2023-05-15 15:28:40,541 INFO [finetune.py:1062] (1/2) Training started 2023-05-15 15:28:40,542 INFO [finetune.py:1072] (1/2) Device: cuda:1 2023-05-15 15:28:40,545 INFO [finetune.py:1081] (1/2) {'frame_shift_ms': 10.0, 'allowed_excess_duration_ratio': 0.1, 'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.23.4', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'a23383c5a381713b51e9014f3f05d096f8aceec3', 'k2-git-date': 'Wed Apr 26 15:33:33 2023', 'lhotse-version': '1.14.0.dev+git.b61b917.dirty', 'torch-version': '1.13.1', 'torch-cuda-available': True, 'torch-cuda-version': '11.6', 'python-version': '3.1', 'icefall-git-branch': 'master', 'icefall-git-sha1': '45c13e9-dirty', 'icefall-git-date': 'Mon Apr 24 15:00:02 2023', 'icefall-path': '/k2-dev/yangyifan/icefall-master', 'k2-path': '/k2-dev/yangyifan/anaconda3/envs/icefall/lib/python3.10/site-packages/k2-1.23.4.dev20230427+cuda11.6.torch1.13.1-py3.10-linux-x86_64.egg/k2/__init__.py', 'lhotse-path': '/k2-dev/yangyifan/anaconda3/envs/icefall/lib/python3.10/site-packages/lhotse-1.14.0.dev0+git.b61b917.dirty-py3.10.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-7-1218101249-5d97868c7c-v8ngc', 'IP address': '10.177.77.18'}, 'world_size': 2, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 20, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7/exp_giga_finetune'), 'bpe_model': 'icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11/data/lang_bpe_500/bpe.model', 'base_lr': 0.005, 'lr_batches': 100000.0, 'lr_epochs': 100.0, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 30, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,4,3,2,4', 'feedforward_dims': '1024,1024,2048,2048,1024', 'nhead': '8,8,8,8,8', 'encoder_dims': '384,384,384,384,384', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '256,256,256,256,256', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'do_finetune': True, 'use_mux': True, 'init_modules': None, 'finetune_ckpt': 'icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11/exp/pretrained.pt', 'manifest_dir': PosixPath('data/fbank'), '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, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'subset': 'S', 'small_dev': False, 'blank_id': 0, 'vocab_size': 500} 2023-05-15 15:28:40,546 INFO [finetune.py:1083] (1/2) About to create model 2023-05-15 15:28:41,274 INFO [zipformer.py:178] (1/2) 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. 2023-05-15 15:28:41,293 INFO [finetune.py:1087] (1/2) Number of model parameters: 70369391 2023-05-15 15:28:41,293 INFO [finetune.py:639] (1/2) Loading checkpoint from icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11/exp/pretrained.pt 2023-05-15 15:28:43,851 INFO [finetune.py:1109] (1/2) Using DDP 2023-05-15 15:28:44,730 INFO [asr_datamodule.py:425] (1/2) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts 2023-05-15 15:28:44,733 INFO [gigaspeech.py:389] (1/2) About to get train_S cuts 2023-05-15 15:28:44,733 INFO [gigaspeech.py:216] (1/2) Enable MUSAN 2023-05-15 15:28:44,733 INFO [gigaspeech.py:217] (1/2) About to get Musan cuts 2023-05-15 15:28:47,176 INFO [gigaspeech.py:241] (1/2) Enable SpecAugment 2023-05-15 15:28:47,176 INFO [gigaspeech.py:242] (1/2) Time warp factor: 80 2023-05-15 15:28:47,177 INFO [gigaspeech.py:252] (1/2) Num frame mask: 10 2023-05-15 15:28:47,177 INFO [gigaspeech.py:265] (1/2) About to create train dataset 2023-05-15 15:28:47,177 INFO [gigaspeech.py:291] (1/2) Using DynamicBucketingSampler. 2023-05-15 15:28:53,081 INFO [gigaspeech.py:306] (1/2) About to create train dataloader 2023-05-15 15:28:53,081 INFO [gigaspeech.py:396] (1/2) About to get dev cuts 2023-05-15 15:28:53,083 INFO [gigaspeech.py:337] (1/2) About to create dev dataset 2023-05-15 15:28:53,502 INFO [gigaspeech.py:354] (1/2) About to create dev dataloader 2023-05-15 15:29:16,361 INFO [finetune.py:992] (1/2) Epoch 1, batch 0, loss[loss=0.5445, simple_loss=0.5134, pruned_loss=0.3104, over 12346.00 frames. ], tot_loss[loss=0.5445, simple_loss=0.5134, pruned_loss=0.3104, over 12346.00 frames. ], batch size: 36, lr: 2.50e-03, grad_scale: 2.0 2023-05-15 15:29:16,362 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 15:29:34,131 INFO [finetune.py:1026] (1/2) Epoch 1, validation: loss=0.4712, simple_loss=0.4526, pruned_loss=0.1872, over 1020973.00 frames. 2023-05-15 15:29:34,131 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 9271MB 2023-05-15 15:29:39,295 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100007.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:29:56,806 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100028.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:29:59,998 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1982, 2.2048, 2.7805, 3.1858, 2.9826, 3.2241, 2.8482, 2.5098], device='cuda:1'), covar=tensor([0.0048, 0.0303, 0.0157, 0.0050, 0.0115, 0.0070, 0.0084, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0120, 0.0105, 0.0074, 0.0099, 0.0108, 0.0083, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:30:02,991 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100036.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:30:14,100 INFO [finetune.py:992] (1/2) Epoch 1, batch 50, loss[loss=0.23, simple_loss=0.23, pruned_loss=0.02445, over 12302.00 frames. ], tot_loss[loss=0.248, simple_loss=0.246, pruned_loss=0.03358, over 536007.26 frames. ], batch size: 34, lr: 2.75e-03, grad_scale: 2.0 2023-05-15 15:30:27,247 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100068.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 15:30:33,284 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100076.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:30:52,160 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 1.779e+02 2.045e+02 2.486e+02 5.674e+02, threshold=4.090e+02, percent-clipped=0.0 2023-05-15 15:30:52,189 INFO [finetune.py:992] (1/2) Epoch 1, batch 100, loss[loss=0.2553, simple_loss=0.2578, pruned_loss=0.02795, over 12268.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.2433, pruned_loss=0.03033, over 951577.19 frames. ], batch size: 37, lr: 3.00e-03, grad_scale: 2.0 2023-05-15 15:30:52,565 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 2023-05-15 15:31:09,359 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100124.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:31:09,436 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100124.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 15:31:10,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9020, 2.4653, 3.3683, 2.8683, 3.2289, 3.0333, 2.3875, 3.3463], device='cuda:1'), covar=tensor([0.0111, 0.0272, 0.0111, 0.0208, 0.0147, 0.0138, 0.0278, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0198, 0.0176, 0.0180, 0.0201, 0.0156, 0.0186, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:31:20,066 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4039, 5.3832, 5.2353, 5.3147, 5.0117, 5.2847, 5.3356, 5.6006], device='cuda:1'), covar=tensor([0.0148, 0.0096, 0.0144, 0.0197, 0.0590, 0.0253, 0.0091, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0193, 0.0190, 0.0243, 0.0246, 0.0206, 0.0176, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0006], device='cuda:1') 2023-05-15 15:31:29,622 INFO [finetune.py:992] (1/2) Epoch 1, batch 150, loss[loss=0.249, simple_loss=0.2534, pruned_loss=0.03078, over 12193.00 frames. ], tot_loss[loss=0.244, simple_loss=0.245, pruned_loss=0.03096, over 1266269.40 frames. ], batch size: 35, lr: 3.25e-03, grad_scale: 2.0 2023-05-15 15:31:32,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3827, 4.9564, 5.3371, 4.7204, 4.9769, 4.7897, 5.3863, 5.0571], device='cuda:1'), covar=tensor([0.0216, 0.0298, 0.0218, 0.0203, 0.0292, 0.0225, 0.0149, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0252, 0.0269, 0.0246, 0.0244, 0.0243, 0.0220, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-05-15 15:31:34,578 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2508, 4.1401, 4.2179, 4.6534, 2.8416, 4.0250, 2.6157, 4.1016], device='cuda:1'), covar=tensor([0.1622, 0.0591, 0.0712, 0.0451, 0.1302, 0.0552, 0.1842, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0266, 0.0307, 0.0374, 0.0247, 0.0240, 0.0265, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004], device='cuda:1') 2023-05-15 15:31:41,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-15 15:31:46,484 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100172.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 15:32:06,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-05-15 15:32:08,954 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 1.926e+02 2.279e+02 2.718e+02 1.176e+03, threshold=4.559e+02, percent-clipped=2.0 2023-05-15 15:32:08,974 INFO [finetune.py:992] (1/2) Epoch 1, batch 200, loss[loss=0.2022, simple_loss=0.2098, pruned_loss=0.01518, over 12364.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.2459, pruned_loss=0.03147, over 1508411.73 frames. ], batch size: 30, lr: 3.50e-03, grad_scale: 2.0 2023-05-15 15:32:26,245 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100224.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:32:34,432 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2383, 2.5942, 3.7910, 3.0980, 3.6342, 3.2346, 2.5826, 3.7051], device='cuda:1'), covar=tensor([0.0090, 0.0280, 0.0101, 0.0213, 0.0099, 0.0140, 0.0281, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0197, 0.0175, 0.0179, 0.0200, 0.0155, 0.0185, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:32:36,605 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100238.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:32:41,862 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100245.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:32:46,197 INFO [finetune.py:992] (1/2) Epoch 1, batch 250, loss[loss=0.2155, simple_loss=0.2243, pruned_loss=0.02514, over 12338.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.2458, pruned_loss=0.0319, over 1698718.67 frames. ], batch size: 30, lr: 3.75e-03, grad_scale: 2.0 2023-05-15 15:33:01,966 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100272.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:33:17,716 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100293.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:33:21,527 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9870, 4.9627, 4.8456, 4.8609, 4.5805, 4.9304, 4.9687, 5.1307], device='cuda:1'), covar=tensor([0.0148, 0.0102, 0.0123, 0.0265, 0.0630, 0.0267, 0.0111, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0191, 0.0188, 0.0240, 0.0244, 0.0203, 0.0175, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0006], device='cuda:1') 2023-05-15 15:33:22,367 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100299.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:33:23,640 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 1.969e+02 2.308e+02 2.769e+02 4.182e+02, threshold=4.615e+02, percent-clipped=0.0 2023-05-15 15:33:23,659 INFO [finetune.py:992] (1/2) Epoch 1, batch 300, loss[loss=0.237, simple_loss=0.2499, pruned_loss=0.02531, over 11554.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.2469, pruned_loss=0.03213, over 1856851.29 frames. ], batch size: 48, lr: 4.00e-03, grad_scale: 2.0 2023-05-15 15:33:49,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.14 vs. limit=5.0 2023-05-15 15:33:51,808 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100336.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:34:02,757 INFO [finetune.py:992] (1/2) Epoch 1, batch 350, loss[loss=0.2286, simple_loss=0.2406, pruned_loss=0.03559, over 12263.00 frames. ], tot_loss[loss=0.24, simple_loss=0.2475, pruned_loss=0.0322, over 1969926.55 frames. ], batch size: 32, lr: 4.25e-03, grad_scale: 2.0 2023-05-15 15:34:02,972 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0396, 4.8343, 4.9401, 4.9142, 4.7209, 4.9785, 4.8944, 2.8879], device='cuda:1'), covar=tensor([0.0082, 0.0051, 0.0068, 0.0061, 0.0049, 0.0068, 0.0061, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0079, 0.0073, 0.0061, 0.0091, 0.0078, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 15:34:12,115 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 15:34:27,748 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100384.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:34:40,987 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.106e+02 2.479e+02 2.883e+02 4.026e+02, threshold=4.958e+02, percent-clipped=0.0 2023-05-15 15:34:41,007 INFO [finetune.py:992] (1/2) Epoch 1, batch 400, loss[loss=0.2289, simple_loss=0.2434, pruned_loss=0.03548, over 12253.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.2467, pruned_loss=0.03211, over 2059081.14 frames. ], batch size: 32, lr: 4.50e-03, grad_scale: 4.0 2023-05-15 15:34:41,972 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100402.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:34:55,685 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4988, 3.2314, 4.8262, 2.6865, 2.7345, 3.7086, 3.0684, 3.8629], device='cuda:1'), covar=tensor([0.0391, 0.1084, 0.0233, 0.1064, 0.1899, 0.1207, 0.1358, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0239, 0.0249, 0.0188, 0.0249, 0.0302, 0.0234, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:35:18,807 INFO [finetune.py:992] (1/2) Epoch 1, batch 450, loss[loss=0.2301, simple_loss=0.248, pruned_loss=0.03333, over 12348.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.2454, pruned_loss=0.03196, over 2137433.25 frames. ], batch size: 35, lr: 4.75e-03, grad_scale: 4.0 2023-05-15 15:35:28,094 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100463.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:35:57,652 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.141e+02 2.472e+02 2.964e+02 7.480e+02, threshold=4.945e+02, percent-clipped=1.0 2023-05-15 15:35:57,672 INFO [finetune.py:992] (1/2) Epoch 1, batch 500, loss[loss=0.2197, simple_loss=0.2401, pruned_loss=0.02981, over 12337.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.2483, pruned_loss=0.03311, over 2186562.86 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 4.0 2023-05-15 15:36:35,196 INFO [finetune.py:992] (1/2) Epoch 1, batch 550, loss[loss=0.2312, simple_loss=0.2559, pruned_loss=0.03051, over 12253.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.2492, pruned_loss=0.0336, over 2231373.77 frames. ], batch size: 32, lr: 5.00e-03, grad_scale: 4.0 2023-05-15 15:36:42,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-15 15:36:56,988 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6723, 2.8214, 3.3870, 4.4954, 2.5832, 4.5870, 4.5223, 4.7152], device='cuda:1'), covar=tensor([0.0088, 0.1022, 0.0409, 0.0104, 0.1044, 0.0196, 0.0152, 0.0059], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0207, 0.0195, 0.0117, 0.0192, 0.0188, 0.0178, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:37:07,494 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100594.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:37:13,164 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.361e+02 2.670e+02 3.184e+02 7.098e+02, threshold=5.341e+02, percent-clipped=2.0 2023-05-15 15:37:13,184 INFO [finetune.py:992] (1/2) Epoch 1, batch 600, loss[loss=0.2229, simple_loss=0.2484, pruned_loss=0.03203, over 12163.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.2496, pruned_loss=0.03355, over 2272820.22 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 4.0 2023-05-15 15:37:32,509 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5036, 4.9476, 4.2168, 5.2411, 4.8226, 3.1889, 4.4854, 3.2454], device='cuda:1'), covar=tensor([0.0734, 0.0568, 0.1306, 0.0280, 0.0894, 0.1429, 0.0807, 0.2974], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0381, 0.0362, 0.0266, 0.0372, 0.0272, 0.0342, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:37:51,910 INFO [finetune.py:992] (1/2) Epoch 1, batch 650, loss[loss=0.2364, simple_loss=0.2607, pruned_loss=0.04608, over 12312.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.2497, pruned_loss=0.0338, over 2300342.36 frames. ], batch size: 34, lr: 5.00e-03, grad_scale: 4.0 2023-05-15 15:38:00,998 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100663.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:38:28,975 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.560e+02 2.965e+02 3.612e+02 8.428e+02, threshold=5.929e+02, percent-clipped=4.0 2023-05-15 15:38:28,994 INFO [finetune.py:992] (1/2) Epoch 1, batch 700, loss[loss=0.2165, simple_loss=0.248, pruned_loss=0.02881, over 10642.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.251, pruned_loss=0.0347, over 2312670.05 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 4.0 2023-05-15 15:38:30,722 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4681, 2.6997, 3.5539, 4.4071, 3.9070, 4.3440, 3.8755, 2.8406], device='cuda:1'), covar=tensor([0.0032, 0.0345, 0.0156, 0.0038, 0.0108, 0.0070, 0.0085, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0124, 0.0108, 0.0077, 0.0103, 0.0111, 0.0086, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:38:36,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-15 15:38:36,547 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100711.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:38:50,758 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100730.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:39:06,034 INFO [finetune.py:992] (1/2) Epoch 1, batch 750, loss[loss=0.2031, simple_loss=0.2337, pruned_loss=0.03021, over 12121.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.2519, pruned_loss=0.03539, over 2331993.33 frames. ], batch size: 30, lr: 5.00e-03, grad_scale: 4.0 2023-05-15 15:39:11,456 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100758.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 15:39:37,537 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100791.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:39:45,144 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.417e+02 2.979e+02 3.608e+02 6.266e+02, threshold=5.958e+02, percent-clipped=1.0 2023-05-15 15:39:45,164 INFO [finetune.py:992] (1/2) Epoch 1, batch 800, loss[loss=0.2102, simple_loss=0.2475, pruned_loss=0.02673, over 12185.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.2516, pruned_loss=0.03562, over 2338961.75 frames. ], batch size: 35, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:40:22,070 INFO [finetune.py:992] (1/2) Epoch 1, batch 850, loss[loss=0.1999, simple_loss=0.2352, pruned_loss=0.0305, over 12182.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.2528, pruned_loss=0.03676, over 2340520.73 frames. ], batch size: 29, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:40:54,151 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100894.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:40:59,141 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.549e+02 3.052e+02 3.672e+02 5.649e+02, threshold=6.104e+02, percent-clipped=0.0 2023-05-15 15:40:59,160 INFO [finetune.py:992] (1/2) Epoch 1, batch 900, loss[loss=0.1993, simple_loss=0.2362, pruned_loss=0.0324, over 12036.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2533, pruned_loss=0.03718, over 2343021.07 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:41:01,574 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7469, 3.2294, 5.1128, 2.6102, 2.8240, 3.9540, 3.2577, 3.9482], device='cuda:1'), covar=tensor([0.0412, 0.1161, 0.0295, 0.1323, 0.2015, 0.1387, 0.1426, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0238, 0.0248, 0.0187, 0.0248, 0.0298, 0.0233, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:41:28,191 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2547, 5.0386, 5.2774, 5.2272, 4.5031, 4.6153, 4.7281, 5.0659], device='cuda:1'), covar=tensor([0.0840, 0.0810, 0.0675, 0.0687, 0.2767, 0.2027, 0.0593, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0523, 0.0672, 0.0563, 0.0642, 0.0857, 0.0758, 0.0546, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0006, 0.0006, 0.0004, 0.0004], device='cuda:1') 2023-05-15 15:41:30,880 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100942.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:41:33,341 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2117, 4.7114, 4.0289, 4.8792, 4.4005, 2.9260, 4.1702, 3.1414], device='cuda:1'), covar=tensor([0.0713, 0.0532, 0.1279, 0.0300, 0.0945, 0.1574, 0.0961, 0.2786], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0378, 0.0359, 0.0264, 0.0369, 0.0270, 0.0340, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:41:37,568 INFO [finetune.py:992] (1/2) Epoch 1, batch 950, loss[loss=0.223, simple_loss=0.2633, pruned_loss=0.04218, over 12332.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2527, pruned_loss=0.03688, over 2353714.17 frames. ], batch size: 30, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:41:50,616 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9304, 3.9070, 3.8269, 3.8901, 3.6261, 3.8641, 3.8557, 4.0382], device='cuda:1'), covar=tensor([0.0220, 0.0170, 0.0220, 0.0306, 0.0648, 0.0393, 0.0198, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0192, 0.0187, 0.0240, 0.0242, 0.0202, 0.0176, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0006], device='cuda:1') 2023-05-15 15:42:09,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3107, 5.0017, 5.2527, 5.1545, 4.9669, 5.2132, 5.1693, 3.1976], device='cuda:1'), covar=tensor([0.0083, 0.0059, 0.0054, 0.0056, 0.0057, 0.0069, 0.0055, 0.0491], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0079, 0.0080, 0.0074, 0.0062, 0.0093, 0.0080, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 15:42:15,814 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.497e+02 2.927e+02 3.305e+02 5.891e+02, threshold=5.853e+02, percent-clipped=0.0 2023-05-15 15:42:15,833 INFO [finetune.py:992] (1/2) Epoch 1, batch 1000, loss[loss=0.211, simple_loss=0.2558, pruned_loss=0.03496, over 11108.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2523, pruned_loss=0.03676, over 2356295.51 frames. ], batch size: 55, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:42:44,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-15 15:42:52,940 INFO [finetune.py:992] (1/2) Epoch 1, batch 1050, loss[loss=0.2199, simple_loss=0.2695, pruned_loss=0.03703, over 12146.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2525, pruned_loss=0.03713, over 2359453.61 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:42:58,425 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101058.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:43:20,391 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101086.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 15:43:31,227 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.509e+02 2.900e+02 3.477e+02 6.199e+02, threshold=5.799e+02, percent-clipped=2.0 2023-05-15 15:43:31,246 INFO [finetune.py:992] (1/2) Epoch 1, batch 1100, loss[loss=0.187, simple_loss=0.2338, pruned_loss=0.02952, over 12190.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2531, pruned_loss=0.03784, over 2344630.99 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:43:35,095 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101106.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:43:35,945 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101107.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:44:08,189 INFO [finetune.py:992] (1/2) Epoch 1, batch 1150, loss[loss=0.2277, simple_loss=0.2768, pruned_loss=0.04947, over 10722.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2534, pruned_loss=0.03798, over 2349041.26 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:44:20,934 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101168.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:44:20,991 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8257, 2.6598, 4.0078, 4.1884, 2.9273, 2.8152, 2.8132, 2.3757], device='cuda:1'), covar=tensor([0.1324, 0.2562, 0.0498, 0.0392, 0.1039, 0.1858, 0.2189, 0.3146], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0390, 0.0274, 0.0299, 0.0260, 0.0291, 0.0371, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:44:21,772 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7172, 3.1663, 5.0492, 2.6271, 2.6505, 3.8182, 2.9966, 3.9890], device='cuda:1'), covar=tensor([0.0418, 0.1164, 0.0354, 0.1192, 0.2162, 0.1462, 0.1577, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0239, 0.0248, 0.0187, 0.0248, 0.0298, 0.0234, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:44:45,566 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.377e+02 2.892e+02 3.477e+02 1.072e+03, threshold=5.784e+02, percent-clipped=5.0 2023-05-15 15:44:45,595 INFO [finetune.py:992] (1/2) Epoch 1, batch 1200, loss[loss=0.1985, simple_loss=0.2464, pruned_loss=0.04077, over 11314.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2549, pruned_loss=0.03833, over 2349920.10 frames. ], batch size: 25, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:45:11,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 15:45:13,777 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.3363, 6.2985, 6.1976, 5.5101, 5.4013, 6.2316, 5.8801, 5.7104], device='cuda:1'), covar=tensor([0.0519, 0.0728, 0.0414, 0.1259, 0.0546, 0.0529, 0.1132, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0540, 0.0483, 0.0611, 0.0386, 0.0694, 0.0768, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 15:45:23,998 INFO [finetune.py:992] (1/2) Epoch 1, batch 1250, loss[loss=0.2128, simple_loss=0.2731, pruned_loss=0.03785, over 12059.00 frames. ], tot_loss[loss=0.209, simple_loss=0.254, pruned_loss=0.03811, over 2358728.44 frames. ], batch size: 42, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:45:31,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6419, 2.5547, 3.3005, 4.5271, 2.4101, 4.4621, 4.5376, 4.7588], device='cuda:1'), covar=tensor([0.0117, 0.1212, 0.0476, 0.0128, 0.1207, 0.0245, 0.0123, 0.0062], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0211, 0.0196, 0.0119, 0.0193, 0.0190, 0.0177, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:45:32,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-15 15:46:00,732 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.665e+02 3.108e+02 3.746e+02 6.531e+02, threshold=6.215e+02, percent-clipped=2.0 2023-05-15 15:46:00,761 INFO [finetune.py:992] (1/2) Epoch 1, batch 1300, loss[loss=0.2095, simple_loss=0.2607, pruned_loss=0.04904, over 12338.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2545, pruned_loss=0.03792, over 2366368.44 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:46:21,603 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9307, 3.4773, 5.2965, 2.8642, 2.9479, 4.0309, 3.2037, 4.0656], device='cuda:1'), covar=tensor([0.0416, 0.0981, 0.0199, 0.1096, 0.1906, 0.1136, 0.1372, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0236, 0.0246, 0.0186, 0.0246, 0.0295, 0.0231, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:46:37,426 INFO [finetune.py:992] (1/2) Epoch 1, batch 1350, loss[loss=0.204, simple_loss=0.2656, pruned_loss=0.03961, over 11662.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2551, pruned_loss=0.0382, over 2366997.85 frames. ], batch size: 48, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:46:39,175 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2728, 4.5319, 4.0112, 4.8709, 4.3887, 2.7708, 4.0970, 3.0359], device='cuda:1'), covar=tensor([0.0711, 0.0749, 0.1297, 0.0310, 0.1050, 0.1661, 0.1131, 0.2960], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0382, 0.0364, 0.0268, 0.0374, 0.0273, 0.0342, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:47:04,844 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101386.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:47:16,296 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.435e+02 2.915e+02 3.611e+02 6.080e+02, threshold=5.831e+02, percent-clipped=0.0 2023-05-15 15:47:16,316 INFO [finetune.py:992] (1/2) Epoch 1, batch 1400, loss[loss=0.1785, simple_loss=0.2418, pruned_loss=0.0293, over 12019.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2555, pruned_loss=0.0386, over 2357215.72 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:47:40,801 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101434.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:47:52,838 INFO [finetune.py:992] (1/2) Epoch 1, batch 1450, loss[loss=0.2124, simple_loss=0.2729, pruned_loss=0.05116, over 12308.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.255, pruned_loss=0.03831, over 2367307.02 frames. ], batch size: 34, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:48:01,766 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101463.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:48:29,430 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.709e+02 3.317e+02 3.771e+02 6.828e+02, threshold=6.633e+02, percent-clipped=3.0 2023-05-15 15:48:29,449 INFO [finetune.py:992] (1/2) Epoch 1, batch 1500, loss[loss=0.1831, simple_loss=0.2545, pruned_loss=0.03105, over 12280.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2558, pruned_loss=0.03877, over 2368894.14 frames. ], batch size: 37, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:49:07,512 INFO [finetune.py:992] (1/2) Epoch 1, batch 1550, loss[loss=0.1976, simple_loss=0.2674, pruned_loss=0.04251, over 12361.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2562, pruned_loss=0.03897, over 2364298.73 frames. ], batch size: 35, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:49:23,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-15 15:49:45,134 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.752e+02 3.202e+02 3.781e+02 7.212e+02, threshold=6.405e+02, percent-clipped=2.0 2023-05-15 15:49:45,154 INFO [finetune.py:992] (1/2) Epoch 1, batch 1600, loss[loss=0.1566, simple_loss=0.2231, pruned_loss=0.02776, over 12164.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2551, pruned_loss=0.03903, over 2363364.85 frames. ], batch size: 29, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:49:58,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-15 15:50:15,191 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7414, 3.3155, 3.9280, 4.7727, 4.3790, 4.7533, 4.2513, 3.2402], device='cuda:1'), covar=tensor([0.0021, 0.0259, 0.0096, 0.0024, 0.0063, 0.0049, 0.0076, 0.0294], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0121, 0.0106, 0.0075, 0.0101, 0.0109, 0.0085, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:50:23,067 INFO [finetune.py:992] (1/2) Epoch 1, batch 1650, loss[loss=0.1861, simple_loss=0.2597, pruned_loss=0.03982, over 12091.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2555, pruned_loss=0.03924, over 2369532.21 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:50:48,472 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101684.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:51:00,878 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.787e+02 3.159e+02 3.724e+02 8.656e+02, threshold=6.317e+02, percent-clipped=3.0 2023-05-15 15:51:00,897 INFO [finetune.py:992] (1/2) Epoch 1, batch 1700, loss[loss=0.1929, simple_loss=0.2732, pruned_loss=0.04152, over 12139.00 frames. ], tot_loss[loss=0.1899, simple_loss=0.2556, pruned_loss=0.03901, over 2379014.91 frames. ], batch size: 38, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:51:02,583 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3039, 3.5654, 3.1619, 3.2222, 2.8918, 2.8063, 3.4586, 2.2153], device='cuda:1'), covar=tensor([0.0353, 0.0095, 0.0139, 0.0132, 0.0258, 0.0229, 0.0093, 0.0366], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0160, 0.0152, 0.0180, 0.0203, 0.0197, 0.0158, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:51:33,493 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101745.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:51:37,775 INFO [finetune.py:992] (1/2) Epoch 1, batch 1750, loss[loss=0.1636, simple_loss=0.245, pruned_loss=0.02913, over 12283.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2567, pruned_loss=0.03961, over 2376616.95 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:51:46,637 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101763.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:52:06,446 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7383, 4.4222, 4.6041, 4.6123, 4.4393, 4.6257, 4.5705, 2.6832], device='cuda:1'), covar=tensor([0.0096, 0.0071, 0.0079, 0.0068, 0.0055, 0.0088, 0.0074, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0079, 0.0080, 0.0075, 0.0062, 0.0092, 0.0081, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 15:52:15,369 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.759e+02 3.294e+02 3.844e+02 7.750e+02, threshold=6.588e+02, percent-clipped=3.0 2023-05-15 15:52:15,388 INFO [finetune.py:992] (1/2) Epoch 1, batch 1800, loss[loss=0.1948, simple_loss=0.2698, pruned_loss=0.051, over 11835.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2564, pruned_loss=0.03951, over 2376630.68 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:52:22,681 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101811.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:52:52,283 INFO [finetune.py:992] (1/2) Epoch 1, batch 1850, loss[loss=0.2039, simple_loss=0.291, pruned_loss=0.05098, over 11869.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2574, pruned_loss=0.03989, over 2377608.79 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:53:19,801 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2489, 2.7511, 3.6727, 3.1785, 3.6059, 3.2657, 2.5941, 3.6306], device='cuda:1'), covar=tensor([0.0108, 0.0260, 0.0160, 0.0225, 0.0121, 0.0160, 0.0310, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0197, 0.0174, 0.0178, 0.0198, 0.0155, 0.0186, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 15:53:22,707 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4790, 5.2698, 5.3739, 5.4173, 5.0617, 5.0472, 4.9476, 5.3679], device='cuda:1'), covar=tensor([0.0583, 0.0498, 0.0621, 0.0519, 0.1815, 0.1249, 0.0473, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0662, 0.0561, 0.0636, 0.0848, 0.0750, 0.0541, 0.0486], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0006, 0.0006, 0.0004, 0.0004], device='cuda:1') 2023-05-15 15:53:29,158 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 2.693e+02 3.375e+02 4.011e+02 6.163e+02, threshold=6.750e+02, percent-clipped=0.0 2023-05-15 15:53:29,177 INFO [finetune.py:992] (1/2) Epoch 1, batch 1900, loss[loss=0.1707, simple_loss=0.26, pruned_loss=0.03587, over 12014.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2575, pruned_loss=0.04, over 2374101.07 frames. ], batch size: 40, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:53:35,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-15 15:53:43,447 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101920.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:53:55,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-15 15:54:06,199 INFO [finetune.py:992] (1/2) Epoch 1, batch 1950, loss[loss=0.171, simple_loss=0.2624, pruned_loss=0.03742, over 12307.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2579, pruned_loss=0.04009, over 2381905.00 frames. ], batch size: 34, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 15:54:29,099 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101981.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 15:54:30,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 15:54:33,275 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101987.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:54:36,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0 2023-05-15 15:54:47,259 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.827e+02 2.821e+02 3.357e+02 4.284e+02 8.103e+02, threshold=6.713e+02, percent-clipped=5.0 2023-05-15 15:54:47,288 INFO [finetune.py:992] (1/2) Epoch 1, batch 2000, loss[loss=0.1703, simple_loss=0.2608, pruned_loss=0.03988, over 11748.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2573, pruned_loss=0.04037, over 2372977.19 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:54:51,993 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6634, 3.3158, 5.1129, 2.6373, 2.8529, 3.9295, 3.2247, 3.8990], device='cuda:1'), covar=tensor([0.0412, 0.1091, 0.0221, 0.1167, 0.1882, 0.1196, 0.1329, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0236, 0.0247, 0.0186, 0.0246, 0.0293, 0.0232, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:54:55,715 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 15:55:15,791 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102040.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:55:18,135 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2605, 4.1612, 2.5668, 2.3862, 3.6890, 2.3620, 3.7286, 2.9927], device='cuda:1'), covar=tensor([0.0567, 0.0685, 0.1082, 0.1420, 0.0247, 0.1216, 0.0419, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0261, 0.0177, 0.0200, 0.0142, 0.0182, 0.0200, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 15:55:21,741 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102048.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:55:23,736 INFO [finetune.py:992] (1/2) Epoch 1, batch 2050, loss[loss=0.1875, simple_loss=0.2798, pruned_loss=0.04766, over 12056.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2568, pruned_loss=0.03989, over 2380616.68 frames. ], batch size: 37, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:55:40,098 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4257, 2.5660, 3.7159, 4.4019, 4.0110, 4.4347, 3.8935, 2.7212], device='cuda:1'), covar=tensor([0.0030, 0.0352, 0.0110, 0.0035, 0.0095, 0.0051, 0.0084, 0.0361], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0121, 0.0105, 0.0076, 0.0100, 0.0109, 0.0084, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:56:01,302 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.025e+02 2.708e+02 3.160e+02 3.820e+02 1.513e+03, threshold=6.321e+02, percent-clipped=2.0 2023-05-15 15:56:01,321 INFO [finetune.py:992] (1/2) Epoch 1, batch 2100, loss[loss=0.148, simple_loss=0.2273, pruned_loss=0.03435, over 11994.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2565, pruned_loss=0.03953, over 2381348.12 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:56:20,550 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8340, 3.4143, 5.2292, 2.7822, 2.8209, 4.1739, 3.3845, 4.1493], device='cuda:1'), covar=tensor([0.0381, 0.1052, 0.0241, 0.1135, 0.1910, 0.1086, 0.1183, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0236, 0.0248, 0.0186, 0.0246, 0.0294, 0.0232, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:56:38,906 INFO [finetune.py:992] (1/2) Epoch 1, batch 2150, loss[loss=0.1442, simple_loss=0.2303, pruned_loss=0.02907, over 12171.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2563, pruned_loss=0.03965, over 2382128.17 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:56:47,954 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2908, 4.9537, 5.0906, 5.0870, 4.9803, 5.0757, 5.0616, 2.7829], device='cuda:1'), covar=tensor([0.0067, 0.0047, 0.0055, 0.0046, 0.0041, 0.0069, 0.0057, 0.0630], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0079, 0.0080, 0.0075, 0.0062, 0.0093, 0.0081, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 15:57:15,764 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.922e+02 3.570e+02 4.264e+02 7.831e+02, threshold=7.141e+02, percent-clipped=3.0 2023-05-15 15:57:15,784 INFO [finetune.py:992] (1/2) Epoch 1, batch 2200, loss[loss=0.1873, simple_loss=0.2851, pruned_loss=0.04479, over 11608.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2576, pruned_loss=0.04033, over 2372408.32 frames. ], batch size: 48, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:57:31,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-15 15:57:53,007 INFO [finetune.py:992] (1/2) Epoch 1, batch 2250, loss[loss=0.1732, simple_loss=0.2658, pruned_loss=0.04034, over 12083.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2581, pruned_loss=0.04038, over 2367066.27 frames. ], batch size: 42, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:58:12,329 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102276.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 15:58:30,596 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.636e+02 3.287e+02 3.960e+02 8.443e+02, threshold=6.574e+02, percent-clipped=2.0 2023-05-15 15:58:30,615 INFO [finetune.py:992] (1/2) Epoch 1, batch 2300, loss[loss=0.1733, simple_loss=0.2654, pruned_loss=0.04059, over 12079.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.258, pruned_loss=0.04059, over 2365511.77 frames. ], batch size: 32, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:58:33,058 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6644, 3.3123, 5.1215, 2.7265, 2.6169, 4.0524, 3.1713, 4.1058], device='cuda:1'), covar=tensor([0.0421, 0.1146, 0.0266, 0.1174, 0.2118, 0.1122, 0.1514, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0236, 0.0249, 0.0187, 0.0248, 0.0296, 0.0234, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 15:58:59,506 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102340.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:59:01,683 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102343.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:59:07,415 INFO [finetune.py:992] (1/2) Epoch 1, batch 2350, loss[loss=0.1797, simple_loss=0.2713, pruned_loss=0.04401, over 12270.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2575, pruned_loss=0.04067, over 2369050.54 frames. ], batch size: 37, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:59:24,172 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1662, 4.9190, 5.0014, 5.0112, 4.7750, 4.9736, 4.9197, 2.7082], device='cuda:1'), covar=tensor([0.0094, 0.0045, 0.0066, 0.0056, 0.0053, 0.0077, 0.0060, 0.0684], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0079, 0.0081, 0.0075, 0.0063, 0.0094, 0.0082, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 15:59:34,852 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102388.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 15:59:44,626 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.777e+02 3.297e+02 3.992e+02 1.048e+03, threshold=6.593e+02, percent-clipped=3.0 2023-05-15 15:59:44,645 INFO [finetune.py:992] (1/2) Epoch 1, batch 2400, loss[loss=0.1779, simple_loss=0.2658, pruned_loss=0.04506, over 12040.00 frames. ], tot_loss[loss=0.171, simple_loss=0.258, pruned_loss=0.04079, over 2368174.04 frames. ], batch size: 40, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 15:59:50,438 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4926, 4.8653, 2.8844, 2.6234, 4.0870, 2.7827, 4.1456, 3.4017], device='cuda:1'), covar=tensor([0.0638, 0.0535, 0.1207, 0.1559, 0.0252, 0.1142, 0.0395, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0264, 0.0180, 0.0204, 0.0144, 0.0183, 0.0202, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:00:21,727 INFO [finetune.py:992] (1/2) Epoch 1, batch 2450, loss[loss=0.1719, simple_loss=0.2645, pruned_loss=0.03969, over 12341.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2578, pruned_loss=0.04055, over 2372457.33 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:00:58,513 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.773e+02 3.289e+02 4.075e+02 7.904e+02, threshold=6.577e+02, percent-clipped=1.0 2023-05-15 16:00:58,534 INFO [finetune.py:992] (1/2) Epoch 1, batch 2500, loss[loss=0.1371, simple_loss=0.2155, pruned_loss=0.02933, over 12188.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2569, pruned_loss=0.03998, over 2374899.98 frames. ], batch size: 29, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:01:05,364 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4562, 5.0292, 5.4118, 4.8795, 5.1222, 4.9380, 5.4360, 5.0254], device='cuda:1'), covar=tensor([0.0228, 0.0314, 0.0221, 0.0203, 0.0245, 0.0228, 0.0176, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0252, 0.0264, 0.0241, 0.0241, 0.0239, 0.0219, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-05-15 16:01:25,287 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2435, 4.0751, 4.1502, 4.5915, 3.4018, 4.0786, 2.6125, 4.0747], device='cuda:1'), covar=tensor([0.1565, 0.0692, 0.0825, 0.0608, 0.1026, 0.0522, 0.1877, 0.1415], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0270, 0.0310, 0.0375, 0.0252, 0.0243, 0.0267, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004], device='cuda:1') 2023-05-15 16:01:35,495 INFO [finetune.py:992] (1/2) Epoch 1, batch 2550, loss[loss=0.2077, simple_loss=0.287, pruned_loss=0.06419, over 8185.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2572, pruned_loss=0.04022, over 2364697.64 frames. ], batch size: 99, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:01:54,635 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102576.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 16:02:12,855 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 3.014e+02 3.479e+02 4.250e+02 9.382e+02, threshold=6.957e+02, percent-clipped=5.0 2023-05-15 16:02:12,876 INFO [finetune.py:992] (1/2) Epoch 1, batch 2600, loss[loss=0.188, simple_loss=0.2806, pruned_loss=0.04773, over 12069.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2576, pruned_loss=0.04029, over 2370843.85 frames. ], batch size: 42, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:02:29,670 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102624.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:02:39,608 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3576, 4.9495, 5.3427, 4.7415, 5.0196, 4.7787, 5.3735, 4.8672], device='cuda:1'), covar=tensor([0.0246, 0.0346, 0.0230, 0.0200, 0.0258, 0.0276, 0.0192, 0.0356], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0256, 0.0269, 0.0245, 0.0244, 0.0243, 0.0222, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-05-15 16:02:43,271 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102643.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:02:48,811 INFO [finetune.py:992] (1/2) Epoch 1, batch 2650, loss[loss=0.1627, simple_loss=0.2496, pruned_loss=0.03792, over 12026.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2576, pruned_loss=0.0403, over 2374410.25 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:03:11,901 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-15 16:03:18,667 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102691.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:03:25,832 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 2.634e+02 3.104e+02 3.917e+02 7.520e+02, threshold=6.208e+02, percent-clipped=2.0 2023-05-15 16:03:25,856 INFO [finetune.py:992] (1/2) Epoch 1, batch 2700, loss[loss=0.2062, simple_loss=0.2914, pruned_loss=0.06051, over 12001.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2579, pruned_loss=0.04055, over 2376639.33 frames. ], batch size: 40, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:03:57,370 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8751, 5.7672, 5.6177, 5.0815, 5.0936, 5.7866, 5.3111, 5.1469], device='cuda:1'), covar=tensor([0.0710, 0.0928, 0.0677, 0.1391, 0.0653, 0.0664, 0.1567, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0603, 0.0537, 0.0480, 0.0608, 0.0389, 0.0689, 0.0760, 0.0546], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:04:02,836 INFO [finetune.py:992] (1/2) Epoch 1, batch 2750, loss[loss=0.1536, simple_loss=0.2385, pruned_loss=0.03433, over 12247.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2588, pruned_loss=0.04093, over 2372413.24 frames. ], batch size: 32, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:04:39,870 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.779e+02 3.278e+02 3.944e+02 1.142e+03, threshold=6.557e+02, percent-clipped=3.0 2023-05-15 16:04:39,889 INFO [finetune.py:992] (1/2) Epoch 1, batch 2800, loss[loss=0.161, simple_loss=0.2508, pruned_loss=0.03567, over 12294.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2583, pruned_loss=0.04062, over 2372329.60 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:05:07,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-05-15 16:05:16,518 INFO [finetune.py:992] (1/2) Epoch 1, batch 2850, loss[loss=0.1771, simple_loss=0.2671, pruned_loss=0.04354, over 12054.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2578, pruned_loss=0.04054, over 2371475.09 frames. ], batch size: 37, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:05:16,726 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102851.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:05:43,543 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102887.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:05:53,843 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.152e+02 2.894e+02 3.348e+02 4.267e+02 1.176e+03, threshold=6.696e+02, percent-clipped=6.0 2023-05-15 16:05:53,862 INFO [finetune.py:992] (1/2) Epoch 1, batch 2900, loss[loss=0.1971, simple_loss=0.2889, pruned_loss=0.05263, over 12013.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2575, pruned_loss=0.04063, over 2353761.13 frames. ], batch size: 40, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:06:01,936 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102912.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:06:28,077 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102948.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:06:30,074 INFO [finetune.py:992] (1/2) Epoch 1, batch 2950, loss[loss=0.1821, simple_loss=0.273, pruned_loss=0.04563, over 10438.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2572, pruned_loss=0.04041, over 2362574.54 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:06:41,839 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6326, 3.0994, 3.8550, 4.7348, 4.3070, 4.5861, 4.1632, 3.2584], device='cuda:1'), covar=tensor([0.0032, 0.0279, 0.0109, 0.0024, 0.0075, 0.0069, 0.0075, 0.0287], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0121, 0.0106, 0.0076, 0.0101, 0.0111, 0.0085, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:07:07,068 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.662e+02 3.202e+02 3.925e+02 7.031e+02, threshold=6.403e+02, percent-clipped=1.0 2023-05-15 16:07:07,092 INFO [finetune.py:992] (1/2) Epoch 1, batch 3000, loss[loss=0.1948, simple_loss=0.285, pruned_loss=0.05226, over 11853.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2572, pruned_loss=0.0403, over 2374134.53 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:07:07,093 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 16:07:15,456 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9935, 2.2337, 3.4049, 2.8050, 3.3409, 3.0114, 2.2745, 3.3660], device='cuda:1'), covar=tensor([0.0094, 0.0280, 0.0086, 0.0178, 0.0101, 0.0158, 0.0264, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0196, 0.0174, 0.0178, 0.0197, 0.0154, 0.0184, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:07:16,471 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9206, 5.8773, 5.8093, 5.1736, 5.3482, 5.8774, 5.2270, 5.4146], device='cuda:1'), covar=tensor([0.0692, 0.0902, 0.0482, 0.1297, 0.0389, 0.0604, 0.1699, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0536, 0.0480, 0.0608, 0.0389, 0.0685, 0.0759, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:07:25,014 INFO [finetune.py:1026] (1/2) Epoch 1, validation: loss=0.3762, simple_loss=0.4347, pruned_loss=0.1589, over 1020973.00 frames. 2023-05-15 16:07:25,015 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 16:08:00,959 INFO [finetune.py:992] (1/2) Epoch 1, batch 3050, loss[loss=0.1739, simple_loss=0.2625, pruned_loss=0.04262, over 10536.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.257, pruned_loss=0.04054, over 2378575.75 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:08:18,847 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9584, 3.4221, 5.1998, 2.6281, 2.9249, 3.8858, 3.3405, 3.9698], device='cuda:1'), covar=tensor([0.0319, 0.1040, 0.0340, 0.1209, 0.2006, 0.1463, 0.1272, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0236, 0.0248, 0.0186, 0.0247, 0.0294, 0.0233, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:08:35,957 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103098.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:08:37,998 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.894e+02 3.540e+02 4.389e+02 8.495e+02, threshold=7.080e+02, percent-clipped=4.0 2023-05-15 16:08:38,018 INFO [finetune.py:992] (1/2) Epoch 1, batch 3100, loss[loss=0.1411, simple_loss=0.227, pruned_loss=0.02761, over 12115.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2575, pruned_loss=0.04061, over 2375568.22 frames. ], batch size: 30, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:09:01,333 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103132.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:09:08,064 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8184, 2.6762, 3.9971, 4.1386, 2.9661, 2.7183, 2.7996, 2.2968], device='cuda:1'), covar=tensor([0.1327, 0.2756, 0.0531, 0.0414, 0.1020, 0.1926, 0.2324, 0.3436], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0384, 0.0271, 0.0297, 0.0258, 0.0285, 0.0365, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:09:15,010 INFO [finetune.py:992] (1/2) Epoch 1, batch 3150, loss[loss=0.1437, simple_loss=0.2334, pruned_loss=0.02704, over 12298.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2571, pruned_loss=0.04038, over 2378217.12 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:09:21,062 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103159.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:09:45,859 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103193.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:09:51,644 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.858e+02 3.403e+02 4.168e+02 1.314e+03, threshold=6.807e+02, percent-clipped=6.0 2023-05-15 16:09:51,663 INFO [finetune.py:992] (1/2) Epoch 1, batch 3200, loss[loss=0.1864, simple_loss=0.2823, pruned_loss=0.04519, over 12335.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2575, pruned_loss=0.04048, over 2382185.91 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:09:56,062 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103207.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:10:19,365 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4160, 4.7026, 4.2481, 4.9879, 4.6072, 3.0495, 4.4510, 3.2034], device='cuda:1'), covar=tensor([0.0666, 0.0704, 0.1149, 0.0289, 0.0861, 0.1475, 0.0786, 0.2793], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0378, 0.0355, 0.0266, 0.0366, 0.0268, 0.0337, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:10:22,764 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103243.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:10:28,473 INFO [finetune.py:992] (1/2) Epoch 1, batch 3250, loss[loss=0.1662, simple_loss=0.2586, pruned_loss=0.03688, over 12096.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2575, pruned_loss=0.03991, over 2382289.37 frames. ], batch size: 42, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:11:05,081 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.862e+02 3.260e+02 4.103e+02 6.638e+02, threshold=6.520e+02, percent-clipped=1.0 2023-05-15 16:11:05,100 INFO [finetune.py:992] (1/2) Epoch 1, batch 3300, loss[loss=0.1466, simple_loss=0.2275, pruned_loss=0.03288, over 11999.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2577, pruned_loss=0.04046, over 2379456.54 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:11:41,104 INFO [finetune.py:992] (1/2) Epoch 1, batch 3350, loss[loss=0.1703, simple_loss=0.2523, pruned_loss=0.04419, over 12342.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2579, pruned_loss=0.0405, over 2378998.07 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:11:48,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-15 16:11:52,668 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103366.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:11:56,349 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3262, 4.7248, 2.8918, 2.4460, 4.1736, 2.2492, 4.0432, 3.2864], device='cuda:1'), covar=tensor([0.0657, 0.0441, 0.1146, 0.1721, 0.0266, 0.1492, 0.0452, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0257, 0.0176, 0.0200, 0.0142, 0.0179, 0.0200, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:12:18,310 INFO [finetune.py:992] (1/2) Epoch 1, batch 3400, loss[loss=0.1825, simple_loss=0.2757, pruned_loss=0.04464, over 12313.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.258, pruned_loss=0.04067, over 2382187.82 frames. ], batch size: 34, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:12:19,026 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 2.898e+02 3.310e+02 3.849e+02 7.246e+02, threshold=6.620e+02, percent-clipped=1.0 2023-05-15 16:12:20,750 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4093, 4.8180, 2.9911, 2.7374, 4.1555, 2.4232, 4.0489, 3.4521], device='cuda:1'), covar=tensor([0.0541, 0.0400, 0.0907, 0.1267, 0.0216, 0.1161, 0.0404, 0.0623], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0257, 0.0176, 0.0200, 0.0142, 0.0179, 0.0199, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:12:33,407 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-15 16:12:38,104 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103427.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 16:12:55,372 INFO [finetune.py:992] (1/2) Epoch 1, batch 3450, loss[loss=0.1677, simple_loss=0.2603, pruned_loss=0.03754, over 12104.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2576, pruned_loss=0.04044, over 2385151.65 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:12:57,631 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103454.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:13:22,124 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103488.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:13:27,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-15 16:13:31,340 INFO [finetune.py:992] (1/2) Epoch 1, batch 3500, loss[loss=0.2087, simple_loss=0.288, pruned_loss=0.06467, over 11857.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2579, pruned_loss=0.04103, over 2380774.74 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:13:32,042 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.723e+02 3.181e+02 3.899e+02 6.414e+02, threshold=6.363e+02, percent-clipped=0.0 2023-05-15 16:13:35,863 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103507.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:13:45,404 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1296, 2.4795, 3.6552, 3.0220, 3.4769, 3.1221, 2.5457, 3.6653], device='cuda:1'), covar=tensor([0.0099, 0.0265, 0.0099, 0.0214, 0.0124, 0.0160, 0.0290, 0.0071], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0191, 0.0168, 0.0173, 0.0193, 0.0151, 0.0179, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:14:02,702 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103543.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:14:09,189 INFO [finetune.py:992] (1/2) Epoch 1, batch 3550, loss[loss=0.1678, simple_loss=0.261, pruned_loss=0.03724, over 10410.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2578, pruned_loss=0.04064, over 2374197.53 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:14:12,121 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103555.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:14:29,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2928, 6.0075, 5.5420, 5.5569, 6.0624, 5.3706, 5.6726, 5.6637], device='cuda:1'), covar=tensor([0.1381, 0.0896, 0.0784, 0.1769, 0.0903, 0.2197, 0.1493, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0472, 0.0365, 0.0414, 0.0449, 0.0417, 0.0383, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:14:38,331 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103591.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:14:45,738 INFO [finetune.py:992] (1/2) Epoch 1, batch 3600, loss[loss=0.1616, simple_loss=0.2507, pruned_loss=0.03627, over 12194.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2574, pruned_loss=0.04054, over 2376514.62 frames. ], batch size: 35, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:14:46,422 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.881e+02 3.349e+02 4.086e+02 6.568e+02, threshold=6.698e+02, percent-clipped=1.0 2023-05-15 16:15:21,478 INFO [finetune.py:992] (1/2) Epoch 1, batch 3650, loss[loss=0.1862, simple_loss=0.2584, pruned_loss=0.05698, over 8605.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2582, pruned_loss=0.04069, over 2371674.48 frames. ], batch size: 98, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:15:34,812 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3987, 3.9047, 3.9876, 4.3779, 3.2463, 3.8648, 2.7029, 4.0068], device='cuda:1'), covar=tensor([0.1407, 0.0662, 0.0843, 0.0619, 0.0942, 0.0560, 0.1658, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0266, 0.0305, 0.0368, 0.0248, 0.0240, 0.0266, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:15:59,304 INFO [finetune.py:992] (1/2) Epoch 1, batch 3700, loss[loss=0.1535, simple_loss=0.243, pruned_loss=0.032, over 12112.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2576, pruned_loss=0.04044, over 2374726.94 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:16:00,021 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.124e+02 2.853e+02 3.301e+02 3.898e+02 6.815e+02, threshold=6.603e+02, percent-clipped=1.0 2023-05-15 16:16:14,383 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103722.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 16:16:27,697 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103740.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:16:35,398 INFO [finetune.py:992] (1/2) Epoch 1, batch 3750, loss[loss=0.1989, simple_loss=0.2852, pruned_loss=0.05633, over 12028.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2589, pruned_loss=0.04124, over 2371485.91 frames. ], batch size: 40, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:16:37,671 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103754.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:16:45,509 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:17:02,138 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103788.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:17:12,314 INFO [finetune.py:992] (1/2) Epoch 1, batch 3800, loss[loss=0.1784, simple_loss=0.2743, pruned_loss=0.04121, over 12057.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2593, pruned_loss=0.04148, over 2366274.65 frames. ], batch size: 40, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:17:12,531 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103801.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:17:13,033 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.959e+02 3.693e+02 4.385e+02 1.112e+03, threshold=7.387e+02, percent-clipped=2.0 2023-05-15 16:17:13,133 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103802.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:17:30,363 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103826.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:17:37,451 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103836.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:17:48,647 INFO [finetune.py:992] (1/2) Epoch 1, batch 3850, loss[loss=0.1893, simple_loss=0.2722, pruned_loss=0.05315, over 12119.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2593, pruned_loss=0.04161, over 2367351.03 frames. ], batch size: 38, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:18:07,101 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103876.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:18:24,773 INFO [finetune.py:992] (1/2) Epoch 1, batch 3900, loss[loss=0.1607, simple_loss=0.255, pruned_loss=0.03322, over 10627.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2591, pruned_loss=0.04158, over 2365388.46 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:18:25,491 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.744e+02 3.382e+02 4.503e+02 9.689e+02, threshold=6.763e+02, percent-clipped=2.0 2023-05-15 16:18:31,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2023-05-15 16:18:50,820 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103937.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:19:01,526 INFO [finetune.py:992] (1/2) Epoch 1, batch 3950, loss[loss=0.191, simple_loss=0.2743, pruned_loss=0.05384, over 12136.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2598, pruned_loss=0.04208, over 2359223.13 frames. ], batch size: 38, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:19:29,863 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103990.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:19:35,515 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103997.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:19:41,600 INFO [finetune.py:992] (1/2) Epoch 1, batch 4000, loss[loss=0.1378, simple_loss=0.2213, pruned_loss=0.02711, over 12235.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2597, pruned_loss=0.04201, over 2365395.44 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:19:42,215 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.918e+02 3.442e+02 3.941e+02 5.823e+02, threshold=6.884e+02, percent-clipped=0.0 2023-05-15 16:19:56,597 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104022.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 16:19:56,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3748, 4.6775, 2.8638, 2.2622, 4.1545, 2.3402, 4.0173, 3.1299], device='cuda:1'), covar=tensor([0.0580, 0.0442, 0.0958, 0.1631, 0.0248, 0.1233, 0.0374, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0251, 0.0174, 0.0198, 0.0139, 0.0176, 0.0195, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:20:16,031 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104049.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:20:17,282 INFO [finetune.py:992] (1/2) Epoch 1, batch 4050, loss[loss=0.159, simple_loss=0.2543, pruned_loss=0.03187, over 12204.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2591, pruned_loss=0.04144, over 2368420.45 frames. ], batch size: 35, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:20:17,550 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104051.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:20:22,654 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104058.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:20:31,156 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104070.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:20:50,336 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104096.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:20:53,833 INFO [finetune.py:992] (1/2) Epoch 1, batch 4100, loss[loss=0.1796, simple_loss=0.2757, pruned_loss=0.04168, over 12347.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2586, pruned_loss=0.04117, over 2379543.60 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:20:54,470 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.086e+02 2.927e+02 3.555e+02 4.282e+02 8.168e+02, threshold=7.110e+02, percent-clipped=2.0 2023-05-15 16:21:00,267 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104110.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:21:08,133 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104121.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:21:17,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 2023-05-15 16:21:27,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7137, 3.7076, 3.4335, 3.2601, 3.0795, 2.7992, 3.6705, 2.3660], device='cuda:1'), covar=tensor([0.0280, 0.0116, 0.0116, 0.0166, 0.0282, 0.0282, 0.0084, 0.0395], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0159, 0.0152, 0.0179, 0.0201, 0.0198, 0.0156, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:21:30,619 INFO [finetune.py:992] (1/2) Epoch 1, batch 4150, loss[loss=0.1729, simple_loss=0.263, pruned_loss=0.04134, over 11702.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2596, pruned_loss=0.04161, over 2371750.11 frames. ], batch size: 48, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:21:36,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-15 16:21:52,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 16:22:06,921 INFO [finetune.py:992] (1/2) Epoch 1, batch 4200, loss[loss=0.1818, simple_loss=0.2674, pruned_loss=0.0481, over 10963.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2595, pruned_loss=0.042, over 2364496.90 frames. ], batch size: 70, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:22:07,611 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 2.892e+02 3.381e+02 4.316e+02 9.943e+02, threshold=6.762e+02, percent-clipped=2.0 2023-05-15 16:22:18,495 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3858, 2.1808, 3.8244, 4.3866, 3.9861, 4.3512, 3.9553, 2.8489], device='cuda:1'), covar=tensor([0.0034, 0.0482, 0.0105, 0.0034, 0.0089, 0.0070, 0.0098, 0.0395], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0122, 0.0107, 0.0077, 0.0103, 0.0112, 0.0088, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:22:22,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-15 16:22:28,935 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104232.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:22:30,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-05-15 16:22:43,273 INFO [finetune.py:992] (1/2) Epoch 1, batch 4250, loss[loss=0.1899, simple_loss=0.2824, pruned_loss=0.04872, over 11164.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2597, pruned_loss=0.04198, over 2367750.16 frames. ], batch size: 55, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:23:19,535 INFO [finetune.py:992] (1/2) Epoch 1, batch 4300, loss[loss=0.1547, simple_loss=0.2547, pruned_loss=0.0274, over 12055.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2605, pruned_loss=0.04207, over 2371920.78 frames. ], batch size: 37, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:23:20,204 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.993e+02 3.570e+02 4.476e+02 1.466e+03, threshold=7.140e+02, percent-clipped=3.0 2023-05-15 16:23:52,404 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104346.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:23:55,866 INFO [finetune.py:992] (1/2) Epoch 1, batch 4350, loss[loss=0.1591, simple_loss=0.2445, pruned_loss=0.03681, over 12173.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2606, pruned_loss=0.04186, over 2378247.14 frames. ], batch size: 29, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:23:57,361 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:24:29,166 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104396.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:24:32,907 INFO [finetune.py:992] (1/2) Epoch 1, batch 4400, loss[loss=0.1494, simple_loss=0.2312, pruned_loss=0.0338, over 11378.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2602, pruned_loss=0.04179, over 2382678.93 frames. ], batch size: 25, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:24:33,594 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.824e+02 3.318e+02 3.804e+02 6.419e+02, threshold=6.637e+02, percent-clipped=0.0 2023-05-15 16:24:35,794 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104405.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:24:47,380 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104421.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:25:04,580 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104444.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:25:09,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-15 16:25:09,477 INFO [finetune.py:992] (1/2) Epoch 1, batch 4450, loss[loss=0.1806, simple_loss=0.2688, pruned_loss=0.04616, over 11779.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2595, pruned_loss=0.04128, over 2387249.02 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:25:19,232 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-15 16:25:22,510 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104469.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:25:45,044 INFO [finetune.py:992] (1/2) Epoch 1, batch 4500, loss[loss=0.1624, simple_loss=0.2495, pruned_loss=0.03764, over 12179.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2595, pruned_loss=0.04149, over 2380369.68 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:25:45,761 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.830e+02 3.188e+02 3.845e+02 1.100e+03, threshold=6.377e+02, percent-clipped=3.0 2023-05-15 16:26:07,372 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104532.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 16:26:21,451 INFO [finetune.py:992] (1/2) Epoch 1, batch 4550, loss[loss=0.1569, simple_loss=0.2428, pruned_loss=0.03551, over 12086.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2599, pruned_loss=0.04165, over 2377660.43 frames. ], batch size: 32, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:26:32,962 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3618, 5.1284, 5.2440, 5.3087, 4.9080, 4.9403, 4.7741, 5.2910], device='cuda:1'), covar=tensor([0.0524, 0.0545, 0.0770, 0.0495, 0.1933, 0.1256, 0.0488, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0653, 0.0556, 0.0623, 0.0824, 0.0733, 0.0533, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0006, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 16:26:42,701 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104580.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:26:50,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-15 16:26:58,146 INFO [finetune.py:992] (1/2) Epoch 1, batch 4600, loss[loss=0.1645, simple_loss=0.2451, pruned_loss=0.04198, over 12130.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2602, pruned_loss=0.04202, over 2373009.79 frames. ], batch size: 30, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:26:58,777 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.795e+02 3.549e+02 4.331e+02 1.029e+03, threshold=7.098e+02, percent-clipped=4.0 2023-05-15 16:27:02,674 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7732, 3.7699, 3.3760, 3.3369, 2.9098, 2.9629, 3.7383, 2.5214], device='cuda:1'), covar=tensor([0.0294, 0.0119, 0.0153, 0.0169, 0.0375, 0.0299, 0.0095, 0.0402], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0164, 0.0156, 0.0184, 0.0208, 0.0205, 0.0161, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:27:30,615 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104646.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:27:34,054 INFO [finetune.py:992] (1/2) Epoch 1, batch 4650, loss[loss=0.1829, simple_loss=0.2657, pruned_loss=0.05001, over 12163.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2605, pruned_loss=0.04209, over 2377283.30 frames. ], batch size: 39, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:27:35,717 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104653.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:28:05,976 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104694.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:28:10,893 INFO [finetune.py:992] (1/2) Epoch 1, batch 4700, loss[loss=0.1597, simple_loss=0.2453, pruned_loss=0.03706, over 12022.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2592, pruned_loss=0.04154, over 2379931.37 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:28:10,953 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104701.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:28:11,572 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.857e+02 3.257e+02 4.088e+02 7.241e+02, threshold=6.513e+02, percent-clipped=1.0 2023-05-15 16:28:13,920 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104705.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:28:20,428 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4287, 4.8137, 2.9718, 2.7700, 4.1551, 2.8477, 4.1687, 3.3418], device='cuda:1'), covar=tensor([0.0530, 0.0346, 0.0882, 0.1202, 0.0214, 0.0934, 0.0340, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0250, 0.0175, 0.0197, 0.0139, 0.0178, 0.0196, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:28:40,638 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7312, 4.9356, 3.1480, 2.4815, 4.4269, 2.6163, 4.2039, 3.2847], device='cuda:1'), covar=tensor([0.0417, 0.0390, 0.0775, 0.1534, 0.0238, 0.1215, 0.0377, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0250, 0.0174, 0.0197, 0.0138, 0.0177, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:28:47,411 INFO [finetune.py:992] (1/2) Epoch 1, batch 4750, loss[loss=0.1599, simple_loss=0.2559, pruned_loss=0.03189, over 12305.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2591, pruned_loss=0.04167, over 2382378.11 frames. ], batch size: 34, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:28:48,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-15 16:28:48,940 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104753.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:29:19,033 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0329, 2.1421, 3.4084, 4.0267, 3.7110, 3.9944, 3.6643, 2.7846], device='cuda:1'), covar=tensor([0.0037, 0.0413, 0.0134, 0.0035, 0.0099, 0.0078, 0.0084, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0122, 0.0106, 0.0077, 0.0102, 0.0111, 0.0087, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:29:23,532 INFO [finetune.py:992] (1/2) Epoch 1, batch 4800, loss[loss=0.2014, simple_loss=0.2921, pruned_loss=0.05531, over 12102.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2605, pruned_loss=0.0422, over 2371719.91 frames. ], batch size: 38, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:29:24,199 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.849e+02 3.513e+02 4.430e+02 7.695e+02, threshold=7.026e+02, percent-clipped=3.0 2023-05-15 16:29:27,124 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7813, 3.9664, 3.6760, 4.2695, 3.9871, 2.7415, 3.7108, 2.9419], device='cuda:1'), covar=tensor([0.0763, 0.0880, 0.1239, 0.0523, 0.0983, 0.1531, 0.0994, 0.2703], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0381, 0.0356, 0.0266, 0.0367, 0.0269, 0.0339, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:29:59,425 INFO [finetune.py:992] (1/2) Epoch 1, batch 4850, loss[loss=0.2417, simple_loss=0.3218, pruned_loss=0.08084, over 7938.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.26, pruned_loss=0.04228, over 2366847.07 frames. ], batch size: 98, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:30:35,658 INFO [finetune.py:992] (1/2) Epoch 1, batch 4900, loss[loss=0.1498, simple_loss=0.2286, pruned_loss=0.03547, over 12256.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2603, pruned_loss=0.04223, over 2369075.05 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:30:36,272 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 3.072e+02 3.542e+02 4.222e+02 6.583e+02, threshold=7.085e+02, percent-clipped=0.0 2023-05-15 16:30:51,729 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.73 vs. limit=5.0 2023-05-15 16:31:11,169 INFO [finetune.py:992] (1/2) Epoch 1, batch 4950, loss[loss=0.2295, simple_loss=0.2998, pruned_loss=0.0796, over 7858.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2609, pruned_loss=0.04247, over 2358387.30 frames. ], batch size: 98, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:31:45,454 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 2023-05-15 16:31:48,108 INFO [finetune.py:992] (1/2) Epoch 1, batch 5000, loss[loss=0.1613, simple_loss=0.2502, pruned_loss=0.0362, over 12261.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2605, pruned_loss=0.04203, over 2362761.06 frames. ], batch size: 32, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:31:48,785 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.772e+02 3.423e+02 4.530e+02 7.566e+02, threshold=6.846e+02, percent-clipped=4.0 2023-05-15 16:32:24,728 INFO [finetune.py:992] (1/2) Epoch 1, batch 5050, loss[loss=0.2001, simple_loss=0.2839, pruned_loss=0.05814, over 7822.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2611, pruned_loss=0.04226, over 2355691.03 frames. ], batch size: 97, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:32:29,241 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7157, 2.6618, 3.3720, 4.4509, 2.4652, 4.6796, 4.5772, 4.8027], device='cuda:1'), covar=tensor([0.0085, 0.1053, 0.0396, 0.0117, 0.1061, 0.0165, 0.0118, 0.0060], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0204, 0.0193, 0.0117, 0.0188, 0.0182, 0.0173, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:32:29,985 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2179, 2.6264, 3.6968, 3.2091, 3.5927, 3.2062, 2.7468, 3.7529], device='cuda:1'), covar=tensor([0.0105, 0.0283, 0.0114, 0.0191, 0.0114, 0.0165, 0.0255, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0193, 0.0169, 0.0173, 0.0194, 0.0153, 0.0182, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:32:50,131 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-15 16:33:00,457 INFO [finetune.py:992] (1/2) Epoch 1, batch 5100, loss[loss=0.1494, simple_loss=0.2253, pruned_loss=0.03678, over 12020.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2607, pruned_loss=0.04223, over 2363783.70 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:33:01,105 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.904e+02 3.598e+02 4.298e+02 1.348e+03, threshold=7.197e+02, percent-clipped=3.0 2023-05-15 16:33:36,747 INFO [finetune.py:992] (1/2) Epoch 1, batch 5150, loss[loss=0.2179, simple_loss=0.2972, pruned_loss=0.06932, over 7654.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2606, pruned_loss=0.04231, over 2359773.46 frames. ], batch size: 98, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:33:43,436 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2975, 4.7027, 2.9439, 2.7592, 3.9270, 2.3767, 3.9825, 3.2217], device='cuda:1'), covar=tensor([0.0668, 0.0471, 0.0983, 0.1343, 0.0301, 0.1356, 0.0464, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0251, 0.0174, 0.0196, 0.0139, 0.0178, 0.0195, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:33:56,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.88 vs. limit=5.0 2023-05-15 16:34:13,370 INFO [finetune.py:992] (1/2) Epoch 1, batch 5200, loss[loss=0.1662, simple_loss=0.2577, pruned_loss=0.03738, over 10614.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2599, pruned_loss=0.04246, over 2356496.73 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:34:14,034 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.131e+02 2.916e+02 3.523e+02 4.295e+02 9.519e+02, threshold=7.045e+02, percent-clipped=2.0 2023-05-15 16:34:23,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-15 16:34:33,628 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0951, 3.5478, 3.6431, 4.0578, 2.8923, 3.5275, 2.3785, 3.5457], device='cuda:1'), covar=tensor([0.1575, 0.0755, 0.0915, 0.0584, 0.1059, 0.0640, 0.1842, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0265, 0.0306, 0.0368, 0.0246, 0.0241, 0.0262, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:34:47,455 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2106, 2.0444, 2.4724, 2.2540, 2.4603, 2.3147, 1.9235, 2.4734], device='cuda:1'), covar=tensor([0.0096, 0.0239, 0.0163, 0.0194, 0.0150, 0.0153, 0.0238, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0192, 0.0168, 0.0173, 0.0193, 0.0152, 0.0181, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:34:48,637 INFO [finetune.py:992] (1/2) Epoch 1, batch 5250, loss[loss=0.1612, simple_loss=0.2622, pruned_loss=0.03009, over 12140.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2605, pruned_loss=0.04268, over 2359679.52 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:35:24,857 INFO [finetune.py:992] (1/2) Epoch 1, batch 5300, loss[loss=0.1718, simple_loss=0.2578, pruned_loss=0.04292, over 12261.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2609, pruned_loss=0.04271, over 2365201.22 frames. ], batch size: 32, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:35:25,566 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.212e+02 2.987e+02 3.418e+02 4.113e+02 6.916e+02, threshold=6.836e+02, percent-clipped=0.0 2023-05-15 16:35:44,394 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7210, 3.9827, 3.6869, 3.5822, 3.2115, 3.3086, 3.9017, 2.6077], device='cuda:1'), covar=tensor([0.0300, 0.0091, 0.0125, 0.0119, 0.0251, 0.0205, 0.0086, 0.0354], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0162, 0.0154, 0.0180, 0.0205, 0.0202, 0.0161, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:36:01,576 INFO [finetune.py:992] (1/2) Epoch 1, batch 5350, loss[loss=0.1828, simple_loss=0.279, pruned_loss=0.04335, over 11796.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2607, pruned_loss=0.04304, over 2366262.39 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:36:11,871 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2227, 5.2268, 5.0919, 5.1516, 4.7753, 5.2276, 5.1679, 5.4240], device='cuda:1'), covar=tensor([0.0178, 0.0102, 0.0148, 0.0217, 0.0605, 0.0174, 0.0107, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0185, 0.0181, 0.0228, 0.0230, 0.0195, 0.0170, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 16:36:28,430 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105388.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:36:35,862 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2163, 3.9734, 4.2392, 4.6265, 3.2694, 4.0559, 2.8330, 4.0504], device='cuda:1'), covar=tensor([0.1610, 0.0798, 0.0811, 0.0581, 0.1093, 0.0594, 0.1660, 0.1390], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0266, 0.0305, 0.0369, 0.0246, 0.0240, 0.0262, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:36:38,030 INFO [finetune.py:992] (1/2) Epoch 1, batch 5400, loss[loss=0.1656, simple_loss=0.26, pruned_loss=0.03559, over 12023.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2595, pruned_loss=0.04268, over 2364162.75 frames. ], batch size: 42, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:36:38,764 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.689e+02 3.380e+02 4.059e+02 1.017e+03, threshold=6.760e+02, percent-clipped=1.0 2023-05-15 16:37:13,326 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105449.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 16:37:14,582 INFO [finetune.py:992] (1/2) Epoch 1, batch 5450, loss[loss=0.1936, simple_loss=0.2846, pruned_loss=0.05133, over 10203.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2605, pruned_loss=0.04298, over 2360740.79 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:37:37,032 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0120, 4.6668, 4.8083, 4.8822, 4.6033, 4.8575, 4.7633, 2.6539], device='cuda:1'), covar=tensor([0.0072, 0.0063, 0.0073, 0.0049, 0.0050, 0.0080, 0.0074, 0.0653], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0077, 0.0071, 0.0059, 0.0087, 0.0077, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:1') 2023-05-15 16:37:40,668 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105486.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:37:44,051 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7999, 5.7579, 5.5514, 5.1308, 5.0450, 5.7228, 5.2570, 5.0660], device='cuda:1'), covar=tensor([0.0806, 0.1071, 0.0682, 0.1365, 0.0804, 0.0728, 0.1758, 0.1090], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0518, 0.0472, 0.0588, 0.0376, 0.0672, 0.0733, 0.0533], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:37:50,078 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2507, 4.8860, 5.2241, 4.6334, 4.8860, 4.6824, 5.2589, 4.8971], device='cuda:1'), covar=tensor([0.0231, 0.0297, 0.0224, 0.0217, 0.0276, 0.0271, 0.0177, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0249, 0.0263, 0.0239, 0.0239, 0.0238, 0.0220, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:37:51,359 INFO [finetune.py:992] (1/2) Epoch 1, batch 5500, loss[loss=0.178, simple_loss=0.2727, pruned_loss=0.04161, over 12175.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2606, pruned_loss=0.04282, over 2363438.47 frames. ], batch size: 35, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:37:51,976 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 2.780e+02 3.412e+02 4.141e+02 8.244e+02, threshold=6.823e+02, percent-clipped=2.0 2023-05-15 16:37:52,164 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105502.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:38:04,195 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1408, 5.9447, 5.4460, 5.4778, 6.0458, 5.3865, 5.7198, 5.6312], device='cuda:1'), covar=tensor([0.1358, 0.0823, 0.1007, 0.1778, 0.0804, 0.1876, 0.1420, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0474, 0.0367, 0.0419, 0.0444, 0.0416, 0.0384, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:38:24,664 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105547.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:38:27,279 INFO [finetune.py:992] (1/2) Epoch 1, batch 5550, loss[loss=0.1741, simple_loss=0.2692, pruned_loss=0.03952, over 12156.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.261, pruned_loss=0.04328, over 2360680.97 frames. ], batch size: 36, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:38:35,957 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105563.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:38:56,929 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3687, 3.4679, 3.1791, 3.1257, 2.6826, 2.5555, 3.3982, 2.2467], device='cuda:1'), covar=tensor([0.0351, 0.0113, 0.0167, 0.0160, 0.0349, 0.0330, 0.0112, 0.0416], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0161, 0.0153, 0.0181, 0.0206, 0.0201, 0.0160, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:39:04,055 INFO [finetune.py:992] (1/2) Epoch 1, batch 5600, loss[loss=0.237, simple_loss=0.3018, pruned_loss=0.0861, over 8567.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.261, pruned_loss=0.04324, over 2363414.62 frames. ], batch size: 98, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:39:04,724 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.813e+02 3.398e+02 3.932e+02 6.738e+02, threshold=6.795e+02, percent-clipped=0.0 2023-05-15 16:39:39,739 INFO [finetune.py:992] (1/2) Epoch 1, batch 5650, loss[loss=0.1846, simple_loss=0.2672, pruned_loss=0.05103, over 11998.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2608, pruned_loss=0.04327, over 2368034.84 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:39:50,924 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.80 vs. limit=5.0 2023-05-15 16:40:09,394 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0606, 6.0594, 5.8567, 5.2848, 5.1370, 5.9764, 5.5662, 5.2968], device='cuda:1'), covar=tensor([0.0696, 0.0823, 0.0544, 0.1272, 0.0619, 0.0622, 0.1323, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0513, 0.0470, 0.0588, 0.0372, 0.0666, 0.0731, 0.0531], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:40:16,554 INFO [finetune.py:992] (1/2) Epoch 1, batch 5700, loss[loss=0.1709, simple_loss=0.2453, pruned_loss=0.04822, over 11809.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2621, pruned_loss=0.04403, over 2359858.77 frames. ], batch size: 26, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:40:17,228 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.090e+02 2.924e+02 3.415e+02 4.302e+02 8.366e+02, threshold=6.830e+02, percent-clipped=2.0 2023-05-15 16:40:48,296 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105744.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 16:40:50,520 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7415, 4.4236, 4.5300, 4.5664, 4.3825, 4.6152, 4.5905, 2.5333], device='cuda:1'), covar=tensor([0.0086, 0.0069, 0.0088, 0.0064, 0.0064, 0.0085, 0.0075, 0.0686], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0076, 0.0078, 0.0072, 0.0060, 0.0088, 0.0078, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:1') 2023-05-15 16:40:50,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-15 16:40:53,144 INFO [finetune.py:992] (1/2) Epoch 1, batch 5750, loss[loss=0.1944, simple_loss=0.2783, pruned_loss=0.0553, over 11253.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2622, pruned_loss=0.04359, over 2370904.72 frames. ], batch size: 55, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:41:22,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105792.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:41:27,446 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105799.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:41:28,903 INFO [finetune.py:992] (1/2) Epoch 1, batch 5800, loss[loss=0.1436, simple_loss=0.2261, pruned_loss=0.03048, over 12010.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2625, pruned_loss=0.0436, over 2370885.49 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:41:29,591 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.183e+02 2.921e+02 3.626e+02 4.206e+02 7.946e+02, threshold=7.252e+02, percent-clipped=5.0 2023-05-15 16:41:34,758 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9022, 4.5679, 4.8462, 4.2997, 4.6012, 4.3720, 4.8646, 4.4675], device='cuda:1'), covar=tensor([0.0238, 0.0344, 0.0281, 0.0258, 0.0275, 0.0274, 0.0220, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0251, 0.0266, 0.0243, 0.0240, 0.0241, 0.0221, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1') 2023-05-15 16:41:59,186 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105842.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:42:05,522 INFO [finetune.py:992] (1/2) Epoch 1, batch 5850, loss[loss=0.1656, simple_loss=0.254, pruned_loss=0.03858, over 12303.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04374, over 2373039.00 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:42:07,148 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105853.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:42:10,775 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105858.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:42:12,320 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105860.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:42:33,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5721, 4.7810, 3.0045, 2.3159, 4.2322, 2.2141, 4.0921, 3.2442], device='cuda:1'), covar=tensor([0.0418, 0.0395, 0.0883, 0.1652, 0.0202, 0.1481, 0.0389, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0255, 0.0177, 0.0200, 0.0140, 0.0182, 0.0197, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:42:42,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.33 vs. limit=5.0 2023-05-15 16:42:42,334 INFO [finetune.py:992] (1/2) Epoch 1, batch 5900, loss[loss=0.1752, simple_loss=0.2666, pruned_loss=0.04192, over 10525.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2622, pruned_loss=0.04368, over 2379140.60 frames. ], batch size: 68, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:42:43,031 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.033e+02 2.888e+02 3.561e+02 4.128e+02 1.057e+03, threshold=7.123e+02, percent-clipped=1.0 2023-05-15 16:43:10,000 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5595, 2.7749, 3.2733, 4.4173, 2.4814, 4.5852, 4.4541, 4.6849], device='cuda:1'), covar=tensor([0.0107, 0.0994, 0.0441, 0.0119, 0.1086, 0.0174, 0.0134, 0.0073], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0191, 0.0115, 0.0187, 0.0180, 0.0172, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:43:17,583 INFO [finetune.py:992] (1/2) Epoch 1, batch 5950, loss[loss=0.1757, simple_loss=0.2565, pruned_loss=0.04746, over 12166.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04399, over 2379728.16 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:43:24,204 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1041, 3.9089, 2.5548, 2.2170, 3.4836, 2.2729, 3.5986, 2.7711], device='cuda:1'), covar=tensor([0.0639, 0.0634, 0.0961, 0.1479, 0.0307, 0.1247, 0.0442, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0254, 0.0176, 0.0200, 0.0140, 0.0181, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:43:37,246 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1862, 4.4863, 3.8685, 4.8879, 4.4183, 2.8572, 4.2230, 3.0204], device='cuda:1'), covar=tensor([0.0828, 0.0780, 0.1393, 0.0313, 0.1046, 0.1552, 0.0878, 0.2844], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0378, 0.0354, 0.0265, 0.0365, 0.0266, 0.0337, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:43:40,749 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4670, 2.6072, 3.1944, 4.3793, 2.3177, 4.4698, 4.4199, 4.6149], device='cuda:1'), covar=tensor([0.0106, 0.0993, 0.0436, 0.0100, 0.1112, 0.0180, 0.0108, 0.0062], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0191, 0.0116, 0.0187, 0.0180, 0.0172, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:43:48,447 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8567, 3.7520, 3.7545, 3.8410, 3.4618, 3.8514, 3.8176, 3.9588], device='cuda:1'), covar=tensor([0.0223, 0.0169, 0.0200, 0.0290, 0.0735, 0.0299, 0.0190, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0185, 0.0183, 0.0229, 0.0233, 0.0198, 0.0171, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 16:43:57,612 INFO [finetune.py:992] (1/2) Epoch 1, batch 6000, loss[loss=0.1768, simple_loss=0.2719, pruned_loss=0.04089, over 11642.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2629, pruned_loss=0.04436, over 2372990.52 frames. ], batch size: 48, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:43:57,612 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 16:44:17,038 INFO [finetune.py:1026] (1/2) Epoch 1, validation: loss=0.3651, simple_loss=0.4266, pruned_loss=0.1518, over 1020973.00 frames. 2023-05-15 16:44:17,039 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 16:44:17,700 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.870e+02 3.370e+02 4.278e+02 8.951e+02, threshold=6.741e+02, percent-clipped=4.0 2023-05-15 16:44:48,384 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106044.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:44:53,233 INFO [finetune.py:992] (1/2) Epoch 1, batch 6050, loss[loss=0.1739, simple_loss=0.2603, pruned_loss=0.04371, over 12287.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2625, pruned_loss=0.04418, over 2367028.99 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 16.0 2023-05-15 16:45:07,325 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-15 16:45:12,262 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106077.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:45:23,597 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106092.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:45:28,942 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1904, 4.3132, 4.2482, 4.6767, 3.1003, 4.0160, 2.8908, 4.0608], device='cuda:1'), covar=tensor([0.1716, 0.0610, 0.0846, 0.0526, 0.1167, 0.0593, 0.1643, 0.1657], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0263, 0.0302, 0.0363, 0.0244, 0.0239, 0.0259, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:45:30,129 INFO [finetune.py:992] (1/2) Epoch 1, batch 6100, loss[loss=0.1919, simple_loss=0.2694, pruned_loss=0.05721, over 11718.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2615, pruned_loss=0.04358, over 2375661.77 frames. ], batch size: 48, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:45:31,568 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.834e+02 3.255e+02 3.982e+02 7.817e+02, threshold=6.509e+02, percent-clipped=2.0 2023-05-15 16:45:56,473 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1262, 3.8867, 4.0582, 4.3585, 3.1359, 3.9239, 2.5360, 4.1204], device='cuda:1'), covar=tensor([0.1694, 0.0740, 0.0855, 0.0672, 0.1033, 0.0615, 0.1877, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0263, 0.0303, 0.0364, 0.0244, 0.0239, 0.0260, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:45:57,107 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6139, 4.5600, 4.5242, 4.5393, 4.1513, 4.5724, 4.5681, 4.7921], device='cuda:1'), covar=tensor([0.0237, 0.0138, 0.0163, 0.0270, 0.0750, 0.0315, 0.0165, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0184, 0.0181, 0.0227, 0.0231, 0.0197, 0.0169, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 16:45:57,163 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106138.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:00,615 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106142.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:04,912 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106148.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:06,972 INFO [finetune.py:992] (1/2) Epoch 1, batch 6150, loss[loss=0.1739, simple_loss=0.2571, pruned_loss=0.04533, over 12285.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2617, pruned_loss=0.04354, over 2374619.76 frames. ], batch size: 33, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:46:09,911 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106155.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:12,150 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106158.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:34,895 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106190.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:43,159 INFO [finetune.py:992] (1/2) Epoch 1, batch 6200, loss[loss=0.1968, simple_loss=0.2744, pruned_loss=0.05966, over 11870.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2625, pruned_loss=0.04393, over 2368020.58 frames. ], batch size: 44, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:46:44,521 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.121e+02 2.817e+02 3.317e+02 3.981e+02 7.574e+02, threshold=6.635e+02, percent-clipped=2.0 2023-05-15 16:46:46,764 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106206.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:46:57,062 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2473, 5.1553, 5.2107, 5.2917, 4.9202, 4.9270, 4.8050, 5.2327], device='cuda:1'), covar=tensor([0.0620, 0.0421, 0.0655, 0.0446, 0.1469, 0.1199, 0.0432, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0632, 0.0545, 0.0602, 0.0792, 0.0713, 0.0525, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0006, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 16:47:09,144 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7534, 3.6792, 3.4363, 3.3814, 3.0377, 2.9899, 3.7150, 2.6267], device='cuda:1'), covar=tensor([0.0270, 0.0200, 0.0156, 0.0150, 0.0278, 0.0291, 0.0102, 0.0307], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0160, 0.0153, 0.0178, 0.0201, 0.0197, 0.0158, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:47:19,652 INFO [finetune.py:992] (1/2) Epoch 1, batch 6250, loss[loss=0.1707, simple_loss=0.2438, pruned_loss=0.04885, over 11998.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2619, pruned_loss=0.04365, over 2372518.44 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:47:31,833 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106268.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:47:55,376 INFO [finetune.py:992] (1/2) Epoch 1, batch 6300, loss[loss=0.1997, simple_loss=0.2961, pruned_loss=0.05162, over 11450.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2636, pruned_loss=0.04434, over 2371305.01 frames. ], batch size: 55, lr: 5.00e-03, grad_scale: 8.0 2023-05-15 16:47:56,626 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.327e+02 3.079e+02 3.553e+02 4.255e+02 1.115e+03, threshold=7.105e+02, percent-clipped=3.0 2023-05-15 16:48:08,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-15 16:48:15,422 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106329.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:48:20,403 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106336.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:48:23,357 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3700, 4.6870, 2.8677, 2.6336, 3.9183, 2.7177, 3.9651, 3.3048], device='cuda:1'), covar=tensor([0.0606, 0.0513, 0.1081, 0.1426, 0.0268, 0.1133, 0.0449, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0252, 0.0175, 0.0198, 0.0138, 0.0180, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 16:48:30,962 INFO [finetune.py:992] (1/2) Epoch 1, batch 6350, loss[loss=0.194, simple_loss=0.2875, pruned_loss=0.0503, over 11619.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2629, pruned_loss=0.04381, over 2374883.16 frames. ], batch size: 48, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:48:41,861 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3141, 3.2415, 3.1854, 3.1131, 2.7742, 2.6979, 3.3406, 2.1039], device='cuda:1'), covar=tensor([0.0307, 0.0128, 0.0132, 0.0136, 0.0309, 0.0263, 0.0099, 0.0379], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0160, 0.0153, 0.0180, 0.0201, 0.0198, 0.0159, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:48:44,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-15 16:49:04,713 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106397.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:49:07,692 INFO [finetune.py:992] (1/2) Epoch 1, batch 6400, loss[loss=0.1791, simple_loss=0.2787, pruned_loss=0.03977, over 11263.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2632, pruned_loss=0.04392, over 2370328.14 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:49:09,152 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.851e+02 3.219e+02 3.942e+02 7.404e+02, threshold=6.437e+02, percent-clipped=1.0 2023-05-15 16:49:25,407 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7708, 2.7238, 4.6955, 5.0621, 3.5021, 2.6434, 3.0650, 1.9757], device='cuda:1'), covar=tensor([0.1286, 0.2962, 0.0338, 0.0238, 0.0772, 0.1859, 0.2274, 0.4088], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0373, 0.0264, 0.0291, 0.0252, 0.0279, 0.0356, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:49:30,883 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106433.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:49:42,322 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106448.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:49:44,366 INFO [finetune.py:992] (1/2) Epoch 1, batch 6450, loss[loss=0.2127, simple_loss=0.2928, pruned_loss=0.06625, over 10553.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04356, over 2371688.78 frames. ], batch size: 69, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:49:46,716 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106454.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:49:47,371 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106455.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:50:12,208 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2984, 4.6039, 4.0694, 4.9679, 4.4807, 2.8774, 4.2513, 3.1021], device='cuda:1'), covar=tensor([0.0802, 0.0773, 0.1428, 0.0362, 0.1063, 0.1589, 0.1047, 0.3056], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0380, 0.0355, 0.0266, 0.0367, 0.0266, 0.0338, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:50:16,410 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106496.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:50:19,976 INFO [finetune.py:992] (1/2) Epoch 1, batch 6500, loss[loss=0.16, simple_loss=0.2504, pruned_loss=0.03483, over 12285.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2625, pruned_loss=0.04336, over 2374371.48 frames. ], batch size: 33, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:50:21,365 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.770e+02 3.404e+02 3.913e+02 8.890e+02, threshold=6.807e+02, percent-clipped=2.0 2023-05-15 16:50:21,452 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106503.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:50:30,190 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106515.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:50:56,140 INFO [finetune.py:992] (1/2) Epoch 1, batch 6550, loss[loss=0.2137, simple_loss=0.2977, pruned_loss=0.06484, over 12068.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2635, pruned_loss=0.0438, over 2367915.67 frames. ], batch size: 42, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:51:32,711 INFO [finetune.py:992] (1/2) Epoch 1, batch 6600, loss[loss=0.1875, simple_loss=0.2608, pruned_loss=0.05705, over 11873.