, pruned_loss=0.06543, over 8328.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2821, pruned_loss=0.05798, over 1612273.62 frames. ], batch size: 26, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:37:53,341 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.387e+02 2.942e+02 3.576e+02 8.277e+02, threshold=5.883e+02, percent-clipped=4.0 +2023-02-07 11:37:56,795 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212916.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:38:25,583 INFO [train.py:901] (2/4) Epoch 27, batch 2800, loss[loss=0.1796, simple_loss=0.2549, pruned_loss=0.05214, over 7788.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2821, pruned_loss=0.0582, over 1609150.67 frames. ], batch size: 19, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:38:43,388 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7740, 2.1193, 3.6106, 1.7686, 1.8149, 3.5586, 0.6961, 2.1995], + device='cuda:2'), covar=tensor([0.1349, 0.1178, 0.0202, 0.1603, 0.2259, 0.0254, 0.1845, 0.1252], + device='cuda:2'), in_proj_covar=tensor([0.0199, 0.0205, 0.0134, 0.0224, 0.0277, 0.0145, 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') +2023-02-07 11:38:46,066 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212986.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:38:58,529 INFO [train.py:901] (2/4) Epoch 27, batch 2850, loss[loss=0.2965, simple_loss=0.3611, pruned_loss=0.1159, over 8072.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2835, pruned_loss=0.05847, over 1615715.92 frames. ], batch size: 21, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:39:02,632 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 2.420e+02 3.040e+02 3.738e+02 9.771e+02, threshold=6.080e+02, percent-clipped=4.0 +2023-02-07 11:39:02,865 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213011.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:39:03,499 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7798, 1.4854, 1.7277, 1.3868, 0.9755, 1.5128, 1.5982, 1.5031], + device='cuda:2'), covar=tensor([0.0551, 0.1252, 0.1579, 0.1491, 0.0581, 0.1443, 0.0681, 0.0635], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0162, 0.0101, 0.0164, 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') +2023-02-07 11:39:33,419 INFO [train.py:901] (2/4) Epoch 27, batch 2900, loss[loss=0.1855, simple_loss=0.2564, pruned_loss=0.05736, over 7540.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2825, pruned_loss=0.05791, over 1613306.95 frames. ], batch size: 18, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:39:36,853 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213060.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:39:46,823 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213073.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:40:08,833 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.00 vs. limit=5.0 +2023-02-07 11:40:09,153 INFO [train.py:901] (2/4) Epoch 27, batch 2950, loss[loss=0.2075, simple_loss=0.2925, pruned_loss=0.06124, over 7975.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2828, pruned_loss=0.05767, over 1617100.29 frames. ], batch size: 21, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:40:12,524 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 11:40:13,188 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.292e+02 2.734e+02 3.601e+02 6.803e+02, threshold=5.467e+02, percent-clipped=1.0 +2023-02-07 11:40:30,260 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213136.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:40:42,826 INFO [train.py:901] (2/4) Epoch 27, batch 3000, loss[loss=0.1675, simple_loss=0.2481, pruned_loss=0.04348, over 7701.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2831, pruned_loss=0.05805, over 1613629.45 frames. ], batch size: 18, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:40:42,826 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 11:40:56,478 INFO [train.py:935] (2/4) Epoch 27, validation: loss=0.171, simple_loss=0.2706, pruned_loss=0.03572, over 944034.00 frames. +2023-02-07 11:40:56,479 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 11:41:08,330 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213172.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:10,343 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213175.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:19,868 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213188.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:25,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213197.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:30,817 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 11:41:31,705 INFO [train.py:901] (2/4) Epoch 27, batch 3050, loss[loss=0.1915, simple_loss=0.2898, pruned_loss=0.04656, over 8355.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2835, pruned_loss=0.05815, over 1614695.47 frames. ], batch size: 24, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:41:36,535 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.283e+02 2.877e+02 3.649e+02 6.604e+02, threshold=5.754e+02, percent-clipped=7.0 +2023-02-07 11:41:40,805 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5647, 1.2597, 2.3852, 1.3073, 2.2044, 2.5232, 2.7130, 2.1612], + device='cuda:2'), covar=tensor([0.1019, 0.1438, 0.0433, 0.2077, 0.0724, 0.0391, 0.0750, 0.0616], + device='cuda:2'), in_proj_covar=tensor([0.0306, 0.0328, 0.0293, 0.0321, 0.0321, 0.0279, 0.0438, 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') +2023-02-07 11:42:04,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213251.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:42:06,427 INFO [train.py:901] (2/4) Epoch 27, batch 3100, loss[loss=0.2232, simple_loss=0.3017, pruned_loss=0.07229, over 8676.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2844, pruned_loss=0.0588, over 1615510.16 frames. ], batch size: 34, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:42:29,241 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 11:42:40,164 INFO [train.py:901] (2/4) Epoch 27, batch 3150, loss[loss=0.1754, simple_loss=0.2532, pruned_loss=0.04874, over 7414.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2835, pruned_loss=0.05867, over 1613033.82 frames. ], batch size: 17, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:42:44,217 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.557e+02 3.186e+02 3.836e+02 1.080e+03, threshold=6.372e+02, percent-clipped=6.0 +2023-02-07 11:42:57,226 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1037, 1.9552, 2.5078, 2.0802, 2.4879, 2.2245, 2.0583, 1.3660], + device='cuda:2'), covar=tensor([0.5751, 0.4978, 0.2048, 0.4094, 0.2724, 0.3329, 0.1988, 0.5579], + device='cuda:2'), in_proj_covar=tensor([0.0965, 0.1016, 0.0828, 0.0989, 0.1022, 0.0927, 0.0770, 0.0850], + 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') +2023-02-07 11:43:15,298 INFO [train.py:901] (2/4) Epoch 27, batch 3200, loss[loss=0.2093, simple_loss=0.2961, pruned_loss=0.06122, over 8467.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2841, pruned_loss=0.05883, over 1619352.74 frames. ], batch size: 25, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:43:38,614 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5937, 4.6243, 4.1304, 2.0761, 4.0512, 4.3195, 4.0896, 4.1210], + device='cuda:2'), covar=tensor([0.0683, 0.0504, 0.1027, 0.4532, 0.0801, 0.0790, 0.1202, 0.0705], + device='cuda:2'), in_proj_covar=tensor([0.0542, 0.0460, 0.0447, 0.0557, 0.0441, 0.0464, 0.0437, 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') +2023-02-07 11:43:46,356 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2355, 2.1015, 2.6374, 2.2461, 2.6893, 2.3353, 2.1573, 1.5137], + device='cuda:2'), covar=tensor([0.5864, 0.5162, 0.2277, 0.4213, 0.2914, 0.3420, 0.2009, 0.6007], + device='cuda:2'), in_proj_covar=tensor([0.0968, 0.1021, 0.0831, 0.0992, 0.1026, 0.0930, 0.0772, 0.0854], + 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') +2023-02-07 11:43:48,885 INFO [train.py:901] (2/4) Epoch 27, batch 3250, loss[loss=0.1786, simple_loss=0.2525, pruned_loss=0.05231, over 7204.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2824, pruned_loss=0.05801, over 1617858.22 frames. ], batch size: 16, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:43:52,804 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.478e+02 2.885e+02 3.413e+02 5.983e+02, threshold=5.770e+02, percent-clipped=0.0 +2023-02-07 11:44:07,298 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:17,675 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213444.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:25,453 INFO [train.py:901] (2/4) Epoch 27, batch 3300, loss[loss=0.1899, simple_loss=0.2744, pruned_loss=0.05272, over 7818.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2821, pruned_loss=0.05772, over 1618504.58 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:44:26,265 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213456.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:35,238 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213469.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:59,357 INFO [train.py:901] (2/4) Epoch 27, batch 3350, loss[loss=0.1908, simple_loss=0.2679, pruned_loss=0.05683, over 6841.