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04371, over 2373672.03 frames. ], batch size: 26, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:51:34,109 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.440e+02 3.236e+02 3.772e+02 4.434e+02 3.589e+03, threshold=7.544e+02, percent-clipped=5.0 2023-05-15 16:51:49,167 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106624.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:51:50,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-15 16:51:59,120 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5124, 5.4905, 5.2784, 4.9297, 4.8736, 5.4374, 5.0841, 4.8134], device='cuda:1'), covar=tensor([0.0640, 0.0752, 0.0623, 0.1180, 0.0865, 0.0714, 0.1289, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0514, 0.0467, 0.0585, 0.0375, 0.0668, 0.0723, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:52:03,565 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6312, 4.7837, 4.3213, 5.0729, 4.8172, 3.2232, 4.5663, 3.3827], device='cuda:1'), covar=tensor([0.0592, 0.0637, 0.1190, 0.0369, 0.0821, 0.1327, 0.0761, 0.2614], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0378, 0.0353, 0.0264, 0.0364, 0.0265, 0.0335, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:52:07,767 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6040, 2.1713, 3.0208, 2.5725, 2.8449, 2.8067, 2.2268, 2.9580], device='cuda:1'), covar=tensor([0.0100, 0.0284, 0.0133, 0.0184, 0.0130, 0.0129, 0.0245, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0194, 0.0169, 0.0174, 0.0193, 0.0152, 0.0182, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:52:08,319 INFO [finetune.py:992] (1/2) Epoch 1, batch 6650, loss[loss=0.1783, simple_loss=0.2733, pruned_loss=0.04165, over 10701.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2636, pruned_loss=0.04383, over 2373739.83 frames. ], batch size: 68, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:52:38,327 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106692.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:52:44,672 INFO [finetune.py:992] (1/2) Epoch 1, batch 6700, loss[loss=0.1397, simple_loss=0.22, pruned_loss=0.02974, over 11993.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2634, pruned_loss=0.04342, over 2375584.91 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:52:45,605 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1005, 2.1579, 2.7258, 3.0913, 1.9970, 3.2422, 3.1217, 3.2428], device='cuda:1'), covar=tensor([0.0190, 0.1027, 0.0443, 0.0168, 0.1009, 0.0271, 0.0260, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0202, 0.0189, 0.0115, 0.0186, 0.0181, 0.0170, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:52:46,071 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.078e+02 3.048e+02 3.636e+02 4.318e+02 9.262e+02, threshold=7.272e+02, percent-clipped=2.0 2023-05-15 16:53:08,308 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106733.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:53:21,020 INFO [finetune.py:992] (1/2) Epoch 1, batch 6750, loss[loss=0.1501, simple_loss=0.2364, pruned_loss=0.03189, over 12081.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2632, pruned_loss=0.04351, over 2367152.47 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:53:40,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-15 16:53:42,613 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106781.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:53:56,872 INFO [finetune.py:992] (1/2) Epoch 1, batch 6800, loss[loss=0.1802, simple_loss=0.2752, pruned_loss=0.04255, over 12190.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2634, pruned_loss=0.04341, over 2371777.48 frames. ], batch size: 35, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:53:57,155 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5163, 4.8286, 4.4086, 5.1664, 4.8065, 3.1992, 4.5350, 3.2987], device='cuda:1'), covar=tensor([0.0682, 0.0663, 0.1080, 0.0270, 0.0913, 0.1312, 0.0763, 0.2780], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0378, 0.0354, 0.0263, 0.0364, 0.0265, 0.0335, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:53:58,277 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.775e+02 3.272e+02 4.008e+02 8.494e+02, threshold=6.545e+02, percent-clipped=1.0 2023-05-15 16:54:03,950 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:54:24,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-05-15 16:54:27,549 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6041, 3.7572, 3.3431, 3.4111, 3.0818, 2.9582, 3.7490, 2.5070], device='cuda:1'), covar=tensor([0.0337, 0.0121, 0.0183, 0.0163, 0.0338, 0.0298, 0.0113, 0.0380], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0160, 0.0154, 0.0179, 0.0201, 0.0197, 0.0160, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 16:54:33,087 INFO [finetune.py:992] (1/2) Epoch 1, batch 6850, loss[loss=0.1959, simple_loss=0.2895, pruned_loss=0.05115, over 12124.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2624, pruned_loss=0.04311, over 2378316.65 frames. ], batch size: 38, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:55:09,347 INFO [finetune.py:992] (1/2) Epoch 1, batch 6900, loss[loss=0.1829, simple_loss=0.2715, pruned_loss=0.0471, over 11221.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2623, pruned_loss=0.04312, over 2381537.69 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:55:10,777 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.008e+02 2.903e+02 3.472e+02 4.212e+02 6.915e+02, threshold=6.944e+02, percent-clipped=1.0 2023-05-15 16:55:13,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2023-05-15 16:55:25,415 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-15 16:55:25,825 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106924.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:55:44,747 INFO [finetune.py:992] (1/2) Epoch 1, batch 6950, loss[loss=0.1885, simple_loss=0.2746, pruned_loss=0.05122, over 10648.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2625, pruned_loss=0.04323, over 2377984.45 frames. ], batch size: 68, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:55:54,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-15 16:56:00,676 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106972.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:56:15,025 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106992.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:56:21,515 INFO [finetune.py:992] (1/2) Epoch 1, batch 7000, loss[loss=0.1646, simple_loss=0.2659, pruned_loss=0.0317, over 12276.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2632, pruned_loss=0.04343, over 2379399.23 frames. ], batch size: 37, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:56:22,938 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.089e+02 2.961e+02 3.430e+02 4.386e+02 1.005e+03, threshold=6.860e+02, percent-clipped=4.0 2023-05-15 16:56:37,948 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3499, 4.8977, 5.3166, 4.6402, 4.9181, 4.8439, 5.3346, 4.9991], device='cuda:1'), covar=tensor([0.0232, 0.0344, 0.0241, 0.0254, 0.0285, 0.0219, 0.0200, 0.0253], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0248, 0.0261, 0.0237, 0.0238, 0.0238, 0.0218, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 16:56:49,834 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107040.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:56:57,648 INFO [finetune.py:992] (1/2) Epoch 1, batch 7050, loss[loss=0.1666, simple_loss=0.2543, pruned_loss=0.03948, over 12247.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.263, pruned_loss=0.04385, over 2374217.75 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:57:33,393 INFO [finetune.py:992] (1/2) Epoch 1, batch 7100, loss[loss=0.1679, simple_loss=0.2556, pruned_loss=0.0401, over 12031.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2637, pruned_loss=0.04419, over 2372869.98 frames. ], batch size: 31, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:57:34,804 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.881e+02 3.375e+02 4.211e+02 1.027e+03, threshold=6.751e+02, percent-clipped=1.0 2023-05-15 16:57:40,059 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107110.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:58:01,627 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107140.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:58:09,323 INFO [finetune.py:992] (1/2) Epoch 1, batch 7150, loss[loss=0.217, simple_loss=0.2965, pruned_loss=0.06875, over 12104.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.263, pruned_loss=0.04376, over 2374206.00 frames. ], batch size: 38, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:58:10,230 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107152.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:58:14,437 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107158.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 16:58:42,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-15 16:58:46,314 INFO [finetune.py:992] (1/2) Epoch 1, batch 7200, loss[loss=0.1699, simple_loss=0.2532, pruned_loss=0.04333, over 12242.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04379, over 2368235.12 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:58:46,575 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107201.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 16:58:47,844 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.124e+02 2.811e+02 3.297e+02 3.942e+02 8.760e+02, threshold=6.593e+02, percent-clipped=1.0 2023-05-15 16:58:55,081 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107213.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 16:59:03,255 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1196, 3.6212, 5.3289, 2.7510, 2.9806, 4.1062, 3.4886, 4.0863], device='cuda:1'), covar=tensor([0.0276, 0.0917, 0.0199, 0.1043, 0.1665, 0.1078, 0.1157, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0234, 0.0243, 0.0182, 0.0240, 0.0287, 0.0231, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 16:59:21,834 INFO [finetune.py:992] (1/2) Epoch 1, batch 7250, loss[loss=0.1828, simple_loss=0.2691, pruned_loss=0.04824, over 12157.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2638, pruned_loss=0.04409, over 2365734.62 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:59:57,167 INFO [finetune.py:992] (1/2) Epoch 1, batch 7300, loss[loss=0.2164, simple_loss=0.3061, pruned_loss=0.06337, over 11346.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2643, pruned_loss=0.04442, over 2362808.51 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 16:59:58,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 2.821e+02 3.345e+02 4.009e+02 7.143e+02, threshold=6.691e+02, percent-clipped=2.0 2023-05-15 17:00:04,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-15 17:00:09,663 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0973, 4.4137, 4.0589, 4.8531, 4.3642, 2.7307, 3.9898, 3.0294], device='cuda:1'), covar=tensor([0.0794, 0.0767, 0.1192, 0.0334, 0.1062, 0.1656, 0.1050, 0.2836], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0375, 0.0350, 0.0263, 0.0361, 0.0263, 0.0331, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:00:13,698 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2238, 6.0139, 5.4938, 5.5581, 6.0629, 5.4602, 5.6962, 5.6191], device='cuda:1'), covar=tensor([0.1363, 0.0792, 0.0998, 0.1854, 0.0875, 0.1921, 0.1448, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0474, 0.0366, 0.0419, 0.0447, 0.0421, 0.0380, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:00:33,505 INFO [finetune.py:992] (1/2) Epoch 1, batch 7350, loss[loss=0.155, simple_loss=0.2433, pruned_loss=0.03338, over 12138.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2638, pruned_loss=0.04414, over 2367614.85 frames. ], batch size: 30, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:00:41,564 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2779, 2.5766, 3.7402, 3.1515, 3.6496, 3.1696, 2.6155, 3.7773], device='cuda:1'), covar=tensor([0.0090, 0.0271, 0.0114, 0.0186, 0.0097, 0.0160, 0.0276, 0.0078], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0189, 0.0166, 0.0170, 0.0190, 0.0148, 0.0179, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:01:02,657 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8232, 2.4748, 3.5230, 3.7290, 2.9895, 2.6900, 2.6369, 2.3563], device='cuda:1'), covar=tensor([0.0967, 0.2362, 0.0511, 0.0385, 0.0672, 0.1498, 0.2052, 0.2830], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0374, 0.0269, 0.0291, 0.0252, 0.0280, 0.0354, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:01:09,727 INFO [finetune.py:992] (1/2) Epoch 1, batch 7400, loss[loss=0.147, simple_loss=0.23, pruned_loss=0.03204, over 12274.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2651, pruned_loss=0.04512, over 2354005.52 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:01:11,044 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.953e+02 3.402e+02 4.337e+02 2.849e+03, threshold=6.805e+02, percent-clipped=4.0 2023-05-15 17:01:44,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 2023-05-15 17:01:45,478 INFO [finetune.py:992] (1/2) Epoch 1, batch 7450, loss[loss=0.1735, simple_loss=0.2652, pruned_loss=0.04088, over 12025.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2644, pruned_loss=0.04491, over 2359464.12 frames. ], batch size: 31, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:02:18,641 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107496.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:02:21,938 INFO [finetune.py:992] (1/2) Epoch 1, batch 7500, loss[loss=0.1999, simple_loss=0.2774, pruned_loss=0.06115, over 12084.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2638, pruned_loss=0.04488, over 2360241.24 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:02:23,395 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 3.043e+02 3.581e+02 4.065e+02 1.110e+03, threshold=7.161e+02, percent-clipped=3.0 2023-05-15 17:02:27,169 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107508.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 17:02:28,600 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107510.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:02:52,890 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.17 vs. limit=5.0 2023-05-15 17:02:58,038 INFO [finetune.py:992] (1/2) Epoch 1, batch 7550, loss[loss=0.1795, simple_loss=0.2755, pruned_loss=0.04178, over 12161.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2636, pruned_loss=0.04503, over 2359583.73 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:03:12,506 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107571.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:03:29,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-05-15 17:03:34,050 INFO [finetune.py:992] (1/2) Epoch 1, batch 7600, loss[loss=0.168, simple_loss=0.2619, pruned_loss=0.03702, over 12358.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.04444, over 2370809.22 frames. ], batch size: 35, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:03:35,431 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.960e+02 3.577e+02 4.566e+02 1.199e+03, threshold=7.155e+02, percent-clipped=4.0 2023-05-15 17:04:10,184 INFO [finetune.py:992] (1/2) Epoch 1, batch 7650, loss[loss=0.2068, simple_loss=0.2936, pruned_loss=0.05998, over 10664.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2638, pruned_loss=0.0452, over 2355698.03 frames. ], batch size: 68, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:04:12,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-05-15 17:04:14,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-05-15 17:04:16,140 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107659.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:04:46,010 INFO [finetune.py:992] (1/2) Epoch 1, batch 7700, loss[loss=0.1451, simple_loss=0.2325, pruned_loss=0.02884, over 12277.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2637, pruned_loss=0.04502, over 2359356.65 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:04:47,419 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 3.089e+02 3.767e+02 4.573e+02 1.081e+03, threshold=7.534e+02, percent-clipped=2.0 2023-05-15 17:04:59,958 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107720.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:05:12,036 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107737.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:05:21,971 INFO [finetune.py:992] (1/2) Epoch 1, batch 7750, loss[loss=0.1633, simple_loss=0.2692, pruned_loss=0.02869, over 12341.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2634, pruned_loss=0.04471, over 2366787.02 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:05:38,651 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-15 17:05:46,137 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-15 17:05:54,781 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107796.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:05:56,308 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107798.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 17:05:58,492 INFO [finetune.py:992] (1/2) Epoch 1, batch 7800, loss[loss=0.1583, simple_loss=0.2437, pruned_loss=0.03645, over 12126.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2637, pruned_loss=0.04474, over 2369845.89 frames. ], batch size: 30, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:05:59,954 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.891e+02 3.488e+02 4.279e+02 1.043e+03, threshold=6.975e+02, percent-clipped=3.0 2023-05-15 17:06:03,603 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107808.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:06:07,886 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2774, 4.5450, 2.7886, 2.5128, 3.8548, 2.2960, 3.9406, 2.9030], device='cuda:1'), covar=tensor([0.0618, 0.0528, 0.0955, 0.1387, 0.0232, 0.1314, 0.0391, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0254, 0.0177, 0.0199, 0.0141, 0.0182, 0.0195, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:06:14,835 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1369, 2.4155, 3.6469, 2.9814, 3.4112, 3.0873, 2.3790, 3.5778], device='cuda:1'), covar=tensor([0.0091, 0.0275, 0.0108, 0.0184, 0.0107, 0.0150, 0.0273, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0191, 0.0168, 0.0170, 0.0192, 0.0150, 0.0180, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:06:29,478 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107844.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:06:34,425 INFO [finetune.py:992] (1/2) Epoch 1, batch 7850, loss[loss=0.1471, simple_loss=0.2318, pruned_loss=0.03122, over 11991.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2645, pruned_loss=0.04508, over 2373097.27 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:06:37,933 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107856.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:06:45,078 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107866.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:07:10,030 INFO [finetune.py:992] (1/2) Epoch 1, batch 7900, loss[loss=0.1722, simple_loss=0.2527, pruned_loss=0.04582, over 12344.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2638, pruned_loss=0.04474, over 2371676.57 frames. ], batch size: 30, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:07:11,496 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.826e+02 3.338e+02 3.958e+02 7.049e+02, threshold=6.676e+02, percent-clipped=1.0 2023-05-15 17:07:13,930 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2486, 4.3883, 3.9916, 4.9510, 4.4540, 2.8599, 4.1552, 3.0068], device='cuda:1'), covar=tensor([0.0711, 0.0912, 0.1278, 0.0317, 0.0982, 0.1533, 0.0932, 0.3028], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0375, 0.0350, 0.0262, 0.0361, 0.0264, 0.0330, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:07:44,177 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1225, 2.7102, 3.7283, 3.1003, 3.5163, 3.1828, 2.5116, 3.6271], device='cuda:1'), covar=tensor([0.0108, 0.0295, 0.0144, 0.0199, 0.0114, 0.0159, 0.0305, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0192, 0.0169, 0.0172, 0.0193, 0.0151, 0.0181, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:07:46,145 INFO [finetune.py:992] (1/2) Epoch 1, batch 7950, loss[loss=0.1459, simple_loss=0.228, pruned_loss=0.03185, over 12224.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2639, pruned_loss=0.04487, over 2370336.05 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:08:13,721 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3099, 4.9047, 5.2828, 4.6815, 4.9248, 4.7636, 5.3117, 4.9285], device='cuda:1'), covar=tensor([0.0208, 0.0302, 0.0241, 0.0208, 0.0280, 0.0252, 0.0154, 0.0260], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0241, 0.0255, 0.0231, 0.0230, 0.0231, 0.0212, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 17:08:25,921 INFO [finetune.py:992] (1/2) Epoch 1, batch 8000, loss[loss=0.1906, simple_loss=0.284, pruned_loss=0.04862, over 11102.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2638, pruned_loss=0.04497, over 2361250.96 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:08:27,370 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.835e+02 3.457e+02 4.309e+02 8.802e+02, threshold=6.915e+02, percent-clipped=2.0 2023-05-15 17:08:35,767 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108015.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:09:01,599 INFO [finetune.py:992] (1/2) Epoch 1, batch 8050, loss[loss=0.1904, simple_loss=0.2761, pruned_loss=0.05239, over 11827.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2632, pruned_loss=0.04467, over 2367788.95 frames. ], batch size: 44, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:09:28,279 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2905, 4.6897, 2.7494, 2.6083, 3.9883, 2.5376, 4.0069, 3.1265], device='cuda:1'), covar=tensor([0.0627, 0.0398, 0.1049, 0.1380, 0.0249, 0.1189, 0.0384, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0251, 0.0175, 0.0196, 0.0141, 0.0179, 0.0192, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:09:32,367 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108093.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 17:09:37,837 INFO [finetune.py:992] (1/2) Epoch 1, batch 8100, loss[loss=0.154, simple_loss=0.2376, pruned_loss=0.03521, over 12315.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04457, over 2370386.67 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:09:39,366 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.748e+02 3.364e+02 4.081e+02 7.047e+02, threshold=6.728e+02, percent-clipped=1.0 2023-05-15 17:09:39,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-15 17:10:14,071 INFO [finetune.py:992] (1/2) Epoch 1, batch 8150, loss[loss=0.1809, simple_loss=0.2734, pruned_loss=0.04416, over 12115.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04461, over 2373876.73 frames. ], batch size: 38, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:10:24,909 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108166.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:10:44,228 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3060, 4.7357, 2.7416, 2.7875, 3.9936, 2.5783, 3.9140, 3.1345], device='cuda:1'), covar=tensor([0.0660, 0.0392, 0.1122, 0.1305, 0.0245, 0.1230, 0.0424, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0251, 0.0175, 0.0196, 0.0141, 0.0179, 0.0192, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:10:50,003 INFO [finetune.py:992] (1/2) Epoch 1, batch 8200, loss[loss=0.166, simple_loss=0.2575, pruned_loss=0.03722, over 12157.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.263, pruned_loss=0.04418, over 2375076.90 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:10:51,341 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.756e+02 3.279e+02 4.378e+02 1.560e+03, threshold=6.558e+02, percent-clipped=3.0 2023-05-15 17:10:59,733 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:11:26,359 INFO [finetune.py:992] (1/2) Epoch 1, batch 8250, loss[loss=0.1856, simple_loss=0.2774, pruned_loss=0.04691, over 11784.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.263, pruned_loss=0.04419, over 2375532.91 frames. ], batch size: 44, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:12:02,237 INFO [finetune.py:992] (1/2) Epoch 1, batch 8300, loss[loss=0.1796, simple_loss=0.2721, pruned_loss=0.04351, over 12364.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2644, pruned_loss=0.04502, over 2366504.37 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:12:03,699 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.245e+02 2.893e+02 3.434e+02 4.076e+02 8.565e+02, threshold=6.869e+02, percent-clipped=3.0 2023-05-15 17:12:12,307 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108315.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:12:19,415 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108325.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:12:28,661 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5343, 3.5777, 3.3337, 3.1958, 2.8600, 2.8526, 3.5715, 2.3917], device='cuda:1'), covar=tensor([0.0322, 0.0124, 0.0155, 0.0158, 0.0382, 0.0294, 0.0118, 0.0378], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0162, 0.0153, 0.0180, 0.0201, 0.0195, 0.0159, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:12:37,671 INFO [finetune.py:992] (1/2) Epoch 1, batch 8350, loss[loss=0.1561, simple_loss=0.241, pruned_loss=0.03565, over 12336.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2632, pruned_loss=0.04476, over 2366416.15 frames. ], batch size: 30, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:12:40,855 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0408, 4.1482, 4.0235, 4.4390, 3.0711, 3.9207, 2.6711, 4.0312], device='cuda:1'), covar=tensor([0.1640, 0.0624, 0.0875, 0.0692, 0.1031, 0.0598, 0.1728, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0264, 0.0303, 0.0364, 0.0243, 0.0241, 0.0259, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:12:46,230 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108363.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:12:48,459 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0632, 2.4921, 3.5658, 3.0534, 3.3456, 3.1085, 2.4241, 3.4916], device='cuda:1'), covar=tensor([0.0107, 0.0316, 0.0146, 0.0203, 0.0155, 0.0150, 0.0320, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0195, 0.0171, 0.0174, 0.0195, 0.0152, 0.0183, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:13:03,406 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108386.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:13:08,367 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108393.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:13:14,220 INFO [finetune.py:992] (1/2) Epoch 1, batch 8400, loss[loss=0.1872, simple_loss=0.2684, pruned_loss=0.05301, over 11249.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.263, pruned_loss=0.04492, over 2367577.77 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:13:16,188 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.951e+02 3.305e+02 4.157e+02 8.866e+02, threshold=6.611e+02, percent-clipped=2.0 2023-05-15 17:13:31,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-15 17:13:36,781 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108432.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:13:42,961 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108441.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:13:49,915 INFO [finetune.py:992] (1/2) Epoch 1, batch 8450, loss[loss=0.1689, simple_loss=0.2587, pruned_loss=0.03957, over 11828.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2638, pruned_loss=0.045, over 2370585.51 frames. ], batch size: 44, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:13:52,313 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1774, 2.6203, 3.7606, 3.1948, 3.5201, 3.2306, 2.6060, 3.6538], device='cuda:1'), covar=tensor([0.0087, 0.0276, 0.0107, 0.0183, 0.0120, 0.0145, 0.0293, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0193, 0.0169, 0.0173, 0.0193, 0.0150, 0.0181, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:14:20,080 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108493.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 17:14:25,546 INFO [finetune.py:992] (1/2) Epoch 1, batch 8500, loss[loss=0.1822, simple_loss=0.2787, pruned_loss=0.04284, over 12111.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2639, pruned_loss=0.04513, over 2369668.44 frames. ], batch size: 38, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:14:27,007 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 3.104e+02 3.426e+02 4.387e+02 8.910e+02, threshold=6.853e+02, percent-clipped=3.0 2023-05-15 17:14:50,935 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6585, 3.8517, 3.5141, 3.3142, 3.0920, 2.9137, 3.7737, 2.5832], device='cuda:1'), covar=tensor([0.0296, 0.0096, 0.0124, 0.0141, 0.0335, 0.0270, 0.0086, 0.0367], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0160, 0.0152, 0.0179, 0.0200, 0.0194, 0.0157, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:15:02,688 INFO [finetune.py:992] (1/2) Epoch 1, batch 8550, loss[loss=0.2892, simple_loss=0.3521, pruned_loss=0.1132, over 8290.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.264, pruned_loss=0.04484, over 2370420.13 frames. ], batch size: 98, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:15:15,722 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0750, 3.8925, 3.8272, 4.2730, 2.8854, 3.7514, 2.5571, 3.8676], device='cuda:1'), covar=tensor([0.1731, 0.0738, 0.1018, 0.0604, 0.1136, 0.0637, 0.1843, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0263, 0.0303, 0.0364, 0.0243, 0.0240, 0.0258, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:15:21,526 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108577.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:15:38,744 INFO [finetune.py:992] (1/2) Epoch 1, batch 8600, loss[loss=0.1814, simple_loss=0.2737, pruned_loss=0.04453, over 12142.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2633, pruned_loss=0.04485, over 2372725.81 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:15:40,128 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.862e+02 3.397e+02 4.304e+02 9.483e+02, threshold=6.794e+02, percent-clipped=4.0 2023-05-15 17:15:42,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-15 17:15:56,148 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6599, 2.8336, 3.8551, 4.6696, 4.2636, 4.5936, 4.0481, 3.3108], device='cuda:1'), covar=tensor([0.0021, 0.0335, 0.0099, 0.0028, 0.0082, 0.0062, 0.0087, 0.0284], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0120, 0.0105, 0.0076, 0.0099, 0.0111, 0.0087, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:16:05,558 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108638.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:16:14,953 INFO [finetune.py:992] (1/2) Epoch 1, batch 8650, loss[loss=0.2102, simple_loss=0.3062, pruned_loss=0.05708, over 11580.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2634, pruned_loss=0.04492, over 2364118.16 frames. ], batch size: 48, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:16:36,702 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108681.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:16:37,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-05-15 17:16:43,786 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9501, 4.9041, 4.7537, 4.8305, 4.4174, 4.8841, 4.9006, 5.1315], device='cuda:1'), covar=tensor([0.0194, 0.0118, 0.0173, 0.0231, 0.0740, 0.0247, 0.0144, 0.0145], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0185, 0.0184, 0.0226, 0.0232, 0.0198, 0.0169, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 17:16:51,190 INFO [finetune.py:992] (1/2) Epoch 1, batch 8700, loss[loss=0.1939, simple_loss=0.2844, pruned_loss=0.0517, over 12050.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2626, pruned_loss=0.04439, over 2373080.35 frames. ], batch size: 37, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:16:52,536 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.611e+02 3.200e+02 3.810e+02 1.059e+03, threshold=6.400e+02, percent-clipped=5.0 2023-05-15 17:17:26,424 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4358, 4.9751, 5.3933, 4.7564, 5.0022, 4.7884, 5.4447, 4.9625], device='cuda:1'), covar=tensor([0.0220, 0.0333, 0.0245, 0.0227, 0.0295, 0.0305, 0.0193, 0.0278], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0244, 0.0261, 0.0234, 0.0236, 0.0236, 0.0216, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 17:17:26,918 INFO [finetune.py:992] (1/2) Epoch 1, batch 8750, loss[loss=0.159, simple_loss=0.2494, pruned_loss=0.03427, over 12085.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2626, pruned_loss=0.04461, over 2369073.08 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:17:33,135 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-15 17:17:53,173 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108788.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 17:17:55,492 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108791.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:18:03,516 INFO [finetune.py:992] (1/2) Epoch 1, batch 8800, loss[loss=0.1593, simple_loss=0.248, pruned_loss=0.03527, over 12261.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.261, pruned_loss=0.044, over 2369478.31 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:18:04,938 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 2.996e+02 3.587e+02 4.204e+02 6.716e+02, threshold=7.173e+02, percent-clipped=2.0 2023-05-15 17:18:23,156 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6252, 2.7722, 4.4149, 4.6731, 3.0541, 2.5378, 2.8401, 2.0153], device='cuda:1'), covar=tensor([0.1306, 0.2762, 0.0411, 0.0327, 0.1022, 0.1905, 0.2376, 0.3603], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0370, 0.0266, 0.0288, 0.0251, 0.0275, 0.0351, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:18:34,397 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8087, 4.7458, 4.6179, 4.7600, 4.2944, 4.8892, 4.8010, 5.0081], device='cuda:1'), covar=tensor([0.0171, 0.0138, 0.0184, 0.0239, 0.0750, 0.0224, 0.0138, 0.0147], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0185, 0.0183, 0.0225, 0.0232, 0.0197, 0.0169, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 17:18:39,807 INFO [finetune.py:992] (1/2) Epoch 1, batch 8850, loss[loss=0.1454, simple_loss=0.2203, pruned_loss=0.03531, over 12271.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2616, pruned_loss=0.04382, over 2366224.36 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:18:40,754 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108852.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:18:58,698 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3637, 4.9017, 5.3129, 4.6693, 4.9593, 4.7157, 5.3520, 4.9940], device='cuda:1'), covar=tensor([0.0210, 0.0318, 0.0243, 0.0223, 0.0287, 0.0280, 0.0189, 0.0257], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0244, 0.0261, 0.0233, 0.0235, 0.0236, 0.0216, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 17:19:15,769 INFO [finetune.py:992] (1/2) Epoch 1, batch 8900, loss[loss=0.1569, simple_loss=0.2495, pruned_loss=0.03217, over 12027.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2613, pruned_loss=0.04357, over 2372815.10 frames. ], batch size: 31, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:19:17,230 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 3.094e+02 3.535e+02 4.369e+02 7.573e+02, threshold=7.071e+02, percent-clipped=1.0 2023-05-15 17:19:38,501 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108933.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:19:51,747 INFO [finetune.py:992] (1/2) Epoch 1, batch 8950, loss[loss=0.1699, simple_loss=0.2622, pruned_loss=0.03884, over 12287.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2616, pruned_loss=0.04379, over 2377059.93 frames. ], batch size: 33, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:20:12,936 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108981.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:20:22,083 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3156, 6.1044, 5.6848, 5.5695, 6.1723, 5.3862, 5.8158, 5.6780], device='cuda:1'), covar=tensor([0.1516, 0.0997, 0.0997, 0.2030, 0.0999, 0.2375, 0.1519, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0468, 0.0367, 0.0419, 0.0444, 0.0418, 0.0378, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:20:28,037 INFO [finetune.py:992] (1/2) Epoch 1, batch 9000, loss[loss=0.1926, simple_loss=0.279, pruned_loss=0.05311, over 11605.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.262, pruned_loss=0.0436, over 2377661.98 frames. ], batch size: 48, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:20:28,037 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 17:20:45,217 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1011, 4.5280, 2.7081, 2.4158, 4.0206, 2.4626, 3.9282, 2.8218], device='cuda:1'), covar=tensor([0.0594, 0.0215, 0.0944, 0.1524, 0.0127, 0.1111, 0.0292, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0252, 0.0176, 0.0199, 0.0141, 0.0179, 0.0193, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:20:46,577 INFO [finetune.py:1026] (1/2) Epoch 1, validation: loss=0.357, simple_loss=0.4207, pruned_loss=0.1467, over 1020973.00 frames. 2023-05-15 17:20:46,578 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 17:20:48,013 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.087e+02 2.872e+02 3.471e+02 4.611e+02 1.446e+03, threshold=6.943e+02, percent-clipped=2.0 2023-05-15 17:21:06,570 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109029.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:21:12,388 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109037.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 17:21:17,346 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0881, 5.0571, 4.8275, 5.1169, 3.9248, 5.2318, 5.0514, 5.2768], device='cuda:1'), covar=tensor([0.0251, 0.0152, 0.0207, 0.0221, 0.1213, 0.0249, 0.0177, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0183, 0.0182, 0.0225, 0.0231, 0.0196, 0.0169, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 17:21:22,769 INFO [finetune.py:992] (1/2) Epoch 1, batch 9050, loss[loss=0.2125, simple_loss=0.2961, pruned_loss=0.06444, over 12357.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2625, pruned_loss=0.04349, over 2388610.60 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:21:27,186 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2244, 5.0195, 5.0806, 5.1562, 4.7661, 4.8531, 4.7010, 5.0959], device='cuda:1'), covar=tensor([0.0546, 0.0559, 0.0704, 0.0555, 0.1782, 0.1275, 0.0526, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0630, 0.0536, 0.0593, 0.0784, 0.0713, 0.0522, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0006, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 17:21:42,908 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9250, 4.8959, 4.7021, 4.8284, 4.3554, 4.9315, 4.8806, 5.1160], device='cuda:1'), covar=tensor([0.0184, 0.0117, 0.0202, 0.0254, 0.0807, 0.0272, 0.0144, 0.0150], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0183, 0.0182, 0.0225, 0.0231, 0.0196, 0.0169, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 17:21:49,820 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109088.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:21:56,937 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109098.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:21:58,838 INFO [finetune.py:992] (1/2) Epoch 1, batch 9100, loss[loss=0.1626, simple_loss=0.2514, pruned_loss=0.03688, over 12283.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04396, over 2379302.74 frames. ], batch size: 33, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:22:00,289 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 3.137e+02 3.755e+02 4.480e+02 1.020e+03, threshold=7.511e+02, percent-clipped=5.0 2023-05-15 17:22:23,926 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109136.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:22:31,614 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109147.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:22:34,378 INFO [finetune.py:992] (1/2) Epoch 1, batch 9150, loss[loss=0.176, simple_loss=0.2734, pruned_loss=0.03931, over 11825.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2633, pruned_loss=0.04424, over 2380754.42 frames. ], batch size: 44, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:22:38,908 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8863, 3.5506, 5.2486, 2.8156, 2.8574, 4.0941, 3.3722, 4.0626], device='cuda:1'), covar=tensor([0.0380, 0.0916, 0.0218, 0.1035, 0.1731, 0.1119, 0.1238, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0226, 0.0237, 0.0175, 0.0232, 0.0279, 0.0224, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:23:05,509 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109194.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:23:10,695 INFO [finetune.py:992] (1/2) Epoch 1, batch 9200, loss[loss=0.2175, simple_loss=0.3044, pruned_loss=0.06532, over 11188.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2634, pruned_loss=0.0441, over 2375232.27 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:23:12,102 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.734e+02 3.400e+02 4.293e+02 1.083e+03, threshold=6.801e+02, percent-clipped=3.0 2023-05-15 17:23:34,104 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109233.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:23:46,645 INFO [finetune.py:992] (1/2) Epoch 1, batch 9250, loss[loss=0.1952, simple_loss=0.284, pruned_loss=0.05321, over 12269.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2641, pruned_loss=0.04454, over 2378035.99 frames. ], batch size: 37, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:23:49,792 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109255.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:23:59,045 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4344, 4.3321, 4.3435, 4.7563, 3.2300, 3.9794, 2.7796, 4.2298], device='cuda:1'), covar=tensor([0.1490, 0.0535, 0.0714, 0.0517, 0.1012, 0.0595, 0.1621, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0260, 0.0299, 0.0359, 0.0241, 0.0237, 0.0256, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:24:00,616 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-15 17:24:08,071 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109281.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:24:17,293 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109294.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:24:22,165 INFO [finetune.py:992] (1/2) Epoch 1, batch 9300, loss[loss=0.1928, simple_loss=0.2841, pruned_loss=0.05074, over 12038.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2645, pruned_loss=0.04425, over 2377555.10 frames. ], batch size: 40, lr: 4.99e-03, grad_scale: 16.0 2023-05-15 17:24:23,561 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.953e+02 3.400e+02 3.968e+02 5.850e+02, threshold=6.799e+02, percent-clipped=0.0 2023-05-15 17:24:28,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 2023-05-15 17:24:34,383 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109318.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:24:58,236 INFO [finetune.py:992] (1/2) Epoch 1, batch 9350, loss[loss=0.1877, simple_loss=0.282, pruned_loss=0.04675, over 12375.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2646, pruned_loss=0.04441, over 2378688.38 frames. ], batch size: 38, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:25:01,377 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109355.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:25:06,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 17:25:18,321 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109379.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 17:25:25,467 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 17:25:28,646 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109393.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 17:25:34,456 INFO [finetune.py:992] (1/2) Epoch 1, batch 9400, loss[loss=0.2025, simple_loss=0.2894, pruned_loss=0.05778, over 11173.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2642, pruned_loss=0.04418, over 2380277.46 frames. ], batch size: 55, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:25:36,644 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 3.155e+02 3.540e+02 4.244e+02 9.825e+02, threshold=7.080e+02, percent-clipped=5.0 2023-05-15 17:26:07,473 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109447.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:26:10,165 INFO [finetune.py:992] (1/2) Epoch 1, batch 9450, loss[loss=0.1432, simple_loss=0.225, pruned_loss=0.03066, over 12194.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2651, pruned_loss=0.04502, over 2370564.78 frames. ], batch size: 29, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:26:11,081 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6594, 2.6798, 3.3871, 4.3133, 2.4476, 4.6051, 4.5346, 4.7868], device='cuda:1'), covar=tensor([0.0094, 0.1075, 0.0419, 0.0145, 0.1083, 0.0202, 0.0149, 0.0065], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0203, 0.0189, 0.0115, 0.0188, 0.0180, 0.0171, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:26:21,064 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0169, 2.2977, 3.0345, 3.7957, 2.1177, 4.0860, 3.9330, 4.1442], device='cuda:1'), covar=tensor([0.0136, 0.1115, 0.0422, 0.0166, 0.1176, 0.0199, 0.0178, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0203, 0.0190, 0.0115, 0.0188, 0.0181, 0.0171, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:26:26,146 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9078, 2.9028, 4.8488, 5.0013, 3.2627, 2.7169, 3.0477, 2.1783], device='cuda:1'), covar=tensor([0.1251, 0.2611, 0.0323, 0.0269, 0.0895, 0.1785, 0.2329, 0.3478], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0371, 0.0267, 0.0290, 0.0251, 0.0276, 0.0352, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:26:42,320 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109495.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:26:46,333 INFO [finetune.py:992] (1/2) Epoch 1, batch 9500, loss[loss=0.1609, simple_loss=0.2563, pruned_loss=0.03276, over 12105.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2654, pruned_loss=0.04494, over 2361639.74 frames. ], batch size: 39, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:26:48,425 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.725e+02 3.238e+02 3.980e+02 6.999e+02, threshold=6.476e+02, percent-clipped=0.0 2023-05-15 17:27:10,840 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7852, 2.3272, 3.2649, 3.7996, 3.4598, 3.7646, 3.3114, 2.7108], device='cuda:1'), covar=tensor([0.0037, 0.0343, 0.0123, 0.0034, 0.0119, 0.0068, 0.0096, 0.0306], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0121, 0.0104, 0.0076, 0.0100, 0.0111, 0.0087, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:27:19,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-05-15 17:27:22,064 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109550.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:27:22,704 INFO [finetune.py:992] (1/2) Epoch 1, batch 9550, loss[loss=0.187, simple_loss=0.2734, pruned_loss=0.05032, over 12155.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2642, pruned_loss=0.04445, over 2362962.42 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:27:39,378 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-15 17:27:42,690 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109579.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:27:58,356 INFO [finetune.py:992] (1/2) Epoch 1, batch 9600, loss[loss=0.1707, simple_loss=0.2684, pruned_loss=0.03647, over 11580.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2638, pruned_loss=0.04428, over 2372212.63 frames. ], batch size: 48, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:28:00,542 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.787e+02 3.129e+02 4.200e+02 2.494e+03, threshold=6.258e+02, percent-clipped=6.0 2023-05-15 17:28:12,079 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3077, 4.8071, 5.2794, 4.5732, 4.9368, 4.6838, 5.3324, 4.9670], device='cuda:1'), covar=tensor([0.0225, 0.0379, 0.0234, 0.0261, 0.0309, 0.0308, 0.0187, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0245, 0.0260, 0.0234, 0.0236, 0.0235, 0.0216, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 17:28:27,041 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109640.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:28:34,087 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109650.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:28:34,759 INFO [finetune.py:992] (1/2) Epoch 1, batch 9650, loss[loss=0.1313, simple_loss=0.2112, pruned_loss=0.02572, over 12188.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2634, pruned_loss=0.04448, over 2363301.19 frames. ], batch size: 29, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:28:40,494 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109659.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:28:51,846 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109674.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 17:29:05,291 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109693.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:29:08,990 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0603, 4.0291, 4.2454, 4.5304, 3.2609, 3.8519, 2.6770, 4.0800], device='cuda:1'), covar=tensor([0.1731, 0.0706, 0.0740, 0.0637, 0.0950, 0.0685, 0.1703, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0256, 0.0294, 0.0355, 0.0237, 0.0235, 0.0252, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:29:10,843 INFO [finetune.py:992] (1/2) Epoch 1, batch 9700, loss[loss=0.1638, simple_loss=0.2495, pruned_loss=0.03909, over 12028.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2635, pruned_loss=0.04473, over 2366210.04 frames. ], batch size: 31, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:29:10,954 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3391, 5.1284, 5.2447, 5.2795, 4.7032, 4.8056, 4.8458, 5.1874], device='cuda:1'), covar=tensor([0.0742, 0.0784, 0.0797, 0.0755, 0.2563, 0.1505, 0.0602, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0624, 0.0534, 0.0592, 0.0777, 0.0709, 0.0516, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 17:29:12,852 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.182e+02 3.051e+02 3.582e+02 4.337e+02 1.129e+03, threshold=7.164e+02, percent-clipped=10.0 2023-05-15 17:29:22,160 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3167, 4.6656, 4.1514, 5.1151, 4.5997, 3.1592, 4.3498, 3.1334], device='cuda:1'), covar=tensor([0.0704, 0.0752, 0.1426, 0.0290, 0.1042, 0.1352, 0.0859, 0.2848], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0371, 0.0349, 0.0260, 0.0356, 0.0261, 0.0332, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:29:24,204 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109720.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:29:37,550 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0301, 4.5633, 4.0332, 4.2358, 4.7122, 4.0566, 4.2662, 4.2093], device='cuda:1'), covar=tensor([0.1341, 0.1134, 0.1459, 0.1868, 0.1044, 0.1883, 0.1727, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0469, 0.0368, 0.0416, 0.0446, 0.0420, 0.0379, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:29:38,958 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109741.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:29:46,026 INFO [finetune.py:992] (1/2) Epoch 1, batch 9750, loss[loss=0.178, simple_loss=0.2649, pruned_loss=0.04555, over 11872.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2627, pruned_loss=0.04445, over 2371436.53 frames. ], batch size: 44, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:29:49,001 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5572, 2.8447, 3.7302, 4.6331, 3.9284, 4.6495, 3.9315, 3.3591], device='cuda:1'), covar=tensor([0.0027, 0.0293, 0.0118, 0.0024, 0.0104, 0.0043, 0.0087, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0118, 0.0102, 0.0073, 0.0098, 0.0108, 0.0084, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:29:59,449 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2485, 3.1354, 2.9954, 2.9754, 2.6430, 2.5972, 3.0879, 2.0008], device='cuda:1'), covar=tensor([0.0327, 0.0146, 0.0155, 0.0134, 0.0363, 0.0266, 0.0126, 0.0412], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0164, 0.0154, 0.0182, 0.0206, 0.0197, 0.0161, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:30:22,016 INFO [finetune.py:992] (1/2) Epoch 1, batch 9800, loss[loss=0.2731, simple_loss=0.3354, pruned_loss=0.1054, over 8150.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2642, pruned_loss=0.0451, over 2367214.09 frames. ], batch size: 97, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:30:24,134 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.923e+02 3.465e+02 4.220e+02 9.009e+02, threshold=6.931e+02, percent-clipped=2.0 2023-05-15 17:30:24,357 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1382, 2.3705, 3.4549, 4.1624, 3.6200, 4.0899, 3.5893, 2.5893], device='cuda:1'), covar=tensor([0.0035, 0.0367, 0.0137, 0.0033, 0.0119, 0.0071, 0.0098, 0.0383], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0119, 0.0103, 0.0074, 0.0099, 0.0109, 0.0085, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:30:57,283 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109850.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:30:57,892 INFO [finetune.py:992] (1/2) Epoch 1, batch 9850, loss[loss=0.1717, simple_loss=0.266, pruned_loss=0.03876, over 12285.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2642, pruned_loss=0.04518, over 2366749.74 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:31:22,079 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2593, 4.7082, 4.1733, 5.1054, 4.6182, 2.6063, 4.2179, 2.9612], device='cuda:1'), covar=tensor([0.0749, 0.0732, 0.1158, 0.0300, 0.0911, 0.1699, 0.1029, 0.2987], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0368, 0.0348, 0.0258, 0.0355, 0.0260, 0.0330, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:31:31,187 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109898.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:31:33,196 INFO [finetune.py:992] (1/2) Epoch 1, batch 9900, loss[loss=0.1642, simple_loss=0.2566, pruned_loss=0.03593, over 12268.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2643, pruned_loss=0.04486, over 2371966.50 frames. ], batch size: 33, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:31:35,259 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.138e+02 2.877e+02 3.253e+02 4.045e+02 6.927e+02, threshold=6.506e+02, percent-clipped=0.0 2023-05-15 17:31:57,917 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109935.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:32:08,930 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109950.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:32:09,477 INFO [finetune.py:992] (1/2) Epoch 1, batch 9950, loss[loss=0.1613, simple_loss=0.2501, pruned_loss=0.03622, over 12087.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2638, pruned_loss=0.04509, over 2359272.03 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:32:18,885 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5457, 4.0489, 3.7113, 4.3413, 3.1741, 4.0791, 2.4129, 4.2966], device='cuda:1'), covar=tensor([0.1167, 0.0583, 0.1103, 0.0754, 0.0888, 0.0466, 0.1636, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0257, 0.0293, 0.0356, 0.0238, 0.0234, 0.0252, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:32:26,017 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4876, 4.8197, 3.0557, 2.8169, 4.0597, 2.7144, 4.0563, 3.2932], device='cuda:1'), covar=tensor([0.0578, 0.0466, 0.0941, 0.1278, 0.0264, 0.1130, 0.0435, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0251, 0.0176, 0.0196, 0.0140, 0.0179, 0.0192, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:32:26,678 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109974.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:32:43,527 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109998.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:32:49,158 INFO [finetune.py:992] (1/2) Epoch 1, batch 10000, loss[loss=0.1758, simple_loss=0.267, pruned_loss=0.04225, over 12338.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2644, pruned_loss=0.04502, over 2366055.95 frames. ], batch size: 36, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:32:51,300 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.816e+02 3.590e+02 4.301e+02 9.697e+02, threshold=7.180e+02, percent-clipped=4.0 2023-05-15 17:32:59,323 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110015.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:33:00,825 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110017.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:33:04,258 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110022.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:33:06,723 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-15 17:33:20,091 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110044.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:33:24,896 INFO [finetune.py:992] (1/2) Epoch 1, batch 10050, loss[loss=0.1586, simple_loss=0.2379, pruned_loss=0.03964, over 12175.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04463, over 2374172.05 frames. ], batch size: 29, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:33:44,470 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110078.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:34:01,137 INFO [finetune.py:992] (1/2) Epoch 1, batch 10100, loss[loss=0.1902, simple_loss=0.2811, pruned_loss=0.04968, over 11765.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2641, pruned_loss=0.04458, over 2368262.48 frames. ], batch size: 44, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:34:03,244 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.129e+02 2.928e+02 3.602e+02 4.397e+02 7.962e+02, threshold=7.204e+02, percent-clipped=1.0 2023-05-15 17:34:04,120 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110105.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:34:07,637 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110110.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:34:09,902 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2113, 2.4696, 3.6456, 3.0912, 3.4758, 3.1861, 2.4421, 3.5903], device='cuda:1'), covar=tensor([0.0106, 0.0297, 0.0113, 0.0202, 0.0126, 0.0139, 0.0314, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0192, 0.0170, 0.0175, 0.0192, 0.0151, 0.0182, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:34:11,338 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110115.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:34:35,969 INFO [finetune.py:992] (1/2) Epoch 1, batch 10150, loss[loss=0.1707, simple_loss=0.2586, pruned_loss=0.04138, over 12158.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2638, pruned_loss=0.0444, over 2377671.93 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:34:37,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-15 17:34:50,635 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110171.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:34:54,130 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110176.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:35:12,210 INFO [finetune.py:992] (1/2) Epoch 1, batch 10200, loss[loss=0.2037, simple_loss=0.2878, pruned_loss=0.05983, over 12135.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2647, pruned_loss=0.04452, over 2371762.09 frames. ], batch size: 38, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:35:14,277 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.856e+02 3.463e+02 4.178e+02 6.748e+02, threshold=6.925e+02, percent-clipped=0.0 2023-05-15 17:35:36,342 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110235.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:35:47,037 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1118, 4.9532, 5.0389, 5.0992, 4.7392, 4.7807, 4.6639, 5.0339], device='cuda:1'), covar=tensor([0.0596, 0.0581, 0.0694, 0.0520, 0.1840, 0.1167, 0.0523, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0619, 0.0530, 0.0590, 0.0768, 0.0705, 0.0513, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 17:35:48,304 INFO [finetune.py:992] (1/2) Epoch 1, batch 10250, loss[loss=0.1878, simple_loss=0.2883, pruned_loss=0.04367, over 12189.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2637, pruned_loss=0.04377, over 2375078.28 frames. ], batch size: 35, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:35:57,582 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6369, 4.6274, 4.4959, 4.1525, 4.3061, 4.6371, 4.3293, 4.1624], device='cuda:1'), covar=tensor([0.0822, 0.0895, 0.0624, 0.1173, 0.1982, 0.0724, 0.1296, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0507, 0.0457, 0.0581, 0.0370, 0.0654, 0.0715, 0.0520], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 17:36:11,096 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:36:11,216 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:36:18,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.29 vs. limit=5.0 2023-05-15 17:36:23,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.01 vs. limit=5.0 2023-05-15 17:36:23,999 INFO [finetune.py:992] (1/2) Epoch 1, batch 10300, loss[loss=0.1738, simple_loss=0.2666, pruned_loss=0.04045, over 12188.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04365, over 2374497.98 frames. ], batch size: 31, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:36:25,570 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110303.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:36:26,051 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.961e+02 3.396e+02 3.978e+02 7.164e+02, threshold=6.793e+02, percent-clipped=1.0 2023-05-15 17:36:34,003 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110315.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:36:36,099 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110318.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:36:55,374 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110344.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:37:00,259 INFO [finetune.py:992] (1/2) Epoch 1, batch 10350, loss[loss=0.1966, simple_loss=0.2761, pruned_loss=0.05861, over 12134.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2629, pruned_loss=0.04359, over 2372358.54 frames. ], batch size: 39, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:37:08,665 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110363.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:37:09,533 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110364.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:37:15,797 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110373.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:37:20,118 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:37:23,625 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3224, 4.6038, 3.9960, 5.0052, 4.5698, 2.8312, 4.3664, 3.1112], device='cuda:1'), covar=tensor([0.0671, 0.0794, 0.1291, 0.0321, 0.0920, 0.1475, 0.0826, 0.2915], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0366, 0.0346, 0.0259, 0.0354, 0.0258, 0.0328, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:37:35,892 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110400.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:37:36,486 INFO [finetune.py:992] (1/2) Epoch 1, batch 10400, loss[loss=0.1703, simple_loss=0.266, pruned_loss=0.03726, over 12351.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2642, pruned_loss=0.04437, over 2371815.44 frames. ], batch size: 35, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:37:38,659 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 3.067e+02 3.653e+02 4.594e+02 8.110e+02, threshold=7.306e+02, percent-clipped=3.0 2023-05-15 17:37:47,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-15 17:37:56,817 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3341, 3.4050, 3.5715, 4.0256, 2.8939, 3.3835, 2.5390, 3.2913], device='cuda:1'), covar=tensor([0.1471, 0.0845, 0.0961, 0.0654, 0.1069, 0.0710, 0.1627, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0259, 0.0294, 0.0356, 0.0238, 0.0238, 0.0253, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:38:11,407 INFO [finetune.py:992] (1/2) Epoch 1, batch 10450, loss[loss=0.1885, simple_loss=0.2762, pruned_loss=0.05038, over 12115.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2644, pruned_loss=0.04472, over 2364539.58 frames. ], batch size: 39, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:38:21,850 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110466.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:38:25,343 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110471.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:38:43,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-15 17:38:47,247 INFO [finetune.py:992] (1/2) Epoch 1, batch 10500, loss[loss=0.1797, simple_loss=0.2704, pruned_loss=0.04448, over 12292.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2645, pruned_loss=0.04491, over 2362380.55 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:38:49,440 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 3.055e+02 3.705e+02 4.613e+02 1.350e+03, threshold=7.409e+02, percent-clipped=4.0 2023-05-15 17:39:06,122 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-15 17:39:22,907 INFO [finetune.py:992] (1/2) Epoch 1, batch 10550, loss[loss=0.161, simple_loss=0.246, pruned_loss=0.03797, over 11793.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2638, pruned_loss=0.04448, over 2364304.33 frames. ], batch size: 26, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:39:58,881 INFO [finetune.py:992] (1/2) Epoch 1, batch 10600, loss[loss=0.1731, simple_loss=0.2715, pruned_loss=0.03737, over 12267.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2633, pruned_loss=0.04407, over 2367245.34 frames. ], batch size: 37, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:40:00,988 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.036e+02 2.797e+02 3.485e+02 4.135e+02 9.594e+02, threshold=6.971e+02, percent-clipped=1.0 2023-05-15 17:40:06,147 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110611.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:40:06,807 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1322, 5.8801, 5.5769, 5.3652, 5.9953, 5.2867, 5.5900, 5.5035], device='cuda:1'), covar=tensor([0.1373, 0.0874, 0.0765, 0.1903, 0.0862, 0.1783, 0.1505, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0462, 0.0363, 0.0417, 0.0444, 0.0413, 0.0374, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:40:26,941 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110639.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:40:27,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-15 17:40:29,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-15 17:40:35,371 INFO [finetune.py:992] (1/2) Epoch 1, batch 10650, loss[loss=0.1834, simple_loss=0.2783, pruned_loss=0.04432, over 12303.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.04439, over 2366500.42 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:40:41,032 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110659.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:40:50,332 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110672.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:40:50,969 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110673.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:40:52,296 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110674.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:41:10,613 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110700.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:41:11,217 INFO [finetune.py:992] (1/2) Epoch 1, batch 10700, loss[loss=0.1438, simple_loss=0.2252, pruned_loss=0.03119, over 12008.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2642, pruned_loss=0.04473, over 2357219.85 frames. ], batch size: 28, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:41:13,394 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.805e+02 3.521e+02 4.283e+02 1.056e+03, threshold=7.043e+02, percent-clipped=6.0 2023-05-15 17:41:25,433 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110721.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:41:36,282 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110736.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:41:44,713 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110748.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:41:46,805 INFO [finetune.py:992] (1/2) Epoch 1, batch 10750, loss[loss=0.1719, simple_loss=0.2711, pruned_loss=0.03632, over 12107.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2643, pruned_loss=0.04456, over 2363262.88 frames. ], batch size: 33, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:41:57,945 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110766.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:42:01,531 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110771.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:42:05,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-15 17:42:20,585 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110797.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:42:23,496 INFO [finetune.py:992] (1/2) Epoch 1, batch 10800, loss[loss=0.1716, simple_loss=0.2611, pruned_loss=0.04109, over 12091.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2646, pruned_loss=0.0446, over 2366302.96 frames. ], batch size: 33, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:42:25,613 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 3.029e+02 3.519e+02 4.086e+02 7.256e+02, threshold=7.039e+02, percent-clipped=1.0 2023-05-15 17:42:32,741 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110814.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:42:36,334 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110819.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:42:59,704 INFO [finetune.py:992] (1/2) Epoch 1, batch 10850, loss[loss=0.1771, simple_loss=0.2602, pruned_loss=0.04703, over 12177.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2644, pruned_loss=0.04469, over 2367667.56 frames. ], batch size: 31, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:43:05,821 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7505, 4.5035, 4.1497, 4.0883, 4.5972, 3.9119, 4.2603, 3.9753], device='cuda:1'), covar=tensor([0.1853, 0.1149, 0.1472, 0.2150, 0.1263, 0.2400, 0.1612, 0.1578], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0463, 0.0365, 0.0420, 0.0448, 0.0418, 0.0377, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:43:12,529 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:43:36,033 INFO [finetune.py:992] (1/2) Epoch 1, batch 10900, loss[loss=0.1511, simple_loss=0.2459, pruned_loss=0.02815, over 12251.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2637, pruned_loss=0.04434, over 2373124.82 frames. ], batch size: 32, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:43:38,036 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.219e+02 2.906e+02 3.302e+02 4.025e+02 7.029e+02, threshold=6.603e+02, percent-clipped=0.0 2023-05-15 17:43:53,038 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8711, 4.4877, 4.7959, 4.2107, 4.4971, 4.2937, 4.8419, 4.5467], device='cuda:1'), covar=tensor([0.0250, 0.0313, 0.0272, 0.0254, 0.0306, 0.0275, 0.0215, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0240, 0.0257, 0.0231, 0.0233, 0.0233, 0.0213, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 17:43:56,609 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110929.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:03,720 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110939.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:12,053 INFO [finetune.py:992] (1/2) Epoch 1, batch 10950, loss[loss=0.1721, simple_loss=0.2652, pruned_loss=0.03956, over 12313.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2651, pruned_loss=0.04526, over 2371532.57 frames. ], batch size: 34, lr: 4.99e-03, grad_scale: 8.0 2023-05-15 17:44:17,168 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110958.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:17,811 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110959.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:19,362 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3396, 4.5973, 4.1055, 4.9507, 4.5766, 2.7219, 4.2062, 3.0283], device='cuda:1'), covar=tensor([0.0605, 0.0675, 0.1188, 0.0297, 0.0798, 0.1498, 0.0850, 0.2837], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0372, 0.0352, 0.0262, 0.0360, 0.0264, 0.0333, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:44:23,450 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110967.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:29,185 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110974.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:30,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-05-15 17:44:38,265 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110987.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:48,252 INFO [finetune.py:992] (1/2) Epoch 1, batch 11000, loss[loss=0.1847, simple_loss=0.2764, pruned_loss=0.04652, over 12340.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2679, pruned_loss=0.04662, over 2364559.91 frames. ], batch size: 36, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:44:50,408 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 3.036e+02 3.496e+02 4.197e+02 8.188e+02, threshold=6.991e+02, percent-clipped=4.0 2023-05-15 17:44:52,567 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111007.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:44:57,590 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3198, 3.1320, 3.2474, 3.6473, 2.7426, 3.1409, 2.5149, 3.0753], device='cuda:1'), covar=tensor([0.1259, 0.0782, 0.0821, 0.0558, 0.0879, 0.0643, 0.1358, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0260, 0.0298, 0.0358, 0.0240, 0.0238, 0.0254, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:45:01,082 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111019.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:45:03,076 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111022.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:45:11,467 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0708, 2.4814, 3.4560, 2.9426, 3.3774, 3.0427, 2.3805, 3.4934], device='cuda:1'), covar=tensor([0.0087, 0.0276, 0.0120, 0.0207, 0.0107, 0.0142, 0.0288, 0.0090], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0190, 0.0169, 0.0174, 0.0190, 0.0150, 0.0181, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:45:23,273 INFO [finetune.py:992] (1/2) Epoch 1, batch 11050, loss[loss=0.1646, simple_loss=0.2472, pruned_loss=0.04102, over 12146.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2717, pruned_loss=0.04942, over 2336677.59 frames. ], batch size: 30, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:45:54,102 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111092.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 17:46:00,253 INFO [finetune.py:992] (1/2) Epoch 1, batch 11100, loss[loss=0.161, simple_loss=0.2454, pruned_loss=0.03828, over 12175.00 frames. ], tot_loss[loss=0.1896, simple_loss=0.2753, pruned_loss=0.05192, over 2295069.47 frames. ], batch size: 31, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:46:02,291 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.324e+02 3.326e+02 4.024e+02 5.027e+02 9.650e+02, threshold=8.047e+02, percent-clipped=3.0 2023-05-15 17:46:07,643 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6846, 2.6750, 3.9448, 4.2468, 3.3374, 2.8173, 2.8728, 2.1113], device='cuda:1'), covar=tensor([0.1278, 0.2576, 0.0469, 0.0298, 0.0732, 0.1603, 0.2143, 0.3825], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0363, 0.0262, 0.0285, 0.0246, 0.0271, 0.0345, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:46:36,138 INFO [finetune.py:992] (1/2) Epoch 1, batch 11150, loss[loss=0.3399, simple_loss=0.3864, pruned_loss=0.1467, over 6928.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2823, pruned_loss=0.05683, over 2222429.83 frames. ], batch size: 101, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:46:41,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-15 17:47:11,552 INFO [finetune.py:992] (1/2) Epoch 1, batch 11200, loss[loss=0.1471, simple_loss=0.2321, pruned_loss=0.031, over 12005.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.29, pruned_loss=0.06262, over 2157973.77 frames. ], batch size: 28, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:47:13,736 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 3.669e+02 4.380e+02 5.080e+02 1.411e+03, threshold=8.760e+02, percent-clipped=3.0 2023-05-15 17:47:28,133 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111224.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:47:28,917 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111225.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:47:47,220 INFO [finetune.py:992] (1/2) Epoch 1, batch 11250, loss[loss=0.2572, simple_loss=0.3314, pruned_loss=0.09149, over 10323.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.299, pruned_loss=0.06957, over 2071477.18 frames. ], batch size: 68, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:47:58,435 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111267.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:48:11,591 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111286.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:48:22,094 INFO [finetune.py:992] (1/2) Epoch 1, batch 11300, loss[loss=0.2777, simple_loss=0.3352, pruned_loss=0.1101, over 7437.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3046, pruned_loss=0.07388, over 2011021.89 frames. ], batch size: 99, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:48:24,236 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.301e+02 3.763e+02 4.460e+02 5.493e+02 1.220e+03, threshold=8.920e+02, percent-clipped=6.0 2023-05-15 17:48:31,856 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111314.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:48:32,580 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111315.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:48:57,063 INFO [finetune.py:992] (1/2) Epoch 1, batch 11350, loss[loss=0.3244, simple_loss=0.3726, pruned_loss=0.1381, over 6448.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3107, pruned_loss=0.07809, over 1969593.08 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 16.0 2023-05-15 17:49:01,900 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0183, 4.9062, 4.9351, 4.9686, 4.8475, 5.0669, 4.8770, 2.8126], device='cuda:1'), covar=tensor([0.0103, 0.0052, 0.0075, 0.0054, 0.0048, 0.0064, 0.0067, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0070, 0.0074, 0.0067, 0.0056, 0.0082, 0.0073, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:49:26,209 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111392.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 17:49:32,423 INFO [finetune.py:992] (1/2) Epoch 1, batch 11400, loss[loss=0.2581, simple_loss=0.3316, pruned_loss=0.09233, over 11530.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3157, pruned_loss=0.08185, over 1932757.12 frames. ], batch size: 48, lr: 4.98e-03, grad_scale: 16.0 2023-05-15 17:49:34,495 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.662e+02 3.744e+02 4.367e+02 5.298e+02 9.715e+02, threshold=8.735e+02, percent-clipped=1.0 2023-05-15 17:50:00,475 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111440.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:50:01,269 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1063, 2.3625, 3.3704, 2.9548, 3.2566, 3.0617, 2.2084, 3.3853], device='cuda:1'), covar=tensor([0.0083, 0.0281, 0.0077, 0.0191, 0.0097, 0.0130, 0.0304, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0183, 0.0158, 0.0165, 0.0180, 0.0142, 0.0173, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:50:07,891 INFO [finetune.py:992] (1/2) Epoch 1, batch 11450, loss[loss=0.2348, simple_loss=0.318, pruned_loss=0.07578, over 10175.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3192, pruned_loss=0.0848, over 1887265.36 frames. ], batch size: 68, lr: 4.98e-03, grad_scale: 16.0 2023-05-15 17:50:19,797 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111468.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:50:42,439 INFO [finetune.py:992] (1/2) Epoch 1, batch 11500, loss[loss=0.2688, simple_loss=0.3252, pruned_loss=0.1063, over 7204.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3219, pruned_loss=0.0869, over 1867012.73 frames. ], batch size: 98, lr: 4.98e-03, grad_scale: 16.0 2023-05-15 17:50:44,444 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.242e+02 3.745e+02 4.269e+02 5.103e+02 9.433e+02, threshold=8.538e+02, percent-clipped=0.0 2023-05-15 17:50:58,778 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111524.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:51:02,184 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111529.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:51:12,834 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1772, 4.1955, 2.7952, 2.3018, 3.6883, 2.3348, 3.8081, 2.8263], device='cuda:1'), covar=tensor([0.0759, 0.0476, 0.1169, 0.1904, 0.0227, 0.1660, 0.0405, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0239, 0.0174, 0.0195, 0.0135, 0.0177, 0.0186, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:51:16,797 INFO [finetune.py:992] (1/2) Epoch 1, batch 11550, loss[loss=0.2576, simple_loss=0.3348, pruned_loss=0.09019, over 11831.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3237, pruned_loss=0.08865, over 1859231.98 frames. ], batch size: 44, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:51:22,415 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-05-15 17:51:32,047 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111572.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:51:38,272 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111581.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:51:52,036 INFO [finetune.py:992] (1/2) Epoch 1, batch 11600, loss[loss=0.2385, simple_loss=0.3237, pruned_loss=0.07668, over 10457.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3252, pruned_loss=0.08995, over 1843085.80 frames. ], batch size: 68, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:51:54,675 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.870e+02 3.662e+02 4.318e+02 5.254e+02 1.085e+03, threshold=8.637e+02, percent-clipped=4.0 2023-05-15 17:52:01,755 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111614.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:52:28,660 INFO [finetune.py:992] (1/2) Epoch 1, batch 11650, loss[loss=0.2509, simple_loss=0.3268, pruned_loss=0.08749, over 6480.00 frames. ], tot_loss[loss=0.253, simple_loss=0.325, pruned_loss=0.09055, over 1811507.82 frames. ], batch size: 97, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:52:36,717 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111662.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:53:02,993 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7408, 2.7141, 3.9979, 4.1316, 3.0428, 2.7137, 2.7482, 2.0437], device='cuda:1'), covar=tensor([0.1277, 0.2358, 0.0451, 0.0392, 0.0855, 0.1799, 0.2311, 0.3861], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0355, 0.0253, 0.0277, 0.0240, 0.0267, 0.0342, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:53:04,035 INFO [finetune.py:992] (1/2) Epoch 1, batch 11700, loss[loss=0.2555, simple_loss=0.3183, pruned_loss=0.09636, over 6474.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3254, pruned_loss=0.09183, over 1766289.25 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:53:06,691 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.650e+02 3.667e+02 3.963e+02 4.752e+02 8.265e+02, threshold=7.925e+02, percent-clipped=0.0 2023-05-15 17:53:08,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-15 17:53:38,228 INFO [finetune.py:992] (1/2) Epoch 1, batch 11750, loss[loss=0.2465, simple_loss=0.3131, pruned_loss=0.08998, over 6652.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3257, pruned_loss=0.09251, over 1745317.47 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:54:06,447 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111791.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 17:54:13,731 INFO [finetune.py:992] (1/2) Epoch 1, batch 11800, loss[loss=0.3623, simple_loss=0.4009, pruned_loss=0.1619, over 6957.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3296, pruned_loss=0.09578, over 1716063.47 frames. ], batch size: 99, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:54:14,627 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7652, 3.7946, 3.7474, 3.8370, 3.6612, 3.6724, 3.6655, 3.7603], device='cuda:1'), covar=tensor([0.0792, 0.0578, 0.1044, 0.0666, 0.1546, 0.1181, 0.0526, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0563, 0.0480, 0.0533, 0.0688, 0.0635, 0.0465, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-15 17:54:16,475 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 3.032e+02 3.813e+02 4.515e+02 5.504e+02 1.380e+03, threshold=9.030e+02, percent-clipped=6.0 2023-05-15 17:54:30,321 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111824.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:54:48,313 INFO [finetune.py:992] (1/2) Epoch 1, batch 11850, loss[loss=0.2656, simple_loss=0.3318, pruned_loss=0.0997, over 7076.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3311, pruned_loss=0.09647, over 1711045.30 frames. ], batch size: 101, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:54:49,116 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111852.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 17:55:06,183 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2843, 2.8354, 2.8488, 2.8021, 2.5701, 2.2351, 2.9739, 1.9661], device='cuda:1'), covar=tensor([0.0345, 0.0151, 0.0149, 0.0147, 0.0279, 0.0321, 0.0113, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0148, 0.0140, 0.0167, 0.0187, 0.0183, 0.0145, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 17:55:10,108 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111881.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:55:23,520 INFO [finetune.py:992] (1/2) Epoch 1, batch 11900, loss[loss=0.2261, simple_loss=0.3101, pruned_loss=0.07106, over 10205.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3291, pruned_loss=0.09441, over 1698570.09 frames. ], batch size: 68, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:55:26,248 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.186e+02 3.677e+02 4.314e+02 4.976e+02 1.689e+03, threshold=8.628e+02, percent-clipped=4.0 2023-05-15 17:55:29,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 17:55:43,162 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111929.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:55:57,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1729, 3.1679, 4.4703, 2.5672, 2.7155, 3.5213, 3.1046, 3.5223], device='cuda:1'), covar=tensor([0.0393, 0.1025, 0.0187, 0.1100, 0.1682, 0.1117, 0.1196, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0213, 0.0210, 0.0169, 0.0221, 0.0255, 0.0210, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:55:58,572 INFO [finetune.py:992] (1/2) Epoch 1, batch 11950, loss[loss=0.2408, simple_loss=0.3029, pruned_loss=0.08934, over 7138.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3253, pruned_loss=0.09087, over 1706186.49 frames. ], batch size: 102, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:56:21,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.09 vs. limit=5.0 2023-05-15 17:56:37,220 INFO [finetune.py:992] (1/2) Epoch 1, batch 12000, loss[loss=0.247, simple_loss=0.3143, pruned_loss=0.08979, over 7250.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3198, pruned_loss=0.08601, over 1710241.70 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:56:37,221 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 17:56:53,941 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9514, 2.2280, 2.3431, 2.3135, 2.0561, 1.9001, 2.3458, 1.7130], device='cuda:1'), covar=tensor([0.0286, 0.0147, 0.0124, 0.0149, 0.0296, 0.0234, 0.0122, 0.0385], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0146, 0.0137, 0.0165, 0.0185, 0.0180, 0.0142, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 17:56:55,919 INFO [finetune.py:1026] (1/2) Epoch 1, validation: loss=0.2946, simple_loss=0.3701, pruned_loss=0.1095, over 1020973.00 frames. 2023-05-15 17:56:55,920 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 17:56:58,669 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.354e+02 3.236e+02 3.961e+02 5.158e+02 1.207e+03, threshold=7.921e+02, percent-clipped=1.0 2023-05-15 17:57:12,716 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-15 17:57:30,931 INFO [finetune.py:992] (1/2) Epoch 1, batch 12050, loss[loss=0.2464, simple_loss=0.3342, pruned_loss=0.07927, over 11142.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3161, pruned_loss=0.08284, over 1707760.56 frames. ], batch size: 55, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:57:40,890 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112066.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:58:03,542 INFO [finetune.py:992] (1/2) Epoch 1, batch 12100, loss[loss=0.2207, simple_loss=0.2931, pruned_loss=0.07417, over 7048.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3142, pruned_loss=0.08086, over 1719012.17 frames. ], batch size: 101, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:58:04,463 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5594, 2.9905, 3.8567, 2.3160, 2.5527, 3.2335, 2.9137, 3.2265], device='cuda:1'), covar=tensor([0.0442, 0.0952, 0.0257, 0.1235, 0.1646, 0.0985, 0.1164, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0211, 0.0207, 0.0167, 0.0218, 0.0251, 0.0207, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 17:58:06,084 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.361e+02 3.222e+02 3.757e+02 4.693e+02 1.119e+03, threshold=7.514e+02, percent-clipped=4.0 2023-05-15 17:58:18,417 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112124.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:58:20,344 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112127.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:58:24,980 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4389, 3.1801, 3.1644, 3.4340, 2.8640, 3.2050, 2.6944, 2.9496], device='cuda:1'), covar=tensor([0.1517, 0.0710, 0.0928, 0.0539, 0.0936, 0.0672, 0.1479, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0253, 0.0286, 0.0338, 0.0232, 0.0230, 0.0248, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 17:58:33,047 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112147.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 17:58:35,557 INFO [finetune.py:992] (1/2) Epoch 1, batch 12150, loss[loss=0.2284, simple_loss=0.3116, pruned_loss=0.07263, over 10587.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3145, pruned_loss=0.08136, over 1707695.37 frames. ], batch size: 70, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:58:49,608 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112172.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 17:59:04,277 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7592, 3.6900, 3.6741, 3.7623, 3.5061, 3.7860, 3.7644, 3.8735], device='cuda:1'), covar=tensor([0.0197, 0.0151, 0.0174, 0.0263, 0.0531, 0.0267, 0.0179, 0.0195], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0153, 0.0152, 0.0187, 0.0193, 0.0164, 0.0142, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 17:59:08,149 INFO [finetune.py:992] (1/2) Epoch 1, batch 12200, loss[loss=0.2525, simple_loss=0.3186, pruned_loss=0.09324, over 7051.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.315, pruned_loss=0.08214, over 1679355.46 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:59:10,535 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.169e+02 3.288e+02 3.899e+02 4.371e+02 8.916e+02, threshold=7.798e+02, percent-clipped=1.0 2023-05-15 17:59:26,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-15 17:59:54,047 INFO [finetune.py:992] (1/2) Epoch 2, batch 0, loss[loss=0.1815, simple_loss=0.2687, pruned_loss=0.04715, over 12282.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2687, pruned_loss=0.04715, over 12282.00 frames. ], batch size: 33, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 17:59:54,048 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 18:00:11,581 INFO [finetune.py:1026] (1/2) Epoch 2, validation: loss=0.2987, simple_loss=0.3712, pruned_loss=0.1132, over 1020973.00 frames. 2023-05-15 18:00:11,582 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 18:00:24,222 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5546, 2.7782, 3.9213, 2.2570, 2.4033, 3.1369, 2.8107, 3.2284], device='cuda:1'), covar=tensor([0.0513, 0.1250, 0.0288, 0.1374, 0.2026, 0.1322, 0.1413, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0213, 0.0207, 0.0170, 0.0221, 0.0254, 0.0210, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:00:47,303 INFO [finetune.py:992] (1/2) Epoch 2, batch 50, loss[loss=0.1634, simple_loss=0.2443, pruned_loss=0.04128, over 11992.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2736, pruned_loss=0.05029, over 532717.23 frames. ], batch size: 28, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:00:58,182 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0143, 2.2696, 3.5718, 2.8884, 3.2950, 3.1046, 2.2453, 3.3330], device='cuda:1'), covar=tensor([0.0140, 0.0441, 0.0143, 0.0284, 0.0174, 0.0195, 0.0441, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0141, 0.0179, 0.0148, 0.0158, 0.0172, 0.0136, 0.0168, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:01:01,428 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.225e+02 3.270e+02 3.935e+02 4.941e+02 1.663e+03, threshold=7.870e+02, percent-clipped=3.0 2023-05-15 18:01:22,488 INFO [finetune.py:992] (1/2) Epoch 2, batch 100, loss[loss=0.1612, simple_loss=0.2471, pruned_loss=0.03764, over 12348.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2721, pruned_loss=0.04922, over 944425.37 frames. ], batch size: 30, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:01:58,370 INFO [finetune.py:992] (1/2) Epoch 2, batch 150, loss[loss=0.2578, simple_loss=0.3234, pruned_loss=0.09611, over 8095.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2718, pruned_loss=0.04922, over 1251012.68 frames. ], batch size: 98, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:02:01,464 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112389.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:02:11,979 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-05-15 18:02:12,808 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 2.983e+02 3.524e+02 4.187e+02 7.539e+02, threshold=7.047e+02, percent-clipped=0.0 2023-05-15 18:02:25,634 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112422.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:02:34,893 INFO [finetune.py:992] (1/2) Epoch 2, batch 200, loss[loss=0.1603, simple_loss=0.2494, pruned_loss=0.03567, over 12131.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2691, pruned_loss=0.04747, over 1505791.37 frames. ], batch size: 30, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:02:39,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2815, 4.7199, 4.0786, 5.1158, 4.8349, 2.9678, 4.5260, 3.1315], device='cuda:1'), covar=tensor([0.0774, 0.0749, 0.1351, 0.0346, 0.0828, 0.1530, 0.0840, 0.3139], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0354, 0.0331, 0.0238, 0.0342, 0.0256, 0.0317, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:02:41,195 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1485, 5.8854, 5.5055, 5.4194, 5.9872, 5.2023, 5.6108, 5.5167], device='cuda:1'), covar=tensor([0.1426, 0.0918, 0.0920, 0.2124, 0.1017, 0.2421, 0.1541, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0428, 0.0343, 0.0390, 0.0414, 0.0391, 0.0349, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:02:43,303 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112447.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 18:02:45,614 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112450.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:03:10,365 INFO [finetune.py:992] (1/2) Epoch 2, batch 250, loss[loss=0.1494, simple_loss=0.2314, pruned_loss=0.03373, over 12017.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2683, pruned_loss=0.04693, over 1699014.90 frames. ], batch size: 28, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:03:18,027 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0838, 2.1671, 2.7847, 3.0316, 2.9877, 3.1281, 2.9139, 2.4967], device='cuda:1'), covar=tensor([0.0061, 0.0321, 0.0130, 0.0052, 0.0100, 0.0074, 0.0077, 0.0278], device='cuda:1'), in_proj_covar=tensor([0.0074, 0.0115, 0.0095, 0.0070, 0.0092, 0.0104, 0.0080, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:03:18,522 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112495.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 18:03:24,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-05-15 18:03:25,433 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.114e+02 2.915e+02 3.506e+02 4.058e+02 6.967e+02, threshold=7.011e+02, percent-clipped=0.0 2023-05-15 18:03:46,663 INFO [finetune.py:992] (1/2) Epoch 2, batch 300, loss[loss=0.1749, simple_loss=0.2645, pruned_loss=0.04268, over 12097.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2684, pruned_loss=0.04685, over 1850953.94 frames. ], batch size: 33, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:04:06,060 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112562.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:04:17,256 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112577.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:04:21,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-15 18:04:22,784 INFO [finetune.py:992] (1/2) Epoch 2, batch 350, loss[loss=0.1557, simple_loss=0.2515, pruned_loss=0.02997, over 12290.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.268, pruned_loss=0.04652, over 1967112.27 frames. ], batch size: 33, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:04:29,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-15 18:04:37,321 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.824e+02 3.340e+02 4.009e+02 1.325e+03, threshold=6.681e+02, percent-clipped=2.0 2023-05-15 18:04:42,414 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1405, 5.1157, 4.9219, 4.9967, 4.6011, 5.0974, 5.0958, 5.3446], device='cuda:1'), covar=tensor([0.0219, 0.0127, 0.0191, 0.0247, 0.0732, 0.0255, 0.0146, 0.0151], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0169, 0.0169, 0.0208, 0.0214, 0.0182, 0.0157, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1') 2023-05-15 18:04:45,983 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112617.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:04:50,398 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112623.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:04:58,233 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0207, 2.3142, 3.5220, 2.9579, 3.2904, 3.0559, 2.2386, 3.4100], device='cuda:1'), covar=tensor([0.0095, 0.0297, 0.0104, 0.0213, 0.0121, 0.0134, 0.0310, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0186, 0.0157, 0.0166, 0.0181, 0.0143, 0.0176, 0.0153], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:04:58,641 INFO [finetune.py:992] (1/2) Epoch 2, batch 400, loss[loss=0.1971, simple_loss=0.2903, pruned_loss=0.05197, over 11997.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2666, pruned_loss=0.04569, over 2064394.50 frames. ], batch size: 42, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:05:01,058 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112638.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 18:05:29,936 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112678.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 18:05:34,528 INFO [finetune.py:992] (1/2) Epoch 2, batch 450, loss[loss=0.2174, simple_loss=0.2966, pruned_loss=0.06913, over 11638.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2658, pruned_loss=0.04553, over 2140446.18 frames. ], batch size: 48, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:05:48,798 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 2.826e+02 3.576e+02 4.360e+02 8.764e+02, threshold=7.152e+02, percent-clipped=3.0 2023-05-15 18:06:00,933 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112722.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:06:10,657 INFO [finetune.py:992] (1/2) Epoch 2, batch 500, loss[loss=0.215, simple_loss=0.3042, pruned_loss=0.06289, over 7911.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2663, pruned_loss=0.04597, over 2193247.75 frames. ], batch size: 98, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:06:17,966 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112745.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:06:35,541 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112770.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:06:46,040 INFO [finetune.py:992] (1/2) Epoch 2, batch 550, loss[loss=0.1608, simple_loss=0.2587, pruned_loss=0.03142, over 12291.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2653, pruned_loss=0.04523, over 2233947.18 frames. ], batch size: 37, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:06:55,201 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1814, 4.8522, 4.9537, 5.0204, 4.8831, 4.9971, 4.8824, 2.6664], device='cuda:1'), covar=tensor([0.0087, 0.0066, 0.0095, 0.0058, 0.0045, 0.0075, 0.0101, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0068, 0.0072, 0.0066, 0.0053, 0.0080, 0.0071, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:07:01,100 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.848e+02 3.230e+02 3.802e+02 1.094e+03, threshold=6.461e+02, percent-clipped=1.0 2023-05-15 18:07:22,414 INFO [finetune.py:992] (1/2) Epoch 2, batch 600, loss[loss=0.177, simple_loss=0.2671, pruned_loss=0.04341, over 12247.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2646, pruned_loss=0.04462, over 2270048.71 frames. ], batch size: 37, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:07:34,153 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-15 18:07:39,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-15 18:07:58,199 INFO [finetune.py:992] (1/2) Epoch 2, batch 650, loss[loss=0.1551, simple_loss=0.2436, pruned_loss=0.03332, over 12024.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2653, pruned_loss=0.04494, over 2292550.08 frames. ], batch size: 31, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:08:12,428 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.261e+02 2.862e+02 3.449e+02 4.206e+02 6.989e+02, threshold=6.897e+02, percent-clipped=2.0 2023-05-15 18:08:21,919 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112918.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:08:24,124 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112921.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:08:27,090 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7354, 3.3387, 5.1732, 2.6813, 2.8590, 3.9001, 3.1279, 4.0052], device='cuda:1'), covar=tensor([0.0439, 0.1175, 0.0239, 0.1161, 0.1954, 0.1323, 0.1472, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0224, 0.0224, 0.0177, 0.0234, 0.0271, 0.0221, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:08:30,669 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6711, 3.6350, 3.4739, 3.5130, 3.0878, 3.1248, 3.8686, 2.4027], device='cuda:1'), covar=tensor([0.0328, 0.0149, 0.0164, 0.0132, 0.0332, 0.0260, 0.0098, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0153, 0.0143, 0.0171, 0.0192, 0.0185, 0.0147, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:08:32,641 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112933.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 18:08:33,933 INFO [finetune.py:992] (1/2) Epoch 2, batch 700, loss[loss=0.1654, simple_loss=0.256, pruned_loss=0.03736, over 12079.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2642, pruned_loss=0.04435, over 2319065.57 frames. ], batch size: 32, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:08:39,724 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5459, 2.4538, 3.3501, 4.3826, 2.5899, 4.4629, 4.4241, 4.6204], device='cuda:1'), covar=tensor([0.0088, 0.1160, 0.0398, 0.0101, 0.1052, 0.0180, 0.0131, 0.0059], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0201, 0.0183, 0.0108, 0.0188, 0.0172, 0.0164, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:09:01,395 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112973.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 18:09:07,923 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112982.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:09:09,724 INFO [finetune.py:992] (1/2) Epoch 2, batch 750, loss[loss=0.1714, simple_loss=0.2624, pruned_loss=0.04016, over 12422.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2639, pruned_loss=0.04396, over 2337640.27 frames. ], batch size: 32, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:09:24,420 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.175e+02 2.857e+02 3.349e+02 4.061e+02 6.781e+02, threshold=6.697e+02, percent-clipped=0.0 2023-05-15 18:09:46,145 INFO [finetune.py:992] (1/2) Epoch 2, batch 800, loss[loss=0.2084, simple_loss=0.2995, pruned_loss=0.05866, over 12024.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.264, pruned_loss=0.04379, over 2351544.27 frames. ], batch size: 42, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:09:53,416 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113045.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:10:02,681 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1927, 2.3692, 3.6197, 4.3000, 3.7601, 4.2244, 3.7823, 2.9487], device='cuda:1'), covar=tensor([0.0036, 0.0382, 0.0118, 0.0028, 0.0105, 0.0056, 0.0092, 0.0327], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0120, 0.0099, 0.0073, 0.0096, 0.0108, 0.0084, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:10:05,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-05-15 18:10:21,377 INFO [finetune.py:992] (1/2) Epoch 2, batch 850, loss[loss=0.1658, simple_loss=0.2566, pruned_loss=0.03753, over 11539.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2644, pruned_loss=0.04434, over 2350791.94 frames. ], batch size: 48, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:10:27,746 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113093.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:10:36,173 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 3.121e+02 3.766e+02 4.532e+02 1.104e+03, threshold=7.531e+02, percent-clipped=12.0 2023-05-15 18:10:46,938 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5453, 5.3675, 5.3383, 5.4787, 5.1109, 5.1152, 4.8198, 5.5173], device='cuda:1'), covar=tensor([0.0555, 0.0534, 0.0799, 0.0576, 0.1923, 0.1225, 0.0598, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0587, 0.0501, 0.0553, 0.0721, 0.0663, 0.0485, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 18:10:57,558 INFO [finetune.py:992] (1/2) Epoch 2, batch 900, loss[loss=0.1403, simple_loss=0.2222, pruned_loss=0.02923, over 12169.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2643, pruned_loss=0.04409, over 2360493.91 frames. ], batch size: 29, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:11:13,977 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0290, 4.9235, 4.8361, 4.9832, 4.4295, 5.0642, 5.0058, 5.3247], device='cuda:1'), covar=tensor([0.0253, 0.0165, 0.0230, 0.0265, 0.0805, 0.0248, 0.0162, 0.0146], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0176, 0.0176, 0.0218, 0.0224, 0.0191, 0.0164, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 18:11:33,565 INFO [finetune.py:992] (1/2) Epoch 2, batch 950, loss[loss=0.1849, simple_loss=0.2674, pruned_loss=0.05116, over 12094.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2631, pruned_loss=0.04335, over 2370864.40 frames. ], batch size: 32, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:11:48,150 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.177e+02 2.906e+02 3.318e+02 3.826e+02 7.167e+02, threshold=6.636e+02, percent-clipped=0.0 2023-05-15 18:11:57,342 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113218.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:11:59,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-05-15 18:12:01,048 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2171, 2.4955, 3.7329, 3.1527, 3.5706, 3.2873, 2.4124, 3.7110], device='cuda:1'), covar=tensor([0.0094, 0.0295, 0.0125, 0.0224, 0.0116, 0.0140, 0.0297, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0191, 0.0163, 0.0171, 0.0185, 0.0147, 0.0180, 0.0157], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:12:08,161 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113233.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:12:08,885 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113234.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:12:09,389 INFO [finetune.py:992] (1/2) Epoch 2, batch 1000, loss[loss=0.1818, simple_loss=0.2738, pruned_loss=0.04491, over 11826.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2632, pruned_loss=0.04342, over 2375493.86 frames. ], batch size: 44, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:12:32,398 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113266.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:12:35,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-15 18:12:37,491 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113273.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 18:12:40,319 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113277.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:12:43,111 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113281.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:12:45,830 INFO [finetune.py:992] (1/2) Epoch 2, batch 1050, loss[loss=0.1914, simple_loss=0.2789, pruned_loss=0.05194, over 10629.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2622, pruned_loss=0.04312, over 2376645.76 frames. ], batch size: 68, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:12:53,242 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:12:57,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-15 18:13:00,150 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.902e+02 3.523e+02 4.453e+02 1.735e+03, threshold=7.046e+02, percent-clipped=8.0 2023-05-15 18:13:12,202 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113321.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:13:22,317 INFO [finetune.py:992] (1/2) Epoch 2, batch 1100, loss[loss=0.1602, simple_loss=0.2408, pruned_loss=0.03974, over 12354.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2622, pruned_loss=0.04333, over 2376698.05 frames. ], batch size: 30, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:13:57,242 INFO [finetune.py:992] (1/2) Epoch 2, batch 1150, loss[loss=0.1731, simple_loss=0.2658, pruned_loss=0.0402, over 12105.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04371, over 2382918.35 frames. ], batch size: 39, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:14:05,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-05-15 18:14:12,257 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.863e+02 3.318e+02 3.829e+02 6.998e+02, threshold=6.637e+02, percent-clipped=0.0 2023-05-15 18:14:28,051 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0719, 4.3442, 3.8873, 4.6162, 4.3203, 2.7939, 4.1422, 2.8674], device='cuda:1'), covar=tensor([0.0781, 0.0790, 0.1342, 0.0439, 0.0987, 0.1518, 0.0816, 0.3265], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0363, 0.0342, 0.0250, 0.0351, 0.0260, 0.0326, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:14:33,302 INFO [finetune.py:992] (1/2) Epoch 2, batch 1200, loss[loss=0.1662, simple_loss=0.2667, pruned_loss=0.03287, over 12350.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2627, pruned_loss=0.04358, over 2384004.35 frames. ], batch size: 35, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:15:09,608 INFO [finetune.py:992] (1/2) Epoch 2, batch 1250, loss[loss=0.1662, simple_loss=0.2607, pruned_loss=0.03583, over 12259.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2628, pruned_loss=0.04347, over 2383462.56 frames. ], batch size: 37, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:15:23,696 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.287e+02 2.869e+02 3.347e+02 3.921e+02 1.251e+03, threshold=6.694e+02, percent-clipped=4.0 2023-05-15 18:15:31,763 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5646, 2.5036, 3.7064, 4.6559, 4.0483, 4.5929, 4.1055, 3.0873], device='cuda:1'), covar=tensor([0.0027, 0.0374, 0.0108, 0.0026, 0.0089, 0.0061, 0.0061, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0120, 0.0099, 0.0073, 0.0095, 0.0108, 0.0083, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:15:45,172 INFO [finetune.py:992] (1/2) Epoch 2, batch 1300, loss[loss=0.1856, simple_loss=0.2794, pruned_loss=0.04588, over 11815.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2628, pruned_loss=0.0433, over 2377139.21 frames. ], batch size: 44, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:16:15,892 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113577.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:16:18,822 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2521, 3.6174, 3.7356, 4.1913, 2.9157, 3.6056, 2.4651, 3.6447], device='cuda:1'), covar=tensor([0.1712, 0.0861, 0.0961, 0.0704, 0.1199, 0.0718, 0.1941, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0268, 0.0300, 0.0360, 0.0244, 0.0244, 0.0260, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:16:19,437 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4170, 5.2719, 5.3133, 5.4418, 5.0120, 5.0182, 4.8501, 5.3968], device='cuda:1'), covar=tensor([0.0697, 0.0510, 0.0782, 0.0507, 0.1948, 0.1310, 0.0548, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0608, 0.0518, 0.0575, 0.0752, 0.0683, 0.0501, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 18:16:21,373 INFO [finetune.py:992] (1/2) Epoch 2, batch 1350, loss[loss=0.1448, simple_loss=0.2271, pruned_loss=0.03123, over 12021.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2636, pruned_loss=0.04364, over 2373919.65 frames. ], batch size: 31, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:16:25,050 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113590.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:16:36,578 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 2.988e+02 3.437e+02 3.961e+02 7.082e+02, threshold=6.873e+02, percent-clipped=1.0 2023-05-15 18:16:44,493 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113617.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:16:50,814 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113625.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:16:58,006 INFO [finetune.py:992] (1/2) Epoch 2, batch 1400, loss[loss=0.1688, simple_loss=0.2606, pruned_loss=0.03852, over 12152.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2621, pruned_loss=0.04307, over 2370635.79 frames. ], batch size: 36, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:17:05,923 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113646.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:17:28,232 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113678.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:17:33,064 INFO [finetune.py:992] (1/2) Epoch 2, batch 1450, loss[loss=0.157, simple_loss=0.2416, pruned_loss=0.0362, over 12349.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2626, pruned_loss=0.04352, over 2368582.24 frames. ], batch size: 30, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:17:48,637 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.119e+02 2.823e+02 3.362e+02 3.969e+02 1.640e+03, threshold=6.724e+02, percent-clipped=5.0 2023-05-15 18:17:49,552 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113707.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:18:08,656 INFO [finetune.py:992] (1/2) Epoch 2, batch 1500, loss[loss=0.1955, simple_loss=0.2895, pruned_loss=0.05071, over 12067.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2642, pruned_loss=0.04416, over 2358755.11 frames. ], batch size: 42, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:18:13,064 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9204, 3.4449, 5.1668, 2.6449, 2.9169, 3.9655, 3.4409, 3.9204], device='cuda:1'), covar=tensor([0.0372, 0.1061, 0.0291, 0.1268, 0.1885, 0.1336, 0.1298, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0228, 0.0231, 0.0181, 0.0238, 0.0277, 0.0225, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:18:30,853 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5387, 2.5172, 3.2209, 4.3948, 2.3332, 4.4325, 4.4461, 4.6033], device='cuda:1'), covar=tensor([0.0095, 0.1096, 0.0430, 0.0115, 0.1266, 0.0183, 0.0106, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0202, 0.0184, 0.0109, 0.0188, 0.0173, 0.0166, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:18:44,187 INFO [finetune.py:992] (1/2) Epoch 2, batch 1550, loss[loss=0.1921, simple_loss=0.2803, pruned_loss=0.05192, over 12379.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.265, pruned_loss=0.04439, over 2368581.54 frames. ], batch size: 38, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:18:46,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-15 18:18:59,276 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.900e+02 3.326e+02 4.041e+02 9.467e+02, threshold=6.652e+02, percent-clipped=1.0 2023-05-15 18:19:20,721 INFO [finetune.py:992] (1/2) Epoch 2, batch 1600, loss[loss=0.2333, simple_loss=0.3113, pruned_loss=0.07766, over 8423.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2646, pruned_loss=0.04417, over 2369036.09 frames. ], batch size: 98, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:19:46,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.40 vs. limit=5.0 2023-05-15 18:19:47,370 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0520, 5.9757, 5.7385, 5.3156, 5.0540, 5.9209, 5.4518, 5.2598], device='cuda:1'), covar=tensor([0.0682, 0.0895, 0.0674, 0.1473, 0.0657, 0.0695, 0.1569, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0505, 0.0463, 0.0591, 0.0376, 0.0654, 0.0714, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 18:19:56,677 INFO [finetune.py:992] (1/2) Epoch 2, batch 1650, loss[loss=0.1709, simple_loss=0.2596, pruned_loss=0.04104, over 12073.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2644, pruned_loss=0.04383, over 2374794.86 frames. ], batch size: 42, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:20:00,391 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113890.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:20:11,667 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.174e+02 2.961e+02 3.386e+02 4.032e+02 9.524e+02, threshold=6.773e+02, percent-clipped=3.0 2023-05-15 18:20:28,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.39 vs. limit=5.0 2023-05-15 18:20:33,103 INFO [finetune.py:992] (1/2) Epoch 2, batch 1700, loss[loss=0.1795, simple_loss=0.2769, pruned_loss=0.04101, over 12110.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2644, pruned_loss=0.04392, over 2381466.18 frames. ], batch size: 38, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:20:35,350 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113938.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:20:59,855 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113973.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:21:08,327 INFO [finetune.py:992] (1/2) Epoch 2, batch 1750, loss[loss=0.1743, simple_loss=0.2558, pruned_loss=0.04647, over 12184.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2634, pruned_loss=0.04345, over 2388779.64 frames. ], batch size: 31, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:21:16,894 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2041, 5.0554, 5.1488, 5.1968, 4.8057, 4.8395, 4.6262, 5.1943], device='cuda:1'), covar=tensor([0.0703, 0.0567, 0.0687, 0.0541, 0.1704, 0.1260, 0.0560, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0605, 0.0517, 0.0575, 0.0753, 0.0687, 0.0502, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 18:21:24,136 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114002.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:21:26,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-05-15 18:21:26,780 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 2.960e+02 3.494e+02 4.137e+02 7.553e+02, threshold=6.989e+02, percent-clipped=3.0 2023-05-15 18:21:38,128 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0394, 4.5994, 4.7451, 4.8017, 4.7039, 4.8879, 4.7347, 2.5897], device='cuda:1'), covar=tensor([0.0069, 0.0061, 0.0074, 0.0057, 0.0045, 0.0064, 0.0070, 0.0718], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0073, 0.0077, 0.0070, 0.0057, 0.0085, 0.0075, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0004], device='cuda:1') 2023-05-15 18:21:46,729 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3476, 4.5586, 4.2398, 4.9336, 4.6012, 2.8398, 4.1854, 3.0896], device='cuda:1'), covar=tensor([0.0713, 0.0739, 0.1102, 0.0386, 0.0909, 0.1553, 0.0960, 0.2945], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0358, 0.0335, 0.0246, 0.0347, 0.0255, 0.0320, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:21:47,100 INFO [finetune.py:992] (1/2) Epoch 2, batch 1800, loss[loss=0.1978, simple_loss=0.2865, pruned_loss=0.05448, over 12145.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2638, pruned_loss=0.04379, over 2388401.64 frames. ], batch size: 34, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:22:22,822 INFO [finetune.py:992] (1/2) Epoch 2, batch 1850, loss[loss=0.1946, simple_loss=0.2856, pruned_loss=0.05185, over 12365.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.264, pruned_loss=0.04373, over 2393107.03 frames. ], batch size: 38, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:22:26,711 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7847, 2.3464, 3.2046, 2.6907, 3.0671, 2.9614, 2.1641, 3.1951], device='cuda:1'), covar=tensor([0.0101, 0.0285, 0.0109, 0.0213, 0.0126, 0.0137, 0.0284, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0152, 0.0190, 0.0165, 0.0171, 0.0186, 0.0147, 0.0180, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:22:37,776 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.825e+02 2.921e+02 3.335e+02 4.221e+02 7.050e+02, threshold=6.670e+02, percent-clipped=1.0 2023-05-15 18:22:48,443 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8022, 3.4433, 5.1149, 2.6689, 2.8692, 3.9531, 3.3544, 3.9258], device='cuda:1'), covar=tensor([0.0380, 0.0982, 0.0239, 0.1161, 0.1845, 0.1300, 0.1252, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0231, 0.0234, 0.0183, 0.0240, 0.0282, 0.0228, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:22:52,717 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114127.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:22:54,399 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-15 18:22:59,074 INFO [finetune.py:992] (1/2) Epoch 2, batch 1900, loss[loss=0.1798, simple_loss=0.2656, pruned_loss=0.04699, over 12191.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2628, pruned_loss=0.04318, over 2393522.40 frames. ], batch size: 35, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:23:34,603 INFO [finetune.py:992] (1/2) Epoch 2, batch 1950, loss[loss=0.1465, simple_loss=0.2293, pruned_loss=0.03189, over 12264.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2631, pruned_loss=0.04327, over 2385934.69 frames. ], batch size: 28, lr: 4.98e-03, grad_scale: 8.0 2023-05-15 18:23:36,817 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114188.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:23:49,659 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.996e+02 3.482e+02 4.338e+02 7.908e+02, threshold=6.963e+02, percent-clipped=3.0 2023-05-15 18:23:58,179 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2039, 4.7530, 5.1379, 4.4994, 4.8876, 4.5720, 5.1990, 4.9332], device='cuda:1'), covar=tensor([0.0240, 0.0353, 0.0277, 0.0240, 0.0245, 0.0259, 0.0173, 0.0260], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0231, 0.0250, 0.0225, 0.0223, 0.0228, 0.0206, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 18:24:10,858 INFO [finetune.py:992] (1/2) Epoch 2, batch 2000, loss[loss=0.2011, simple_loss=0.2843, pruned_loss=0.05895, over 12274.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2639, pruned_loss=0.04387, over 2377693.06 frames. ], batch size: 37, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:24:20,017 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2233, 4.8223, 5.1598, 4.4971, 4.8850, 4.5631, 5.2333, 4.8953], device='cuda:1'), covar=tensor([0.0216, 0.0305, 0.0270, 0.0246, 0.0268, 0.0288, 0.0168, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0231, 0.0250, 0.0225, 0.0223, 0.0227, 0.0206, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 18:24:20,062 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3193, 2.3032, 3.5023, 4.4222, 3.8062, 4.3679, 3.8352, 3.0475], device='cuda:1'), covar=tensor([0.0027, 0.0373, 0.0130, 0.0031, 0.0101, 0.0063, 0.0076, 0.0291], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0118, 0.0098, 0.0072, 0.0095, 0.0106, 0.0083, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:24:38,085 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114273.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:24:46,219 INFO [finetune.py:992] (1/2) Epoch 2, batch 2050, loss[loss=0.1914, simple_loss=0.2896, pruned_loss=0.04662, over 12351.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2639, pruned_loss=0.04401, over 2375487.61 frames. ], batch size: 35, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:24:59,028 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114302.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:25:01,716 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.702e+02 3.223e+02 3.685e+02 5.606e+02, threshold=6.446e+02, percent-clipped=0.0 2023-05-15 18:25:01,972 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3835, 3.3620, 3.2118, 3.1108, 2.7734, 2.5323, 3.3820, 2.1903], device='cuda:1'), covar=tensor([0.0374, 0.0143, 0.0160, 0.0174, 0.0346, 0.0331, 0.0113, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0157, 0.0145, 0.0174, 0.0195, 0.0189, 0.0152, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:25:12,497 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114321.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:25:22,475 INFO [finetune.py:992] (1/2) Epoch 2, batch 2100, loss[loss=0.171, simple_loss=0.2706, pruned_loss=0.0357, over 12137.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2645, pruned_loss=0.04405, over 2372424.55 frames. ], batch size: 36, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:25:28,503 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 18:25:32,957 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:25:39,939 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1721, 5.8955, 5.4266, 5.4725, 6.0267, 5.3101, 5.6360, 5.5667], device='cuda:1'), covar=tensor([0.1342, 0.1014, 0.0918, 0.2344, 0.1093, 0.2262, 0.1403, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0328, 0.0452, 0.0359, 0.0409, 0.0437, 0.0408, 0.0370, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:25:47,066 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1058, 4.6821, 4.8145, 4.9296, 4.7134, 4.9471, 4.8430, 2.7558], device='cuda:1'), covar=tensor([0.0075, 0.0058, 0.0082, 0.0057, 0.0047, 0.0073, 0.0063, 0.0676], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0073, 0.0077, 0.0070, 0.0057, 0.0085, 0.0075, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:25:58,390 INFO [finetune.py:992] (1/2) Epoch 2, batch 2150, loss[loss=0.22, simple_loss=0.3015, pruned_loss=0.06921, over 8520.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2638, pruned_loss=0.04397, over 2375245.70 frames. ], batch size: 97, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:26:13,347 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.133e+02 3.063e+02 3.595e+02 4.621e+02 1.301e+03, threshold=7.190e+02, percent-clipped=5.0 2023-05-15 18:26:23,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-05-15 18:26:34,486 INFO [finetune.py:992] (1/2) Epoch 2, batch 2200, loss[loss=0.1903, simple_loss=0.2776, pruned_loss=0.05149, over 10489.00 frames. ], tot_loss[loss=0.176, simple_loss=0.264, pruned_loss=0.04403, over 2375428.62 frames. ], batch size: 69, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:27:08,549 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114483.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:27:09,930 INFO [finetune.py:992] (1/2) Epoch 2, batch 2250, loss[loss=0.1494, simple_loss=0.2277, pruned_loss=0.03555, over 12340.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2641, pruned_loss=0.04376, over 2373666.03 frames. ], batch size: 30, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:27:25,725 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.121e+02 2.927e+02 3.341e+02 4.008e+02 6.848e+02, threshold=6.682e+02, percent-clipped=0.0 2023-05-15 18:27:29,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-05-15 18:27:35,895 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-15 18:27:46,133 INFO [finetune.py:992] (1/2) Epoch 2, batch 2300, loss[loss=0.1752, simple_loss=0.2689, pruned_loss=0.04075, over 12280.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2642, pruned_loss=0.04397, over 2371156.22 frames. ], batch size: 37, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:27:48,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-05-15 18:27:49,977 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5693, 3.7471, 3.4948, 3.5055, 2.9541, 2.8751, 3.7799, 2.4177], device='cuda:1'), covar=tensor([0.0318, 0.0102, 0.0120, 0.0119, 0.0300, 0.0250, 0.0092, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0158, 0.0145, 0.0174, 0.0195, 0.0190, 0.0152, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:28:21,601 INFO [finetune.py:992] (1/2) Epoch 2, batch 2350, loss[loss=0.1775, simple_loss=0.2761, pruned_loss=0.03946, over 12302.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2638, pruned_loss=0.04386, over 2378576.12 frames. ], batch size: 34, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:28:36,735 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.032e+02 2.856e+02 3.351e+02 4.065e+02 8.891e+02, threshold=6.701e+02, percent-clipped=2.0 2023-05-15 18:28:57,361 INFO [finetune.py:992] (1/2) Epoch 2, batch 2400, loss[loss=0.1697, simple_loss=0.2417, pruned_loss=0.0489, over 11393.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2634, pruned_loss=0.04379, over 2367094.64 frames. ], batch size: 25, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:29:33,897 INFO [finetune.py:992] (1/2) Epoch 2, batch 2450, loss[loss=0.1731, simple_loss=0.2616, pruned_loss=0.04231, over 12089.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2617, pruned_loss=0.04337, over 2370627.37 frames. ], batch size: 32, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:29:48,848 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.690e+02 3.055e+02 4.005e+02 7.689e+02, threshold=6.109e+02, percent-clipped=1.0 2023-05-15 18:29:56,842 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114716.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:30:10,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-15 18:30:10,136 INFO [finetune.py:992] (1/2) Epoch 2, batch 2500, loss[loss=0.1843, simple_loss=0.2788, pruned_loss=0.04489, over 12141.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2621, pruned_loss=0.04351, over 2375674.19 frames. ], batch size: 39, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:30:40,272 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114777.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:30:44,416 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114783.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:30:45,770 INFO [finetune.py:992] (1/2) Epoch 2, batch 2550, loss[loss=0.1579, simple_loss=0.2429, pruned_loss=0.03642, over 12077.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04392, over 2371852.64 frames. ], batch size: 32, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:31:01,295 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.981e+02 2.977e+02 3.410e+02 4.074e+02 7.350e+02, threshold=6.819e+02, percent-clipped=2.0 2023-05-15 18:31:19,230 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114831.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:31:21,915 INFO [finetune.py:992] (1/2) Epoch 2, batch 2600, loss[loss=0.2002, simple_loss=0.2942, pruned_loss=0.05315, over 11583.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04392, over 2371287.89 frames. ], batch size: 48, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:31:48,330 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114871.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:31:57,993 INFO [finetune.py:992] (1/2) Epoch 2, batch 2650, loss[loss=0.1583, simple_loss=0.2421, pruned_loss=0.0373, over 12183.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2628, pruned_loss=0.04402, over 2368017.08 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:32:12,518 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.207e+02 2.908e+02 3.486e+02 4.320e+02 1.243e+03, threshold=6.972e+02, percent-clipped=4.0 2023-05-15 18:32:12,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0475, 4.7450, 4.9383, 4.9103, 4.6480, 4.9893, 4.7736, 2.7811], device='cuda:1'), covar=tensor([0.0115, 0.0069, 0.0083, 0.0063, 0.0056, 0.0088, 0.0088, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0073, 0.0077, 0.0070, 0.0057, 0.0085, 0.0075, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:32:30,973 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114932.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:32:32,963 INFO [finetune.py:992] (1/2) Epoch 2, batch 2700, loss[loss=0.1752, simple_loss=0.2645, pruned_loss=0.04296, over 12344.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2627, pruned_loss=0.04384, over 2375296.62 frames. ], batch size: 36, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:32:34,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-15 18:33:05,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-05-15 18:33:09,019 INFO [finetune.py:992] (1/2) Epoch 2, batch 2750, loss[loss=0.1629, simple_loss=0.2516, pruned_loss=0.03712, over 12180.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2625, pruned_loss=0.04348, over 2381364.35 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:33:24,489 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.889e+02 3.342e+02 4.054e+02 5.766e+02, threshold=6.684e+02, percent-clipped=0.0 2023-05-15 18:33:45,721 INFO [finetune.py:992] (1/2) Epoch 2, batch 2800, loss[loss=0.1659, simple_loss=0.2586, pruned_loss=0.03659, over 10397.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2615, pruned_loss=0.04316, over 2375621.85 frames. ], batch size: 68, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:34:10,685 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9648, 2.1349, 2.2633, 2.2436, 2.0145, 1.6963, 2.2889, 1.7573], device='cuda:1'), covar=tensor([0.0256, 0.0178, 0.0136, 0.0142, 0.0294, 0.0219, 0.0130, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0158, 0.0146, 0.0176, 0.0196, 0.0191, 0.0153, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:34:11,973 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115072.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:34:21,198 INFO [finetune.py:992] (1/2) Epoch 2, batch 2850, loss[loss=0.1568, simple_loss=0.2382, pruned_loss=0.03769, over 12182.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2617, pruned_loss=0.04329, over 2375164.10 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:34:22,101 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115086.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:34:36,496 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 2.884e+02 3.510e+02 4.253e+02 7.197e+02, threshold=7.019e+02, percent-clipped=0.0 2023-05-15 18:34:37,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1686, 4.4583, 4.1031, 4.9278, 4.6216, 2.9054, 4.3011, 2.9875], device='cuda:1'), covar=tensor([0.0893, 0.0860, 0.1396, 0.0386, 0.0947, 0.1594, 0.0898, 0.3192], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0370, 0.0349, 0.0256, 0.0360, 0.0262, 0.0330, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:34:56,915 INFO [finetune.py:992] (1/2) Epoch 2, batch 2900, loss[loss=0.1668, simple_loss=0.2449, pruned_loss=0.04434, over 12004.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2616, pruned_loss=0.04347, over 2376132.56 frames. ], batch size: 28, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:35:05,609 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115147.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:35:25,202 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-15 18:35:27,137 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-15 18:35:33,029 INFO [finetune.py:992] (1/2) Epoch 2, batch 2950, loss[loss=0.1578, simple_loss=0.2475, pruned_loss=0.03406, over 12346.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2606, pruned_loss=0.04307, over 2384942.36 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:35:48,187 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.064e+02 2.824e+02 3.306e+02 3.914e+02 8.627e+02, threshold=6.612e+02, percent-clipped=2.0 2023-05-15 18:36:03,055 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115227.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:36:08,592 INFO [finetune.py:992] (1/2) Epoch 2, batch 3000, loss[loss=0.1565, simple_loss=0.2328, pruned_loss=0.0401, over 12023.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2604, pruned_loss=0.04289, over 2378627.09 frames. ], batch size: 28, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:36:08,592 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 18:36:26,382 INFO [finetune.py:1026] (1/2) Epoch 2, validation: loss=0.3401, simple_loss=0.4115, pruned_loss=0.1344, over 1020973.00 frames. 2023-05-15 18:36:26,383 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 18:37:02,839 INFO [finetune.py:992] (1/2) Epoch 2, batch 3050, loss[loss=0.1801, simple_loss=0.2688, pruned_loss=0.04569, over 12032.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2614, pruned_loss=0.04316, over 2367554.75 frames. ], batch size: 40, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:37:17,766 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 2.803e+02 3.228e+02 3.994e+02 5.351e+02, threshold=6.456e+02, percent-clipped=0.0 2023-05-15 18:37:29,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-15 18:37:38,444 INFO [finetune.py:992] (1/2) Epoch 2, batch 3100, loss[loss=0.1614, simple_loss=0.2565, pruned_loss=0.03318, over 12295.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2611, pruned_loss=0.04284, over 2372150.20 frames. ], batch size: 33, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:37:54,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.23 vs. limit=5.0 2023-05-15 18:38:04,175 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115370.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:38:04,853 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115371.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:38:05,489 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115372.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:38:14,443 INFO [finetune.py:992] (1/2) Epoch 2, batch 3150, loss[loss=0.165, simple_loss=0.2502, pruned_loss=0.03989, over 12160.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2612, pruned_loss=0.04287, over 2367221.55 frames. ], batch size: 36, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:38:29,741 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.799e+02 3.424e+02 4.301e+02 1.225e+03, threshold=6.849e+02, percent-clipped=7.0 2023-05-15 18:38:34,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-15 18:38:40,572 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115420.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:38:44,963 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2129, 2.4060, 3.1136, 4.1341, 2.3600, 4.1723, 4.0804, 4.3249], device='cuda:1'), covar=tensor([0.0143, 0.1045, 0.0394, 0.0107, 0.1134, 0.0189, 0.0161, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0199, 0.0182, 0.0109, 0.0186, 0.0172, 0.0167, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:38:45,660 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5316, 2.9662, 3.8677, 2.3792, 2.5741, 3.1259, 2.9831, 3.2589], device='cuda:1'), covar=tensor([0.0485, 0.0903, 0.0353, 0.1150, 0.1643, 0.1173, 0.1051, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0230, 0.0233, 0.0181, 0.0238, 0.0279, 0.0225, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:38:48,396 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115431.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:38:49,043 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115432.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:38:50,951 INFO [finetune.py:992] (1/2) Epoch 2, batch 3200, loss[loss=0.2041, simple_loss=0.2849, pruned_loss=0.0617, over 12379.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2612, pruned_loss=0.04306, over 2370178.38 frames. ], batch size: 38, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:38:55,956 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115442.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:39:09,718 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3068, 2.9139, 3.8764, 3.2767, 3.6850, 3.3851, 2.7881, 3.8477], device='cuda:1'), covar=tensor([0.0093, 0.0229, 0.0099, 0.0183, 0.0141, 0.0136, 0.0261, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0188, 0.0165, 0.0170, 0.0187, 0.0147, 0.0178, 0.0157], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:39:09,753 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9285, 3.2743, 5.3408, 2.8265, 2.8735, 3.7621, 3.4836, 3.9181], device='cuda:1'), covar=tensor([0.0595, 0.1206, 0.0232, 0.1139, 0.1978, 0.1624, 0.1194, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0230, 0.0232, 0.0181, 0.0237, 0.0278, 0.0225, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:39:26,324 INFO [finetune.py:992] (1/2) Epoch 2, batch 3250, loss[loss=0.1685, simple_loss=0.254, pruned_loss=0.04148, over 12252.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2613, pruned_loss=0.04305, over 2374635.50 frames. ], batch size: 32, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:39:41,622 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.258e+02 3.049e+02 3.540e+02 3.983e+02 9.276e+02, threshold=7.080e+02, percent-clipped=3.0 2023-05-15 18:39:43,919 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0561, 3.7196, 5.3824, 2.6725, 2.9302, 3.9113, 3.5016, 4.0503], device='cuda:1'), covar=tensor([0.0281, 0.0855, 0.0176, 0.1069, 0.1759, 0.1186, 0.1064, 0.1048], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0230, 0.0232, 0.0180, 0.0236, 0.0277, 0.0224, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:39:56,759 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115527.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:40:02,260 INFO [finetune.py:992] (1/2) Epoch 2, batch 3300, loss[loss=0.1936, simple_loss=0.2757, pruned_loss=0.05581, over 10547.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2615, pruned_loss=0.04307, over 2364358.56 frames. ], batch size: 69, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:40:31,688 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115575.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:40:38,836 INFO [finetune.py:992] (1/2) Epoch 2, batch 3350, loss[loss=0.1547, simple_loss=0.2444, pruned_loss=0.03249, over 12022.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2611, pruned_loss=0.04285, over 2366461.47 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 16.0 2023-05-15 18:40:54,235 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.063e+02 2.912e+02 3.332e+02 3.940e+02 7.157e+02, threshold=6.663e+02, percent-clipped=1.0 2023-05-15 18:41:14,825 INFO [finetune.py:992] (1/2) Epoch 2, batch 3400, loss[loss=0.1819, simple_loss=0.2717, pruned_loss=0.04609, over 12062.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2605, pruned_loss=0.04253, over 2374939.46 frames. ], batch size: 42, lr: 4.97e-03, grad_scale: 16.0 2023-05-15 18:41:24,203 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5068, 5.4542, 5.3067, 4.8611, 4.8552, 5.4521, 5.0371, 4.8552], device='cuda:1'), covar=tensor([0.0673, 0.0843, 0.0599, 0.1286, 0.1036, 0.0698, 0.1457, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0573, 0.0504, 0.0460, 0.0577, 0.0374, 0.0651, 0.0714, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 18:41:51,025 INFO [finetune.py:992] (1/2) Epoch 2, batch 3450, loss[loss=0.1658, simple_loss=0.2544, pruned_loss=0.03866, over 12132.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2615, pruned_loss=0.04307, over 2373675.77 frames. ], batch size: 30, lr: 4.97e-03, grad_scale: 16.0 2023-05-15 18:42:06,753 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 2.941e+02 3.439e+02 4.034e+02 9.078e+02, threshold=6.877e+02, percent-clipped=3.0 2023-05-15 18:42:21,123 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115726.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:42:21,728 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115727.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:42:26,830 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4422, 5.0719, 5.3729, 4.7246, 5.1150, 4.6160, 5.3672, 5.1830], device='cuda:1'), covar=tensor([0.0344, 0.0376, 0.0442, 0.0252, 0.0293, 0.0337, 0.0338, 0.0253], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0237, 0.0256, 0.0231, 0.0233, 0.0234, 0.0213, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 18:42:27,380 INFO [finetune.py:992] (1/2) Epoch 2, batch 3500, loss[loss=0.1724, simple_loss=0.2537, pruned_loss=0.04553, over 12195.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2611, pruned_loss=0.04301, over 2380837.68 frames. ], batch size: 29, lr: 4.97e-03, grad_scale: 16.0 2023-05-15 18:42:32,370 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115742.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:43:02,731 INFO [finetune.py:992] (1/2) Epoch 2, batch 3550, loss[loss=0.1702, simple_loss=0.262, pruned_loss=0.03921, over 11591.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2615, pruned_loss=0.04326, over 2375984.19 frames. ], batch size: 48, lr: 4.97e-03, grad_scale: 16.0 2023-05-15 18:43:06,352 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115790.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:43:18,426 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.268e+02 3.032e+02 3.512e+02 4.471e+02 9.547e+02, threshold=7.024e+02, percent-clipped=4.0 2023-05-15 18:43:25,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-15 18:43:29,971 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9716, 4.9395, 4.8053, 4.8328, 4.4540, 4.9313, 4.9812, 5.1542], device='cuda:1'), covar=tensor([0.0198, 0.0140, 0.0170, 0.0301, 0.0767, 0.0230, 0.0146, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0183, 0.0181, 0.0225, 0.0230, 0.0195, 0.0168, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 18:43:38,974 INFO [finetune.py:992] (1/2) Epoch 2, batch 3600, loss[loss=0.1781, simple_loss=0.2697, pruned_loss=0.04319, over 12368.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2622, pruned_loss=0.04348, over 2369898.85 frames. ], batch size: 36, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:44:14,850 INFO [finetune.py:992] (1/2) Epoch 2, batch 3650, loss[loss=0.1612, simple_loss=0.2426, pruned_loss=0.03986, over 12329.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2628, pruned_loss=0.04356, over 2370538.12 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:44:20,038 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115892.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 18:44:30,539 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.927e+02 3.317e+02 4.260e+02 7.259e+02, threshold=6.633e+02, percent-clipped=3.0 2023-05-15 18:44:50,651 INFO [finetune.py:992] (1/2) Epoch 2, batch 3700, loss[loss=0.1702, simple_loss=0.2663, pruned_loss=0.03711, over 12363.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2621, pruned_loss=0.04312, over 2375972.88 frames. ], batch size: 35, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:44:51,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.71 vs. limit=5.0 2023-05-15 18:45:04,485 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115953.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 18:45:27,081 INFO [finetune.py:992] (1/2) Epoch 2, batch 3750, loss[loss=0.1761, simple_loss=0.2601, pruned_loss=0.04605, over 12191.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2619, pruned_loss=0.04337, over 2377421.99 frames. ], batch size: 35, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:45:46,262 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.103e+02 3.010e+02 3.508e+02 3.936e+02 8.193e+02, threshold=7.016e+02, percent-clipped=1.0 2023-05-15 18:45:47,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2023-05-15 18:45:51,427 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116014.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:45:53,514 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0805, 6.0899, 5.8311, 5.2971, 5.1536, 6.0308, 5.5740, 5.2890], device='cuda:1'), covar=tensor([0.0777, 0.0740, 0.0600, 0.1470, 0.0608, 0.0669, 0.1469, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0504, 0.0465, 0.0583, 0.0376, 0.0654, 0.0722, 0.0521], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 18:45:59,346 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4612, 4.3479, 4.4246, 4.4125, 4.1550, 4.4827, 4.3107, 2.5529], device='cuda:1'), covar=tensor([0.0123, 0.0072, 0.0097, 0.0075, 0.0070, 0.0098, 0.0099, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0070, 0.0056, 0.0085, 0.0075, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:45:59,902 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116026.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:46:00,704 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116027.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:46:06,046 INFO [finetune.py:992] (1/2) Epoch 2, batch 3800, loss[loss=0.2065, simple_loss=0.2851, pruned_loss=0.0639, over 11809.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2616, pruned_loss=0.04351, over 2371413.92 frames. ], batch size: 44, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:46:06,953 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4017, 2.2938, 3.0777, 4.3244, 2.2779, 4.3941, 4.2777, 4.5150], device='cuda:1'), covar=tensor([0.0128, 0.1180, 0.0455, 0.0126, 0.1199, 0.0165, 0.0153, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0203, 0.0185, 0.0112, 0.0190, 0.0175, 0.0171, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:46:11,926 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1943, 5.1122, 5.0570, 4.9895, 4.6950, 5.1224, 5.1435, 5.2517], device='cuda:1'), covar=tensor([0.0200, 0.0143, 0.0139, 0.0239, 0.0611, 0.0206, 0.0128, 0.0170], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0183, 0.0181, 0.0225, 0.0231, 0.0196, 0.0169, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 18:46:33,678 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116074.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:46:34,434 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116075.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:46:34,560 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116075.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 18:46:40,606 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2917, 2.5799, 3.8611, 3.2135, 3.6329, 3.3128, 2.5207, 3.7188], device='cuda:1'), covar=tensor([0.0108, 0.0289, 0.0108, 0.0198, 0.0112, 0.0160, 0.0280, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0191, 0.0167, 0.0172, 0.0190, 0.0148, 0.0180, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:46:41,064 INFO [finetune.py:992] (1/2) Epoch 2, batch 3850, loss[loss=0.1661, simple_loss=0.2481, pruned_loss=0.04205, over 12127.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.262, pruned_loss=0.04384, over 2366404.91 frames. ], batch size: 30, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:46:41,365 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2339, 4.5142, 4.0953, 4.9870, 4.6565, 2.8549, 4.3021, 3.0354], device='cuda:1'), covar=tensor([0.0865, 0.0805, 0.1416, 0.0315, 0.0970, 0.1487, 0.0893, 0.3085], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0367, 0.0348, 0.0257, 0.0357, 0.0260, 0.0330, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:46:50,860 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5034, 2.2516, 2.9468, 2.5945, 2.8224, 2.7305, 2.0809, 2.9506], device='cuda:1'), covar=tensor([0.0096, 0.0259, 0.0136, 0.0202, 0.0173, 0.0159, 0.0290, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0190, 0.0167, 0.0171, 0.0190, 0.0147, 0.0180, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:46:55,795 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116105.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:46:56,951 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 3.031e+02 3.770e+02 4.514e+02 9.662e+02, threshold=7.540e+02, percent-clipped=3.0 2023-05-15 18:47:03,532 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116116.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:47:17,413 INFO [finetune.py:992] (1/2) Epoch 2, batch 3900, loss[loss=0.1739, simple_loss=0.2536, pruned_loss=0.04707, over 12344.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2618, pruned_loss=0.04345, over 2375346.32 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:47:30,746 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0467, 2.3391, 2.3875, 2.3186, 2.1571, 1.8438, 2.3548, 1.7734], device='cuda:1'), covar=tensor([0.0268, 0.0146, 0.0125, 0.0146, 0.0273, 0.0206, 0.0128, 0.0310], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0154, 0.0144, 0.0171, 0.0194, 0.0187, 0.0150, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:47:31,391 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6243, 4.4892, 4.5581, 4.6091, 4.3113, 4.3601, 4.1799, 4.5412], device='cuda:1'), covar=tensor([0.0553, 0.0545, 0.0687, 0.0545, 0.1735, 0.1076, 0.0532, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0628, 0.0537, 0.0594, 0.0779, 0.0700, 0.0516, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 18:47:39,376 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116166.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:47:41,380 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1097, 5.9021, 5.3583, 5.4532, 6.0128, 5.2423, 5.6065, 5.5592], device='cuda:1'), covar=tensor([0.1417, 0.0796, 0.0950, 0.1929, 0.0926, 0.2172, 0.1430, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0436, 0.0347, 0.0396, 0.0422, 0.0396, 0.0358, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:47:47,207 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116177.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:47:52,791 INFO [finetune.py:992] (1/2) Epoch 2, batch 3950, loss[loss=0.1472, simple_loss=0.2305, pruned_loss=0.0319, over 12020.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2623, pruned_loss=0.04358, over 2371130.99 frames. ], batch size: 28, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:48:08,575 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.779e+02 3.406e+02 4.033e+02 7.927e+02, threshold=6.812e+02, percent-clipped=2.0 2023-05-15 18:48:28,885 INFO [finetune.py:992] (1/2) Epoch 2, batch 4000, loss[loss=0.1761, simple_loss=0.2624, pruned_loss=0.04495, over 12111.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2628, pruned_loss=0.04367, over 2370003.66 frames. ], batch size: 33, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:48:38,165 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116248.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 18:48:56,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 18:49:04,981 INFO [finetune.py:992] (1/2) Epoch 2, batch 4050, loss[loss=0.1805, simple_loss=0.274, pruned_loss=0.04346, over 12137.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2628, pruned_loss=0.04375, over 2372832.79 frames. ], batch size: 39, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:49:15,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 18:49:16,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-05-15 18:49:20,558 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.857e+02 3.441e+02 3.984e+02 7.490e+02, threshold=6.881e+02, percent-clipped=3.0 2023-05-15 18:49:31,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-15 18:49:32,062 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116323.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:49:40,135 INFO [finetune.py:992] (1/2) Epoch 2, batch 4100, loss[loss=0.1916, simple_loss=0.284, pruned_loss=0.04962, over 12281.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04356, over 2379657.35 frames. ], batch size: 37, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:49:45,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9508, 4.2270, 3.9079, 4.6429, 4.2623, 2.7146, 3.8383, 2.8705], device='cuda:1'), covar=tensor([0.0755, 0.0735, 0.1099, 0.0353, 0.0945, 0.1478, 0.1046, 0.2819], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0369, 0.0349, 0.0257, 0.0358, 0.0261, 0.0332, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:50:05,280 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116370.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 18:50:16,042 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116384.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:50:16,524 INFO [finetune.py:992] (1/2) Epoch 2, batch 4150, loss[loss=0.1484, simple_loss=0.2408, pruned_loss=0.02803, over 12361.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2628, pruned_loss=0.04388, over 2363620.75 frames. ], batch size: 35, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:50:32,530 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.865e+02 3.386e+02 4.375e+02 9.798e+02, threshold=6.772e+02, percent-clipped=6.0 2023-05-15 18:50:51,313 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.36 vs. limit=5.0 2023-05-15 18:50:53,135 INFO [finetune.py:992] (1/2) Epoch 2, batch 4200, loss[loss=0.1969, simple_loss=0.2908, pruned_loss=0.05151, over 11643.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04363, over 2364495.08 frames. ], batch size: 48, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:51:00,804 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116446.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:51:11,256 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116461.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:51:18,566 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2678, 2.8464, 2.9138, 2.7911, 2.5193, 2.3573, 2.9535, 2.0281], device='cuda:1'), covar=tensor([0.0296, 0.0134, 0.0139, 0.0141, 0.0292, 0.0234, 0.0112, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0154, 0.0144, 0.0173, 0.0194, 0.0186, 0.0151, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:51:19,200 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116472.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:51:22,918 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9659, 4.8188, 4.9017, 4.9513, 4.7565, 4.9508, 4.8994, 2.6420], device='cuda:1'), covar=tensor([0.0092, 0.0055, 0.0066, 0.0053, 0.0048, 0.0075, 0.0060, 0.0646], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0075, 0.0069, 0.0056, 0.0085, 0.0074, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:51:28,455 INFO [finetune.py:992] (1/2) Epoch 2, batch 4250, loss[loss=0.1535, simple_loss=0.2404, pruned_loss=0.0333, over 12293.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2624, pruned_loss=0.0438, over 2356312.72 frames. ], batch size: 33, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:51:37,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8305, 2.8762, 4.9065, 4.9962, 2.9911, 2.8786, 3.0594, 2.2714], device='cuda:1'), covar=tensor([0.1265, 0.2742, 0.0330, 0.0345, 0.1085, 0.1741, 0.2399, 0.3466], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0365, 0.0261, 0.0285, 0.0248, 0.0272, 0.0348, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:51:39,234 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8294, 3.3189, 5.1252, 2.6449, 2.7942, 3.6629, 3.0692, 3.7628], device='cuda:1'), covar=tensor([0.0334, 0.1139, 0.0225, 0.1199, 0.1879, 0.1469, 0.1402, 0.0994], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0227, 0.0231, 0.0179, 0.0233, 0.0276, 0.0222, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:51:42,766 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116505.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:51:43,996 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.921e+02 3.307e+02 4.423e+02 2.186e+03, threshold=6.613e+02, percent-clipped=3.0 2023-05-15 18:51:44,261 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116507.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 18:51:56,937 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9884, 5.8959, 5.4624, 5.4093, 5.9614, 5.3161, 5.5361, 5.5023], device='cuda:1'), covar=tensor([0.1542, 0.0980, 0.1005, 0.2323, 0.1168, 0.2424, 0.1860, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0445, 0.0356, 0.0406, 0.0434, 0.0406, 0.0367, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:52:04,676 INFO [finetune.py:992] (1/2) Epoch 2, batch 4300, loss[loss=0.1691, simple_loss=0.2511, pruned_loss=0.04357, over 12195.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2617, pruned_loss=0.04362, over 2361886.92 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:52:13,901 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116548.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 18:52:26,798 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116566.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:52:27,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2270, 3.1809, 3.1730, 2.9636, 2.6616, 2.5385, 3.1748, 1.9645], device='cuda:1'), covar=tensor([0.0378, 0.0127, 0.0123, 0.0148, 0.0281, 0.0239, 0.0098, 0.0414], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0154, 0.0144, 0.0173, 0.0194, 0.0187, 0.0151, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 18:52:40,707 INFO [finetune.py:992] (1/2) Epoch 2, batch 4350, loss[loss=0.146, simple_loss=0.2257, pruned_loss=0.03319, over 11781.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2622, pruned_loss=0.0439, over 2357404.27 frames. ], batch size: 26, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:52:48,469 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116596.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 18:52:56,313 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.055e+02 3.059e+02 3.531e+02 4.344e+02 1.283e+03, threshold=7.062e+02, percent-clipped=4.0 2023-05-15 18:53:04,263 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116618.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:53:13,535 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4682, 4.6689, 4.1743, 5.0627, 4.7758, 2.9874, 4.5240, 3.1747], device='cuda:1'), covar=tensor([0.0687, 0.0817, 0.1242, 0.0370, 0.0952, 0.1528, 0.0821, 0.3032], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0366, 0.0344, 0.0254, 0.0355, 0.0258, 0.0327, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:53:16,088 INFO [finetune.py:992] (1/2) Epoch 2, batch 4400, loss[loss=0.1542, simple_loss=0.244, pruned_loss=0.03225, over 12301.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2625, pruned_loss=0.04397, over 2365744.88 frames. ], batch size: 34, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:53:29,923 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9619, 4.2980, 3.7849, 4.6645, 4.2813, 2.7912, 4.1084, 2.9632], device='cuda:1'), covar=tensor([0.0720, 0.0733, 0.1190, 0.0369, 0.0947, 0.1515, 0.0798, 0.2923], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0365, 0.0343, 0.0253, 0.0354, 0.0258, 0.0326, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:53:41,565 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116670.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 18:53:47,795 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:53:47,939 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:53:51,914 INFO [finetune.py:992] (1/2) Epoch 2, batch 4450, loss[loss=0.1702, simple_loss=0.2637, pruned_loss=0.03835, over 11214.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2629, pruned_loss=0.04405, over 2370808.75 frames. ], batch size: 55, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:54:05,666 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116704.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:54:07,669 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.725e+02 3.200e+02 3.613e+02 8.473e+02, threshold=6.400e+02, percent-clipped=2.0 2023-05-15 18:54:15,288 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116718.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:54:28,086 INFO [finetune.py:992] (1/2) Epoch 2, batch 4500, loss[loss=0.1978, simple_loss=0.2852, pruned_loss=0.05519, over 12024.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2634, pruned_loss=0.04442, over 2361103.48 frames. ], batch size: 40, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:54:34,049 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-15 18:54:38,924 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3030, 4.7856, 2.9899, 2.7499, 4.0897, 2.6704, 4.0726, 3.3070], device='cuda:1'), covar=tensor([0.0682, 0.0619, 0.0987, 0.1450, 0.0255, 0.1211, 0.0431, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0247, 0.0173, 0.0196, 0.0135, 0.0179, 0.0192, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:54:46,715 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116761.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:54:49,645 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116765.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:54:54,611 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:55:03,557 INFO [finetune.py:992] (1/2) Epoch 2, batch 4550, loss[loss=0.1634, simple_loss=0.2544, pruned_loss=0.03622, over 12351.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2634, pruned_loss=0.04458, over 2356629.70 frames. ], batch size: 35, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:55:16,604 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116802.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 18:55:16,735 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116802.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:55:20,080 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.993e+02 3.616e+02 4.229e+02 7.475e+02, threshold=7.232e+02, percent-clipped=4.0 2023-05-15 18:55:20,324 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3041, 4.6382, 2.8335, 2.6423, 3.9018, 2.5189, 3.8891, 3.1708], device='cuda:1'), covar=tensor([0.0645, 0.0463, 0.0958, 0.1401, 0.0305, 0.1243, 0.0445, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0247, 0.0173, 0.0196, 0.0135, 0.0179, 0.0192, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:55:21,580 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116809.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:55:29,380 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:55:40,105 INFO [finetune.py:992] (1/2) Epoch 2, batch 4600, loss[loss=0.1679, simple_loss=0.246, pruned_loss=0.04489, over 12336.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2635, pruned_loss=0.04458, over 2351317.32 frames. ], batch size: 31, lr: 4.97e-03, grad_scale: 8.0 2023-05-15 18:55:51,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-15 18:55:59,265 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116861.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:56:00,799 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:56:04,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 2023-05-15 18:56:05,843 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4890, 2.2872, 3.2475, 4.4171, 2.3628, 4.5214, 4.4857, 4.6380], device='cuda:1'), covar=tensor([0.0114, 0.1347, 0.0431, 0.0097, 0.1253, 0.0150, 0.0099, 0.0066], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0206, 0.0187, 0.0114, 0.0192, 0.0178, 0.0171, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:56:16,382 INFO [finetune.py:992] (1/2) Epoch 2, batch 4650, loss[loss=0.1531, simple_loss=0.2379, pruned_loss=0.03416, over 12343.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2627, pruned_loss=0.04429, over 2360042.02 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 18:56:31,905 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.962e+02 3.393e+02 3.992e+02 1.065e+03, threshold=6.785e+02, percent-clipped=1.0 2023-05-15 18:56:34,959 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4246, 4.7911, 2.9448, 2.6184, 4.0665, 2.7182, 3.9787, 3.3770], device='cuda:1'), covar=tensor([0.0634, 0.0574, 0.1009, 0.1521, 0.0247, 0.1173, 0.0503, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0246, 0.0172, 0.0195, 0.0135, 0.0178, 0.0191, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:56:51,495 INFO [finetune.py:992] (1/2) Epoch 2, batch 4700, loss[loss=0.1612, simple_loss=0.2557, pruned_loss=0.03336, over 12279.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2626, pruned_loss=0.04421, over 2356656.38 frames. ], batch size: 37, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 18:56:54,506 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0069, 5.0115, 4.8451, 4.8963, 4.4695, 4.9602, 5.0099, 5.1658], device='cuda:1'), covar=tensor([0.0283, 0.0149, 0.0216, 0.0294, 0.0881, 0.0356, 0.0195, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0181, 0.0181, 0.0223, 0.0227, 0.0195, 0.0168, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 18:57:05,758 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6652, 2.7999, 4.5084, 4.6267, 2.9955, 2.6979, 2.9488, 2.1313], device='cuda:1'), covar=tensor([0.1357, 0.2848, 0.0406, 0.0409, 0.1128, 0.1881, 0.2273, 0.3555], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0369, 0.0264, 0.0288, 0.0251, 0.0277, 0.0353, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:57:06,591 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-15 18:57:19,854 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116974.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:57:23,487 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116979.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:57:27,772 INFO [finetune.py:992] (1/2) Epoch 2, batch 4750, loss[loss=0.1747, simple_loss=0.2699, pruned_loss=0.03971, over 12306.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2624, pruned_loss=0.04372, over 2357845.37 frames. ], batch size: 34, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 18:57:44,290 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.939e+02 3.466e+02 4.060e+02 9.560e+02, threshold=6.933e+02, percent-clipped=2.0 2023-05-15 18:57:58,579 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117027.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:58:03,730 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5450, 2.5551, 3.4524, 4.4821, 2.2926, 4.5477, 4.4780, 4.7490], device='cuda:1'), covar=tensor([0.0179, 0.1152, 0.0365, 0.0117, 0.1233, 0.0184, 0.0155, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0205, 0.0187, 0.0114, 0.0192, 0.0179, 0.0171, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:58:04,252 INFO [finetune.py:992] (1/2) Epoch 2, batch 4800, loss[loss=0.1567, simple_loss=0.2333, pruned_loss=0.04002, over 12296.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2609, pruned_loss=0.04341, over 2361996.45 frames. ], batch size: 28, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 18:58:21,976 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117060.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:58:34,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3236, 4.5528, 4.2217, 4.9712, 4.5792, 3.1354, 4.2142, 3.0977], device='cuda:1'), covar=tensor([0.0733, 0.0748, 0.1107, 0.0288, 0.1016, 0.1337, 0.0959, 0.2838], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0364, 0.0344, 0.0254, 0.0354, 0.0256, 0.0326, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:58:39,423 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.17 vs. limit=5.0 2023-05-15 18:58:39,690 INFO [finetune.py:992] (1/2) Epoch 2, batch 4850, loss[loss=0.1837, simple_loss=0.2673, pruned_loss=0.05004, over 11820.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2618, pruned_loss=0.04385, over 2355339.02 frames. ], batch size: 44, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 18:58:46,278 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4460, 4.6455, 4.2851, 5.1025, 4.7411, 3.1662, 4.2905, 3.1855], device='cuda:1'), covar=tensor([0.0668, 0.0796, 0.1151, 0.0309, 0.0905, 0.1333, 0.1027, 0.2837], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0364, 0.0344, 0.0254, 0.0354, 0.0256, 0.0326, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 18:58:51,358 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8588, 3.2907, 5.2254, 2.7893, 3.0953, 3.8367, 3.4072, 3.9400], device='cuda:1'), covar=tensor([0.0379, 0.1193, 0.0250, 0.1152, 0.1720, 0.1451, 0.1258, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0228, 0.0231, 0.0179, 0.0234, 0.0277, 0.0223, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 18:58:52,683 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117102.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:58:56,075 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.262e+02 2.964e+02 3.334e+02 4.015e+02 7.759e+02, threshold=6.668e+02, percent-clipped=1.0 2023-05-15 18:59:06,188 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9459, 5.6571, 5.2513, 5.3175, 5.8394, 5.0636, 5.3025, 5.2842], device='cuda:1'), covar=tensor([0.1160, 0.0944, 0.0936, 0.1627, 0.0843, 0.1846, 0.1704, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0448, 0.0357, 0.0406, 0.0434, 0.0401, 0.0368, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 18:59:15,516 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5598, 4.5152, 4.5202, 4.5989, 4.3075, 4.3032, 4.1808, 4.5127], device='cuda:1'), covar=tensor([0.0716, 0.0551, 0.0897, 0.0575, 0.1623, 0.1342, 0.0554, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0627, 0.0532, 0.0588, 0.0767, 0.0699, 0.0513, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 18:59:16,082 INFO [finetune.py:992] (1/2) Epoch 2, batch 4900, loss[loss=0.1807, simple_loss=0.2724, pruned_loss=0.04455, over 10472.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.261, pruned_loss=0.04319, over 2362817.36 frames. ], batch size: 68, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 18:59:26,761 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117150.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:59:33,156 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117158.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:59:35,314 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117161.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 18:59:36,622 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3730, 5.2039, 5.2848, 5.3564, 4.9835, 4.9903, 4.7443, 5.2957], device='cuda:1'), covar=tensor([0.0558, 0.0512, 0.0626, 0.0544, 0.1748, 0.1210, 0.0574, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0631, 0.0535, 0.0593, 0.0772, 0.0704, 0.0517, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 18:59:52,166 INFO [finetune.py:992] (1/2) Epoch 2, batch 4950, loss[loss=0.2205, simple_loss=0.2929, pruned_loss=0.0741, over 8194.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2615, pruned_loss=0.04341, over 2364666.03 frames. ], batch size: 98, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:00:08,168 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.817e+02 3.366e+02 4.085e+02 9.670e+02, threshold=6.732e+02, percent-clipped=4.0 2023-05-15 19:00:09,702 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117209.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:00:18,310 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117221.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:00:28,752 INFO [finetune.py:992] (1/2) Epoch 2, batch 5000, loss[loss=0.1682, simple_loss=0.2547, pruned_loss=0.0408, over 12321.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2623, pruned_loss=0.04393, over 2355332.13 frames. ], batch size: 34, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:00:56,708 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117274.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:01:02,458 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117282.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:01:04,411 INFO [finetune.py:992] (1/2) Epoch 2, batch 5050, loss[loss=0.1622, simple_loss=0.2418, pruned_loss=0.04127, over 12338.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2627, pruned_loss=0.0441, over 2358728.17 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:01:08,348 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3419, 4.5315, 4.0750, 4.9780, 4.6129, 3.0151, 4.2771, 3.1839], device='cuda:1'), covar=tensor([0.0722, 0.0774, 0.1324, 0.0347, 0.0979, 0.1404, 0.0958, 0.2732], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0364, 0.0345, 0.0253, 0.0354, 0.0257, 0.0326, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:01:10,960 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2052, 5.2263, 5.1058, 5.1693, 4.7015, 5.2226, 5.2413, 5.3659], device='cuda:1'), covar=tensor([0.0182, 0.0108, 0.0145, 0.0227, 0.0681, 0.0251, 0.0127, 0.0145], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0182, 0.0181, 0.0223, 0.0229, 0.0195, 0.0168, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 19:01:20,677 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 3.022e+02 3.419e+02 3.930e+02 8.423e+02, threshold=6.837e+02, percent-clipped=1.0 2023-05-15 19:01:31,456 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117322.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:01:36,484 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117329.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:01:40,334 INFO [finetune.py:992] (1/2) Epoch 2, batch 5100, loss[loss=0.1705, simple_loss=0.259, pruned_loss=0.04097, over 12348.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.263, pruned_loss=0.04365, over 2367001.59 frames. ], batch size: 36, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:01:44,162 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6975, 2.7510, 4.4552, 4.6668, 3.1555, 2.6920, 3.0839, 1.9549], device='cuda:1'), covar=tensor([0.1325, 0.2567, 0.0388, 0.0348, 0.0989, 0.1883, 0.2099, 0.3574], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0369, 0.0263, 0.0286, 0.0250, 0.0275, 0.0352, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:01:48,291 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6078, 4.3508, 4.4519, 4.5703, 4.3763, 4.4873, 4.4899, 2.7989], device='cuda:1'), covar=tensor([0.0086, 0.0070, 0.0078, 0.0061, 0.0046, 0.0091, 0.0069, 0.0567], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0070, 0.0056, 0.0086, 0.0075, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:01:58,289 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117360.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:02:16,503 INFO [finetune.py:992] (1/2) Epoch 2, batch 5150, loss[loss=0.1882, simple_loss=0.2585, pruned_loss=0.05891, over 12141.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2622, pruned_loss=0.04353, over 2367851.43 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:02:20,334 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117390.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:02:32,754 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.982e+02 3.418e+02 4.155e+02 8.976e+02, threshold=6.836e+02, percent-clipped=2.0 2023-05-15 19:02:33,546 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117408.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:02:52,218 INFO [finetune.py:992] (1/2) Epoch 2, batch 5200, loss[loss=0.1882, simple_loss=0.2757, pruned_loss=0.05031, over 11320.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2626, pruned_loss=0.04353, over 2357525.34 frames. ], batch size: 55, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:03:09,005 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117458.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:03:16,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9795, 5.7146, 5.3282, 5.3231, 5.8181, 5.0337, 5.3194, 5.3485], device='cuda:1'), covar=tensor([0.1150, 0.0862, 0.0890, 0.1790, 0.0929, 0.2174, 0.1847, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0451, 0.0361, 0.0410, 0.0439, 0.0407, 0.0371, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:03:28,090 INFO [finetune.py:992] (1/2) Epoch 2, batch 5250, loss[loss=0.157, simple_loss=0.2381, pruned_loss=0.03794, over 12348.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.04443, over 2348985.09 frames. ], batch size: 31, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:03:35,841 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3144, 4.8463, 5.3229, 4.6449, 4.9684, 4.6964, 5.3410, 4.9786], device='cuda:1'), covar=tensor([0.0264, 0.0313, 0.0227, 0.0227, 0.0262, 0.0272, 0.0187, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0234, 0.0250, 0.0229, 0.0228, 0.0230, 0.0209, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:03:42,933 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117506.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:03:43,579 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.118e+02 2.779e+02 3.417e+02 4.487e+02 9.626e+02, threshold=6.834e+02, percent-clipped=5.0 2023-05-15 19:04:04,133 INFO [finetune.py:992] (1/2) Epoch 2, batch 5300, loss[loss=0.171, simple_loss=0.2529, pruned_loss=0.0445, over 12172.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2632, pruned_loss=0.04442, over 2363861.10 frames. ], batch size: 31, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:04:24,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-15 19:04:33,721 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117577.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:04:39,283 INFO [finetune.py:992] (1/2) Epoch 2, batch 5350, loss[loss=0.144, simple_loss=0.2328, pruned_loss=0.02755, over 12002.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2635, pruned_loss=0.04461, over 2361386.11 frames. ], batch size: 28, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:04:48,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-15 19:04:55,726 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.946e+02 3.399e+02 4.247e+02 1.173e+03, threshold=6.799e+02, percent-clipped=4.0 2023-05-15 19:05:15,437 INFO [finetune.py:992] (1/2) Epoch 2, batch 5400, loss[loss=0.1705, simple_loss=0.2631, pruned_loss=0.03899, over 12282.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2631, pruned_loss=0.04465, over 2364658.95 frames. ], batch size: 37, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:05:41,988 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1344, 1.9867, 2.4305, 2.2698, 2.3836, 2.4239, 1.8904, 2.4507], device='cuda:1'), covar=tensor([0.0082, 0.0247, 0.0158, 0.0163, 0.0140, 0.0131, 0.0246, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0190, 0.0168, 0.0172, 0.0190, 0.0147, 0.0181, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:05:51,782 INFO [finetune.py:992] (1/2) Epoch 2, batch 5450, loss[loss=0.1676, simple_loss=0.2606, pruned_loss=0.03724, over 12196.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2641, pruned_loss=0.04485, over 2359549.25 frames. ], batch size: 35, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:05:51,882 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117685.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:05:56,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3168, 4.9630, 5.2324, 5.1162, 4.9821, 5.1122, 5.1034, 2.8964], device='cuda:1'), covar=tensor([0.0091, 0.0058, 0.0047, 0.0055, 0.0036, 0.0076, 0.0060, 0.0594], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0073, 0.0076, 0.0070, 0.0057, 0.0086, 0.0074, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:06:07,355 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.792e+02 3.327e+02 3.916e+02 6.376e+02, threshold=6.653e+02, percent-clipped=0.0 2023-05-15 19:06:27,150 INFO [finetune.py:992] (1/2) Epoch 2, batch 5500, loss[loss=0.1539, simple_loss=0.2435, pruned_loss=0.03208, over 12113.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2632, pruned_loss=0.04431, over 2370372.78 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:06:30,838 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5747, 4.6784, 4.3879, 5.1044, 4.7205, 2.9223, 4.3280, 3.1467], device='cuda:1'), covar=tensor([0.0694, 0.0788, 0.1298, 0.0371, 0.1157, 0.1629, 0.1035, 0.3028], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0366, 0.0347, 0.0256, 0.0356, 0.0259, 0.0328, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:06:42,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.04 vs. limit=5.0 2023-05-15 19:06:44,809 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3798, 4.8087, 2.9263, 2.9532, 4.0777, 2.5241, 4.0224, 3.1573], device='cuda:1'), covar=tensor([0.0594, 0.0344, 0.1031, 0.1168, 0.0229, 0.1243, 0.0416, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0246, 0.0173, 0.0195, 0.0135, 0.0179, 0.0191, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:06:55,165 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8529, 3.5863, 3.7207, 3.8186, 3.7607, 3.8660, 3.7625, 2.6297], device='cuda:1'), covar=tensor([0.0087, 0.0081, 0.0094, 0.0067, 0.0057, 0.0090, 0.0078, 0.0574], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0073, 0.0076, 0.0070, 0.0057, 0.0085, 0.0074, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:07:03,035 INFO [finetune.py:992] (1/2) Epoch 2, batch 5550, loss[loss=0.1623, simple_loss=0.2433, pruned_loss=0.04061, over 12088.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2633, pruned_loss=0.04446, over 2367160.09 frames. ], batch size: 32, lr: 4.96e-03, grad_scale: 8.0 2023-05-15 19:07:03,914 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0001, 5.9813, 5.7381, 5.2401, 5.0719, 5.9196, 5.5311, 5.3283], device='cuda:1'), covar=tensor([0.0614, 0.0760, 0.0580, 0.1412, 0.0639, 0.0611, 0.1255, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0511, 0.0466, 0.0582, 0.0381, 0.0661, 0.0721, 0.0521], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:07:19,627 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.047e+02 2.985e+02 3.531e+02 4.289e+02 1.406e+03, threshold=7.062e+02, percent-clipped=6.0 2023-05-15 19:07:39,501 INFO [finetune.py:992] (1/2) Epoch 2, batch 5600, loss[loss=0.1397, simple_loss=0.2187, pruned_loss=0.03039, over 12251.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2627, pruned_loss=0.04445, over 2361143.46 frames. ], batch size: 28, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:08:09,117 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117877.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:08:10,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-15 19:08:15,298 INFO [finetune.py:992] (1/2) Epoch 2, batch 5650, loss[loss=0.2015, simple_loss=0.2844, pruned_loss=0.05931, over 12001.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2631, pruned_loss=0.04449, over 2364670.77 frames. ], batch size: 42, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:08:30,774 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.114e+02 2.959e+02 3.524e+02 4.205e+02 8.301e+02, threshold=7.048e+02, percent-clipped=3.0 2023-05-15 19:08:43,539 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117925.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:08:50,585 INFO [finetune.py:992] (1/2) Epoch 2, batch 5700, loss[loss=0.1488, simple_loss=0.2281, pruned_loss=0.03471, over 11877.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2644, pruned_loss=0.04542, over 2353327.21 frames. ], batch size: 26, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:09:26,514 INFO [finetune.py:992] (1/2) Epoch 2, batch 5750, loss[loss=0.164, simple_loss=0.2476, pruned_loss=0.04026, over 12080.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2629, pruned_loss=0.04453, over 2368440.27 frames. ], batch size: 32, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:09:26,708 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117985.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:09:45,348 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.023e+02 2.769e+02 3.355e+02 4.095e+02 8.466e+02, threshold=6.710e+02, percent-clipped=3.0 2023-05-15 19:10:00,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.72 vs. limit=5.0 2023-05-15 19:10:04,809 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118033.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:10:06,179 INFO [finetune.py:992] (1/2) Epoch 2, batch 5800, loss[loss=0.1853, simple_loss=0.2709, pruned_loss=0.04985, over 12095.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2636, pruned_loss=0.04493, over 2368490.54 frames. ], batch size: 32, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:10:09,926 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118040.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:10:15,436 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1628, 6.1303, 5.9263, 5.4068, 5.2057, 6.0967, 5.6399, 5.3907], device='cuda:1'), covar=tensor([0.0707, 0.0836, 0.0526, 0.1364, 0.0560, 0.0671, 0.1534, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0508, 0.0466, 0.0585, 0.0380, 0.0662, 0.0720, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:10:41,892 INFO [finetune.py:992] (1/2) Epoch 2, batch 5850, loss[loss=0.1548, simple_loss=0.2357, pruned_loss=0.03692, over 12134.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2636, pruned_loss=0.04469, over 2372241.43 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:10:54,021 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118101.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:10:58,038 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 2.830e+02 3.539e+02 4.021e+02 7.766e+02, threshold=7.079e+02, percent-clipped=1.0 2023-05-15 19:11:17,738 INFO [finetune.py:992] (1/2) Epoch 2, batch 5900, loss[loss=0.1949, simple_loss=0.2865, pruned_loss=0.05164, over 11341.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2639, pruned_loss=0.04489, over 2368280.29 frames. ], batch size: 55, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:11:41,806 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118168.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:11:42,017 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2023-05-15 19:11:53,734 INFO [finetune.py:992] (1/2) Epoch 2, batch 5950, loss[loss=0.1557, simple_loss=0.2403, pruned_loss=0.03559, over 12369.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2636, pruned_loss=0.04481, over 2367525.98 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:11:56,802 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6109, 4.9798, 2.9912, 2.4667, 4.2459, 2.5724, 4.1427, 3.3644], device='cuda:1'), covar=tensor([0.0598, 0.0340, 0.1069, 0.1638, 0.0272, 0.1384, 0.0371, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0245, 0.0173, 0.0195, 0.0136, 0.0179, 0.0191, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:12:09,706 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 2.846e+02 3.350e+02 3.911e+02 9.426e+02, threshold=6.700e+02, percent-clipped=2.0 2023-05-15 19:12:25,185 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118229.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 19:12:29,091 INFO [finetune.py:992] (1/2) Epoch 2, batch 6000, loss[loss=0.2747, simple_loss=0.3323, pruned_loss=0.1086, over 8097.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2636, pruned_loss=0.04506, over 2365116.18 frames. ], batch size: 97, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:12:29,091 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 19:12:47,236 INFO [finetune.py:1026] (1/2) Epoch 2, validation: loss=0.3365, simple_loss=0.4084, pruned_loss=0.1323, over 1020973.00 frames. 2023-05-15 19:12:47,237 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 19:13:01,318 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9377, 4.7697, 4.8036, 4.8419, 4.6775, 4.9523, 4.8104, 2.5446], device='cuda:1'), covar=tensor([0.0123, 0.0077, 0.0090, 0.0080, 0.0063, 0.0080, 0.0082, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0070, 0.0057, 0.0086, 0.0075, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:13:11,066 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118269.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:13:22,961 INFO [finetune.py:992] (1/2) Epoch 2, batch 6050, loss[loss=0.1756, simple_loss=0.2647, pruned_loss=0.04331, over 12373.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2634, pruned_loss=0.04489, over 2371967.76 frames. ], batch size: 38, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:13:26,055 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118289.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:13:38,428 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.286e+02 2.970e+02 3.436e+02 4.188e+02 8.999e+02, threshold=6.871e+02, percent-clipped=4.0 2023-05-15 19:13:53,933 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118329.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:13:54,722 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118330.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:13:54,808 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3589, 4.6911, 4.1209, 5.0052, 4.6777, 2.9796, 4.3734, 3.0850], device='cuda:1'), covar=tensor([0.0676, 0.0732, 0.1391, 0.0291, 0.0857, 0.1445, 0.0844, 0.2916], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0363, 0.0344, 0.0253, 0.0353, 0.0256, 0.0324, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:13:58,734 INFO [finetune.py:992] (1/2) Epoch 2, batch 6100, loss[loss=0.1769, simple_loss=0.2658, pruned_loss=0.04405, over 12272.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2652, pruned_loss=0.04575, over 2368796.49 frames. ], batch size: 37, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:14:03,180 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3958, 4.7813, 2.9634, 2.6300, 3.9928, 2.7205, 4.0611, 3.3336], device='cuda:1'), covar=tensor([0.0620, 0.0381, 0.1023, 0.1361, 0.0236, 0.1142, 0.0386, 0.0727], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0243, 0.0172, 0.0194, 0.0136, 0.0178, 0.0189, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:14:09,644 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118350.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:14:34,233 INFO [finetune.py:992] (1/2) Epoch 2, batch 6150, loss[loss=0.1642, simple_loss=0.2528, pruned_loss=0.03784, over 12148.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2649, pruned_loss=0.04518, over 2377700.04 frames. ], batch size: 36, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:14:38,150 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118390.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:14:42,280 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118396.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:14:50,159 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.033e+02 2.985e+02 3.360e+02 4.001e+02 1.245e+03, threshold=6.721e+02, percent-clipped=3.0 2023-05-15 19:15:10,798 INFO [finetune.py:992] (1/2) Epoch 2, batch 6200, loss[loss=0.2078, simple_loss=0.2995, pruned_loss=0.05803, over 12052.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2646, pruned_loss=0.04472, over 2385922.35 frames. ], batch size: 42, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:15:46,835 INFO [finetune.py:992] (1/2) Epoch 2, batch 6250, loss[loss=0.1642, simple_loss=0.2552, pruned_loss=0.03658, over 12075.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2645, pruned_loss=0.04464, over 2385421.08 frames. ], batch size: 32, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:16:02,308 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.962e+02 3.437e+02 4.041e+02 7.964e+02, threshold=6.874e+02, percent-clipped=1.0 2023-05-15 19:16:09,368 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9764, 3.7839, 3.9959, 3.6554, 3.8457, 3.6073, 3.9513, 3.5943], device='cuda:1'), covar=tensor([0.0319, 0.0302, 0.0291, 0.0244, 0.0281, 0.0301, 0.0256, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0237, 0.0257, 0.0232, 0.0232, 0.0234, 0.0212, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:16:14,272 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118524.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 19:16:21,811 INFO [finetune.py:992] (1/2) Epoch 2, batch 6300, loss[loss=0.2281, simple_loss=0.3061, pruned_loss=0.07508, over 10429.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2644, pruned_loss=0.04463, over 2384141.11 frames. ], batch size: 68, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:16:42,657 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4987, 4.7597, 4.2047, 5.2043, 4.7271, 3.0193, 4.4596, 3.2641], device='cuda:1'), covar=tensor([0.0686, 0.0676, 0.1242, 0.0306, 0.0948, 0.1400, 0.0890, 0.2756], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0362, 0.0342, 0.0253, 0.0351, 0.0255, 0.0323, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:16:57,697 INFO [finetune.py:992] (1/2) Epoch 2, batch 6350, loss[loss=0.1599, simple_loss=0.2589, pruned_loss=0.03046, over 12031.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2646, pruned_loss=0.04457, over 2384091.92 frames. ], batch size: 31, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:17:08,886 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0768, 3.9692, 4.0581, 4.5715, 3.0759, 3.8370, 2.6937, 3.8413], device='cuda:1'), covar=tensor([0.1901, 0.0850, 0.0910, 0.0581, 0.1195, 0.0739, 0.1897, 0.1868], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0264, 0.0298, 0.0358, 0.0240, 0.0241, 0.0257, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 19:17:13,583 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 3.006e+02 3.545e+02 4.353e+02 1.068e+03, threshold=7.091e+02, percent-clipped=4.0 2023-05-15 19:17:26,335 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118625.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:17:31,552 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:17:34,133 INFO [finetune.py:992] (1/2) Epoch 2, batch 6400, loss[loss=0.1893, simple_loss=0.2852, pruned_loss=0.04668, over 11441.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2652, pruned_loss=0.04477, over 2377131.70 frames. ], batch size: 55, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:17:41,340 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:18:09,982 INFO [finetune.py:992] (1/2) Epoch 2, batch 6450, loss[loss=0.1739, simple_loss=0.2475, pruned_loss=0.05016, over 11791.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2639, pruned_loss=0.04444, over 2378719.28 frames. ], batch size: 26, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:18:10,067 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118685.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:18:14,952 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118692.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:18:17,514 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118696.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:18:25,825 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.933e+02 3.395e+02 3.970e+02 8.348e+02, threshold=6.791e+02, percent-clipped=2.0 2023-05-15 19:18:45,819 INFO [finetune.py:992] (1/2) Epoch 2, batch 6500, loss[loss=0.1713, simple_loss=0.2677, pruned_loss=0.03744, over 12190.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2647, pruned_loss=0.0449, over 2376386.58 frames. ], batch size: 35, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:18:52,320 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118744.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:19:21,874 INFO [finetune.py:992] (1/2) Epoch 2, batch 6550, loss[loss=0.1658, simple_loss=0.2636, pruned_loss=0.03403, over 12299.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2652, pruned_loss=0.04545, over 2367558.33 frames. ], batch size: 34, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:19:37,719 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 3.114e+02 3.790e+02 4.411e+02 9.201e+02, threshold=7.579e+02, percent-clipped=1.0 2023-05-15 19:19:49,780 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118824.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:19:57,509 INFO [finetune.py:992] (1/2) Epoch 2, batch 6600, loss[loss=0.1824, simple_loss=0.2784, pruned_loss=0.04314, over 12350.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2653, pruned_loss=0.04552, over 2361635.17 frames. ], batch size: 35, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:20:24,659 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118872.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:20:33,870 INFO [finetune.py:992] (1/2) Epoch 2, batch 6650, loss[loss=0.2046, simple_loss=0.297, pruned_loss=0.05613, over 11805.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2656, pruned_loss=0.04569, over 2360301.70 frames. ], batch size: 44, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:20:49,520 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 2.777e+02 3.194e+02 3.859e+02 9.715e+02, threshold=6.388e+02, percent-clipped=2.0 2023-05-15 19:21:03,206 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118925.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:21:08,220 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1168, 2.0476, 2.7354, 3.1179, 2.1120, 3.2337, 3.0611, 3.2826], device='cuda:1'), covar=tensor([0.0145, 0.1017, 0.0385, 0.0141, 0.0951, 0.0279, 0.0288, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0198, 0.0179, 0.0112, 0.0183, 0.0171, 0.0166, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:21:10,091 INFO [finetune.py:992] (1/2) Epoch 2, batch 6700, loss[loss=0.1837, simple_loss=0.2718, pruned_loss=0.0478, over 12372.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2659, pruned_loss=0.04574, over 2360030.00 frames. ], batch size: 36, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:21:17,285 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118945.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:21:37,276 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118973.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:21:45,620 INFO [finetune.py:992] (1/2) Epoch 2, batch 6750, loss[loss=0.1584, simple_loss=0.2466, pruned_loss=0.03512, over 12125.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2657, pruned_loss=0.04557, over 2364335.19 frames. ], batch size: 30, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:21:45,737 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118985.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:21:47,124 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118987.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:21:51,331 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:21:52,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-15 19:21:54,370 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1216, 3.6833, 3.4143, 3.1715, 2.9139, 2.9407, 3.6306, 1.9684], device='cuda:1'), covar=tensor([0.0421, 0.0084, 0.0106, 0.0158, 0.0288, 0.0213, 0.0073, 0.0453], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0156, 0.0144, 0.0174, 0.0195, 0.0188, 0.0154, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:22:02,211 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.416e+02 2.939e+02 3.508e+02 3.989e+02 7.684e+02, threshold=7.016e+02, percent-clipped=3.0 2023-05-15 19:22:20,811 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119033.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:22:22,147 INFO [finetune.py:992] (1/2) Epoch 2, batch 6800, loss[loss=0.1904, simple_loss=0.2911, pruned_loss=0.04481, over 12355.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2665, pruned_loss=0.04592, over 2368661.01 frames. ], batch size: 36, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:22:25,351 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-05-15 19:22:43,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-05-15 19:22:50,981 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119075.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:22:57,922 INFO [finetune.py:992] (1/2) Epoch 2, batch 6850, loss[loss=0.2122, simple_loss=0.299, pruned_loss=0.0627, over 10469.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.265, pruned_loss=0.04524, over 2373749.92 frames. ], batch size: 68, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:23:03,886 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:23:13,457 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.191e+02 3.028e+02 3.541e+02 4.177e+02 1.306e+03, threshold=7.081e+02, percent-clipped=3.0 2023-05-15 19:23:31,857 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4037, 4.2871, 4.2232, 4.6494, 3.3542, 4.0975, 2.9787, 4.2147], device='cuda:1'), covar=tensor([0.1533, 0.0585, 0.0747, 0.0489, 0.0908, 0.0535, 0.1467, 0.1542], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0261, 0.0292, 0.0352, 0.0236, 0.0238, 0.0253, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 19:23:33,001 INFO [finetune.py:992] (1/2) Epoch 2, batch 6900, loss[loss=0.1624, simple_loss=0.2429, pruned_loss=0.04093, over 12283.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2652, pruned_loss=0.04527, over 2366449.14 frames. ], batch size: 28, lr: 4.96e-03, grad_scale: 16.0 2023-05-15 19:23:33,924 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119136.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:23:47,262 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119154.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 19:23:50,894 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6655, 2.7832, 3.3867, 4.5278, 2.5474, 4.7490, 4.5606, 4.7896], device='cuda:1'), covar=tensor([0.0091, 0.0972, 0.0398, 0.0155, 0.1062, 0.0145, 0.0140, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0199, 0.0180, 0.0113, 0.0185, 0.0172, 0.0167, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:24:08,877 INFO [finetune.py:992] (1/2) Epoch 2, batch 6950, loss[loss=0.1852, simple_loss=0.2776, pruned_loss=0.04644, over 10497.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2653, pruned_loss=0.0454, over 2357122.04 frames. ], batch size: 68, lr: 4.95e-03, grad_scale: 16.0 2023-05-15 19:24:25,181 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 2.934e+02 3.415e+02 4.070e+02 9.985e+02, threshold=6.831e+02, percent-clipped=1.0 2023-05-15 19:24:44,690 INFO [finetune.py:992] (1/2) Epoch 2, batch 7000, loss[loss=0.1978, simple_loss=0.279, pruned_loss=0.05829, over 12356.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2642, pruned_loss=0.04488, over 2367208.78 frames. ], batch size: 38, lr: 4.95e-03, grad_scale: 16.0 2023-05-15 19:25:20,155 INFO [finetune.py:992] (1/2) Epoch 2, batch 7050, loss[loss=0.1718, simple_loss=0.2521, pruned_loss=0.0458, over 12003.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2641, pruned_loss=0.0448, over 2365551.12 frames. ], batch size: 28, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:25:21,828 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119287.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:25:37,231 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.828e+02 3.255e+02 4.157e+02 7.416e+02, threshold=6.509e+02, percent-clipped=2.0 2023-05-15 19:25:49,691 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7479, 2.8981, 4.5230, 4.7847, 2.8739, 2.7436, 3.0565, 2.0546], device='cuda:1'), covar=tensor([0.1292, 0.2548, 0.0425, 0.0340, 0.1114, 0.1876, 0.2193, 0.3620], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0363, 0.0261, 0.0284, 0.0247, 0.0274, 0.0345, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:25:55,777 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1332, 6.1008, 5.8736, 5.3404, 5.2255, 6.0341, 5.6049, 5.3668], device='cuda:1'), covar=tensor([0.0582, 0.0791, 0.0512, 0.1293, 0.0496, 0.0618, 0.1377, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0499, 0.0460, 0.0571, 0.0372, 0.0652, 0.0707, 0.0513], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:25:57,116 INFO [finetune.py:992] (1/2) Epoch 2, batch 7100, loss[loss=0.2024, simple_loss=0.2997, pruned_loss=0.05255, over 11799.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2641, pruned_loss=0.04469, over 2363572.96 frames. ], batch size: 44, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:25:57,189 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119335.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:26:32,150 INFO [finetune.py:992] (1/2) Epoch 2, batch 7150, loss[loss=0.1717, simple_loss=0.2617, pruned_loss=0.0408, over 12093.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2649, pruned_loss=0.04528, over 2363374.24 frames. ], batch size: 33, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:26:48,753 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.175e+02 2.928e+02 3.393e+02 4.118e+02 7.925e+02, threshold=6.787e+02, percent-clipped=1.0 2023-05-15 19:27:04,978 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119431.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:27:07,705 INFO [finetune.py:992] (1/2) Epoch 2, batch 7200, loss[loss=0.2063, simple_loss=0.2963, pruned_loss=0.0581, over 12210.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2643, pruned_loss=0.04498, over 2369341.78 frames. ], batch size: 35, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:27:18,432 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119449.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 19:27:20,067 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-05-15 19:27:23,445 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-15 19:27:44,278 INFO [finetune.py:992] (1/2) Epoch 2, batch 7250, loss[loss=0.1497, simple_loss=0.2398, pruned_loss=0.02982, over 11837.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2658, pruned_loss=0.04569, over 2357505.91 frames. ], batch size: 26, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:28:00,674 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.223e+02 2.955e+02 3.322e+02 4.173e+02 9.420e+02, threshold=6.644e+02, percent-clipped=4.0 2023-05-15 19:28:05,088 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119514.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:28:19,663 INFO [finetune.py:992] (1/2) Epoch 2, batch 7300, loss[loss=0.1969, simple_loss=0.2727, pruned_loss=0.06057, over 12119.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2648, pruned_loss=0.04534, over 2361928.26 frames. ], batch size: 33, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:28:28,436 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3959, 3.4565, 3.1415, 3.2287, 2.9077, 2.7102, 3.5505, 2.1664], device='cuda:1'), covar=tensor([0.0319, 0.0141, 0.0141, 0.0132, 0.0294, 0.0299, 0.0104, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0157, 0.0147, 0.0175, 0.0196, 0.0191, 0.0156, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:28:48,199 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119575.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:28:55,085 INFO [finetune.py:992] (1/2) Epoch 2, batch 7350, loss[loss=0.1975, simple_loss=0.2827, pruned_loss=0.05614, over 12061.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.265, pruned_loss=0.04566, over 2360893.42 frames. ], batch size: 42, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:29:12,256 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.761e+02 3.326e+02 4.003e+02 6.449e+02, threshold=6.652e+02, percent-clipped=0.0 2023-05-15 19:29:17,068 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4405, 5.2536, 5.3750, 5.4234, 5.0003, 5.0428, 4.9541, 5.3599], device='cuda:1'), covar=tensor([0.0567, 0.0523, 0.0655, 0.0498, 0.1781, 0.1201, 0.0499, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0626, 0.0532, 0.0588, 0.0763, 0.0704, 0.0512, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 19:29:31,790 INFO [finetune.py:992] (1/2) Epoch 2, batch 7400, loss[loss=0.1982, simple_loss=0.2861, pruned_loss=0.05509, over 12015.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2655, pruned_loss=0.04583, over 2349111.40 frames. ], batch size: 40, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:29:53,438 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119665.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:30:04,087 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6024, 5.0941, 5.6042, 4.9089, 5.1309, 4.9656, 5.6095, 5.1948], device='cuda:1'), covar=tensor([0.0199, 0.0269, 0.0206, 0.0190, 0.0261, 0.0248, 0.0179, 0.0196], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0239, 0.0258, 0.0231, 0.0234, 0.0232, 0.0214, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:30:07,448 INFO [finetune.py:992] (1/2) Epoch 2, batch 7450, loss[loss=0.1763, simple_loss=0.2693, pruned_loss=0.04165, over 12056.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.265, pruned_loss=0.04565, over 2355384.11 frames. ], batch size: 40, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:30:19,084 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1649, 3.8695, 3.8536, 4.2424, 2.8707, 3.7146, 2.4959, 3.9792], device='cuda:1'), covar=tensor([0.1799, 0.0910, 0.1149, 0.0776, 0.1262, 0.0795, 0.2107, 0.1530], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0263, 0.0296, 0.0354, 0.0240, 0.0241, 0.0256, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 19:30:23,544 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.184e+02 3.159e+02 3.777e+02 4.498e+02 9.403e+02, threshold=7.554e+02, percent-clipped=2.0 2023-05-15 19:30:36,255 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119726.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:30:39,687 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119731.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:30:42,495 INFO [finetune.py:992] (1/2) Epoch 2, batch 7500, loss[loss=0.1694, simple_loss=0.2643, pruned_loss=0.03726, over 12107.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2641, pruned_loss=0.04531, over 2364778.10 frames. ], batch size: 42, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:30:53,292 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119749.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 19:31:14,751 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119779.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:31:18,898 INFO [finetune.py:992] (1/2) Epoch 2, batch 7550, loss[loss=0.1508, simple_loss=0.2374, pruned_loss=0.03213, over 12248.