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2812, pruned_loss=0.05712, over 1616169.89 frames. ], batch size: 15, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:45:00,994 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213507.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:45:03,443 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.606e+02 3.102e+02 3.998e+02 8.787e+02, threshold=6.203e+02, percent-clipped=8.0 +2023-02-07 11:45:18,446 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213532.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:45:34,519 INFO [train.py:901] (2/4) Epoch 27, batch 3400, loss[loss=0.1901, simple_loss=0.2816, pruned_loss=0.04926, over 8642.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2812, pruned_loss=0.05715, over 1617164.72 frames. ], batch size: 34, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:45:54,958 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-07 11:45:55,324 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213584.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:46:09,627 INFO [train.py:901] (2/4) Epoch 27, batch 3450, loss[loss=0.2351, simple_loss=0.3176, pruned_loss=0.07624, over 7117.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2823, pruned_loss=0.05769, over 1616187.99 frames. ], batch size: 71, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:46:13,709 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.297e+02 2.616e+02 3.439e+02 9.820e+02, threshold=5.232e+02, percent-clipped=1.0 +2023-02-07 11:46:13,977 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1997, 2.1037, 2.7426, 2.3199, 2.8123, 2.2634, 2.1446, 1.7142], + device='cuda:2'), covar=tensor([0.6043, 0.5051, 0.2118, 0.3945, 0.2512, 0.3174, 0.1969, 0.5328], + device='cuda:2'), in_proj_covar=tensor([0.0964, 0.1019, 0.0829, 0.0988, 0.1019, 0.0929, 0.0770, 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') +2023-02-07 11:46:44,529 INFO [train.py:901] (2/4) Epoch 27, batch 3500, loss[loss=0.2032, simple_loss=0.2911, pruned_loss=0.05762, over 8248.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2821, pruned_loss=0.05754, over 1615340.34 frames. ], batch size: 24, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:47:11,028 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 11:47:20,453 INFO [train.py:901] (2/4) Epoch 27, batch 3550, loss[loss=0.2302, simple_loss=0.3141, pruned_loss=0.07318, over 8263.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2827, pruned_loss=0.0578, over 1616380.68 frames. ], batch size: 24, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:47:24,356 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.484e+02 3.157e+02 3.893e+02 8.912e+02, threshold=6.313e+02, percent-clipped=7.0 +2023-02-07 11:47:55,118 INFO [train.py:901] (2/4) Epoch 27, batch 3600, loss[loss=0.2225, simple_loss=0.3075, pruned_loss=0.0688, over 8471.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2836, pruned_loss=0.05778, over 1617776.97 frames. ], batch size: 25, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:48:31,565 INFO [train.py:901] (2/4) Epoch 27, batch 3650, loss[loss=0.2243, simple_loss=0.3072, pruned_loss=0.0707, over 8636.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.284, pruned_loss=0.05808, over 1621545.61 frames. ], batch size: 34, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:48:35,649 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.435e+02 3.005e+02 4.000e+02 1.001e+03, threshold=6.009e+02, percent-clipped=1.0 +2023-02-07 11:49:05,228 INFO [train.py:901] (2/4) Epoch 27, batch 3700, loss[loss=0.1753, simple_loss=0.2758, pruned_loss=0.03742, over 8629.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2839, pruned_loss=0.05782, over 1618331.01 frames. ], batch size: 39, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:49:11,355 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 11:49:27,322 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 +2023-02-07 11:49:40,755 INFO [train.py:901] (2/4) Epoch 27, batch 3750, loss[loss=0.183, simple_loss=0.2497, pruned_loss=0.05813, over 7703.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2839, pruned_loss=0.05793, over 1621981.80 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:49:44,676 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.240e+02 2.670e+02 3.453e+02 6.024e+02, threshold=5.340e+02, percent-clipped=1.0 +2023-02-07 11:49:57,384 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213928.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:49:58,184 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1473, 2.2716, 2.2624, 1.6912, 2.3860, 1.8535, 1.7053, 2.1330], + device='cuda:2'), covar=tensor([0.0624, 0.0379, 0.0307, 0.0596, 0.0437, 0.0657, 0.0791, 0.0417], + device='cuda:2'), in_proj_covar=tensor([0.0470, 0.0406, 0.0361, 0.0457, 0.0392, 0.0546, 0.0402, 0.0437], + device='cuda:2'), out_proj_covar=tensor([1.2462e-04, 1.0556e-04, 9.4331e-05, 1.1968e-04, 1.0257e-04, 1.5229e-04, + 1.0730e-04, 1.1460e-04], device='cuda:2') +2023-02-07 11:50:15,110 INFO [train.py:901] (2/4) Epoch 27, batch 3800, loss[loss=0.2044, simple_loss=0.2909, pruned_loss=0.05898, over 8294.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2819, pruned_loss=0.05733, over 1615547.34 frames. ], batch size: 23, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:50:29,642 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 11:50:51,696 INFO [train.py:901] (2/4) Epoch 27, batch 3850, loss[loss=0.2373, simple_loss=0.3195, pruned_loss=0.07759, over 8348.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2815, pruned_loss=0.05739, over 1613065.09 frames. ], batch size: 26, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:50:55,678 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.325e+02 2.987e+02 3.815e+02 9.366e+02, threshold=5.974e+02, percent-clipped=6.0 +2023-02-07 11:51:19,434 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214043.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:51:21,264 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 11:51:27,207 INFO [train.py:901] (2/4) Epoch 27, batch 3900, loss[loss=0.2051, simple_loss=0.2951, pruned_loss=0.05754, over 8464.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2817, pruned_loss=0.05766, over 1612418.05 frames. ], batch size: 25, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:52:00,516 INFO [train.py:901] (2/4) Epoch 27, batch 3950, loss[loss=0.1887, simple_loss=0.2716, pruned_loss=0.0529, over 7813.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2825, pruned_loss=0.05753, over 1612456.24 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:52:04,374 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.377e+02 2.717e+02 3.364e+02 5.097e+02, threshold=5.435e+02, percent-clipped=0.0 +2023-02-07 11:52:30,737 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214147.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:52:36,016 INFO [train.py:901] (2/4) Epoch 27, batch 4000, loss[loss=0.1739, simple_loss=0.2601, pruned_loss=0.04391, over 8228.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2825, pruned_loss=0.05746, over 1614910.80 frames. ], batch size: 22, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:52:55,322 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6670, 2.0747, 3.2261, 1.5434, 2.3825, 2.1721, 1.7277, 2.5080], + device='cuda:2'), covar=tensor([0.1850, 0.2805, 0.0881, 0.4691, 0.1935, 0.3098, 0.2512, 0.2264], + device='cuda:2'), in_proj_covar=tensor([0.0535, 0.0630, 0.0562, 0.0665, 0.0656, 0.0604, 0.0559, 0.0642], + 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') +2023-02-07 11:53:10,789 INFO [train.py:901] (2/4) Epoch 27, batch 4050, loss[loss=0.1818, simple_loss=0.2541, pruned_loss=0.05473, over 7695.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2825, pruned_loss=0.05781, over 1615276.81 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:53:14,930 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.379e+02 2.958e+02 3.648e+02 7.596e+02, threshold=5.915e+02, percent-clipped=3.0 +2023-02-07 11:53:31,866 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214236.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:53:45,411 INFO [train.py:901] (2/4) Epoch 27, batch 4100, loss[loss=0.2177, simple_loss=0.3005, pruned_loss=0.06742, over 8625.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2821, pruned_loss=0.05745, over 1613919.68 frames. ], batch size: 34, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:54:00,512 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 11:54:17,261 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214299.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:54:21,105 INFO [train.py:901] (2/4) Epoch 27, batch 4150, loss[loss=0.2135, simple_loss=0.306, pruned_loss=0.06046, over 8363.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.282, pruned_loss=0.05717, over 1612348.70 frames. ], batch size: 24, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:54:23,224 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3890, 4.3736, 3.9709, 1.9887, 3.8764, 4.1308, 3.9201, 3.8920], + device='cuda:2'), covar=tensor([0.0702, 0.0512, 0.0950, 0.4418, 0.0831, 0.0798, 0.1228, 0.0727], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0459, 0.0446, 0.0555, 0.0441, 0.0465, 0.0439, 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') +2023-02-07 11:54:25,184 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.522e+02 2.957e+02 3.518e+02 6.524e+02, threshold=5.913e+02, percent-clipped=2.0 +2023-02-07 11:54:34,234 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:54:55,413 INFO [train.py:901] (2/4) Epoch 27, batch 4200, loss[loss=0.1859, simple_loss=0.2719, pruned_loss=0.04996, over 8084.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2829, pruned_loss=0.05717, over 1614243.18 frames. ], batch size: 21, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:54:56,403 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.87 vs. limit=5.0 +2023-02-07 11:55:10,584 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 +2023-02-07 11:55:15,716 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 11:55:30,769 INFO [train.py:901] (2/4) Epoch 27, batch 4250, loss[loss=0.2027, simple_loss=0.2966, pruned_loss=0.05442, over 8446.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2833, pruned_loss=0.0576, over 1615812.75 frames. ], batch size: 25, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:55:34,789 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.351e+02 2.930e+02 3.605e+02 8.966e+02, threshold=5.860e+02, percent-clipped=4.0 +2023-02-07 11:55:40,122 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 11:56:04,677 INFO [train.py:901] (2/4) Epoch 27, batch 4300, loss[loss=0.2101, simple_loss=0.2975, pruned_loss=0.06132, over 8464.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.283, pruned_loss=0.05771, over 1615254.13 frames. ], batch size: 27, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:56:09,564 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2023-02-07 11:56:15,639 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.08 vs. limit=5.0 +2023-02-07 11:56:18,079 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1388, 1.3791, 1.6689, 1.3324, 0.7448, 1.4613, 1.2191, 1.0769], + device='cuda:2'), covar=tensor([0.0655, 0.1211, 0.1645, 0.1402, 0.0547, 0.1391, 0.0691, 0.0710], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0154, 0.0190, 0.0161, 0.0101, 0.0164, 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') +2023-02-07 11:56:29,278 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=214491.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:56:40,081 INFO [train.py:901] (2/4) Epoch 27, batch 4350, loss[loss=0.2541, simple_loss=0.3248, pruned_loss=0.09168, over 6652.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.05878, over 1609302.35 frames. ], batch size: 71, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:56:44,881 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.457e+02 2.989e+02 4.041e+02 8.697e+02, threshold=5.978e+02, percent-clipped=4.0 +2023-02-07 11:57:11,489 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 11:57:14,883 INFO [train.py:901] (2/4) Epoch 27, batch 4400, loss[loss=0.1954, simple_loss=0.2742, pruned_loss=0.05829, over 8232.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2827, pruned_loss=0.05778, over 1608648.17 frames. ], batch size: 22, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:57:17,080 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2366, 2.4797, 2.5828, 1.5642, 3.0376, 1.8350, 1.5139, 2.3796], + device='cuda:2'), covar=tensor([0.0968, 0.0521, 0.0457, 0.0887, 0.0521, 0.1020, 0.1038, 0.0559], + device='cuda:2'), in_proj_covar=tensor([0.0473, 0.0411, 0.0364, 0.0461, 0.0395, 0.0550, 0.0403, 0.0439], + device='cuda:2'), out_proj_covar=tensor([1.2533e-04, 1.0673e-04, 9.4917e-05, 1.2070e-04, 1.0355e-04, 1.5371e-04, + 1.0771e-04, 1.1520e-04], device='cuda:2') +2023-02-07 11:57:32,006 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=214580.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:57:46,529 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.17 vs. limit=5.0 +2023-02-07 11:57:49,419 INFO [train.py:901] (2/4) Epoch 27, batch 4450, loss[loss=0.3096, simple_loss=0.3648, pruned_loss=0.1272, over 6955.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2821, pruned_loss=0.05728, over 1608460.47 frames. ], batch size: 72, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:57:50,290 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214606.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:57:51,441 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 11:57:52,862 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0423, 2.0580, 1.8638, 2.6079, 1.3199, 1.6344, 2.0220, 2.2306], + device='cuda:2'), covar=tensor([0.0676, 0.0742, 0.0825, 0.0429, 0.1074, 0.1258, 0.0772, 0.0701], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0195, 0.0243, 0.0211, 0.0202, 0.0245, 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') +2023-02-07 11:57:53,336 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.435e+02 2.910e+02 3.675e+02 1.096e+03, threshold=5.821e+02, percent-clipped=3.0 +2023-02-07 11:58:25,056 INFO [train.py:901] (2/4) Epoch 27, batch 4500, loss[loss=0.1708, simple_loss=0.2488, pruned_loss=0.04643, over 7817.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2817, pruned_loss=0.05693, over 1612223.63 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:58:44,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8112, 2.5451, 4.1421, 1.6871, 3.1572, 2.4126, 2.0125, 3.1676], + device='cuda:2'), covar=tensor([0.1971, 0.2642, 0.0934, 0.4716, 0.1862, 0.3271, 0.2520, 0.2353], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0632, 0.0563, 0.0667, 0.0658, 0.0607, 0.0562, 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') +2023-02-07 11:58:49,087 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 11:58:51,959 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214695.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:58:58,564 INFO [train.py:901] (2/4) Epoch 27, batch 4550, loss[loss=0.1769, simple_loss=0.272, pruned_loss=0.04094, over 8698.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2813, pruned_loss=0.05654, over 1617903.50 frames. ], batch size: 39, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:59:03,198 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.415e+02 2.259e+02 2.795e+02 3.667e+02 7.490e+02, threshold=5.591e+02, percent-clipped=6.0 +2023-02-07 11:59:34,572 INFO [train.py:901] (2/4) Epoch 27, batch 4600, loss[loss=0.2267, simple_loss=0.3051, pruned_loss=0.07418, over 6811.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2828, pruned_loss=0.0577, over 1614639.12 frames. ], batch size: 72, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:00:07,854 INFO [train.py:901] (2/4) Epoch 27, batch 4650, loss[loss=0.1688, simple_loss=0.2553, pruned_loss=0.04111, over 8076.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2825, pruned_loss=0.05764, over 1616169.34 frames. ], batch size: 21, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:00:11,913 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.452e+02 3.083e+02 3.974e+02 1.018e+03, threshold=6.165e+02, percent-clipped=5.0 +2023-02-07 12:00:20,129 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0856, 1.8444, 2.3300, 2.0214, 2.3324, 2.1772, 2.0024, 1.2040], + device='cuda:2'), covar=tensor([0.5902, 0.5095, 0.2003, 0.4092, 0.2514, 0.3377, 0.1932, 0.5310], + device='cuda:2'), in_proj_covar=tensor([0.0960, 0.1013, 0.0824, 0.0985, 0.1017, 0.0923, 0.0766, 0.0845], + 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') +2023-02-07 12:00:43,652 INFO [train.py:901] (2/4) Epoch 27, batch 4700, loss[loss=0.2098, simple_loss=0.2797, pruned_loss=0.06996, over 7661.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2829, pruned_loss=0.05804, over 1615804.10 frames. ], batch size: 19, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:00:48,565 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214862.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:01:05,467 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214887.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:01:17,192 INFO [train.py:901] (2/4) Epoch 27, batch 4750, loss[loss=0.2318, simple_loss=0.2926, pruned_loss=0.08552, over 7682.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2826, pruned_loss=0.05797, over 1612227.19 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:01:21,117 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.444e+02 3.016e+02 3.790e+02 1.117e+03, threshold=6.032e+02, percent-clipped=6.0 +2023-02-07 12:01:42,264 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 12:01:45,072 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 12:01:49,024 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214951.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:01:51,303 INFO [train.py:901] (2/4) Epoch 27, batch 4800, loss[loss=0.2722, simple_loss=0.3288, pruned_loss=0.1078, over 7159.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2828, pruned_loss=0.05809, over 1608773.35 frames. ], batch size: 71, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:02:07,902 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:02:27,023 INFO [train.py:901] (2/4) Epoch 27, batch 4850, loss[loss=0.268, simple_loss=0.3453, pruned_loss=0.09533, over 8827.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2837, pruned_loss=0.05847, over 1613132.10 frames. ], batch size: 40, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:02:31,200 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.294e+02 2.705e+02 3.274e+02 6.085e+02, threshold=5.409e+02, percent-clipped=1.0 +2023-02-07 12:02:36,648 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 12:02:47,194 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 12:02:54,782 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-07 12:03:01,819 INFO [train.py:901] (2/4) Epoch 27, batch 4900, loss[loss=0.18, simple_loss=0.2622, pruned_loss=0.04892, over 7701.