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2642, pruned_loss=0.04551, over 2371646.32 frames. ], batch size: 32, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:31:20,686 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5802, 4.0634, 4.0515, 4.4787, 3.0639, 3.9249, 2.6058, 4.0864], device='cuda:1'), covar=tensor([0.1542, 0.0730, 0.0893, 0.0684, 0.1122, 0.0646, 0.1823, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0263, 0.0295, 0.0354, 0.0239, 0.0240, 0.0255, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 19:31:24,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7594, 2.6407, 4.0612, 4.3525, 2.8391, 2.7302, 2.7705, 2.0650], device='cuda:1'), covar=tensor([0.1278, 0.2571, 0.0564, 0.0387, 0.1120, 0.1870, 0.2432, 0.3597], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0366, 0.0263, 0.0287, 0.0249, 0.0275, 0.0348, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:31:27,597 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119797.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:31:35,739 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 3.021e+02 3.531e+02 4.316e+02 1.223e+03, threshold=7.062e+02, percent-clipped=7.0 2023-05-15 19:31:40,851 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8636, 3.2134, 5.0972, 2.6364, 2.7853, 3.9177, 3.3775, 3.7862], device='cuda:1'), covar=tensor([0.0353, 0.1145, 0.0320, 0.1111, 0.1855, 0.1195, 0.1222, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0225, 0.0231, 0.0177, 0.0232, 0.0275, 0.0221, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:31:54,868 INFO [finetune.py:992] (1/2) Epoch 2, batch 7600, loss[loss=0.2004, simple_loss=0.2876, pruned_loss=0.05655, over 11642.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2636, pruned_loss=0.04505, over 2376024.67 frames. ], batch size: 48, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:32:06,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0 2023-05-15 19:32:19,735 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119870.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:32:30,986 INFO [finetune.py:992] (1/2) Epoch 2, batch 7650, loss[loss=0.1917, simple_loss=0.2812, pruned_loss=0.05109, over 11806.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2646, pruned_loss=0.04548, over 2367629.10 frames. ], batch size: 44, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:32:35,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-15 19:32:47,879 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.981e+02 2.810e+02 3.505e+02 4.378e+02 1.736e+03, threshold=7.010e+02, percent-clipped=5.0 2023-05-15 19:32:59,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6656, 2.6127, 4.3667, 4.6166, 2.9964, 2.5054, 2.8316, 2.0502], device='cuda:1'), covar=tensor([0.1289, 0.2764, 0.0434, 0.0349, 0.1018, 0.1947, 0.2293, 0.3429], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0366, 0.0264, 0.0287, 0.0249, 0.0275, 0.0348, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:33:07,244 INFO [finetune.py:992] (1/2) Epoch 2, batch 7700, loss[loss=0.1554, simple_loss=0.2413, pruned_loss=0.03472, over 12024.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2644, pruned_loss=0.04502, over 2372093.31 frames. ], batch size: 31, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:33:16,783 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3258, 4.5570, 4.0743, 5.0002, 4.6115, 3.0660, 4.3709, 3.0832], device='cuda:1'), covar=tensor([0.0678, 0.0843, 0.1187, 0.0296, 0.0963, 0.1396, 0.0856, 0.2890], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0368, 0.0343, 0.0254, 0.0354, 0.0256, 0.0324, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:33:42,684 INFO [finetune.py:992] (1/2) Epoch 2, batch 7750, loss[loss=0.1713, simple_loss=0.2688, pruned_loss=0.03691, over 12300.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2638, pruned_loss=0.04471, over 2377484.81 frames. ], batch size: 34, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:34:02,462 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.817e+02 3.479e+02 4.094e+02 1.236e+03, threshold=6.958e+02, percent-clipped=2.0 2023-05-15 19:34:11,852 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120021.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:34:15,010 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-05-15 19:34:22,380 INFO [finetune.py:992] (1/2) Epoch 2, batch 7800, loss[loss=0.1521, simple_loss=0.2302, pruned_loss=0.03697, over 12343.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2635, pruned_loss=0.04465, over 2378244.72 frames. ], batch size: 30, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:34:58,424 INFO [finetune.py:992] (1/2) Epoch 2, batch 7850, loss[loss=0.1912, simple_loss=0.2759, pruned_loss=0.05326, over 11185.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.264, pruned_loss=0.04522, over 2365213.08 frames. ], batch size: 55, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:35:14,934 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.826e+02 3.357e+02 4.136e+02 8.252e+02, threshold=6.714e+02, percent-clipped=1.0 2023-05-15 19:35:33,895 INFO [finetune.py:992] (1/2) Epoch 2, batch 7900, loss[loss=0.1877, simple_loss=0.275, pruned_loss=0.05017, over 11808.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2648, pruned_loss=0.04561, over 2368656.95 frames. ], batch size: 44, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:35:58,920 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120170.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:36:10,285 INFO [finetune.py:992] (1/2) Epoch 2, batch 7950, loss[loss=0.1858, simple_loss=0.2731, pruned_loss=0.04923, over 12100.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2648, pruned_loss=0.04507, over 2371823.77 frames. ], batch size: 38, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:36:27,561 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.839e+02 3.382e+02 4.302e+02 1.919e+03, threshold=6.765e+02, percent-clipped=5.0 2023-05-15 19:36:33,826 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-15 19:36:34,935 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=120218.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:36:46,915 INFO [finetune.py:992] (1/2) Epoch 2, batch 8000, loss[loss=0.1805, simple_loss=0.2687, pruned_loss=0.04616, over 12209.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2646, pruned_loss=0.04486, over 2373501.10 frames. ], batch size: 35, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:37:15,617 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8351, 3.4524, 5.0795, 2.5707, 2.8026, 3.7600, 3.1981, 3.9109], device='cuda:1'), covar=tensor([0.0397, 0.1033, 0.0389, 0.1192, 0.1920, 0.1452, 0.1333, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0226, 0.0232, 0.0178, 0.0233, 0.0277, 0.0221, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:37:22,464 INFO [finetune.py:992] (1/2) Epoch 2, batch 8050, loss[loss=0.1973, simple_loss=0.2857, pruned_loss=0.05441, over 12151.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2645, pruned_loss=0.04491, over 2372231.15 frames. ], batch size: 36, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:37:38,853 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.033e+02 2.935e+02 3.464e+02 4.476e+02 1.396e+03, threshold=6.927e+02, percent-clipped=8.0 2023-05-15 19:37:43,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-15 19:37:48,300 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120321.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:37:59,593 INFO [finetune.py:992] (1/2) Epoch 2, batch 8100, loss[loss=0.1623, simple_loss=0.2478, pruned_loss=0.03843, over 12414.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2644, pruned_loss=0.04545, over 2366667.69 frames. ], batch size: 32, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:38:06,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.62 vs. limit=5.0 2023-05-15 19:38:23,977 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=120369.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:38:35,478 INFO [finetune.py:992] (1/2) Epoch 2, batch 8150, loss[loss=0.1717, simple_loss=0.2673, pruned_loss=0.03799, over 12351.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2628, pruned_loss=0.04466, over 2377069.21 frames. ], batch size: 35, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:38:36,430 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6894, 3.2590, 5.0644, 2.6369, 2.8367, 3.8657, 3.1689, 4.0565], device='cuda:1'), covar=tensor([0.0485, 0.1125, 0.0245, 0.1148, 0.1843, 0.1357, 0.1293, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0225, 0.0230, 0.0177, 0.0232, 0.0273, 0.0219, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:38:51,828 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.925e+02 3.403e+02 4.122e+02 5.583e+02, threshold=6.806e+02, percent-clipped=0.0 2023-05-15 19:39:10,922 INFO [finetune.py:992] (1/2) Epoch 2, batch 8200, loss[loss=0.1827, simple_loss=0.2683, pruned_loss=0.04854, over 12092.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2644, pruned_loss=0.04501, over 2380244.18 frames. ], batch size: 32, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:39:17,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-15 19:39:47,519 INFO [finetune.py:992] (1/2) Epoch 2, batch 8250, loss[loss=0.1753, simple_loss=0.2647, pruned_loss=0.04296, over 12092.00 frames. ], tot_loss[loss=0.178, simple_loss=0.265, pruned_loss=0.04552, over 2372865.38 frames. ], batch size: 32, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:40:03,628 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.102e+02 3.007e+02 3.486e+02 4.183e+02 6.928e+02, threshold=6.971e+02, percent-clipped=1.0 2023-05-15 19:40:22,810 INFO [finetune.py:992] (1/2) Epoch 2, batch 8300, loss[loss=0.1845, simple_loss=0.2648, pruned_loss=0.05212, over 12124.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2655, pruned_loss=0.046, over 2369982.86 frames. ], batch size: 33, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:40:51,776 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0343, 2.4591, 3.7879, 3.0453, 3.6237, 3.1479, 2.4733, 3.6865], device='cuda:1'), covar=tensor([0.0110, 0.0328, 0.0099, 0.0219, 0.0118, 0.0152, 0.0320, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0191, 0.0168, 0.0172, 0.0191, 0.0147, 0.0181, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:40:57,807 INFO [finetune.py:992] (1/2) Epoch 2, batch 8350, loss[loss=0.1826, simple_loss=0.269, pruned_loss=0.04808, over 12300.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2666, pruned_loss=0.04657, over 2364704.03 frames. ], batch size: 34, lr: 4.95e-03, grad_scale: 4.0 2023-05-15 19:41:09,645 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1859, 2.4443, 3.6772, 3.0613, 3.4506, 3.1747, 2.6086, 3.6169], device='cuda:1'), covar=tensor([0.0088, 0.0282, 0.0106, 0.0190, 0.0131, 0.0142, 0.0257, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0189, 0.0166, 0.0171, 0.0190, 0.0146, 0.0180, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:41:15,160 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.128e+02 3.117e+02 3.632e+02 4.219e+02 7.197e+02, threshold=7.264e+02, percent-clipped=1.0 2023-05-15 19:41:34,977 INFO [finetune.py:992] (1/2) Epoch 2, batch 8400, loss[loss=0.1686, simple_loss=0.2669, pruned_loss=0.03518, over 12289.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2656, pruned_loss=0.04593, over 2373026.79 frames. ], batch size: 34, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:42:04,222 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120676.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:42:10,697 INFO [finetune.py:992] (1/2) Epoch 2, batch 8450, loss[loss=0.1727, simple_loss=0.2702, pruned_loss=0.03757, over 12349.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2653, pruned_loss=0.04591, over 2373861.35 frames. ], batch size: 35, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:42:20,949 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1338, 3.8782, 5.4330, 2.5628, 3.1699, 4.0460, 3.5961, 4.1808], device='cuda:1'), covar=tensor([0.0284, 0.0791, 0.0190, 0.1219, 0.1624, 0.1219, 0.1058, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0223, 0.0229, 0.0176, 0.0231, 0.0272, 0.0218, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:42:27,721 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.008e+02 2.780e+02 3.356e+02 3.939e+02 8.101e+02, threshold=6.712e+02, percent-clipped=2.0 2023-05-15 19:42:45,163 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2162, 4.3471, 3.9535, 4.8228, 4.3546, 2.6643, 4.0640, 3.1299], device='cuda:1'), covar=tensor([0.0761, 0.0862, 0.1258, 0.0374, 0.1144, 0.1567, 0.0926, 0.2720], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0367, 0.0345, 0.0256, 0.0353, 0.0257, 0.0328, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:42:46,282 INFO [finetune.py:992] (1/2) Epoch 2, batch 8500, loss[loss=0.21, simple_loss=0.3015, pruned_loss=0.05928, over 10525.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2671, pruned_loss=0.04692, over 2354161.14 frames. ], batch size: 68, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:42:47,978 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120737.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:43:10,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 19:43:23,291 INFO [finetune.py:992] (1/2) Epoch 2, batch 8550, loss[loss=0.1828, simple_loss=0.2785, pruned_loss=0.04356, over 11647.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.266, pruned_loss=0.0462, over 2367479.67 frames. ], batch size: 48, lr: 4.95e-03, grad_scale: 4.0 2023-05-15 19:43:31,370 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1610, 2.1771, 2.8815, 3.1317, 2.9428, 3.1282, 2.7747, 2.4811], device='cuda:1'), covar=tensor([0.0048, 0.0299, 0.0129, 0.0060, 0.0122, 0.0091, 0.0113, 0.0284], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0117, 0.0099, 0.0074, 0.0099, 0.0108, 0.0085, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 19:43:40,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-05-15 19:43:41,268 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.862e+02 3.409e+02 3.902e+02 6.283e+02, threshold=6.819e+02, percent-clipped=0.0 2023-05-15 19:43:59,042 INFO [finetune.py:992] (1/2) Epoch 2, batch 8600, loss[loss=0.1521, simple_loss=0.2334, pruned_loss=0.03544, over 12177.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2651, pruned_loss=0.04587, over 2363964.45 frames. ], batch size: 29, lr: 4.95e-03, grad_scale: 4.0 2023-05-15 19:44:02,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1475, 4.9064, 5.0160, 5.0969, 4.8061, 5.0217, 4.9601, 2.8178], device='cuda:1'), covar=tensor([0.0083, 0.0050, 0.0061, 0.0050, 0.0049, 0.0073, 0.0069, 0.0609], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0071, 0.0074, 0.0069, 0.0056, 0.0085, 0.0074, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:44:12,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3755, 3.8702, 3.4537, 3.4553, 3.0321, 2.9820, 3.7840, 2.1943], device='cuda:1'), covar=tensor([0.0399, 0.0089, 0.0111, 0.0126, 0.0257, 0.0263, 0.0081, 0.0428], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0157, 0.0148, 0.0176, 0.0200, 0.0193, 0.0157, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:44:34,720 INFO [finetune.py:992] (1/2) Epoch 2, batch 8650, loss[loss=0.1632, simple_loss=0.2605, pruned_loss=0.0329, over 12345.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2653, pruned_loss=0.04598, over 2354809.83 frames. ], batch size: 36, lr: 4.95e-03, grad_scale: 4.0 2023-05-15 19:44:53,804 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.163e+02 2.876e+02 3.374e+02 4.073e+02 7.321e+02, threshold=6.747e+02, percent-clipped=1.0 2023-05-15 19:45:08,212 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0425, 4.9709, 4.8288, 4.8934, 4.4869, 4.8989, 4.9784, 5.1534], device='cuda:1'), covar=tensor([0.0173, 0.0116, 0.0156, 0.0251, 0.0703, 0.0296, 0.0132, 0.0149], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0179, 0.0180, 0.0224, 0.0228, 0.0198, 0.0166, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 19:45:11,517 INFO [finetune.py:992] (1/2) Epoch 2, batch 8700, loss[loss=0.1781, simple_loss=0.2656, pruned_loss=0.04534, over 12390.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.266, pruned_loss=0.04637, over 2356376.99 frames. ], batch size: 38, lr: 4.95e-03, grad_scale: 4.0 2023-05-15 19:45:46,658 INFO [finetune.py:992] (1/2) Epoch 2, batch 8750, loss[loss=0.1986, simple_loss=0.291, pruned_loss=0.05311, over 11647.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2662, pruned_loss=0.04653, over 2359323.62 frames. ], batch size: 48, lr: 4.95e-03, grad_scale: 4.0 2023-05-15 19:45:54,923 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1090, 5.0539, 4.9298, 5.0812, 4.2915, 5.0192, 5.0689, 5.2442], device='cuda:1'), covar=tensor([0.0202, 0.0136, 0.0176, 0.0253, 0.1003, 0.0322, 0.0153, 0.0167], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0180, 0.0181, 0.0225, 0.0228, 0.0199, 0.0167, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 19:45:59,502 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121003.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:46:04,182 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.962e+02 3.419e+02 4.419e+02 8.057e+02, threshold=6.838e+02, percent-clipped=2.0 2023-05-15 19:46:11,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-05-15 19:46:18,285 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9903, 5.8248, 5.4125, 5.3624, 5.9146, 5.1995, 5.5295, 5.4978], device='cuda:1'), covar=tensor([0.1378, 0.0915, 0.0913, 0.2001, 0.0943, 0.2025, 0.1441, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0452, 0.0360, 0.0414, 0.0433, 0.0408, 0.0372, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:46:19,667 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121032.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:46:21,778 INFO [finetune.py:992] (1/2) Epoch 2, batch 8800, loss[loss=0.2427, simple_loss=0.3119, pruned_loss=0.08675, over 7943.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2657, pruned_loss=0.04639, over 2360772.30 frames. ], batch size: 97, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:46:43,724 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121064.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:46:47,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-15 19:46:58,479 INFO [finetune.py:992] (1/2) Epoch 2, batch 8850, loss[loss=0.1808, simple_loss=0.2588, pruned_loss=0.05138, over 12111.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2661, pruned_loss=0.04692, over 2352536.48 frames. ], batch size: 33, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:47:16,533 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 3.010e+02 3.613e+02 4.361e+02 8.894e+02, threshold=7.226e+02, percent-clipped=1.0 2023-05-15 19:47:34,450 INFO [finetune.py:992] (1/2) Epoch 2, batch 8900, loss[loss=0.1746, simple_loss=0.2634, pruned_loss=0.04294, over 12037.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2656, pruned_loss=0.04671, over 2348613.18 frames. ], batch size: 31, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:47:41,001 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6258, 2.7050, 4.4952, 4.7686, 3.3809, 2.6590, 2.8848, 1.9339], device='cuda:1'), covar=tensor([0.1265, 0.2871, 0.0405, 0.0304, 0.0830, 0.1798, 0.2351, 0.3680], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0363, 0.0264, 0.0285, 0.0247, 0.0272, 0.0343, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:47:55,167 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121164.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:48:10,523 INFO [finetune.py:992] (1/2) Epoch 2, batch 8950, loss[loss=0.1725, simple_loss=0.2656, pruned_loss=0.03972, over 11706.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2645, pruned_loss=0.04606, over 2359323.40 frames. ], batch size: 48, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:48:29,066 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.756e+02 3.209e+02 4.008e+02 1.639e+03, threshold=6.418e+02, percent-clipped=3.0 2023-05-15 19:48:29,249 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4439, 5.0023, 5.4543, 4.7401, 5.1073, 4.8055, 5.4618, 5.0985], device='cuda:1'), covar=tensor([0.0226, 0.0286, 0.0211, 0.0207, 0.0271, 0.0270, 0.0161, 0.0257], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0233, 0.0252, 0.0227, 0.0230, 0.0228, 0.0209, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:48:39,893 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121225.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:48:46,848 INFO [finetune.py:992] (1/2) Epoch 2, batch 9000, loss[loss=0.1568, simple_loss=0.2486, pruned_loss=0.03247, over 12142.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.264, pruned_loss=0.04528, over 2366965.99 frames. ], batch size: 34, lr: 4.95e-03, grad_scale: 8.0 2023-05-15 19:48:46,848 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 19:48:57,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9553, 4.8042, 4.9807, 5.0154, 4.4565, 4.6386, 4.5087, 4.9141], device='cuda:1'), covar=tensor([0.0834, 0.0641, 0.0703, 0.0561, 0.2141, 0.1306, 0.0644, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0623, 0.0534, 0.0581, 0.0769, 0.0706, 0.0514, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 19:48:58,373 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7183, 5.1972, 4.8831, 4.9203, 5.3122, 4.5498, 4.7160, 4.8813], device='cuda:1'), covar=tensor([0.1270, 0.1132, 0.1038, 0.1627, 0.0887, 0.2166, 0.2031, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0454, 0.0360, 0.0417, 0.0434, 0.0408, 0.0374, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 19:49:05,158 INFO [finetune.py:1026] (1/2) Epoch 2, validation: loss=0.3439, simple_loss=0.4128, pruned_loss=0.1375, over 1020973.00 frames. 2023-05-15 19:49:05,159 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 19:49:40,797 INFO [finetune.py:992] (1/2) Epoch 2, batch 9050, loss[loss=0.1645, simple_loss=0.2416, pruned_loss=0.04367, over 11980.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2641, pruned_loss=0.04547, over 2369267.85 frames. ], batch size: 28, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:49:58,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 2.921e+02 3.503e+02 4.414e+02 9.528e+02, threshold=7.007e+02, percent-clipped=6.0 2023-05-15 19:49:59,796 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121311.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 19:50:14,610 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121332.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:50:16,684 INFO [finetune.py:992] (1/2) Epoch 2, batch 9100, loss[loss=0.204, simple_loss=0.2882, pruned_loss=0.05994, over 12095.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2656, pruned_loss=0.04622, over 2361117.50 frames. ], batch size: 38, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 19:50:33,426 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121359.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:50:42,851 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121372.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 19:50:48,314 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121380.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:50:51,933 INFO [finetune.py:992] (1/2) Epoch 2, batch 9150, loss[loss=0.1518, simple_loss=0.2351, pruned_loss=0.03425, over 12139.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.266, pruned_loss=0.04623, over 2369268.59 frames. ], batch size: 30, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 19:50:52,148 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5496, 2.3763, 3.8367, 4.5585, 3.9977, 4.4920, 3.9683, 3.2109], device='cuda:1'), covar=tensor([0.0025, 0.0362, 0.0088, 0.0030, 0.0088, 0.0064, 0.0081, 0.0286], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0116, 0.0097, 0.0072, 0.0097, 0.0107, 0.0083, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 19:51:07,992 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121407.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:51:10,594 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 3.099e+02 3.650e+02 4.725e+02 1.105e+03, threshold=7.299e+02, percent-clipped=6.0 2023-05-15 19:51:27,883 INFO [finetune.py:992] (1/2) Epoch 2, batch 9200, loss[loss=0.1829, simple_loss=0.2671, pruned_loss=0.04937, over 12292.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2664, pruned_loss=0.04652, over 2365665.69 frames. ], batch size: 33, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:51:51,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-15 19:51:52,005 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121468.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:52:03,945 INFO [finetune.py:992] (1/2) Epoch 2, batch 9250, loss[loss=0.1881, simple_loss=0.2823, pruned_loss=0.04695, over 11187.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.2672, pruned_loss=0.04664, over 2359947.90 frames. ], batch size: 55, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:52:22,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.873e+02 2.914e+02 3.402e+02 4.168e+02 8.635e+02, threshold=6.805e+02, percent-clipped=2.0 2023-05-15 19:52:28,796 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121520.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:52:37,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 19:52:39,441 INFO [finetune.py:992] (1/2) Epoch 2, batch 9300, loss[loss=0.136, simple_loss=0.2307, pruned_loss=0.02067, over 12359.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2664, pruned_loss=0.04587, over 2367903.33 frames. ], batch size: 30, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:53:15,969 INFO [finetune.py:992] (1/2) Epoch 2, batch 9350, loss[loss=0.182, simple_loss=0.2759, pruned_loss=0.04405, over 10433.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2661, pruned_loss=0.04566, over 2368952.35 frames. ], batch size: 68, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:53:27,599 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9422, 4.1927, 3.7760, 4.5851, 4.1724, 2.7822, 3.9169, 2.9073], device='cuda:1'), covar=tensor([0.0804, 0.0924, 0.1414, 0.0354, 0.1205, 0.1482, 0.0911, 0.2953], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0365, 0.0343, 0.0254, 0.0352, 0.0255, 0.0327, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:53:29,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 19:53:34,340 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.845e+02 3.346e+02 4.203e+02 7.244e+02, threshold=6.692e+02, percent-clipped=1.0 2023-05-15 19:53:51,432 INFO [finetune.py:992] (1/2) Epoch 2, batch 9400, loss[loss=0.1589, simple_loss=0.2509, pruned_loss=0.03347, over 12364.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2645, pruned_loss=0.04471, over 2379992.89 frames. ], batch size: 35, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:54:08,423 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121659.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:54:13,993 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121667.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 19:54:16,583 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-15 19:54:17,061 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7865, 3.5012, 5.0824, 2.6237, 2.7227, 3.9667, 3.2578, 4.0718], device='cuda:1'), covar=tensor([0.0423, 0.0965, 0.0236, 0.1114, 0.1821, 0.1016, 0.1249, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0226, 0.0231, 0.0178, 0.0232, 0.0275, 0.0221, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:54:21,153 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4725, 4.9967, 5.4834, 4.7952, 5.0975, 4.8279, 5.4946, 5.0646], device='cuda:1'), covar=tensor([0.0245, 0.0339, 0.0242, 0.0208, 0.0284, 0.0269, 0.0178, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0236, 0.0254, 0.0230, 0.0233, 0.0231, 0.0211, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 19:54:26,753 INFO [finetune.py:992] (1/2) Epoch 2, batch 9450, loss[loss=0.1475, simple_loss=0.2343, pruned_loss=0.03034, over 11780.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2631, pruned_loss=0.04409, over 2385215.93 frames. ], batch size: 26, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:54:42,343 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121707.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:54:45,053 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 2.877e+02 3.491e+02 4.071e+02 9.031e+02, threshold=6.982e+02, percent-clipped=1.0 2023-05-15 19:55:03,144 INFO [finetune.py:992] (1/2) Epoch 2, batch 9500, loss[loss=0.2141, simple_loss=0.2985, pruned_loss=0.06485, over 12366.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2637, pruned_loss=0.04427, over 2382897.38 frames. ], batch size: 35, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:55:23,184 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121763.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:55:38,600 INFO [finetune.py:992] (1/2) Epoch 2, batch 9550, loss[loss=0.1475, simple_loss=0.2307, pruned_loss=0.03213, over 12003.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.264, pruned_loss=0.04408, over 2386359.69 frames. ], batch size: 28, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:55:55,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-15 19:55:57,236 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.787e+02 3.335e+02 3.979e+02 8.470e+02, threshold=6.670e+02, percent-clipped=2.0 2023-05-15 19:56:03,765 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121820.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:56:14,068 INFO [finetune.py:992] (1/2) Epoch 2, batch 9600, loss[loss=0.1866, simple_loss=0.2762, pruned_loss=0.0485, over 12015.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2638, pruned_loss=0.04397, over 2377316.69 frames. ], batch size: 40, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:56:24,361 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3425, 4.5081, 4.2182, 5.0315, 4.6326, 2.8756, 4.3782, 3.0905], device='cuda:1'), covar=tensor([0.0669, 0.0769, 0.1188, 0.0285, 0.0858, 0.1426, 0.0894, 0.2836], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0362, 0.0343, 0.0253, 0.0351, 0.0254, 0.0327, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:56:37,782 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121868.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:56:51,191 INFO [finetune.py:992] (1/2) Epoch 2, batch 9650, loss[loss=0.1505, simple_loss=0.2313, pruned_loss=0.03484, over 12335.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2618, pruned_loss=0.04344, over 2383288.09 frames. ], batch size: 31, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:57:09,351 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.125e+02 3.093e+02 3.583e+02 4.110e+02 1.152e+03, threshold=7.167e+02, percent-clipped=4.0 2023-05-15 19:57:11,688 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7231, 2.9963, 4.6074, 4.6913, 3.0575, 2.7033, 2.9315, 2.0436], device='cuda:1'), covar=tensor([0.1275, 0.2689, 0.0389, 0.0383, 0.1021, 0.1802, 0.2313, 0.3581], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0368, 0.0267, 0.0289, 0.0250, 0.0276, 0.0348, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:57:26,314 INFO [finetune.py:992] (1/2) Epoch 2, batch 9700, loss[loss=0.1519, simple_loss=0.2349, pruned_loss=0.03445, over 12330.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2618, pruned_loss=0.04402, over 2381774.69 frames. ], batch size: 30, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:57:37,160 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121950.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:57:39,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-15 19:57:45,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 2023-05-15 19:57:49,146 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121967.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 19:58:01,736 INFO [finetune.py:992] (1/2) Epoch 2, batch 9750, loss[loss=0.1846, simple_loss=0.2721, pruned_loss=0.04855, over 12141.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2617, pruned_loss=0.0437, over 2386719.99 frames. ], batch size: 34, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:58:23,194 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.816e+02 3.449e+02 4.237e+02 2.383e+03, threshold=6.898e+02, percent-clipped=6.0 2023-05-15 19:58:23,420 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122011.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:58:26,901 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122015.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 19:58:37,412 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2350, 4.5576, 3.9104, 4.9433, 4.4523, 2.7256, 4.2007, 3.1093], device='cuda:1'), covar=tensor([0.0816, 0.0745, 0.1379, 0.0326, 0.0978, 0.1541, 0.0920, 0.2865], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0363, 0.0344, 0.0255, 0.0354, 0.0256, 0.0328, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:58:41,409 INFO [finetune.py:992] (1/2) Epoch 2, batch 9800, loss[loss=0.2506, simple_loss=0.3299, pruned_loss=0.0857, over 10615.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2633, pruned_loss=0.04437, over 2374045.94 frames. ], batch size: 68, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:58:45,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-15 19:59:01,456 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122063.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:59:16,905 INFO [finetune.py:992] (1/2) Epoch 2, batch 9850, loss[loss=0.1614, simple_loss=0.2418, pruned_loss=0.04048, over 12116.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2635, pruned_loss=0.04499, over 2363556.52 frames. ], batch size: 30, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 19:59:19,958 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4504, 2.5554, 3.2673, 4.4206, 2.2092, 4.4546, 4.3285, 4.6191], device='cuda:1'), covar=tensor([0.0105, 0.1094, 0.0358, 0.0106, 0.1151, 0.0164, 0.0145, 0.0065], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0198, 0.0183, 0.0112, 0.0181, 0.0170, 0.0167, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 19:59:35,149 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.944e+02 3.381e+02 4.045e+02 6.935e+02, threshold=6.762e+02, percent-clipped=1.0 2023-05-15 19:59:35,239 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122111.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:59:39,750 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122117.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 19:59:52,108 INFO [finetune.py:992] (1/2) Epoch 2, batch 9900, loss[loss=0.1531, simple_loss=0.2416, pruned_loss=0.03236, over 12135.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2636, pruned_loss=0.045, over 2353541.96 frames. ], batch size: 30, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:00:24,021 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122178.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 20:00:28,802 INFO [finetune.py:992] (1/2) Epoch 2, batch 9950, loss[loss=0.1854, simple_loss=0.2782, pruned_loss=0.04633, over 11732.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2633, pruned_loss=0.04471, over 2360671.34 frames. ], batch size: 44, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:00:35,832 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122195.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:00:47,440 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.921e+02 3.481e+02 4.108e+02 1.157e+03, threshold=6.963e+02, percent-clipped=3.0 2023-05-15 20:01:02,019 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6427, 4.3329, 4.2499, 4.7654, 3.3033, 4.1040, 2.8393, 4.3695], device='cuda:1'), covar=tensor([0.1357, 0.0600, 0.0915, 0.0544, 0.1012, 0.0535, 0.1637, 0.1437], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0262, 0.0297, 0.0353, 0.0239, 0.0239, 0.0256, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:01:04,629 INFO [finetune.py:992] (1/2) Epoch 2, batch 10000, loss[loss=0.1596, simple_loss=0.2406, pruned_loss=0.03934, over 12323.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2632, pruned_loss=0.04496, over 2358243.78 frames. ], batch size: 30, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:01:19,701 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122256.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:01:21,850 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7872, 1.9688, 3.5362, 2.8028, 3.3015, 2.8818, 2.0101, 3.3752], device='cuda:1'), covar=tensor([0.0144, 0.0403, 0.0115, 0.0239, 0.0126, 0.0192, 0.0365, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0189, 0.0166, 0.0171, 0.0190, 0.0147, 0.0180, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:01:40,204 INFO [finetune.py:992] (1/2) Epoch 2, batch 10050, loss[loss=0.165, simple_loss=0.259, pruned_loss=0.03547, over 12185.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2632, pruned_loss=0.04457, over 2364014.81 frames. ], batch size: 35, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:01:50,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-15 20:01:56,593 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122306.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:01:59,984 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.902e+02 3.287e+02 3.882e+02 8.143e+02, threshold=6.574e+02, percent-clipped=1.0 2023-05-15 20:02:05,999 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122319.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:02:10,142 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3900, 2.3214, 3.2121, 4.3512, 2.1208, 4.4575, 4.3404, 4.5397], device='cuda:1'), covar=tensor([0.0124, 0.1305, 0.0410, 0.0163, 0.1314, 0.0190, 0.0167, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0198, 0.0182, 0.0112, 0.0180, 0.0170, 0.0165, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:02:10,184 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1049, 3.9235, 4.1903, 4.5373, 3.2071, 3.8514, 2.6523, 4.0686], device='cuda:1'), covar=tensor([0.1664, 0.0803, 0.0754, 0.0589, 0.1024, 0.0659, 0.1825, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0261, 0.0295, 0.0353, 0.0237, 0.0238, 0.0255, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:02:17,044 INFO [finetune.py:992] (1/2) Epoch 2, batch 10100, loss[loss=0.1684, simple_loss=0.2476, pruned_loss=0.0446, over 12196.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2636, pruned_loss=0.04455, over 2366569.10 frames. ], batch size: 31, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:02:38,871 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5326, 5.1771, 4.7310, 4.8281, 5.3019, 4.6503, 4.7744, 4.7684], device='cuda:1'), covar=tensor([0.1329, 0.0991, 0.1079, 0.1747, 0.0970, 0.2005, 0.1924, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0451, 0.0357, 0.0410, 0.0432, 0.0406, 0.0372, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:02:48,837 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122380.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:02:52,156 INFO [finetune.py:992] (1/2) Epoch 2, batch 10150, loss[loss=0.1762, simple_loss=0.2705, pruned_loss=0.04098, over 12349.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2641, pruned_loss=0.04467, over 2363971.95 frames. ], batch size: 36, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:03:11,564 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.036e+02 2.956e+02 3.518e+02 4.206e+02 6.663e+02, threshold=7.036e+02, percent-clipped=1.0 2023-05-15 20:03:16,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-15 20:03:21,240 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2004, 2.1043, 2.7392, 3.2248, 2.1168, 3.3112, 3.2148, 3.3041], device='cuda:1'), covar=tensor([0.0129, 0.0949, 0.0377, 0.0129, 0.0929, 0.0212, 0.0236, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0185, 0.0113, 0.0182, 0.0172, 0.0167, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:03:28,121 INFO [finetune.py:992] (1/2) Epoch 2, batch 10200, loss[loss=0.1528, simple_loss=0.2415, pruned_loss=0.03204, over 12091.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2646, pruned_loss=0.04496, over 2359652.23 frames. ], batch size: 32, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:03:28,309 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8942, 4.6444, 4.8162, 4.7712, 4.5732, 4.8147, 4.7101, 2.8174], device='cuda:1'), covar=tensor([0.0122, 0.0054, 0.0077, 0.0068, 0.0051, 0.0091, 0.0083, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0077, 0.0071, 0.0058, 0.0088, 0.0076, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:03:52,396 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122468.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:03:55,745 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122473.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 20:04:04,155 INFO [finetune.py:992] (1/2) Epoch 2, batch 10250, loss[loss=0.189, simple_loss=0.281, pruned_loss=0.04851, over 12378.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2641, pruned_loss=0.04466, over 2366370.67 frames. ], batch size: 38, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:04:23,103 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.148e+02 2.763e+02 3.392e+02 4.054e+02 1.109e+03, threshold=6.785e+02, percent-clipped=4.0 2023-05-15 20:04:28,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9139, 4.5207, 4.8630, 4.3303, 4.5692, 4.3408, 4.8800, 4.5829], device='cuda:1'), covar=tensor([0.0241, 0.0298, 0.0246, 0.0209, 0.0290, 0.0299, 0.0234, 0.0410], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0236, 0.0252, 0.0228, 0.0232, 0.0229, 0.0210, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:04:35,496 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122529.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:04:37,016 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8303, 3.8636, 3.5408, 3.3932, 3.1916, 2.9948, 3.8244, 2.5408], device='cuda:1'), covar=tensor([0.0269, 0.0095, 0.0132, 0.0148, 0.0281, 0.0267, 0.0081, 0.0361], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0155, 0.0146, 0.0177, 0.0197, 0.0190, 0.0157, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:04:39,628 INFO [finetune.py:992] (1/2) Epoch 2, batch 10300, loss[loss=0.1811, simple_loss=0.2699, pruned_loss=0.04622, over 12304.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.264, pruned_loss=0.04471, over 2362302.66 frames. ], batch size: 34, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:04:51,341 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122551.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:05:03,401 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8728, 4.5977, 4.7643, 4.7733, 4.6646, 4.8004, 4.6216, 2.8631], device='cuda:1'), covar=tensor([0.0095, 0.0055, 0.0065, 0.0058, 0.0040, 0.0080, 0.0069, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0072, 0.0076, 0.0071, 0.0057, 0.0087, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:05:16,328 INFO [finetune.py:992] (1/2) Epoch 2, batch 10350, loss[loss=0.1359, simple_loss=0.2199, pruned_loss=0.02594, over 12278.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2624, pruned_loss=0.0439, over 2366726.18 frames. ], batch size: 28, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:05:32,330 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122606.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:05:36,469 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.820e+02 3.400e+02 4.097e+02 7.939e+02, threshold=6.800e+02, percent-clipped=1.0 2023-05-15 20:05:49,885 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4521, 3.6174, 3.2800, 3.1118, 2.9326, 2.6843, 3.5289, 2.2765], device='cuda:1'), covar=tensor([0.0340, 0.0092, 0.0121, 0.0160, 0.0317, 0.0310, 0.0092, 0.0409], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0157, 0.0148, 0.0178, 0.0199, 0.0192, 0.0158, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:05:52,632 INFO [finetune.py:992] (1/2) Epoch 2, batch 10400, loss[loss=0.1883, simple_loss=0.2867, pruned_loss=0.04493, over 12061.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04368, over 2376224.11 frames. ], batch size: 42, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:06:06,084 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122654.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:06:06,882 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9438, 5.8728, 5.6471, 5.1825, 5.1316, 5.8503, 5.3790, 5.2202], device='cuda:1'), covar=tensor([0.0711, 0.0937, 0.0678, 0.1477, 0.0639, 0.0601, 0.1405, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0514, 0.0468, 0.0587, 0.0385, 0.0662, 0.0722, 0.0524], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:06:15,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7619, 3.8540, 3.4275, 3.2625, 3.0560, 2.8749, 3.7640, 2.5219], device='cuda:1'), covar=tensor([0.0290, 0.0090, 0.0128, 0.0148, 0.0322, 0.0277, 0.0091, 0.0355], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0155, 0.0146, 0.0177, 0.0197, 0.0191, 0.0156, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:06:21,007 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122675.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:06:27,897 INFO [finetune.py:992] (1/2) Epoch 2, batch 10450, loss[loss=0.2496, simple_loss=0.3228, pruned_loss=0.0882, over 8250.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2641, pruned_loss=0.04462, over 2371253.38 frames. ], batch size: 98, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:06:46,908 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.972e+02 3.516e+02 4.240e+02 1.276e+03, threshold=7.032e+02, percent-clipped=3.0 2023-05-15 20:07:03,786 INFO [finetune.py:992] (1/2) Epoch 2, batch 10500, loss[loss=0.1882, simple_loss=0.279, pruned_loss=0.04864, over 12335.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2642, pruned_loss=0.04454, over 2368035.53 frames. ], batch size: 35, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:07:31,077 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122773.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 20:07:32,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2023-05-15 20:07:39,214 INFO [finetune.py:992] (1/2) Epoch 2, batch 10550, loss[loss=0.1828, simple_loss=0.2728, pruned_loss=0.04639, over 12041.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2644, pruned_loss=0.0449, over 2366745.53 frames. ], batch size: 40, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:07:40,978 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-15 20:07:57,038 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-15 20:07:58,689 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 2.924e+02 3.382e+02 3.985e+02 8.472e+02, threshold=6.763e+02, percent-clipped=1.0 2023-05-15 20:08:05,227 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122821.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:08:07,426 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122824.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:08:15,045 INFO [finetune.py:992] (1/2) Epoch 2, batch 10600, loss[loss=0.1606, simple_loss=0.2571, pruned_loss=0.03203, over 12305.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2648, pruned_loss=0.04493, over 2363702.56 frames. ], batch size: 34, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:08:26,831 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122851.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:08:42,476 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122872.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:08:51,536 INFO [finetune.py:992] (1/2) Epoch 2, batch 10650, loss[loss=0.1424, simple_loss=0.2244, pruned_loss=0.03017, over 12262.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2652, pruned_loss=0.04545, over 2358460.27 frames. ], batch size: 28, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:09:02,348 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122899.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:09:11,590 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.882e+02 3.250e+02 4.087e+02 1.166e+03, threshold=6.501e+02, percent-clipped=2.0 2023-05-15 20:09:20,749 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1760, 4.4880, 3.9340, 4.8767, 4.3370, 2.7661, 4.0668, 2.9212], device='cuda:1'), covar=tensor([0.0809, 0.0764, 0.1346, 0.0328, 0.1213, 0.1603, 0.0987, 0.2998], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0366, 0.0346, 0.0257, 0.0355, 0.0257, 0.0330, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:09:26,378 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122933.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:09:27,612 INFO [finetune.py:992] (1/2) Epoch 2, batch 10700, loss[loss=0.202, simple_loss=0.2895, pruned_loss=0.05729, over 11670.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2647, pruned_loss=0.04532, over 2363290.48 frames. ], batch size: 48, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:09:40,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-15 20:09:48,913 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4416, 5.2941, 5.4033, 5.4233, 5.0205, 5.0954, 4.9693, 5.3390], device='cuda:1'), covar=tensor([0.0628, 0.0473, 0.0599, 0.0525, 0.1853, 0.1259, 0.0476, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0622, 0.0530, 0.0584, 0.0766, 0.0700, 0.0510, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 20:09:55,996 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122975.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:10:03,053 INFO [finetune.py:992] (1/2) Epoch 2, batch 10750, loss[loss=0.1817, simple_loss=0.271, pruned_loss=0.04618, over 12283.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2647, pruned_loss=0.04501, over 2367941.02 frames. ], batch size: 33, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:10:06,206 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4468, 2.3467, 4.3789, 4.7029, 3.1663, 2.4046, 2.8246, 1.7481], device='cuda:1'), covar=tensor([0.1571, 0.3452, 0.0452, 0.0316, 0.0912, 0.2101, 0.2580, 0.4531], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0364, 0.0262, 0.0284, 0.0246, 0.0272, 0.0344, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:10:23,399 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.914e+02 3.371e+02 4.358e+02 9.799e+02, threshold=6.743e+02, percent-clipped=2.0 2023-05-15 20:10:31,371 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123023.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:10:39,765 INFO [finetune.py:992] (1/2) Epoch 2, batch 10800, loss[loss=0.1483, simple_loss=0.2356, pruned_loss=0.03054, over 11996.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2647, pruned_loss=0.04491, over 2371865.65 frames. ], batch size: 28, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:11:15,866 INFO [finetune.py:992] (1/2) Epoch 2, batch 10850, loss[loss=0.2777, simple_loss=0.3302, pruned_loss=0.1126, over 7632.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2642, pruned_loss=0.04492, over 2363533.52 frames. ], batch size: 97, lr: 4.94e-03, grad_scale: 8.0 2023-05-15 20:11:26,653 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6647, 2.7034, 3.8038, 4.8297, 4.1989, 4.7377, 4.1320, 3.4566], device='cuda:1'), covar=tensor([0.0027, 0.0302, 0.0108, 0.0023, 0.0089, 0.0046, 0.0078, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0118, 0.0099, 0.0074, 0.0098, 0.0108, 0.0086, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:11:36,884 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 3.070e+02 3.614e+02 4.156e+02 8.622e+02, threshold=7.228e+02, percent-clipped=2.0 2023-05-15 20:11:44,120 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123124.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:11:51,909 INFO [finetune.py:992] (1/2) Epoch 2, batch 10900, loss[loss=0.1692, simple_loss=0.2615, pruned_loss=0.03841, over 12192.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2656, pruned_loss=0.04558, over 2361664.15 frames. ], batch size: 35, lr: 4.94e-03, grad_scale: 4.0 2023-05-15 20:12:13,029 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9513, 4.6944, 4.8661, 4.8839, 4.5915, 4.8595, 4.7237, 2.7320], device='cuda:1'), covar=tensor([0.0105, 0.0063, 0.0077, 0.0061, 0.0054, 0.0087, 0.0118, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0072, 0.0076, 0.0070, 0.0057, 0.0086, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:12:18,586 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123172.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:12:28,238 INFO [finetune.py:992] (1/2) Epoch 2, batch 10950, loss[loss=0.1852, simple_loss=0.2764, pruned_loss=0.04701, over 12021.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2657, pruned_loss=0.046, over 2353416.93 frames. ], batch size: 40, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:12:47,333 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2102, 3.8609, 4.1057, 4.5561, 2.9554, 3.9854, 2.6478, 4.0891], device='cuda:1'), covar=tensor([0.1527, 0.0780, 0.0894, 0.0493, 0.1112, 0.0530, 0.1660, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0295, 0.0349, 0.0236, 0.0236, 0.0252, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:12:49,146 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.930e+02 3.524e+02 4.092e+02 6.133e+02, threshold=7.048e+02, percent-clipped=1.0 2023-05-15 20:12:59,172 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123228.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:13:01,377 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0605, 2.4501, 3.1060, 3.9769, 2.2393, 4.1245, 3.9775, 4.1926], device='cuda:1'), covar=tensor([0.0107, 0.1051, 0.0394, 0.0101, 0.1112, 0.0158, 0.0180, 0.0063], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0201, 0.0185, 0.0114, 0.0183, 0.0171, 0.0167, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:13:03,914 INFO [finetune.py:992] (1/2) Epoch 2, batch 11000, loss[loss=0.2107, simple_loss=0.2948, pruned_loss=0.06326, over 12018.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2677, pruned_loss=0.04714, over 2339637.03 frames. ], batch size: 40, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:13:10,512 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1919, 2.1876, 3.6105, 3.0810, 3.5098, 3.2534, 2.4115, 3.4846], device='cuda:1'), covar=tensor([0.0094, 0.0323, 0.0127, 0.0178, 0.0113, 0.0127, 0.0296, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0188, 0.0166, 0.0170, 0.0188, 0.0145, 0.0178, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:13:39,342 INFO [finetune.py:992] (1/2) Epoch 2, batch 11050, loss[loss=0.2278, simple_loss=0.3178, pruned_loss=0.0689, over 11089.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2709, pruned_loss=0.04926, over 2304048.39 frames. ], batch size: 55, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:13:40,574 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 20:13:50,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1940, 2.0666, 3.6844, 3.1059, 3.4739, 3.1825, 2.4171, 3.5404], device='cuda:1'), covar=tensor([0.0105, 0.0397, 0.0111, 0.0197, 0.0135, 0.0154, 0.0329, 0.0104], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0188, 0.0165, 0.0169, 0.0187, 0.0145, 0.0178, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:13:58,422 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7728, 4.4617, 4.4639, 4.6411, 4.4139, 4.6510, 4.5409, 2.5965], device='cuda:1'), covar=tensor([0.0070, 0.0061, 0.0095, 0.0058, 0.0051, 0.0090, 0.0081, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0070, 0.0057, 0.0086, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:14:00,283 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 3.317e+02 4.016e+02 4.800e+02 7.007e+02, threshold=8.032e+02, percent-clipped=0.0 2023-05-15 20:14:00,477 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123314.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:14:08,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0982, 4.7981, 4.8589, 5.0288, 4.7608, 4.9433, 4.9053, 2.8706], device='cuda:1'), covar=tensor([0.0070, 0.0062, 0.0080, 0.0048, 0.0046, 0.0077, 0.0078, 0.0613], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0070, 0.0057, 0.0086, 0.0075, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:14:14,847 INFO [finetune.py:992] (1/2) Epoch 2, batch 11100, loss[loss=0.252, simple_loss=0.3373, pruned_loss=0.08333, over 11102.00 frames. ], tot_loss[loss=0.1905, simple_loss=0.2761, pruned_loss=0.05244, over 2267390.14 frames. ], batch size: 55, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:14:18,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 2023-05-15 20:14:40,475 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3476, 3.4941, 3.2750, 3.5906, 3.4365, 2.5859, 3.3138, 2.8398], device='cuda:1'), covar=tensor([0.0711, 0.0970, 0.1184, 0.0618, 0.1019, 0.1429, 0.0924, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0357, 0.0337, 0.0250, 0.0347, 0.0252, 0.0323, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:14:43,701 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123375.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:14:50,128 INFO [finetune.py:992] (1/2) Epoch 2, batch 11150, loss[loss=0.3277, simple_loss=0.384, pruned_loss=0.1357, over 7070.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2834, pruned_loss=0.05745, over 2218175.68 frames. ], batch size: 99, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:15:11,056 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.003e+02 3.729e+02 4.554e+02 5.583e+02 9.300e+02, threshold=9.109e+02, percent-clipped=4.0 2023-05-15 20:15:20,873 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123428.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:15:25,746 INFO [finetune.py:992] (1/2) Epoch 2, batch 11200, loss[loss=0.3367, simple_loss=0.387, pruned_loss=0.1431, over 6962.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2906, pruned_loss=0.06262, over 2160980.04 frames. ], batch size: 99, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:16:01,415 INFO [finetune.py:992] (1/2) Epoch 2, batch 11250, loss[loss=0.2224, simple_loss=0.3066, pruned_loss=0.06909, over 11816.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.3, pruned_loss=0.06953, over 2076248.72 frames. ], batch size: 44, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:16:04,463 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123489.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:16:14,425 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2427, 6.0204, 5.5671, 5.6838, 6.0729, 5.5016, 5.6827, 5.7101], device='cuda:1'), covar=tensor([0.1029, 0.0722, 0.0878, 0.1563, 0.0794, 0.1682, 0.1335, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0319, 0.0433, 0.0345, 0.0398, 0.0421, 0.0398, 0.0354, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:16:21,091 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.415e+02 3.857e+02 4.379e+02 5.227e+02 1.226e+03, threshold=8.757e+02, percent-clipped=1.0 2023-05-15 20:16:27,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9068, 4.6548, 4.7047, 4.8380, 4.5558, 4.8033, 4.6816, 2.5216], device='cuda:1'), covar=tensor([0.0092, 0.0056, 0.0084, 0.0058, 0.0046, 0.0111, 0.0072, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0070, 0.0074, 0.0068, 0.0056, 0.0083, 0.0073, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:16:31,180 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123528.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:16:36,542 INFO [finetune.py:992] (1/2) Epoch 2, batch 11300, loss[loss=0.2406, simple_loss=0.3238, pruned_loss=0.07871, over 10939.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3054, pruned_loss=0.07314, over 2029783.20 frames. ], batch size: 55, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:17:04,531 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123576.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:17:10,546 INFO [finetune.py:992] (1/2) Epoch 2, batch 11350, loss[loss=0.2796, simple_loss=0.3649, pruned_loss=0.09711, over 10489.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3102, pruned_loss=0.07632, over 1983639.57 frames. ], batch size: 68, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:17:31,983 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.677e+02 3.491e+02 4.298e+02 4.901e+02 8.727e+02, threshold=8.596e+02, percent-clipped=0.0 2023-05-15 20:17:46,099 INFO [finetune.py:992] (1/2) Epoch 2, batch 11400, loss[loss=0.3328, simple_loss=0.386, pruned_loss=0.1398, over 6618.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3152, pruned_loss=0.08028, over 1909732.74 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:17:51,782 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7599, 3.6451, 3.6442, 3.8341, 3.4779, 3.8227, 3.7999, 3.9277], device='cuda:1'), covar=tensor([0.0208, 0.0175, 0.0175, 0.0245, 0.0611, 0.0314, 0.0166, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0163, 0.0163, 0.0204, 0.0207, 0.0181, 0.0151, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-15 20:18:10,405 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123670.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:18:21,023 INFO [finetune.py:992] (1/2) Epoch 2, batch 11450, loss[loss=0.3004, simple_loss=0.368, pruned_loss=0.1164, over 6941.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3189, pruned_loss=0.08342, over 1874611.11 frames. ], batch size: 97, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:18:41,104 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.333e+02 3.687e+02 4.414e+02 5.331e+02 1.195e+03, threshold=8.827e+02, percent-clipped=1.0 2023-05-15 20:18:54,876 INFO [finetune.py:992] (1/2) Epoch 2, batch 11500, loss[loss=0.2463, simple_loss=0.3268, pruned_loss=0.08292, over 10188.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3209, pruned_loss=0.08493, over 1851487.37 frames. ], batch size: 68, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:19:29,384 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123784.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:19:29,952 INFO [finetune.py:992] (1/2) Epoch 2, batch 11550, loss[loss=0.2555, simple_loss=0.3314, pruned_loss=0.08983, over 6757.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.323, pruned_loss=0.08702, over 1816679.50 frames. ], batch size: 101, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:19:50,523 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.512e+02 3.528e+02 4.263e+02 5.111e+02 8.080e+02, threshold=8.525e+02, percent-clipped=0.0 2023-05-15 20:20:04,557 INFO [finetune.py:992] (1/2) Epoch 2, batch 11600, loss[loss=0.2484, simple_loss=0.3243, pruned_loss=0.08628, over 10217.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.324, pruned_loss=0.0885, over 1787815.65 frames. ], batch size: 68, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:20:06,350 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-05-15 20:20:32,237 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123872.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:20:33,039 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7722, 2.5328, 3.5085, 3.6030, 2.9563, 2.7034, 2.5940, 2.3914], device='cuda:1'), covar=tensor([0.1043, 0.2351, 0.0560, 0.0370, 0.0709, 0.1525, 0.2142, 0.3073], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0349, 0.0251, 0.0272, 0.0235, 0.0263, 0.0334, 0.0337], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:20:33,693 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7008, 3.6236, 3.6261, 3.7389, 3.4438, 3.7236, 3.7190, 3.8200], device='cuda:1'), covar=tensor([0.0193, 0.0158, 0.0173, 0.0251, 0.0582, 0.0314, 0.0157, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0157, 0.0157, 0.0198, 0.0201, 0.0174, 0.0145, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-15 20:20:41,339 INFO [finetune.py:992] (1/2) Epoch 2, batch 11650, loss[loss=0.1974, simple_loss=0.2761, pruned_loss=0.0594, over 6363.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3235, pruned_loss=0.08872, over 1769792.49 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:21:01,766 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.751e+02 3.666e+02 4.379e+02 5.316e+02 1.010e+03, threshold=8.758e+02, percent-clipped=1.0 2023-05-15 20:21:15,342 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123933.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:21:16,388 INFO [finetune.py:992] (1/2) Epoch 2, batch 11700, loss[loss=0.2716, simple_loss=0.3335, pruned_loss=0.1049, over 6532.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3229, pruned_loss=0.08914, over 1745719.66 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:21:30,000 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6504, 5.5372, 5.5697, 5.6109, 5.2765, 5.3256, 5.2155, 5.6495], device='cuda:1'), covar=tensor([0.0704, 0.0507, 0.0764, 0.0654, 0.1763, 0.1208, 0.0503, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0567, 0.0492, 0.0532, 0.0685, 0.0640, 0.0472, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:21:40,777 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123970.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:21:50,547 INFO [finetune.py:992] (1/2) Epoch 2, batch 11750, loss[loss=0.2137, simple_loss=0.3022, pruned_loss=0.06261, over 11044.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3216, pruned_loss=0.08847, over 1733926.37 frames. ], batch size: 55, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:21:52,055 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8879, 2.1082, 2.6411, 2.8723, 2.8890, 2.9284, 2.7926, 2.4874], device='cuda:1'), covar=tensor([0.0049, 0.0300, 0.0139, 0.0056, 0.0087, 0.0091, 0.0084, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0074, 0.0114, 0.0093, 0.0070, 0.0092, 0.0104, 0.0081, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:22:14,226 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.778e+02 3.583e+02 4.299e+02 5.119e+02 9.573e+02, threshold=8.597e+02, percent-clipped=1.0 2023-05-15 20:22:17,080 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124018.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:22:29,156 INFO [finetune.py:992] (1/2) Epoch 2, batch 11800, loss[loss=0.2845, simple_loss=0.3486, pruned_loss=0.1102, over 6760.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3241, pruned_loss=0.0908, over 1704358.63 frames. ], batch size: 102, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:22:30,099 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124036.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:22:46,192 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5101, 4.4493, 4.3859, 4.0253, 4.1401, 4.4964, 4.1908, 4.1393], device='cuda:1'), covar=tensor([0.0734, 0.0918, 0.0613, 0.1179, 0.2153, 0.0757, 0.1350, 0.0984], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0476, 0.0430, 0.0535, 0.0357, 0.0598, 0.0652, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:22:48,169 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124062.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:23:03,233 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124084.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:23:03,781 INFO [finetune.py:992] (1/2) Epoch 2, batch 11850, loss[loss=0.2995, simple_loss=0.3502, pruned_loss=0.1244, over 6446.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.327, pruned_loss=0.09275, over 1684487.49 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:23:12,210 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124097.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:23:24,116 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.378e+02 3.700e+02 4.417e+02 5.163e+02 1.393e+03, threshold=8.834e+02, percent-clipped=2.0 2023-05-15 20:23:30,457 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124123.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:23:36,443 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124132.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:23:38,327 INFO [finetune.py:992] (1/2) Epoch 2, batch 11900, loss[loss=0.2955, simple_loss=0.3555, pruned_loss=0.1178, over 6777.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3259, pruned_loss=0.09133, over 1673433.27 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:23:51,470 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2316, 4.2310, 2.5540, 2.3208, 3.7321, 2.3473, 3.8576, 3.0047], device='cuda:1'), covar=tensor([0.0679, 0.0464, 0.1174, 0.1716, 0.0227, 0.1531, 0.0360, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0228, 0.0167, 0.0190, 0.0128, 0.0174, 0.0181, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:23:56,889 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0115, 1.9789, 2.3942, 2.1097, 2.2791, 2.3160, 1.7402, 2.2521], device='cuda:1'), covar=tensor([0.0098, 0.0257, 0.0137, 0.0178, 0.0157, 0.0151, 0.0269, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0176, 0.0148, 0.0157, 0.0171, 0.0135, 0.0166, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:24:00,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-15 20:24:12,515 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8313, 2.4366, 3.4830, 3.6090, 2.8003, 2.7587, 2.5734, 2.4759], device='cuda:1'), covar=tensor([0.1012, 0.2553, 0.0537, 0.0408, 0.0755, 0.1527, 0.2344, 0.2994], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0348, 0.0249, 0.0270, 0.0235, 0.0264, 0.0334, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:24:13,582 INFO [finetune.py:992] (1/2) Epoch 2, batch 11950, loss[loss=0.2331, simple_loss=0.3082, pruned_loss=0.07895, over 6994.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3234, pruned_loss=0.0888, over 1670208.20 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:24:34,129 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.275e+02 3.145e+02 3.876e+02 4.654e+02 9.978e+02, threshold=7.752e+02, percent-clipped=1.0 2023-05-15 20:24:41,252 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6511, 2.5494, 4.2011, 4.3811, 3.0697, 2.7031, 2.8821, 2.0268], device='cuda:1'), covar=tensor([0.1419, 0.3102, 0.0440, 0.0347, 0.0951, 0.2074, 0.2490, 0.4328], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0348, 0.0249, 0.0269, 0.0234, 0.0263, 0.0334, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:24:43,694 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124228.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:24:48,921 INFO [finetune.py:992] (1/2) Epoch 2, batch 12000, loss[loss=0.272, simple_loss=0.3249, pruned_loss=0.1096, over 6913.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3182, pruned_loss=0.08452, over 1666175.01 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:24:48,921 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 20:25:03,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8837, 2.2340, 3.3010, 3.8135, 3.6639, 3.8800, 3.6022, 2.5628], device='cuda:1'), covar=tensor([0.0031, 0.0347, 0.0125, 0.0034, 0.0079, 0.0048, 0.0078, 0.0365], device='cuda:1'), in_proj_covar=tensor([0.0073, 0.0113, 0.0092, 0.0069, 0.0090, 0.0102, 0.0080, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:25:03,435 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4219, 4.1915, 4.0514, 4.4890, 3.0414, 4.0732, 2.7602, 3.8729], device='cuda:1'), covar=tensor([0.1753, 0.0576, 0.0896, 0.0417, 0.1150, 0.0546, 0.1919, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0249, 0.0281, 0.0331, 0.0228, 0.0228, 0.0248, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:25:06,676 INFO [finetune.py:1026] (1/2) Epoch 2, validation: loss=0.2947, simple_loss=0.3701, pruned_loss=0.1097, over 1020973.00 frames. 2023-05-15 20:25:06,677 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 20:25:32,582 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124272.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:25:41,016 INFO [finetune.py:992] (1/2) Epoch 2, batch 12050, loss[loss=0.2193, simple_loss=0.3004, pruned_loss=0.06914, over 12289.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3134, pruned_loss=0.08095, over 1656684.48 frames. ], batch size: 37, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:25:46,798 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7104, 3.8990, 3.6666, 4.3265, 3.9025, 2.5548, 3.6842, 2.9151], device='cuda:1'), covar=tensor([0.0953, 0.1111, 0.1252, 0.0311, 0.1209, 0.1916, 0.1004, 0.3283], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0344, 0.0321, 0.0232, 0.0331, 0.0249, 0.0311, 0.0337], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:26:00,643 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.298e+02 3.069e+02 3.589e+02 4.447e+02 1.626e+03, threshold=7.177e+02, percent-clipped=4.0 2023-05-15 20:26:12,884 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124333.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:26:14,009 INFO [finetune.py:992] (1/2) Epoch 2, batch 12100, loss[loss=0.2366, simple_loss=0.3107, pruned_loss=0.08131, over 7261.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3117, pruned_loss=0.07912, over 1666074.50 frames. ], batch size: 98, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:26:22,606 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124348.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 20:26:46,132 INFO [finetune.py:992] (1/2) Epoch 2, batch 12150, loss[loss=0.2284, simple_loss=0.3059, pruned_loss=0.07548, over 7167.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3117, pruned_loss=0.07837, over 1688898.75 frames. ], batch size: 100, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:26:50,647 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124392.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:27:02,208 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124409.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 20:27:05,398 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 3.308e+02 3.785e+02 4.469e+02 7.285e+02, threshold=7.570e+02, percent-clipped=1.0 2023-05-15 20:27:08,020 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124418.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:27:16,830 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124432.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:27:18,577 INFO [finetune.py:992] (1/2) Epoch 2, batch 12200, loss[loss=0.2667, simple_loss=0.3451, pruned_loss=0.09412, over 6813.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3136, pruned_loss=0.08, over 1665669.03 frames. ], batch size: 99, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:27:34,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-15 20:28:02,416 INFO [finetune.py:992] (1/2) Epoch 3, batch 0, loss[loss=0.1699, simple_loss=0.2567, pruned_loss=0.04152, over 12030.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2567, pruned_loss=0.04152, over 12030.00 frames. ], batch size: 28, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:28:02,416 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 20:28:19,572 INFO [finetune.py:1026] (1/2) Epoch 3, validation: loss=0.2957, simple_loss=0.3691, pruned_loss=0.1112, over 1020973.00 frames. 2023-05-15 20:28:19,573 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 20:28:37,725 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124493.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:28:52,935 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.479e+02 3.480e+02 3.921e+02 4.591e+02 7.457e+02, threshold=7.842e+02, percent-clipped=0.0 2023-05-15 20:28:55,731 INFO [finetune.py:992] (1/2) Epoch 3, batch 50, loss[loss=0.1921, simple_loss=0.2795, pruned_loss=0.05239, over 12351.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2789, pruned_loss=0.05217, over 539767.62 frames. ], batch size: 35, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:29:02,944 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124528.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:29:09,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-15 20:29:31,893 INFO [finetune.py:992] (1/2) Epoch 3, batch 100, loss[loss=0.1938, simple_loss=0.2862, pruned_loss=0.05066, over 12160.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2717, pruned_loss=0.04948, over 949966.69 frames. ], batch size: 36, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:29:36,715 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-15 20:29:37,632 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124576.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:30:05,511 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 2.969e+02 3.564e+02 3.993e+02 8.882e+02, threshold=7.127e+02, percent-clipped=1.0 2023-05-15 20:30:07,666 INFO [finetune.py:992] (1/2) Epoch 3, batch 150, loss[loss=0.1625, simple_loss=0.2468, pruned_loss=0.0391, over 12183.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2729, pruned_loss=0.04869, over 1269261.25 frames. ], batch size: 29, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:30:14,821 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124628.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:30:22,105 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124638.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:30:32,157 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5050, 4.6856, 4.2258, 5.0239, 4.6538, 2.8018, 4.3216, 3.0449], device='cuda:1'), covar=tensor([0.0721, 0.0828, 0.1354, 0.0396, 0.0956, 0.1857, 0.0965, 0.3579], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0353, 0.0333, 0.0239, 0.0339, 0.0255, 0.0320, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:30:40,562 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124663.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:30:43,770 INFO [finetune.py:992] (1/2) Epoch 3, batch 200, loss[loss=0.195, simple_loss=0.2822, pruned_loss=0.05392, over 12365.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2711, pruned_loss=0.04827, over 1505483.30 frames. ], batch size: 38, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:31:01,539 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124692.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:31:06,559 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124699.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:31:09,993 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124704.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 20:31:17,437 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.889e+02 3.397e+02 4.137e+02 7.935e+02, threshold=6.795e+02, percent-clipped=1.0 2023-05-15 20:31:19,616 INFO [finetune.py:992] (1/2) Epoch 3, batch 250, loss[loss=0.2008, simple_loss=0.2977, pruned_loss=0.05198, over 12153.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2698, pruned_loss=0.04742, over 1702239.99 frames. ], batch size: 34, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:31:19,769 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124718.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:31:24,111 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124724.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:31:24,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-15 20:31:30,542 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1124, 5.0855, 4.8320, 4.9693, 4.6872, 5.0349, 5.0267, 5.3963], device='cuda:1'), covar=tensor([0.0291, 0.0143, 0.0216, 0.0324, 0.0646, 0.0287, 0.0164, 0.0156], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0158, 0.0160, 0.0199, 0.0201, 0.0175, 0.0147, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-15 20:31:35,445 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124740.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:31:53,867 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124766.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:31:55,239 INFO [finetune.py:992] (1/2) Epoch 3, batch 300, loss[loss=0.1836, simple_loss=0.2654, pruned_loss=0.0509, over 12132.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2679, pruned_loss=0.04645, over 1859281.68 frames. ], batch size: 38, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:32:03,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-05-15 20:32:09,684 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124788.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:32:29,654 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.943e+02 3.581e+02 4.310e+02 1.397e+03, threshold=7.162e+02, percent-clipped=2.0 2023-05-15 20:32:31,888 INFO [finetune.py:992] (1/2) Epoch 3, batch 350, loss[loss=0.151, simple_loss=0.2401, pruned_loss=0.03098, over 12340.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2673, pruned_loss=0.0464, over 1967089.17 frames. ], batch size: 30, lr: 4.93e-03, grad_scale: 4.0 2023-05-15 20:32:33,550 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124820.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:33:07,896 INFO [finetune.py:992] (1/2) Epoch 3, batch 400, loss[loss=0.1812, simple_loss=0.2649, pruned_loss=0.04879, over 12167.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2662, pruned_loss=0.04563, over 2068757.96 frames. ], batch size: 31, lr: 4.93e-03, grad_scale: 8.0 2023-05-15 20:33:17,343 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124881.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:33:25,761 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2816, 4.8453, 5.2129, 4.5708, 4.8483, 4.6628, 5.3273, 4.9487], device='cuda:1'), covar=tensor([0.0270, 0.0329, 0.0298, 0.0246, 0.0362, 0.0290, 0.0190, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0227, 0.0244, 0.0220, 0.0222, 0.0219, 0.0199, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:33:41,351 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.875e+02 3.470e+02 4.271e+02 8.495e+02, threshold=6.940e+02, percent-clipped=2.0 2023-05-15 20:33:43,536 INFO [finetune.py:992] (1/2) Epoch 3, batch 450, loss[loss=0.1361, simple_loss=0.2227, pruned_loss=0.02472, over 12092.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.266, pruned_loss=0.04525, over 2143000.42 frames. ], batch size: 32, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:33:50,760 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124928.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:34:19,817 INFO [finetune.py:992] (1/2) Epoch 3, batch 500, loss[loss=0.1511, simple_loss=0.238, pruned_loss=0.03208, over 12255.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.2669, pruned_loss=0.04591, over 2188642.28 frames. ], batch size: 32, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:34:25,341 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124976.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:34:38,969 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124994.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:34:44,324 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4110, 5.1456, 5.3707, 5.3757, 4.9258, 5.0307, 4.8450, 5.3021], device='cuda:1'), covar=tensor([0.0676, 0.0548, 0.0690, 0.0548, 0.1970, 0.1156, 0.0514, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0586, 0.0505, 0.0547, 0.0717, 0.0657, 0.0482, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0004], device='cuda:1') 2023-05-15 20:34:46,521 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125004.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 20:34:54,251 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.166e+02 2.875e+02 3.419e+02 4.165e+02 9.341e+02, threshold=6.838e+02, percent-clipped=3.0 2023-05-15 20:34:56,420 INFO [finetune.py:992] (1/2) Epoch 3, batch 550, loss[loss=0.173, simple_loss=0.266, pruned_loss=0.04007, over 12371.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.266, pruned_loss=0.04552, over 2236462.96 frames. ], batch size: 38, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:34:57,249 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125019.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:35:20,761 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125052.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 20:35:32,288 INFO [finetune.py:992] (1/2) Epoch 3, batch 600, loss[loss=0.1476, simple_loss=0.2259, pruned_loss=0.03462, over 12274.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2642, pruned_loss=0.04461, over 2275061.19 frames. ], batch size: 28, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:35:46,467 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125088.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:36:06,021 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.348e+02 2.876e+02 3.279e+02 4.102e+02 1.110e+03, threshold=6.559e+02, percent-clipped=4.0 2023-05-15 20:36:07,642 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6594, 2.2158, 3.0193, 2.6752, 2.8889, 2.7689, 2.0264, 2.9119], device='cuda:1'), covar=tensor([0.0092, 0.0268, 0.0115, 0.0202, 0.0135, 0.0146, 0.0321, 0.0110], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0187, 0.0159, 0.0168, 0.0183, 0.0144, 0.0179, 0.0154], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:36:08,149 INFO [finetune.py:992] (1/2) Epoch 3, batch 650, loss[loss=0.1831, simple_loss=0.2703, pruned_loss=0.04796, over 10718.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2662, pruned_loss=0.0452, over 2294845.68 frames. ], batch size: 68, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:36:08,966 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0106, 5.7115, 5.2972, 5.2274, 5.7998, 5.0748, 5.4678, 5.2992], device='cuda:1'), covar=tensor([0.1489, 0.1107, 0.0939, 0.2169, 0.1017, 0.2364, 0.1714, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0436, 0.0346, 0.0396, 0.0415, 0.0399, 0.0354, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:36:15,349 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125128.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:36:21,003 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125136.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:36:33,171 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0559, 6.0029, 5.8125, 5.4012, 5.1143, 6.0135, 5.6049, 5.3751], device='cuda:1'), covar=tensor([0.0613, 0.0851, 0.0608, 0.1318, 0.0631, 0.0592, 0.1349, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0495, 0.0454, 0.0564, 0.0368, 0.0628, 0.0683, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:36:44,335 INFO [finetune.py:992] (1/2) Epoch 3, batch 700, loss[loss=0.1853, simple_loss=0.2757, pruned_loss=0.04745, over 12175.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2651, pruned_loss=0.04507, over 2310555.58 frames. ], batch size: 36, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:36:49,949 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125176.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:36:55,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3966, 4.6536, 2.7367, 2.4382, 4.0472, 2.3463, 3.9236, 3.2038], device='cuda:1'), covar=tensor([0.0625, 0.0565, 0.1091, 0.1689, 0.0280, 0.1495, 0.0469, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0239, 0.0171, 0.0194, 0.0133, 0.0178, 0.0187, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:36:59,350 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125189.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:37:17,937 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 2.879e+02 3.347e+02 4.098e+02 6.743e+02, threshold=6.693e+02, percent-clipped=1.0 2023-05-15 20:37:20,147 INFO [finetune.py:992] (1/2) Epoch 3, batch 750, loss[loss=0.1698, simple_loss=0.2516, pruned_loss=0.04401, over 12268.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2657, pruned_loss=0.04498, over 2331410.42 frames. ], batch size: 28, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:37:56,256 INFO [finetune.py:992] (1/2) Epoch 3, batch 800, loss[loss=0.1784, simple_loss=0.266, pruned_loss=0.04543, over 12139.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2662, pruned_loss=0.04508, over 2333855.83 frames. ], batch size: 34, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:38:15,504 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125294.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:38:30,182 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 2.921e+02 3.420e+02 4.045e+02 6.574e+02, threshold=6.840e+02, percent-clipped=0.0 2023-05-15 20:38:32,333 INFO [finetune.py:992] (1/2) Epoch 3, batch 850, loss[loss=0.17, simple_loss=0.2616, pruned_loss=0.03919, over 11219.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2659, pruned_loss=0.04514, over 2339475.12 frames. ], batch size: 55, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:38:33,075 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125319.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:38:49,457 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125342.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:38:59,415 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125356.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:39:06,979 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125367.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:39:07,600 INFO [finetune.py:992] (1/2) Epoch 3, batch 900, loss[loss=0.1784, simple_loss=0.2667, pruned_loss=0.04503, over 12044.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2652, pruned_loss=0.04487, over 2343141.89 frames. ], batch size: 37, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:39:13,537 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125376.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:39:42,146 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.973e+02 3.414e+02 3.855e+02 6.034e+02, threshold=6.827e+02, percent-clipped=0.0 2023-05-15 20:39:43,779 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125417.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 20:39:44,254 INFO [finetune.py:992] (1/2) Epoch 3, batch 950, loss[loss=0.2036, simple_loss=0.2861, pruned_loss=0.06051, over 11613.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2653, pruned_loss=0.04466, over 2350199.00 frames. ], batch size: 48, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:39:46,449 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125421.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:39:58,417 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125437.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:40:06,149 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1717, 6.1480, 5.9039, 5.4603, 5.1125, 6.1220, 5.6703, 5.4093], device='cuda:1'), covar=tensor([0.0667, 0.0794, 0.0602, 0.1468, 0.0625, 0.0661, 0.1540, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0502, 0.0459, 0.0570, 0.0372, 0.0634, 0.0694, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:40:20,374 INFO [finetune.py:992] (1/2) Epoch 3, batch 1000, loss[loss=0.1631, simple_loss=0.2428, pruned_loss=0.04168, over 11394.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2648, pruned_loss=0.04444, over 2360273.46 frames. ], batch size: 25, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:40:26,242 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125476.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:40:30,406 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125482.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:40:31,750 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125484.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:40:46,443 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-15 20:40:54,381 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.897e+02 3.324e+02 3.887e+02 6.575e+02, threshold=6.648e+02, percent-clipped=0.0 2023-05-15 20:40:55,819 INFO [finetune.py:992] (1/2) Epoch 3, batch 1050, loss[loss=0.1962, simple_loss=0.2882, pruned_loss=0.05209, over 12041.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2651, pruned_loss=0.04431, over 2369403.27 frames. ], batch size: 42, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:41:00,157 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125524.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:41:31,870 INFO [finetune.py:992] (1/2) Epoch 3, batch 1100, loss[loss=0.1947, simple_loss=0.2754, pruned_loss=0.05701, over 12025.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2646, pruned_loss=0.04459, over 2369988.62 frames. ], batch size: 31, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:41:51,033 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2027, 4.4449, 2.7388, 2.5517, 3.8081, 2.4464, 3.8580, 3.0229], device='cuda:1'), covar=tensor([0.0681, 0.0613, 0.1080, 0.1521, 0.0307, 0.1428, 0.0435, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0244, 0.0174, 0.0197, 0.0135, 0.0181, 0.0189, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:41:57,554 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125603.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:42:02,563 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1501, 3.7201, 5.3359, 2.8539, 2.9440, 3.9064, 3.4349, 3.9922], device='cuda:1'), covar=tensor([0.0301, 0.0953, 0.0295, 0.1113, 0.1881, 0.1430, 0.1178, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0224, 0.0222, 0.0174, 0.0230, 0.0268, 0.0218, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:42:06,362 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 3.110e+02 3.758e+02 4.115e+02 6.117e+02, threshold=7.515e+02, percent-clipped=0.0 2023-05-15 20:42:07,865 INFO [finetune.py:992] (1/2) Epoch 3, batch 1150, loss[loss=0.1443, simple_loss=0.222, pruned_loss=0.03333, over 12273.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2649, pruned_loss=0.04507, over 2368820.92 frames. ], batch size: 28, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:42:13,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-15 20:42:17,967 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7798, 4.4271, 4.5047, 4.5746, 4.3790, 4.6642, 4.4921, 2.5522], device='cuda:1'), covar=tensor([0.0083, 0.0061, 0.0085, 0.0057, 0.0055, 0.0082, 0.0071, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0069, 0.0074, 0.0067, 0.0055, 0.0082, 0.0072, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:42:21,469 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0584, 2.6468, 3.3250, 4.0285, 3.7051, 4.0445, 3.6560, 2.8272], device='cuda:1'), covar=tensor([0.0030, 0.0308, 0.0143, 0.0036, 0.0095, 0.0058, 0.0105, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0117, 0.0096, 0.0071, 0.0095, 0.0107, 0.0084, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:42:27,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5387, 5.0979, 5.4596, 4.8782, 5.0818, 4.9079, 5.5746, 5.1550], device='cuda:1'), covar=tensor([0.0202, 0.0272, 0.0232, 0.0202, 0.0287, 0.0258, 0.0143, 0.0225], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0232, 0.0249, 0.0224, 0.0226, 0.0224, 0.0203, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:42:40,329 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125664.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:42:42,960 INFO [finetune.py:992] (1/2) Epoch 3, batch 1200, loss[loss=0.1765, simple_loss=0.2639, pruned_loss=0.04458, over 12341.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.265, pruned_loss=0.04496, over 2371469.24 frames. ], batch size: 31, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:42:47,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6509, 2.7291, 3.5827, 4.5708, 4.0215, 4.5755, 4.0264, 3.0253], device='cuda:1'), covar=tensor([0.0020, 0.0334, 0.0131, 0.0048, 0.0099, 0.0066, 0.0105, 0.0370], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0117, 0.0096, 0.0071, 0.0095, 0.0106, 0.0084, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:43:15,052 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125712.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 20:43:18,354 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.924e+02 3.346e+02 4.080e+02 7.279e+02, threshold=6.692e+02, percent-clipped=0.0 2023-05-15 20:43:18,964 INFO [finetune.py:992] (1/2) Epoch 3, batch 1250, loss[loss=0.2173, simple_loss=0.2954, pruned_loss=0.06958, over 12313.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2646, pruned_loss=0.04505, over 2375132.99 frames. ], batch size: 34, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:43:24,735 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125725.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:43:29,599 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125732.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:43:39,571 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0706, 2.3608, 3.6297, 3.1242, 3.4541, 3.2224, 2.4464, 3.6185], device='cuda:1'), covar=tensor([0.0127, 0.0332, 0.0154, 0.0211, 0.0140, 0.0126, 0.0302, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0189, 0.0163, 0.0170, 0.0185, 0.0144, 0.0179, 0.0156], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:43:55,033 INFO [finetune.py:992] (1/2) Epoch 3, batch 1300, loss[loss=0.1987, simple_loss=0.2862, pruned_loss=0.05557, over 8089.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2641, pruned_loss=0.04476, over 2373382.34 frames. ], batch size: 98, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:44:01,537 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125777.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:44:06,484 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125784.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:44:07,928 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125786.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:44:29,861 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 3.119e+02 3.684e+02 4.625e+02 9.118e+02, threshold=7.367e+02, percent-clipped=4.0 2023-05-15 20:44:30,598 INFO [finetune.py:992] (1/2) Epoch 3, batch 1350, loss[loss=0.207, simple_loss=0.2918, pruned_loss=0.06106, over 12204.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2646, pruned_loss=0.04476, over 2376320.78 frames. ], batch size: 35, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:44:40,907 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125832.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:44:49,692 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3491, 4.9208, 5.2945, 4.6579, 4.9024, 4.7407, 5.3866, 5.0239], device='cuda:1'), covar=tensor([0.0280, 0.0338, 0.0278, 0.0252, 0.0368, 0.0269, 0.0177, 0.0294], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0234, 0.0250, 0.0226, 0.0227, 0.0225, 0.0204, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:44:59,475 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1379, 5.9175, 5.4952, 5.4321, 5.9692, 5.3874, 5.6441, 5.5261], device='cuda:1'), covar=tensor([0.1293, 0.0827, 0.0849, 0.1842, 0.1000, 0.1937, 0.1442, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0438, 0.0348, 0.0399, 0.0419, 0.0403, 0.0353, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:45:06,291 INFO [finetune.py:992] (1/2) Epoch 3, batch 1400, loss[loss=0.1683, simple_loss=0.2577, pruned_loss=0.0394, over 12305.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2655, pruned_loss=0.04473, over 2383797.85 frames. ], batch size: 34, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:45:41,508 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.155e+02 2.857e+02 3.363e+02 4.209e+02 6.773e+02, threshold=6.726e+02, percent-clipped=0.0 2023-05-15 20:45:42,194 INFO [finetune.py:992] (1/2) Epoch 3, batch 1450, loss[loss=0.2104, simple_loss=0.2904, pruned_loss=0.06517, over 12046.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2654, pruned_loss=0.04476, over 2385737.22 frames. ], batch size: 42, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:46:11,341 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125959.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:46:17,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0979, 4.8501, 4.9446, 4.9914, 4.8215, 4.9925, 4.8866, 2.8474], device='cuda:1'), covar=tensor([0.0107, 0.0045, 0.0064, 0.0052, 0.0038, 0.0081, 0.0067, 0.0585], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0070, 0.0075, 0.0068, 0.0055, 0.0083, 0.0073, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:46:17,786 INFO [finetune.py:992] (1/2) Epoch 3, batch 1500, loss[loss=0.1591, simple_loss=0.2446, pruned_loss=0.03679, over 12027.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2653, pruned_loss=0.0448, over 2381027.68 frames. ], batch size: 31, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:46:24,181 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2613, 4.9661, 5.0728, 5.1067, 4.9696, 5.1704, 4.9825, 2.9584], device='cuda:1'), covar=tensor([0.0102, 0.0044, 0.0059, 0.0045, 0.0035, 0.0070, 0.0062, 0.0523], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0070, 0.0074, 0.0067, 0.0055, 0.0083, 0.0073, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:46:39,635 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125998.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:46:52,443 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126012.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 20:46:55,824 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.152e+02 2.827e+02 3.591e+02 4.249e+02 9.010e+02, threshold=7.182e+02, percent-clipped=2.0 2023-05-15 20:46:56,546 INFO [finetune.py:992] (1/2) Epoch 3, batch 1550, loss[loss=0.1599, simple_loss=0.254, pruned_loss=0.03286, over 12192.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2653, pruned_loss=0.04492, over 2384504.71 frames. ], batch size: 35, lr: 4.92e-03, grad_scale: 4.0 2023-05-15 20:47:07,449 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126032.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:47:26,519 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126059.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:47:27,038 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126060.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:47:32,257 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1439, 2.6239, 3.6268, 3.2006, 3.5602, 3.2209, 2.5496, 3.6399], device='cuda:1'), covar=tensor([0.0091, 0.0265, 0.0121, 0.0168, 0.0097, 0.0148, 0.0281, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0188, 0.0164, 0.0170, 0.0185, 0.0144, 0.0180, 0.0157], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:47:32,758 INFO [finetune.py:992] (1/2) Epoch 3, batch 1600, loss[loss=0.1475, simple_loss=0.241, pruned_loss=0.02699, over 12041.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2658, pruned_loss=0.0453, over 2382916.52 frames. ], batch size: 31, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:47:32,930 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2865, 4.9393, 4.9723, 5.1580, 4.8780, 5.1507, 4.8917, 2.8623], device='cuda:1'), covar=tensor([0.0077, 0.0043, 0.0069, 0.0049, 0.0039, 0.0065, 0.0077, 0.0560], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0070, 0.0075, 0.0068, 0.0055, 0.0083, 0.0073, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:47:32,957 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6893, 2.5135, 3.3205, 4.5512, 2.3605, 4.6143, 4.6356, 4.8339], device='cuda:1'), covar=tensor([0.0103, 0.1110, 0.0397, 0.0151, 0.1270, 0.0205, 0.0128, 0.0069], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0199, 0.0179, 0.0110, 0.0184, 0.0167, 0.0162, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:47:39,327 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126077.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:47:40,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1499, 5.0906, 4.9476, 5.0326, 4.5669, 5.1647, 5.1352, 5.3769], device='cuda:1'), covar=tensor([0.0185, 0.0127, 0.0179, 0.0252, 0.0750, 0.0255, 0.0133, 0.0145], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0177, 0.0179, 0.0222, 0.0224, 0.0195, 0.0165, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 20:47:41,405 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126080.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:47:41,798 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-15 20:47:42,188 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126081.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:47:54,874 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3412, 2.2651, 3.4945, 4.2997, 3.7619, 4.3057, 3.6942, 2.9631], device='cuda:1'), covar=tensor([0.0024, 0.0362, 0.0122, 0.0032, 0.0101, 0.0043, 0.0095, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0117, 0.0096, 0.0071, 0.0096, 0.0106, 0.0084, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 20:48:05,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-15 20:48:07,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-15 20:48:07,482 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.864e+02 2.888e+02 3.272e+02 4.126e+02 7.200e+02, threshold=6.545e+02, percent-clipped=1.0 2023-05-15 20:48:08,159 INFO [finetune.py:992] (1/2) Epoch 3, batch 1650, loss[loss=0.1859, simple_loss=0.2738, pruned_loss=0.04902, over 12354.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2653, pruned_loss=0.04488, over 2384046.00 frames. ], batch size: 36, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:48:13,761 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126125.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:48:29,192 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.33 vs. limit=5.0 2023-05-15 20:48:42,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0098, 2.2661, 3.6073, 3.0259, 3.4034, 3.1133, 2.4857, 3.4915], device='cuda:1'), covar=tensor([0.0124, 0.0329, 0.0126, 0.0219, 0.0125, 0.0156, 0.0302, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0189, 0.0165, 0.0171, 0.0186, 0.0145, 0.0180, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:48:44,870 INFO [finetune.py:992] (1/2) Epoch 3, batch 1700, loss[loss=0.176, simple_loss=0.267, pruned_loss=0.04249, over 12355.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.265, pruned_loss=0.04498, over 2374957.09 frames. ], batch size: 36, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:49:20,342 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 3.003e+02 3.423e+02 3.859e+02 6.319e+02, threshold=6.845e+02, percent-clipped=0.0 2023-05-15 20:49:21,057 INFO [finetune.py:992] (1/2) Epoch 3, batch 1750, loss[loss=0.1642, simple_loss=0.2481, pruned_loss=0.04015, over 12371.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.264, pruned_loss=0.04432, over 2381720.76 frames. ], batch size: 30, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:49:49,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-05-15 20:49:50,192 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126259.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:49:56,497 INFO [finetune.py:992] (1/2) Epoch 3, batch 1800, loss[loss=0.156, simple_loss=0.2318, pruned_loss=0.04012, over 11717.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2636, pruned_loss=0.04402, over 2380607.60 frames. ], batch size: 26, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:50:24,805 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126307.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:50:31,954 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.025e+02 2.885e+02 3.436e+02 3.955e+02 8.404e+02, threshold=6.872e+02, percent-clipped=1.0 2023-05-15 20:50:32,655 INFO [finetune.py:992] (1/2) Epoch 3, batch 1850, loss[loss=0.1997, simple_loss=0.2947, pruned_loss=0.05236, over 10544.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2641, pruned_loss=0.04399, over 2383393.87 frames. ], batch size: 68, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:50:58,769 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126354.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:51:08,437 INFO [finetune.py:992] (1/2) Epoch 3, batch 1900, loss[loss=0.2018, simple_loss=0.2843, pruned_loss=0.05962, over 12308.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2656, pruned_loss=0.04447, over 2377589.44 frames. ], batch size: 34, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:51:17,205 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126380.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:51:17,887 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126381.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:51:43,429 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.796e+02 3.344e+02 4.119e+02 6.290e+02, threshold=6.688e+02, percent-clipped=0.0 2023-05-15 20:51:44,199 INFO [finetune.py:992] (1/2) Epoch 3, batch 1950, loss[loss=0.167, simple_loss=0.249, pruned_loss=0.04252, over 12330.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2646, pruned_loss=0.04437, over 2379610.81 frames. ], batch size: 30, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:51:52,636 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126429.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:52:01,290 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126441.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:52:07,617 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126450.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:52:10,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-15 20:52:21,048 INFO [finetune.py:992] (1/2) Epoch 3, batch 2000, loss[loss=0.1693, simple_loss=0.2606, pruned_loss=0.03895, over 12145.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2647, pruned_loss=0.04404, over 2383338.66 frames. ], batch size: 34, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:52:32,601 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5954, 3.2606, 4.9859, 2.7036, 2.8773, 3.8793, 3.1092, 3.8663], device='cuda:1'), covar=tensor([0.0464, 0.1127, 0.0301, 0.1155, 0.1804, 0.1217, 0.1369, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0227, 0.0227, 0.0175, 0.0232, 0.0271, 0.0222, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:52:42,410 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2070, 6.0996, 6.0034, 5.5306, 5.3058, 6.1300, 5.6657, 5.4427], device='cuda:1'), covar=tensor([0.0709, 0.1004, 0.0578, 0.1468, 0.0576, 0.0670, 0.1544, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0508, 0.0468, 0.0583, 0.0379, 0.0653, 0.0712, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:52:51,827 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126511.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:52:55,870 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.900e+02 3.495e+02 4.173e+02 1.166e+03, threshold=6.989e+02, percent-clipped=2.0 2023-05-15 20:52:56,573 INFO [finetune.py:992] (1/2) Epoch 3, batch 2050, loss[loss=0.1723, simple_loss=0.2645, pruned_loss=0.04006, over 12114.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.265, pruned_loss=0.04431, over 2385525.98 frames. ], batch size: 39, lr: 4.92e-03, grad_scale: 8.0 2023-05-15 20:53:12,457 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8139, 5.6550, 5.2002, 5.1678, 5.7293, 5.0078, 5.2705, 5.1473], device='cuda:1'), covar=tensor([0.1472, 0.1031, 0.1048, 0.1815, 0.0906, 0.2317, 0.1808, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0445, 0.0352, 0.0404, 0.0423, 0.0406, 0.0359, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:53:29,411 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4290, 4.9268, 5.3681, 4.6828, 4.9205, 4.7582, 5.4017, 5.1116], device='cuda:1'), covar=tensor([0.0219, 0.0332, 0.0229, 0.0221, 0.0349, 0.0320, 0.0190, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0235, 0.0251, 0.0227, 0.0229, 0.0226, 0.0208, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 20:53:32,073 INFO [finetune.py:992] (1/2) Epoch 3, batch 2100, loss[loss=0.2043, simple_loss=0.2908, pruned_loss=0.05892, over 12123.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2654, pruned_loss=0.04451, over 2388902.76 frames. ], batch size: 39, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 20:54:08,263 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.812e+02 2.866e+02 3.267e+02 3.861e+02 7.719e+02, threshold=6.534e+02, percent-clipped=2.0 2023-05-15 20:54:08,946 INFO [finetune.py:992] (1/2) Epoch 3, batch 2150, loss[loss=0.1554, simple_loss=0.2363, pruned_loss=0.0372, over 12026.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2651, pruned_loss=0.04411, over 2394906.74 frames. ], batch size: 31, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 20:54:34,819 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126654.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:54:44,579 INFO [finetune.py:992] (1/2) Epoch 3, batch 2200, loss[loss=0.1553, simple_loss=0.237, pruned_loss=0.03674, over 12421.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2649, pruned_loss=0.04409, over 2393311.41 frames. ], batch size: 32, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 20:55:08,784 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126702.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:55:20,246 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.895e+02 3.584e+02 4.587e+02 8.110e+02, threshold=7.168e+02, percent-clipped=5.0 2023-05-15 20:55:20,266 INFO [finetune.py:992] (1/2) Epoch 3, batch 2250, loss[loss=0.1613, simple_loss=0.2457, pruned_loss=0.03846, over 12074.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2645, pruned_loss=0.04384, over 2388025.53 frames. ], batch size: 32, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 20:55:33,780 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126736.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:55:52,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-15 20:55:56,830 INFO [finetune.py:992] (1/2) Epoch 3, batch 2300, loss[loss=0.1802, simple_loss=0.2689, pruned_loss=0.04574, over 12186.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.264, pruned_loss=0.0437, over 2387472.03 frames. ], batch size: 35, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 20:56:00,537 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8618, 3.0846, 4.6799, 5.0440, 3.1404, 2.9401, 3.1653, 2.3742], device='cuda:1'), covar=tensor([0.1281, 0.2751, 0.0470, 0.0316, 0.1052, 0.1781, 0.2266, 0.3573], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0367, 0.0260, 0.0285, 0.0248, 0.0275, 0.0346, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:56:13,326 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5700, 3.1487, 5.0274, 2.3117, 2.5161, 3.7664, 3.0115, 3.7636], device='cuda:1'), covar=tensor([0.0457, 0.1174, 0.0236, 0.1355, 0.2089, 0.1194, 0.1445, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0226, 0.0227, 0.0175, 0.0232, 0.0271, 0.0221, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:56:24,198 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126806.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:56:31,347 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126816.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:56:32,500 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.251e+02 3.050e+02 3.373e+02 4.203e+02 1.390e+03, threshold=6.745e+02, percent-clipped=2.0 2023-05-15 20:56:32,524 INFO [finetune.py:992] (1/2) Epoch 3, batch 2350, loss[loss=0.1798, simple_loss=0.2694, pruned_loss=0.04515, over 11527.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2638, pruned_loss=0.04367, over 2382775.39 frames. ], batch size: 48, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 20:56:52,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-15 20:56:59,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-15 20:56:59,482 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5361, 4.2375, 4.2182, 4.4208, 4.3018, 4.4119, 4.2548, 2.4409], device='cuda:1'), covar=tensor([0.0181, 0.0095, 0.0149, 0.0108, 0.0074, 0.0173, 0.0130, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0069, 0.0056, 0.0085, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:57:04,400 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6043, 2.1605, 3.2776, 4.4241, 2.3835, 4.5274, 4.5013, 4.7314], device='cuda:1'), covar=tensor([0.0119, 0.1285, 0.0462, 0.0127, 0.1223, 0.0215, 0.0129, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0199, 0.0180, 0.0110, 0.0183, 0.0168, 0.0162, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:57:07,782 INFO [finetune.py:992] (1/2) Epoch 3, batch 2400, loss[loss=0.1792, simple_loss=0.265, pruned_loss=0.04668, over 12126.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2643, pruned_loss=0.04399, over 2381582.55 frames. ], batch size: 38, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 20:57:15,076 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126877.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:57:23,535 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9822, 2.3266, 3.3753, 2.9389, 3.2888, 3.0512, 2.3146, 3.3612], device='cuda:1'), covar=tensor([0.0107, 0.0315, 0.0150, 0.0206, 0.0133, 0.0168, 0.0323, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0190, 0.0166, 0.0172, 0.0187, 0.0147, 0.0181, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 20:57:44,841 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.886e+02 3.409e+02 3.999e+02 2.480e+03, threshold=6.818e+02, percent-clipped=0.0 2023-05-15 20:57:44,861 INFO [finetune.py:992] (1/2) Epoch 3, batch 2450, loss[loss=0.1698, simple_loss=0.2545, pruned_loss=0.0425, over 12349.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2626, pruned_loss=0.04349, over 2377835.59 frames. ], batch size: 31, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 20:58:20,051 INFO [finetune.py:992] (1/2) Epoch 3, batch 2500, loss[loss=0.1856, simple_loss=0.2599, pruned_loss=0.05564, over 12027.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04367, over 2378743.54 frames. ], batch size: 31, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 20:58:55,962 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.205e+02 2.860e+02 3.434e+02 4.041e+02 7.879e+02, threshold=6.868e+02, percent-clipped=5.0 2023-05-15 20:58:55,986 INFO [finetune.py:992] (1/2) Epoch 3, batch 2550, loss[loss=0.1619, simple_loss=0.249, pruned_loss=0.03745, over 12106.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2634, pruned_loss=0.04391, over 2383497.26 frames. ], batch size: 33, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 20:59:09,044 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1439, 5.1055, 5.0226, 5.0437, 4.6386, 5.2529, 5.1717, 5.4305], device='cuda:1'), covar=tensor([0.0179, 0.0120, 0.0141, 0.0261, 0.0664, 0.0209, 0.0110, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0178, 0.0178, 0.0222, 0.0226, 0.0197, 0.0164, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 20:59:09,755 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127036.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:59:32,993 INFO [finetune.py:992] (1/2) Epoch 3, batch 2600, loss[loss=0.1451, simple_loss=0.2256, pruned_loss=0.03233, over 12274.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2632, pruned_loss=0.04394, over 2384160.11 frames. ], batch size: 28, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 20:59:39,326 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7841, 5.5390, 5.0415, 5.0813, 5.6263, 4.8935, 5.1796, 5.0608], device='cuda:1'), covar=tensor([0.1265, 0.1008, 0.1079, 0.1913, 0.0940, 0.2350, 0.1578, 0.1184], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0438, 0.0346, 0.0397, 0.0418, 0.0397, 0.0357, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 20:59:44,390 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127084.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 20:59:57,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-15 21:00:00,115 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127106.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:00:08,665 INFO [finetune.py:992] (1/2) Epoch 3, batch 2650, loss[loss=0.2498, simple_loss=0.3193, pruned_loss=0.09014, over 7605.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2621, pruned_loss=0.04335, over 2381796.68 frames. ], batch size: 98, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 21:00:09,357 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.855e+02 3.326e+02 4.115e+02 6.905e+02, threshold=6.652e+02, percent-clipped=2.0 2023-05-15 21:00:34,237 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127154.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:00:44,059 INFO [finetune.py:992] (1/2) Epoch 3, batch 2700, loss[loss=0.1726, simple_loss=0.2711, pruned_loss=0.037, over 12159.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2628, pruned_loss=0.04366, over 2376881.70 frames. ], batch size: 34, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 21:00:47,083 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127172.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:01:21,321 INFO [finetune.py:992] (1/2) Epoch 3, batch 2750, loss[loss=0.1902, simple_loss=0.2781, pruned_loss=0.05112, over 12145.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2618, pruned_loss=0.04344, over 2375116.13 frames. ], batch size: 39, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 21:01:21,935 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.156e+02 3.036e+02 3.627e+02 4.430e+02 3.094e+03, threshold=7.254e+02, percent-clipped=6.0 2023-05-15 21:01:43,138 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127249.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:01:46,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 21:01:56,098 INFO [finetune.py:992] (1/2) Epoch 3, batch 2800, loss[loss=0.1739, simple_loss=0.2605, pruned_loss=0.04368, over 12338.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2621, pruned_loss=0.04347, over 2377101.62 frames. ], batch size: 35, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:02:02,682 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2520, 4.6973, 2.7639, 2.4415, 4.0010, 2.5233, 4.0038, 3.1499], device='cuda:1'), covar=tensor([0.0668, 0.0511, 0.1149, 0.1592, 0.0262, 0.1323, 0.0443, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0244, 0.0172, 0.0195, 0.0135, 0.0178, 0.0190, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:02:10,350 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2348, 2.0757, 3.0488, 4.1427, 2.0018, 4.3214, 4.1974, 4.3314], device='cuda:1'), covar=tensor([0.0126, 0.1272, 0.0436, 0.0111, 0.1309, 0.0187, 0.0145, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0202, 0.0181, 0.0111, 0.0185, 0.0171, 0.0165, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:02:26,083 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127310.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:02:32,275 INFO [finetune.py:992] (1/2) Epoch 3, batch 2850, loss[loss=0.1812, simple_loss=0.2751, pruned_loss=0.04365, over 12179.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.263, pruned_loss=0.04361, over 2378140.59 frames. ], batch size: 35, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:02:32,982 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.250e+02 2.893e+02 3.266e+02 4.085e+02 1.029e+03, threshold=6.532e+02, percent-clipped=2.0 2023-05-15 21:02:42,821 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9717, 2.3563, 3.5604, 2.9283, 3.3280, 3.0697, 2.3149, 3.4335], device='cuda:1'), covar=tensor([0.0114, 0.0332, 0.0121, 0.0240, 0.0120, 0.0161, 0.0352, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0191, 0.0169, 0.0174, 0.0190, 0.0149, 0.0182, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:03:08,423 INFO [finetune.py:992] (1/2) Epoch 3, batch 2900, loss[loss=0.1837, simple_loss=0.2693, pruned_loss=0.04902, over 11192.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2632, pruned_loss=0.04351, over 2374946.83 frames. ], batch size: 55, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:03:44,260 INFO [finetune.py:992] (1/2) Epoch 3, batch 2950, loss[loss=0.1723, simple_loss=0.2639, pruned_loss=0.04035, over 12309.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2625, pruned_loss=0.04347, over 2374552.13 frames. ], batch size: 34, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:03:44,893 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.714e+02 3.322e+02 3.797e+02 1.040e+03, threshold=6.645e+02, percent-clipped=2.0 2023-05-15 21:03:50,799 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127427.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:04:15,674 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2074, 5.2261, 5.0662, 5.0857, 4.6863, 5.2163, 5.0939, 5.4145], device='cuda:1'), covar=tensor([0.0183, 0.0099, 0.0137, 0.0239, 0.0661, 0.0248, 0.0143, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0177, 0.0178, 0.0222, 0.0226, 0.0197, 0.0166, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 21:04:20,326 INFO [finetune.py:992] (1/2) Epoch 3, batch 3000, loss[loss=0.2066, simple_loss=0.2888, pruned_loss=0.06218, over 12061.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2635, pruned_loss=0.04379, over 2376982.72 frames. ], batch size: 37, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:04:20,327 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 21:04:38,161 INFO [finetune.py:1026] (1/2) Epoch 3, validation: loss=0.3349, simple_loss=0.4074, pruned_loss=0.1313, over 1020973.00 frames. 2023-05-15 21:04:38,161 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 21:04:40,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-15 21:04:41,108 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127472.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:04:46,166 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5533, 3.6689, 3.3924, 3.2813, 2.9918, 2.7804, 3.6606, 2.4871], device='cuda:1'), covar=tensor([0.0344, 0.0144, 0.0147, 0.0173, 0.0387, 0.0340, 0.0118, 0.0391], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0155, 0.0144, 0.0175, 0.0199, 0.0192, 0.0154, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 21:04:52,480 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127488.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 21:05:13,615 INFO [finetune.py:992] (1/2) Epoch 3, batch 3050, loss[loss=0.1875, simple_loss=0.2625, pruned_loss=0.05631, over 12147.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2621, pruned_loss=0.04362, over 2377369.57 frames. ], batch size: 36, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:05:14,239 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.817e+02 3.491e+02 4.167e+02 6.780e+02, threshold=6.983e+02, percent-clipped=2.0 2023-05-15 21:05:15,045 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127520.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:05:21,517 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127529.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:05:49,168 INFO [finetune.py:992] (1/2) Epoch 3, batch 3100, loss[loss=0.2065, simple_loss=0.2902, pruned_loss=0.06143, over 8403.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2633, pruned_loss=0.04417, over 2379250.38 frames. ], batch size: 97, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:06:06,141 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127590.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:06:07,465 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0505, 2.0740, 2.6922, 3.0500, 2.8652, 3.0570, 2.8356, 2.4522], device='cuda:1'), covar=tensor([0.0060, 0.0371, 0.0162, 0.0061, 0.0141, 0.0102, 0.0094, 0.0328], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0122, 0.0100, 0.0074, 0.0099, 0.0109, 0.0088, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:06:10,367 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6263, 3.7588, 3.4354, 3.3900, 3.1056, 2.9327, 3.8059, 2.6083], device='cuda:1'), covar=tensor([0.0317, 0.0115, 0.0131, 0.0143, 0.0332, 0.0322, 0.0103, 0.0355], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0155, 0.0144, 0.0175, 0.0199, 0.0191, 0.0153, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 21:06:16,958 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127605.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:06:26,021 INFO [finetune.py:992] (1/2) Epoch 3, batch 3150, loss[loss=0.184, simple_loss=0.2791, pruned_loss=0.04446, over 12073.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2633, pruned_loss=0.04421, over 2372323.82 frames. ], batch size: 42, lr: 4.91e-03, grad_scale: 4.0 2023-05-15 21:06:27,412 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 2.701e+02 3.163e+02 3.791e+02 6.573e+02, threshold=6.327e+02, percent-clipped=0.0 2023-05-15 21:06:40,441 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5702, 2.1040, 3.2692, 4.3619, 2.3177, 4.6286, 4.5779, 4.6398], device='cuda:1'), covar=tensor([0.0104, 0.1278, 0.0432, 0.0127, 0.1186, 0.0166, 0.0109, 0.0090], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0201, 0.0182, 0.0112, 0.0184, 0.0171, 0.0165, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:06:57,483 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127662.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:06:59,622 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1991, 4.8442, 5.0157, 5.1251, 4.7075, 5.0154, 4.8656, 2.9751], device='cuda:1'), covar=tensor([0.0068, 0.0060, 0.0060, 0.0046, 0.0057, 0.0076, 0.0081, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0076, 0.0070, 0.0056, 0.0085, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:07:01,602 INFO [finetune.py:992] (1/2) Epoch 3, batch 3200, loss[loss=0.1788, simple_loss=0.2674, pruned_loss=0.0451, over 12062.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2625, pruned_loss=0.04394, over 2366748.06 frames. ], batch size: 40, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:07:37,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-15 21:07:37,610 INFO [finetune.py:992] (1/2) Epoch 3, batch 3250, loss[loss=0.1992, simple_loss=0.2815, pruned_loss=0.05842, over 11139.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.04444, over 2372548.65 frames. ], batch size: 55, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:07:39,049 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.130e+02 2.912e+02 3.431e+02 4.040e+02 1.283e+03, threshold=6.861e+02, percent-clipped=3.0 2023-05-15 21:07:41,411 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127723.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:07:55,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0390, 3.9410, 2.5118, 2.1040, 3.4720, 2.0763, 3.5640, 2.7498], device='cuda:1'), covar=tensor([0.0707, 0.0581, 0.1198, 0.1705, 0.0276, 0.1547, 0.0495, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0246, 0.0175, 0.0196, 0.0135, 0.0178, 0.0192, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:08:13,974 INFO [finetune.py:992] (1/2) Epoch 3, batch 3300, loss[loss=0.1685, simple_loss=0.2495, pruned_loss=0.04371, over 12331.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2633, pruned_loss=0.04412, over 2375957.39 frames. ], batch size: 30, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:08:22,190 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127779.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:08:25,040 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127783.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:08:37,451 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8420, 2.2389, 3.3904, 2.8679, 3.2453, 3.0361, 2.1689, 3.3029], device='cuda:1'), covar=tensor([0.0112, 0.0301, 0.0137, 0.0208, 0.0116, 0.0144, 0.0288, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0192, 0.0167, 0.0174, 0.0189, 0.0148, 0.0181, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:08:44,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-05-15 21:08:50,073 INFO [finetune.py:992] (1/2) Epoch 3, batch 3350, loss[loss=0.1396, simple_loss=0.223, pruned_loss=0.02808, over 11991.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2619, pruned_loss=0.04344, over 2378457.63 frames. ], batch size: 28, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:08:51,510 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.744e+02 3.185e+02 3.891e+02 6.795e+02, threshold=6.371e+02, percent-clipped=0.0 2023-05-15 21:08:53,176 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9306, 2.3302, 2.2243, 2.2710, 2.0158, 1.9050, 2.3200, 1.7120], device='cuda:1'), covar=tensor([0.0273, 0.0157, 0.0149, 0.0154, 0.0303, 0.0217, 0.0135, 0.0324], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0155, 0.0144, 0.0175, 0.0199, 0.0191, 0.0153, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 21:09:05,939 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127840.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:09:06,581 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4524, 5.0197, 5.4023, 4.7923, 4.9919, 4.8608, 5.4320, 5.0326], device='cuda:1'), covar=tensor([0.0231, 0.0293, 0.0275, 0.0206, 0.0307, 0.0257, 0.0200, 0.0284], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0236, 0.0255, 0.0228, 0.0232, 0.0230, 0.0210, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:09:13,039 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127850.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:09:26,311 INFO [finetune.py:992] (1/2) Epoch 3, batch 3400, loss[loss=0.1615, simple_loss=0.2618, pruned_loss=0.03062, over 12363.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2629, pruned_loss=0.04369, over 2374006.68 frames. ], batch size: 36, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:09:29,108 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127872.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:09:36,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5471, 2.8995, 3.7223, 4.5677, 4.0636, 4.5080, 4.0429, 3.2136], device='cuda:1'), covar=tensor([0.0024, 0.0304, 0.0111, 0.0031, 0.0079, 0.0062, 0.0077, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0120, 0.0099, 0.0073, 0.0098, 0.0107, 0.0087, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:09:38,873 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127885.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:09:52,941 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127905.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:09:57,175 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127911.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:10:01,963 INFO [finetune.py:992] (1/2) Epoch 3, batch 3450, loss[loss=0.2123, simple_loss=0.2773, pruned_loss=0.07363, over 8270.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2634, pruned_loss=0.04395, over 2376318.14 frames. ], batch size: 98, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:10:03,383 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.048e+02 3.039e+02 3.509e+02 4.157e+02 8.157e+02, threshold=7.017e+02, percent-clipped=2.0 2023-05-15 21:10:12,794 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127933.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:10:16,594 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-15 21:10:26,956 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127953.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:10:37,298 INFO [finetune.py:992] (1/2) Epoch 3, batch 3500, loss[loss=0.1678, simple_loss=0.2548, pruned_loss=0.04038, over 12032.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2635, pruned_loss=0.04377, over 2378464.36 frames. ], batch size: 31, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:10:44,494 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4315, 2.2963, 3.2391, 4.2288, 2.5275, 4.4273, 4.3823, 4.5735], device='cuda:1'), covar=tensor([0.0120, 0.1227, 0.0424, 0.0164, 0.1083, 0.0219, 0.0129, 0.0078], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0201, 0.0181, 0.0112, 0.0184, 0.0170, 0.0163, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:11:17,004 INFO [finetune.py:992] (1/2) Epoch 3, batch 3550, loss[loss=0.1546, simple_loss=0.2381, pruned_loss=0.03555, over 12114.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.263, pruned_loss=0.04353, over 2371646.06 frames. ], batch size: 30, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:11:17,096 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128018.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:11:18,463 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.920e+02 2.802e+02 3.301e+02 3.996e+02 6.263e+02, threshold=6.602e+02, percent-clipped=0.0 2023-05-15 21:11:35,684 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1820, 4.7733, 5.1309, 4.5488, 4.7587, 4.5234, 5.1957, 4.8277], device='cuda:1'), covar=tensor([0.0219, 0.0298, 0.0268, 0.0211, 0.0308, 0.0260, 0.0180, 0.0290], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0239, 0.0259, 0.0230, 0.0234, 0.0232, 0.0213, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:11:53,311 INFO [finetune.py:992] (1/2) Epoch 3, batch 3600, loss[loss=0.1778, simple_loss=0.2583, pruned_loss=0.04864, over 12204.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2619, pruned_loss=0.04308, over 2380956.37 frames. ], batch size: 29, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:12:04,085 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128083.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:12:28,276 INFO [finetune.py:992] (1/2) Epoch 3, batch 3650, loss[loss=0.1753, simple_loss=0.2608, pruned_loss=0.0449, over 12147.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2631, pruned_loss=0.04336, over 2368284.97 frames. ], batch size: 36, lr: 4.91e-03, grad_scale: 8.0 2023-05-15 21:12:29,751 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.114e+02 2.958e+02 3.551e+02 4.110e+02 1.232e+03, threshold=7.103e+02, percent-clipped=2.0 2023-05-15 21:12:37,713 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128131.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:12:40,613 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128135.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:13:05,093 INFO [finetune.py:992] (1/2) Epoch 3, batch 3700, loss[loss=0.2005, simple_loss=0.291, pruned_loss=0.055, over 10600.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2631, pruned_loss=0.04367, over 2364784.82 frames. ], batch size: 68, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:13:17,190 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128185.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:13:32,302 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128206.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:13:37,311 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2407, 5.9964, 5.5846, 5.5990, 6.0726, 5.5036, 5.7060, 5.6461], device='cuda:1'), covar=tensor([0.1391, 0.0890, 0.0897, 0.1962, 0.0895, 0.2132, 0.1359, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0328, 0.0452, 0.0354, 0.0408, 0.0431, 0.0406, 0.0367, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:13:40,811 INFO [finetune.py:992] (1/2) Epoch 3, batch 3750, loss[loss=0.1882, simple_loss=0.2824, pruned_loss=0.04704, over 12295.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2626, pruned_loss=0.04367, over 2364350.38 frames. ], batch size: 37, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:13:42,274 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 2.975e+02 3.523e+02 4.137e+02 1.386e+03, threshold=7.045e+02, percent-clipped=3.0 2023-05-15 21:13:47,991 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128228.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:13:51,407 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128233.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:14:11,306 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1474, 4.2211, 4.0918, 4.4686, 2.9324, 4.0256, 2.6626, 4.0739], device='cuda:1'), covar=tensor([0.1553, 0.0548, 0.0915, 0.0651, 0.1094, 0.0571, 0.1701, 0.1619], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0258, 0.0289, 0.0343, 0.0235, 0.0234, 0.0252, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:14:15,973 INFO [finetune.py:992] (1/2) Epoch 3, batch 3800, loss[loss=0.1639, simple_loss=0.242, pruned_loss=0.04288, over 11438.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2615, pruned_loss=0.04354, over 2367317.58 frames. ], batch size: 25, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:14:53,011 INFO [finetune.py:992] (1/2) Epoch 3, batch 3850, loss[loss=0.1459, simple_loss=0.2209, pruned_loss=0.03542, over 12185.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2612, pruned_loss=0.04321, over 2372864.55 frames. ], batch size: 29, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:14:53,154 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128318.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:14:54,346 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.143e+02 3.113e+02 3.561e+02 4.163e+02 1.278e+03, threshold=7.122e+02, percent-clipped=0.0 2023-05-15 21:14:54,531 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128320.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:15:27,208 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128366.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:15:28,616 INFO [finetune.py:992] (1/2) Epoch 3, batch 3900, loss[loss=0.153, simple_loss=0.2457, pruned_loss=0.03017, over 12352.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2607, pruned_loss=0.04307, over 2378877.15 frames. ], batch size: 35, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:15:32,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-15 21:15:38,093 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128381.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:15:51,961 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-05-15 21:15:55,276 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7396, 2.8783, 5.2467, 2.4049, 2.4837, 4.1029, 2.8816, 4.0019], device='cuda:1'), covar=tensor([0.0365, 0.1314, 0.0233, 0.1243, 0.1963, 0.1023, 0.1536, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0226, 0.0228, 0.0176, 0.0231, 0.0272, 0.0221, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:15:58,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-15 21:15:59,516 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6136, 2.6700, 4.0892, 4.3350, 2.8004, 2.6202, 2.7742, 2.1412], device='cuda:1'), covar=tensor([0.1341, 0.2812, 0.0557, 0.0444, 0.1124, 0.1950, 0.2447, 0.3560], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0368, 0.0260, 0.0285, 0.0250, 0.0276, 0.0345, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:16:00,959 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128413.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:16:04,428 INFO [finetune.py:992] (1/2) Epoch 3, batch 3950, loss[loss=0.1384, simple_loss=0.2299, pruned_loss=0.02345, over 12174.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2612, pruned_loss=0.0433, over 2374368.00 frames. ], batch size: 29, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:16:05,834 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.899e+02 3.271e+02 3.921e+02 6.752e+02, threshold=6.543e+02, percent-clipped=1.0 2023-05-15 21:16:16,773 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128435.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:16:28,522 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1260, 5.0974, 4.9661, 5.0347, 4.6307, 5.0852, 5.0197, 5.3111], device='cuda:1'), covar=tensor([0.0170, 0.0128, 0.0145, 0.0263, 0.0640, 0.0237, 0.0136, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0183, 0.0182, 0.0228, 0.0231, 0.0202, 0.0170, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 21:16:29,357 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6470, 2.8145, 4.4836, 4.7231, 2.9571, 2.7006, 2.9649, 2.0566], device='cuda:1'), covar=tensor([0.1341, 0.2717, 0.0460, 0.0345, 0.1114, 0.1943, 0.2414, 0.3686], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0368, 0.0260, 0.0285, 0.0250, 0.0275, 0.0345, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:16:41,043 INFO [finetune.py:992] (1/2) Epoch 3, batch 4000, loss[loss=0.1719, simple_loss=0.2456, pruned_loss=0.04906, over 11779.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2612, pruned_loss=0.04341, over 2379435.82 frames. ], batch size: 26, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:16:45,570 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128474.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:16:49,173 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5204, 2.8587, 3.6269, 4.4745, 3.9177, 4.4981, 3.7976, 3.1566], device='cuda:1'), covar=tensor([0.0029, 0.0309, 0.0147, 0.0057, 0.0122, 0.0072, 0.0112, 0.0321], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0124, 0.0102, 0.0076, 0.0101, 0.0111, 0.0090, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:16:51,895 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128483.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:17:00,552 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6842, 2.9939, 4.8337, 4.9709, 2.9887, 2.8370, 3.0649, 2.1648], device='cuda:1'), covar=tensor([0.1361, 0.2749, 0.0339, 0.0347, 0.1066, 0.1811, 0.2435, 0.3492], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0368, 0.0261, 0.0285, 0.0250, 0.0276, 0.0345, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:17:08,227 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128506.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:17:16,405 INFO [finetune.py:992] (1/2) Epoch 3, batch 4050, loss[loss=0.1815, simple_loss=0.2756, pruned_loss=0.04365, over 12201.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2625, pruned_loss=0.0441, over 2360358.63 frames. ], batch size: 35, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:17:17,795 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.064e+02 2.990e+02 3.571e+02 4.197e+02 7.894e+02, threshold=7.143e+02, percent-clipped=2.0 2023-05-15 21:17:23,699 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128528.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:17:32,852 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8907, 5.8565, 5.6587, 5.1207, 5.0675, 5.7664, 5.3454, 5.1801], device='cuda:1'), covar=tensor([0.0641, 0.0788, 0.0635, 0.1432, 0.0690, 0.0700, 0.1581, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0504, 0.0472, 0.0582, 0.0383, 0.0657, 0.0719, 0.0529], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:17:41,899 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128554.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:17:51,683 INFO [finetune.py:992] (1/2) Epoch 3, batch 4100, loss[loss=0.1682, simple_loss=0.2567, pruned_loss=0.03981, over 12295.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2631, pruned_loss=0.04427, over 2366214.83 frames. ], batch size: 33, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:17:57,527 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128576.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:18:09,877 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 2023-05-15 21:18:20,734 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0512, 4.6657, 4.7925, 4.9944, 4.6701, 4.9333, 4.8320, 2.9600], device='cuda:1'), covar=tensor([0.0093, 0.0066, 0.0080, 0.0054, 0.0045, 0.0075, 0.0070, 0.0569], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0075, 0.0068, 0.0056, 0.0084, 0.0074, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:18:29,037 INFO [finetune.py:992] (1/2) Epoch 3, batch 4150, loss[loss=0.2326, simple_loss=0.3075, pruned_loss=0.07885, over 7896.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2633, pruned_loss=0.04423, over 2363847.03 frames. ], batch size: 97, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:18:31,156 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.169e+02 2.865e+02 3.448e+02 4.142e+02 7.233e+02, threshold=6.896e+02, percent-clipped=1.0 2023-05-15 21:18:40,182 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-15 21:19:04,688 INFO [finetune.py:992] (1/2) Epoch 3, batch 4200, loss[loss=0.1647, simple_loss=0.2516, pruned_loss=0.03894, over 12275.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2634, pruned_loss=0.04451, over 2355964.06 frames. ], batch size: 33, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:19:10,597 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128676.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:19:19,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-15 21:19:22,217 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2158, 4.6209, 2.8226, 2.5450, 4.0140, 2.0866, 3.8863, 2.9691], device='cuda:1'), covar=tensor([0.0707, 0.0439, 0.1179, 0.1622, 0.0263, 0.1683, 0.0437, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0246, 0.0175, 0.0196, 0.0136, 0.0177, 0.0192, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:19:39,207 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4621, 4.9407, 3.0931, 2.9656, 4.1229, 2.6161, 4.1208, 3.3747], device='cuda:1'), covar=tensor([0.0671, 0.0476, 0.1118, 0.1388, 0.0316, 0.1349, 0.0399, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0247, 0.0175, 0.0197, 0.0137, 0.0178, 0.0192, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:19:40,408 INFO [finetune.py:992] (1/2) Epoch 3, batch 4250, loss[loss=0.2069, simple_loss=0.2952, pruned_loss=0.05929, over 11828.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.263, pruned_loss=0.04439, over 2364736.12 frames. ], batch size: 44, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:19:42,610 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.169e+02 2.988e+02 3.450e+02 4.097e+02 6.207e+02, threshold=6.901e+02, percent-clipped=0.0 2023-05-15 21:19:50,654 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0994, 5.9735, 5.5264, 5.6000, 6.0547, 5.4308, 5.5552, 5.5977], device='cuda:1'), covar=tensor([0.1237, 0.0817, 0.0855, 0.1398, 0.0874, 0.1785, 0.1736, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0449, 0.0353, 0.0402, 0.0431, 0.0402, 0.0366, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:19:54,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-15 21:20:17,478 INFO [finetune.py:992] (1/2) Epoch 3, batch 4300, loss[loss=0.1857, simple_loss=0.2726, pruned_loss=0.04942, over 12344.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2625, pruned_loss=0.04404, over 2371539.60 frames. ], batch size: 36, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:20:18,279 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128769.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:20:19,660 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2223, 6.1137, 5.9892, 5.5062, 5.2619, 6.1189, 5.6613, 5.4939], device='cuda:1'), covar=tensor([0.0596, 0.0883, 0.0560, 0.1419, 0.0612, 0.0672, 0.1513, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0501, 0.0469, 0.0578, 0.0380, 0.0650, 0.0714, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:20:26,787 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128781.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:20:39,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 21:20:52,968 INFO [finetune.py:992] (1/2) Epoch 3, batch 4350, loss[loss=0.1796, simple_loss=0.2756, pruned_loss=0.04186, over 12046.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.262, pruned_loss=0.04394, over 2375461.69 frames. ], batch size: 42, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:20:55,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 2023-05-15 21:20:55,189 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.239e+02 2.702e+02 3.202e+02 3.965e+02 7.883e+02, threshold=6.404e+02, percent-clipped=2.0 2023-05-15 21:20:56,801 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6550, 2.7362, 4.3832, 4.6559, 3.0463, 2.6818, 2.8600, 2.0606], device='cuda:1'), covar=tensor([0.1294, 0.2816, 0.0468, 0.0377, 0.1012, 0.1906, 0.2508, 0.3611], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0362, 0.0258, 0.0280, 0.0247, 0.0273, 0.0341, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:21:01,047 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1489, 2.6349, 3.7321, 3.1633, 3.5448, 3.1947, 2.4484, 3.6234], device='cuda:1'), covar=tensor([0.0118, 0.0277, 0.0121, 0.0192, 0.0122, 0.0149, 0.0332, 0.0098], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0192, 0.0167, 0.0176, 0.0192, 0.0148, 0.0181, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:21:10,218 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128842.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:21:28,677 INFO [finetune.py:992] (1/2) Epoch 3, batch 4400, loss[loss=0.1636, simple_loss=0.2583, pruned_loss=0.03443, over 12153.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2626, pruned_loss=0.04379, over 2382874.39 frames. ], batch size: 36, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:21:45,244 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128890.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:22:02,347 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4710, 2.2250, 3.2083, 4.2475, 2.4055, 4.4640, 4.3777, 4.5867], device='cuda:1'), covar=tensor([0.0162, 0.1206, 0.0430, 0.0190, 0.1140, 0.0171, 0.0146, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0200, 0.0179, 0.0113, 0.0182, 0.0171, 0.0163, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:22:05,702 INFO [finetune.py:992] (1/2) Epoch 3, batch 4450, loss[loss=0.1614, simple_loss=0.2453, pruned_loss=0.03876, over 12174.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2629, pruned_loss=0.04399, over 2371951.34 frames. ], batch size: 31, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:22:07,804 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.215e+02 3.079e+02 3.640e+02 4.177e+02 8.310e+02, threshold=7.280e+02, percent-clipped=2.0 2023-05-15 21:22:10,092 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5899, 5.4164, 5.5158, 5.5861, 5.1736, 5.1945, 4.9977, 5.4674], device='cuda:1'), covar=tensor([0.0567, 0.0472, 0.0633, 0.0474, 0.1519, 0.1240, 0.0497, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0617, 0.0526, 0.0578, 0.0757, 0.0687, 0.0505, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 21:22:29,542 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128951.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 21:22:41,333 INFO [finetune.py:992] (1/2) Epoch 3, batch 4500, loss[loss=0.154, simple_loss=0.25, pruned_loss=0.02904, over 12307.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2617, pruned_loss=0.0435, over 2373289.69 frames. ], batch size: 34, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:22:46,904 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128976.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:23:02,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 21:23:17,342 INFO [finetune.py:992] (1/2) Epoch 3, batch 4550, loss[loss=0.1916, simple_loss=0.2856, pruned_loss=0.04884, over 12354.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04457, over 2364708.02 frames. ], batch size: 36, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:23:19,513 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 3.068e+02 3.552e+02 4.377e+02 9.495e+02, threshold=7.104e+02, percent-clipped=2.0 2023-05-15 21:23:21,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-15 21:23:21,787 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129024.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:23:53,376 INFO [finetune.py:992] (1/2) Epoch 3, batch 4600, loss[loss=0.1752, simple_loss=0.2672, pruned_loss=0.04162, over 12270.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2637, pruned_loss=0.04442, over 2367521.54 frames. ], batch size: 33, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:23:54,161 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129069.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:24:04,858 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9292, 3.5193, 5.2415, 2.7970, 2.9259, 3.9499, 3.5623, 4.0588], device='cuda:1'), covar=tensor([0.0409, 0.1002, 0.0205, 0.1144, 0.1752, 0.1202, 0.1140, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0225, 0.0228, 0.0175, 0.0231, 0.0274, 0.0221, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:24:28,065 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129117.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:24:28,707 INFO [finetune.py:992] (1/2) Epoch 3, batch 4650, loss[loss=0.1858, simple_loss=0.2851, pruned_loss=0.04331, over 12154.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2638, pruned_loss=0.04435, over 2373660.89 frames. ], batch size: 36, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:24:30,890 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 2.760e+02 3.230e+02 3.906e+02 6.042e+02, threshold=6.460e+02, percent-clipped=0.0 2023-05-15 21:24:42,440 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129137.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:24:50,115 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0843, 6.0107, 5.8065, 5.4132, 5.1340, 5.9540, 5.5246, 5.2583], device='cuda:1'), covar=tensor([0.0599, 0.0800, 0.0586, 0.1365, 0.0618, 0.0622, 0.1480, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0563, 0.0500, 0.0465, 0.0576, 0.0379, 0.0648, 0.0708, 0.0523], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:24:51,396 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1725, 6.0907, 5.6146, 5.6673, 6.1533, 5.4181, 5.7827, 5.6293], device='cuda:1'), covar=tensor([0.1448, 0.0883, 0.1031, 0.1894, 0.0851, 0.2085, 0.1496, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0458, 0.0361, 0.0414, 0.0441, 0.0413, 0.0373, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:24:59,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-15 21:25:04,894 INFO [finetune.py:992] (1/2) Epoch 3, batch 4700, loss[loss=0.1853, simple_loss=0.2793, pruned_loss=0.04566, over 12389.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2633, pruned_loss=0.04418, over 2382094.00 frames. ], batch size: 38, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:25:41,425 INFO [finetune.py:992] (1/2) Epoch 3, batch 4750, loss[loss=0.1573, simple_loss=0.2414, pruned_loss=0.0366, over 12090.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2632, pruned_loss=0.04423, over 2379934.69 frames. ], batch size: 32, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:25:43,567 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.721e+02 3.330e+02 3.956e+02 7.224e+02, threshold=6.660e+02, percent-clipped=1.0 2023-05-15 21:26:01,325 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129246.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 21:26:16,858 INFO [finetune.py:992] (1/2) Epoch 3, batch 4800, loss[loss=0.1833, simple_loss=0.2794, pruned_loss=0.04354, over 12338.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2632, pruned_loss=0.04401, over 2385146.91 frames. ], batch size: 35, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:26:22,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9965, 3.1729, 5.3973, 2.7362, 2.5533, 4.2470, 3.1138, 4.1588], device='cuda:1'), covar=tensor([0.0336, 0.1211, 0.0210, 0.1200, 0.1988, 0.1105, 0.1423, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0228, 0.0230, 0.0177, 0.0233, 0.0277, 0.0224, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:26:43,883 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3143, 3.2567, 3.3370, 3.6723, 2.7386, 3.2689, 2.5482, 3.1379], device='cuda:1'), covar=tensor([0.1265, 0.0638, 0.0749, 0.0565, 0.0827, 0.0603, 0.1424, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0257, 0.0290, 0.0344, 0.0234, 0.0233, 0.0253, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:26:52,929 INFO [finetune.py:992] (1/2) Epoch 3, batch 4850, loss[loss=0.1904, simple_loss=0.2795, pruned_loss=0.05068, over 12359.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2632, pruned_loss=0.04413, over 2383672.04 frames. ], batch size: 35, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:26:54,621 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1236, 2.4659, 3.7619, 2.9988, 3.4401, 3.1046, 2.3815, 3.6241], device='cuda:1'), covar=tensor([0.0110, 0.0309, 0.0127, 0.0217, 0.0124, 0.0179, 0.0339, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0190, 0.0166, 0.0172, 0.0190, 0.0146, 0.0180, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:26:55,086 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.802e+02 3.179e+02 3.791e+02 1.091e+03, threshold=6.357e+02, percent-clipped=2.0 2023-05-15 21:26:56,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-15 21:27:08,687 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9605, 5.9369, 5.7172, 5.2515, 5.0932, 5.8560, 5.3967, 5.1867], device='cuda:1'), covar=tensor([0.0691, 0.0824, 0.0602, 0.1355, 0.0573, 0.0655, 0.1634, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0499, 0.0465, 0.0577, 0.0378, 0.0650, 0.0710, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:27:17,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-15 21:27:29,163 INFO [finetune.py:992] (1/2) Epoch 3, batch 4900, loss[loss=0.1924, simple_loss=0.2796, pruned_loss=0.05262, over 11517.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2628, pruned_loss=0.0444, over 2373291.68 frames. ], batch size: 48, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:27:46,937 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4199, 4.8733, 2.9767, 2.5760, 4.2889, 2.4994, 4.0912, 3.4219], device='cuda:1'), covar=tensor([0.0608, 0.0445, 0.1061, 0.1472, 0.0217, 0.1313, 0.0396, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0245, 0.0174, 0.0196, 0.0135, 0.0177, 0.0190, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:27:47,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-15 21:28:01,354 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129413.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:28:04,576 INFO [finetune.py:992] (1/2) Epoch 3, batch 4950, loss[loss=0.1457, simple_loss=0.2248, pruned_loss=0.03327, over 12029.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2623, pruned_loss=0.04417, over 2378353.18 frames. ], batch size: 28, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:28:07,404 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 2.853e+02 3.450e+02 4.427e+02 1.611e+03, threshold=6.900e+02, percent-clipped=5.0 2023-05-15 21:28:18,201 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129437.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:28:41,015 INFO [finetune.py:992] (1/2) Epoch 3, batch 5000, loss[loss=0.2063, simple_loss=0.2836, pruned_loss=0.06451, over 8068.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2617, pruned_loss=0.04376, over 2375042.17 frames. ], batch size: 99, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:28:45,508 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129474.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 21:28:53,902 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129485.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:28:54,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 21:29:15,473 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4249, 2.2559, 3.2324, 4.2245, 2.4260, 4.3785, 4.3077, 4.5237], device='cuda:1'), covar=tensor([0.0100, 0.1125, 0.0387, 0.0156, 0.1064, 0.0181, 0.0130, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0197, 0.0176, 0.0111, 0.0180, 0.0168, 0.0160, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:29:17,366 INFO [finetune.py:992] (1/2) Epoch 3, batch 5050, loss[loss=0.1619, simple_loss=0.2371, pruned_loss=0.0434, over 11985.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2612, pruned_loss=0.04348, over 2379543.66 frames. ], batch size: 28, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:29:19,528 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0562, 5.9584, 5.5017, 5.5607, 6.0434, 5.3902, 5.4838, 5.5447], device='cuda:1'), covar=tensor([0.1345, 0.0883, 0.0944, 0.1673, 0.0930, 0.2100, 0.1977, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0330, 0.0454, 0.0357, 0.0411, 0.0434, 0.0412, 0.0368, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:29:20,129 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.734e+02 3.226e+02 3.747e+02 5.556e+02, threshold=6.452e+02, percent-clipped=0.0 2023-05-15 21:29:37,807 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129546.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 21:29:53,342 INFO [finetune.py:992] (1/2) Epoch 3, batch 5100, loss[loss=0.1982, simple_loss=0.289, pruned_loss=0.05367, over 11262.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2611, pruned_loss=0.04352, over 2378512.10 frames. ], batch size: 55, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:30:04,727 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2548, 4.4569, 2.7726, 2.5164, 3.8772, 2.3482, 3.8457, 3.1583], device='cuda:1'), covar=tensor([0.0609, 0.0487, 0.0976, 0.1377, 0.0241, 0.1322, 0.0430, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0248, 0.0175, 0.0198, 0.0137, 0.0179, 0.0193, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:30:12,239 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129594.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:30:25,269 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129612.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:30:29,420 INFO [finetune.py:992] (1/2) Epoch 3, batch 5150, loss[loss=0.1637, simple_loss=0.237, pruned_loss=0.04516, over 12339.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2628, pruned_loss=0.04352, over 2379169.84 frames. ], batch size: 30, lr: 4.90e-03, grad_scale: 4.0 2023-05-15 21:30:32,846 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.725e+02 3.094e+02 3.822e+02 6.071e+02, threshold=6.187e+02, percent-clipped=0.0 2023-05-15 21:31:05,840 INFO [finetune.py:992] (1/2) Epoch 3, batch 5200, loss[loss=0.1619, simple_loss=0.2458, pruned_loss=0.03897, over 12193.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2634, pruned_loss=0.04367, over 2377423.78 frames. ], batch size: 31, lr: 4.90e-03, grad_scale: 8.0 2023-05-15 21:31:09,587 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129673.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:31:28,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-15 21:31:41,155 INFO [finetune.py:992] (1/2) Epoch 3, batch 5250, loss[loss=0.1896, simple_loss=0.2778, pruned_loss=0.05065, over 12006.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.263, pruned_loss=0.04378, over 2382390.28 frames. ], batch size: 40, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:31:43,947 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.785e+02 3.303e+02 4.067e+02 7.536e+02, threshold=6.606e+02, percent-clipped=4.0 2023-05-15 21:32:17,404 INFO [finetune.py:992] (1/2) Epoch 3, batch 5300, loss[loss=0.1729, simple_loss=0.27, pruned_loss=0.03795, over 12147.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04395, over 2372378.10 frames. ], batch size: 34, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:32:18,207 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129769.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:32:34,694 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129791.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:32:53,844 INFO [finetune.py:992] (1/2) Epoch 3, batch 5350, loss[loss=0.1827, simple_loss=0.2745, pruned_loss=0.04547, over 12105.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.0446, over 2365235.91 frames. ], batch size: 38, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:32:56,577 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.064e+02 2.918e+02 3.504e+02 4.192e+02 1.918e+03, threshold=7.007e+02, percent-clipped=5.0 2023-05-15 21:33:13,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1549, 5.8477, 5.4405, 5.4530, 5.9857, 5.2401, 5.5488, 5.5227], device='cuda:1'), covar=tensor([0.1190, 0.0809, 0.0875, 0.1942, 0.0841, 0.2074, 0.1482, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0455, 0.0361, 0.0415, 0.0437, 0.0416, 0.0368, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:33:18,131 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129852.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:33:23,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-15 21:33:29,123 INFO [finetune.py:992] (1/2) Epoch 3, batch 5400, loss[loss=0.1573, simple_loss=0.2381, pruned_loss=0.03825, over 11780.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2633, pruned_loss=0.04441, over 2371277.67 frames. ], batch size: 26, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:33:40,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.24 vs. limit=5.0 2023-05-15 21:33:53,727 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8200, 2.8681, 4.7422, 4.9871, 3.0593, 2.8911, 3.0476, 2.3044], device='cuda:1'), covar=tensor([0.1252, 0.2745, 0.0400, 0.0329, 0.1079, 0.1807, 0.2471, 0.3404], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0370, 0.0264, 0.0285, 0.0251, 0.0278, 0.0345, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:34:06,043 INFO [finetune.py:992] (1/2) Epoch 3, batch 5450, loss[loss=0.2024, simple_loss=0.2837, pruned_loss=0.06053, over 8565.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2646, pruned_loss=0.04456, over 2360748.59 frames. ], batch size: 98, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:34:08,846 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.118e+02 2.953e+02 3.315e+02 4.012e+02 8.490e+02, threshold=6.630e+02, percent-clipped=3.0 2023-05-15 21:34:38,745 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7344, 2.5338, 3.4378, 4.5884, 2.8100, 4.6342, 4.6235, 4.8579], device='cuda:1'), covar=tensor([0.0099, 0.1105, 0.0375, 0.0106, 0.0938, 0.0166, 0.0124, 0.0057], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0203, 0.0182, 0.0113, 0.0185, 0.0174, 0.0165, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:34:41,407 INFO [finetune.py:992] (1/2) Epoch 3, batch 5500, loss[loss=0.1608, simple_loss=0.2508, pruned_loss=0.03538, over 12342.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2636, pruned_loss=0.04416, over 2367335.57 frames. ], batch size: 31, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:34:41,488 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129968.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:34:41,581 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2117, 2.0420, 2.6770, 3.1796, 2.1667, 3.3029, 3.1723, 3.3214], device='cuda:1'), covar=tensor([0.0139, 0.0992, 0.0404, 0.0149, 0.0936, 0.0243, 0.0267, 0.0103], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0203, 0.0182, 0.0113, 0.0185, 0.0173, 0.0165, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:35:07,662 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130000.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:35:09,109 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130002.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:35:20,235 INFO [finetune.py:992] (1/2) Epoch 3, batch 5550, loss[loss=0.1889, simple_loss=0.2806, pruned_loss=0.04862, over 11105.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2643, pruned_loss=0.04477, over 2356249.62 frames. ], batch size: 55, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:35:23,116 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 2.826e+02 3.312e+02 3.995e+02 7.380e+02, threshold=6.624e+02, percent-clipped=1.0 2023-05-15 21:35:44,925 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2407, 2.5029, 3.8194, 3.1225, 3.5396, 3.2814, 2.5158, 3.6987], device='cuda:1'), covar=tensor([0.0108, 0.0310, 0.0106, 0.0200, 0.0118, 0.0153, 0.0306, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0190, 0.0167, 0.0171, 0.0191, 0.0146, 0.0180, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:35:45,609 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1973, 4.5792, 2.6982, 2.4043, 3.8729, 2.3423, 3.8397, 2.9388], device='cuda:1'), covar=tensor([0.0701, 0.0484, 0.1201, 0.1537, 0.0273, 0.1458, 0.0476, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0249, 0.0176, 0.0196, 0.0138, 0.0179, 0.0193, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:35:51,812 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130061.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:35:53,256 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130063.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 21:35:56,621 INFO [finetune.py:992] (1/2) Epoch 3, batch 5600, loss[loss=0.1576, simple_loss=0.2378, pruned_loss=0.03873, over 12006.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2634, pruned_loss=0.0445, over 2361024.29 frames. ], batch size: 28, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:35:57,491 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130069.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:36:08,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-15 21:36:18,097 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5661, 5.0122, 5.5162, 4.8217, 5.0737, 4.8552, 5.5096, 5.1708], device='cuda:1'), covar=tensor([0.0189, 0.0350, 0.0236, 0.0226, 0.0288, 0.0305, 0.0202, 0.0243], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0240, 0.0259, 0.0235, 0.0234, 0.0235, 0.0215, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:36:31,267 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130117.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 21:36:31,841 INFO [finetune.py:992] (1/2) Epoch 3, batch 5650, loss[loss=0.1798, simple_loss=0.2747, pruned_loss=0.04246, over 12120.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2641, pruned_loss=0.0449, over 2355152.83 frames. ], batch size: 39, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:36:34,700 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.738e+02 3.427e+02 4.073e+02 1.202e+03, threshold=6.854e+02, percent-clipped=1.0 2023-05-15 21:36:52,383 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130147.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:36:59,136 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-15 21:37:03,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2918, 4.7055, 2.8139, 2.5531, 3.9972, 2.4633, 4.0094, 3.1526], device='cuda:1'), covar=tensor([0.0589, 0.0344, 0.1111, 0.1408, 0.0279, 0.1281, 0.0381, 0.0718], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0246, 0.0174, 0.0195, 0.0136, 0.0178, 0.0191, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:37:07,261 INFO [finetune.py:992] (1/2) Epoch 3, batch 5700, loss[loss=0.1943, simple_loss=0.2831, pruned_loss=0.0528, over 11224.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2642, pruned_loss=0.04468, over 2366286.39 frames. ], batch size: 55, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:37:21,745 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5444, 2.2499, 3.3781, 4.3323, 2.3660, 4.4726, 4.4138, 4.6292], device='cuda:1'), covar=tensor([0.0111, 0.1162, 0.0376, 0.0118, 0.1109, 0.0151, 0.0113, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0200, 0.0181, 0.0113, 0.0183, 0.0172, 0.0163, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:37:44,815 INFO [finetune.py:992] (1/2) Epoch 3, batch 5750, loss[loss=0.1588, simple_loss=0.2513, pruned_loss=0.03316, over 12087.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2641, pruned_loss=0.04436, over 2371751.40 frames. ], batch size: 32, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:37:47,430 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.153e+02 3.005e+02 3.316e+02 3.944e+02 6.494e+02, threshold=6.631e+02, percent-clipped=0.0 2023-05-15 21:38:20,129 INFO [finetune.py:992] (1/2) Epoch 3, batch 5800, loss[loss=0.1666, simple_loss=0.2605, pruned_loss=0.03638, over 12013.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2637, pruned_loss=0.04429, over 2373480.63 frames. ], batch size: 40, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:38:20,236 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130268.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:38:28,865 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6935, 2.8801, 4.5973, 4.8379, 2.8128, 2.7266, 2.8714, 2.1650], device='cuda:1'), covar=tensor([0.1280, 0.2817, 0.0390, 0.0334, 0.1171, 0.1910, 0.2470, 0.3566], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0374, 0.0265, 0.0288, 0.0254, 0.0280, 0.0347, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:38:39,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-05-15 21:38:49,559 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1860, 4.6831, 5.1579, 4.4859, 4.7136, 4.5356, 5.1527, 4.8941], device='cuda:1'), covar=tensor([0.0263, 0.0373, 0.0265, 0.0265, 0.0361, 0.0358, 0.0231, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0239, 0.0258, 0.0233, 0.0233, 0.0233, 0.0213, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:38:54,446 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130316.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:38:55,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-05-15 21:38:55,828 INFO [finetune.py:992] (1/2) Epoch 3, batch 5850, loss[loss=0.1881, simple_loss=0.2748, pruned_loss=0.05073, over 12126.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2649, pruned_loss=0.04533, over 2360997.73 frames. ], batch size: 38, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:38:58,542 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.062e+02 2.888e+02 3.485e+02 4.475e+02 7.584e+02, threshold=6.969e+02, percent-clipped=2.0 2023-05-15 21:39:16,420 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1236, 5.0983, 4.9224, 5.0643, 4.5815, 5.1190, 5.1497, 5.4141], device='cuda:1'), covar=tensor([0.0186, 0.0135, 0.0192, 0.0247, 0.0844, 0.0224, 0.0130, 0.0130], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0179, 0.0180, 0.0224, 0.0228, 0.0198, 0.0165, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 21:39:23,959 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130356.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:39:25,434 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130358.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 21:39:32,432 INFO [finetune.py:992] (1/2) Epoch 3, batch 5900, loss[loss=0.2572, simple_loss=0.3075, pruned_loss=0.1034, over 7605.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2647, pruned_loss=0.04542, over 2361486.66 frames. ], batch size: 98, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:39:43,804 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130384.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:40:08,138 INFO [finetune.py:992] (1/2) Epoch 3, batch 5950, loss[loss=0.1708, simple_loss=0.2516, pruned_loss=0.04504, over 11795.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2646, pruned_loss=0.04575, over 2358346.17 frames. ], batch size: 26, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:40:10,519 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8728, 3.3926, 5.0892, 2.4754, 2.8511, 3.8336, 3.0882, 3.9622], device='cuda:1'), covar=tensor([0.0332, 0.1074, 0.0339, 0.1162, 0.1790, 0.1441, 0.1386, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0226, 0.0229, 0.0175, 0.0230, 0.0275, 0.0221, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:40:10,940 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.936e+02 3.497e+02 4.249e+02 9.499e+02, threshold=6.995e+02, percent-clipped=1.0 2023-05-15 21:40:18,354 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130432.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:40:22,588 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130438.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:40:27,585 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130445.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:40:28,992 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130447.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:40:43,774 INFO [finetune.py:992] (1/2) Epoch 3, batch 6000, loss[loss=0.1635, simple_loss=0.2513, pruned_loss=0.03778, over 12078.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2654, pruned_loss=0.04518, over 2362482.86 frames. ], batch size: 32, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:40:43,774 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 21:41:01,522 INFO [finetune.py:1026] (1/2) Epoch 3, validation: loss=0.3355, simple_loss=0.4069, pruned_loss=0.1321, over 1020973.00 frames. 2023-05-15 21:41:01,523 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 21:41:12,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-15 21:41:19,530 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130493.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:41:20,829 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130495.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:41:23,862 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130499.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:41:37,112 INFO [finetune.py:992] (1/2) Epoch 3, batch 6050, loss[loss=0.2666, simple_loss=0.3314, pruned_loss=0.1009, over 8361.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.265, pruned_loss=0.04508, over 2358942.20 frames. ], batch size: 97, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:41:39,970 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 2.959e+02 3.443e+02 4.363e+02 1.639e+03, threshold=6.886e+02, percent-clipped=4.0 2023-05-15 21:41:58,641 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130548.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:42:04,286 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3404, 3.4542, 3.1433, 3.1191, 2.8174, 2.5839, 3.4543, 2.1502], device='cuda:1'), covar=tensor([0.0354, 0.0098, 0.0144, 0.0138, 0.0317, 0.0305, 0.0102, 0.0427], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0156, 0.0148, 0.0175, 0.0200, 0.0190, 0.0155, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:42:09,402 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4346, 4.8769, 3.0884, 2.9212, 4.1562, 2.7094, 4.1425, 3.3648], device='cuda:1'), covar=tensor([0.0667, 0.0435, 0.1037, 0.1250, 0.0318, 0.1256, 0.0418, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0245, 0.0173, 0.0194, 0.0136, 0.0177, 0.0190, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:42:10,767 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130565.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:42:12,582 INFO [finetune.py:992] (1/2) Epoch 3, batch 6100, loss[loss=0.176, simple_loss=0.2674, pruned_loss=0.04228, over 11830.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2644, pruned_loss=0.04514, over 2364060.96 frames. ], batch size: 44, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:42:37,719 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5352, 2.3405, 3.3509, 4.4202, 2.2873, 4.4795, 4.4387, 4.6256], device='cuda:1'), covar=tensor([0.0078, 0.1147, 0.0419, 0.0110, 0.1226, 0.0163, 0.0142, 0.0056], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0199, 0.0181, 0.0112, 0.0182, 0.0170, 0.0163, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:42:43,464 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130609.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 21:42:49,482 INFO [finetune.py:992] (1/2) Epoch 3, batch 6150, loss[loss=0.2021, simple_loss=0.2885, pruned_loss=0.05783, over 12118.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2635, pruned_loss=0.04483, over 2369241.17 frames. ], batch size: 39, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:42:52,329 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.993e+02 3.524e+02 4.331e+02 7.227e+02, threshold=7.047e+02, percent-clipped=1.0 2023-05-15 21:42:55,365 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130626.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:42:58,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5805, 2.9372, 3.9037, 2.3322, 2.6413, 3.1314, 2.8752, 3.2943], device='cuda:1'), covar=tensor([0.0462, 0.0914, 0.0300, 0.1110, 0.1450, 0.1236, 0.1086, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0225, 0.0227, 0.0174, 0.0229, 0.0273, 0.0220, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:43:03,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-15 21:43:16,728 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130656.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:43:18,022 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130658.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 21:43:20,889 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4587, 4.7875, 2.9751, 2.7753, 4.0876, 2.7894, 3.9867, 3.3606], device='cuda:1'), covar=tensor([0.0591, 0.0446, 0.1012, 0.1372, 0.0275, 0.1095, 0.0492, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0244, 0.0173, 0.0194, 0.0135, 0.0176, 0.0191, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:43:24,952 INFO [finetune.py:992] (1/2) Epoch 3, batch 6200, loss[loss=0.1735, simple_loss=0.2614, pruned_loss=0.04275, over 12365.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2639, pruned_loss=0.04513, over 2359662.74 frames. ], batch size: 36, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:43:50,272 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130704.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:43:51,744 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130706.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:44:00,969 INFO [finetune.py:992] (1/2) Epoch 3, batch 6250, loss[loss=0.1751, simple_loss=0.2733, pruned_loss=0.03843, over 12343.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2631, pruned_loss=0.04452, over 2372321.70 frames. ], batch size: 36, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:44:03,770 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.911e+02 3.273e+02 3.794e+02 8.028e+02, threshold=6.546e+02, percent-clipped=1.0 2023-05-15 21:44:16,540 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130740.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:44:18,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-15 21:44:34,330 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3922, 3.6595, 3.2176, 3.1692, 2.7577, 2.6202, 3.6381, 2.1441], device='cuda:1'), covar=tensor([0.0335, 0.0098, 0.0140, 0.0158, 0.0377, 0.0335, 0.0105, 0.0482], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0157, 0.0150, 0.0177, 0.0202, 0.0192, 0.0156, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:44:36,993 INFO [finetune.py:992] (1/2) Epoch 3, batch 6300, loss[loss=0.1692, simple_loss=0.2624, pruned_loss=0.038, over 11225.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2637, pruned_loss=0.04518, over 2360348.93 frames. ], batch size: 55, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:44:51,363 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130788.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:44:53,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.99 vs. limit=5.0 2023-05-15 21:44:55,556 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130794.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:45:12,714 INFO [finetune.py:992] (1/2) Epoch 3, batch 6350, loss[loss=0.1802, simple_loss=0.2713, pruned_loss=0.04453, over 12307.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2637, pruned_loss=0.04479, over 2376015.05 frames. ], batch size: 34, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:45:15,395 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.159e+02 2.855e+02 3.388e+02 4.377e+02 6.922e+02, threshold=6.775e+02, percent-clipped=1.0 2023-05-15 21:45:32,039 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2634, 4.3535, 2.7114, 2.3112, 3.7927, 2.4133, 3.8249, 2.9350], device='cuda:1'), covar=tensor([0.0631, 0.0595, 0.1052, 0.1444, 0.0297, 0.1279, 0.0472, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0245, 0.0172, 0.0195, 0.0136, 0.0177, 0.0191, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 21:45:41,867 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9570, 4.9148, 4.8309, 4.9773, 3.8878, 5.0019, 5.0235, 5.1517], device='cuda:1'), covar=tensor([0.0263, 0.0184, 0.0208, 0.0287, 0.1174, 0.0275, 0.0163, 0.0206], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0179, 0.0179, 0.0225, 0.0227, 0.0197, 0.0165, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 21:45:48,722 INFO [finetune.py:992] (1/2) Epoch 3, batch 6400, loss[loss=0.1785, simple_loss=0.2699, pruned_loss=0.04352, over 12173.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2646, pruned_loss=0.04497, over 2376066.88 frames. ], batch size: 31, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:46:15,221 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130904.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 21:46:17,553 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5296, 2.4485, 4.4868, 4.8003, 3.1176, 2.5053, 2.9282, 1.7377], device='cuda:1'), covar=tensor([0.1460, 0.3465, 0.0392, 0.0301, 0.0936, 0.2078, 0.2405, 0.4497], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0369, 0.0263, 0.0285, 0.0250, 0.0276, 0.0344, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:46:25,309 INFO [finetune.py:992] (1/2) Epoch 3, batch 6450, loss[loss=0.1399, simple_loss=0.2283, pruned_loss=0.0258, over 12186.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2638, pruned_loss=0.04415, over 2384565.11 frames. ], batch size: 29, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:46:27,456 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130921.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:46:28,032 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.906e+02 2.956e+02 3.435e+02 4.132e+02 8.465e+02, threshold=6.870e+02, percent-clipped=2.0 2023-05-15 21:46:31,694 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2480, 5.1730, 5.0275, 5.0634, 4.7806, 5.1349, 5.1857, 5.3817], device='cuda:1'), covar=tensor([0.0139, 0.0115, 0.0157, 0.0257, 0.0631, 0.0249, 0.0118, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0180, 0.0179, 0.0225, 0.0228, 0.0199, 0.0166, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 21:47:00,295 INFO [finetune.py:992] (1/2) Epoch 3, batch 6500, loss[loss=0.2139, simple_loss=0.2967, pruned_loss=0.06552, over 12059.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2635, pruned_loss=0.04424, over 2383442.63 frames. ], batch size: 37, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:47:00,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.62 vs. limit=5.0 2023-05-15 21:47:23,590 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3503, 4.2024, 4.1307, 4.4766, 3.1291, 3.9512, 2.5800, 3.9990], device='cuda:1'), covar=tensor([0.1510, 0.0628, 0.1029, 0.0652, 0.1070, 0.0616, 0.1811, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0256, 0.0291, 0.0347, 0.0236, 0.0235, 0.0252, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:47:36,999 INFO [finetune.py:992] (1/2) Epoch 3, batch 6550, loss[loss=0.1893, simple_loss=0.2932, pruned_loss=0.04268, over 12124.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04368, over 2379037.50 frames. ], batch size: 36, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:47:39,229 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2525, 5.1730, 5.0301, 5.1004, 4.7826, 5.1585, 5.1968, 5.4779], device='cuda:1'), covar=tensor([0.0160, 0.0115, 0.0171, 0.0240, 0.0666, 0.0260, 0.0133, 0.0125], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0181, 0.0181, 0.0228, 0.0231, 0.0200, 0.0167, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 21:47:39,771 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.885e+02 3.348e+02 4.078e+02 8.437e+02, threshold=6.696e+02, percent-clipped=3.0 2023-05-15 21:47:53,153 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131040.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:48:12,997 INFO [finetune.py:992] (1/2) Epoch 3, batch 6600, loss[loss=0.1718, simple_loss=0.2666, pruned_loss=0.03856, over 12341.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2632, pruned_loss=0.04367, over 2377157.13 frames. ], batch size: 35, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:48:27,152 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131088.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:48:27,271 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131088.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:48:31,394 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131094.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:48:48,249 INFO [finetune.py:992] (1/2) Epoch 3, batch 6650, loss[loss=0.163, simple_loss=0.2568, pruned_loss=0.03459, over 12152.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.264, pruned_loss=0.04442, over 2374103.63 frames. ], batch size: 36, lr: 4.89e-03, grad_scale: 8.0 2023-05-15 21:48:51,061 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 3.035e+02 3.644e+02 4.405e+02 1.056e+03, threshold=7.288e+02, percent-clipped=5.0 2023-05-15 21:48:58,617 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-15 21:48:59,119 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0405, 5.9729, 5.7297, 5.3787, 5.1405, 5.8961, 5.5035, 5.2706], device='cuda:1'), covar=tensor([0.0664, 0.0923, 0.0628, 0.1405, 0.0586, 0.0657, 0.1489, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0505, 0.0462, 0.0577, 0.0377, 0.0654, 0.0711, 0.0524], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:49:01,287 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131136.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:49:05,549 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131142.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:49:24,680 INFO [finetune.py:992] (1/2) Epoch 3, batch 6700, loss[loss=0.1829, simple_loss=0.2767, pruned_loss=0.04459, over 12024.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.263, pruned_loss=0.04376, over 2378944.98 frames. ], batch size: 42, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:49:29,157 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131174.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:49:43,388 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1599, 6.1159, 5.9035, 5.4373, 5.1391, 6.0055, 5.6026, 5.4358], device='cuda:1'), covar=tensor([0.0534, 0.0691, 0.0513, 0.1408, 0.0649, 0.0645, 0.1272, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0569, 0.0507, 0.0465, 0.0579, 0.0378, 0.0655, 0.0713, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 21:49:51,660 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131204.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 21:50:01,491 INFO [finetune.py:992] (1/2) Epoch 3, batch 6750, loss[loss=0.2161, simple_loss=0.3035, pruned_loss=0.06436, over 10574.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2635, pruned_loss=0.04382, over 2374895.22 frames. ], batch size: 68, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:50:03,666 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131221.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:50:04,170 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.729e+02 3.205e+02 3.799e+02 1.174e+03, threshold=6.410e+02, percent-clipped=3.0 2023-05-15 21:50:13,644 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131235.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 21:50:25,474 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131252.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:50:36,909 INFO [finetune.py:992] (1/2) Epoch 3, batch 6800, loss[loss=0.1661, simple_loss=0.2506, pruned_loss=0.04081, over 12351.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2643, pruned_loss=0.04423, over 2367749.44 frames. ], batch size: 31, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:50:37,596 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131269.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:50:52,366 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3765, 2.4014, 3.6757, 4.2619, 4.0066, 4.1605, 3.7490, 3.0807], device='cuda:1'), covar=tensor([0.0023, 0.0383, 0.0103, 0.0041, 0.0074, 0.0085, 0.0104, 0.0307], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0119, 0.0100, 0.0073, 0.0098, 0.0108, 0.0088, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:51:12,862 INFO [finetune.py:992] (1/2) Epoch 3, batch 6850, loss[loss=0.1736, simple_loss=0.2591, pruned_loss=0.04407, over 12091.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2653, pruned_loss=0.0449, over 2360113.57 frames. ], batch size: 32, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:51:15,648 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 3.070e+02 3.537e+02 4.360e+02 1.019e+03, threshold=7.074e+02, percent-clipped=8.0 2023-05-15 21:51:26,013 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8972, 1.8607, 3.4196, 2.8554, 3.3076, 3.1001, 2.1142, 3.3004], device='cuda:1'), covar=tensor([0.0113, 0.0440, 0.0137, 0.0208, 0.0128, 0.0142, 0.0346, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0190, 0.0169, 0.0171, 0.0192, 0.0147, 0.0180, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:51:48,833 INFO [finetune.py:992] (1/2) Epoch 3, batch 6900, loss[loss=0.1734, simple_loss=0.2649, pruned_loss=0.04097, over 12036.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2645, pruned_loss=0.04454, over 2364804.98 frames. ], batch size: 37, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:51:51,512 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-05-15 21:52:17,993 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131409.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:52:19,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.88 vs. limit=5.0 2023-05-15 21:52:21,603 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8344, 2.9699, 5.2778, 2.4427, 2.5573, 4.2245, 3.0711, 4.1044], device='cuda:1'), covar=tensor([0.0372, 0.1326, 0.0210, 0.1289, 0.2090, 0.1120, 0.1486, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0223, 0.0228, 0.0174, 0.0230, 0.0274, 0.0219, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:52:22,932 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131416.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:52:24,191 INFO [finetune.py:992] (1/2) Epoch 3, batch 6950, loss[loss=0.157, simple_loss=0.244, pruned_loss=0.03501, over 12084.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2647, pruned_loss=0.04446, over 2366457.13 frames. ], batch size: 32, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:52:27,692 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 3.023e+02 3.604e+02 4.404e+02 1.310e+03, threshold=7.207e+02, percent-clipped=2.0 2023-05-15 21:52:45,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-15 21:53:00,836 INFO [finetune.py:992] (1/2) Epoch 3, batch 7000, loss[loss=0.1948, simple_loss=0.2783, pruned_loss=0.05565, over 12090.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2642, pruned_loss=0.04466, over 2372739.05 frames. ], batch size: 33, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:53:02,411 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131470.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:53:07,408 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131477.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:53:17,940 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131492.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:53:35,921 INFO [finetune.py:992] (1/2) Epoch 3, batch 7050, loss[loss=0.1875, simple_loss=0.2797, pruned_loss=0.04766, over 11654.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04463, over 2373172.56 frames. ], batch size: 48, lr: 4.88e-03, grad_scale: 4.0 2023-05-15 21:53:40,216 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.138e+02 2.887e+02 3.480e+02 4.466e+02 8.621e+02, threshold=6.960e+02, percent-clipped=3.0 2023-05-15 21:53:44,689 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131530.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 21:53:54,764 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0369, 4.2586, 3.7221, 4.6330, 4.2292, 2.7034, 3.9202, 3.0810], device='cuda:1'), covar=tensor([0.0735, 0.0796, 0.1379, 0.0478, 0.1021, 0.1577, 0.0966, 0.2813], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0364, 0.0343, 0.0255, 0.0351, 0.0257, 0.0328, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:54:01,184 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131553.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:54:11,248 INFO [finetune.py:992] (1/2) Epoch 3, batch 7100, loss[loss=0.1865, simple_loss=0.2682, pruned_loss=0.05237, over 12258.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2639, pruned_loss=0.0449, over 2380512.74 frames. ], batch size: 32, lr: 4.88e-03, grad_scale: 4.0 2023-05-15 21:54:48,030 INFO [finetune.py:992] (1/2) Epoch 3, batch 7150, loss[loss=0.1802, simple_loss=0.2502, pruned_loss=0.0551, over 12287.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2644, pruned_loss=0.04529, over 2372271.78 frames. ], batch size: 28, lr: 4.88e-03, grad_scale: 4.0 2023-05-15 21:54:52,333 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.829e+02 3.273e+02 4.121e+02 5.730e+02, threshold=6.545e+02, percent-clipped=0.0 2023-05-15 21:55:03,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-15 21:55:23,462 INFO [finetune.py:992] (1/2) Epoch 3, batch 7200, loss[loss=0.1767, simple_loss=0.2697, pruned_loss=0.04184, over 11747.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2647, pruned_loss=0.04552, over 2373818.89 frames. ], batch size: 44, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:55:58,541 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4809, 5.2840, 5.4361, 5.4574, 5.0738, 5.0689, 4.8251, 5.4133], device='cuda:1'), covar=tensor([0.0624, 0.0533, 0.0610, 0.0541, 0.1615, 0.1188, 0.0603, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0633, 0.0541, 0.0591, 0.0772, 0.0706, 0.0516, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 21:55:59,063 INFO [finetune.py:992] (1/2) Epoch 3, batch 7250, loss[loss=0.17, simple_loss=0.2624, pruned_loss=0.03876, over 12341.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2653, pruned_loss=0.04562, over 2374463.68 frames. ], batch size: 31, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:55:59,972 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6057, 5.4565, 5.5806, 5.6122, 5.2137, 5.2121, 4.9931, 5.5542], device='cuda:1'), covar=tensor([0.0678, 0.0491, 0.0615, 0.0516, 0.1601, 0.1164, 0.0535, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0633, 0.0541, 0.0591, 0.0772, 0.0706, 0.0516, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 21:56:03,156 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.970e+02 2.906e+02 3.464e+02 4.213e+02 7.151e+02, threshold=6.928e+02, percent-clipped=2.0 2023-05-15 21:56:19,721 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3304, 3.3353, 3.0951, 3.0960, 2.7466, 2.5688, 3.3402, 1.9713], device='cuda:1'), covar=tensor([0.0368, 0.0134, 0.0175, 0.0176, 0.0332, 0.0330, 0.0127, 0.0486], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0155, 0.0149, 0.0177, 0.0202, 0.0192, 0.0156, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 21:56:33,714 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131765.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:56:35,728 INFO [finetune.py:992] (1/2) Epoch 3, batch 7300, loss[loss=0.1616, simple_loss=0.2358, pruned_loss=0.04365, over 12187.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2645, pruned_loss=0.04542, over 2376179.41 frames. ], batch size: 29, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:56:38,619 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131772.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:57:11,676 INFO [finetune.py:992] (1/2) Epoch 3, batch 7350, loss[loss=0.1768, simple_loss=0.276, pruned_loss=0.03879, over 12363.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.264, pruned_loss=0.04482, over 2380218.03 frames. ], batch size: 36, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:57:15,910 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 2.957e+02 3.600e+02 4.199e+02 8.110e+02, threshold=7.200e+02, percent-clipped=1.0 2023-05-15 21:57:20,306 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131830.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:57:33,420 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131848.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:57:39,897 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131857.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:57:47,476 INFO [finetune.py:992] (1/2) Epoch 3, batch 7400, loss[loss=0.1764, simple_loss=0.2732, pruned_loss=0.03977, over 12273.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2629, pruned_loss=0.04437, over 2374791.64 frames. ], batch size: 37, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:57:55,333 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131878.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:58:24,322 INFO [finetune.py:992] (1/2) Epoch 3, batch 7450, loss[loss=0.1788, simple_loss=0.2722, pruned_loss=0.0427, over 11877.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2633, pruned_loss=0.04497, over 2372670.81 frames. ], batch size: 44, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:58:24,538 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131918.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:58:28,562 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.924e+02 3.430e+02 4.215e+02 1.172e+03, threshold=6.861e+02, percent-clipped=1.0 2023-05-15 21:58:43,915 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-05-15 21:58:59,886 INFO [finetune.py:992] (1/2) Epoch 3, batch 7500, loss[loss=0.1683, simple_loss=0.2431, pruned_loss=0.04672, over 12005.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2637, pruned_loss=0.04501, over 2376650.49 frames. ], batch size: 28, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:59:00,145 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9980, 4.0246, 4.0025, 4.3325, 3.0606, 4.0012, 2.6202, 4.0459], device='cuda:1'), covar=tensor([0.1652, 0.0639, 0.0949, 0.0674, 0.1066, 0.0525, 0.1634, 0.1373], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0258, 0.0292, 0.0349, 0.0238, 0.0235, 0.0252, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 21:59:39,392 INFO [finetune.py:992] (1/2) Epoch 3, batch 7550, loss[loss=0.2005, simple_loss=0.2904, pruned_loss=0.05536, over 12278.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2645, pruned_loss=0.04526, over 2374678.39 frames. ], batch size: 37, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 21:59:43,565 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 3.013e+02 3.558e+02 4.169e+02 8.465e+02, threshold=7.117e+02, percent-clipped=1.0 2023-05-15 21:59:48,691 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132031.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 21:59:57,862 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1599, 4.1889, 4.1081, 4.4374, 3.1169, 4.1690, 2.8047, 4.1530], device='cuda:1'), covar=tensor([0.1543, 0.0621, 0.0879, 0.0603, 0.1072, 0.0483, 0.1539, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0256, 0.0290, 0.0348, 0.0237, 0.0234, 0.0251, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:00:05,735 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132054.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:00:13,497 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132065.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:00:15,544 INFO [finetune.py:992] (1/2) Epoch 3, batch 7600, loss[loss=0.1886, simple_loss=0.2816, pruned_loss=0.04782, over 12271.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.265, pruned_loss=0.04514, over 2375539.16 frames. ], batch size: 37, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:00:18,572 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132072.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:00:32,896 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132092.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:00:40,229 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-15 22:00:47,651 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132113.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:00:49,246 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132115.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 22:00:51,025 INFO [finetune.py:992] (1/2) Epoch 3, batch 7650, loss[loss=0.1997, simple_loss=0.2938, pruned_loss=0.05278, over 12147.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2644, pruned_loss=0.04436, over 2382047.73 frames. ], batch size: 36, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:00:52,530 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132120.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:00:55,402 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.907e+02 3.519e+02 4.376e+02 6.204e+02, threshold=7.039e+02, percent-clipped=0.0 2023-05-15 22:01:00,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-15 22:01:06,190 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2046, 4.5009, 3.9476, 4.9314, 4.3700, 2.8296, 4.0550, 3.0384], device='cuda:1'), covar=tensor([0.0731, 0.0838, 0.1309, 0.0298, 0.1039, 0.1485, 0.0972, 0.2829], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0366, 0.0343, 0.0255, 0.0353, 0.0257, 0.0327, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:01:12,524 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132148.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:01:16,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-15 22:01:27,165 INFO [finetune.py:992] (1/2) Epoch 3, batch 7700, loss[loss=0.2084, simple_loss=0.2918, pruned_loss=0.06254, over 11388.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2651, pruned_loss=0.04508, over 2379323.34 frames. ], batch size: 55, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:01:37,250 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8480, 4.5443, 4.7236, 4.7974, 4.5874, 4.8230, 4.6242, 2.8724], device='cuda:1'), covar=tensor([0.0093, 0.0064, 0.0072, 0.0054, 0.0052, 0.0083, 0.0067, 0.0571], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0072, 0.0075, 0.0068, 0.0056, 0.0084, 0.0074, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:01:47,626 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132196.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:01:59,757 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132213.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:02:03,243 INFO [finetune.py:992] (1/2) Epoch 3, batch 7750, loss[loss=0.1963, simple_loss=0.2893, pruned_loss=0.05166, over 12028.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2643, pruned_loss=0.04453, over 2382760.13 frames. ], batch size: 42, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:02:07,536 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.978e+02 3.505e+02 4.131e+02 6.274e+02, threshold=7.010e+02, percent-clipped=0.0 2023-05-15 22:02:19,892 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6166, 2.6660, 4.0999, 4.3719, 2.8688, 2.6411, 2.8210, 2.2217], device='cuda:1'), covar=tensor([0.1364, 0.2677, 0.0526, 0.0400, 0.1115, 0.1888, 0.2324, 0.3450], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0370, 0.0265, 0.0285, 0.0250, 0.0278, 0.0345, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:02:38,763 INFO [finetune.py:992] (1/2) Epoch 3, batch 7800, loss[loss=0.1537, simple_loss=0.2437, pruned_loss=0.03184, over 12183.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2635, pruned_loss=0.04429, over 2373253.92 frames. ], batch size: 31, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:03:14,555 INFO [finetune.py:992] (1/2) Epoch 3, batch 7850, loss[loss=0.1931, simple_loss=0.2889, pruned_loss=0.04868, over 12156.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2635, pruned_loss=0.04439, over 2374259.84 frames. ], batch size: 36, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:03:18,731 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.804e+02 3.373e+02 4.263e+02 8.998e+02, threshold=6.746e+02, percent-clipped=4.0 2023-05-15 22:03:45,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.54 vs. limit=2.0 2023-05-15 22:03:49,913 INFO [finetune.py:992] (1/2) Epoch 3, batch 7900, loss[loss=0.1663, simple_loss=0.2428, pruned_loss=0.04495, over 12115.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2645, pruned_loss=0.04499, over 2380126.66 frames. ], batch size: 30, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:03:55,909 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0182, 4.9697, 4.8508, 4.9304, 4.5105, 5.0173, 4.9640, 5.2167], device='cuda:1'), covar=tensor([0.0187, 0.0134, 0.0174, 0.0263, 0.0707, 0.0291, 0.0150, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0184, 0.0183, 0.0231, 0.0232, 0.0201, 0.0166, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 22:04:03,505 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132387.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:04:20,188 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132410.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 22:04:25,739 INFO [finetune.py:992] (1/2) Epoch 3, batch 7950, loss[loss=0.192, simple_loss=0.2827, pruned_loss=0.05068, over 12206.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2644, pruned_loss=0.04534, over 2374837.38 frames. ], batch size: 35, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:04:30,096 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.816e+02 3.359e+02 4.037e+02 1.124e+03, threshold=6.718e+02, percent-clipped=1.0 2023-05-15 22:04:46,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-15 22:05:01,809 INFO [finetune.py:992] (1/2) Epoch 3, batch 8000, loss[loss=0.1614, simple_loss=0.2469, pruned_loss=0.03795, over 12350.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2638, pruned_loss=0.04466, over 2382679.39 frames. ], batch size: 31, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:05:07,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-15 22:05:34,472 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132513.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:05:37,692 INFO [finetune.py:992] (1/2) Epoch 3, batch 8050, loss[loss=0.1565, simple_loss=0.2466, pruned_loss=0.03325, over 12288.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2646, pruned_loss=0.04523, over 2381678.69 frames. ], batch size: 33, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:05:41,977 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 3.034e+02 3.557e+02 4.293e+02 8.713e+02, threshold=7.114e+02, percent-clipped=4.0 2023-05-15 22:05:49,299 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8334, 2.7684, 3.9389, 4.7476, 4.2014, 4.7749, 4.1727, 3.3804], device='cuda:1'), covar=tensor([0.0020, 0.0353, 0.0108, 0.0030, 0.0100, 0.0058, 0.0075, 0.0299], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0119, 0.0101, 0.0074, 0.0099, 0.0109, 0.0087, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:06:08,228 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132561.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:06:13,220 INFO [finetune.py:992] (1/2) Epoch 3, batch 8100, loss[loss=0.1701, simple_loss=0.259, pruned_loss=0.04062, over 12126.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2647, pruned_loss=0.04544, over 2373421.38 frames. ], batch size: 33, lr: 4.88e-03, grad_scale: 8.0 2023-05-15 22:06:49,188 INFO [finetune.py:992] (1/2) Epoch 3, batch 8150, loss[loss=0.2, simple_loss=0.2866, pruned_loss=0.05673, over 12293.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2647, pruned_loss=0.04511, over 2380709.95 frames. ], batch size: 34, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:06:53,457 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.840e+02 3.244e+02 3.868e+02 7.076e+02, threshold=6.489e+02, percent-clipped=0.0 2023-05-15 22:07:19,864 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132660.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:07:25,383 INFO [finetune.py:992] (1/2) Epoch 3, batch 8200, loss[loss=0.1555, simple_loss=0.2416, pruned_loss=0.03472, over 12353.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2646, pruned_loss=0.04546, over 2377959.82 frames. ], batch size: 30, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:07:39,182 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132687.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:07:55,485 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132710.0, num_to_drop=1, layers_to_drop={2} 2023-05-15 22:08:01,008 INFO [finetune.py:992] (1/2) Epoch 3, batch 8250, loss[loss=0.3077, simple_loss=0.359, pruned_loss=0.1282, over 7784.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2649, pruned_loss=0.04568, over 2372037.86 frames. ], batch size: 98, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:08:03,359 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132721.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:08:05,195 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 2.844e+02 3.378e+02 4.059e+02 1.339e+03, threshold=6.755e+02, percent-clipped=5.0 2023-05-15 22:08:06,903 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6069, 2.7943, 3.3027, 4.4508, 2.5707, 4.5454, 4.5941, 4.7563], device='cuda:1'), covar=tensor([0.0091, 0.1002, 0.0415, 0.0120, 0.1113, 0.0156, 0.0113, 0.0055], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0199, 0.0182, 0.0114, 0.0181, 0.0171, 0.0165, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:08:12,929 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132735.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:08:20,920 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132746.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:08:29,998 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132758.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:08:36,927 INFO [finetune.py:992] (1/2) Epoch 3, batch 8300, loss[loss=0.1641, simple_loss=0.2513, pruned_loss=0.0385, over 12336.