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05877, over 1613888.42 frames. ], batch size: 18, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:03:37,410 INFO [train.py:901] (2/4) Epoch 27, batch 4950, loss[loss=0.2053, simple_loss=0.2847, pruned_loss=0.06299, over 8344.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2843, pruned_loss=0.05917, over 1608341.53 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:03:41,314 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.323e+02 2.858e+02 3.502e+02 9.819e+02, threshold=5.716e+02, percent-clipped=5.0 +2023-02-07 12:04:10,551 INFO [train.py:901] (2/4) Epoch 27, batch 5000, loss[loss=0.1932, simple_loss=0.2627, pruned_loss=0.06184, over 7704.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2848, pruned_loss=0.05959, over 1600656.53 frames. ], batch size: 18, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:04:47,112 INFO [train.py:901] (2/4) Epoch 27, batch 5050, loss[loss=0.2681, simple_loss=0.3412, pruned_loss=0.09746, over 8776.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2844, pruned_loss=0.05921, over 1601842.42 frames. ], batch size: 30, lr: 2.78e-03, grad_scale: 32.0 +2023-02-07 12:04:50,989 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.415e+02 2.920e+02 3.667e+02 5.760e+02, threshold=5.840e+02, percent-clipped=1.0 +2023-02-07 12:04:53,741 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2886, 1.7647, 1.2730, 2.8102, 1.2619, 1.1881, 2.0103, 1.8334], + device='cuda:2'), covar=tensor([0.1592, 0.1244, 0.1997, 0.0392, 0.1305, 0.2179, 0.0900, 0.1021], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0243, 0.0212, 0.0203, 0.0246, 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') +2023-02-07 12:05:10,151 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 12:05:15,550 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215248.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:05:20,039 INFO [train.py:901] (2/4) Epoch 27, batch 5100, loss[loss=0.2245, simple_loss=0.3126, pruned_loss=0.06822, over 8358.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2861, pruned_loss=0.06026, over 1602816.90 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 32.0 +2023-02-07 12:05:23,562 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0336, 2.1897, 1.9290, 2.8026, 1.4283, 1.6962, 2.1133, 2.2347], + device='cuda:2'), covar=tensor([0.0708, 0.0704, 0.0774, 0.0357, 0.0923, 0.1183, 0.0634, 0.0682], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0243, 0.0212, 0.0203, 0.0246, 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') +2023-02-07 12:05:54,329 INFO [train.py:901] (2/4) Epoch 27, batch 5150, loss[loss=0.1816, simple_loss=0.2612, pruned_loss=0.05094, over 7660.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2842, pruned_loss=0.0593, over 1599328.96 frames. ], batch size: 19, lr: 2.78e-03, grad_scale: 32.0 +2023-02-07 12:05:59,266 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.449e+02 2.868e+02 3.492e+02 6.640e+02, threshold=5.736e+02, percent-clipped=1.0 +2023-02-07 12:06:12,726 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 12:06:22,709 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215343.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:06:28,150 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215351.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:06:30,706 INFO [train.py:901] (2/4) Epoch 27, batch 5200, loss[loss=0.2047, simple_loss=0.2947, pruned_loss=0.0573, over 8354.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.284, pruned_loss=0.05916, over 1604820.75 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:06:59,160 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 +2023-02-07 12:07:05,197 INFO [train.py:901] (2/4) Epoch 27, batch 5250, loss[loss=0.1714, simple_loss=0.2452, pruned_loss=0.04877, over 7443.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2832, pruned_loss=0.05876, over 1606248.93 frames. ], batch size: 17, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:07:09,821 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.358e+02 2.790e+02 3.638e+02 8.125e+02, threshold=5.579e+02, percent-clipped=3.0 +2023-02-07 12:07:12,598 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 12:07:40,340 INFO [train.py:901] (2/4) Epoch 27, batch 5300, loss[loss=0.2035, simple_loss=0.2811, pruned_loss=0.06295, over 7917.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2833, pruned_loss=0.05835, over 1612805.84 frames. ], batch size: 20, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:08:13,769 INFO [train.py:901] (2/4) Epoch 27, batch 5350, loss[loss=0.2129, simple_loss=0.3076, pruned_loss=0.05908, over 8365.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2836, pruned_loss=0.0585, over 1615319.32 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:08:18,669 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.455e+02 2.847e+02 3.988e+02 1.267e+03, threshold=5.693e+02, percent-clipped=12.0 +2023-02-07 12:08:24,722 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 12:08:25,711 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215521.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:08:48,885 INFO [train.py:901] (2/4) Epoch 27, batch 5400, loss[loss=0.1888, simple_loss=0.2814, pruned_loss=0.04813, over 8751.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.283, pruned_loss=0.05798, over 1615706.49 frames. ], batch size: 34, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:08:57,739 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215566.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:09:08,913 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:09:14,843 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215592.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:09:23,343 INFO [train.py:901] (2/4) Epoch 27, batch 5450, loss[loss=0.1957, simple_loss=0.2908, pruned_loss=0.05028, over 8510.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2821, pruned_loss=0.05766, over 1613357.14 frames. ], batch size: 28, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:09:27,901 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.540e+02 3.136e+02 3.819e+02 8.555e+02, threshold=6.272e+02, percent-clipped=5.0 +2023-02-07 12:09:56,882 INFO [train.py:901] (2/4) Epoch 27, batch 5500, loss[loss=0.1761, simple_loss=0.2497, pruned_loss=0.05126, over 7274.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2834, pruned_loss=0.0587, over 1615611.51 frames. ], batch size: 16, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:09:56,921 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 12:10:05,866 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6648, 1.7937, 5.8576, 2.1259, 5.2526, 4.9386, 5.3340, 5.2521], + device='cuda:2'), covar=tensor([0.0570, 0.4553, 0.0384, 0.4016, 0.0949, 0.0818, 0.0539, 0.0534], + device='cuda:2'), in_proj_covar=tensor([0.0674, 0.0664, 0.0733, 0.0654, 0.0741, 0.0632, 0.0633, 0.0711], + 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') +2023-02-07 12:10:20,606 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215687.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:10:25,967 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215695.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:10:32,543 INFO [train.py:901] (2/4) Epoch 27, batch 5550, loss[loss=0.1861, simple_loss=0.2785, pruned_loss=0.04689, over 8740.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2825, pruned_loss=0.05844, over 1612877.51 frames. ], batch size: 30, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:10:34,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215707.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:10:37,243 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.445e+02 2.973e+02 3.969e+02 8.778e+02, threshold=5.947e+02, percent-clipped=4.0 +2023-02-07 12:10:37,654 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 +2023-02-07 12:10:52,200 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7646, 1.5389, 3.1770, 1.4752, 2.3274, 3.3952, 3.5014, 2.9367], + device='cuda:2'), covar=tensor([0.1245, 0.1714, 0.0327, 0.2103, 0.0952, 0.0254, 0.0669, 0.0527], + device='cuda:2'), in_proj_covar=tensor([0.0306, 0.0325, 0.0291, 0.0320, 0.0321, 0.0277, 0.0437, 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') +2023-02-07 12:11:06,807 INFO [train.py:901] (2/4) Epoch 27, batch 5600, loss[loss=0.1755, simple_loss=0.2509, pruned_loss=0.05007, over 7226.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2828, pruned_loss=0.05844, over 1611184.08 frames. ], batch size: 16, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:11:40,023 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215802.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:11:41,824 INFO [train.py:901] (2/4) Epoch 27, batch 5650, loss[loss=0.1952, simple_loss=0.2668, pruned_loss=0.06175, over 7808.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2834, pruned_loss=0.05868, over 1613886.74 frames. ], batch size: 19, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:11:42,241 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 12:11:46,137 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215810.