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.266, pruned_loss=0.04628, over 2373112.60 frames. ], batch size: 31, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:08:48,596 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9036, 3.3362, 5.0994, 2.6789, 2.7937, 3.8062, 3.3261, 3.6102], device='cuda:1'), covar=tensor([0.0342, 0.1052, 0.0266, 0.1086, 0.1884, 0.1478, 0.1304, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0226, 0.0232, 0.0175, 0.0232, 0.0280, 0.0223, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:09:05,173 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6668, 2.5605, 3.8134, 4.6576, 4.1478, 4.5026, 3.9669, 3.4823], device='cuda:1'), covar=tensor([0.0019, 0.0349, 0.0094, 0.0022, 0.0078, 0.0065, 0.0079, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0117, 0.0099, 0.0073, 0.0097, 0.0107, 0.0086, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:09:05,885 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132807.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:09:13,340 INFO [finetune.py:992] (1/2) Epoch 3, batch 8350, loss[loss=0.1429, simple_loss=0.2246, pruned_loss=0.03062, over 11997.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2664, pruned_loss=0.04616, over 2373947.64 frames. ], batch size: 28, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:09:17,509 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.966e+02 3.421e+02 4.323e+02 8.228e+02, threshold=6.842e+02, percent-clipped=4.0 2023-05-15 22:09:48,605 INFO [finetune.py:992] (1/2) Epoch 3, batch 8400, loss[loss=0.1715, simple_loss=0.2591, pruned_loss=0.04192, over 12302.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2662, pruned_loss=0.04592, over 2374303.85 frames. ], batch size: 34, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:10:25,115 INFO [finetune.py:992] (1/2) Epoch 3, batch 8450, loss[loss=0.1721, simple_loss=0.2506, pruned_loss=0.04681, over 11797.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2652, pruned_loss=0.04565, over 2377391.76 frames. ], batch size: 26, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:10:29,297 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.961e+02 3.332e+02 4.096e+02 8.306e+02, threshold=6.665e+02, percent-clipped=5.0 2023-05-15 22:10:59,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-15 22:11:00,959 INFO [finetune.py:992] (1/2) Epoch 3, batch 8500, loss[loss=0.1573, simple_loss=0.2379, pruned_loss=0.0384, over 12004.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2658, pruned_loss=0.04596, over 2372885.00 frames. ], batch size: 28, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:11:35,610 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133016.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:11:36,946 INFO [finetune.py:992] (1/2) Epoch 3, batch 8550, loss[loss=0.2649, simple_loss=0.3267, pruned_loss=0.1015, over 7971.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2659, pruned_loss=0.04577, over 2368348.17 frames. ], batch size: 98, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:11:41,140 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.845e+02 3.482e+02 4.287e+02 2.922e+03, threshold=6.963e+02, percent-clipped=3.0 2023-05-15 22:12:13,341 INFO [finetune.py:992] (1/2) Epoch 3, batch 8600, loss[loss=0.1575, simple_loss=0.2458, pruned_loss=0.03459, over 12253.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2655, pruned_loss=0.04569, over 2364817.52 frames. ], batch size: 32, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:12:37,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1043, 6.0436, 5.8529, 5.4027, 5.1327, 6.0397, 5.5685, 5.3820], device='cuda:1'), covar=tensor([0.0632, 0.0909, 0.0591, 0.1263, 0.0619, 0.0597, 0.1463, 0.0901], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0512, 0.0473, 0.0583, 0.0384, 0.0663, 0.0718, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:12:38,106 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133102.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:12:49,046 INFO [finetune.py:992] (1/2) Epoch 3, batch 8650, loss[loss=0.213, simple_loss=0.2991, pruned_loss=0.0635, over 10482.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2655, pruned_loss=0.04562, over 2370087.82 frames. ], batch size: 68, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:12:53,339 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 3.020e+02 3.510e+02 4.241e+02 9.799e+02, threshold=7.020e+02, percent-clipped=1.0 2023-05-15 22:13:09,831 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1225, 2.3779, 3.6451, 3.0412, 3.4305, 3.2101, 2.4969, 3.5464], device='cuda:1'), covar=tensor([0.0109, 0.0315, 0.0097, 0.0202, 0.0116, 0.0135, 0.0286, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0191, 0.0168, 0.0172, 0.0192, 0.0147, 0.0181, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:13:24,480 INFO [finetune.py:992] (1/2) Epoch 3, batch 8700, loss[loss=0.1747, simple_loss=0.2653, pruned_loss=0.04201, over 12279.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2646, pruned_loss=0.04562, over 2365860.57 frames. ], batch size: 37, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:13:45,705 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9337, 5.9365, 5.7229, 5.2579, 5.0787, 5.8542, 5.3986, 5.2022], device='cuda:1'), covar=tensor([0.0656, 0.0715, 0.0543, 0.1253, 0.0725, 0.0580, 0.1266, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0570, 0.0512, 0.0473, 0.0583, 0.0386, 0.0664, 0.0720, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:14:01,469 INFO [finetune.py:992] (1/2) Epoch 3, batch 8750, loss[loss=0.193, simple_loss=0.2882, pruned_loss=0.04894, over 11151.00 frames. ], tot_loss[loss=0.178, simple_loss=0.265, pruned_loss=0.04547, over 2367739.50 frames. ], batch size: 55, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:14:05,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.998e+02 2.947e+02 3.495e+02 4.217e+02 7.819e+02, threshold=6.990e+02, percent-clipped=0.0 2023-05-15 22:14:37,132 INFO [finetune.py:992] (1/2) Epoch 3, batch 8800, loss[loss=0.1548, simple_loss=0.2349, pruned_loss=0.0374, over 12246.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2645, pruned_loss=0.04505, over 2369054.17 frames. ], batch size: 28, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:14:39,460 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3257, 1.9596, 3.2285, 4.1272, 1.9566, 4.2518, 4.3231, 4.4541], device='cuda:1'), covar=tensor([0.0109, 0.1286, 0.0398, 0.0191, 0.1237, 0.0193, 0.0144, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0199, 0.0182, 0.0115, 0.0182, 0.0173, 0.0166, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:15:05,054 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6287, 3.7547, 3.3785, 3.2705, 3.0205, 2.9804, 3.7501, 2.3718], device='cuda:1'), covar=tensor([0.0339, 0.0133, 0.0147, 0.0154, 0.0364, 0.0302, 0.0098, 0.0416], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0154, 0.0147, 0.0175, 0.0201, 0.0191, 0.0153, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 22:15:11,443 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133316.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:15:12,696 INFO [finetune.py:992] (1/2) Epoch 3, batch 8850, loss[loss=0.1766, simple_loss=0.2593, pruned_loss=0.04696, over 12100.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2646, pruned_loss=0.04509, over 2368108.45 frames. ], batch size: 33, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:15:16,896 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.778e+02 3.188e+02 3.991e+02 7.101e+02, threshold=6.375e+02, percent-clipped=2.0 2023-05-15 22:15:33,478 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1731, 2.5458, 3.7004, 3.1169, 3.5299, 3.3220, 2.6824, 3.6988], device='cuda:1'), covar=tensor([0.0122, 0.0285, 0.0138, 0.0234, 0.0126, 0.0157, 0.0303, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0192, 0.0171, 0.0174, 0.0194, 0.0148, 0.0184, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:15:44,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5764, 4.7867, 4.1883, 5.2205, 4.7208, 2.7561, 4.3295, 3.0437], device='cuda:1'), covar=tensor([0.0602, 0.0637, 0.1217, 0.0286, 0.0934, 0.1523, 0.1040, 0.2917], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0368, 0.0347, 0.0257, 0.0357, 0.0261, 0.0331, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:15:46,748 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133364.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:15:47,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-15 22:15:49,496 INFO [finetune.py:992] (1/2) Epoch 3, batch 8900, loss[loss=0.2639, simple_loss=0.3348, pruned_loss=0.09654, over 8075.00 frames. ], tot_loss[loss=0.178, simple_loss=0.265, pruned_loss=0.04549, over 2360862.54 frames. ], batch size: 100, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:15:53,927 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133374.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 22:15:54,776 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3762, 4.6007, 4.1229, 5.0406, 4.7055, 3.0816, 4.2873, 3.1088], device='cuda:1'), covar=tensor([0.0714, 0.0851, 0.1305, 0.0344, 0.0906, 0.1381, 0.0971, 0.2954], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0368, 0.0347, 0.0257, 0.0357, 0.0261, 0.0331, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:15:55,361 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133376.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:16:00,314 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9021, 4.5474, 4.0511, 4.1716, 4.6781, 4.0854, 4.2515, 3.9783], device='cuda:1'), covar=tensor([0.1658, 0.1102, 0.1463, 0.1876, 0.0983, 0.1966, 0.1502, 0.1655], device='cuda:1'), in_proj_covar=tensor([0.0331, 0.0460, 0.0358, 0.0415, 0.0437, 0.0415, 0.0366, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:16:08,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-15 22:16:14,079 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133402.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:16:25,126 INFO [finetune.py:992] (1/2) Epoch 3, batch 8950, loss[loss=0.1767, simple_loss=0.2669, pruned_loss=0.0433, over 10491.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2647, pruned_loss=0.04531, over 2356367.77 frames. ], batch size: 68, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:16:29,471 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.233e+02 2.921e+02 3.651e+02 4.216e+02 9.647e+02, threshold=7.301e+02, percent-clipped=4.0 2023-05-15 22:16:37,806 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133435.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 22:16:39,139 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133437.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:16:47,947 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133450.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:17:00,526 INFO [finetune.py:992] (1/2) Epoch 3, batch 9000, loss[loss=0.1916, simple_loss=0.2855, pruned_loss=0.04882, over 12155.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2644, pruned_loss=0.04542, over 2363701.48 frames. ], batch size: 39, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:17:00,527 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 22:17:18,365 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7697, 3.6788, 3.8105, 3.8218, 3.5087, 3.5806, 3.5304, 3.7224], device='cuda:1'), covar=tensor([0.0850, 0.0882, 0.0859, 0.0695, 0.1967, 0.1465, 0.0677, 0.1000], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0639, 0.0550, 0.0596, 0.0775, 0.0712, 0.0518, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 22:17:20,672 INFO [finetune.py:1026] (1/2) Epoch 3, validation: loss=0.3381, simple_loss=0.4084, pruned_loss=0.1339, over 1020973.00 frames. 2023-05-15 22:17:20,673 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 22:17:25,213 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133474.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:17:56,186 INFO [finetune.py:992] (1/2) Epoch 3, batch 9050, loss[loss=0.1698, simple_loss=0.2672, pruned_loss=0.03618, over 11764.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2636, pruned_loss=0.04488, over 2370853.35 frames. ], batch size: 44, lr: 4.87e-03, grad_scale: 16.0 2023-05-15 22:17:57,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.77 vs. limit=5.0 2023-05-15 22:18:00,448 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.048e+02 2.817e+02 3.375e+02 4.169e+02 1.335e+03, threshold=6.751e+02, percent-clipped=5.0 2023-05-15 22:18:08,473 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4349, 2.2547, 3.1214, 4.2990, 2.4469, 4.3652, 4.4502, 4.6070], device='cuda:1'), covar=tensor([0.0110, 0.1178, 0.0425, 0.0135, 0.1075, 0.0155, 0.0132, 0.0069], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0199, 0.0182, 0.0114, 0.0182, 0.0172, 0.0166, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:18:08,499 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133535.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:18:31,248 INFO [finetune.py:992] (1/2) Epoch 3, batch 9100, loss[loss=0.1687, simple_loss=0.2528, pruned_loss=0.04235, over 12169.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2637, pruned_loss=0.04478, over 2376892.65 frames. ], batch size: 29, lr: 4.87e-03, grad_scale: 16.0 2023-05-15 22:18:32,029 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9304, 4.5489, 4.1650, 4.0995, 4.6284, 3.9997, 4.1878, 4.0144], device='cuda:1'), covar=tensor([0.1411, 0.1045, 0.1203, 0.1960, 0.1054, 0.1988, 0.1670, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0464, 0.0361, 0.0418, 0.0442, 0.0421, 0.0370, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:18:34,290 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6513, 3.7959, 3.3681, 3.3157, 3.0045, 2.9104, 3.7311, 2.4643], device='cuda:1'), covar=tensor([0.0303, 0.0104, 0.0172, 0.0132, 0.0312, 0.0249, 0.0094, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0152, 0.0146, 0.0172, 0.0197, 0.0188, 0.0152, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 22:18:55,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3796, 4.1418, 4.1194, 4.5016, 3.0092, 3.9292, 2.5669, 4.0501], device='cuda:1'), covar=tensor([0.1710, 0.0672, 0.0904, 0.0686, 0.1201, 0.0647, 0.1890, 0.1404], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0256, 0.0290, 0.0350, 0.0237, 0.0235, 0.0254, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:19:08,105 INFO [finetune.py:992] (1/2) Epoch 3, batch 9150, loss[loss=0.1811, simple_loss=0.2733, pruned_loss=0.04439, over 11199.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2642, pruned_loss=0.04568, over 2358367.19 frames. ], batch size: 55, lr: 4.87e-03, grad_scale: 16.0 2023-05-15 22:19:12,163 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.917e+02 3.668e+02 4.519e+02 9.899e+02, threshold=7.336e+02, percent-clipped=4.0 2023-05-15 22:19:15,961 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0167, 2.3197, 3.5835, 2.9594, 3.3736, 3.2022, 2.3438, 3.4777], device='cuda:1'), covar=tensor([0.0118, 0.0328, 0.0113, 0.0225, 0.0130, 0.0134, 0.0323, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0191, 0.0170, 0.0173, 0.0192, 0.0146, 0.0182, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:19:32,067 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133652.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:19:43,292 INFO [finetune.py:992] (1/2) Epoch 3, batch 9200, loss[loss=0.1483, simple_loss=0.2469, pruned_loss=0.02482, over 12149.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2637, pruned_loss=0.04497, over 2363917.23 frames. ], batch size: 34, lr: 4.87e-03, grad_scale: 16.0 2023-05-15 22:20:15,940 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133713.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:20:19,357 INFO [finetune.py:992] (1/2) Epoch 3, batch 9250, loss[loss=0.1755, simple_loss=0.2716, pruned_loss=0.03965, over 11644.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.264, pruned_loss=0.04465, over 2369914.42 frames. ], batch size: 48, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:20:25,095 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.732e+02 3.186e+02 3.762e+02 2.059e+03, threshold=6.372e+02, percent-clipped=2.0 2023-05-15 22:20:28,703 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133730.0, num_to_drop=1, layers_to_drop={3} 2023-05-15 22:20:30,175 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133732.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:20:56,408 INFO [finetune.py:992] (1/2) Epoch 3, batch 9300, loss[loss=0.1845, simple_loss=0.2747, pruned_loss=0.04716, over 12384.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2635, pruned_loss=0.04478, over 2367114.14 frames. ], batch size: 38, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:21:20,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-05-15 22:21:31,714 INFO [finetune.py:992] (1/2) Epoch 3, batch 9350, loss[loss=0.1785, simple_loss=0.2739, pruned_loss=0.04153, over 11615.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2632, pruned_loss=0.04479, over 2367756.96 frames. ], batch size: 48, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:21:36,748 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 2.993e+02 3.452e+02 4.472e+02 7.220e+02, threshold=6.904e+02, percent-clipped=5.0 2023-05-15 22:21:37,675 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3956, 4.1320, 4.1887, 4.5143, 3.1433, 4.0372, 2.8698, 4.1591], device='cuda:1'), covar=tensor([0.1524, 0.0687, 0.0868, 0.0625, 0.1131, 0.0557, 0.1674, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0255, 0.0291, 0.0349, 0.0238, 0.0235, 0.0255, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:21:40,254 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133830.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:21:40,412 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0951, 2.2641, 3.5951, 3.0351, 3.3935, 3.2239, 2.3710, 3.5632], device='cuda:1'), covar=tensor([0.0099, 0.0323, 0.0100, 0.0191, 0.0107, 0.0120, 0.0289, 0.0093], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0190, 0.0169, 0.0172, 0.0191, 0.0147, 0.0181, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:21:43,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-15 22:21:55,776 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1450, 4.8541, 4.9473, 5.0038, 4.8425, 4.9709, 4.8850, 2.7958], device='cuda:1'), covar=tensor([0.0085, 0.0052, 0.0073, 0.0050, 0.0042, 0.0076, 0.0071, 0.0623], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0077, 0.0070, 0.0057, 0.0086, 0.0076, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:22:07,699 INFO [finetune.py:992] (1/2) Epoch 3, batch 9400, loss[loss=0.176, simple_loss=0.2676, pruned_loss=0.04225, over 12147.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2643, pruned_loss=0.04537, over 2360147.95 frames. ], batch size: 34, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:22:43,371 INFO [finetune.py:992] (1/2) Epoch 3, batch 9450, loss[loss=0.2023, simple_loss=0.2896, pruned_loss=0.05747, over 11380.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2639, pruned_loss=0.04513, over 2367147.38 frames. ], batch size: 55, lr: 4.87e-03, grad_scale: 8.0 2023-05-15 22:22:48,357 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.031e+02 2.929e+02 3.427e+02 4.452e+02 8.212e+02, threshold=6.855e+02, percent-clipped=3.0 2023-05-15 22:23:06,112 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8549, 5.8256, 5.6476, 5.0864, 5.0383, 5.7526, 5.3592, 5.2591], device='cuda:1'), covar=tensor([0.0757, 0.0949, 0.0614, 0.1479, 0.0667, 0.0721, 0.1449, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0517, 0.0479, 0.0588, 0.0388, 0.0668, 0.0723, 0.0533], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:23:18,714 INFO [finetune.py:992] (1/2) Epoch 3, batch 9500, loss[loss=0.1468, simple_loss=0.2329, pruned_loss=0.03032, over 12349.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2642, pruned_loss=0.04467, over 2369029.53 frames. ], batch size: 30, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:23:39,411 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133997.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:23:51,101 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134008.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:23:54,449 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-05-15 22:23:58,378 INFO [finetune.py:992] (1/2) Epoch 3, batch 9550, loss[loss=0.1571, simple_loss=0.2566, pruned_loss=0.02882, over 12180.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2646, pruned_loss=0.04443, over 2373537.55 frames. ], batch size: 35, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:24:03,377 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 3.023e+02 3.557e+02 4.480e+02 3.253e+03, threshold=7.114e+02, percent-clipped=3.0 2023-05-15 22:24:07,127 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134030.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 22:24:08,588 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134032.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:24:18,279 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-05-15 22:24:27,692 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134058.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:24:34,662 INFO [finetune.py:992] (1/2) Epoch 3, batch 9600, loss[loss=0.1984, simple_loss=0.2904, pruned_loss=0.05321, over 10274.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2633, pruned_loss=0.04388, over 2379834.04 frames. ], batch size: 68, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:24:42,010 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134078.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 22:24:43,359 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134080.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:24:52,670 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0307, 4.7225, 5.0315, 4.9762, 4.1348, 4.2997, 4.4857, 4.7691], device='cuda:1'), covar=tensor([0.0880, 0.1328, 0.0782, 0.0967, 0.3779, 0.2094, 0.0766, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0632, 0.0540, 0.0594, 0.0771, 0.0705, 0.0514, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 22:25:10,212 INFO [finetune.py:992] (1/2) Epoch 3, batch 9650, loss[loss=0.1613, simple_loss=0.2453, pruned_loss=0.03865, over 12183.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2646, pruned_loss=0.04456, over 2375631.14 frames. ], batch size: 31, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:25:14,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-15 22:25:15,048 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.666e+02 3.292e+02 4.068e+02 8.915e+02, threshold=6.584e+02, percent-clipped=1.0 2023-05-15 22:25:18,790 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134130.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:25:46,051 INFO [finetune.py:992] (1/2) Epoch 3, batch 9700, loss[loss=0.1542, simple_loss=0.2428, pruned_loss=0.03283, over 12179.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2644, pruned_loss=0.0447, over 2369790.37 frames. ], batch size: 31, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:25:53,291 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134178.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:26:22,419 INFO [finetune.py:992] (1/2) Epoch 3, batch 9750, loss[loss=0.2504, simple_loss=0.3107, pruned_loss=0.09503, over 8446.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2635, pruned_loss=0.04461, over 2370280.45 frames. ], batch size: 98, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:26:27,014 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9854, 3.4340, 5.2958, 2.9329, 2.8954, 4.1389, 3.4428, 4.1052], device='cuda:1'), covar=tensor([0.0347, 0.1022, 0.0177, 0.0988, 0.1786, 0.1142, 0.1114, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0224, 0.0232, 0.0173, 0.0232, 0.0276, 0.0220, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:26:27,471 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 2.841e+02 3.593e+02 4.266e+02 1.145e+03, threshold=7.185e+02, percent-clipped=4.0 2023-05-15 22:26:34,336 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-15 22:26:39,048 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2568, 4.1917, 4.1595, 4.5673, 3.2340, 3.8440, 2.9927, 4.0902], device='cuda:1'), covar=tensor([0.1712, 0.0636, 0.0878, 0.0683, 0.1046, 0.0723, 0.1581, 0.1563], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0256, 0.0292, 0.0349, 0.0238, 0.0236, 0.0255, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:26:57,999 INFO [finetune.py:992] (1/2) Epoch 3, batch 9800, loss[loss=0.1647, simple_loss=0.2523, pruned_loss=0.03851, over 12027.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2629, pruned_loss=0.04403, over 2372861.31 frames. ], batch size: 31, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:27:24,343 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134304.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:27:27,234 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134308.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:27:28,001 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134309.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:27:34,142 INFO [finetune.py:992] (1/2) Epoch 3, batch 9850, loss[loss=0.1636, simple_loss=0.2494, pruned_loss=0.03894, over 12268.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2633, pruned_loss=0.04419, over 2371757.68 frames. ], batch size: 28, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:27:39,277 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 3.023e+02 3.597e+02 4.284e+02 7.979e+02, threshold=7.194e+02, percent-clipped=2.0 2023-05-15 22:27:58,669 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1242, 4.2267, 2.5645, 2.2824, 3.5691, 2.3286, 3.7225, 2.8545], device='cuda:1'), covar=tensor([0.0669, 0.0570, 0.1139, 0.1698, 0.0335, 0.1432, 0.0505, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0247, 0.0174, 0.0197, 0.0138, 0.0177, 0.0192, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:27:59,977 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134353.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:28:02,160 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134356.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:28:08,732 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134365.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:28:10,638 INFO [finetune.py:992] (1/2) Epoch 3, batch 9900, loss[loss=0.1599, simple_loss=0.2419, pruned_loss=0.03896, over 12347.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2628, pruned_loss=0.04429, over 2369026.67 frames. ], batch size: 30, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:28:12,243 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134370.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:28:18,798 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0820, 2.0264, 2.3942, 2.1592, 2.3542, 2.3970, 1.8661, 2.3882], device='cuda:1'), covar=tensor([0.0089, 0.0220, 0.0144, 0.0164, 0.0110, 0.0139, 0.0227, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0190, 0.0170, 0.0172, 0.0193, 0.0148, 0.0182, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:28:43,299 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3087, 4.8569, 5.3059, 4.6267, 4.9232, 4.6499, 5.3053, 4.9370], device='cuda:1'), covar=tensor([0.0223, 0.0338, 0.0228, 0.0243, 0.0296, 0.0304, 0.0193, 0.0251], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0239, 0.0260, 0.0233, 0.0232, 0.0235, 0.0212, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:28:46,530 INFO [finetune.py:992] (1/2) Epoch 3, batch 9950, loss[loss=0.1631, simple_loss=0.2499, pruned_loss=0.0381, over 12076.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2622, pruned_loss=0.04394, over 2372204.44 frames. ], batch size: 32, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:28:51,536 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.773e+02 3.327e+02 3.836e+02 9.753e+02, threshold=6.655e+02, percent-clipped=1.0 2023-05-15 22:29:22,938 INFO [finetune.py:992] (1/2) Epoch 3, batch 10000, loss[loss=0.1789, simple_loss=0.2759, pruned_loss=0.04096, over 11771.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2635, pruned_loss=0.0446, over 2362539.15 frames. ], batch size: 44, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:29:45,823 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3208, 6.1230, 5.6742, 5.6641, 6.1411, 5.5195, 5.7312, 5.7197], device='cuda:1'), covar=tensor([0.1354, 0.0882, 0.0770, 0.1968, 0.0910, 0.2014, 0.1438, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0465, 0.0361, 0.0415, 0.0440, 0.0417, 0.0369, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:29:59,102 INFO [finetune.py:992] (1/2) Epoch 3, batch 10050, loss[loss=0.1457, simple_loss=0.2342, pruned_loss=0.02854, over 12314.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2633, pruned_loss=0.04451, over 2369525.34 frames. ], batch size: 34, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:30:04,118 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.679e+02 3.237e+02 3.824e+02 1.026e+03, threshold=6.474e+02, percent-clipped=1.0 2023-05-15 22:30:15,029 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4620, 5.2315, 5.3817, 5.3915, 4.9892, 5.0343, 4.8341, 5.3813], device='cuda:1'), covar=tensor([0.0595, 0.0523, 0.0613, 0.0628, 0.1811, 0.1308, 0.0555, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0625, 0.0534, 0.0589, 0.0757, 0.0692, 0.0509, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 22:30:34,529 INFO [finetune.py:992] (1/2) Epoch 3, batch 10100, loss[loss=0.1993, simple_loss=0.293, pruned_loss=0.05283, over 12151.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2642, pruned_loss=0.04492, over 2366996.38 frames. ], batch size: 38, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:30:39,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2315, 5.1958, 5.0697, 5.0463, 4.6902, 5.2409, 5.1498, 5.4209], device='cuda:1'), covar=tensor([0.0201, 0.0113, 0.0157, 0.0258, 0.0718, 0.0206, 0.0159, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0179, 0.0179, 0.0227, 0.0229, 0.0199, 0.0163, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 22:31:11,397 INFO [finetune.py:992] (1/2) Epoch 3, batch 10150, loss[loss=0.1616, simple_loss=0.251, pruned_loss=0.0361, over 12345.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2635, pruned_loss=0.04488, over 2362095.66 frames. ], batch size: 36, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:31:16,371 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.738e+02 2.878e+02 3.408e+02 4.078e+02 1.033e+03, threshold=6.816e+02, percent-clipped=2.0 2023-05-15 22:31:27,539 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.06 vs. limit=5.0 2023-05-15 22:31:33,988 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8035, 2.8760, 3.3333, 4.5931, 2.8026, 4.6450, 4.6982, 4.9117], device='cuda:1'), covar=tensor([0.0068, 0.0966, 0.0392, 0.0105, 0.1063, 0.0157, 0.0111, 0.0058], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0204, 0.0185, 0.0117, 0.0188, 0.0175, 0.0171, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:31:37,546 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134653.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:31:42,470 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134660.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:31:45,994 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134665.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:31:47,999 INFO [finetune.py:992] (1/2) Epoch 3, batch 10200, loss[loss=0.1765, simple_loss=0.2673, pruned_loss=0.04282, over 11670.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2643, pruned_loss=0.04512, over 2362306.79 frames. ], batch size: 48, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:32:07,215 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9811, 4.5994, 4.0777, 4.2548, 4.6785, 4.1093, 4.3155, 4.1030], device='cuda:1'), covar=tensor([0.1358, 0.1008, 0.1294, 0.1766, 0.0986, 0.2004, 0.1456, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0470, 0.0363, 0.0417, 0.0442, 0.0419, 0.0371, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:32:11,352 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134701.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:32:23,037 INFO [finetune.py:992] (1/2) Epoch 3, batch 10250, loss[loss=0.1978, simple_loss=0.2858, pruned_loss=0.05491, over 10610.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2648, pruned_loss=0.04531, over 2357415.71 frames. ], batch size: 68, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:32:27,937 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.962e+02 3.516e+02 4.268e+02 7.106e+02, threshold=7.031e+02, percent-clipped=2.0 2023-05-15 22:32:29,222 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2023-05-15 22:32:59,178 INFO [finetune.py:992] (1/2) Epoch 3, batch 10300, loss[loss=0.1931, simple_loss=0.279, pruned_loss=0.05359, over 12061.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2649, pruned_loss=0.04531, over 2363070.38 frames. ], batch size: 37, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:33:03,090 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7474, 2.6404, 3.3954, 4.5623, 2.7057, 4.6662, 4.7532, 4.8368], device='cuda:1'), covar=tensor([0.0081, 0.1210, 0.0408, 0.0121, 0.1103, 0.0175, 0.0097, 0.0066], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0204, 0.0185, 0.0117, 0.0187, 0.0175, 0.0171, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:33:35,689 INFO [finetune.py:992] (1/2) Epoch 3, batch 10350, loss[loss=0.1928, simple_loss=0.2864, pruned_loss=0.0496, over 11171.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2643, pruned_loss=0.04502, over 2373589.33 frames. ], batch size: 55, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:33:40,537 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 2.825e+02 3.250e+02 4.015e+02 7.345e+02, threshold=6.499e+02, percent-clipped=1.0 2023-05-15 22:34:02,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-15 22:34:07,243 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134862.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:34:11,317 INFO [finetune.py:992] (1/2) Epoch 3, batch 10400, loss[loss=0.202, simple_loss=0.2854, pruned_loss=0.05936, over 11634.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2649, pruned_loss=0.04488, over 2377743.56 frames. ], batch size: 48, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:34:26,400 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8781, 4.6948, 4.8331, 4.8505, 4.4658, 4.4977, 4.3596, 4.7583], device='cuda:1'), covar=tensor([0.0578, 0.0554, 0.0653, 0.0605, 0.1626, 0.1226, 0.0524, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0640, 0.0542, 0.0597, 0.0776, 0.0707, 0.0519, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 22:34:47,511 INFO [finetune.py:992] (1/2) Epoch 3, batch 10450, loss[loss=0.1581, simple_loss=0.245, pruned_loss=0.0356, over 12290.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2638, pruned_loss=0.0443, over 2377045.49 frames. ], batch size: 33, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:34:51,216 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134923.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:34:52,403 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.107e+02 2.965e+02 3.504e+02 4.131e+02 9.829e+02, threshold=7.008e+02, percent-clipped=3.0 2023-05-15 22:35:06,158 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6266, 4.8400, 4.3754, 5.2432, 4.8704, 3.2214, 4.5419, 3.1374], device='cuda:1'), covar=tensor([0.0648, 0.0657, 0.1232, 0.0316, 0.0906, 0.1329, 0.0814, 0.3030], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0364, 0.0346, 0.0258, 0.0356, 0.0260, 0.0330, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:35:18,031 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134960.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:35:21,483 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134965.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:35:23,496 INFO [finetune.py:992] (1/2) Epoch 3, batch 10500, loss[loss=0.1619, simple_loss=0.2435, pruned_loss=0.04013, over 12181.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2642, pruned_loss=0.04439, over 2376992.35 frames. ], batch size: 29, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:35:52,440 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135008.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:35:56,062 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135013.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:35:59,384 INFO [finetune.py:992] (1/2) Epoch 3, batch 10550, loss[loss=0.1583, simple_loss=0.2573, pruned_loss=0.02964, over 12199.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2631, pruned_loss=0.04391, over 2379622.48 frames. ], batch size: 35, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:36:04,295 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.851e+02 3.385e+02 4.106e+02 8.437e+02, threshold=6.769e+02, percent-clipped=3.0 2023-05-15 22:36:11,690 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7955, 4.7853, 4.6880, 4.6894, 3.8683, 4.8308, 4.8080, 4.9503], device='cuda:1'), covar=tensor([0.0214, 0.0152, 0.0177, 0.0300, 0.1073, 0.0279, 0.0154, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0179, 0.0179, 0.0228, 0.0228, 0.0201, 0.0164, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 22:36:35,662 INFO [finetune.py:992] (1/2) Epoch 3, batch 10600, loss[loss=0.19, simple_loss=0.2844, pruned_loss=0.04776, over 10527.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2624, pruned_loss=0.04356, over 2377893.18 frames. ], batch size: 68, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:37:11,786 INFO [finetune.py:992] (1/2) Epoch 3, batch 10650, loss[loss=0.1694, simple_loss=0.2609, pruned_loss=0.03894, over 12343.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2629, pruned_loss=0.04409, over 2378695.82 frames. ], batch size: 35, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:37:11,940 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9628, 5.9581, 5.6538, 5.2934, 5.1623, 5.8887, 5.4561, 5.2021], device='cuda:1'), covar=tensor([0.0731, 0.0867, 0.0679, 0.1389, 0.0605, 0.0648, 0.1483, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0521, 0.0484, 0.0592, 0.0387, 0.0669, 0.0727, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:37:16,770 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.600e+02 3.182e+02 4.069e+02 7.657e+02, threshold=6.365e+02, percent-clipped=1.0 2023-05-15 22:37:47,376 INFO [finetune.py:992] (1/2) Epoch 3, batch 10700, loss[loss=0.1682, simple_loss=0.2551, pruned_loss=0.04063, over 12399.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2622, pruned_loss=0.0436, over 2382974.39 frames. ], batch size: 32, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:37:49,752 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135171.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:38:08,031 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0146, 5.9706, 5.7325, 5.1843, 5.1361, 5.9016, 5.4565, 5.2699], device='cuda:1'), covar=tensor([0.0658, 0.0786, 0.0565, 0.1384, 0.0617, 0.0689, 0.1475, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0570, 0.0513, 0.0479, 0.0586, 0.0382, 0.0661, 0.0721, 0.0529], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:38:17,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5621, 2.4831, 3.6880, 4.5047, 3.9198, 4.5450, 3.8421, 3.3137], device='cuda:1'), covar=tensor([0.0027, 0.0368, 0.0125, 0.0029, 0.0119, 0.0053, 0.0087, 0.0281], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0118, 0.0101, 0.0074, 0.0099, 0.0109, 0.0088, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:38:17,820 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2360, 4.4967, 2.7228, 2.4299, 3.8050, 2.4480, 3.9124, 3.1490], device='cuda:1'), covar=tensor([0.0684, 0.0559, 0.1180, 0.1612, 0.0325, 0.1331, 0.0487, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0248, 0.0175, 0.0199, 0.0138, 0.0179, 0.0193, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:38:24,709 INFO [finetune.py:992] (1/2) Epoch 3, batch 10750, loss[loss=0.1512, simple_loss=0.2321, pruned_loss=0.0352, over 12119.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2635, pruned_loss=0.04411, over 2380062.68 frames. ], batch size: 30, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:38:24,796 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135218.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:38:29,768 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.207e+02 3.060e+02 3.393e+02 4.113e+02 1.124e+03, threshold=6.785e+02, percent-clipped=1.0 2023-05-15 22:38:34,921 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135232.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:39:00,492 INFO [finetune.py:992] (1/2) Epoch 3, batch 10800, loss[loss=0.1886, simple_loss=0.2768, pruned_loss=0.05021, over 10433.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2642, pruned_loss=0.0446, over 2364713.79 frames. ], batch size: 68, lr: 4.86e-03, grad_scale: 8.0 2023-05-15 22:39:36,143 INFO [finetune.py:992] (1/2) Epoch 3, batch 10850, loss[loss=0.1611, simple_loss=0.2443, pruned_loss=0.03888, over 12283.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2645, pruned_loss=0.04464, over 2365333.92 frames. ], batch size: 28, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:39:41,719 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.242e+02 2.957e+02 3.427e+02 4.068e+02 7.800e+02, threshold=6.853e+02, percent-clipped=3.0 2023-05-15 22:39:55,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8770, 3.3825, 5.3189, 2.8207, 2.8985, 3.9349, 3.5331, 4.1313], device='cuda:1'), covar=tensor([0.0456, 0.1124, 0.0261, 0.1061, 0.1836, 0.1288, 0.1166, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0228, 0.0235, 0.0176, 0.0234, 0.0278, 0.0223, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:40:13,858 INFO [finetune.py:992] (1/2) Epoch 3, batch 10900, loss[loss=0.1664, simple_loss=0.2552, pruned_loss=0.03879, over 12301.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2644, pruned_loss=0.04483, over 2367591.71 frames. ], batch size: 33, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:40:26,577 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4271, 5.1829, 5.2518, 5.3590, 4.9263, 5.0187, 4.8314, 5.2512], device='cuda:1'), covar=tensor([0.0558, 0.0571, 0.0726, 0.0569, 0.1927, 0.1123, 0.0470, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0626, 0.0537, 0.0585, 0.0763, 0.0694, 0.0512, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 22:40:37,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5555, 4.8776, 3.1960, 2.8673, 4.0764, 2.6570, 4.1753, 3.5149], device='cuda:1'), covar=tensor([0.0532, 0.0374, 0.0984, 0.1359, 0.0235, 0.1146, 0.0362, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0246, 0.0173, 0.0198, 0.0137, 0.0178, 0.0191, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:40:40,602 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8962, 2.9756, 4.7694, 4.9139, 3.1615, 2.8095, 3.0664, 2.1223], device='cuda:1'), covar=tensor([0.1205, 0.2679, 0.0348, 0.0328, 0.1025, 0.1803, 0.2191, 0.3468], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0371, 0.0266, 0.0288, 0.0251, 0.0278, 0.0349, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:40:49,340 INFO [finetune.py:992] (1/2) Epoch 3, batch 10950, loss[loss=0.2086, simple_loss=0.2905, pruned_loss=0.06336, over 12261.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2637, pruned_loss=0.04492, over 2369952.29 frames. ], batch size: 37, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:40:54,094 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.889e+02 3.575e+02 4.378e+02 6.782e+02, threshold=7.149e+02, percent-clipped=0.0 2023-05-15 22:41:01,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-05-15 22:41:24,967 INFO [finetune.py:992] (1/2) Epoch 3, batch 11000, loss[loss=0.1667, simple_loss=0.249, pruned_loss=0.04217, over 12271.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2655, pruned_loss=0.04593, over 2357091.87 frames. ], batch size: 28, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:41:29,854 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0111, 5.9838, 5.7781, 5.3500, 5.0937, 5.9556, 5.4818, 5.3865], device='cuda:1'), covar=tensor([0.0649, 0.0852, 0.0617, 0.1460, 0.0611, 0.0571, 0.1366, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0512, 0.0481, 0.0586, 0.0383, 0.0660, 0.0719, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:41:49,756 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6965, 4.6589, 4.5608, 4.1799, 4.3646, 4.6753, 4.3243, 4.2175], device='cuda:1'), covar=tensor([0.0641, 0.0785, 0.0602, 0.1301, 0.1090, 0.0689, 0.1324, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0509, 0.0479, 0.0583, 0.0381, 0.0657, 0.0715, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 22:42:01,228 INFO [finetune.py:992] (1/2) Epoch 3, batch 11050, loss[loss=0.1532, simple_loss=0.2321, pruned_loss=0.03711, over 12250.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2685, pruned_loss=0.04783, over 2322570.88 frames. ], batch size: 28, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:42:01,403 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135518.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:42:06,056 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 3.260e+02 3.848e+02 4.589e+02 1.215e+03, threshold=7.697e+02, percent-clipped=4.0 2023-05-15 22:42:07,363 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135527.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:42:15,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 2023-05-15 22:42:35,348 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135566.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:42:36,713 INFO [finetune.py:992] (1/2) Epoch 3, batch 11100, loss[loss=0.1706, simple_loss=0.261, pruned_loss=0.0401, over 12148.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2734, pruned_loss=0.05117, over 2270786.90 frames. ], batch size: 36, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:43:12,094 INFO [finetune.py:992] (1/2) Epoch 3, batch 11150, loss[loss=0.3027, simple_loss=0.3583, pruned_loss=0.1235, over 6754.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2802, pruned_loss=0.05572, over 2205383.89 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:43:13,136 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-15 22:43:16,874 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 3.344e+02 4.365e+02 5.107e+02 8.408e+02, threshold=8.730e+02, percent-clipped=1.0 2023-05-15 22:43:20,652 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135630.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:43:27,976 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2701, 3.1241, 3.0056, 3.2996, 2.6422, 3.1224, 2.6484, 3.0428], device='cuda:1'), covar=tensor([0.1279, 0.0695, 0.0647, 0.0380, 0.0787, 0.0645, 0.1205, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0252, 0.0285, 0.0341, 0.0232, 0.0230, 0.0248, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:43:47,809 INFO [finetune.py:992] (1/2) Epoch 3, batch 11200, loss[loss=0.262, simple_loss=0.3366, pruned_loss=0.09366, over 11709.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2881, pruned_loss=0.06114, over 2145369.51 frames. ], batch size: 44, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:44:02,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.22 vs. limit=5.0 2023-05-15 22:44:03,973 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135691.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:44:09,823 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8435, 2.5801, 3.4504, 3.5411, 2.9375, 2.7913, 2.6807, 2.4724], device='cuda:1'), covar=tensor([0.0932, 0.2035, 0.0519, 0.0380, 0.0696, 0.1571, 0.1853, 0.2717], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0366, 0.0263, 0.0282, 0.0247, 0.0276, 0.0347, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:44:22,509 INFO [finetune.py:992] (1/2) Epoch 3, batch 11250, loss[loss=0.1841, simple_loss=0.2785, pruned_loss=0.04487, over 12270.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2953, pruned_loss=0.06582, over 2090546.14 frames. ], batch size: 37, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:44:27,927 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.526e+02 3.507e+02 4.335e+02 5.323e+02 9.145e+02, threshold=8.671e+02, percent-clipped=1.0 2023-05-15 22:44:57,276 INFO [finetune.py:992] (1/2) Epoch 3, batch 11300, loss[loss=0.2236, simple_loss=0.3134, pruned_loss=0.0669, over 10118.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3028, pruned_loss=0.07077, over 2033408.32 frames. ], batch size: 68, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:45:02,260 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9880, 2.1662, 2.6555, 2.9054, 2.8629, 3.0067, 2.7522, 2.4423], device='cuda:1'), covar=tensor([0.0052, 0.0320, 0.0162, 0.0056, 0.0098, 0.0083, 0.0094, 0.0291], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0118, 0.0099, 0.0073, 0.0097, 0.0108, 0.0087, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:45:33,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-05-15 22:45:33,283 INFO [finetune.py:992] (1/2) Epoch 3, batch 11350, loss[loss=0.309, simple_loss=0.37, pruned_loss=0.124, over 6895.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3073, pruned_loss=0.07432, over 1969467.28 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:45:37,862 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.640e+02 3.541e+02 4.178e+02 4.948e+02 7.953e+02, threshold=8.356e+02, percent-clipped=0.0 2023-05-15 22:45:39,454 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135827.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:45:51,996 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-15 22:46:07,753 INFO [finetune.py:992] (1/2) Epoch 3, batch 11400, loss[loss=0.2793, simple_loss=0.3343, pruned_loss=0.1121, over 7097.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3104, pruned_loss=0.07646, over 1944825.82 frames. ], batch size: 99, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:46:13,228 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135875.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:46:25,963 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8022, 2.2718, 3.2697, 3.7413, 3.5681, 3.6895, 3.4764, 2.3926], device='cuda:1'), covar=tensor([0.0036, 0.0401, 0.0135, 0.0040, 0.0081, 0.0074, 0.0081, 0.0441], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0116, 0.0098, 0.0072, 0.0095, 0.0107, 0.0085, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:46:42,738 INFO [finetune.py:992] (1/2) Epoch 3, batch 11450, loss[loss=0.2985, simple_loss=0.3578, pruned_loss=0.1196, over 7064.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3143, pruned_loss=0.07909, over 1915258.40 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:46:47,387 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.608e+02 3.568e+02 4.271e+02 4.917e+02 1.252e+03, threshold=8.543e+02, percent-clipped=1.0 2023-05-15 22:47:16,862 INFO [finetune.py:992] (1/2) Epoch 3, batch 11500, loss[loss=0.261, simple_loss=0.3257, pruned_loss=0.09819, over 6805.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3171, pruned_loss=0.08172, over 1875333.12 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:47:26,930 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135982.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:47:29,621 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135986.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:47:35,198 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135994.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:47:36,620 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8065, 2.4232, 3.3028, 3.7549, 3.5853, 3.7471, 3.4415, 2.5154], device='cuda:1'), covar=tensor([0.0035, 0.0340, 0.0120, 0.0042, 0.0082, 0.0062, 0.0098, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0116, 0.0097, 0.0072, 0.0095, 0.0106, 0.0085, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:47:48,231 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2862, 4.6237, 4.1674, 5.0031, 4.5401, 2.7406, 4.3457, 3.2715], device='cuda:1'), covar=tensor([0.0793, 0.0779, 0.1321, 0.0342, 0.1054, 0.1892, 0.1001, 0.2835], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0349, 0.0329, 0.0241, 0.0340, 0.0251, 0.0315, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:47:55,319 INFO [finetune.py:992] (1/2) Epoch 3, batch 11550, loss[loss=0.3043, simple_loss=0.3534, pruned_loss=0.1276, over 7110.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3187, pruned_loss=0.08374, over 1831729.08 frames. ], batch size: 104, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:47:57,637 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8294, 3.7082, 3.8331, 3.5900, 3.7062, 3.5905, 3.8087, 3.4790], device='cuda:1'), covar=tensor([0.0321, 0.0316, 0.0302, 0.0204, 0.0280, 0.0291, 0.0258, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0220, 0.0240, 0.0218, 0.0215, 0.0217, 0.0196, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:48:00,054 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.813e+02 3.551e+02 4.092e+02 4.768e+02 1.093e+03, threshold=8.184e+02, percent-clipped=1.0 2023-05-15 22:48:02,306 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7054, 2.9491, 4.2898, 4.5486, 3.1741, 2.7146, 2.8063, 2.0103], device='cuda:1'), covar=tensor([0.1247, 0.2296, 0.0384, 0.0311, 0.0911, 0.1904, 0.2395, 0.3746], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0357, 0.0256, 0.0275, 0.0241, 0.0269, 0.0340, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:48:13,011 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136043.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:48:21,191 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136055.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:48:30,362 INFO [finetune.py:992] (1/2) Epoch 3, batch 11600, loss[loss=0.2663, simple_loss=0.3382, pruned_loss=0.09718, over 6772.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3204, pruned_loss=0.08527, over 1809195.54 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 16.0 2023-05-15 22:48:52,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.10 vs. limit=5.0 2023-05-15 22:49:05,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-15 22:49:05,621 INFO [finetune.py:992] (1/2) Epoch 3, batch 11650, loss[loss=0.2876, simple_loss=0.3294, pruned_loss=0.1229, over 7391.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3202, pruned_loss=0.08612, over 1794412.32 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:49:11,213 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.566e+02 3.696e+02 4.157e+02 5.000e+02 3.157e+03, threshold=8.314e+02, percent-clipped=5.0 2023-05-15 22:49:22,690 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136142.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:49:40,488 INFO [finetune.py:992] (1/2) Epoch 3, batch 11700, loss[loss=0.2645, simple_loss=0.3297, pruned_loss=0.0996, over 6276.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3203, pruned_loss=0.08699, over 1754434.96 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:49:45,453 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8458, 3.7303, 3.8438, 3.5874, 3.7067, 3.5685, 3.8040, 3.5259], device='cuda:1'), covar=tensor([0.0364, 0.0355, 0.0345, 0.0243, 0.0307, 0.0289, 0.0327, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0218, 0.0237, 0.0216, 0.0213, 0.0215, 0.0194, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:49:50,643 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5688, 3.4099, 3.5517, 3.6517, 3.5200, 3.6565, 3.5177, 2.6514], device='cuda:1'), covar=tensor([0.0088, 0.0081, 0.0108, 0.0079, 0.0066, 0.0101, 0.0086, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0067, 0.0071, 0.0064, 0.0052, 0.0080, 0.0069, 0.0084], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:50:05,167 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136203.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:50:15,095 INFO [finetune.py:992] (1/2) Epoch 3, batch 11750, loss[loss=0.2044, simple_loss=0.299, pruned_loss=0.05493, over 10080.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.321, pruned_loss=0.08813, over 1722799.60 frames. ], batch size: 68, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:50:20,310 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 3.625e+02 4.141e+02 4.770e+02 9.665e+02, threshold=8.282e+02, percent-clipped=2.0 2023-05-15 22:50:24,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-05-15 22:50:50,277 INFO [finetune.py:992] (1/2) Epoch 3, batch 11800, loss[loss=0.2903, simple_loss=0.3418, pruned_loss=0.1194, over 6404.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.323, pruned_loss=0.08958, over 1716729.73 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:51:02,669 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136286.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:51:25,568 INFO [finetune.py:992] (1/2) Epoch 3, batch 11850, loss[loss=0.2777, simple_loss=0.3378, pruned_loss=0.1088, over 6719.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3247, pruned_loss=0.09061, over 1695460.67 frames. ], batch size: 99, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:51:30,897 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.669e+02 3.750e+02 4.377e+02 5.279e+02 9.259e+02, threshold=8.753e+02, percent-clipped=2.0 2023-05-15 22:51:36,270 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136334.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:51:37,056 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136335.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:51:38,922 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136338.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:51:47,154 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136350.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:52:00,484 INFO [finetune.py:992] (1/2) Epoch 3, batch 11900, loss[loss=0.2558, simple_loss=0.3237, pruned_loss=0.09389, over 7290.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.324, pruned_loss=0.08863, over 1704012.68 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:52:05,519 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3085, 3.1530, 3.1424, 3.3895, 2.7516, 3.1409, 2.5540, 2.7514], device='cuda:1'), covar=tensor([0.1819, 0.0923, 0.0947, 0.0564, 0.1082, 0.0832, 0.1858, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0253, 0.0281, 0.0335, 0.0231, 0.0228, 0.0249, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:52:19,644 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136396.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:52:34,984 INFO [finetune.py:992] (1/2) Epoch 3, batch 11950, loss[loss=0.2461, simple_loss=0.3154, pruned_loss=0.08846, over 7160.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3213, pruned_loss=0.0861, over 1691752.41 frames. ], batch size: 98, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:52:41,189 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.345e+02 3.513e+02 4.159e+02 5.094e+02 8.737e+02, threshold=8.317e+02, percent-clipped=0.0 2023-05-15 22:53:09,688 INFO [finetune.py:992] (1/2) Epoch 3, batch 12000, loss[loss=0.2098, simple_loss=0.2993, pruned_loss=0.0602, over 11756.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3166, pruned_loss=0.08201, over 1713872.07 frames. ], batch size: 49, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:53:09,688 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 22:53:28,729 INFO [finetune.py:1026] (1/2) Epoch 3, validation: loss=0.2937, simple_loss=0.369, pruned_loss=0.1092, over 1020973.00 frames. 2023-05-15 22:53:28,730 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 22:53:48,922 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136498.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:54:02,715 INFO [finetune.py:992] (1/2) Epoch 3, batch 12050, loss[loss=0.2125, simple_loss=0.2893, pruned_loss=0.06783, over 7209.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3124, pruned_loss=0.07867, over 1721165.07 frames. ], batch size: 99, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:54:08,051 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.114e+02 2.988e+02 3.634e+02 4.408e+02 6.442e+02, threshold=7.268e+02, percent-clipped=0.0 2023-05-15 22:54:13,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-05-15 22:54:17,366 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7697, 4.1075, 3.6251, 4.3359, 3.8261, 2.6718, 3.8744, 2.9220], device='cuda:1'), covar=tensor([0.0859, 0.0777, 0.1376, 0.0355, 0.1294, 0.1735, 0.0970, 0.3198], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0341, 0.0322, 0.0232, 0.0332, 0.0247, 0.0309, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:54:35,715 INFO [finetune.py:992] (1/2) Epoch 3, batch 12100, loss[loss=0.2421, simple_loss=0.3349, pruned_loss=0.07465, over 10139.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.311, pruned_loss=0.07702, over 1734917.06 frames. ], batch size: 68, lr: 4.85e-03, grad_scale: 8.0 2023-05-15 22:54:58,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-15 22:54:59,159 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.8609, 1.8088, 2.1151, 1.9764, 2.1511, 2.1312, 1.6263, 2.1883], device='cuda:1'), covar=tensor([0.0107, 0.0318, 0.0138, 0.0203, 0.0169, 0.0170, 0.0301, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0177, 0.0150, 0.0157, 0.0175, 0.0135, 0.0169, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:55:07,958 INFO [finetune.py:992] (1/2) Epoch 3, batch 12150, loss[loss=0.2228, simple_loss=0.3115, pruned_loss=0.06702, over 11576.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3111, pruned_loss=0.07698, over 1730013.55 frames. ], batch size: 48, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:55:12,828 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 3.271e+02 3.722e+02 4.794e+02 9.063e+02, threshold=7.443e+02, percent-clipped=4.0 2023-05-15 22:55:13,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-05-15 22:55:14,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-15 22:55:20,450 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136638.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:55:24,266 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136643.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:55:28,843 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136650.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:55:39,638 INFO [finetune.py:992] (1/2) Epoch 3, batch 12200, loss[loss=0.2687, simple_loss=0.3221, pruned_loss=0.1077, over 6859.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3117, pruned_loss=0.07803, over 1694028.48 frames. ], batch size: 99, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:55:50,572 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136686.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:55:50,737 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3861, 3.1707, 3.1625, 3.3740, 2.6446, 3.1782, 2.4817, 2.8403], device='cuda:1'), covar=tensor([0.1493, 0.0679, 0.0831, 0.0571, 0.1012, 0.0677, 0.1637, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0253, 0.0282, 0.0334, 0.0231, 0.0230, 0.0251, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:55:53,519 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136691.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:55:57,898 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136698.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:56:26,419 INFO [finetune.py:992] (1/2) Epoch 4, batch 0, loss[loss=0.2304, simple_loss=0.3105, pruned_loss=0.07514, over 12292.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3105, pruned_loss=0.07514, over 12292.00 frames. ], batch size: 33, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:56:26,420 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 22:56:39,655 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5167, 3.3567, 3.4091, 3.4940, 3.2715, 3.5280, 3.4549, 3.5791], device='cuda:1'), covar=tensor([0.0175, 0.0159, 0.0156, 0.0289, 0.0493, 0.0236, 0.0191, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0148, 0.0149, 0.0190, 0.0189, 0.0166, 0.0137, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-15 22:56:44,023 INFO [finetune.py:1026] (1/2) Epoch 4, validation: loss=0.2991, simple_loss=0.3703, pruned_loss=0.1139, over 1020973.00 frames. 2023-05-15 22:56:44,024 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 22:56:45,674 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136704.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:57:00,965 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.149e+02 3.364e+02 3.979e+02 4.865e+02 1.021e+03, threshold=7.957e+02, percent-clipped=3.0 2023-05-15 22:57:20,547 INFO [finetune.py:992] (1/2) Epoch 4, batch 50, loss[loss=0.1746, simple_loss=0.2634, pruned_loss=0.04288, over 12137.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2719, pruned_loss=0.04792, over 539633.78 frames. ], batch size: 38, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:57:23,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.02 vs. limit=5.0 2023-05-15 22:57:26,272 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9230, 5.7463, 5.6988, 5.0872, 4.9499, 5.9589, 5.2586, 5.3848], device='cuda:1'), covar=tensor([0.1179, 0.1521, 0.1159, 0.2839, 0.1086, 0.1244, 0.3310, 0.1545], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0476, 0.0448, 0.0540, 0.0357, 0.0609, 0.0656, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:57:44,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-15 22:57:48,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-15 22:57:53,394 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136798.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:57:56,277 INFO [finetune.py:992] (1/2) Epoch 4, batch 100, loss[loss=0.1714, simple_loss=0.2509, pruned_loss=0.0459, over 11815.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2713, pruned_loss=0.04794, over 946180.77 frames. ], batch size: 26, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:57:59,321 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136806.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:58:13,295 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.875e+02 3.507e+02 4.053e+02 1.106e+03, threshold=7.015e+02, percent-clipped=2.0 2023-05-15 22:58:27,418 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136846.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:58:31,618 INFO [finetune.py:992] (1/2) Epoch 4, batch 150, loss[loss=0.1864, simple_loss=0.2707, pruned_loss=0.05104, over 12132.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2704, pruned_loss=0.04721, over 1272016.22 frames. ], batch size: 39, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:58:40,389 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3662, 5.1890, 5.2606, 5.3437, 4.9531, 4.9730, 4.8283, 5.2964], device='cuda:1'), covar=tensor([0.0789, 0.0667, 0.0833, 0.0690, 0.2086, 0.1319, 0.0536, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0571, 0.0496, 0.0534, 0.0687, 0.0636, 0.0470, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-15 22:58:42,687 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136867.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 22:59:07,973 INFO [finetune.py:992] (1/2) Epoch 4, batch 200, loss[loss=0.1799, simple_loss=0.2655, pruned_loss=0.04716, over 12108.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.27, pruned_loss=0.04744, over 1516524.86 frames. ], batch size: 33, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 22:59:18,932 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7308, 4.5076, 4.2671, 4.7051, 3.4195, 4.2681, 3.0624, 4.4972], device='cuda:1'), covar=tensor([0.1467, 0.0546, 0.1052, 0.0710, 0.0935, 0.0516, 0.1568, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0256, 0.0286, 0.0340, 0.0233, 0.0232, 0.0253, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 22:59:25,716 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 3.078e+02 3.576e+02 4.250e+02 9.794e+02, threshold=7.152e+02, percent-clipped=4.0 2023-05-15 22:59:28,238 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5121, 4.7444, 4.2188, 5.0946, 4.6201, 3.1620, 4.4641, 3.1816], device='cuda:1'), covar=tensor([0.0691, 0.0734, 0.1384, 0.0380, 0.1152, 0.1478, 0.0867, 0.3050], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0352, 0.0332, 0.0239, 0.0343, 0.0254, 0.0318, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 22:59:44,105 INFO [finetune.py:992] (1/2) Epoch 4, batch 250, loss[loss=0.1772, simple_loss=0.2718, pruned_loss=0.04127, over 12028.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2702, pruned_loss=0.04735, over 1705771.50 frames. ], batch size: 31, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:00:00,721 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6041, 4.7888, 4.2921, 5.1688, 4.7642, 3.1132, 4.4797, 3.2180], device='cuda:1'), covar=tensor([0.0660, 0.0803, 0.1256, 0.0301, 0.1002, 0.1556, 0.0907, 0.3078], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0356, 0.0336, 0.0241, 0.0346, 0.0257, 0.0321, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:00:11,989 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136991.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:00:17,648 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136999.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:00:20,042 INFO [finetune.py:992] (1/2) Epoch 4, batch 300, loss[loss=0.1662, simple_loss=0.2449, pruned_loss=0.04377, over 12352.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2681, pruned_loss=0.0468, over 1850045.50 frames. ], batch size: 30, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:00:27,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2023-05-15 23:00:36,828 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.970e+02 3.404e+02 3.901e+02 9.506e+02, threshold=6.809e+02, percent-clipped=2.0 2023-05-15 23:00:37,747 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9856, 4.7719, 4.8717, 4.9281, 4.5999, 4.8955, 4.7721, 2.5507], device='cuda:1'), covar=tensor([0.0086, 0.0058, 0.0071, 0.0056, 0.0051, 0.0088, 0.0071, 0.0736], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0066, 0.0071, 0.0063, 0.0052, 0.0080, 0.0069, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:00:46,928 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137039.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:00:56,327 INFO [finetune.py:992] (1/2) Epoch 4, batch 350, loss[loss=0.1712, simple_loss=0.253, pruned_loss=0.04473, over 12178.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2675, pruned_loss=0.0463, over 1977299.96 frames. ], batch size: 29, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:01:32,269 INFO [finetune.py:992] (1/2) Epoch 4, batch 400, loss[loss=0.1581, simple_loss=0.2449, pruned_loss=0.03561, over 12171.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2666, pruned_loss=0.0456, over 2067516.52 frames. ], batch size: 31, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:01:49,110 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.920e+02 2.863e+02 3.332e+02 3.985e+02 7.745e+02, threshold=6.664e+02, percent-clipped=2.0 2023-05-15 23:01:50,808 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8152, 3.4788, 5.1120, 2.7357, 3.0101, 3.8282, 3.3894, 3.9489], device='cuda:1'), covar=tensor([0.0415, 0.1041, 0.0254, 0.1153, 0.1765, 0.1362, 0.1217, 0.0995], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0221, 0.0219, 0.0174, 0.0228, 0.0266, 0.0217, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:02:05,465 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6023, 3.5331, 3.2748, 3.1580, 2.9192, 2.7591, 3.4936, 2.2681], device='cuda:1'), covar=tensor([0.0319, 0.0119, 0.0129, 0.0168, 0.0339, 0.0321, 0.0110, 0.0454], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0145, 0.0137, 0.0166, 0.0185, 0.0179, 0.0143, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 23:02:07,350 INFO [finetune.py:992] (1/2) Epoch 4, batch 450, loss[loss=0.1658, simple_loss=0.2419, pruned_loss=0.04487, over 12174.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2655, pruned_loss=0.04536, over 2146339.06 frames. ], batch size: 29, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:02:14,463 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137162.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:02:43,495 INFO [finetune.py:992] (1/2) Epoch 4, batch 500, loss[loss=0.164, simple_loss=0.2519, pruned_loss=0.03809, over 12288.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2659, pruned_loss=0.04533, over 2199928.64 frames. ], batch size: 33, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:02:45,270 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7205, 2.8211, 4.6870, 4.9544, 3.0111, 2.6713, 3.0005, 2.1534], device='cuda:1'), covar=tensor([0.1491, 0.3090, 0.0420, 0.0377, 0.1216, 0.2274, 0.2614, 0.4077], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0367, 0.0259, 0.0280, 0.0247, 0.0277, 0.0347, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:02:47,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-15 23:02:57,959 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2418, 4.0178, 3.9494, 4.3583, 2.9601, 3.8549, 2.5151, 4.0726], device='cuda:1'), covar=tensor([0.1679, 0.0658, 0.0927, 0.0587, 0.1103, 0.0613, 0.1814, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0255, 0.0284, 0.0339, 0.0233, 0.0232, 0.0251, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:03:00,589 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.872e+02 3.437e+02 3.927e+02 1.123e+03, threshold=6.874e+02, percent-clipped=1.0 2023-05-15 23:03:12,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2023-05-15 23:03:19,538 INFO [finetune.py:992] (1/2) Epoch 4, batch 550, loss[loss=0.206, simple_loss=0.2978, pruned_loss=0.0571, over 11985.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2664, pruned_loss=0.04529, over 2237929.07 frames. ], batch size: 40, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:03:52,656 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137299.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:03:54,762 INFO [finetune.py:992] (1/2) Epoch 4, batch 600, loss[loss=0.1638, simple_loss=0.2508, pruned_loss=0.03844, over 12343.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2671, pruned_loss=0.04564, over 2258522.49 frames. ], batch size: 31, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:04:11,614 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.075e+02 2.902e+02 3.445e+02 4.026e+02 7.738e+02, threshold=6.890e+02, percent-clipped=2.0 2023-05-15 23:04:27,064 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137347.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:04:30,491 INFO [finetune.py:992] (1/2) Epoch 4, batch 650, loss[loss=0.186, simple_loss=0.269, pruned_loss=0.05151, over 12133.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2659, pruned_loss=0.04517, over 2290106.02 frames. ], batch size: 39, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:04:30,750 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6035, 3.6914, 3.3720, 3.2857, 2.9724, 2.8617, 3.7187, 2.3749], device='cuda:1'), covar=tensor([0.0302, 0.0103, 0.0123, 0.0158, 0.0350, 0.0304, 0.0100, 0.0415], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0148, 0.0139, 0.0169, 0.0189, 0.0183, 0.0145, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 23:05:06,776 INFO [finetune.py:992] (1/2) Epoch 4, batch 700, loss[loss=0.1864, simple_loss=0.2754, pruned_loss=0.04873, over 12120.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2657, pruned_loss=0.04515, over 2309443.84 frames. ], batch size: 38, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:05:10,610 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2102, 5.1531, 4.9083, 4.9827, 4.5805, 5.0703, 5.0966, 5.3323], device='cuda:1'), covar=tensor([0.0200, 0.0116, 0.0186, 0.0283, 0.0788, 0.0249, 0.0141, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0168, 0.0169, 0.0217, 0.0215, 0.0187, 0.0153, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-15 23:05:23,830 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.978e+02 3.456e+02 4.081e+02 5.922e+02, threshold=6.912e+02, percent-clipped=0.0 2023-05-15 23:05:26,960 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5157, 3.5049, 3.2837, 3.1036, 2.9003, 2.5709, 3.5207, 2.3304], device='cuda:1'), covar=tensor([0.0303, 0.0116, 0.0118, 0.0148, 0.0339, 0.0342, 0.0104, 0.0359], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0148, 0.0140, 0.0169, 0.0188, 0.0183, 0.0145, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 23:05:42,332 INFO [finetune.py:992] (1/2) Epoch 4, batch 750, loss[loss=0.1533, simple_loss=0.2376, pruned_loss=0.03453, over 12034.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2656, pruned_loss=0.04523, over 2325252.29 frames. ], batch size: 31, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:05:45,337 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2450, 5.2007, 4.8947, 4.9993, 4.6465, 5.0567, 5.1204, 5.3280], device='cuda:1'), covar=tensor([0.0259, 0.0116, 0.0209, 0.0325, 0.0742, 0.0370, 0.0139, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0171, 0.0171, 0.0220, 0.0218, 0.0190, 0.0155, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-15 23:05:49,431 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137462.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:06:04,256 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4435, 2.6164, 3.1414, 4.2474, 2.3148, 4.3845, 4.4886, 4.5691], device='cuda:1'), covar=tensor([0.0101, 0.1166, 0.0452, 0.0113, 0.1205, 0.0206, 0.0101, 0.0060], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0197, 0.0178, 0.0110, 0.0180, 0.0165, 0.0161, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:06:06,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.48 vs. limit=5.0 2023-05-15 23:06:18,129 INFO [finetune.py:992] (1/2) Epoch 4, batch 800, loss[loss=0.1988, simple_loss=0.2868, pruned_loss=0.05538, over 12092.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2663, pruned_loss=0.04569, over 2328712.75 frames. ], batch size: 38, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:06:24,142 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137510.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:06:36,142 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 2.963e+02 3.442e+02 4.311e+02 7.987e+02, threshold=6.885e+02, percent-clipped=2.0 2023-05-15 23:06:46,479 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8980, 2.3731, 3.2772, 2.8516, 3.2730, 3.0900, 2.4098, 3.3439], device='cuda:1'), covar=tensor([0.0118, 0.0299, 0.0153, 0.0257, 0.0137, 0.0156, 0.0299, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0190, 0.0165, 0.0171, 0.0190, 0.0145, 0.0182, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:06:54,379 INFO [finetune.py:992] (1/2) Epoch 4, batch 850, loss[loss=0.1796, simple_loss=0.2688, pruned_loss=0.04524, over 12033.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2661, pruned_loss=0.04556, over 2346987.04 frames. ], batch size: 40, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:07:17,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-15 23:07:29,858 INFO [finetune.py:992] (1/2) Epoch 4, batch 900, loss[loss=0.1892, simple_loss=0.2818, pruned_loss=0.04831, over 12036.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2657, pruned_loss=0.04527, over 2356986.34 frames. ], batch size: 42, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:07:46,880 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.861e+02 3.473e+02 4.370e+02 1.849e+03, threshold=6.947e+02, percent-clipped=5.0 2023-05-15 23:08:06,293 INFO [finetune.py:992] (1/2) Epoch 4, batch 950, loss[loss=0.1957, simple_loss=0.2835, pruned_loss=0.05389, over 10727.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2657, pruned_loss=0.04512, over 2366239.70 frames. ], batch size: 68, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:08:17,790 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6561, 2.6933, 4.3688, 4.4832, 2.8411, 2.5841, 2.7194, 1.9904], device='cuda:1'), covar=tensor([0.1341, 0.2950, 0.0439, 0.0422, 0.1109, 0.2025, 0.2702, 0.3936], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0368, 0.0260, 0.0280, 0.0248, 0.0277, 0.0347, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:08:42,204 INFO [finetune.py:992] (1/2) Epoch 4, batch 1000, loss[loss=0.191, simple_loss=0.2853, pruned_loss=0.04836, over 12321.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2659, pruned_loss=0.04515, over 2365789.47 frames. ], batch size: 34, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:08:59,200 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.918e+02 3.501e+02 4.066e+02 1.012e+03, threshold=7.003e+02, percent-clipped=3.0 2023-05-15 23:09:05,778 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4120, 5.1732, 5.2583, 5.3336, 4.9577, 4.9452, 4.8210, 5.2668], device='cuda:1'), covar=tensor([0.0597, 0.0562, 0.0784, 0.0611, 0.1838, 0.1492, 0.0569, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0616, 0.0534, 0.0577, 0.0751, 0.0690, 0.0501, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:09:17,644 INFO [finetune.py:992] (1/2) Epoch 4, batch 1050, loss[loss=0.19, simple_loss=0.282, pruned_loss=0.04903, over 10443.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2644, pruned_loss=0.04447, over 2373947.27 frames. ], batch size: 68, lr: 4.84e-03, grad_scale: 8.0 2023-05-15 23:09:25,549 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9437, 5.0533, 4.7707, 4.7983, 4.4639, 4.9309, 4.9496, 5.2255], device='cuda:1'), covar=tensor([0.0244, 0.0119, 0.0189, 0.0345, 0.0778, 0.0339, 0.0154, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0173, 0.0173, 0.0224, 0.0221, 0.0193, 0.0159, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1') 2023-05-15 23:09:26,224 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1118, 4.7911, 5.1102, 4.3672, 4.7633, 4.4311, 5.0882, 4.8989], device='cuda:1'), covar=tensor([0.0307, 0.0358, 0.0366, 0.0312, 0.0336, 0.0344, 0.0305, 0.0300], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0233, 0.0252, 0.0229, 0.0228, 0.0231, 0.0208, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:09:53,715 INFO [finetune.py:992] (1/2) Epoch 4, batch 1100, loss[loss=0.1791, simple_loss=0.2621, pruned_loss=0.04805, over 12334.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2652, pruned_loss=0.04473, over 2383078.49 frames. ], batch size: 30, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:10:11,422 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 2.964e+02 3.480e+02 4.053e+02 1.106e+03, threshold=6.959e+02, percent-clipped=4.0 2023-05-15 23:10:29,693 INFO [finetune.py:992] (1/2) Epoch 4, batch 1150, loss[loss=0.154, simple_loss=0.2445, pruned_loss=0.03178, over 12122.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2648, pruned_loss=0.04428, over 2388100.06 frames. ], batch size: 33, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:10:58,430 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137892.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:11:00,364 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137895.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:11:04,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-15 23:11:04,989 INFO [finetune.py:992] (1/2) Epoch 4, batch 1200, loss[loss=0.1488, simple_loss=0.2307, pruned_loss=0.03342, over 11990.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2638, pruned_loss=0.04398, over 2394441.46 frames. ], batch size: 28, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:11:05,226 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1597, 4.1429, 2.6764, 2.4169, 3.5809, 2.3113, 3.6930, 2.8253], device='cuda:1'), covar=tensor([0.0682, 0.0774, 0.1115, 0.1550, 0.0402, 0.1481, 0.0560, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0245, 0.0173, 0.0199, 0.0136, 0.0181, 0.0191, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:11:22,096 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.876e+02 3.340e+02 4.046e+02 2.321e+03, threshold=6.680e+02, percent-clipped=1.0 2023-05-15 23:11:41,363 INFO [finetune.py:992] (1/2) Epoch 4, batch 1250, loss[loss=0.1677, simple_loss=0.25, pruned_loss=0.0427, over 12016.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2638, pruned_loss=0.04389, over 2393329.91 frames. ], batch size: 31, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:11:42,255 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137953.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:11:44,174 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137956.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:12:20,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-05-15 23:12:21,033 INFO [finetune.py:992] (1/2) Epoch 4, batch 1300, loss[loss=0.1942, simple_loss=0.2828, pruned_loss=0.05277, over 12134.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2638, pruned_loss=0.04397, over 2388014.59 frames. ], batch size: 39, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:12:38,027 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.233e+02 2.960e+02 3.353e+02 3.979e+02 8.906e+02, threshold=6.705e+02, percent-clipped=5.0 2023-05-15 23:12:54,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-15 23:12:56,442 INFO [finetune.py:992] (1/2) Epoch 4, batch 1350, loss[loss=0.1965, simple_loss=0.2916, pruned_loss=0.0507, over 12276.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.264, pruned_loss=0.04394, over 2386515.98 frames. ], batch size: 37, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:13:15,182 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5423, 2.4380, 3.2823, 4.4686, 2.3540, 4.4597, 4.5595, 4.7287], device='cuda:1'), covar=tensor([0.0122, 0.1203, 0.0420, 0.0113, 0.1206, 0.0189, 0.0125, 0.0061], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0201, 0.0183, 0.0112, 0.0184, 0.0170, 0.0165, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:13:32,558 INFO [finetune.py:992] (1/2) Epoch 4, batch 1400, loss[loss=0.1636, simple_loss=0.2534, pruned_loss=0.03692, over 12361.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2646, pruned_loss=0.04426, over 2394547.79 frames. ], batch size: 36, lr: 4.83e-03, grad_scale: 8.0 2023-05-15 23:13:50,457 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.116e+02 2.931e+02 3.271e+02 3.889e+02 6.456e+02, threshold=6.542e+02, percent-clipped=0.0 2023-05-15 23:14:08,833 INFO [finetune.py:992] (1/2) Epoch 4, batch 1450, loss[loss=0.1893, simple_loss=0.2777, pruned_loss=0.05042, over 12116.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2632, pruned_loss=0.04377, over 2387583.91 frames. ], batch size: 39, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:14:32,012 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0572, 3.6733, 5.4111, 2.7481, 2.9874, 3.9022, 3.4832, 4.1298], device='cuda:1'), covar=tensor([0.0347, 0.0949, 0.0198, 0.1095, 0.1796, 0.1284, 0.1166, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0223, 0.0223, 0.0176, 0.0232, 0.0273, 0.0220, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:14:44,733 INFO [finetune.py:992] (1/2) Epoch 4, batch 1500, loss[loss=0.1574, simple_loss=0.2352, pruned_loss=0.03983, over 12290.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04364, over 2389587.64 frames. ], batch size: 28, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:14:53,541 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5119, 5.3682, 5.4471, 5.5237, 5.1090, 5.1447, 4.9906, 5.4562], device='cuda:1'), covar=tensor([0.0661, 0.0521, 0.0683, 0.0536, 0.1841, 0.1387, 0.0521, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0624, 0.0534, 0.0584, 0.0764, 0.0696, 0.0506, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:15:02,647 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.957e+02 3.712e+02 4.461e+02 1.458e+03, threshold=7.425e+02, percent-clipped=4.0 2023-05-15 23:15:18,378 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138248.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:15:20,512 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138251.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:15:21,187 INFO [finetune.py:992] (1/2) Epoch 4, batch 1550, loss[loss=0.1591, simple_loss=0.2534, pruned_loss=0.03241, over 12355.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2625, pruned_loss=0.04347, over 2384522.46 frames. ], batch size: 35, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:15:34,729 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138270.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:15:57,252 INFO [finetune.py:992] (1/2) Epoch 4, batch 1600, loss[loss=0.1698, simple_loss=0.256, pruned_loss=0.04181, over 12087.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2629, pruned_loss=0.04377, over 2383909.94 frames. ], batch size: 32, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:16:11,513 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138322.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:16:14,172 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.128e+02 2.944e+02 3.290e+02 4.014e+02 5.311e+02, threshold=6.580e+02, percent-clipped=0.0 2023-05-15 23:16:18,001 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138331.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:16:30,592 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4808, 2.5424, 3.1940, 4.3331, 2.1601, 4.4058, 4.4499, 4.6268], device='cuda:1'), covar=tensor([0.0119, 0.1197, 0.0470, 0.0142, 0.1331, 0.0246, 0.0126, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0203, 0.0187, 0.0114, 0.0187, 0.0174, 0.0167, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:16:32,552 INFO [finetune.py:992] (1/2) Epoch 4, batch 1650, loss[loss=0.1961, simple_loss=0.2864, pruned_loss=0.05289, over 12047.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2629, pruned_loss=0.04362, over 2381981.77 frames. ], batch size: 37, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:16:55,446 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138383.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:17:08,935 INFO [finetune.py:992] (1/2) Epoch 4, batch 1700, loss[loss=0.1835, simple_loss=0.2666, pruned_loss=0.05013, over 12247.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2644, pruned_loss=0.04425, over 2376260.10 frames. ], batch size: 32, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:17:26,531 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 2.850e+02 3.347e+02 4.033e+02 6.121e+02, threshold=6.694e+02, percent-clipped=0.0 2023-05-15 23:17:44,969 INFO [finetune.py:992] (1/2) Epoch 4, batch 1750, loss[loss=0.1774, simple_loss=0.2497, pruned_loss=0.05259, over 12254.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2646, pruned_loss=0.04435, over 2372767.96 frames. ], batch size: 32, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:17:46,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-15 23:17:56,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-15 23:18:17,352 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4333, 4.9903, 5.3874, 4.7537, 5.0248, 4.8325, 5.4565, 5.0605], device='cuda:1'), covar=tensor([0.0212, 0.0278, 0.0237, 0.0219, 0.0312, 0.0257, 0.0163, 0.0193], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0240, 0.0258, 0.0233, 0.0232, 0.0237, 0.0213, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:18:20,203 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8358, 3.6665, 3.6763, 3.7735, 3.4290, 3.9240, 3.8589, 4.0036], device='cuda:1'), covar=tensor([0.0206, 0.0179, 0.0184, 0.0383, 0.0718, 0.0295, 0.0171, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0177, 0.0177, 0.0229, 0.0225, 0.0197, 0.0162, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 23:18:20,747 INFO [finetune.py:992] (1/2) Epoch 4, batch 1800, loss[loss=0.2761, simple_loss=0.3305, pruned_loss=0.1109, over 8059.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2634, pruned_loss=0.04393, over 2373963.33 frames. ], batch size: 99, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:18:37,977 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 2.878e+02 3.516e+02 4.441e+02 6.847e+02, threshold=7.032e+02, percent-clipped=2.0 2023-05-15 23:18:54,031 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138548.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:18:56,001 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138551.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:18:56,583 INFO [finetune.py:992] (1/2) Epoch 4, batch 1850, loss[loss=0.1826, simple_loss=0.2786, pruned_loss=0.0433, over 12046.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2645, pruned_loss=0.0445, over 2370452.36 frames. ], batch size: 40, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:19:11,987 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-15 23:19:28,499 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138596.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:19:30,585 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138599.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:19:32,998 INFO [finetune.py:992] (1/2) Epoch 4, batch 1900, loss[loss=0.1797, simple_loss=0.2591, pruned_loss=0.05015, over 12173.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.264, pruned_loss=0.04434, over 2375481.86 frames. ], batch size: 31, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:19:49,916 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.190e+02 2.936e+02 3.314e+02 4.022e+02 9.144e+02, threshold=6.629e+02, percent-clipped=5.0 2023-05-15 23:19:50,074 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138626.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:20:08,463 INFO [finetune.py:992] (1/2) Epoch 4, batch 1950, loss[loss=0.1875, simple_loss=0.2776, pruned_loss=0.0487, over 12190.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2637, pruned_loss=0.04438, over 2379194.88 frames. ], batch size: 35, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:20:27,595 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138678.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:20:29,729 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138681.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:20:33,247 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138686.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:20:41,000 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138697.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:20:44,389 INFO [finetune.py:992] (1/2) Epoch 4, batch 2000, loss[loss=0.1518, simple_loss=0.2374, pruned_loss=0.0331, over 11810.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2631, pruned_loss=0.04425, over 2380274.72 frames. ], batch size: 26, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:20:45,311 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6120, 2.6247, 4.3710, 4.6568, 2.8991, 2.5295, 2.7869, 2.0225], device='cuda:1'), covar=tensor([0.1365, 0.3280, 0.0445, 0.0341, 0.1103, 0.2106, 0.2537, 0.3885], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0370, 0.0263, 0.0283, 0.0250, 0.0280, 0.0348, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:21:02,091 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.852e+02 3.308e+02 4.055e+02 7.312e+02, threshold=6.616e+02, percent-clipped=3.0 2023-05-15 23:21:13,610 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138742.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:21:17,197 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138747.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:21:18,694 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0994, 4.3790, 3.9003, 4.8238, 4.4107, 2.7717, 4.0863, 3.0217], device='cuda:1'), covar=tensor([0.0847, 0.0839, 0.1403, 0.0371, 0.0961, 0.1516, 0.0967, 0.2830], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0358, 0.0339, 0.0249, 0.0347, 0.0253, 0.0324, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:21:20,479 INFO [finetune.py:992] (1/2) Epoch 4, batch 2050, loss[loss=0.1901, simple_loss=0.2792, pruned_loss=0.05055, over 11247.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2627, pruned_loss=0.04443, over 2380242.28 frames. ], batch size: 55, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:21:24,912 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138758.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:21:34,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5184, 5.3727, 5.3654, 4.8533, 4.8109, 5.5578, 4.6443, 4.9900], device='cuda:1'), covar=tensor([0.1429, 0.1766, 0.1210, 0.2784, 0.1287, 0.1267, 0.3683, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0513, 0.0484, 0.0580, 0.0384, 0.0656, 0.0719, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:21:56,177 INFO [finetune.py:992] (1/2) Epoch 4, batch 2100, loss[loss=0.1797, simple_loss=0.2631, pruned_loss=0.04818, over 12287.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2634, pruned_loss=0.04476, over 2374048.18 frames. ], batch size: 33, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:22:13,176 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.939e+02 2.809e+02 3.283e+02 4.095e+02 9.466e+02, threshold=6.566e+02, percent-clipped=3.0 2023-05-15 23:22:32,001 INFO [finetune.py:992] (1/2) Epoch 4, batch 2150, loss[loss=0.1661, simple_loss=0.2513, pruned_loss=0.04047, over 12083.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2629, pruned_loss=0.04432, over 2380539.09 frames. ], batch size: 32, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:22:36,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-05-15 23:22:41,460 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4593, 4.8046, 2.9730, 2.6849, 4.2313, 2.6611, 4.1814, 3.3466], device='cuda:1'), covar=tensor([0.0618, 0.0587, 0.1038, 0.1444, 0.0257, 0.1258, 0.0427, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0241, 0.0171, 0.0197, 0.0134, 0.0177, 0.0188, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:23:08,026 INFO [finetune.py:992] (1/2) Epoch 4, batch 2200, loss[loss=0.1666, simple_loss=0.2568, pruned_loss=0.03823, over 12118.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.262, pruned_loss=0.04364, over 2379665.28 frames. ], batch size: 33, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:23:13,084 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138909.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:23:22,393 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5726, 2.3669, 3.4178, 4.4022, 2.4635, 4.3844, 4.5186, 4.6862], device='cuda:1'), covar=tensor([0.0122, 0.1234, 0.0382, 0.0130, 0.1181, 0.0277, 0.0158, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0199, 0.0183, 0.0112, 0.0184, 0.0173, 0.0165, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:23:24,911 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.112e+02 2.850e+02 3.488e+02 4.120e+02 8.663e+02, threshold=6.977e+02, percent-clipped=4.0 2023-05-15 23:23:25,053 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138926.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:23:32,997 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9443, 3.3682, 5.1833, 2.6753, 2.8669, 3.9323, 3.4292, 3.9789], device='cuda:1'), covar=tensor([0.0314, 0.1101, 0.0255, 0.1145, 0.1793, 0.1198, 0.1213, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0228, 0.0229, 0.0179, 0.0235, 0.0277, 0.0224, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:23:43,431 INFO [finetune.py:992] (1/2) Epoch 4, batch 2250, loss[loss=0.1803, simple_loss=0.2762, pruned_loss=0.04223, over 12155.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2629, pruned_loss=0.04416, over 2380695.19 frames. ], batch size: 34, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:23:56,421 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138970.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 23:23:59,170 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138974.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:24:02,134 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138978.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:24:19,920 INFO [finetune.py:992] (1/2) Epoch 4, batch 2300, loss[loss=0.159, simple_loss=0.2485, pruned_loss=0.0348, over 12190.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2626, pruned_loss=0.04411, over 2382731.83 frames. ], batch size: 31, lr: 4.83e-03, grad_scale: 16.0 2023-05-15 23:24:26,361 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6934, 2.9130, 4.5096, 4.6822, 3.0190, 2.7779, 3.0292, 2.1210], device='cuda:1'), covar=tensor([0.1376, 0.2832, 0.0444, 0.0396, 0.1051, 0.1824, 0.2282, 0.3521], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0369, 0.0263, 0.0283, 0.0249, 0.0279, 0.0348, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:24:37,532 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.146e+02 2.924e+02 3.589e+02 4.110e+02 6.308e+02, threshold=7.178e+02, percent-clipped=0.0 2023-05-15 23:24:37,625 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139026.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:24:39,754 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5243, 5.1134, 5.5180, 4.8076, 5.1072, 4.8651, 5.5441, 5.1514], device='cuda:1'), covar=tensor([0.0189, 0.0237, 0.0185, 0.0200, 0.0285, 0.0238, 0.0181, 0.0234], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0241, 0.0258, 0.0236, 0.0233, 0.0237, 0.0214, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:24:40,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-15 23:24:45,414 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139037.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:24:48,885 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139042.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:24:56,060 INFO [finetune.py:992] (1/2) Epoch 4, batch 2350, loss[loss=0.1568, simple_loss=0.2497, pruned_loss=0.03198, over 12278.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.263, pruned_loss=0.04389, over 2386973.41 frames. ], batch size: 33, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:24:56,824 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139053.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:25:06,899 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4661, 3.6031, 3.2607, 3.1719, 2.8947, 2.7331, 3.6344, 2.2655], device='cuda:1'), covar=tensor([0.0333, 0.0120, 0.0173, 0.0163, 0.0336, 0.0318, 0.0104, 0.0413], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0155, 0.0148, 0.0176, 0.0199, 0.0191, 0.0153, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 23:25:26,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-15 23:25:31,484 INFO [finetune.py:992] (1/2) Epoch 4, batch 2400, loss[loss=0.1756, simple_loss=0.2663, pruned_loss=0.04245, over 12068.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2627, pruned_loss=0.04349, over 2382786.92 frames. ], batch size: 40, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:25:48,261 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 2.841e+02 3.377e+02 3.980e+02 6.478e+02, threshold=6.753e+02, percent-clipped=0.0 2023-05-15 23:26:08,255 INFO [finetune.py:992] (1/2) Epoch 4, batch 2450, loss[loss=0.1638, simple_loss=0.2344, pruned_loss=0.04661, over 12293.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2624, pruned_loss=0.04384, over 2375371.91 frames. ], batch size: 28, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:26:43,800 INFO [finetune.py:992] (1/2) Epoch 4, batch 2500, loss[loss=0.204, simple_loss=0.2955, pruned_loss=0.05623, over 10485.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2632, pruned_loss=0.04425, over 2373552.89 frames. ], batch size: 68, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:26:45,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-15 23:27:00,672 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.089e+02 2.899e+02 3.236e+02 3.898e+02 7.132e+02, threshold=6.472e+02, percent-clipped=1.0 2023-05-15 23:27:06,509 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3901, 5.1431, 5.2548, 5.3306, 4.8941, 5.0042, 4.7224, 5.2719], device='cuda:1'), covar=tensor([0.0586, 0.0634, 0.0747, 0.0645, 0.2087, 0.1321, 0.0635, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0634, 0.0539, 0.0598, 0.0778, 0.0702, 0.0515, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:27:09,451 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139238.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:27:10,959 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1504, 2.3798, 3.6850, 2.9753, 3.4653, 3.1161, 2.4116, 3.5635], device='cuda:1'), covar=tensor([0.0097, 0.0311, 0.0113, 0.0217, 0.0116, 0.0157, 0.0330, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0191, 0.0169, 0.0171, 0.0194, 0.0147, 0.0184, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:27:19,336 INFO [finetune.py:992] (1/2) Epoch 4, batch 2550, loss[loss=0.1598, simple_loss=0.2453, pruned_loss=0.03717, over 12259.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2639, pruned_loss=0.04439, over 2373281.39 frames. ], batch size: 32, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:27:28,651 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139265.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 23:27:36,492 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139276.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:27:47,012 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139289.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:27:54,162 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139299.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:27:56,012 INFO [finetune.py:992] (1/2) Epoch 4, batch 2600, loss[loss=0.1896, simple_loss=0.2717, pruned_loss=0.05375, over 12297.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2637, pruned_loss=0.04426, over 2370441.34 frames. ], batch size: 34, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:28:13,102 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.986e+02 3.375e+02 3.883e+02 7.968e+02, threshold=6.751e+02, percent-clipped=3.0 2023-05-15 23:28:21,079 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139337.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:28:21,192 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139337.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 23:28:24,445 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139342.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:28:30,076 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139350.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:28:31,312 INFO [finetune.py:992] (1/2) Epoch 4, batch 2650, loss[loss=0.1767, simple_loss=0.2679, pruned_loss=0.04279, over 11611.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2639, pruned_loss=0.04419, over 2364519.53 frames. ], batch size: 48, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:28:32,248 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139353.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:28:51,372 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9898, 4.1619, 4.0999, 4.4220, 2.9863, 3.9968, 2.6332, 4.0786], device='cuda:1'), covar=tensor([0.1751, 0.0634, 0.0972, 0.0656, 0.1075, 0.0563, 0.1865, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0260, 0.0294, 0.0352, 0.0237, 0.0237, 0.0258, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:28:54,712 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139385.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:28:58,206 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139390.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:29:06,362 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139401.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:29:07,038 INFO [finetune.py:992] (1/2) Epoch 4, batch 2700, loss[loss=0.1498, simple_loss=0.2437, pruned_loss=0.02797, over 12241.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2634, pruned_loss=0.04373, over 2375852.38 frames. ], batch size: 32, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:29:24,185 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 2.943e+02 3.380e+02 4.126e+02 9.261e+02, threshold=6.761e+02, percent-clipped=2.0 2023-05-15 23:29:43,194 INFO [finetune.py:992] (1/2) Epoch 4, batch 2750, loss[loss=0.1766, simple_loss=0.2717, pruned_loss=0.0408, over 12356.