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:11:47,203 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.361e+02 2.799e+02 3.308e+02 5.877e+02, threshold=5.598e+02, percent-clipped=0.0 +2023-02-07 12:12:05,002 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 12:12:11,226 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215847.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:12:16,647 INFO [train.py:901] (2/4) Epoch 27, batch 5700, loss[loss=0.203, simple_loss=0.283, pruned_loss=0.06151, over 8522.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05878, over 1612256.52 frames. ], batch size: 28, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:12:23,685 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:12:51,905 INFO [train.py:901] (2/4) Epoch 27, batch 5750, loss[loss=0.1966, simple_loss=0.2845, pruned_loss=0.05437, over 8565.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2825, pruned_loss=0.05817, over 1606207.09 frames. ], batch size: 31, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:12:56,019 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215910.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:12:57,194 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.395e+02 2.899e+02 3.864e+02 7.116e+02, threshold=5.798e+02, percent-clipped=7.0 +2023-02-07 12:13:07,809 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215927.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:13:10,342 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 12:13:26,396 INFO [train.py:901] (2/4) Epoch 27, batch 5800, loss[loss=0.2123, simple_loss=0.2831, pruned_loss=0.07072, over 7205.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2828, pruned_loss=0.05863, over 1606351.21 frames. ], batch size: 16, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:13:31,935 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215963.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:13:32,842 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-07 12:13:40,545 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1721, 1.6426, 1.6776, 1.4672, 1.0529, 1.5035, 1.7880, 1.7204], + device='cuda:2'), covar=tensor([0.0563, 0.1180, 0.1629, 0.1446, 0.0619, 0.1495, 0.0744, 0.0631], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0152, 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') +2023-02-07 12:13:43,178 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215980.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:13:47,613 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6940, 1.6367, 2.3268, 1.2768, 1.1974, 2.3738, 0.3921, 1.3701], + device='cuda:2'), covar=tensor([0.1723, 0.1161, 0.0391, 0.1376, 0.2471, 0.0409, 0.1888, 0.1347], + device='cuda:2'), in_proj_covar=tensor([0.0199, 0.0203, 0.0135, 0.0222, 0.0275, 0.0145, 0.0170, 0.0197], + 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') +2023-02-07 12:13:48,929 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215988.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:14:01,068 INFO [train.py:901] (2/4) Epoch 27, batch 5850, loss[loss=0.1571, simple_loss=0.2332, pruned_loss=0.04053, over 7524.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2833, pruned_loss=0.05824, over 1607280.72 frames. ], batch size: 18, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:14:05,651 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.506e+02 2.428e+02 2.871e+02 3.760e+02 7.078e+02, threshold=5.742e+02, percent-clipped=9.0 +2023-02-07 12:14:15,358 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216025.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:14:27,446 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216042.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:14:30,139 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216046.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:14:36,738 INFO [train.py:901] (2/4) Epoch 27, batch 5900, loss[loss=0.2024, simple_loss=0.2777, pruned_loss=0.06359, over 7963.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2831, pruned_loss=0.05845, over 1606322.00 frames. ], batch size: 21, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:14:38,967 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216058.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:14:44,154 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216066.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:14:53,613 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 12:14:55,481 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216083.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:15:00,978 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:15:10,147 INFO [train.py:901] (2/4) Epoch 27, batch 5950, loss[loss=0.1771, simple_loss=0.2604, pruned_loss=0.04693, over 7780.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2829, pruned_loss=0.05803, over 1605663.08 frames. ], batch size: 19, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:15:15,769 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.411e+02 2.864e+02 3.625e+02 8.908e+02, threshold=5.728e+02, percent-clipped=5.0 +2023-02-07 12:15:46,872 INFO [train.py:901] (2/4) Epoch 27, batch 6000, loss[loss=0.1937, simple_loss=0.294, pruned_loss=0.04676, over 8534.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2839, pruned_loss=0.0587, over 1600053.66 frames. ], batch size: 28, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:15:46,873 INFO [train.py:926] (2/4) Computing validation loss +2023-02-07 12:15:59,961 INFO [train.py:935] (2/4) Epoch 27, validation: loss=0.1711, simple_loss=0.2711, pruned_loss=0.03554, over 944034.00 frames. +2023-02-07 12:15:59,962 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6789MB +2023-02-07 12:16:25,736 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=216191.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:16:35,311 INFO [train.py:901] (2/4) Epoch 27, batch 6050, loss[loss=0.1946, simple_loss=0.2924, pruned_loss=0.04841, over 8527.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2833, pruned_loss=0.05836, over 1603854.92 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:16:40,126 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.565e+02 3.207e+02 4.227e+02 9.285e+02, threshold=6.415e+02, percent-clipped=9.0 +2023-02-07 12:16:56,671 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216236.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:09,862 INFO [train.py:901] (2/4) Epoch 27, batch 6100, loss[loss=0.1949, simple_loss=0.2844, pruned_loss=0.05268, over 8461.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2834, pruned_loss=0.05833, over 1600886.96 frames. ], batch size: 39, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:17:14,015 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216261.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:28,455 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216281.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:17:39,621 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216298.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:44,965 INFO [train.py:901] (2/4) Epoch 27, batch 6150, loss[loss=0.201, simple_loss=0.2933, pruned_loss=0.05436, over 8579.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2827, pruned_loss=0.05783, over 1604102.04 frames. ], batch size: 31, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:17:44,980 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 12:17:45,827 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216306.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:45,850 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216306.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:17:49,762 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.311e+02 2.985e+02 4.036e+02 8.594e+02, threshold=5.970e+02, percent-clipped=2.0 +2023-02-07 12:17:57,192 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216323.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:18:18,513 INFO [train.py:901] (2/4) Epoch 27, batch 6200, loss[loss=0.1991, simple_loss=0.2575, pruned_loss=0.07032, over 7707.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2828, pruned_loss=0.05848, over 1604092.00 frames. ], batch size: 18, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:18:28,740 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 12:18:42,476 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=216390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:18:45,887 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5504, 1.4924, 1.8220, 1.2627, 1.2519, 1.8385, 0.2556, 1.2399], + device='cuda:2'), covar=tensor([0.1560, 0.1229, 0.0442, 0.0849, 0.2253, 0.0450, 0.1830, 0.1198], + device='cuda:2'), in_proj_covar=tensor([0.0200, 0.0205, 0.0136, 0.0222, 0.0275, 0.0145, 0.0171, 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') +2023-02-07 12:18:53,897 INFO [train.py:901] (2/4) Epoch 27, batch 6250, loss[loss=0.1978, simple_loss=0.2777, pruned_loss=0.05893, over 8035.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2824, pruned_loss=0.05836, over 1604751.72 frames. ], batch size: 22, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:18:58,452 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.470e+02 2.901e+02 3.405e+02 7.374e+02, threshold=5.803e+02, percent-clipped=1.0 +2023-02-07 12:19:27,775 INFO [train.py:901] (2/4) Epoch 27, batch 6300, loss[loss=0.1892, simple_loss=0.2853, pruned_loss=0.04652, over 8372.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2827, pruned_loss=0.05818, over 1609013.10 frames. ], batch size: 24, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:19:54,184 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6427, 1.7611, 2.0092, 1.6598, 1.1999, 1.7759, 2.3399, 1.9412], + device='cuda:2'), covar=tensor([0.0489, 0.1196, 0.1623, 0.1397, 0.0578, 0.1454, 0.0623, 0.0636], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0152, 0.0190, 0.0161, 0.0101, 0.0163, 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') +2023-02-07 12:19:56,913 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.18 vs. limit=5.0 +2023-02-07 12:20:01,964 INFO [train.py:901] (2/4) Epoch 27, batch 6350, loss[loss=0.2432, simple_loss=0.3116, pruned_loss=0.08743, over 8444.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2828, pruned_loss=0.05832, over 1610481.62 frames. ], batch size: 49, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:20:02,156 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216505.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:20:07,863 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.511e+02 2.994e+02 4.018e+02 7.521e+02, threshold=5.987e+02, percent-clipped=5.0 +2023-02-07 12:20:23,600 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.04 vs. limit=5.0 +2023-02-07 12:20:30,861 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2790, 1.6979, 4.5983, 1.9956, 2.6238, 5.1436, 5.2682, 4.4625], + device='cuda:2'), covar=tensor([0.1236, 0.1823, 0.0225, 0.1969, 0.1013, 0.0157, 0.0566, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0325, 0.0291, 0.0318, 0.0321, 0.0276, 0.0437, 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') +2023-02-07 12:20:36,660 INFO [train.py:901] (2/4) Epoch 27, batch 6400, loss[loss=0.1993, simple_loss=0.2872, pruned_loss=0.05571, over 8540.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2833, pruned_loss=0.05871, over 1609262.23 frames. ], batch size: 39, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:20:41,612 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216562.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:20:53,520 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0527, 2.1655, 1.9895, 2.9651, 1.3978, 1.6519, 2.1851, 2.2349], + device='cuda:2'), covar=tensor([0.0764, 0.0791, 0.0794, 0.0314, 0.1039, 0.1278, 0.0789, 0.0782], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0194, 0.0243, 0.0211, 0.0203, 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') +2023-02-07 12:20:58,196 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216587.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:21:10,669 INFO [train.py:901] (2/4) Epoch 27, batch 6450, loss[loss=0.1828, simple_loss=0.2707, pruned_loss=0.0475, over 7645.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2822, pruned_loss=0.05793, over 1610926.13 frames. ], batch size: 19, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:21:13,554 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216609.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:21:16,162 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.469e+02 2.882e+02 3.609e+02 7.919e+02, threshold=5.765e+02, percent-clipped=2.0 +2023-02-07 12:21:46,015 INFO [train.py:901] (2/4) Epoch 27, batch 6500, loss[loss=0.1845, simple_loss=0.2701, pruned_loss=0.04938, over 8078.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2828, pruned_loss=0.05836, over 1613014.27 frames. ], batch size: 21, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:21:58,517 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216673.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:22:19,973 INFO [train.py:901] (2/4) Epoch 27, batch 6550, loss[loss=0.2397, simple_loss=0.3148, pruned_loss=0.08228, over 8192.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2819, pruned_loss=0.05821, over 1611587.14 frames. ], batch size: 23, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:22:25,118 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.544e+02 2.876e+02 3.743e+02 6.730e+02, threshold=5.752e+02, percent-clipped=5.0 +2023-02-07 12:22:32,762 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:22:48,281 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.97 vs. limit=5.0 +2023-02-07 12:22:52,673 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9658, 1.5147, 1.7150, 1.4521, 1.0518, 1.4654, 1.7742, 1.5415], + device='cuda:2'), covar=tensor([0.0583, 0.1314, 0.1695, 0.1483, 0.0621, 0.1547, 0.0728, 0.0674], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0101, 0.0163, 0.0113, 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') +2023-02-07 12:22:54,583 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 12:22:55,719 INFO [train.py:901] (2/4) Epoch 27, batch 6600, loss[loss=0.2858, simple_loss=0.3548, pruned_loss=0.1084, over 8464.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2825, pruned_loss=0.05825, over 1615227.13 frames. ], batch size: 29, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:22:59,346 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2623, 2.0605, 2.6827, 2.2340, 2.6531, 2.3124, 2.1308, 1.5074], + device='cuda:2'), covar=tensor([0.5792, 0.5447, 0.2142, 0.3817, 0.2611, 0.3295, 0.1932, 0.5702], + device='cuda:2'), in_proj_covar=tensor([0.0966, 0.1021, 0.0832, 0.0992, 0.1029, 0.0930, 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') +2023-02-07 12:23:00,003 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:23:12,957 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 12:23:17,025 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216786.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:23:29,548 INFO [train.py:901] (2/4) Epoch 27, batch 6650, loss[loss=0.1817, simple_loss=0.2697, pruned_loss=0.04681, over 8187.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2822, pruned_loss=0.05804, over 1614579.45 frames. ], batch size: 23, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:23:34,796 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.571e+02 3.099e+02 3.859e+02 9.745e+02, threshold=6.199e+02, percent-clipped=7.0 +2023-02-07 12:24:03,776 INFO [train.py:901] (2/4) Epoch 27, batch 6700, loss[loss=0.2423, simple_loss=0.3101, pruned_loss=0.08726, over 8504.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.283, pruned_loss=0.05831, over 1615693.34 frames. ], batch size: 26, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:24:38,627 INFO [train.py:901] (2/4) Epoch 27, batch 6750, loss[loss=0.2026, simple_loss=0.2749, pruned_loss=0.06519, over 7781.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2842, pruned_loss=0.05891, over 1620436.02 frames. ], batch size: 19, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:24:42,858 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1276, 2.4173, 2.5823, 1.5645, 2.8743, 1.7417, 1.5572, 2.2161], + device='cuda:2'), covar=tensor([0.0909, 0.0468, 0.0399, 0.0912, 0.0487, 0.0998, 0.1049, 0.0602], + device='cuda:2'), in_proj_covar=tensor([0.0470, 0.0409, 0.0362, 0.0457, 0.0393, 0.0552, 0.0401, 0.0438], + device='cuda:2'), out_proj_covar=tensor([1.2446e-04, 1.0608e-04, 9.4260e-05, 1.1943e-04, 1.0299e-04, 1.5394e-04, + 1.0712e-04, 1.1497e-04], device='cuda:2') +2023-02-07 12:24:43,899 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.479e+02 3.006e+02 3.687e+02 6.813e+02, threshold=6.012e+02, percent-clipped=1.0 +2023-02-07 12:25:05,485 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 +2023-02-07 12:25:11,169 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=216953.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:25:12,353 INFO [train.py:901] (2/4) Epoch 27, batch 6800, loss[loss=0.1904, simple_loss=0.2838, pruned_loss=0.04853, over 7913.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2852, pruned_loss=0.05908, over 1624280.75 frames. ], batch size: 20, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:25:24,096 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 12:25:32,325 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216983.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:25:35,736 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 12:25:47,373 INFO [train.py:901] (2/4) Epoch 27, batch 6850, loss[loss=0.2465, simple_loss=0.3138, pruned_loss=0.08959, over 7977.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2852, pruned_loss=0.05928, over 1621012.21 frames. ], batch size: 21, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:25:52,582 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.408e+02 3.097e+02 3.751e+02 9.876e+02, threshold=6.193e+02, percent-clipped=4.0 +2023-02-07 12:25:55,361 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217017.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:26:11,011 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 12:26:21,085 INFO [train.py:901] (2/4) Epoch 27, batch 6900, loss[loss=0.2029, simple_loss=0.2737, pruned_loss=0.06599, over 7788.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2849, pruned_loss=0.0593, over 1614389.03 frames. ], batch size: 19, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:26:29,751 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217067.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:26:30,513 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217068.