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2642, pruned_loss=0.0443, over 2370961.44 frames. ], batch size: 36, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:30:18,350 INFO [finetune.py:992] (1/2) Epoch 4, batch 2800, loss[loss=0.1594, simple_loss=0.2422, pruned_loss=0.03835, over 12365.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.263, pruned_loss=0.04399, over 2379517.34 frames. ], batch size: 30, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:30:35,534 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.134e+02 2.967e+02 3.317e+02 4.034e+02 8.886e+02, threshold=6.634e+02, percent-clipped=2.0 2023-05-15 23:30:35,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-15 23:30:41,660 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-15 23:30:53,851 INFO [finetune.py:992] (1/2) Epoch 4, batch 2850, loss[loss=0.162, simple_loss=0.2437, pruned_loss=0.04012, over 12020.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2627, pruned_loss=0.04366, over 2377534.78 frames. ], batch size: 28, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:31:03,359 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139565.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 23:31:08,378 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139571.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:31:10,624 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3427, 4.0926, 3.8185, 4.1854, 3.1787, 3.8147, 2.5488, 4.2705], device='cuda:1'), covar=tensor([0.1377, 0.0613, 0.1361, 0.0911, 0.0879, 0.0584, 0.1719, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0263, 0.0296, 0.0356, 0.0240, 0.0241, 0.0260, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:31:16,961 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139582.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:31:25,450 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139594.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:31:31,359 INFO [finetune.py:992] (1/2) Epoch 4, batch 2900, loss[loss=0.1572, simple_loss=0.2445, pruned_loss=0.03499, over 12089.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2623, pruned_loss=0.04338, over 2381710.01 frames. ], batch size: 33, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:31:32,879 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2960, 4.9418, 5.2848, 4.6778, 4.9941, 4.7055, 5.3347, 5.0224], device='cuda:1'), covar=tensor([0.0229, 0.0321, 0.0256, 0.0236, 0.0282, 0.0304, 0.0199, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0246, 0.0264, 0.0239, 0.0235, 0.0238, 0.0217, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:31:39,018 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139613.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:31:42,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0555, 5.0767, 4.8868, 4.9526, 4.5308, 5.0525, 5.0328, 5.3430], device='cuda:1'), covar=tensor([0.0291, 0.0121, 0.0199, 0.0271, 0.0795, 0.0301, 0.0133, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0179, 0.0180, 0.0228, 0.0229, 0.0200, 0.0163, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 23:31:48,250 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.018e+02 2.682e+02 3.469e+02 4.078e+02 1.175e+03, threshold=6.939e+02, percent-clipped=1.0 2023-05-15 23:31:52,699 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139632.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 23:31:52,828 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139632.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:32:00,595 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139643.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:32:01,917 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139645.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:32:06,700 INFO [finetune.py:992] (1/2) Epoch 4, batch 2950, loss[loss=0.1621, simple_loss=0.2399, pruned_loss=0.04217, over 12189.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2623, pruned_loss=0.04332, over 2383197.65 frames. ], batch size: 29, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:32:28,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-05-15 23:32:42,127 INFO [finetune.py:992] (1/2) Epoch 4, batch 3000, loss[loss=0.1809, simple_loss=0.281, pruned_loss=0.04043, over 12352.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2622, pruned_loss=0.04341, over 2377388.62 frames. ], batch size: 36, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:32:42,127 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-15 23:32:47,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8020, 4.6620, 4.8621, 4.8417, 4.4065, 4.4379, 4.3362, 4.7134], device='cuda:1'), covar=tensor([0.0743, 0.0585, 0.0602, 0.0543, 0.1852, 0.1509, 0.0708, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0631, 0.0540, 0.0595, 0.0776, 0.0701, 0.0515, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:32:49,869 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9826, 2.7801, 2.7326, 2.7184, 2.4291, 2.2911, 2.7447, 1.7297], device='cuda:1'), covar=tensor([0.0334, 0.0114, 0.0103, 0.0129, 0.0259, 0.0217, 0.0102, 0.0385], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0154, 0.0149, 0.0175, 0.0198, 0.0190, 0.0154, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 23:32:50,945 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7151, 4.7289, 4.6414, 4.5949, 4.0273, 4.7108, 4.6989, 4.8619], device='cuda:1'), covar=tensor([0.0242, 0.0118, 0.0204, 0.0328, 0.0836, 0.0336, 0.0187, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0180, 0.0181, 0.0230, 0.0231, 0.0201, 0.0165, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 23:32:57,296 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9614, 3.0425, 4.4626, 2.3801, 2.5413, 3.5740, 3.0357, 3.6623], device='cuda:1'), covar=tensor([0.0617, 0.1138, 0.0273, 0.1228, 0.1906, 0.1172, 0.1356, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0224, 0.0227, 0.0176, 0.0232, 0.0275, 0.0221, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:32:58,426 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4732, 4.1419, 4.4244, 4.0109, 4.1575, 4.0498, 4.4794, 4.4447], device='cuda:1'), covar=tensor([0.0280, 0.0345, 0.0265, 0.0213, 0.0300, 0.0312, 0.0180, 0.0162], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0245, 0.0263, 0.0238, 0.0236, 0.0238, 0.0216, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:33:00,431 INFO [finetune.py:1026] (1/2) Epoch 4, validation: loss=0.3291, simple_loss=0.4045, pruned_loss=0.1269, over 1020973.00 frames. 2023-05-15 23:33:00,432 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-15 23:33:04,898 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4104, 3.4762, 3.1407, 3.2351, 2.9134, 2.7181, 3.4577, 2.3312], device='cuda:1'), covar=tensor([0.0346, 0.0141, 0.0201, 0.0138, 0.0349, 0.0287, 0.0123, 0.0381], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0154, 0.0148, 0.0175, 0.0198, 0.0190, 0.0154, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-15 23:33:17,215 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 2.801e+02 3.363e+02 4.100e+02 7.836e+02, threshold=6.726e+02, percent-clipped=4.0 2023-05-15 23:33:27,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-15 23:33:35,834 INFO [finetune.py:992] (1/2) Epoch 4, batch 3050, loss[loss=0.1617, simple_loss=0.2434, pruned_loss=0.04007, over 12180.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.262, pruned_loss=0.04352, over 2377668.42 frames. ], batch size: 29, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:33:37,484 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2364, 2.3608, 3.7450, 3.1523, 3.6229, 3.3403, 2.5150, 3.7281], device='cuda:1'), covar=tensor([0.0105, 0.0332, 0.0099, 0.0192, 0.0098, 0.0133, 0.0303, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0192, 0.0169, 0.0171, 0.0195, 0.0148, 0.0185, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:33:47,937 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139769.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:02,300 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139789.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:11,584 INFO [finetune.py:992] (1/2) Epoch 4, batch 3100, loss[loss=0.1826, simple_loss=0.2718, pruned_loss=0.04674, over 12247.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2621, pruned_loss=0.04337, over 2380095.79 frames. ], batch size: 32, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:34:13,213 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139804.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:14,654 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139806.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:26,035 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0786, 5.1324, 4.9977, 5.0878, 4.6184, 5.1741, 5.1396, 5.3531], device='cuda:1'), covar=tensor([0.0255, 0.0112, 0.0163, 0.0250, 0.0727, 0.0197, 0.0134, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0180, 0.0181, 0.0230, 0.0229, 0.0200, 0.0164, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 23:34:29,390 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.867e+02 3.242e+02 3.780e+02 8.522e+02, threshold=6.484e+02, percent-clipped=1.0 2023-05-15 23:34:32,411 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139830.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:47,194 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139850.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:47,222 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139850.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:48,374 INFO [finetune.py:992] (1/2) Epoch 4, batch 3150, loss[loss=0.1906, simple_loss=0.2754, pruned_loss=0.05293, over 12099.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2613, pruned_loss=0.04317, over 2381259.99 frames. ], batch size: 42, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:34:48,537 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0802, 4.8252, 4.9672, 4.9799, 4.7261, 4.8703, 4.8335, 2.8904], device='cuda:1'), covar=tensor([0.0075, 0.0058, 0.0072, 0.0058, 0.0052, 0.0110, 0.0070, 0.0641], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0071, 0.0076, 0.0068, 0.0056, 0.0085, 0.0074, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:34:57,557 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139865.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:34:59,007 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139867.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:17,863 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139894.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:23,439 INFO [finetune.py:992] (1/2) Epoch 4, batch 3200, loss[loss=0.1675, simple_loss=0.2445, pruned_loss=0.04523, over 12015.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.262, pruned_loss=0.04392, over 2381733.04 frames. ], batch size: 28, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:35:30,057 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139911.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:40,345 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.833e+02 3.398e+02 3.955e+02 6.334e+02, threshold=6.796e+02, percent-clipped=0.0 2023-05-15 23:35:41,140 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139927.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:44,796 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139932.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 23:35:48,955 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139938.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:51,735 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139942.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:53,283 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9011, 3.0674, 4.7528, 4.8491, 3.0989, 2.8123, 3.0958, 2.3532], device='cuda:1'), covar=tensor([0.1255, 0.2541, 0.0404, 0.0419, 0.1061, 0.1888, 0.2379, 0.3218], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0368, 0.0263, 0.0284, 0.0252, 0.0280, 0.0348, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:35:53,826 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139945.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:35:56,606 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2732, 5.0879, 5.1816, 5.2462, 4.8478, 4.8865, 4.7130, 5.1298], device='cuda:1'), covar=tensor([0.0558, 0.0497, 0.0639, 0.0502, 0.1589, 0.1107, 0.0515, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0628, 0.0542, 0.0596, 0.0774, 0.0701, 0.0515, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:35:58,566 INFO [finetune.py:992] (1/2) Epoch 4, batch 3250, loss[loss=0.1674, simple_loss=0.2566, pruned_loss=0.03907, over 11229.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2626, pruned_loss=0.0441, over 2380081.31 frames. ], batch size: 55, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:36:11,355 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2062, 5.2763, 5.0749, 5.1968, 4.7226, 5.2241, 5.2605, 5.4936], device='cuda:1'), covar=tensor([0.0207, 0.0113, 0.0195, 0.0257, 0.0699, 0.0317, 0.0137, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0180, 0.0180, 0.0229, 0.0229, 0.0199, 0.0164, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-15 23:36:18,957 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139980.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:36:28,685 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139993.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:36:38,557 INFO [finetune.py:992] (1/2) Epoch 4, batch 3300, loss[loss=0.1879, simple_loss=0.2765, pruned_loss=0.04962, over 12142.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2626, pruned_loss=0.04434, over 2381909.77 frames. ], batch size: 36, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:36:55,093 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 2.787e+02 3.285e+02 4.062e+02 8.144e+02, threshold=6.569e+02, percent-clipped=3.0 2023-05-15 23:37:13,353 INFO [finetune.py:992] (1/2) Epoch 4, batch 3350, loss[loss=0.1908, simple_loss=0.2848, pruned_loss=0.04839, over 12305.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2621, pruned_loss=0.04422, over 2378236.69 frames. ], batch size: 34, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:37:49,410 INFO [finetune.py:992] (1/2) Epoch 4, batch 3400, loss[loss=0.1642, simple_loss=0.2584, pruned_loss=0.03499, over 12325.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2618, pruned_loss=0.04409, over 2376494.37 frames. ], batch size: 34, lr: 4.82e-03, grad_scale: 16.0 2023-05-15 23:37:50,312 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140103.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:37:53,350 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-05-15 23:37:55,236 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140110.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:38:05,630 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140125.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:38:06,249 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.929e+02 3.583e+02 4.170e+02 6.929e+02, threshold=7.166e+02, percent-clipped=2.0 2023-05-15 23:38:06,478 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7634, 2.9636, 3.7865, 4.8375, 4.1523, 4.7302, 4.0125, 3.3763], device='cuda:1'), covar=tensor([0.0025, 0.0337, 0.0123, 0.0026, 0.0100, 0.0073, 0.0094, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0118, 0.0099, 0.0074, 0.0098, 0.0111, 0.0086, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:38:20,401 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140145.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:38:25,303 INFO [finetune.py:992] (1/2) Epoch 4, batch 3450, loss[loss=0.1984, simple_loss=0.3027, pruned_loss=0.04707, over 12021.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2617, pruned_loss=0.04381, over 2384169.29 frames. ], batch size: 40, lr: 4.82e-03, grad_scale: 32.0 2023-05-15 23:38:31,208 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140160.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:38:32,672 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140162.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:38:34,180 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140164.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:38:39,162 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140171.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:39:01,154 INFO [finetune.py:992] (1/2) Epoch 4, batch 3500, loss[loss=0.1688, simple_loss=0.2557, pruned_loss=0.04096, over 12202.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2626, pruned_loss=0.04431, over 2376885.30 frames. ], batch size: 31, lr: 4.82e-03, grad_scale: 32.0 2023-05-15 23:39:04,144 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140206.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:39:18,478 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.882e+02 3.287e+02 3.945e+02 9.868e+02, threshold=6.574e+02, percent-clipped=1.0 2023-05-15 23:39:19,376 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140227.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:39:27,213 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140238.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:39:37,594 INFO [finetune.py:992] (1/2) Epoch 4, batch 3550, loss[loss=0.156, simple_loss=0.2353, pruned_loss=0.03832, over 12186.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2612, pruned_loss=0.04363, over 2379197.90 frames. ], batch size: 29, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:39:42,806 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140259.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:39:53,896 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140275.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:40:02,453 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140286.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:40:13,473 INFO [finetune.py:992] (1/2) Epoch 4, batch 3600, loss[loss=0.1635, simple_loss=0.2574, pruned_loss=0.03475, over 12195.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2617, pruned_loss=0.04382, over 2386286.28 frames. ], batch size: 35, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:40:15,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1783, 2.0141, 2.4059, 2.2301, 2.4046, 2.4463, 1.9108, 2.4454], device='cuda:1'), covar=tensor([0.0099, 0.0235, 0.0157, 0.0170, 0.0155, 0.0147, 0.0270, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0193, 0.0173, 0.0174, 0.0198, 0.0150, 0.0186, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:40:26,653 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140320.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:40:30,984 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1535, 4.9961, 5.0904, 5.1523, 4.7341, 4.8224, 4.6616, 5.1290], device='cuda:1'), covar=tensor([0.0725, 0.0661, 0.0828, 0.0576, 0.1973, 0.1358, 0.0508, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0629, 0.0547, 0.0598, 0.0780, 0.0706, 0.0517, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:40:31,532 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 2.818e+02 3.366e+02 3.811e+02 1.666e+03, threshold=6.733e+02, percent-clipped=4.0 2023-05-15 23:40:48,983 INFO [finetune.py:992] (1/2) Epoch 4, batch 3650, loss[loss=0.1665, simple_loss=0.256, pruned_loss=0.03851, over 10676.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2618, pruned_loss=0.0439, over 2383931.94 frames. ], batch size: 69, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:41:02,002 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1132, 6.0424, 5.5823, 5.5353, 6.0758, 5.4217, 5.6914, 5.5571], device='cuda:1'), covar=tensor([0.1394, 0.0918, 0.0917, 0.2048, 0.0981, 0.2205, 0.1468, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0458, 0.0353, 0.0408, 0.0430, 0.0413, 0.0362, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:41:03,633 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7599, 2.8129, 4.5600, 4.7398, 2.8806, 2.6686, 3.0182, 2.0591], device='cuda:1'), covar=tensor([0.1354, 0.2929, 0.0418, 0.0391, 0.1128, 0.2033, 0.2308, 0.3567], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0370, 0.0262, 0.0286, 0.0252, 0.0279, 0.0347, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:41:06,348 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4235, 5.0532, 5.4330, 4.7482, 5.0352, 4.7988, 5.4314, 4.8963], device='cuda:1'), covar=tensor([0.0217, 0.0281, 0.0217, 0.0209, 0.0299, 0.0278, 0.0188, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0241, 0.0260, 0.0234, 0.0233, 0.0235, 0.0214, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:41:14,741 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140387.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:41:25,429 INFO [finetune.py:992] (1/2) Epoch 4, batch 3700, loss[loss=0.1708, simple_loss=0.2616, pruned_loss=0.04005, over 12286.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2617, pruned_loss=0.04365, over 2377442.82 frames. ], batch size: 33, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:41:39,082 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7670, 3.3428, 5.1504, 2.9725, 2.7163, 3.9892, 3.3959, 4.0673], device='cuda:1'), covar=tensor([0.0434, 0.1164, 0.0313, 0.1028, 0.1938, 0.1208, 0.1286, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0228, 0.0231, 0.0180, 0.0235, 0.0279, 0.0226, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:41:41,781 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140425.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:41:43,076 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.199e+02 2.942e+02 3.360e+02 4.212e+02 8.800e+02, threshold=6.720e+02, percent-clipped=3.0 2023-05-15 23:41:56,678 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140445.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:41:58,626 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140448.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:01,166 INFO [finetune.py:992] (1/2) Epoch 4, batch 3750, loss[loss=0.1587, simple_loss=0.2458, pruned_loss=0.03581, over 12122.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2622, pruned_loss=0.04405, over 2373142.14 frames. ], batch size: 33, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:42:06,189 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140459.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:06,263 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5695, 5.3510, 5.4768, 5.5089, 5.0946, 5.1697, 4.9614, 5.4548], device='cuda:1'), covar=tensor([0.0518, 0.0488, 0.0692, 0.0485, 0.1728, 0.1184, 0.0489, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0632, 0.0547, 0.0598, 0.0778, 0.0706, 0.0517, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:42:07,011 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140460.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:08,477 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140462.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:11,329 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140466.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:16,197 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140473.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:27,540 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140489.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:30,347 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140493.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:35,342 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5647, 2.6272, 3.5757, 4.4899, 3.7905, 4.4885, 3.8958, 3.0561], device='cuda:1'), covar=tensor([0.0022, 0.0331, 0.0142, 0.0039, 0.0111, 0.0058, 0.0092, 0.0318], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0118, 0.0099, 0.0074, 0.0098, 0.0110, 0.0085, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:42:36,576 INFO [finetune.py:992] (1/2) Epoch 4, batch 3800, loss[loss=0.1481, simple_loss=0.2397, pruned_loss=0.02825, over 12155.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2621, pruned_loss=0.04377, over 2374860.92 frames. ], batch size: 36, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:42:39,581 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140506.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:40,896 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140508.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:41,754 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140509.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:42,364 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140510.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:42:45,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 2023-05-15 23:42:54,131 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.899e+02 3.467e+02 4.265e+02 8.161e+02, threshold=6.934e+02, percent-clipped=4.0 2023-05-15 23:43:11,491 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140550.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:43:12,586 INFO [finetune.py:992] (1/2) Epoch 4, batch 3850, loss[loss=0.241, simple_loss=0.2973, pruned_loss=0.09232, over 8237.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.262, pruned_loss=0.04381, over 2368508.35 frames. ], batch size: 98, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:43:13,482 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4129, 5.2160, 5.3218, 5.3681, 4.9524, 5.0594, 4.8200, 5.3074], device='cuda:1'), covar=tensor([0.0600, 0.0551, 0.0743, 0.0553, 0.1786, 0.1196, 0.0512, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0632, 0.0546, 0.0596, 0.0779, 0.0704, 0.0517, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:43:14,115 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140554.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:43:25,446 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140570.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:43:49,023 INFO [finetune.py:992] (1/2) Epoch 4, batch 3900, loss[loss=0.1692, simple_loss=0.2592, pruned_loss=0.03962, over 12116.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2626, pruned_loss=0.04392, over 2373761.98 frames. ], batch size: 30, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:43:55,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-15 23:43:58,502 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140615.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:44:06,951 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.830e+02 3.285e+02 3.854e+02 6.178e+02, threshold=6.570e+02, percent-clipped=0.0 2023-05-15 23:44:19,638 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8760, 2.9910, 4.8827, 5.0392, 3.3791, 3.0483, 3.1449, 2.3024], device='cuda:1'), covar=tensor([0.1346, 0.3046, 0.0391, 0.0322, 0.0997, 0.1768, 0.2465, 0.3554], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0375, 0.0266, 0.0292, 0.0254, 0.0282, 0.0353, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:44:24,233 INFO [finetune.py:992] (1/2) Epoch 4, batch 3950, loss[loss=0.2128, simple_loss=0.3015, pruned_loss=0.06204, over 12061.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2624, pruned_loss=0.04375, over 2375544.84 frames. ], batch size: 42, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:45:00,284 INFO [finetune.py:992] (1/2) Epoch 4, batch 4000, loss[loss=0.1872, simple_loss=0.2708, pruned_loss=0.05178, over 11213.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2631, pruned_loss=0.04408, over 2366469.50 frames. ], batch size: 55, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:45:10,358 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-15 23:45:15,644 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140723.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:45:18,332 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.843e+02 3.354e+02 4.129e+02 8.480e+02, threshold=6.707e+02, percent-clipped=3.0 2023-05-15 23:45:29,861 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140743.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:45:36,189 INFO [finetune.py:992] (1/2) Epoch 4, batch 4050, loss[loss=0.1676, simple_loss=0.2432, pruned_loss=0.04596, over 12282.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.263, pruned_loss=0.04384, over 2376519.73 frames. ], batch size: 28, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:45:41,388 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140759.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:45:44,956 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7258, 2.8603, 3.6905, 4.7313, 3.9726, 4.6418, 4.0659, 3.2607], device='cuda:1'), covar=tensor([0.0026, 0.0311, 0.0120, 0.0034, 0.0103, 0.0057, 0.0079, 0.0300], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0117, 0.0099, 0.0073, 0.0097, 0.0110, 0.0085, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:45:46,305 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140766.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:45:59,165 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140784.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:46:12,212 INFO [finetune.py:992] (1/2) Epoch 4, batch 4100, loss[loss=0.167, simple_loss=0.2575, pruned_loss=0.03822, over 11248.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2623, pruned_loss=0.04365, over 2381661.08 frames. ], batch size: 55, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:46:15,837 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140807.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:46:17,346 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140809.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:46:20,607 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140814.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:46:22,848 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3069, 4.7272, 2.8395, 2.6912, 3.9819, 2.6836, 4.0444, 3.1453], device='cuda:1'), covar=tensor([0.0671, 0.0458, 0.1164, 0.1480, 0.0257, 0.1196, 0.0399, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0249, 0.0174, 0.0201, 0.0138, 0.0179, 0.0193, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:46:29,831 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.854e+02 3.326e+02 3.979e+02 7.859e+02, threshold=6.651e+02, percent-clipped=1.0 2023-05-15 23:46:43,322 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140845.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:46:48,264 INFO [finetune.py:992] (1/2) Epoch 4, batch 4150, loss[loss=0.1762, simple_loss=0.2621, pruned_loss=0.04519, over 12177.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2625, pruned_loss=0.04352, over 2383223.61 frames. ], batch size: 31, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:46:58,298 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140865.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:47:01,868 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140870.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:47:24,241 INFO [finetune.py:992] (1/2) Epoch 4, batch 4200, loss[loss=0.2185, simple_loss=0.3089, pruned_loss=0.06408, over 11350.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2627, pruned_loss=0.04339, over 2383704.79 frames. ], batch size: 55, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:47:33,479 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140915.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:47:34,914 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140917.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:47:41,779 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.886e+02 3.319e+02 3.814e+02 7.969e+02, threshold=6.638e+02, percent-clipped=2.0 2023-05-15 23:47:48,481 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4702, 3.6253, 3.3419, 3.1857, 3.0168, 2.9086, 3.6864, 2.2610], device='cuda:1'), covar=tensor([0.0323, 0.0139, 0.0127, 0.0164, 0.0293, 0.0251, 0.0126, 0.0408], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0159, 0.0151, 0.0178, 0.0200, 0.0191, 0.0155, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:47:56,285 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140947.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:47:59,689 INFO [finetune.py:992] (1/2) Epoch 4, batch 4250, loss[loss=0.1619, simple_loss=0.2459, pruned_loss=0.03894, over 12023.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2627, pruned_loss=0.04336, over 2387361.00 frames. ], batch size: 31, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:48:07,675 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140963.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:48:18,476 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140978.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:48:36,363 INFO [finetune.py:992] (1/2) Epoch 4, batch 4300, loss[loss=0.1893, simple_loss=0.2713, pruned_loss=0.05365, over 12099.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.263, pruned_loss=0.044, over 2370390.00 frames. ], batch size: 33, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:48:41,594 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141008.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:48:42,351 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3540, 3.1785, 4.8029, 2.6002, 2.7984, 3.6759, 3.1978, 3.7805], device='cuda:1'), covar=tensor([0.0517, 0.1125, 0.0304, 0.1117, 0.1679, 0.1344, 0.1247, 0.0965], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0227, 0.0231, 0.0179, 0.0234, 0.0280, 0.0225, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:48:54,922 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.741e+02 3.149e+02 3.897e+02 7.477e+02, threshold=6.298e+02, percent-clipped=1.0 2023-05-15 23:49:06,350 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141043.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:49:12,618 INFO [finetune.py:992] (1/2) Epoch 4, batch 4350, loss[loss=0.144, simple_loss=0.2225, pruned_loss=0.03273, over 12268.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2637, pruned_loss=0.0442, over 2364726.92 frames. ], batch size: 28, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:49:17,154 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6967, 2.6581, 3.9898, 4.1378, 2.9142, 2.6724, 2.7268, 2.1529], device='cuda:1'), covar=tensor([0.1241, 0.2466, 0.0482, 0.0465, 0.1014, 0.1781, 0.2302, 0.3359], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0372, 0.0264, 0.0288, 0.0252, 0.0279, 0.0349, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:49:31,768 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141079.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:49:40,179 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141091.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:49:47,706 INFO [finetune.py:992] (1/2) Epoch 4, batch 4400, loss[loss=0.1711, simple_loss=0.2689, pruned_loss=0.03665, over 12062.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2638, pruned_loss=0.04399, over 2365273.54 frames. ], batch size: 40, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:50:05,759 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 3.044e+02 3.595e+02 4.304e+02 1.021e+03, threshold=7.190e+02, percent-clipped=4.0 2023-05-15 23:50:06,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-15 23:50:17,209 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141143.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:50:18,539 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141145.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:50:24,146 INFO [finetune.py:992] (1/2) Epoch 4, batch 4450, loss[loss=0.1676, simple_loss=0.2494, pruned_loss=0.04294, over 12372.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2649, pruned_loss=0.04444, over 2367222.98 frames. ], batch size: 30, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:50:28,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-15 23:50:33,473 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141165.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:50:33,500 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141165.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:50:52,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-05-15 23:50:53,124 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141193.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:50:59,538 INFO [finetune.py:992] (1/2) Epoch 4, batch 4500, loss[loss=0.1693, simple_loss=0.2615, pruned_loss=0.03857, over 12301.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2655, pruned_loss=0.04493, over 2362049.40 frames. ], batch size: 34, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:51:01,126 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141204.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:51:07,376 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141213.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:51:16,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-05-15 23:51:17,112 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.809e+02 3.455e+02 4.408e+02 1.258e+03, threshold=6.910e+02, percent-clipped=4.0 2023-05-15 23:51:34,778 INFO [finetune.py:992] (1/2) Epoch 4, batch 4550, loss[loss=0.1849, simple_loss=0.2898, pruned_loss=0.04001, over 12131.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2647, pruned_loss=0.04445, over 2366348.64 frames. ], batch size: 36, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:51:50,395 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141273.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:52:11,296 INFO [finetune.py:992] (1/2) Epoch 4, batch 4600, loss[loss=0.172, simple_loss=0.2621, pruned_loss=0.04095, over 12368.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2639, pruned_loss=0.0442, over 2369973.37 frames. ], batch size: 38, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:52:12,005 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141303.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:52:25,417 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2200, 4.8597, 5.0722, 5.1302, 4.7847, 5.0952, 4.9529, 2.7821], device='cuda:1'), covar=tensor([0.0073, 0.0056, 0.0059, 0.0048, 0.0045, 0.0074, 0.0058, 0.0622], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0072, 0.0076, 0.0069, 0.0057, 0.0086, 0.0074, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-15 23:52:28,862 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.669e+02 3.274e+02 3.911e+02 6.807e+02, threshold=6.547e+02, percent-clipped=0.0 2023-05-15 23:52:30,488 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141329.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:52:31,425 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.22 vs. limit=5.0 2023-05-15 23:52:46,733 INFO [finetune.py:992] (1/2) Epoch 4, batch 4650, loss[loss=0.1752, simple_loss=0.2723, pruned_loss=0.03902, over 12055.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2639, pruned_loss=0.04401, over 2373062.43 frames. ], batch size: 37, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:53:05,775 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141379.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:53:11,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0808, 4.0494, 3.8341, 4.0304, 2.9642, 3.7922, 2.5673, 3.7122], device='cuda:1'), covar=tensor([0.1573, 0.0457, 0.0628, 0.0446, 0.0928, 0.0520, 0.1562, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0264, 0.0296, 0.0353, 0.0239, 0.0238, 0.0257, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:53:13,647 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141390.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:53:22,573 INFO [finetune.py:992] (1/2) Epoch 4, batch 4700, loss[loss=0.1527, simple_loss=0.2449, pruned_loss=0.03027, over 12290.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2634, pruned_loss=0.04402, over 2377431.49 frames. ], batch size: 33, lr: 4.81e-03, grad_scale: 16.0 2023-05-15 23:53:33,500 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141417.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 23:53:41,003 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.952e+02 3.312e+02 4.023e+02 9.384e+02, threshold=6.625e+02, percent-clipped=1.0 2023-05-15 23:53:41,091 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141427.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:53:51,069 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141441.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:53:59,444 INFO [finetune.py:992] (1/2) Epoch 4, batch 4750, loss[loss=0.1618, simple_loss=0.247, pruned_loss=0.03832, over 12129.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2631, pruned_loss=0.04385, over 2374341.75 frames. ], batch size: 30, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:54:08,707 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141465.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:54:17,834 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141478.0, num_to_drop=1, layers_to_drop={0} 2023-05-15 23:54:32,532 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141499.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:54:34,605 INFO [finetune.py:992] (1/2) Epoch 4, batch 4800, loss[loss=0.1615, simple_loss=0.239, pruned_loss=0.04195, over 11796.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2633, pruned_loss=0.04388, over 2377037.60 frames. ], batch size: 26, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:54:34,814 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141502.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:54:36,927 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4355, 5.2446, 5.3111, 5.3714, 5.0016, 5.0780, 4.8632, 5.3493], device='cuda:1'), covar=tensor([0.0573, 0.0495, 0.0648, 0.0527, 0.1649, 0.1214, 0.0466, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0633, 0.0550, 0.0589, 0.0777, 0.0701, 0.0513, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-15 23:54:42,299 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141513.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:54:52,291 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.852e+02 3.329e+02 3.958e+02 9.403e+02, threshold=6.658e+02, percent-clipped=2.0 2023-05-15 23:55:09,981 INFO [finetune.py:992] (1/2) Epoch 4, batch 4850, loss[loss=0.2085, simple_loss=0.2986, pruned_loss=0.05916, over 12266.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2633, pruned_loss=0.04393, over 2376072.67 frames. ], batch size: 37, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:55:25,210 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141573.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:55:29,581 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3839, 4.8780, 5.3554, 4.6907, 4.9707, 4.6824, 5.3789, 5.0468], device='cuda:1'), covar=tensor([0.0237, 0.0324, 0.0257, 0.0241, 0.0331, 0.0290, 0.0196, 0.0230], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0246, 0.0263, 0.0240, 0.0239, 0.0242, 0.0219, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:55:31,423 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-05-15 23:55:40,972 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4805, 2.5414, 3.2867, 4.2053, 2.2346, 4.4059, 4.3409, 4.5612], device='cuda:1'), covar=tensor([0.0097, 0.1093, 0.0413, 0.0141, 0.1207, 0.0194, 0.0144, 0.0069], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0201, 0.0185, 0.0116, 0.0187, 0.0175, 0.0169, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:55:46,764 INFO [finetune.py:992] (1/2) Epoch 4, batch 4900, loss[loss=0.1689, simple_loss=0.2461, pruned_loss=0.0458, over 12188.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2633, pruned_loss=0.04403, over 2380802.06 frames. ], batch size: 29, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:55:47,634 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141603.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:56:00,187 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141621.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:56:04,338 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 2.647e+02 3.128e+02 3.697e+02 6.289e+02, threshold=6.257e+02, percent-clipped=0.0 2023-05-15 23:56:11,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 2023-05-15 23:56:21,355 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141651.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:56:22,006 INFO [finetune.py:992] (1/2) Epoch 4, batch 4950, loss[loss=0.172, simple_loss=0.2687, pruned_loss=0.03762, over 12360.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2619, pruned_loss=0.04359, over 2384536.78 frames. ], batch size: 36, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:56:27,145 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1812, 4.0229, 4.0896, 4.4916, 3.1943, 3.9385, 2.6132, 4.1145], device='cuda:1'), covar=tensor([0.1685, 0.0766, 0.0947, 0.0670, 0.1008, 0.0627, 0.1894, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0267, 0.0297, 0.0355, 0.0240, 0.0240, 0.0259, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:56:45,272 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141685.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:56:53,892 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2413, 4.8405, 5.2297, 4.5847, 4.9246, 4.6083, 5.2769, 4.8463], device='cuda:1'), covar=tensor([0.0209, 0.0307, 0.0244, 0.0267, 0.0322, 0.0292, 0.0171, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0245, 0.0263, 0.0239, 0.0238, 0.0240, 0.0218, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-15 23:56:57,377 INFO [finetune.py:992] (1/2) Epoch 4, batch 5000, loss[loss=0.1684, simple_loss=0.2653, pruned_loss=0.03573, over 11252.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04356, over 2381596.01 frames. ], batch size: 55, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:57:15,456 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.970e+02 3.491e+02 4.383e+02 8.620e+02, threshold=6.983e+02, percent-clipped=4.0 2023-05-15 23:57:33,775 INFO [finetune.py:992] (1/2) Epoch 4, batch 5050, loss[loss=0.1841, simple_loss=0.2789, pruned_loss=0.0446, over 11513.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.263, pruned_loss=0.04385, over 2370593.97 frames. ], batch size: 55, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:57:48,595 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141773.0, num_to_drop=1, layers_to_drop={1} 2023-05-15 23:58:05,453 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141797.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:58:06,916 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141799.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:58:09,332 INFO [finetune.py:992] (1/2) Epoch 4, batch 5100, loss[loss=0.1513, simple_loss=0.2329, pruned_loss=0.03485, over 12257.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2622, pruned_loss=0.04358, over 2378424.08 frames. ], batch size: 28, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:58:11,759 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1319, 4.0518, 3.9662, 4.4086, 3.0169, 3.8513, 2.7074, 3.9880], device='cuda:1'), covar=tensor([0.1631, 0.0664, 0.0908, 0.0641, 0.1031, 0.0632, 0.1650, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0266, 0.0298, 0.0354, 0.0240, 0.0240, 0.0259, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-15 23:58:27,076 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.257e+02 2.764e+02 3.422e+02 4.267e+02 1.010e+03, threshold=6.843e+02, percent-clipped=2.0 2023-05-15 23:58:40,770 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141847.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:58:44,288 INFO [finetune.py:992] (1/2) Epoch 4, batch 5150, loss[loss=0.1737, simple_loss=0.2539, pruned_loss=0.04672, over 12067.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2629, pruned_loss=0.04394, over 2378780.26 frames. ], batch size: 32, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:59:21,246 INFO [finetune.py:992] (1/2) Epoch 4, batch 5200, loss[loss=0.1854, simple_loss=0.2798, pruned_loss=0.04552, over 12351.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2625, pruned_loss=0.04381, over 2378312.42 frames. ], batch size: 36, lr: 4.80e-03, grad_scale: 16.0 2023-05-15 23:59:31,086 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141916.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:59:38,870 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.200e+02 2.841e+02 3.341e+02 4.056e+02 6.577e+02, threshold=6.683e+02, percent-clipped=0.0 2023-05-15 23:59:39,813 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141928.0, num_to_drop=0, layers_to_drop=set() 2023-05-15 23:59:46,917 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3071, 2.2935, 3.1139, 4.0894, 1.8832, 4.2423, 4.2077, 4.3723], device='cuda:1'), covar=tensor([0.0118, 0.1160, 0.0451, 0.0158, 0.1441, 0.0216, 0.0178, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0200, 0.0183, 0.0115, 0.0188, 0.0174, 0.0168, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-15 23:59:56,516 INFO [finetune.py:992] (1/2) Epoch 4, batch 5250, loss[loss=0.19, simple_loss=0.2791, pruned_loss=0.05048, over 12117.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2638, pruned_loss=0.04445, over 2373396.42 frames. ], batch size: 38, lr: 4.80e-03, grad_scale: 16.0 2023-05-16 00:00:14,695 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141977.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:00:20,208 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141985.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:00:23,202 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141989.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:00:24,601 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1799, 4.7754, 5.1498, 4.6186, 4.8353, 4.5762, 5.2073, 4.8217], device='cuda:1'), covar=tensor([0.0299, 0.0395, 0.0307, 0.0224, 0.0345, 0.0318, 0.0203, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0248, 0.0265, 0.0240, 0.0238, 0.0242, 0.0219, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:00:35,454 INFO [finetune.py:992] (1/2) Epoch 4, batch 5300, loss[loss=0.1501, simple_loss=0.2295, pruned_loss=0.03536, over 12001.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2635, pruned_loss=0.04403, over 2371705.33 frames. ], batch size: 28, lr: 4.80e-03, grad_scale: 16.0 2023-05-16 00:00:42,655 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0517, 6.0801, 5.8517, 5.2607, 5.0975, 5.8949, 5.5554, 5.3074], device='cuda:1'), covar=tensor([0.0565, 0.0604, 0.0583, 0.1373, 0.0583, 0.0720, 0.1322, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0565, 0.0505, 0.0477, 0.0582, 0.0378, 0.0658, 0.0711, 0.0531], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:00:54,338 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.901e+02 3.592e+02 4.199e+02 7.190e+02, threshold=7.183e+02, percent-clipped=1.0 2023-05-16 00:00:58,586 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142033.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:01:12,238 INFO [finetune.py:992] (1/2) Epoch 4, batch 5350, loss[loss=0.1881, simple_loss=0.2809, pruned_loss=0.04759, over 12199.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04378, over 2372598.66 frames. ], batch size: 35, lr: 4.80e-03, grad_scale: 16.0 2023-05-16 00:01:26,011 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3993, 4.7249, 3.0631, 2.8305, 3.9615, 2.6896, 4.0162, 3.2886], device='cuda:1'), covar=tensor([0.0583, 0.0420, 0.0928, 0.1256, 0.0275, 0.1169, 0.0439, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0249, 0.0172, 0.0198, 0.0138, 0.0178, 0.0192, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:01:27,309 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142073.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 00:01:32,678 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 00:01:43,651 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4239, 2.4915, 3.2612, 4.1722, 2.3067, 4.4119, 4.3950, 4.5050], device='cuda:1'), covar=tensor([0.0127, 0.1164, 0.0416, 0.0168, 0.1238, 0.0169, 0.0142, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0199, 0.0183, 0.0115, 0.0187, 0.0173, 0.0167, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:01:44,251 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142097.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:01:47,675 INFO [finetune.py:992] (1/2) Epoch 4, batch 5400, loss[loss=0.1572, simple_loss=0.2364, pruned_loss=0.03901, over 12185.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2621, pruned_loss=0.04317, over 2378217.57 frames. ], batch size: 29, lr: 4.80e-03, grad_scale: 16.0 2023-05-16 00:02:01,167 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142121.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:02:05,098 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.087e+02 2.960e+02 3.619e+02 4.165e+02 8.260e+02, threshold=7.238e+02, percent-clipped=1.0 2023-05-16 00:02:17,858 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142145.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:02:23,467 INFO [finetune.py:992] (1/2) Epoch 4, batch 5450, loss[loss=0.174, simple_loss=0.2711, pruned_loss=0.03842, over 10511.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.262, pruned_loss=0.04319, over 2380375.76 frames. ], batch size: 68, lr: 4.80e-03, grad_scale: 16.0 2023-05-16 00:02:26,699 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6362, 2.4474, 4.5953, 4.8917, 3.0668, 2.4605, 2.8418, 1.9417], device='cuda:1'), covar=tensor([0.1378, 0.3423, 0.0384, 0.0268, 0.0965, 0.2111, 0.2659, 0.4356], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0366, 0.0259, 0.0284, 0.0249, 0.0275, 0.0344, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:02:55,619 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2778, 5.0599, 5.2002, 5.2182, 4.8311, 4.9615, 4.7116, 5.2163], device='cuda:1'), covar=tensor([0.0593, 0.0560, 0.0680, 0.0543, 0.1866, 0.1149, 0.0546, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0630, 0.0545, 0.0583, 0.0781, 0.0702, 0.0515, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 00:02:59,276 INFO [finetune.py:992] (1/2) Epoch 4, batch 5500, loss[loss=0.1985, simple_loss=0.2843, pruned_loss=0.05635, over 12367.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2626, pruned_loss=0.04328, over 2378119.59 frames. ], batch size: 35, lr: 4.80e-03, grad_scale: 16.0 2023-05-16 00:03:16,920 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 2.754e+02 3.312e+02 4.084e+02 8.741e+02, threshold=6.624e+02, percent-clipped=1.0 2023-05-16 00:03:29,223 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8192, 3.0422, 5.2760, 2.6098, 2.5436, 4.0259, 3.1767, 4.0016], device='cuda:1'), covar=tensor([0.0346, 0.1234, 0.0233, 0.1127, 0.1931, 0.1182, 0.1357, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0225, 0.0229, 0.0177, 0.0230, 0.0277, 0.0221, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:03:34,602 INFO [finetune.py:992] (1/2) Epoch 4, batch 5550, loss[loss=0.1532, simple_loss=0.2318, pruned_loss=0.03723, over 12334.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2627, pruned_loss=0.04336, over 2377447.13 frames. ], batch size: 30, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:03:35,556 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9987, 3.5649, 5.3509, 2.8080, 2.8875, 3.9355, 3.5083, 3.9777], device='cuda:1'), covar=tensor([0.0333, 0.0975, 0.0189, 0.1088, 0.1767, 0.1135, 0.1123, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0225, 0.0229, 0.0177, 0.0230, 0.0277, 0.0221, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:03:48,778 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142272.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:03:51,873 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8634, 2.8069, 4.5056, 4.6985, 2.9139, 2.6247, 2.9624, 2.1048], device='cuda:1'), covar=tensor([0.1229, 0.2661, 0.0435, 0.0359, 0.1109, 0.1980, 0.2357, 0.3533], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0370, 0.0261, 0.0288, 0.0251, 0.0278, 0.0348, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:03:56,029 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0910, 5.0795, 4.9073, 4.9395, 4.4367, 5.0294, 5.0712, 5.2381], device='cuda:1'), covar=tensor([0.0208, 0.0142, 0.0189, 0.0322, 0.0849, 0.0307, 0.0129, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0179, 0.0180, 0.0231, 0.0229, 0.0198, 0.0163, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 00:03:57,374 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142284.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:04:03,150 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0589, 4.7547, 4.9199, 4.9864, 4.7351, 4.9175, 4.8636, 2.8693], device='cuda:1'), covar=tensor([0.0089, 0.0064, 0.0068, 0.0057, 0.0046, 0.0097, 0.0072, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0073, 0.0077, 0.0070, 0.0057, 0.0087, 0.0074, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:04:10,793 INFO [finetune.py:992] (1/2) Epoch 4, batch 5600, loss[loss=0.1862, simple_loss=0.2767, pruned_loss=0.04786, over 12086.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2624, pruned_loss=0.04323, over 2384127.18 frames. ], batch size: 42, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:04:29,383 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.880e+02 2.833e+02 3.192e+02 3.756e+02 1.144e+03, threshold=6.384e+02, percent-clipped=4.0 2023-05-16 00:04:47,038 INFO [finetune.py:992] (1/2) Epoch 4, batch 5650, loss[loss=0.1836, simple_loss=0.2814, pruned_loss=0.04292, over 12126.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2622, pruned_loss=0.04295, over 2387292.84 frames. ], batch size: 38, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:05:03,183 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3175, 4.6114, 2.8368, 2.6123, 3.8489, 2.3585, 3.8710, 3.2040], device='cuda:1'), covar=tensor([0.0596, 0.0539, 0.1085, 0.1467, 0.0291, 0.1372, 0.0476, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0245, 0.0170, 0.0196, 0.0135, 0.0176, 0.0190, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:05:06,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 00:05:22,317 INFO [finetune.py:992] (1/2) Epoch 4, batch 5700, loss[loss=0.1829, simple_loss=0.2729, pruned_loss=0.04642, over 12158.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2636, pruned_loss=0.04362, over 2382769.09 frames. ], batch size: 34, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:05:39,847 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 3.063e+02 3.770e+02 4.357e+02 1.028e+03, threshold=7.540e+02, percent-clipped=5.0 2023-05-16 00:05:58,120 INFO [finetune.py:992] (1/2) Epoch 4, batch 5750, loss[loss=0.2001, simple_loss=0.2863, pruned_loss=0.05699, over 11886.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2634, pruned_loss=0.04401, over 2378497.59 frames. ], batch size: 44, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:05:58,428 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7076, 2.8369, 4.3958, 4.7367, 2.9444, 2.5917, 2.9610, 1.9624], device='cuda:1'), covar=tensor([0.1352, 0.2773, 0.0466, 0.0349, 0.1157, 0.1984, 0.2381, 0.3788], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0375, 0.0265, 0.0291, 0.0255, 0.0282, 0.0353, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:06:33,549 INFO [finetune.py:992] (1/2) Epoch 4, batch 5800, loss[loss=0.1678, simple_loss=0.2621, pruned_loss=0.03676, over 12350.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2637, pruned_loss=0.04409, over 2378395.07 frames. ], batch size: 36, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:06:41,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-05-16 00:06:51,333 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.283e+02 2.997e+02 3.428e+02 4.047e+02 7.003e+02, threshold=6.857e+02, percent-clipped=0.0 2023-05-16 00:07:08,941 INFO [finetune.py:992] (1/2) Epoch 4, batch 5850, loss[loss=0.1668, simple_loss=0.256, pruned_loss=0.03881, over 12295.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2628, pruned_loss=0.04378, over 2383363.40 frames. ], batch size: 34, lr: 4.80e-03, grad_scale: 32.0 2023-05-16 00:07:23,317 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142572.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:07:24,128 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142573.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:07:29,787 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4087, 4.9289, 5.3882, 4.6910, 4.9942, 4.6862, 5.4401, 4.9769], device='cuda:1'), covar=tensor([0.0234, 0.0318, 0.0210, 0.0233, 0.0333, 0.0299, 0.0168, 0.0230], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0244, 0.0259, 0.0238, 0.0236, 0.0237, 0.0216, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:07:31,935 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142584.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:07:45,481 INFO [finetune.py:992] (1/2) Epoch 4, batch 5900, loss[loss=0.1677, simple_loss=0.2526, pruned_loss=0.04137, over 12299.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2631, pruned_loss=0.0441, over 2381507.36 frames. ], batch size: 34, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:07:52,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 00:07:58,827 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142620.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:08:04,595 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.871e+02 3.403e+02 4.114e+02 6.329e+02, threshold=6.806e+02, percent-clipped=0.0 2023-05-16 00:08:07,036 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9353, 2.4402, 3.4204, 2.9825, 3.2440, 3.1125, 2.3353, 3.3810], device='cuda:1'), covar=tensor([0.0118, 0.0290, 0.0132, 0.0232, 0.0149, 0.0161, 0.0345, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0193, 0.0172, 0.0174, 0.0197, 0.0150, 0.0187, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:08:07,556 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142632.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:08:09,013 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142634.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:08:21,687 INFO [finetune.py:992] (1/2) Epoch 4, batch 5950, loss[loss=0.174, simple_loss=0.2601, pruned_loss=0.04397, over 12268.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2627, pruned_loss=0.0439, over 2382301.32 frames. ], batch size: 32, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:08:33,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.25 vs. limit=5.0 2023-05-16 00:08:49,555 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7275, 2.8464, 3.6233, 4.6821, 3.9274, 4.6583, 4.0238, 3.5242], device='cuda:1'), covar=tensor([0.0025, 0.0311, 0.0146, 0.0033, 0.0111, 0.0052, 0.0082, 0.0258], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0119, 0.0102, 0.0074, 0.0099, 0.0111, 0.0086, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:08:57,078 INFO [finetune.py:992] (1/2) Epoch 4, batch 6000, loss[loss=0.1412, simple_loss=0.2222, pruned_loss=0.03016, over 12015.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04456, over 2370657.68 frames. ], batch size: 28, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:08:57,079 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 00:09:12,805 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1593, 2.7305, 3.6088, 2.1421, 2.4711, 3.0648, 2.7509, 3.1732], device='cuda:1'), covar=tensor([0.0595, 0.1043, 0.0340, 0.1293, 0.1584, 0.1054, 0.1073, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0226, 0.0228, 0.0177, 0.0230, 0.0277, 0.0221, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:09:15,291 INFO [finetune.py:1026] (1/2) Epoch 4, validation: loss=0.3275, simple_loss=0.4012, pruned_loss=0.1269, over 1020973.00 frames. 2023-05-16 00:09:15,292 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-16 00:09:30,941 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142724.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:09:33,673 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 2.858e+02 3.331e+02 4.161e+02 8.424e+02, threshold=6.661e+02, percent-clipped=1.0 2023-05-16 00:09:35,162 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5606, 5.3227, 5.4941, 5.4624, 5.0471, 5.1909, 4.9832, 5.4014], device='cuda:1'), covar=tensor([0.0502, 0.0516, 0.0636, 0.0482, 0.1963, 0.1072, 0.0459, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0630, 0.0545, 0.0585, 0.0775, 0.0699, 0.0513, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 00:09:50,256 INFO [finetune.py:992] (1/2) Epoch 4, batch 6050, loss[loss=0.1507, simple_loss=0.2435, pruned_loss=0.02891, over 12340.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2647, pruned_loss=0.04513, over 2364897.77 frames. ], batch size: 30, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:10:13,740 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142785.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:10:26,122 INFO [finetune.py:992] (1/2) Epoch 4, batch 6100, loss[loss=0.1534, simple_loss=0.2369, pruned_loss=0.03494, over 12268.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2642, pruned_loss=0.04474, over 2366191.44 frames. ], batch size: 28, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:10:44,523 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 2.941e+02 3.348e+02 3.892e+02 7.445e+02, threshold=6.695e+02, percent-clipped=1.0 2023-05-16 00:11:02,353 INFO [finetune.py:992] (1/2) Epoch 4, batch 6150, loss[loss=0.1675, simple_loss=0.2536, pruned_loss=0.0407, over 12122.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2638, pruned_loss=0.04452, over 2376475.58 frames. ], batch size: 30, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:11:10,973 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3775, 4.2357, 4.3337, 4.6438, 3.1439, 4.1388, 2.8805, 4.2089], device='cuda:1'), covar=tensor([0.1535, 0.0613, 0.0699, 0.0489, 0.0982, 0.0519, 0.1516, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0264, 0.0294, 0.0349, 0.0240, 0.0238, 0.0256, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:11:19,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-05-16 00:11:27,867 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4983, 3.8379, 3.3194, 3.3396, 3.0755, 2.8537, 3.6647, 2.4244], device='cuda:1'), covar=tensor([0.0352, 0.0090, 0.0170, 0.0144, 0.0320, 0.0301, 0.0114, 0.0413], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0157, 0.0152, 0.0175, 0.0194, 0.0191, 0.0156, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:11:38,060 INFO [finetune.py:992] (1/2) Epoch 4, batch 6200, loss[loss=0.1885, simple_loss=0.2787, pruned_loss=0.04912, over 12362.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2633, pruned_loss=0.04459, over 2376320.53 frames. ], batch size: 38, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:11:56,737 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 2.738e+02 3.128e+02 3.851e+02 9.110e+02, threshold=6.257e+02, percent-clipped=2.0 2023-05-16 00:11:57,652 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142929.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:12:14,018 INFO [finetune.py:992] (1/2) Epoch 4, batch 6250, loss[loss=0.1729, simple_loss=0.2678, pruned_loss=0.03901, over 12283.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2643, pruned_loss=0.04492, over 2377251.74 frames. ], batch size: 33, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:12:49,953 INFO [finetune.py:992] (1/2) Epoch 4, batch 6300, loss[loss=0.1811, simple_loss=0.2697, pruned_loss=0.04625, over 12064.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2642, pruned_loss=0.04487, over 2375471.93 frames. ], batch size: 40, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:12:52,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4138, 2.1450, 3.3516, 4.2979, 2.1754, 4.4198, 4.4120, 4.5611], device='cuda:1'), covar=tensor([0.0174, 0.1347, 0.0385, 0.0146, 0.1321, 0.0198, 0.0137, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0186, 0.0118, 0.0190, 0.0175, 0.0169, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:13:08,920 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.720e+02 3.349e+02 3.998e+02 8.545e+02, threshold=6.697e+02, percent-clipped=2.0 2023-05-16 00:13:26,000 INFO [finetune.py:992] (1/2) Epoch 4, batch 6350, loss[loss=0.1778, simple_loss=0.272, pruned_loss=0.04179, over 12353.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2656, pruned_loss=0.04525, over 2364605.15 frames. ], batch size: 36, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:13:45,795 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143080.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:13:48,128 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1921, 5.1099, 4.9491, 5.0942, 4.6159, 5.1011, 5.1421, 5.3615], device='cuda:1'), covar=tensor([0.0166, 0.0109, 0.0193, 0.0235, 0.0687, 0.0203, 0.0123, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0182, 0.0183, 0.0233, 0.0232, 0.0202, 0.0166, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 00:14:01,317 INFO [finetune.py:992] (1/2) Epoch 4, batch 6400, loss[loss=0.2034, simple_loss=0.2868, pruned_loss=0.05996, over 10450.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2644, pruned_loss=0.04461, over 2368426.76 frames. ], batch size: 68, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:14:19,845 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.964e+02 3.528e+02 4.352e+02 1.519e+03, threshold=7.056e+02, percent-clipped=5.0 2023-05-16 00:14:38,294 INFO [finetune.py:992] (1/2) Epoch 4, batch 6450, loss[loss=0.2358, simple_loss=0.3156, pruned_loss=0.078, over 8466.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2641, pruned_loss=0.04431, over 2372299.55 frames. ], batch size: 97, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:15:10,162 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3234, 4.6618, 4.1627, 4.9853, 4.4937, 2.8882, 4.3741, 3.1985], device='cuda:1'), covar=tensor([0.0684, 0.0673, 0.1194, 0.0356, 0.1021, 0.1499, 0.0817, 0.2892], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0368, 0.0346, 0.0261, 0.0355, 0.0259, 0.0333, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:15:14,510 INFO [finetune.py:992] (1/2) Epoch 4, batch 6500, loss[loss=0.181, simple_loss=0.271, pruned_loss=0.0455, over 12054.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2639, pruned_loss=0.04429, over 2368458.18 frames. ], batch size: 42, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:15:32,844 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.055e+02 2.770e+02 3.120e+02 3.791e+02 9.865e+02, threshold=6.240e+02, percent-clipped=3.0 2023-05-16 00:15:33,719 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143229.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:15:48,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 00:15:49,994 INFO [finetune.py:992] (1/2) Epoch 4, batch 6550, loss[loss=0.1822, simple_loss=0.2797, pruned_loss=0.04235, over 12356.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2635, pruned_loss=0.04375, over 2375210.80 frames. ], batch size: 36, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:16:00,273 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5957, 2.6614, 3.3294, 4.3385, 2.5574, 4.4263, 4.4408, 4.6713], device='cuda:1'), covar=tensor([0.0100, 0.1046, 0.0404, 0.0182, 0.1080, 0.0213, 0.0170, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0186, 0.0118, 0.0189, 0.0175, 0.0170, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:16:08,528 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143277.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:16:16,537 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9213, 3.5642, 5.2945, 2.6652, 2.9277, 3.9492, 3.6049, 3.9726], device='cuda:1'), covar=tensor([0.0399, 0.0976, 0.0262, 0.1111, 0.1727, 0.1240, 0.1033, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0227, 0.0228, 0.0177, 0.0230, 0.0279, 0.0221, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:16:26,975 INFO [finetune.py:992] (1/2) Epoch 4, batch 6600, loss[loss=0.1781, simple_loss=0.2713, pruned_loss=0.04239, over 12061.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2638, pruned_loss=0.04359, over 2376202.17 frames. ], batch size: 37, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:16:45,591 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.893e+02 2.783e+02 3.345e+02 4.038e+02 6.670e+02, threshold=6.691e+02, percent-clipped=2.0 2023-05-16 00:17:02,648 INFO [finetune.py:992] (1/2) Epoch 4, batch 6650, loss[loss=0.1865, simple_loss=0.2767, pruned_loss=0.04815, over 10632.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2636, pruned_loss=0.04385, over 2374543.30 frames. ], batch size: 68, lr: 4.79e-03, grad_scale: 16.0 2023-05-16 00:17:06,480 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2664, 3.7343, 3.6987, 4.1113, 2.8963, 3.5081, 2.4876, 3.5784], device='cuda:1'), covar=tensor([0.1618, 0.0714, 0.1000, 0.0642, 0.1140, 0.0750, 0.1838, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0260, 0.0293, 0.0348, 0.0237, 0.0236, 0.0255, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:17:22,820 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143380.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:17:37,088 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143400.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:17:38,343 INFO [finetune.py:992] (1/2) Epoch 4, batch 6700, loss[loss=0.1755, simple_loss=0.2656, pruned_loss=0.04263, over 12185.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2643, pruned_loss=0.04417, over 2367986.93 frames. ], batch size: 31, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:17:57,446 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143428.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:17:57,979 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.936e+02 3.309e+02 3.963e+02 6.958e+02, threshold=6.618e+02, percent-clipped=1.0 2023-05-16 00:18:14,956 INFO [finetune.py:992] (1/2) Epoch 4, batch 6750, loss[loss=0.1626, simple_loss=0.2502, pruned_loss=0.03752, over 12350.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2644, pruned_loss=0.04444, over 2363844.71 frames. ], batch size: 31, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:18:21,584 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143461.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:18:24,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-05-16 00:18:30,081 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143473.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:18:38,834 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 00:18:50,464 INFO [finetune.py:992] (1/2) Epoch 4, batch 6800, loss[loss=0.1613, simple_loss=0.2487, pruned_loss=0.03694, over 12349.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2643, pruned_loss=0.04416, over 2367682.52 frames. ], batch size: 31, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:19:09,477 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.145e+02 2.927e+02 3.459e+02 4.200e+02 8.756e+02, threshold=6.918e+02, percent-clipped=3.0 2023-05-16 00:19:13,243 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143534.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:19:21,746 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3107, 3.5467, 3.2299, 3.1802, 2.7472, 2.6944, 3.5371, 2.1855], device='cuda:1'), covar=tensor([0.0414, 0.0112, 0.0167, 0.0149, 0.0396, 0.0314, 0.0107, 0.0472], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0161, 0.0153, 0.0178, 0.0200, 0.0194, 0.0159, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:19:25,824 INFO [finetune.py:992] (1/2) Epoch 4, batch 6850, loss[loss=0.2397, simple_loss=0.3167, pruned_loss=0.08134, over 8015.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2637, pruned_loss=0.04419, over 2367414.35 frames. ], batch size: 98, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:19:56,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 00:19:59,409 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2253, 2.4500, 3.0729, 3.9672, 2.4040, 4.0851, 4.0984, 4.3028], device='cuda:1'), covar=tensor([0.0104, 0.1047, 0.0422, 0.0193, 0.1093, 0.0233, 0.0154, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0205, 0.0188, 0.0119, 0.0191, 0.0177, 0.0171, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:20:02,975 INFO [finetune.py:992] (1/2) Epoch 4, batch 6900, loss[loss=0.1582, simple_loss=0.2405, pruned_loss=0.03798, over 11979.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2645, pruned_loss=0.04473, over 2364563.05 frames. ], batch size: 28, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:20:18,886 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1315, 4.7703, 4.8977, 5.0066, 4.8815, 4.9799, 4.9315, 2.9000], device='cuda:1'), covar=tensor([0.0081, 0.0065, 0.0077, 0.0057, 0.0042, 0.0097, 0.0065, 0.0588], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0078, 0.0071, 0.0058, 0.0089, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:20:19,526 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8670, 5.8557, 5.6079, 5.1531, 4.9945, 5.7539, 5.3131, 5.1370], device='cuda:1'), covar=tensor([0.0642, 0.0792, 0.0632, 0.1387, 0.0721, 0.0619, 0.1531, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0509, 0.0477, 0.0588, 0.0380, 0.0659, 0.0725, 0.0533], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:20:22,135 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.832e+02 3.237e+02 3.973e+02 1.176e+03, threshold=6.474e+02, percent-clipped=4.0 2023-05-16 00:20:23,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.55 vs. limit=5.0 2023-05-16 00:20:28,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8414, 2.4154, 3.2153, 3.8702, 3.4426, 3.7203, 3.4734, 2.5751], device='cuda:1'), covar=tensor([0.0040, 0.0335, 0.0143, 0.0035, 0.0098, 0.0077, 0.0085, 0.0333], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0117, 0.0100, 0.0073, 0.0098, 0.0108, 0.0085, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:20:38,460 INFO [finetune.py:992] (1/2) Epoch 4, batch 6950, loss[loss=0.1459, simple_loss=0.2211, pruned_loss=0.03532, over 12268.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2647, pruned_loss=0.04486, over 2356720.42 frames. ], batch size: 28, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:21:13,839 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5679, 2.2926, 3.7745, 4.5733, 4.0328, 4.2093, 4.0278, 3.0179], device='cuda:1'), covar=tensor([0.0035, 0.0473, 0.0130, 0.0028, 0.0104, 0.0098, 0.0079, 0.0361], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0118, 0.0100, 0.0073, 0.0098, 0.0109, 0.0085, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:21:14,357 INFO [finetune.py:992] (1/2) Epoch 4, batch 7000, loss[loss=0.1777, simple_loss=0.2649, pruned_loss=0.0452, over 12144.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2644, pruned_loss=0.0447, over 2363198.87 frames. ], batch size: 34, lr: 4.79e-03, grad_scale: 8.0 2023-05-16 00:21:34,172 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.121e+02 2.902e+02 3.396e+02 4.384e+02 9.697e+02, threshold=6.791e+02, percent-clipped=7.0 2023-05-16 00:21:50,591 INFO [finetune.py:992] (1/2) Epoch 4, batch 7050, loss[loss=0.1687, simple_loss=0.2492, pruned_loss=0.0441, over 11998.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2648, pruned_loss=0.04499, over 2362942.20 frames. ], batch size: 28, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:21:53,608 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143756.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 00:22:16,998 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4916, 4.9652, 3.1158, 2.7548, 4.2863, 2.7674, 4.2271, 3.5217], device='cuda:1'), covar=tensor([0.0608, 0.0403, 0.0939, 0.1375, 0.0285, 0.1196, 0.0350, 0.0686], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0250, 0.0175, 0.0199, 0.0139, 0.0181, 0.0193, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:22:26,498 INFO [finetune.py:992] (1/2) Epoch 4, batch 7100, loss[loss=0.1809, simple_loss=0.2634, pruned_loss=0.04921, over 11978.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2639, pruned_loss=0.04462, over 2361982.11 frames. ], batch size: 28, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:22:45,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 00:22:45,497 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 2.928e+02 3.425e+02 3.906e+02 8.366e+02, threshold=6.851e+02, percent-clipped=2.0 2023-05-16 00:22:45,602 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143829.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:23:02,630 INFO [finetune.py:992] (1/2) Epoch 4, batch 7150, loss[loss=0.1652, simple_loss=0.2569, pruned_loss=0.03675, over 12353.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2643, pruned_loss=0.04471, over 2365131.20 frames. ], batch size: 36, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:23:38,827 INFO [finetune.py:992] (1/2) Epoch 4, batch 7200, loss[loss=0.1733, simple_loss=0.2647, pruned_loss=0.04097, over 12299.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2638, pruned_loss=0.04435, over 2367374.87 frames. ], batch size: 33, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:23:57,998 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.800e+02 3.188e+02 3.874e+02 1.813e+03, threshold=6.376e+02, percent-clipped=2.0 2023-05-16 00:24:14,529 INFO [finetune.py:992] (1/2) Epoch 4, batch 7250, loss[loss=0.1704, simple_loss=0.2577, pruned_loss=0.04157, over 12352.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.0438, over 2373365.18 frames. ], batch size: 35, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:24:36,079 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5329, 5.1439, 5.4862, 4.8545, 5.2279, 4.9383, 5.5561, 5.1591], device='cuda:1'), covar=tensor([0.0199, 0.0307, 0.0238, 0.0247, 0.0240, 0.0269, 0.0174, 0.0198], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0244, 0.0263, 0.0238, 0.0235, 0.0236, 0.0216, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:24:54,151 INFO [finetune.py:992] (1/2) Epoch 4, batch 7300, loss[loss=0.1863, simple_loss=0.2835, pruned_loss=0.04456, over 11730.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2636, pruned_loss=0.04408, over 2373947.92 frames. ], batch size: 48, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:25:13,985 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.750e+02 3.191e+02 3.884e+02 1.605e+03, threshold=6.381e+02, percent-clipped=3.0 2023-05-16 00:25:30,364 INFO [finetune.py:992] (1/2) Epoch 4, batch 7350, loss[loss=0.2608, simple_loss=0.3277, pruned_loss=0.09696, over 7612.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2647, pruned_loss=0.04445, over 2366147.50 frames. ], batch size: 98, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:25:33,427 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144056.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:25:56,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-05-16 00:26:06,065 INFO [finetune.py:992] (1/2) Epoch 4, batch 7400, loss[loss=0.1897, simple_loss=0.2763, pruned_loss=0.05155, over 12354.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2643, pruned_loss=0.04431, over 2362528.73 frames. ], batch size: 35, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:26:07,439 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=144104.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:26:24,675 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.946e+02 3.343e+02 3.974e+02 8.907e+02, threshold=6.685e+02, percent-clipped=2.0 2023-05-16 00:26:24,815 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144129.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 00:26:42,218 INFO [finetune.py:992] (1/2) Epoch 4, batch 7450, loss[loss=0.1921, simple_loss=0.2956, pruned_loss=0.04426, over 12203.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2652, pruned_loss=0.04495, over 2356329.77 frames. ], batch size: 35, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:27:00,276 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=144177.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:27:17,983 INFO [finetune.py:992] (1/2) Epoch 4, batch 7500, loss[loss=0.1585, simple_loss=0.2411, pruned_loss=0.03791, over 11781.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2649, pruned_loss=0.04474, over 2359237.65 frames. ], batch size: 26, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:27:37,169 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.942e+02 3.365e+02 3.938e+02 8.869e+02, threshold=6.730e+02, percent-clipped=6.0 2023-05-16 00:27:42,377 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0028, 4.9677, 4.8494, 4.8703, 4.4564, 4.9792, 4.9556, 5.2402], device='cuda:1'), covar=tensor([0.0164, 0.0130, 0.0163, 0.0264, 0.0740, 0.0249, 0.0140, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0182, 0.0180, 0.0233, 0.0231, 0.0199, 0.0165, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 00:27:53,739 INFO [finetune.py:992] (1/2) Epoch 4, batch 7550, loss[loss=0.162, simple_loss=0.2552, pruned_loss=0.03439, over 12312.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2639, pruned_loss=0.04417, over 2372710.76 frames. ], batch size: 34, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:28:28,322 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8743, 2.2774, 3.2453, 2.8287, 3.1927, 2.9490, 2.2815, 3.2042], device='cuda:1'), covar=tensor([0.0148, 0.0366, 0.0203, 0.0256, 0.0153, 0.0186, 0.0371, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0194, 0.0172, 0.0175, 0.0196, 0.0150, 0.0185, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:28:30,129 INFO [finetune.py:992] (1/2) Epoch 4, batch 7600, loss[loss=0.1903, simple_loss=0.2752, pruned_loss=0.05274, over 12021.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2636, pruned_loss=0.04431, over 2377015.05 frames. ], batch size: 42, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:28:45,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0486, 3.5490, 5.3042, 2.6471, 2.8578, 3.7962, 3.2615, 3.9115], device='cuda:1'), covar=tensor([0.0369, 0.1011, 0.0248, 0.1182, 0.1812, 0.1391, 0.1320, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0229, 0.0232, 0.0179, 0.0233, 0.0283, 0.0224, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:28:49,093 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.903e+02 3.452e+02 4.012e+02 8.262e+02, threshold=6.904e+02, percent-clipped=5.0 2023-05-16 00:29:05,348 INFO [finetune.py:992] (1/2) Epoch 4, batch 7650, loss[loss=0.1671, simple_loss=0.2524, pruned_loss=0.04091, over 12127.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2635, pruned_loss=0.04418, over 2373339.72 frames. ], batch size: 33, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:29:41,341 INFO [finetune.py:992] (1/2) Epoch 4, batch 7700, loss[loss=0.1864, simple_loss=0.284, pruned_loss=0.0444, over 12298.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2638, pruned_loss=0.04448, over 2372295.33 frames. ], batch size: 33, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:29:43,502 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0568, 5.7757, 5.2884, 5.3277, 5.8692, 5.2166, 5.4020, 5.3714], device='cuda:1'), covar=tensor([0.1596, 0.0898, 0.1016, 0.1879, 0.1005, 0.2317, 0.1665, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0463, 0.0365, 0.0415, 0.0444, 0.0419, 0.0377, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:29:47,796 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144411.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:30:01,778 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.806e+02 3.415e+02 3.909e+02 5.886e+02, threshold=6.830e+02, percent-clipped=0.0 2023-05-16 00:30:18,034 INFO [finetune.py:992] (1/2) Epoch 4, batch 7750, loss[loss=0.2042, simple_loss=0.2922, pruned_loss=0.05809, over 12126.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.265, pruned_loss=0.04518, over 2369266.72 frames. ], batch size: 38, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:30:32,155 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144472.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:30:35,694 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7189, 2.5648, 4.9136, 5.1749, 3.2182, 2.6820, 3.0336, 2.0129], device='cuda:1'), covar=tensor([0.1425, 0.3549, 0.0330, 0.0249, 0.0955, 0.2197, 0.2663, 0.4654], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0367, 0.0259, 0.0286, 0.0251, 0.0278, 0.0346, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:30:53,344 INFO [finetune.py:992] (1/2) Epoch 4, batch 7800, loss[loss=0.1609, simple_loss=0.2519, pruned_loss=0.03496, over 12302.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2641, pruned_loss=0.04501, over 2372046.45 frames. ], batch size: 33, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:31:12,464 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.788e+02 3.406e+02 4.091e+02 6.144e+02, threshold=6.812e+02, percent-clipped=0.0 2023-05-16 00:31:28,701 INFO [finetune.py:992] (1/2) Epoch 4, batch 7850, loss[loss=0.1871, simple_loss=0.2777, pruned_loss=0.04823, over 12036.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2645, pruned_loss=0.04498, over 2377980.76 frames. ], batch size: 42, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:31:35,305 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2646, 4.1857, 4.2326, 4.5958, 3.0626, 3.8669, 2.7352, 4.2558], device='cuda:1'), covar=tensor([0.1491, 0.0599, 0.0763, 0.0587, 0.1043, 0.0596, 0.1591, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0257, 0.0290, 0.0347, 0.0236, 0.0236, 0.0252, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:31:47,478 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7645, 2.6898, 4.7358, 4.8104, 2.7222, 2.6901, 2.9392, 2.1686], device='cuda:1'), covar=tensor([0.1308, 0.2872, 0.0365, 0.0413, 0.1211, 0.1921, 0.2461, 0.3578], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0371, 0.0262, 0.0289, 0.0253, 0.0280, 0.0349, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:32:05,927 INFO [finetune.py:992] (1/2) Epoch 4, batch 7900, loss[loss=0.1654, simple_loss=0.244, pruned_loss=0.04342, over 12345.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2638, pruned_loss=0.04455, over 2380741.75 frames. ], batch size: 30, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:32:21,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-16 00:32:25,026 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.084e+02 3.076e+02 3.537e+02 4.366e+02 9.547e+02, threshold=7.073e+02, percent-clipped=4.0 2023-05-16 00:32:41,091 INFO [finetune.py:992] (1/2) Epoch 4, batch 7950, loss[loss=0.1477, simple_loss=0.2294, pruned_loss=0.03295, over 12023.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2638, pruned_loss=0.04473, over 2377314.44 frames. ], batch size: 28, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:32:50,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8557, 5.8873, 5.5664, 5.1805, 4.9340, 5.7652, 5.3588, 5.0399], device='cuda:1'), covar=tensor([0.0822, 0.0816, 0.0672, 0.1545, 0.0761, 0.0767, 0.1601, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0509, 0.0479, 0.0589, 0.0381, 0.0658, 0.0724, 0.0532], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:33:16,683 INFO [finetune.py:992] (1/2) Epoch 4, batch 8000, loss[loss=0.1889, simple_loss=0.278, pruned_loss=0.04992, over 12016.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2647, pruned_loss=0.04521, over 2370549.12 frames. ], batch size: 40, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:33:26,891 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3720, 2.7602, 3.9744, 3.3406, 3.7852, 3.4664, 2.8660, 3.9383], device='cuda:1'), covar=tensor([0.0132, 0.0307, 0.0150, 0.0218, 0.0135, 0.0150, 0.0276, 0.0093], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0195, 0.0174, 0.0177, 0.0199, 0.0151, 0.0186, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:33:37,515 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.930e+02 3.311e+02 4.094e+02 7.677e+02, threshold=6.621e+02, percent-clipped=2.0 2023-05-16 00:33:53,707 INFO [finetune.py:992] (1/2) Epoch 4, batch 8050, loss[loss=0.1973, simple_loss=0.2809, pruned_loss=0.05681, over 12384.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2643, pruned_loss=0.04489, over 2369818.30 frames. ], batch size: 38, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:34:04,563 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144767.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:34:25,299 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 00:34:29,527 INFO [finetune.py:992] (1/2) Epoch 4, batch 8100, loss[loss=0.1761, simple_loss=0.2614, pruned_loss=0.04537, over 12306.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2651, pruned_loss=0.04562, over 2358011.91 frames. ], batch size: 34, lr: 4.78e-03, grad_scale: 8.0 2023-05-16 00:34:37,956 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 00:34:40,635 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5897, 2.3924, 4.2219, 4.6018, 3.1008, 2.4633, 2.7648, 1.9037], device='cuda:1'), covar=tensor([0.1399, 0.3328, 0.0482, 0.0335, 0.1009, 0.2249, 0.2805, 0.4829], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0372, 0.0264, 0.0290, 0.0254, 0.0282, 0.0350, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:34:48,718 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.810e+02 3.253e+02 3.981e+02 7.251e+02, threshold=6.507e+02, percent-clipped=2.0 2023-05-16 00:35:02,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 00:35:04,977 INFO [finetune.py:992] (1/2) Epoch 4, batch 8150, loss[loss=0.1696, simple_loss=0.2574, pruned_loss=0.04094, over 11726.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2645, pruned_loss=0.04518, over 2368673.42 frames. ], batch size: 48, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:35:39,573 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 00:35:42,020 INFO [finetune.py:992] (1/2) Epoch 4, batch 8200, loss[loss=0.1907, simple_loss=0.2801, pruned_loss=0.05064, over 12143.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2647, pruned_loss=0.04493, over 2370615.25 frames. ], batch size: 36, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:35:45,088 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9099, 4.2308, 3.7856, 4.5552, 4.0591, 2.5678, 3.9111, 2.9416], device='cuda:1'), covar=tensor([0.0832, 0.0834, 0.1337, 0.0392, 0.1097, 0.1626, 0.0985, 0.2781], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0368, 0.0345, 0.0261, 0.0354, 0.0259, 0.0334, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:35:55,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.33 vs. limit=5.0 2023-05-16 00:36:01,026 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.917e+02 3.512e+02 3.999e+02 6.164e+02, threshold=7.024e+02, percent-clipped=0.0 2023-05-16 00:36:17,397 INFO [finetune.py:992] (1/2) Epoch 4, batch 8250, loss[loss=0.1698, simple_loss=0.2575, pruned_loss=0.04103, over 12336.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.264, pruned_loss=0.04439, over 2380480.88 frames. ], batch size: 35, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:36:27,052 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-05-16 00:36:48,846 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144996.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:36:53,260 INFO [finetune.py:992] (1/2) Epoch 4, batch 8300, loss[loss=0.2002, simple_loss=0.288, pruned_loss=0.05616, over 12123.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.264, pruned_loss=0.04436, over 2375900.71 frames. ], batch size: 38, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:37:12,465 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145027.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:37:13,773 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 2.880e+02 3.303e+02 3.913e+02 1.017e+03, threshold=6.606e+02, percent-clipped=4.0 2023-05-16 00:37:30,150 INFO [finetune.py:992] (1/2) Epoch 4, batch 8350, loss[loss=0.2042, simple_loss=0.2883, pruned_loss=0.06, over 12131.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.264, pruned_loss=0.04421, over 2379572.37 frames. ], batch size: 38, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:37:33,878 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145057.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:37:39,882 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145066.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:37:40,546 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145067.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:37:55,391 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145088.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:38:05,112 INFO [finetune.py:992] (1/2) Epoch 4, batch 8400, loss[loss=0.1641, simple_loss=0.2616, pruned_loss=0.03335, over 12314.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2643, pruned_loss=0.04427, over 2381836.09 frames. ], batch size: 34, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:38:14,185 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145115.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:38:22,678 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145127.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:38:23,891 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.792e+02 3.296e+02 4.039e+02 8.655e+02, threshold=6.593e+02, percent-clipped=1.0 2023-05-16 00:38:40,184 INFO [finetune.py:992] (1/2) Epoch 4, batch 8450, loss[loss=0.1561, simple_loss=0.2417, pruned_loss=0.03522, over 12129.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2646, pruned_loss=0.04426, over 2378955.89 frames. ], batch size: 30, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:39:01,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 00:39:02,819 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0366, 4.6592, 5.0408, 4.4264, 4.7569, 4.5285, 5.0491, 4.6262], device='cuda:1'), covar=tensor([0.0253, 0.0321, 0.0239, 0.0227, 0.0284, 0.0271, 0.0196, 0.0375], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0243, 0.0260, 0.0235, 0.0234, 0.0235, 0.0214, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:39:16,980 INFO [finetune.py:992] (1/2) Epoch 4, batch 8500, loss[loss=0.2007, simple_loss=0.2903, pruned_loss=0.05561, over 12069.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2644, pruned_loss=0.0444, over 2377956.10 frames. ], batch size: 42, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:39:36,202 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.838e+02 3.322e+02 4.007e+02 1.046e+03, threshold=6.644e+02, percent-clipped=1.0 2023-05-16 00:39:52,532 INFO [finetune.py:992] (1/2) Epoch 4, batch 8550, loss[loss=0.1701, simple_loss=0.2568, pruned_loss=0.04169, over 12087.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2645, pruned_loss=0.04475, over 2369894.56 frames. ], batch size: 32, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:39:55,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 00:40:12,432 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1941, 5.9971, 5.6169, 5.5863, 6.0579, 5.3477, 5.6184, 5.6123], device='cuda:1'), covar=tensor([0.1651, 0.0919, 0.1033, 0.1927, 0.0951, 0.2340, 0.1639, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0463, 0.0367, 0.0416, 0.0445, 0.0422, 0.0376, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:40:14,561 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0515, 5.8466, 5.4577, 5.3965, 5.9354, 5.2907, 5.4667, 5.4477], device='cuda:1'), covar=tensor([0.1554, 0.0929, 0.0994, 0.2068, 0.0922, 0.2082, 0.1638, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0462, 0.0367, 0.0416, 0.0445, 0.0422, 0.0376, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:40:20,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 00:40:28,889 INFO [finetune.py:992] (1/2) Epoch 4, batch 8600, loss[loss=0.1862, simple_loss=0.2791, pruned_loss=0.04668, over 11282.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2649, pruned_loss=0.04497, over 2374777.39 frames. ], batch size: 55, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:40:45,395 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6112, 3.1633, 5.0165, 2.5034, 2.6932, 3.8179, 3.0497, 3.7325], device='cuda:1'), covar=tensor([0.0379, 0.1161, 0.0231, 0.1116, 0.1799, 0.1199, 0.1346, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0225, 0.0229, 0.0175, 0.0229, 0.0277, 0.0221, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:40:48,702 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 3.182e+02 3.638e+02 4.401e+02 1.025e+03, threshold=7.277e+02, percent-clipped=3.0 2023-05-16 00:41:04,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 00:41:05,235 INFO [finetune.py:992] (1/2) Epoch 4, batch 8650, loss[loss=0.1522, simple_loss=0.242, pruned_loss=0.03122, over 11882.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.265, pruned_loss=0.0449, over 2370278.05 frames. ], batch size: 26, lr: 4.77e-03, grad_scale: 8.0 2023-05-16 00:41:05,378 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145352.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:41:18,785 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145371.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:41:27,237 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145383.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:41:40,864 INFO [finetune.py:992] (1/2) Epoch 4, batch 8700, loss[loss=0.2095, simple_loss=0.2944, pruned_loss=0.06226, over 11673.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2655, pruned_loss=0.04484, over 2379268.13 frames. ], batch size: 48, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:41:54,899 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145422.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:41:59,824 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.107e+02 3.044e+02 3.423e+02 4.062e+02 6.784e+02, threshold=6.847e+02, percent-clipped=1.0 2023-05-16 00:42:02,108 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145432.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:42:16,677 INFO [finetune.py:992] (1/2) Epoch 4, batch 8750, loss[loss=0.1999, simple_loss=0.2806, pruned_loss=0.05966, over 11536.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2653, pruned_loss=0.0448, over 2373397.29 frames. ], batch size: 48, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:42:31,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9494, 2.2651, 3.5024, 2.9685, 3.2044, 3.0886, 2.4062, 3.3813], device='cuda:1'), covar=tensor([0.0112, 0.0340, 0.0129, 0.0205, 0.0171, 0.0155, 0.0310, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0196, 0.0173, 0.0176, 0.0199, 0.0152, 0.0186, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:42:53,182 INFO [finetune.py:992] (1/2) Epoch 4, batch 8800, loss[loss=0.1701, simple_loss=0.26, pruned_loss=0.04006, over 12300.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2651, pruned_loss=0.04479, over 2374209.62 frames. ], batch size: 34, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:43:12,284 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.939e+02 2.956e+02 3.531e+02 4.140e+02 9.183e+02, threshold=7.062e+02, percent-clipped=2.0 2023-05-16 00:43:28,581 INFO [finetune.py:992] (1/2) Epoch 4, batch 8850, loss[loss=0.2428, simple_loss=0.3107, pruned_loss=0.08746, over 8306.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2643, pruned_loss=0.0444, over 2374124.12 frames. ], batch size: 98, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:43:59,599 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145594.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:44:05,306 INFO [finetune.py:992] (1/2) Epoch 4, batch 8900, loss[loss=0.1759, simple_loss=0.2692, pruned_loss=0.04134, over 12310.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2635, pruned_loss=0.0442, over 2376027.13 frames. ], batch size: 34, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:44:06,530 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-05-16 00:44:17,588 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.26 vs. limit=5.0 2023-05-16 00:44:25,117 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 2.816e+02 3.227e+02 3.903e+02 1.604e+03, threshold=6.453e+02, percent-clipped=3.0 2023-05-16 00:44:35,357 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145643.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:44:41,704 INFO [finetune.py:992] (1/2) Epoch 4, batch 8950, loss[loss=0.1762, simple_loss=0.2626, pruned_loss=0.04488, over 12339.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2628, pruned_loss=0.04408, over 2368958.60 frames. ], batch size: 30, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:44:41,894 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145652.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:44:43,991 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145655.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 00:45:03,695 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145683.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:45:15,593 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145700.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:45:16,951 INFO [finetune.py:992] (1/2) Epoch 4, batch 9000, loss[loss=0.1812, simple_loss=0.259, pruned_loss=0.05166, over 12266.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2634, pruned_loss=0.04452, over 2363507.31 frames. ], batch size: 28, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:45:16,951 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 00:45:33,389 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7961, 4.6664, 4.9061, 4.8319, 4.3394, 4.4714, 4.3030, 4.6620], device='cuda:1'), covar=tensor([0.0936, 0.0554, 0.0593, 0.0591, 0.2048, 0.1487, 0.0721, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0635, 0.0557, 0.0591, 0.0779, 0.0705, 0.0522, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 00:45:35,140 INFO [finetune.py:1026] (1/2) Epoch 4, validation: loss=0.3366, simple_loss=0.4058, pruned_loss=0.1337, over 1020973.00 frames. 2023-05-16 00:45:35,140 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12574MB 2023-05-16 00:45:36,730 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145704.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:45:41,751 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1956, 2.1523, 2.7732, 3.1760, 2.1726, 3.2439, 3.2031, 3.2948], device='cuda:1'), covar=tensor([0.0174, 0.0878, 0.0391, 0.0171, 0.0959, 0.0268, 0.0243, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0203, 0.0186, 0.0120, 0.0191, 0.0176, 0.0171, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:45:50,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6312, 5.2332, 5.6043, 4.8962, 5.3111, 4.9875, 5.6458, 5.1841], device='cuda:1'), covar=tensor([0.0192, 0.0290, 0.0227, 0.0204, 0.0218, 0.0255, 0.0166, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0244, 0.0262, 0.0238, 0.0235, 0.0236, 0.0215, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:45:50,317 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145722.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:45:53,873 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145727.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:45:55,120 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.863e+02 3.398e+02 3.852e+02 7.519e+02, threshold=6.796e+02, percent-clipped=5.0 2023-05-16 00:45:56,662 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145731.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:46:11,492 INFO [finetune.py:992] (1/2) Epoch 4, batch 9050, loss[loss=0.1534, simple_loss=0.2365, pruned_loss=0.03513, over 12126.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2635, pruned_loss=0.04454, over 2366942.43 frames. ], batch size: 30, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:46:24,193 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145770.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:46:47,017 INFO [finetune.py:992] (1/2) Epoch 4, batch 9100, loss[loss=0.1757, simple_loss=0.2506, pruned_loss=0.05037, over 12174.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2636, pruned_loss=0.0451, over 2364603.99 frames. ], batch size: 31, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:47:00,758 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145821.