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:26:56,816 INFO [train.py:901] (2/4) Epoch 27, batch 6950, loss[loss=0.156, simple_loss=0.2306, pruned_loss=0.04066, over 7717.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2846, pruned_loss=0.05899, over 1616603.18 frames. ], batch size: 18, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:27:02,040 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.303e+02 2.670e+02 3.410e+02 6.861e+02, threshold=5.340e+02, percent-clipped=1.0 +2023-02-07 12:27:15,659 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217132.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:27:19,475 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 12:27:30,833 INFO [train.py:901] (2/4) Epoch 27, batch 7000, loss[loss=0.1769, simple_loss=0.2755, pruned_loss=0.03919, over 8336.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2829, pruned_loss=0.05838, over 1610804.23 frames. ], batch size: 25, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:27:49,474 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217182.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:28:05,137 INFO [train.py:901] (2/4) Epoch 27, batch 7050, loss[loss=0.2149, simple_loss=0.2868, pruned_loss=0.07147, over 8188.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2833, pruned_loss=0.05852, over 1607767.08 frames. ], batch size: 23, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:28:11,287 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.356e+02 3.046e+02 3.591e+02 8.726e+02, threshold=6.092e+02, percent-clipped=6.0 +2023-02-07 12:28:19,539 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8666, 2.0885, 2.1890, 1.3235, 2.3442, 1.6562, 0.7879, 1.9979], + device='cuda:2'), covar=tensor([0.0731, 0.0469, 0.0334, 0.0744, 0.0442, 0.1044, 0.1074, 0.0388], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0408, 0.0363, 0.0458, 0.0393, 0.0550, 0.0401, 0.0439], + device='cuda:2'), out_proj_covar=tensor([1.2496e-04, 1.0575e-04, 9.4613e-05, 1.1965e-04, 1.0285e-04, 1.5345e-04, + 1.0704e-04, 1.1515e-04], device='cuda:2') +2023-02-07 12:28:40,066 INFO [train.py:901] (2/4) Epoch 27, batch 7100, loss[loss=0.2253, simple_loss=0.3006, pruned_loss=0.07503, over 8462.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2825, pruned_loss=0.05804, over 1607580.68 frames. ], batch size: 27, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:29:14,491 INFO [train.py:901] (2/4) Epoch 27, batch 7150, loss[loss=0.1901, simple_loss=0.2754, pruned_loss=0.05237, over 7972.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.0577, over 1610873.58 frames. ], batch size: 21, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:29:18,601 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5246, 2.0023, 3.4181, 1.5097, 1.4826, 3.3994, 0.7243, 1.9824], + device='cuda:2'), covar=tensor([0.1588, 0.1385, 0.0317, 0.1930, 0.2618, 0.0415, 0.1948, 0.1461], + device='cuda:2'), in_proj_covar=tensor([0.0199, 0.0204, 0.0136, 0.0222, 0.0274, 0.0146, 0.0171, 0.0197], + 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') +2023-02-07 12:29:19,666 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.444e+02 3.123e+02 4.113e+02 1.134e+03, threshold=6.246e+02, percent-clipped=7.0 +2023-02-07 12:29:27,891 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:29:29,713 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:29:45,931 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217349.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:29:49,818 INFO [train.py:901] (2/4) Epoch 27, batch 7200, loss[loss=0.2097, simple_loss=0.3106, pruned_loss=0.05441, over 8325.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2834, pruned_loss=0.05772, over 1618340.88 frames. ], batch size: 25, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:29:50,710 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2177, 3.4381, 2.1467, 2.9346, 2.9441, 1.8756, 2.9450, 2.9430], + device='cuda:2'), covar=tensor([0.1686, 0.0410, 0.1304, 0.0770, 0.0724, 0.1606, 0.0988, 0.1170], + device='cuda:2'), in_proj_covar=tensor([0.0363, 0.0244, 0.0345, 0.0316, 0.0306, 0.0350, 0.0353, 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') +2023-02-07 12:30:11,979 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217388.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:30:23,036 INFO [train.py:901] (2/4) Epoch 27, batch 7250, loss[loss=0.2142, simple_loss=0.2735, pruned_loss=0.07745, over 7254.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2837, pruned_loss=0.05747, over 1617038.80 frames. ], batch size: 16, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:30:23,855 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:30:28,402 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.296e+02 2.784e+02 3.610e+02 7.832e+02, threshold=5.568e+02, percent-clipped=2.0 +2023-02-07 12:30:28,617 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217413.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:30:33,321 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1955, 1.0299, 1.2979, 1.0271, 0.9839, 1.3109, 0.1357, 0.9221], + device='cuda:2'), covar=tensor([0.1543, 0.1321, 0.0471, 0.0672, 0.2286, 0.0511, 0.1875, 0.1122], + device='cuda:2'), in_proj_covar=tensor([0.0199, 0.0205, 0.0136, 0.0223, 0.0275, 0.0146, 0.0171, 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') +2023-02-07 12:30:36,047 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1757, 1.2624, 1.5405, 1.2184, 0.7392, 1.3678, 1.0937, 0.9476], + device='cuda:2'), covar=tensor([0.0652, 0.1216, 0.1668, 0.1550, 0.0601, 0.1417, 0.0757, 0.0742], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0161, 0.0101, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 12:30:45,934 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217438.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:30:49,266 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217442.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:30:58,439 INFO [train.py:901] (2/4) Epoch 27, batch 7300, loss[loss=0.1816, simple_loss=0.2672, pruned_loss=0.048, over 7979.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2841, pruned_loss=0.05795, over 1617872.08 frames. ], batch size: 21, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:31:04,044 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217463.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:31:20,395 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3411, 1.5580, 4.6308, 2.1497, 2.5144, 5.1828, 5.2731, 4.5456], + device='cuda:2'), covar=tensor([0.1183, 0.1954, 0.0215, 0.1818, 0.1176, 0.0167, 0.0394, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0307, 0.0324, 0.0292, 0.0318, 0.0322, 0.0277, 0.0439, 0.0307], + 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') +2023-02-07 12:31:33,102 INFO [train.py:901] (2/4) Epoch 27, batch 7350, loss[loss=0.2026, simple_loss=0.299, pruned_loss=0.05314, over 8331.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2841, pruned_loss=0.05778, over 1617837.72 frames. ], batch size: 25, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:31:36,629 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217510.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:31:38,492 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.322e+02 2.888e+02 3.768e+02 6.651e+02, threshold=5.777e+02, percent-clipped=4.0 +2023-02-07 12:31:59,836 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 12:32:07,059 INFO [train.py:901] (2/4) Epoch 27, batch 7400, loss[loss=0.1913, simple_loss=0.2686, pruned_loss=0.05697, over 7415.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2832, pruned_loss=0.05764, over 1614059.95 frames. ], batch size: 17, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:32:19,141 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 12:32:23,953 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1082, 1.5250, 1.7825, 1.4213, 1.1722, 1.5766, 1.8212, 1.4917], + device='cuda:2'), covar=tensor([0.0481, 0.1242, 0.1593, 0.1450, 0.0553, 0.1425, 0.0641, 0.0673], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0152, 0.0189, 0.0161, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-07 12:32:28,669 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4893, 1.9530, 3.1006, 1.4129, 2.3008, 1.9751, 1.6583, 2.3380], + device='cuda:2'), covar=tensor([0.2226, 0.2934, 0.0842, 0.5239, 0.2074, 0.3448, 0.2821, 0.2386], + device='cuda:2'), in_proj_covar=tensor([0.0544, 0.0637, 0.0567, 0.0672, 0.0661, 0.0614, 0.0567, 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') +2023-02-07 12:32:40,306 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.81 vs. limit=5.0 +2023-02-07 12:32:42,453 INFO [train.py:901] (2/4) Epoch 27, batch 7450, loss[loss=0.2607, simple_loss=0.3243, pruned_loss=0.09854, over 8516.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2841, pruned_loss=0.05811, over 1616303.88 frames. ], batch size: 49, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:32:47,768 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.478e+02 3.262e+02 4.062e+02 8.102e+02, threshold=6.523e+02, percent-clipped=5.0 +2023-02-07 12:32:58,357 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 12:33:16,123 INFO [train.py:901] (2/4) Epoch 27, batch 7500, loss[loss=0.2049, simple_loss=0.2898, pruned_loss=0.05997, over 8342.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2838, pruned_loss=0.05764, over 1617635.67 frames. ], batch size: 26, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:33:28,170 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-07 12:33:34,960 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:33:46,545 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:33:51,440 INFO [train.py:901] (2/4) Epoch 27, batch 7550, loss[loss=0.2822, simple_loss=0.3556, pruned_loss=0.1044, over 8244.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2843, pruned_loss=0.05783, over 1618234.21 frames. ], batch size: 24, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:33:56,745 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.428e+02 3.024e+02 3.911e+02 8.560e+02, threshold=6.047e+02, percent-clipped=1.0 +2023-02-07 12:34:01,684 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6108, 1.4154, 2.8989, 1.3787, 2.2908, 3.0673, 3.2402, 2.6713], + device='cuda:2'), covar=tensor([0.1246, 0.1635, 0.0362, 0.2102, 0.0835, 0.0319, 0.0591, 0.0579], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0322, 0.0291, 0.0317, 0.0319, 0.0276, 0.0437, 0.0305], + 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') +2023-02-07 12:34:03,668 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:34:09,234 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-07 12:34:21,845 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:34:25,186 INFO [train.py:901] (2/4) Epoch 27, batch 7600, loss[loss=0.2241, simple_loss=0.2995, pruned_loss=0.0743, over 8083.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2841, pruned_loss=0.05794, over 1619407.51 frames. ], batch size: 21, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:34:53,129 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.28 vs. limit=5.0 +2023-02-07 12:35:01,502 INFO [train.py:901] (2/4) Epoch 27, batch 7650, loss[loss=0.2133, simple_loss=0.3077, pruned_loss=0.05946, over 8240.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2827, pruned_loss=0.05776, over 1615211.52 frames. ], batch size: 24, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:35:06,799 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.541e+02 2.896e+02 3.920e+02 6.720e+02, threshold=5.793e+02, percent-clipped=4.0 +2023-02-07 12:35:18,926 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9032, 3.7865, 3.4593, 1.8069, 3.4073, 3.5093, 3.4224, 3.3385], + device='cuda:2'), covar=tensor([0.0871, 0.0655, 0.1066, 0.4507, 0.1098, 0.1167, 0.1309, 0.0887], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0462, 0.0447, 0.0556, 0.0442, 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') +2023-02-07 12:35:35,074 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217854.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:35:35,677 INFO [train.py:901] (2/4) Epoch 27, batch 7700, loss[loss=0.1963, simple_loss=0.2937, pruned_loss=0.04951, over 8205.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2819, pruned_loss=0.05743, over 1614030.70 frames. ], batch size: 23, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:35:42,374 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:36:05,139 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 12:36:10,562 INFO [train.py:901] (2/4) Epoch 27, batch 7750, loss[loss=0.1889, simple_loss=0.2812, pruned_loss=0.04829, over 8626.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2817, pruned_loss=0.05741, over 1615380.86 frames. ], batch size: 31, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:36:14,308 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 12:36:15,959 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.515e+02 3.033e+02 3.634e+02 8.452e+02, threshold=6.066e+02, percent-clipped=4.0 +2023-02-07 12:36:18,817 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217916.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:36:27,440 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2211, 1.7303, 4.0984, 1.7575, 2.6662, 4.5621, 4.9212, 3.4835], + device='cuda:2'), covar=tensor([0.1627, 0.2246, 0.0453, 0.2672, 0.1279, 0.0353, 0.0642, 0.1179], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0322, 0.0290, 0.0317, 0.0318, 0.0275, 0.0435, 0.0304], + 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') +2023-02-07 12:36:45,569 INFO [train.py:901] (2/4) Epoch 27, batch 7800, loss[loss=0.2132, simple_loss=0.284, pruned_loss=0.07116, over 8297.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2809, pruned_loss=0.05699, over 1614597.22 frames. ], batch size: 23, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:36:55,088 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217969.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:37:04,147 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2434, 3.1309, 2.9063, 1.5801, 2.8333, 2.9318, 2.8044, 2.8579], + device='cuda:2'), covar=tensor([0.1076, 0.0890, 0.1333, 0.4814, 0.1160, 0.1277, 0.1698, 0.1014], + device='cuda:2'), in_proj_covar=tensor([0.0538, 0.0461, 0.0446, 0.0554, 0.0440, 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') +2023-02-07 12:37:19,648 INFO [train.py:901] (2/4) Epoch 27, batch 7850, loss[loss=0.2135, simple_loss=0.3015, pruned_loss=0.06278, over 8355.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2819, pruned_loss=0.0579, over 1614985.65 frames. ], batch size: 26, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:37:24,959 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.277e+02 2.828e+02 3.912e+02 8.712e+02, threshold=5.655e+02, percent-clipped=7.0 +2023-02-07 12:37:25,329 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 12:37:33,507 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218026.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:37:44,209 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6099, 1.9645, 2.9299, 1.5020, 2.2457, 2.0214, 1.6826, 2.2148], + device='cuda:2'), covar=tensor([0.2038, 0.2748, 0.0906, 0.4812, 0.1988, 0.3369, 0.2555, 0.2305], + device='cuda:2'), in_proj_covar=tensor([0.0541, 0.0635, 0.0565, 0.0671, 0.0660, 0.0614, 0.0565, 0.0644], + 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') +2023-02-07 12:37:52,840 INFO [train.py:901] (2/4) Epoch 27, batch 7900, loss[loss=0.1929, simple_loss=0.2728, pruned_loss=0.05646, over 8238.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2811, pruned_loss=0.05751, over 1610510.24 frames. ], batch size: 22, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:38:26,202 INFO [train.py:901] (2/4) Epoch 27, batch 7950, loss[loss=0.1797, simple_loss=0.2623, pruned_loss=0.04858, over 7978.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2815, pruned_loss=0.05782, over 1608851.95 frames. ], batch size: 21, lr: 2.76e-03, grad_scale: 8.0 +2023-02-07 12:38:29,870 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2026, 2.0762, 2.5951, 2.2059, 2.5938, 2.3145, 2.1128, 1.4913], + device='cuda:2'), covar=tensor([0.5812, 0.4946, 0.2269, 0.4001, 0.2686, 0.3343, 0.1930, 0.5862], + device='cuda:2'), in_proj_covar=tensor([0.0960, 0.1019, 0.0827, 0.0988, 0.1022, 0.0926, 0.0766, 0.0849], + 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') +2023-02-07 12:38:31,699 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.553e+02 3.230e+02 4.059e+02 8.354e+02, threshold=6.459e+02, percent-clipped=5.0 +2023-02-07 12:38:35,246 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:37,411 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:50,500 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:51,663 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8205, 1.3199, 2.8369, 1.4113, 2.2643, 3.0709, 3.2324, 2.6155], + device='cuda:2'), covar=tensor([0.1092, 0.1802, 0.0363, 0.2132, 0.0877, 0.0309, 0.0668, 0.0595], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0323, 0.0291, 0.0318, 0.0320, 0.0276, 0.0437, 0.0305], + 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') +2023-02-07 12:38:51,965 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 12:38:53,604 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218146.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:59,466 INFO [train.py:901] (2/4) Epoch 27, batch 8000, loss[loss=0.2332, simple_loss=0.3057, pruned_loss=0.08038, over 8326.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2829, pruned_loss=0.05843, over 1612745.65 frames. ], batch size: 25, lr: 2.76e-03, grad_scale: 8.0 +2023-02-07 12:39:05,387 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1997, 1.4944, 3.4951, 1.4756, 2.4982, 3.8249, 3.9689, 3.2823], + device='cuda:2'), covar=tensor([0.1093, 0.2001, 0.0331, 0.2233, 0.1122, 0.0258, 0.0486, 0.0597], + device='cuda:2'), in_proj_covar=tensor([0.0305, 0.0323, 0.0291, 0.0317, 0.0320, 0.0276, 0.0437, 0.0305], + 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') +2023-02-07 12:39:29,218 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218200.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:39:32,345 INFO [train.py:901] (2/4) Epoch 27, batch 8050, loss[loss=0.1852, simple_loss=0.2827, pruned_loss=0.04385, over 8464.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.05808, over 1609024.61 frames. ], batch size: 25, lr: 2.76e-03, grad_scale: 8.0 +2023-02-07 12:39:38,070 INFO [optim.py:369] (2/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.285e+02 2.948e+02 3.498e+02 7.136e+02, threshold=5.897e+02, percent-clipped=2.0 +2023-02-07 12:39:46,228 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218225.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:39:48,246 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218228.0, num_to_drop=0, layers_to_drop=set()