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:47:01,419 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1682, 5.9917, 5.5715, 5.5350, 6.0526, 5.5213, 5.6189, 5.5807], device='cuda:1'), covar=tensor([0.1479, 0.0926, 0.0980, 0.1839, 0.0892, 0.1885, 0.1565, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0454, 0.0361, 0.0410, 0.0442, 0.0414, 0.0372, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:47:07,046 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.816e+02 3.541e+02 4.634e+02 1.137e+03, threshold=7.082e+02, percent-clipped=3.0 2023-05-16 00:47:23,171 INFO [finetune.py:992] (1/2) Epoch 4, batch 9150, loss[loss=0.1621, simple_loss=0.2538, pruned_loss=0.03524, over 12294.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2639, pruned_loss=0.04532, over 2360569.11 frames. ], batch size: 33, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:47:45,458 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145882.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 00:47:59,146 INFO [finetune.py:992] (1/2) Epoch 4, batch 9200, loss[loss=0.1773, simple_loss=0.2727, pruned_loss=0.041, over 12148.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2635, pruned_loss=0.04483, over 2371099.13 frames. ], batch size: 34, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:48:18,036 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.848e+02 3.273e+02 4.077e+02 7.326e+02, threshold=6.546e+02, percent-clipped=2.0 2023-05-16 00:48:25,326 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8361, 2.3770, 2.9449, 3.6780, 2.3177, 3.7204, 3.7235, 3.9120], device='cuda:1'), covar=tensor([0.0130, 0.1024, 0.0446, 0.0157, 0.1069, 0.0310, 0.0207, 0.0097], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0199, 0.0184, 0.0117, 0.0187, 0.0173, 0.0167, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:48:33,061 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145950.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:48:34,318 INFO [finetune.py:992] (1/2) Epoch 4, batch 9250, loss[loss=0.1915, simple_loss=0.2738, pruned_loss=0.05464, over 12128.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2637, pruned_loss=0.04473, over 2366834.94 frames. ], batch size: 39, lr: 4.77e-03, grad_scale: 16.0 2023-05-16 00:49:08,453 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145999.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:49:14,617 INFO [finetune.py:992] (1/2) Epoch 4, batch 9300, loss[loss=0.1477, simple_loss=0.245, pruned_loss=0.02522, over 12343.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2641, pruned_loss=0.04527, over 2354200.62 frames. ], batch size: 35, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:49:32,644 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146027.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:49:33,858 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.807e+02 3.392e+02 4.099e+02 9.430e+02, threshold=6.784e+02, percent-clipped=2.0 2023-05-16 00:49:49,913 INFO [finetune.py:992] (1/2) Epoch 4, batch 9350, loss[loss=0.1713, simple_loss=0.2492, pruned_loss=0.0467, over 12283.00 frames. ], tot_loss[loss=0.177, simple_loss=0.264, pruned_loss=0.045, over 2357619.98 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:49:58,388 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4666, 5.0100, 5.4405, 4.7507, 5.1255, 4.8529, 5.4495, 5.0345], device='cuda:1'), covar=tensor([0.0205, 0.0336, 0.0234, 0.0216, 0.0261, 0.0265, 0.0195, 0.0226], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0247, 0.0265, 0.0239, 0.0237, 0.0237, 0.0216, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 00:50:06,105 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146075.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:50:25,256 INFO [finetune.py:992] (1/2) Epoch 4, batch 9400, loss[loss=0.2357, simple_loss=0.308, pruned_loss=0.08171, over 8131.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2637, pruned_loss=0.04485, over 2359015.46 frames. ], batch size: 98, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:50:38,802 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 00:50:44,870 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.116e+02 2.827e+02 3.211e+02 4.021e+02 6.482e+02, threshold=6.422e+02, percent-clipped=0.0 2023-05-16 00:50:59,871 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146149.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:51:01,807 INFO [finetune.py:992] (1/2) Epoch 4, batch 9450, loss[loss=0.201, simple_loss=0.2796, pruned_loss=0.06126, over 12147.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2633, pruned_loss=0.04446, over 2370247.00 frames. ], batch size: 39, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:51:07,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-05-16 00:51:19,444 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146177.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 00:51:37,498 INFO [finetune.py:992] (1/2) Epoch 4, batch 9500, loss[loss=0.1652, simple_loss=0.2454, pruned_loss=0.04246, over 11421.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2631, pruned_loss=0.04454, over 2364248.49 frames. ], batch size: 25, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:51:43,461 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146210.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:51:56,948 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.030e+02 3.014e+02 3.420e+02 4.256e+02 8.963e+02, threshold=6.839e+02, percent-clipped=4.0 2023-05-16 00:52:11,959 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146250.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:52:13,244 INFO [finetune.py:992] (1/2) Epoch 4, batch 9550, loss[loss=0.1616, simple_loss=0.25, pruned_loss=0.03656, over 12018.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2628, pruned_loss=0.04428, over 2367097.24 frames. ], batch size: 31, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:52:28,545 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1960, 6.0379, 5.6102, 5.5914, 6.0862, 5.3432, 5.6887, 5.5884], device='cuda:1'), covar=tensor([0.1365, 0.0803, 0.0838, 0.1826, 0.0916, 0.2167, 0.1450, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0459, 0.0363, 0.0410, 0.0449, 0.0419, 0.0375, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:52:28,671 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5712, 2.4592, 3.3303, 4.2990, 2.5135, 4.3952, 4.4825, 4.6278], device='cuda:1'), covar=tensor([0.0128, 0.1202, 0.0389, 0.0151, 0.1260, 0.0205, 0.0145, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0203, 0.0186, 0.0118, 0.0190, 0.0176, 0.0170, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:52:47,039 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146298.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:52:47,851 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146299.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:52:49,828 INFO [finetune.py:992] (1/2) Epoch 4, batch 9600, loss[loss=0.1789, simple_loss=0.2615, pruned_loss=0.04818, over 12280.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2634, pruned_loss=0.04475, over 2362365.52 frames. ], batch size: 37, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:53:08,539 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1659, 4.8752, 5.1103, 5.0237, 4.9009, 5.0136, 4.9541, 2.8178], device='cuda:1'), covar=tensor([0.0090, 0.0056, 0.0058, 0.0053, 0.0038, 0.0085, 0.0055, 0.0589], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0076, 0.0070, 0.0057, 0.0088, 0.0074, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:53:09,056 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.926e+02 3.346e+02 4.034e+02 7.441e+02, threshold=6.692e+02, percent-clipped=2.0 2023-05-16 00:53:22,012 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146347.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:53:22,956 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3666, 4.6060, 4.1556, 4.9727, 4.4576, 2.7729, 4.2414, 3.1239], device='cuda:1'), covar=tensor([0.0688, 0.0736, 0.1204, 0.0388, 0.0990, 0.1491, 0.0844, 0.2886], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0365, 0.0342, 0.0258, 0.0350, 0.0256, 0.0327, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:53:25,561 INFO [finetune.py:992] (1/2) Epoch 4, batch 9650, loss[loss=0.1472, simple_loss=0.2271, pruned_loss=0.03361, over 11999.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2635, pruned_loss=0.04491, over 2362174.73 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:53:55,155 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3943, 3.4791, 3.2701, 3.1099, 2.8830, 2.7024, 3.5069, 2.1491], device='cuda:1'), covar=tensor([0.0356, 0.0160, 0.0149, 0.0164, 0.0347, 0.0337, 0.0113, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0154, 0.0148, 0.0174, 0.0194, 0.0188, 0.0153, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 00:54:00,918 INFO [finetune.py:992] (1/2) Epoch 4, batch 9700, loss[loss=0.1886, simple_loss=0.2899, pruned_loss=0.04371, over 12207.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2635, pruned_loss=0.04492, over 2366811.57 frames. ], batch size: 35, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:54:20,671 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.186e+02 3.038e+02 3.461e+02 3.924e+02 8.082e+02, threshold=6.922e+02, percent-clipped=2.0 2023-05-16 00:54:37,535 INFO [finetune.py:992] (1/2) Epoch 4, batch 9750, loss[loss=0.1765, simple_loss=0.2701, pruned_loss=0.04142, over 12152.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2636, pruned_loss=0.04498, over 2369307.22 frames. ], batch size: 34, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:54:55,396 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146477.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:54:59,013 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1018, 4.8128, 5.0026, 5.0063, 4.8312, 4.9802, 4.9156, 2.9579], device='cuda:1'), covar=tensor([0.0079, 0.0062, 0.0077, 0.0051, 0.0048, 0.0093, 0.0077, 0.0531], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0076, 0.0070, 0.0057, 0.0087, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 00:55:00,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-05-16 00:55:12,873 INFO [finetune.py:992] (1/2) Epoch 4, batch 9800, loss[loss=0.1498, simple_loss=0.2286, pruned_loss=0.03553, over 12004.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2628, pruned_loss=0.04447, over 2366337.59 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:55:13,074 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4938, 2.4560, 3.1909, 4.2286, 2.4696, 4.2811, 4.3682, 4.4727], device='cuda:1'), covar=tensor([0.0085, 0.1038, 0.0385, 0.0127, 0.1096, 0.0187, 0.0119, 0.0071], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0202, 0.0185, 0.0118, 0.0189, 0.0175, 0.0169, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:55:15,109 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146505.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:55:29,386 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146525.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 00:55:32,063 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 2.736e+02 3.214e+02 3.887e+02 6.730e+02, threshold=6.427e+02, percent-clipped=0.0 2023-05-16 00:55:49,193 INFO [finetune.py:992] (1/2) Epoch 4, batch 9850, loss[loss=0.1399, simple_loss=0.2237, pruned_loss=0.02805, over 12345.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2616, pruned_loss=0.04394, over 2368246.29 frames. ], batch size: 30, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:55:51,712 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 00:56:17,818 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146591.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:56:25,561 INFO [finetune.py:992] (1/2) Epoch 4, batch 9900, loss[loss=0.1806, simple_loss=0.2672, pruned_loss=0.04698, over 12156.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.262, pruned_loss=0.04389, over 2369199.87 frames. ], batch size: 34, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:56:26,491 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0718, 3.9659, 4.1013, 4.4480, 2.8776, 3.9028, 2.5823, 4.1363], device='cuda:1'), covar=tensor([0.1690, 0.0707, 0.0854, 0.0623, 0.1203, 0.0627, 0.1829, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0256, 0.0289, 0.0343, 0.0234, 0.0234, 0.0251, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 00:56:42,708 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146626.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:56:44,573 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.744e+02 3.233e+02 3.737e+02 7.432e+02, threshold=6.466e+02, percent-clipped=1.0 2023-05-16 00:56:56,887 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1452, 2.3569, 3.7712, 3.0989, 3.5782, 3.2261, 2.4565, 3.6645], device='cuda:1'), covar=tensor([0.0098, 0.0304, 0.0093, 0.0192, 0.0099, 0.0121, 0.0296, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0194, 0.0172, 0.0173, 0.0195, 0.0150, 0.0183, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:57:00,766 INFO [finetune.py:992] (1/2) Epoch 4, batch 9950, loss[loss=0.1649, simple_loss=0.2559, pruned_loss=0.03695, over 12091.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2629, pruned_loss=0.04463, over 2360272.83 frames. ], batch size: 32, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:57:00,973 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146652.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:57:03,080 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5182, 5.3424, 5.4518, 5.4933, 5.0822, 5.1209, 4.9241, 5.4316], device='cuda:1'), covar=tensor([0.0574, 0.0494, 0.0546, 0.0460, 0.1625, 0.1180, 0.0499, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0639, 0.0547, 0.0589, 0.0776, 0.0705, 0.0513, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 00:57:25,775 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146687.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 00:57:35,568 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5734, 3.4757, 3.2492, 3.0906, 2.8783, 2.7224, 3.5289, 2.2907], device='cuda:1'), covar=tensor([0.0286, 0.0164, 0.0150, 0.0153, 0.0324, 0.0267, 0.0120, 0.0388], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0153, 0.0147, 0.0172, 0.0191, 0.0185, 0.0152, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 00:57:36,725 INFO [finetune.py:992] (1/2) Epoch 4, batch 10000, loss[loss=0.1716, simple_loss=0.2668, pruned_loss=0.03824, over 12125.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2631, pruned_loss=0.04441, over 2368805.37 frames. ], batch size: 39, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:57:55,920 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.009e+02 2.778e+02 3.525e+02 4.163e+02 5.985e+02, threshold=7.050e+02, percent-clipped=0.0 2023-05-16 00:57:57,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 00:58:03,060 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2445, 2.0710, 3.1767, 4.0522, 2.1628, 4.2214, 4.1577, 4.2592], device='cuda:1'), covar=tensor([0.0134, 0.1329, 0.0418, 0.0145, 0.1315, 0.0217, 0.0170, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0202, 0.0186, 0.0119, 0.0191, 0.0176, 0.0170, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 00:58:12,761 INFO [finetune.py:992] (1/2) Epoch 4, batch 10050, loss[loss=0.1735, simple_loss=0.268, pruned_loss=0.03953, over 12027.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2633, pruned_loss=0.04408, over 2376562.44 frames. ], batch size: 31, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:58:48,162 INFO [finetune.py:992] (1/2) Epoch 4, batch 10100, loss[loss=0.1398, simple_loss=0.2247, pruned_loss=0.02743, over 12253.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2638, pruned_loss=0.04398, over 2369843.83 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:58:50,458 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146805.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:59:07,474 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.944e+02 3.443e+02 4.252e+02 7.137e+02, threshold=6.886e+02, percent-clipped=1.0 2023-05-16 00:59:08,470 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4959, 3.6397, 3.3657, 3.2098, 2.9739, 2.8057, 3.6930, 2.3614], device='cuda:1'), covar=tensor([0.0308, 0.0105, 0.0130, 0.0156, 0.0305, 0.0249, 0.0093, 0.0379], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0153, 0.0147, 0.0172, 0.0192, 0.0185, 0.0152, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 00:59:19,129 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146844.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 00:59:22,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 00:59:24,496 INFO [finetune.py:992] (1/2) Epoch 4, batch 10150, loss[loss=0.1543, simple_loss=0.241, pruned_loss=0.03376, over 12293.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2633, pruned_loss=0.04358, over 2366972.96 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 00:59:25,260 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146853.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:00:00,626 INFO [finetune.py:992] (1/2) Epoch 4, batch 10200, loss[loss=0.1399, simple_loss=0.2221, pruned_loss=0.0288, over 11997.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2636, pruned_loss=0.04384, over 2371414.22 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 01:00:03,033 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146905.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:00:14,374 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3156, 4.6759, 4.2588, 5.0816, 4.4626, 3.0404, 4.3175, 3.2182], device='cuda:1'), covar=tensor([0.0828, 0.0772, 0.1283, 0.0366, 0.1113, 0.1440, 0.0897, 0.2999], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0368, 0.0344, 0.0260, 0.0352, 0.0260, 0.0331, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:00:19,676 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.856e+02 3.445e+02 4.302e+02 8.868e+02, threshold=6.890e+02, percent-clipped=4.0 2023-05-16 01:00:32,538 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146947.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:00:36,009 INFO [finetune.py:992] (1/2) Epoch 4, batch 10250, loss[loss=0.193, simple_loss=0.2818, pruned_loss=0.05208, over 12059.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2635, pruned_loss=0.04393, over 2369063.21 frames. ], batch size: 37, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 01:00:42,591 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3277, 5.1908, 5.2757, 5.3237, 4.9341, 4.9471, 4.8033, 5.2748], device='cuda:1'), covar=tensor([0.0583, 0.0493, 0.0702, 0.0508, 0.1630, 0.1173, 0.0466, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0641, 0.0551, 0.0585, 0.0777, 0.0702, 0.0510, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 01:00:57,393 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146982.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 01:01:12,376 INFO [finetune.py:992] (1/2) Epoch 4, batch 10300, loss[loss=0.1598, simple_loss=0.2463, pruned_loss=0.03669, over 12186.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04378, over 2369244.24 frames. ], batch size: 31, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 01:01:19,675 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9141, 4.5681, 4.8997, 4.8266, 4.8361, 4.7553, 4.7221, 2.3873], device='cuda:1'), covar=tensor([0.0169, 0.0095, 0.0102, 0.0083, 0.0057, 0.0164, 0.0101, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0076, 0.0070, 0.0057, 0.0088, 0.0075, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:01:26,652 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9952, 4.6775, 4.9616, 4.2908, 4.6921, 4.3572, 4.9405, 4.6883], device='cuda:1'), covar=tensor([0.0320, 0.0391, 0.0350, 0.0290, 0.0304, 0.0313, 0.0313, 0.0392], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0242, 0.0259, 0.0234, 0.0232, 0.0232, 0.0211, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:01:32,081 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 2.716e+02 3.181e+02 3.895e+02 6.567e+02, threshold=6.362e+02, percent-clipped=0.0 2023-05-16 01:01:48,361 INFO [finetune.py:992] (1/2) Epoch 4, batch 10350, loss[loss=0.1491, simple_loss=0.233, pruned_loss=0.03263, over 11949.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2635, pruned_loss=0.04401, over 2367724.98 frames. ], batch size: 28, lr: 4.76e-03, grad_scale: 16.0 2023-05-16 01:02:17,554 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4938, 2.2421, 3.5627, 4.5092, 3.7521, 4.3550, 3.8268, 3.0470], device='cuda:1'), covar=tensor([0.0030, 0.0395, 0.0148, 0.0029, 0.0123, 0.0065, 0.0099, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0121, 0.0103, 0.0075, 0.0101, 0.0112, 0.0090, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:02:23,873 INFO [finetune.py:992] (1/2) Epoch 4, batch 10400, loss[loss=0.137, simple_loss=0.2202, pruned_loss=0.02687, over 12274.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2629, pruned_loss=0.04344, over 2376461.00 frames. ], batch size: 28, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:02:34,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.57 vs. limit=5.0 2023-05-16 01:02:37,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0690, 2.4671, 3.6887, 3.1033, 3.4622, 3.1629, 2.5722, 3.5640], device='cuda:1'), covar=tensor([0.0119, 0.0307, 0.0112, 0.0189, 0.0127, 0.0147, 0.0312, 0.0103], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0193, 0.0172, 0.0173, 0.0195, 0.0150, 0.0184, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:02:43,169 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.124e+02 2.939e+02 3.483e+02 4.123e+02 8.452e+02, threshold=6.965e+02, percent-clipped=4.0 2023-05-16 01:03:00,260 INFO [finetune.py:992] (1/2) Epoch 4, batch 10450, loss[loss=0.1618, simple_loss=0.2408, pruned_loss=0.0414, over 12018.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2632, pruned_loss=0.04357, over 2371784.33 frames. ], batch size: 28, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:03:35,085 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147200.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:03:36,374 INFO [finetune.py:992] (1/2) Epoch 4, batch 10500, loss[loss=0.1971, simple_loss=0.2836, pruned_loss=0.05526, over 12160.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.263, pruned_loss=0.04327, over 2380612.79 frames. ], batch size: 34, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:03:52,896 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4952, 2.0569, 3.7010, 4.5741, 3.9653, 4.3333, 4.0656, 3.0913], device='cuda:1'), covar=tensor([0.0041, 0.0524, 0.0160, 0.0029, 0.0118, 0.0076, 0.0079, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0122, 0.0103, 0.0075, 0.0101, 0.0112, 0.0090, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:03:55,492 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 3.015e+02 3.764e+02 4.290e+02 1.266e+03, threshold=7.528e+02, percent-clipped=2.0 2023-05-16 01:04:04,336 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147241.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:04:08,427 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147247.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:04:11,828 INFO [finetune.py:992] (1/2) Epoch 4, batch 10550, loss[loss=0.1768, simple_loss=0.2651, pruned_loss=0.04423, over 12351.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2622, pruned_loss=0.04303, over 2376731.43 frames. ], batch size: 35, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:04:33,436 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:04:40,472 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3384, 5.1907, 5.2744, 5.3060, 4.9265, 4.9870, 4.8675, 5.2746], device='cuda:1'), covar=tensor([0.0626, 0.0490, 0.0635, 0.0544, 0.1615, 0.1139, 0.0462, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0638, 0.0550, 0.0586, 0.0772, 0.0697, 0.0505, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 01:04:41,980 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1296, 6.1043, 5.8879, 5.4164, 5.2489, 5.9944, 5.5543, 5.4145], device='cuda:1'), covar=tensor([0.0626, 0.0833, 0.0555, 0.1520, 0.0684, 0.0771, 0.1511, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0585, 0.0524, 0.0491, 0.0604, 0.0392, 0.0676, 0.0747, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 01:04:43,317 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147295.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:04:48,323 INFO [finetune.py:992] (1/2) Epoch 4, batch 10600, loss[loss=0.1829, simple_loss=0.2702, pruned_loss=0.04776, over 12412.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2619, pruned_loss=0.04287, over 2374637.66 frames. ], batch size: 32, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:04:48,530 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147302.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:04:52,433 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-16 01:04:53,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 01:04:54,268 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147310.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:05:08,437 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.841e+02 3.295e+02 4.134e+02 6.793e+02, threshold=6.590e+02, percent-clipped=0.0 2023-05-16 01:05:09,208 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147330.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:05:09,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 01:05:24,633 INFO [finetune.py:992] (1/2) Epoch 4, batch 10650, loss[loss=0.148, simple_loss=0.2332, pruned_loss=0.03137, over 12291.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2618, pruned_loss=0.04293, over 2378802.02 frames. ], batch size: 28, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:05:27,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-05-16 01:05:37,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0894, 4.3244, 3.8922, 4.6209, 4.2010, 2.6759, 3.9257, 2.8342], device='cuda:1'), covar=tensor([0.0702, 0.0816, 0.1211, 0.0480, 0.1017, 0.1607, 0.0976, 0.3141], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0371, 0.0347, 0.0263, 0.0356, 0.0264, 0.0333, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:05:38,494 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147371.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:06:00,681 INFO [finetune.py:992] (1/2) Epoch 4, batch 10700, loss[loss=0.1766, simple_loss=0.2676, pruned_loss=0.0428, over 12296.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2611, pruned_loss=0.04246, over 2382403.97 frames. ], batch size: 34, lr: 4.75e-03, grad_scale: 32.0 2023-05-16 01:06:04,499 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147407.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:06:20,461 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.905e+02 3.370e+02 4.057e+02 8.796e+02, threshold=6.739e+02, percent-clipped=3.0 2023-05-16 01:06:23,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-05-16 01:06:24,178 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147434.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:06:36,933 INFO [finetune.py:992] (1/2) Epoch 4, batch 10750, loss[loss=0.1653, simple_loss=0.2574, pruned_loss=0.03664, over 11820.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2617, pruned_loss=0.04272, over 2382557.15 frames. ], batch size: 44, lr: 4.75e-03, grad_scale: 32.0 2023-05-16 01:06:44,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 01:06:48,875 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147468.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:07:08,035 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147495.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:07:11,626 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147500.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:07:12,880 INFO [finetune.py:992] (1/2) Epoch 4, batch 10800, loss[loss=0.148, simple_loss=0.2337, pruned_loss=0.03113, over 12109.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2622, pruned_loss=0.04297, over 2382355.26 frames. ], batch size: 30, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:07:32,862 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.218e+02 2.861e+02 3.389e+02 3.924e+02 8.164e+02, threshold=6.777e+02, percent-clipped=1.0 2023-05-16 01:07:45,615 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147548.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:07:48,406 INFO [finetune.py:992] (1/2) Epoch 4, batch 10850, loss[loss=0.1798, simple_loss=0.2524, pruned_loss=0.05353, over 12349.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2634, pruned_loss=0.04413, over 2370087.44 frames. ], batch size: 30, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:07:48,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 01:07:58,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3063, 4.7464, 2.8600, 2.9302, 3.9841, 2.5523, 4.0718, 3.3141], device='cuda:1'), covar=tensor([0.0622, 0.0483, 0.1060, 0.1191, 0.0268, 0.1251, 0.0423, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0249, 0.0175, 0.0198, 0.0139, 0.0179, 0.0193, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:08:19,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 01:08:22,349 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147597.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:08:26,830 INFO [finetune.py:992] (1/2) Epoch 4, batch 10900, loss[loss=0.1907, simple_loss=0.2791, pruned_loss=0.05118, over 11363.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.264, pruned_loss=0.04428, over 2362683.51 frames. ], batch size: 55, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:08:46,844 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.788e+02 3.418e+02 4.406e+02 8.127e+02, threshold=6.835e+02, percent-clipped=5.0 2023-05-16 01:09:02,424 INFO [finetune.py:992] (1/2) Epoch 4, batch 10950, loss[loss=0.1952, simple_loss=0.2813, pruned_loss=0.05458, over 11649.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2642, pruned_loss=0.04453, over 2362678.94 frames. ], batch size: 48, lr: 4.75e-03, grad_scale: 16.0 2023-05-16 01:09:07,641 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147659.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:09:12,591 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147666.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:09:38,154 INFO [finetune.py:992] (1/2) Epoch 4, batch 11000, loss[loss=0.1553, simple_loss=0.2434, pruned_loss=0.03358, over 12187.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2675, pruned_loss=0.04655, over 2337990.73 frames. ], batch size: 31, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:09:51,081 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147720.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:09:59,217 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.248e+02 3.094e+02 3.686e+02 4.339e+02 1.027e+03, threshold=7.372e+02, percent-clipped=4.0 2023-05-16 01:10:07,865 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1208, 6.0538, 5.8338, 5.3982, 5.2087, 5.9786, 5.6086, 5.4080], device='cuda:1'), covar=tensor([0.0489, 0.0752, 0.0553, 0.1487, 0.0565, 0.0544, 0.1310, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0512, 0.0481, 0.0591, 0.0380, 0.0657, 0.0731, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 01:10:14,148 INFO [finetune.py:992] (1/2) Epoch 4, batch 11050, loss[loss=0.2428, simple_loss=0.3132, pruned_loss=0.08618, over 8007.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2705, pruned_loss=0.04834, over 2306054.19 frames. ], batch size: 97, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:10:21,780 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147763.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:10:40,572 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147790.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:10:49,709 INFO [finetune.py:992] (1/2) Epoch 4, batch 11100, loss[loss=0.1689, simple_loss=0.2641, pruned_loss=0.03682, over 12343.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.2764, pruned_loss=0.05213, over 2259078.66 frames. ], batch size: 31, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:10:58,213 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1292, 2.4163, 3.5690, 3.0746, 3.3740, 3.1904, 2.5854, 3.5349], device='cuda:1'), covar=tensor([0.0092, 0.0301, 0.0109, 0.0209, 0.0136, 0.0141, 0.0279, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0192, 0.0171, 0.0173, 0.0194, 0.0149, 0.0183, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:10:58,898 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147815.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:11:10,609 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 3.432e+02 4.091e+02 4.867e+02 8.600e+02, threshold=8.181e+02, percent-clipped=2.0 2023-05-16 01:11:16,942 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7433, 3.7563, 3.7326, 3.8184, 3.6034, 3.6092, 3.5423, 3.7423], device='cuda:1'), covar=tensor([0.1125, 0.0713, 0.1380, 0.0703, 0.1688, 0.1364, 0.0586, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0630, 0.0542, 0.0577, 0.0764, 0.0693, 0.0502, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 01:11:25,140 INFO [finetune.py:992] (1/2) Epoch 4, batch 11150, loss[loss=0.2266, simple_loss=0.3097, pruned_loss=0.07174, over 10529.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2823, pruned_loss=0.05657, over 2198920.40 frames. ], batch size: 68, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:11:28,825 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147857.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:11:41,976 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147876.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:11:57,245 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147897.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:12:00,309 INFO [finetune.py:992] (1/2) Epoch 4, batch 11200, loss[loss=0.2637, simple_loss=0.3448, pruned_loss=0.09129, over 10393.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2905, pruned_loss=0.06253, over 2110275.52 frames. ], batch size: 68, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:12:07,344 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147911.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:12:12,126 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147918.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:12:12,940 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8724, 2.6407, 3.4708, 3.5819, 3.0183, 2.7931, 2.7107, 2.5287], device='cuda:1'), covar=tensor([0.0864, 0.1960, 0.0467, 0.0404, 0.0703, 0.1398, 0.1855, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0369, 0.0264, 0.0285, 0.0251, 0.0279, 0.0348, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:12:20,929 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.360e+02 3.735e+02 4.518e+02 5.395e+02 1.112e+03, threshold=9.037e+02, percent-clipped=6.0 2023-05-16 01:12:30,865 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147945.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:12:35,697 INFO [finetune.py:992] (1/2) Epoch 4, batch 11250, loss[loss=0.242, simple_loss=0.3274, pruned_loss=0.07831, over 10529.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2973, pruned_loss=0.06716, over 2052858.23 frames. ], batch size: 68, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:12:39,922 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9230, 4.4916, 4.0812, 4.2586, 4.5885, 3.9746, 4.1825, 4.1155], device='cuda:1'), covar=tensor([0.1423, 0.1064, 0.1375, 0.1608, 0.0989, 0.2165, 0.1536, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0448, 0.0351, 0.0398, 0.0431, 0.0406, 0.0364, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:12:46,127 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147966.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:12:50,168 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147972.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:12:56,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.34 vs. limit=5.0 2023-05-16 01:13:14,696 INFO [finetune.py:992] (1/2) Epoch 4, batch 11300, loss[loss=0.2434, simple_loss=0.3288, pruned_loss=0.07898, over 10911.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3044, pruned_loss=0.07263, over 1977083.38 frames. ], batch size: 55, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:13:23,172 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148014.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:13:23,942 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148015.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:13:35,232 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.605e+02 3.606e+02 4.302e+02 4.951e+02 7.972e+02, threshold=8.603e+02, percent-clipped=0.0 2023-05-16 01:13:50,026 INFO [finetune.py:992] (1/2) Epoch 4, batch 11350, loss[loss=0.1877, simple_loss=0.2762, pruned_loss=0.04957, over 12185.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3086, pruned_loss=0.07531, over 1934049.87 frames. ], batch size: 35, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:13:57,636 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148063.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:14:16,774 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148090.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:14:24,639 INFO [finetune.py:992] (1/2) Epoch 4, batch 11400, loss[loss=0.3008, simple_loss=0.36, pruned_loss=0.1208, over 6995.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3128, pruned_loss=0.07827, over 1903714.69 frames. ], batch size: 98, lr: 4.75e-03, grad_scale: 8.0 2023-05-16 01:14:31,560 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148111.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:14:35,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 01:14:41,163 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-05-16 01:14:44,856 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.328e+02 3.768e+02 4.287e+02 4.920e+02 1.036e+03, threshold=8.574e+02, percent-clipped=2.0 2023-05-16 01:14:49,559 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148138.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:14:59,571 INFO [finetune.py:992] (1/2) Epoch 4, batch 11450, loss[loss=0.2431, simple_loss=0.3228, pruned_loss=0.08168, over 7332.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3157, pruned_loss=0.08073, over 1875537.87 frames. ], batch size: 98, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:15:05,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 01:15:09,891 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6556, 4.5873, 4.4965, 4.5843, 4.1733, 4.7232, 4.7035, 4.8261], device='cuda:1'), covar=tensor([0.0179, 0.0130, 0.0179, 0.0273, 0.0650, 0.0198, 0.0135, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0168, 0.0167, 0.0215, 0.0213, 0.0182, 0.0154, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 01:15:12,470 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148171.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:15:33,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-16 01:15:34,179 INFO [finetune.py:992] (1/2) Epoch 4, batch 11500, loss[loss=0.2742, simple_loss=0.3384, pruned_loss=0.105, over 6815.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3201, pruned_loss=0.08423, over 1830446.96 frames. ], batch size: 101, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:15:42,459 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148213.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:15:45,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 01:15:54,608 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.676e+02 3.747e+02 4.510e+02 5.605e+02 1.553e+03, threshold=9.020e+02, percent-clipped=3.0 2023-05-16 01:16:09,818 INFO [finetune.py:992] (1/2) Epoch 4, batch 11550, loss[loss=0.2656, simple_loss=0.3245, pruned_loss=0.1033, over 6980.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3202, pruned_loss=0.0848, over 1814957.05 frames. ], batch size: 102, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:16:19,906 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148267.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:16:27,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1504, 4.8353, 5.0546, 4.3924, 4.8707, 4.4854, 5.0186, 4.8504], device='cuda:1'), covar=tensor([0.0377, 0.0395, 0.0486, 0.0319, 0.0358, 0.0391, 0.0383, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0226, 0.0244, 0.0221, 0.0219, 0.0218, 0.0196, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:16:33,881 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148287.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:16:44,356 INFO [finetune.py:992] (1/2) Epoch 4, batch 11600, loss[loss=0.2739, simple_loss=0.327, pruned_loss=0.1104, over 6723.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3217, pruned_loss=0.08581, over 1805182.98 frames. ], batch size: 100, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:16:53,069 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148315.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:17:04,084 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.748e+02 3.470e+02 3.977e+02 4.660e+02 1.039e+03, threshold=7.954e+02, percent-clipped=1.0 2023-05-16 01:17:06,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 01:17:17,695 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148348.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:17:20,374 INFO [finetune.py:992] (1/2) Epoch 4, batch 11650, loss[loss=0.2484, simple_loss=0.31, pruned_loss=0.09341, over 6904.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3208, pruned_loss=0.08667, over 1789290.00 frames. ], batch size: 99, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:17:28,883 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148363.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:17:55,232 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7333, 3.4384, 3.6503, 3.7005, 3.6583, 3.7398, 3.6326, 2.6357], device='cuda:1'), covar=tensor([0.0087, 0.0078, 0.0099, 0.0070, 0.0063, 0.0098, 0.0081, 0.0680], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0068, 0.0072, 0.0065, 0.0054, 0.0083, 0.0070, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:17:55,683 INFO [finetune.py:992] (1/2) Epoch 4, batch 11700, loss[loss=0.2797, simple_loss=0.3356, pruned_loss=0.1119, over 7056.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3203, pruned_loss=0.08706, over 1758435.53 frames. ], batch size: 98, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:18:14,528 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148429.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:18:15,795 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.380e+02 3.656e+02 4.264e+02 4.955e+02 6.990e+02, threshold=8.528e+02, percent-clipped=1.0 2023-05-16 01:18:27,339 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7327, 3.7501, 3.6913, 3.7936, 3.5913, 3.6126, 3.6152, 3.7119], device='cuda:1'), covar=tensor([0.0952, 0.0679, 0.1429, 0.0639, 0.1592, 0.1264, 0.0557, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0588, 0.0511, 0.0537, 0.0713, 0.0652, 0.0476, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:18:29,861 INFO [finetune.py:992] (1/2) Epoch 4, batch 11750, loss[loss=0.2288, simple_loss=0.3147, pruned_loss=0.07145, over 10502.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3205, pruned_loss=0.08727, over 1743587.00 frames. ], batch size: 68, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:18:43,422 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148471.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:18:56,992 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148490.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:00,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7484, 3.0654, 2.3358, 2.1724, 2.7873, 2.2881, 2.9400, 2.5570], device='cuda:1'), covar=tensor([0.0533, 0.0569, 0.0876, 0.1398, 0.0258, 0.1043, 0.0436, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0228, 0.0164, 0.0190, 0.0129, 0.0174, 0.0180, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:19:02,408 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148498.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:02,657 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 01:19:04,981 INFO [finetune.py:992] (1/2) Epoch 4, batch 11800, loss[loss=0.2507, simple_loss=0.3358, pruned_loss=0.08284, over 11501.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3225, pruned_loss=0.08918, over 1710824.20 frames. ], batch size: 48, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:19:10,179 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3581, 3.0001, 3.6080, 2.2748, 2.6616, 3.1182, 2.8966, 3.2417], device='cuda:1'), covar=tensor([0.0418, 0.0853, 0.0250, 0.1285, 0.1501, 0.1076, 0.1016, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0217, 0.0213, 0.0169, 0.0221, 0.0263, 0.0210, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:19:12,770 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148513.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:17,495 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148519.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:19,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 01:19:24,878 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6240, 4.5935, 4.5696, 4.6419, 4.3652, 4.3555, 4.3000, 4.5537], device='cuda:1'), covar=tensor([0.0709, 0.0527, 0.0829, 0.0643, 0.1763, 0.1307, 0.0501, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0587, 0.0511, 0.0537, 0.0711, 0.0652, 0.0476, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:19:25,330 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.324e+02 3.595e+02 4.321e+02 4.889e+02 1.131e+03, threshold=8.643e+02, percent-clipped=3.0 2023-05-16 01:19:30,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 01:19:34,296 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3611, 3.1219, 3.1148, 3.3143, 2.8400, 3.1402, 2.5159, 2.8213], device='cuda:1'), covar=tensor([0.1852, 0.0849, 0.0763, 0.0483, 0.1043, 0.0815, 0.1737, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0263, 0.0289, 0.0343, 0.0235, 0.0239, 0.0257, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:19:35,834 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 01:19:40,055 INFO [finetune.py:992] (1/2) Epoch 4, batch 11850, loss[loss=0.2908, simple_loss=0.3492, pruned_loss=0.1162, over 6078.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3237, pruned_loss=0.08943, over 1710393.74 frames. ], batch size: 98, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:19:45,031 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148559.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:46,252 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148561.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:50,369 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148567.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:19:50,458 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0774, 2.1426, 3.1164, 3.9786, 2.3616, 3.9814, 4.0491, 4.1402], device='cuda:1'), covar=tensor([0.0119, 0.1327, 0.0392, 0.0117, 0.1214, 0.0218, 0.0182, 0.0083], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0194, 0.0176, 0.0110, 0.0180, 0.0163, 0.0156, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:19:51,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-16 01:19:55,954 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-05-16 01:20:14,991 INFO [finetune.py:992] (1/2) Epoch 4, batch 11900, loss[loss=0.2333, simple_loss=0.3053, pruned_loss=0.08061, over 6683.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3228, pruned_loss=0.08776, over 1716125.40 frames. ], batch size: 99, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:20:23,829 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:20:27,060 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8657, 4.5197, 4.0721, 4.2595, 4.5510, 4.0093, 4.2492, 4.0355], device='cuda:1'), covar=tensor([0.1632, 0.0934, 0.1144, 0.1662, 0.1019, 0.1947, 0.1451, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0431, 0.0340, 0.0385, 0.0415, 0.0394, 0.0351, 0.0336], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 01:20:34,366 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 3.443e+02 3.857e+02 4.683e+02 7.167e+02, threshold=7.714e+02, percent-clipped=0.0 2023-05-16 01:20:43,189 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148643.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:20:48,514 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148651.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:20:48,966 INFO [finetune.py:992] (1/2) Epoch 4, batch 11950, loss[loss=0.2261, simple_loss=0.2995, pruned_loss=0.07636, over 7183.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3192, pruned_loss=0.08475, over 1707171.11 frames. ], batch size: 99, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:21:24,252 INFO [finetune.py:992] (1/2) Epoch 4, batch 12000, loss[loss=0.1985, simple_loss=0.2875, pruned_loss=0.05476, over 11060.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3141, pruned_loss=0.08048, over 1702572.97 frames. ], batch size: 55, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:21:24,253 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 01:21:37,952 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8159, 2.1368, 2.6555, 2.8098, 2.6896, 2.8322, 2.6474, 2.2606], device='cuda:1'), covar=tensor([0.0063, 0.0247, 0.0145, 0.0053, 0.0081, 0.0075, 0.0106, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0114, 0.0095, 0.0068, 0.0092, 0.0104, 0.0083, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:21:43,114 INFO [finetune.py:1026] (1/2) Epoch 4, validation: loss=0.2922, simple_loss=0.3681, pruned_loss=0.1082, over 1020973.00 frames. 2023-05-16 01:21:43,114 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 01:21:49,997 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148712.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:21:52,661 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148716.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:22:00,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.12 vs. limit=5.0 2023-05-16 01:22:02,469 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.183e+02 3.126e+02 3.672e+02 4.633e+02 1.109e+03, threshold=7.344e+02, percent-clipped=3.0 2023-05-16 01:22:16,797 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5433, 4.5207, 4.4469, 4.0776, 4.2066, 4.5111, 4.2786, 4.1298], device='cuda:1'), covar=tensor([0.0671, 0.0751, 0.0608, 0.1196, 0.1803, 0.0773, 0.1246, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0474, 0.0443, 0.0543, 0.0352, 0.0600, 0.0655, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 01:22:17,274 INFO [finetune.py:992] (1/2) Epoch 4, batch 12050, loss[loss=0.222, simple_loss=0.2953, pruned_loss=0.07431, over 6829.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3093, pruned_loss=0.07683, over 1702855.97 frames. ], batch size: 97, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:22:34,165 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148777.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:22:39,063 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148785.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:22:50,082 INFO [finetune.py:992] (1/2) Epoch 4, batch 12100, loss[loss=0.2353, simple_loss=0.3185, pruned_loss=0.07602, over 11466.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.308, pruned_loss=0.07549, over 1711587.34 frames. ], batch size: 48, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:23:00,385 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2462, 4.9243, 5.2013, 4.6608, 4.9188, 4.7000, 5.1773, 4.9362], device='cuda:1'), covar=tensor([0.0253, 0.0340, 0.0229, 0.0216, 0.0308, 0.0265, 0.0255, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0216, 0.0230, 0.0210, 0.0209, 0.0207, 0.0186, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:23:08,414 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 3.069e+02 3.619e+02 4.456e+02 7.686e+02, threshold=7.237e+02, percent-clipped=2.0 2023-05-16 01:23:19,815 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-16 01:23:21,826 INFO [finetune.py:992] (1/2) Epoch 4, batch 12150, loss[loss=0.2663, simple_loss=0.3352, pruned_loss=0.09865, over 6992.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3091, pruned_loss=0.07574, over 1708363.27 frames. ], batch size: 98, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:23:23,178 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:23:53,746 INFO [finetune.py:992] (1/2) Epoch 4, batch 12200, loss[loss=0.2237, simple_loss=0.2976, pruned_loss=0.07488, over 7110.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3098, pruned_loss=0.07653, over 1695407.57 frames. ], batch size: 100, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:24:11,577 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.560e+02 3.330e+02 3.899e+02 4.842e+02 8.998e+02, threshold=7.799e+02, percent-clipped=1.0 2023-05-16 01:24:39,472 INFO [finetune.py:992] (1/2) Epoch 5, batch 0, loss[loss=0.228, simple_loss=0.3104, pruned_loss=0.07282, over 12377.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3104, pruned_loss=0.07282, over 12377.00 frames. ], batch size: 38, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:24:39,473 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 01:24:57,514 INFO [finetune.py:1026] (1/2) Epoch 5, validation: loss=0.2918, simple_loss=0.3673, pruned_loss=0.1081, over 1020973.00 frames. 2023-05-16 01:24:57,515 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 01:25:03,310 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148943.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:25:17,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 01:25:33,857 INFO [finetune.py:992] (1/2) Epoch 5, batch 50, loss[loss=0.1898, simple_loss=0.2842, pruned_loss=0.04768, over 11835.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2758, pruned_loss=0.05027, over 535454.60 frames. ], batch size: 44, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:25:38,883 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148991.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:25:50,670 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149007.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:25:59,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-16 01:26:07,642 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.225e+02 3.099e+02 3.518e+02 4.160e+02 1.565e+03, threshold=7.036e+02, percent-clipped=2.0 2023-05-16 01:26:10,539 INFO [finetune.py:992] (1/2) Epoch 5, batch 100, loss[loss=0.2037, simple_loss=0.2945, pruned_loss=0.05644, over 11770.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2736, pruned_loss=0.04811, over 945794.37 frames. ], batch size: 44, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:26:37,085 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149072.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:26:46,271 INFO [finetune.py:992] (1/2) Epoch 5, batch 150, loss[loss=0.1853, simple_loss=0.2823, pruned_loss=0.04414, over 12050.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2718, pruned_loss=0.04733, over 1270695.60 frames. ], batch size: 40, lr: 4.74e-03, grad_scale: 8.0 2023-05-16 01:26:46,453 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149085.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:27:19,631 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.062e+02 3.007e+02 3.400e+02 3.916e+02 6.170e+02, threshold=6.801e+02, percent-clipped=0.0 2023-05-16 01:27:20,513 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9597, 5.9965, 5.6666, 5.1953, 5.0130, 5.8054, 5.4921, 5.1894], device='cuda:1'), covar=tensor([0.0687, 0.0714, 0.0662, 0.1470, 0.0607, 0.0802, 0.1496, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0480, 0.0452, 0.0557, 0.0357, 0.0617, 0.0670, 0.0499], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:27:21,185 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149133.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:27:22,584 INFO [finetune.py:992] (1/2) Epoch 5, batch 200, loss[loss=0.1524, simple_loss=0.232, pruned_loss=0.03637, over 12280.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2699, pruned_loss=0.04675, over 1517711.90 frames. ], batch size: 28, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:27:36,999 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149154.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:27:58,881 INFO [finetune.py:992] (1/2) Epoch 5, batch 250, loss[loss=0.1925, simple_loss=0.2822, pruned_loss=0.05136, over 12157.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.2695, pruned_loss=0.04643, over 1715232.66 frames. ], batch size: 36, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:28:11,307 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149202.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:28:31,782 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 2.796e+02 3.378e+02 4.055e+02 7.311e+02, threshold=6.757e+02, percent-clipped=2.0 2023-05-16 01:28:34,670 INFO [finetune.py:992] (1/2) Epoch 5, batch 300, loss[loss=0.1788, simple_loss=0.2725, pruned_loss=0.04262, over 12362.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2691, pruned_loss=0.0466, over 1858166.44 frames. ], batch size: 35, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:28:37,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 01:28:59,848 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2623, 4.6012, 4.0690, 4.9250, 4.4470, 2.9340, 4.2609, 2.9984], device='cuda:1'), covar=tensor([0.0793, 0.0866, 0.1474, 0.0497, 0.1093, 0.1562, 0.0914, 0.3278], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0358, 0.0335, 0.0246, 0.0344, 0.0256, 0.0324, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:29:10,896 INFO [finetune.py:992] (1/2) Epoch 5, batch 350, loss[loss=0.1777, simple_loss=0.2694, pruned_loss=0.04296, over 12345.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2674, pruned_loss=0.04609, over 1980010.02 frames. ], batch size: 38, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:29:15,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.54 vs. limit=2.0 2023-05-16 01:29:27,482 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149307.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:29:39,951 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 01:29:40,400 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2050, 4.7564, 5.1180, 5.1035, 4.9121, 5.1270, 5.0370, 2.9805], device='cuda:1'), covar=tensor([0.0101, 0.0065, 0.0064, 0.0052, 0.0044, 0.0075, 0.0064, 0.0612], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0068, 0.0072, 0.0065, 0.0054, 0.0082, 0.0069, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:29:44,550 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.132e+02 2.873e+02 3.199e+02 3.807e+02 8.710e+02, threshold=6.398e+02, percent-clipped=1.0 2023-05-16 01:29:47,359 INFO [finetune.py:992] (1/2) Epoch 5, batch 400, loss[loss=0.1705, simple_loss=0.2527, pruned_loss=0.0441, over 12165.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2651, pruned_loss=0.04522, over 2075287.66 frames. ], batch size: 29, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:30:01,900 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149355.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:30:10,808 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8291, 4.4522, 4.7373, 4.7897, 4.5886, 4.8338, 4.6741, 2.7516], device='cuda:1'), covar=tensor([0.0142, 0.0070, 0.0088, 0.0058, 0.0052, 0.0079, 0.0071, 0.0676], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0068, 0.0072, 0.0065, 0.0054, 0.0082, 0.0069, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:30:14,311 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149372.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:30:16,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-05-16 01:30:23,333 INFO [finetune.py:992] (1/2) Epoch 5, batch 450, loss[loss=0.1815, simple_loss=0.2732, pruned_loss=0.04486, over 12272.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2648, pruned_loss=0.04513, over 2136633.22 frames. ], batch size: 32, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:30:31,440 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4577, 4.8542, 2.9805, 2.7934, 4.2386, 2.7749, 4.1699, 3.5127], device='cuda:1'), covar=tensor([0.0603, 0.0459, 0.1079, 0.1454, 0.0249, 0.1256, 0.0406, 0.0733], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0234, 0.0168, 0.0195, 0.0132, 0.0176, 0.0183, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:30:49,046 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:30:56,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.946e+02 3.604e+02 4.194e+02 7.984e+02, threshold=7.208e+02, percent-clipped=3.0 2023-05-16 01:30:59,616 INFO [finetune.py:992] (1/2) Epoch 5, batch 500, loss[loss=0.1839, simple_loss=0.2735, pruned_loss=0.04712, over 11601.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2655, pruned_loss=0.04543, over 2190755.15 frames. ], batch size: 48, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:31:35,612 INFO [finetune.py:992] (1/2) Epoch 5, batch 550, loss[loss=0.2365, simple_loss=0.3338, pruned_loss=0.06961, over 12038.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2652, pruned_loss=0.04508, over 2230681.79 frames. ], batch size: 42, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:31:56,683 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149514.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:32:08,208 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.934e+02 3.488e+02 4.250e+02 1.156e+03, threshold=6.977e+02, percent-clipped=3.0 2023-05-16 01:32:10,973 INFO [finetune.py:992] (1/2) Epoch 5, batch 600, loss[loss=0.1561, simple_loss=0.2536, pruned_loss=0.02936, over 12288.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2655, pruned_loss=0.0452, over 2260752.19 frames. ], batch size: 33, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:32:33,174 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149565.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:32:40,354 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149575.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:32:47,897 INFO [finetune.py:992] (1/2) Epoch 5, batch 650, loss[loss=0.1854, simple_loss=0.2779, pruned_loss=0.04641, over 12300.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2649, pruned_loss=0.04461, over 2294460.04 frames. ], batch size: 34, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:33:17,921 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149626.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 01:33:21,111 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.180e+02 2.895e+02 3.340e+02 3.916e+02 6.633e+02, threshold=6.679e+02, percent-clipped=0.0 2023-05-16 01:33:24,019 INFO [finetune.py:992] (1/2) Epoch 5, batch 700, loss[loss=0.1852, simple_loss=0.2671, pruned_loss=0.05171, over 12303.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2643, pruned_loss=0.0441, over 2316959.24 frames. ], batch size: 34, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:33:33,713 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2193, 4.0944, 4.1326, 4.5130, 3.1757, 4.0239, 2.6473, 4.0471], device='cuda:1'), covar=tensor([0.1568, 0.0646, 0.0876, 0.0566, 0.1016, 0.0563, 0.1807, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0258, 0.0290, 0.0343, 0.0234, 0.0236, 0.0255, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:33:59,613 INFO [finetune.py:992] (1/2) Epoch 5, batch 750, loss[loss=0.2197, simple_loss=0.3019, pruned_loss=0.06876, over 10548.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2638, pruned_loss=0.04415, over 2332764.41 frames. ], batch size: 68, lr: 4.73e-03, grad_scale: 8.0 2023-05-16 01:34:33,106 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.109e+02 2.855e+02 3.448e+02 4.582e+02 2.974e+03, threshold=6.895e+02, percent-clipped=9.0 2023-05-16 01:34:36,615 INFO [finetune.py:992] (1/2) Epoch 5, batch 800, loss[loss=0.1848, simple_loss=0.2622, pruned_loss=0.0537, over 12019.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2624, pruned_loss=0.04345, over 2345925.88 frames. ], batch size: 31, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:34:44,581 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149746.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 01:34:59,680 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 01:35:12,264 INFO [finetune.py:992] (1/2) Epoch 5, batch 850, loss[loss=0.1607, simple_loss=0.2479, pruned_loss=0.0367, over 12170.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2625, pruned_loss=0.04374, over 2347164.59 frames. ], batch size: 29, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:35:28,167 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149807.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 01:35:30,183 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8281, 4.8014, 4.6292, 4.7065, 4.3288, 4.8572, 4.8726, 5.0176], device='cuda:1'), covar=tensor([0.0209, 0.0145, 0.0185, 0.0320, 0.0777, 0.0238, 0.0149, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0175, 0.0174, 0.0223, 0.0219, 0.0191, 0.0160, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 01:35:37,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 01:35:45,310 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.781e+02 3.277e+02 3.827e+02 7.458e+02, threshold=6.553e+02, percent-clipped=1.0 2023-05-16 01:35:48,177 INFO [finetune.py:992] (1/2) Epoch 5, batch 900, loss[loss=0.2116, simple_loss=0.2916, pruned_loss=0.06577, over 7797.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2627, pruned_loss=0.0438, over 2352788.48 frames. ], batch size: 98, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:36:13,609 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149870.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:36:20,751 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0317, 5.0342, 4.8436, 4.9335, 4.5380, 5.0798, 5.0327, 5.2903], device='cuda:1'), covar=tensor([0.0240, 0.0116, 0.0152, 0.0246, 0.0687, 0.0221, 0.0127, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0176, 0.0174, 0.0223, 0.0219, 0.0191, 0.0161, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 01:36:23,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-05-16 01:36:24,049 INFO [finetune.py:992] (1/2) Epoch 5, batch 950, loss[loss=0.162, simple_loss=0.2524, pruned_loss=0.03576, over 12273.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2633, pruned_loss=0.04418, over 2365662.89 frames. ], batch size: 33, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:36:31,940 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9986, 5.7901, 5.4174, 5.2889, 5.9493, 5.1819, 5.4399, 5.3496], device='cuda:1'), covar=tensor([0.1623, 0.1087, 0.1055, 0.2077, 0.1022, 0.2411, 0.1737, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0450, 0.0357, 0.0403, 0.0429, 0.0410, 0.0364, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:36:50,128 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149921.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 01:36:57,091 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.907e+02 3.388e+02 4.028e+02 1.015e+03, threshold=6.776e+02, percent-clipped=5.0 2023-05-16 01:36:59,954 INFO [finetune.py:992] (1/2) Epoch 5, batch 1000, loss[loss=0.1744, simple_loss=0.2555, pruned_loss=0.04667, over 12120.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2629, pruned_loss=0.04403, over 2374378.95 frames. ], batch size: 30, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:37:16,698 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6620, 2.3567, 3.3371, 4.4992, 2.2730, 4.4949, 4.5146, 4.6858], device='cuda:1'), covar=tensor([0.0109, 0.1168, 0.0446, 0.0115, 0.1283, 0.0222, 0.0151, 0.0075], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0197, 0.0182, 0.0112, 0.0185, 0.0168, 0.0161, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:37:35,502 INFO [finetune.py:992] (1/2) Epoch 5, batch 1050, loss[loss=0.1783, simple_loss=0.2734, pruned_loss=0.04163, over 12119.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2638, pruned_loss=0.0444, over 2370586.36 frames. ], batch size: 38, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:38:02,753 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-05-16 01:38:12,041 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.704e+02 3.135e+02 3.957e+02 6.595e+02, threshold=6.270e+02, percent-clipped=0.0 2023-05-16 01:38:14,889 INFO [finetune.py:992] (1/2) Epoch 5, batch 1100, loss[loss=0.1522, simple_loss=0.2362, pruned_loss=0.03413, over 12093.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2635, pruned_loss=0.04404, over 2372307.91 frames. ], batch size: 32, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:38:41,183 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9844, 4.5641, 4.9903, 4.3265, 4.6275, 4.3257, 4.9638, 4.6467], device='cuda:1'), covar=tensor([0.0265, 0.0363, 0.0277, 0.0265, 0.0344, 0.0344, 0.0257, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0237, 0.0257, 0.0231, 0.0231, 0.0230, 0.0208, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:38:47,026 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150079.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:38:47,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-05-16 01:38:51,137 INFO [finetune.py:992] (1/2) Epoch 5, batch 1150, loss[loss=0.1941, simple_loss=0.2777, pruned_loss=0.05525, over 12187.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2632, pruned_loss=0.04388, over 2381263.86 frames. ], batch size: 35, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:39:03,257 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150102.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 01:39:23,518 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.882e+02 3.382e+02 3.965e+02 8.013e+02, threshold=6.765e+02, percent-clipped=2.0 2023-05-16 01:39:26,421 INFO [finetune.py:992] (1/2) Epoch 5, batch 1200, loss[loss=0.186, simple_loss=0.28, pruned_loss=0.04602, over 12275.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2637, pruned_loss=0.04386, over 2378439.77 frames. ], batch size: 37, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:39:30,135 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150140.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:39:51,306 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150170.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:40:02,409 INFO [finetune.py:992] (1/2) Epoch 5, batch 1250, loss[loss=0.1809, simple_loss=0.2763, pruned_loss=0.0427, over 11826.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2638, pruned_loss=0.04371, over 2378052.20 frames. ], batch size: 44, lr: 4.73e-03, grad_scale: 16.0 2023-05-16 01:40:27,190 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150218.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:40:29,433 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150221.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:40:36,475 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.917e+02 3.432e+02 3.928e+02 1.096e+03, threshold=6.863e+02, percent-clipped=3.0 2023-05-16 01:40:39,360 INFO [finetune.py:992] (1/2) Epoch 5, batch 1300, loss[loss=0.1558, simple_loss=0.2325, pruned_loss=0.03959, over 12174.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2634, pruned_loss=0.04326, over 2385299.05 frames. ], batch size: 29, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:41:03,803 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150269.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:41:06,774 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8452, 4.6141, 4.5834, 4.8310, 4.6966, 4.8178, 4.7124, 2.8616], device='cuda:1'), covar=tensor([0.0125, 0.0071, 0.0108, 0.0059, 0.0048, 0.0087, 0.0078, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0069, 0.0074, 0.0066, 0.0055, 0.0085, 0.0071, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:41:07,544 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150274.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:41:11,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-05-16 01:41:15,253 INFO [finetune.py:992] (1/2) Epoch 5, batch 1350, loss[loss=0.1525, simple_loss=0.2326, pruned_loss=0.0362, over 12153.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2637, pruned_loss=0.04344, over 2383335.98 frames. ], batch size: 29, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:41:36,349 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0154, 5.0257, 4.8034, 4.8919, 4.5191, 5.1050, 5.0591, 5.2696], device='cuda:1'), covar=tensor([0.0219, 0.0136, 0.0197, 0.0287, 0.0720, 0.0252, 0.0142, 0.0176], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0181, 0.0179, 0.0231, 0.0226, 0.0198, 0.0165, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 01:41:48,907 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.823e+02 3.330e+02 3.848e+02 6.429e+02, threshold=6.661e+02, percent-clipped=0.0 2023-05-16 01:41:51,802 INFO [finetune.py:992] (1/2) Epoch 5, batch 1400, loss[loss=0.2098, simple_loss=0.3002, pruned_loss=0.0597, over 11228.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2648, pruned_loss=0.04372, over 2386107.23 frames. ], batch size: 55, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:41:51,988 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150335.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:42:06,875 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0533, 4.7128, 4.7587, 4.8788, 4.8289, 4.9765, 4.7251, 2.7201], device='cuda:1'), covar=tensor([0.0146, 0.0067, 0.0108, 0.0066, 0.0046, 0.0100, 0.0080, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0070, 0.0075, 0.0067, 0.0055, 0.0086, 0.0072, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:42:27,065 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5154, 3.2781, 5.0124, 2.6186, 2.7913, 3.8254, 3.1677, 3.7914], device='cuda:1'), covar=tensor([0.0456, 0.1095, 0.0305, 0.1169, 0.1780, 0.1234, 0.1307, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0229, 0.0230, 0.0177, 0.0233, 0.0279, 0.0222, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:42:28,127 INFO [finetune.py:992] (1/2) Epoch 5, batch 1450, loss[loss=0.1648, simple_loss=0.2414, pruned_loss=0.04413, over 12354.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2639, pruned_loss=0.04372, over 2385927.77 frames. ], batch size: 31, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:42:32,539 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5382, 3.0436, 4.9592, 2.7078, 2.6730, 3.8657, 3.0879, 3.6084], device='cuda:1'), covar=tensor([0.0368, 0.1265, 0.0307, 0.1111, 0.1837, 0.1166, 0.1356, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0229, 0.0230, 0.0177, 0.0233, 0.0279, 0.0222, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:42:40,552 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150402.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 01:42:56,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 01:43:00,976 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.801e+02 3.263e+02 4.020e+02 6.215e+02, threshold=6.526e+02, percent-clipped=0.0 2023-05-16 01:43:03,765 INFO [finetune.py:992] (1/2) Epoch 5, batch 1500, loss[loss=0.1674, simple_loss=0.2475, pruned_loss=0.04368, over 12138.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.265, pruned_loss=0.04416, over 2386577.24 frames. ], batch size: 30, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:43:03,848 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150435.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:43:14,399 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150450.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 01:43:39,647 INFO [finetune.py:992] (1/2) Epoch 5, batch 1550, loss[loss=0.1765, simple_loss=0.2708, pruned_loss=0.04117, over 12281.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2644, pruned_loss=0.04375, over 2388354.05 frames. ], batch size: 37, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:44:08,860 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0210, 2.3975, 3.5439, 2.8790, 3.3301, 3.1220, 2.4857, 3.4624], device='cuda:1'), covar=tensor([0.0101, 0.0299, 0.0165, 0.0229, 0.0124, 0.0153, 0.0271, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0194, 0.0169, 0.0172, 0.0194, 0.0147, 0.0185, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:44:12,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 01:44:12,913 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.822e+02 3.242e+02 4.160e+02 8.190e+02, threshold=6.485e+02, percent-clipped=2.0 2023-05-16 01:44:15,895 INFO [finetune.py:992] (1/2) Epoch 5, batch 1600, loss[loss=0.1595, simple_loss=0.2474, pruned_loss=0.03579, over 12359.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2634, pruned_loss=0.0435, over 2377834.05 frames. ], batch size: 30, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:44:17,560 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150537.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:44:44,649 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4038, 3.2392, 4.8317, 2.4197, 2.5706, 3.5434, 3.0546, 3.6160], device='cuda:1'), covar=tensor([0.0459, 0.1123, 0.0289, 0.1248, 0.2066, 0.1490, 0.1330, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0228, 0.0230, 0.0177, 0.0234, 0.0279, 0.0221, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:44:51,626 INFO [finetune.py:992] (1/2) Epoch 5, batch 1650, loss[loss=0.1951, simple_loss=0.2928, pruned_loss=0.04868, over 12352.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2644, pruned_loss=0.04429, over 2369647.78 frames. ], batch size: 38, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:44:51,822 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0923, 4.6508, 4.7059, 4.9770, 4.7879, 4.9399, 4.7958, 2.5440], device='cuda:1'), covar=tensor([0.0090, 0.0076, 0.0100, 0.0062, 0.0050, 0.0084, 0.0091, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0070, 0.0074, 0.0067, 0.0055, 0.0085, 0.0072, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:45:01,164 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150598.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:45:09,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5358, 2.3236, 3.2214, 4.2756, 2.1908, 4.3660, 4.4309, 4.4191], device='cuda:1'), covar=tensor([0.0100, 0.1302, 0.0461, 0.0162, 0.1332, 0.0221, 0.0150, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0201, 0.0184, 0.0115, 0.0188, 0.0174, 0.0166, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:45:24,691 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150630.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:45:25,337 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.928e+02 2.880e+02 3.247e+02 4.025e+02 9.318e+02, threshold=6.493e+02, percent-clipped=2.0 2023-05-16 01:45:28,907 INFO [finetune.py:992] (1/2) Epoch 5, batch 1700, loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04392, over 12352.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2645, pruned_loss=0.04429, over 2370378.84 frames. ], batch size: 36, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:45:42,572 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1076, 2.4287, 3.7555, 2.9681, 3.4204, 3.2154, 2.4207, 3.6011], device='cuda:1'), covar=tensor([0.0111, 0.0328, 0.0112, 0.0227, 0.0130, 0.0142, 0.0335, 0.0098], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0194, 0.0169, 0.0172, 0.0194, 0.0148, 0.0186, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:46:04,297 INFO [finetune.py:992] (1/2) Epoch 5, batch 1750, loss[loss=0.1964, simple_loss=0.2878, pruned_loss=0.05253, over 12199.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.265, pruned_loss=0.04466, over 2362318.12 frames. ], batch size: 35, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:46:31,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 01:46:36,882 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.084e+02 2.891e+02 3.374e+02 3.862e+02 1.303e+03, threshold=6.748e+02, percent-clipped=4.0 2023-05-16 01:46:37,638 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150732.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:46:39,493 INFO [finetune.py:992] (1/2) Epoch 5, batch 1800, loss[loss=0.1933, simple_loss=0.2769, pruned_loss=0.05481, over 11806.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2652, pruned_loss=0.04469, over 2372031.73 frames. ], batch size: 44, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:46:39,629 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150735.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:47:06,720 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7854, 2.7680, 4.6756, 4.9322, 3.0888, 2.7878, 3.1593, 2.2023], device='cuda:1'), covar=tensor([0.1383, 0.3243, 0.0422, 0.0342, 0.1103, 0.1977, 0.2355, 0.3808], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0369, 0.0261, 0.0281, 0.0248, 0.0278, 0.0348, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:47:13,416 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150783.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:47:15,455 INFO [finetune.py:992] (1/2) Epoch 5, batch 1850, loss[loss=0.1772, simple_loss=0.2667, pruned_loss=0.04382, over 12038.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2657, pruned_loss=0.04509, over 2367570.18 frames. ], batch size: 42, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:47:21,245 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150793.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 01:47:48,463 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.918e+02 3.629e+02 4.209e+02 8.227e+02, threshold=7.258e+02, percent-clipped=1.0 2023-05-16 01:47:51,368 INFO [finetune.py:992] (1/2) Epoch 5, batch 1900, loss[loss=0.1823, simple_loss=0.2776, pruned_loss=0.04354, over 11210.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2654, pruned_loss=0.04537, over 2367052.80 frames. ], batch size: 55, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:48:10,659 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4358, 4.2022, 4.2972, 4.6377, 3.2929, 4.1810, 2.8391, 4.2372], device='cuda:1'), covar=tensor([0.1473, 0.0614, 0.0747, 0.0519, 0.1013, 0.0525, 0.1555, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0256, 0.0286, 0.0342, 0.0232, 0.0233, 0.0251, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:48:26,695 INFO [finetune.py:992] (1/2) Epoch 5, batch 1950, loss[loss=0.168, simple_loss=0.2666, pruned_loss=0.03469, over 12262.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2644, pruned_loss=0.04455, over 2377678.22 frames. ], batch size: 37, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:48:32,451 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150893.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:48:37,578 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150900.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:48:49,688 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4011, 4.6165, 4.2643, 5.0225, 4.6583, 2.8922, 4.2151, 3.1551], device='cuda:1'), covar=tensor([0.0759, 0.0884, 0.1207, 0.0426, 0.1003, 0.1547, 0.0997, 0.3387], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0364, 0.0343, 0.0255, 0.0352, 0.0260, 0.0328, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:48:59,305 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150930.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:48:59,867 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.802e+02 3.415e+02 4.185e+02 7.760e+02, threshold=6.830e+02, percent-clipped=3.0 2023-05-16 01:49:03,424 INFO [finetune.py:992] (1/2) Epoch 5, batch 2000, loss[loss=0.1579, simple_loss=0.2472, pruned_loss=0.03429, over 12338.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2632, pruned_loss=0.04416, over 2377513.51 frames. ], batch size: 31, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:49:18,062 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7097, 2.6522, 3.3892, 4.5886, 2.4954, 4.5336, 4.5946, 4.7053], device='cuda:1'), covar=tensor([0.0100, 0.1219, 0.0461, 0.0098, 0.1290, 0.0175, 0.0148, 0.0093], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0202, 0.0184, 0.0114, 0.0189, 0.0175, 0.0167, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:49:22,314 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150961.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:49:34,360 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150978.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:49:38,119 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8041, 2.3713, 3.8580, 4.7880, 3.9884, 4.7524, 4.1576, 3.3319], device='cuda:1'), covar=tensor([0.0027, 0.0411, 0.0100, 0.0026, 0.0108, 0.0051, 0.0084, 0.0295], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0120, 0.0099, 0.0071, 0.0098, 0.0110, 0.0087, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:49:39,316 INFO [finetune.py:992] (1/2) Epoch 5, batch 2050, loss[loss=0.1753, simple_loss=0.2595, pruned_loss=0.04557, over 12096.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2622, pruned_loss=0.04377, over 2381216.86 frames. ], batch size: 32, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:50:08,144 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1452, 5.1874, 4.9443, 5.0628, 4.6208, 5.1342, 5.0446, 5.4054], device='cuda:1'), covar=tensor([0.0228, 0.0117, 0.0209, 0.0264, 0.0745, 0.0365, 0.0156, 0.0151], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0179, 0.0179, 0.0231, 0.0227, 0.0197, 0.0163, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1') 2023-05-16 01:50:12,132 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.900e+02 3.345e+02 3.884e+02 7.651e+02, threshold=6.691e+02, percent-clipped=2.0 2023-05-16 01:50:15,079 INFO [finetune.py:992] (1/2) Epoch 5, batch 2100, loss[loss=0.1796, simple_loss=0.2628, pruned_loss=0.04819, over 11263.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2627, pruned_loss=0.0436, over 2379794.38 frames. ], batch size: 55, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:50:51,737 INFO [finetune.py:992] (1/2) Epoch 5, batch 2150, loss[loss=0.1566, simple_loss=0.2408, pruned_loss=0.03619, over 12346.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2627, pruned_loss=0.04331, over 2378222.61 frames. ], batch size: 30, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:50:53,964 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151088.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 01:51:23,879 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.729e+02 3.385e+02 4.097e+02 8.908e+02, threshold=6.771e+02, percent-clipped=1.0 2023-05-16 01:51:25,390 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2097, 6.1535, 5.6275, 5.7350, 6.2335, 5.6912, 5.8258, 5.7629], device='cuda:1'), covar=tensor([0.1575, 0.0981, 0.1175, 0.1974, 0.0934, 0.1775, 0.1463, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0454, 0.0360, 0.0409, 0.0435, 0.0415, 0.0369, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:51:26,696 INFO [finetune.py:992] (1/2) Epoch 5, batch 2200, loss[loss=0.1819, simple_loss=0.2685, pruned_loss=0.04765, over 12108.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2627, pruned_loss=0.04368, over 2372188.25 frames. ], batch size: 33, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:51:48,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 01:52:01,702 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151184.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:52:02,202 INFO [finetune.py:992] (1/2) Epoch 5, batch 2250, loss[loss=0.1467, simple_loss=0.2318, pruned_loss=0.03081, over 11995.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2622, pruned_loss=0.04352, over 2374634.22 frames. ], batch size: 28, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:52:02,500 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6789, 2.8953, 4.4416, 4.8138, 2.8754, 2.7360, 2.8610, 2.0838], device='cuda:1'), covar=tensor([0.1388, 0.2823, 0.0517, 0.0332, 0.1116, 0.1919, 0.2490, 0.3743], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0372, 0.0262, 0.0285, 0.0251, 0.0280, 0.0350, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:52:08,062 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151193.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:52:36,364 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.934e+02 3.390e+02 3.969e+02 1.202e+03, threshold=6.780e+02, percent-clipped=1.0 2023-05-16 01:52:39,304 INFO [finetune.py:992] (1/2) Epoch 5, batch 2300, loss[loss=0.1723, simple_loss=0.2669, pruned_loss=0.03887, over 12073.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2628, pruned_loss=0.04404, over 2365780.10 frames. ], batch size: 42, lr: 4.72e-03, grad_scale: 16.0 2023-05-16 01:52:43,501 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151241.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:52:46,409 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151245.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:52:54,110 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151256.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:53:12,476 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:53:13,884 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3735, 4.9278, 5.2986, 4.5940, 4.9460, 4.6869, 5.2741, 4.9992], device='cuda:1'), covar=tensor([0.0238, 0.0317, 0.0296, 0.0246, 0.0303, 0.0288, 0.0246, 0.0275], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0236, 0.0258, 0.0231, 0.0231, 0.0228, 0.0209, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:53:14,401 INFO [finetune.py:992] (1/2) Epoch 5, batch 2350, loss[loss=0.1836, simple_loss=0.2725, pruned_loss=0.04739, over 12145.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2625, pruned_loss=0.04364, over 2374211.85 frames. ], batch size: 34, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:53:22,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 01:53:23,922 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151298.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:53:39,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 01:53:46,947 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.953e+02 3.417e+02 4.004e+02 7.292e+02, threshold=6.835e+02, percent-clipped=1.0 2023-05-16 01:53:49,775 INFO [finetune.py:992] (1/2) Epoch 5, batch 2400, loss[loss=0.1625, simple_loss=0.2466, pruned_loss=0.03918, over 12291.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2628, pruned_loss=0.0439, over 2365241.26 frames. ], batch size: 33, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:53:55,452 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151343.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:54:07,610 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151359.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:54:26,429 INFO [finetune.py:992] (1/2) Epoch 5, batch 2450, loss[loss=0.169, simple_loss=0.2519, pruned_loss=0.04301, over 12178.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2624, pruned_loss=0.0434, over 2365612.45 frames. ], batch size: 31, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:54:28,603 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151388.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:54:59,076 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.105e+02 2.885e+02 3.251e+02 3.670e+02 6.500e+02, threshold=6.501e+02, percent-clipped=0.0 2023-05-16 01:55:01,839 INFO [finetune.py:992] (1/2) Epoch 5, batch 2500, loss[loss=0.1784, simple_loss=0.2742, pruned_loss=0.04125, over 11841.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04373, over 2363056.83 frames. ], batch size: 44, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:55:02,621 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151436.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:55:19,966 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5250, 4.1388, 4.2604, 4.6603, 3.3466, 4.1494, 2.7545, 4.2168], device='cuda:1'), covar=tensor([0.1404, 0.0652, 0.0860, 0.0591, 0.0946, 0.0511, 0.1728, 0.1449], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0258, 0.0288, 0.0344, 0.0234, 0.0234, 0.0254, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:55:37,282 INFO [finetune.py:992] (1/2) Epoch 5, batch 2550, loss[loss=0.1799, simple_loss=0.2695, pruned_loss=0.04512, over 12352.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2626, pruned_loss=0.04343, over 2373576.72 frames. ], batch size: 36, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:55:55,497 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8743, 2.1451, 3.1684, 3.8262, 3.3794, 3.7469, 3.4520, 2.7553], device='cuda:1'), covar=tensor([0.0036, 0.0381, 0.0139, 0.0040, 0.0131, 0.0078, 0.0090, 0.0322], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0118, 0.0098, 0.0070, 0.0096, 0.0108, 0.0085, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:56:10,860 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 2.895e+02 3.302e+02 4.132e+02 5.921e+02, threshold=6.604e+02, percent-clipped=0.0 2023-05-16 01:56:13,780 INFO [finetune.py:992] (1/2) Epoch 5, batch 2600, loss[loss=0.1675, simple_loss=0.2527, pruned_loss=0.04118, over 12259.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2618, pruned_loss=0.04317, over 2371763.64 frames. ], batch size: 32, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:56:17,467 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151540.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:56:28,801 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151556.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:56:31,564 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1178, 5.9833, 5.6529, 5.5982, 6.0822, 5.3702, 5.6304, 5.6067], device='cuda:1'), covar=tensor([0.1364, 0.0772, 0.0709, 0.1534, 0.0899, 0.1938, 0.1354, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0456, 0.0358, 0.0412, 0.0437, 0.0414, 0.0373, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:56:49,302 INFO [finetune.py:992] (1/2) Epoch 5, batch 2650, loss[loss=0.1693, simple_loss=0.2573, pruned_loss=0.04069, over 12043.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2609, pruned_loss=0.04268, over 2381939.10 frames. ], batch size: 40, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:57:01,048 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1365, 6.0452, 5.6252, 5.6133, 6.1002, 5.5772, 5.5999, 5.6545], device='cuda:1'), covar=tensor([0.1566, 0.0926, 0.0912, 0.1839, 0.0970, 0.1952, 0.1556, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0458, 0.0360, 0.0415, 0.0439, 0.0416, 0.0374, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:57:03,149 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151604.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:57:21,906 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.196e+02 2.905e+02 3.336e+02 3.698e+02 6.432e+02, threshold=6.672e+02, percent-clipped=0.0 2023-05-16 01:57:24,932 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6230, 2.7462, 4.5550, 4.7545, 2.9037, 2.6476, 2.7909, 2.0921], device='cuda:1'), covar=tensor([0.1430, 0.2809, 0.0424, 0.0364, 0.1106, 0.2072, 0.2575, 0.3640], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0372, 0.0263, 0.0286, 0.0251, 0.0281, 0.0351, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:57:25,384 INFO [finetune.py:992] (1/2) Epoch 5, batch 2700, loss[loss=0.1915, simple_loss=0.281, pruned_loss=0.05094, over 12210.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2622, pruned_loss=0.04321, over 2387130.69 frames. ], batch size: 35, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:57:27,662 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151638.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:57:38,725 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151654.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:58:00,893 INFO [finetune.py:992] (1/2) Epoch 5, batch 2750, loss[loss=0.1769, simple_loss=0.2694, pruned_loss=0.04221, over 12190.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2628, pruned_loss=0.04359, over 2383766.57 frames. ], batch size: 35, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:58:03,321 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:58:33,497 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.923e+02 3.297e+02 3.991e+02 1.077e+03, threshold=6.595e+02, percent-clipped=4.0 2023-05-16 01:58:36,345 INFO [finetune.py:992] (1/2) Epoch 5, batch 2800, loss[loss=0.1626, simple_loss=0.255, pruned_loss=0.03508, over 12146.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04367, over 2376856.23 frames. ], batch size: 34, lr: 4.71e-03, grad_scale: 32.0 2023-05-16 01:58:46,295 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151749.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:59:12,507 INFO [finetune.py:992] (1/2) Epoch 5, batch 2850, loss[loss=0.159, simple_loss=0.2408, pruned_loss=0.03861, over 12289.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2629, pruned_loss=0.04387, over 2367118.81 frames. ], batch size: 28, lr: 4.71e-03, grad_scale: 32.0 2023-05-16 01:59:12,758 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7925, 2.8807, 4.7534, 5.0719, 3.3021, 2.7787, 3.0164, 2.3061], device='cuda:1'), covar=tensor([0.1436, 0.2897, 0.0385, 0.0326, 0.1018, 0.2128, 0.2515, 0.3826], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0369, 0.0261, 0.0284, 0.0249, 0.0280, 0.0349, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 01:59:20,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.77 vs. limit=5.0 2023-05-16 01:59:24,052 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151801.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:59:38,148 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2971, 6.1201, 5.7551, 5.7863, 6.2097, 5.5549, 5.8049, 5.7864], device='cuda:1'), covar=tensor([0.1522, 0.0832, 0.0890, 0.1799, 0.0908, 0.1994, 0.1250, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0457, 0.0362, 0.0414, 0.0440, 0.0416, 0.0372, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 01:59:43,209 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7723, 2.2669, 3.7092, 4.7315, 3.9714, 4.6460, 4.1069, 3.2116], device='cuda:1'), covar=tensor([0.0022, 0.0435, 0.0117, 0.0031, 0.0088, 0.0062, 0.0070, 0.0310], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0117, 0.0098, 0.0070, 0.0096, 0.0108, 0.0084, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 01:59:46,491 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 3.095e+02 3.461e+02 4.162e+02 1.203e+03, threshold=6.923e+02, percent-clipped=2.0 2023-05-16 01:59:48,471 INFO [finetune.py:992] (1/2) Epoch 5, batch 2900, loss[loss=0.1421, simple_loss=0.2251, pruned_loss=0.02956, over 11981.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2629, pruned_loss=0.04378, over 2369522.33 frames. ], batch size: 28, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 01:59:51,968 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151840.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 01:59:55,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 02:00:07,621 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151862.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:00:21,599 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0097, 5.9496, 5.7822, 5.3064, 5.1086, 5.9242, 5.4251, 5.2443], device='cuda:1'), covar=tensor([0.0807, 0.1130, 0.0643, 0.1626, 0.0654, 0.0637, 0.1635, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0593, 0.0518, 0.0498, 0.0606, 0.0393, 0.0675, 0.0741, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 02:00:23,588 INFO [finetune.py:992] (1/2) Epoch 5, batch 2950, loss[loss=0.1777, simple_loss=0.2703, pruned_loss=0.04258, over 11853.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2627, pruned_loss=0.04373, over 2375660.57 frames. ], batch size: 44, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:00:23,773 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4446, 2.2267, 3.5567, 4.3460, 3.8303, 4.2646, 4.0036, 2.8050], device='cuda:1'), covar=tensor([0.0028, 0.0419, 0.0124, 0.0035, 0.0104, 0.0072, 0.0079, 0.0366], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0117, 0.0098, 0.0070, 0.0096, 0.0109, 0.0085, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:00:25,809 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151888.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:00:35,171 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2360, 5.2014, 5.0193, 4.6946, 4.7255, 5.1771, 4.7727, 4.5487], device='cuda:1'), covar=tensor([0.0776, 0.1038, 0.0700, 0.1402, 0.0875, 0.0702, 0.1509, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0518, 0.0497, 0.0605, 0.0392, 0.0673, 0.0739, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 02:00:39,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6550, 4.8825, 4.3946, 5.2911, 4.7712, 3.0367, 4.3633, 3.2560], device='cuda:1'), covar=tensor([0.0558, 0.0659, 0.1134, 0.0241, 0.0889, 0.1380, 0.0883, 0.2837], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0363, 0.0342, 0.0257, 0.0349, 0.0258, 0.0326, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:00:57,498 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.810e+02 3.358e+02 4.108e+02 1.104e+03, threshold=6.716e+02, percent-clipped=4.0 2023-05-16 02:00:59,697 INFO [finetune.py:992] (1/2) Epoch 5, batch 3000, loss[loss=0.1488, simple_loss=0.2331, pruned_loss=0.03222, over 12022.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2623, pruned_loss=0.04339, over 2371321.86 frames. ], batch size: 28, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:00:59,697 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 02:01:17,519 INFO [finetune.py:1026] (1/2) Epoch 5, validation: loss=0.328, simple_loss=0.4027, pruned_loss=0.1267, over 1020973.00 frames. 2023-05-16 02:01:17,520 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 02:01:19,832 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151938.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:01:31,066 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151954.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:01:53,017 INFO [finetune.py:992] (1/2) Epoch 5, batch 3050, loss[loss=0.1966, simple_loss=0.2723, pruned_loss=0.06044, over 12140.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2617, pruned_loss=0.04326, over 2378520.47 frames. ], batch size: 39, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:01:53,791 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151986.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:02:00,256 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2716, 1.9948, 2.8550, 3.2335, 2.8807, 3.1563, 3.0125, 2.5280], device='cuda:1'), covar=tensor([0.0050, 0.0374, 0.0138, 0.0048, 0.0134, 0.0096, 0.0090, 0.0281], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0118, 0.0099, 0.0070, 0.0097, 0.0110, 0.0085, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:02:08,628 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152002.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:02:22,201 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2266, 4.4480, 2.6901, 2.4267, 3.8464, 2.3124, 3.8624, 3.0445], device='cuda:1'), covar=tensor([0.0660, 0.0600, 0.0966, 0.1459, 0.0285, 0.1346, 0.0467, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0244, 0.0171, 0.0195, 0.0136, 0.0177, 0.0189, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:02:22,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 02:02:30,578 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.140e+02 2.948e+02 3.429e+02 4.088e+02 7.361e+02, threshold=6.857e+02, percent-clipped=1.0 2023-05-16 02:02:32,721 INFO [finetune.py:992] (1/2) Epoch 5, batch 3100, loss[loss=0.1577, simple_loss=0.2477, pruned_loss=0.03385, over 12015.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2625, pruned_loss=0.04357, over 2373393.35 frames. ], batch size: 31, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:02:39,079 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152044.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:02:48,505 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5156, 2.2163, 3.2562, 4.3746, 2.1631, 4.3260, 4.4248, 4.6158], device='cuda:1'), covar=tensor([0.0134, 0.1232, 0.0432, 0.0122, 0.1230, 0.0213, 0.0119, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0197, 0.0179, 0.0112, 0.0184, 0.0172, 0.0164, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:03:08,397 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6347, 5.1514, 5.5954, 4.8884, 5.1546, 5.0028, 5.6219, 5.1917], device='cuda:1'), covar=tensor([0.0194, 0.0294, 0.0218, 0.0221, 0.0319, 0.0275, 0.0189, 0.0179], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0239, 0.0258, 0.0235, 0.0233, 0.0230, 0.0212, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:03:08,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 02:03:08,994 INFO [finetune.py:992] (1/2) Epoch 5, batch 3150, loss[loss=0.1444, simple_loss=0.2273, pruned_loss=0.0308, over 11458.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2617, pruned_loss=0.04313, over 2372604.27 frames. ], batch size: 25, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:03:09,268 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4731, 4.5941, 4.0362, 4.9472, 4.5796, 2.8636, 4.1747, 3.1263], device='cuda:1'), covar=tensor([0.0651, 0.0683, 0.1205, 0.0343, 0.1025, 0.1466, 0.0921, 0.2920], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0365, 0.0344, 0.0260, 0.0351, 0.0259, 0.0327, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:03:17,561 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3930, 5.1873, 5.3048, 5.3416, 4.9557, 5.0132, 4.7290, 5.3019], device='cuda:1'), covar=tensor([0.0658, 0.0626, 0.0796, 0.0615, 0.2036, 0.1393, 0.0581, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0634, 0.0548, 0.0588, 0.0781, 0.0700, 0.0516, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:03:42,596 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.785e+02 3.310e+02 3.978e+02 6.675e+02, threshold=6.621e+02, percent-clipped=0.0 2023-05-16 02:03:43,582 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2859, 4.6164, 2.7399, 2.2922, 4.0021, 2.1073, 4.0333, 3.0242], device='cuda:1'), covar=tensor([0.0616, 0.0535, 0.1043, 0.1667, 0.0282, 0.1557, 0.0399, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0243, 0.0170, 0.0194, 0.0136, 0.0176, 0.0188, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:03:44,787 INFO [finetune.py:992] (1/2) Epoch 5, batch 3200, loss[loss=0.176, simple_loss=0.2673, pruned_loss=0.04235, over 12253.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2614, pruned_loss=0.04298, over 2379114.41 frames. ], batch size: 32, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:04:00,405 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152157.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:04:20,687 INFO [finetune.py:992] (1/2) Epoch 5, batch 3250, loss[loss=0.1912, simple_loss=0.2934, pruned_loss=0.04446, over 12258.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2615, pruned_loss=0.04313, over 2382465.06 frames. ], batch size: 37, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:04:50,287 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152225.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:04:52,395 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3076, 5.1211, 5.2198, 5.2442, 4.8237, 4.9395, 4.7089, 5.2179], device='cuda:1'), covar=tensor([0.0618, 0.0561, 0.0725, 0.0599, 0.1934, 0.1153, 0.0579, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0632, 0.0550, 0.0590, 0.0783, 0.0697, 0.0516, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:04:55,199 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.801e+02 3.170e+02 3.939e+02 9.922e+02, threshold=6.341e+02, percent-clipped=3.0 2023-05-16 02:04:57,384 INFO [finetune.py:992] (1/2) Epoch 5, batch 3300, loss[loss=0.1934, simple_loss=0.2794, pruned_loss=0.05367, over 12067.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2618, pruned_loss=0.04335, over 2379598.46 frames. ], batch size: 42, lr: 4.71e-03, grad_scale: 16.0 2023-05-16 02:05:24,253 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152273.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:05:32,433 INFO [finetune.py:992] (1/2) Epoch 5, batch 3350, loss[loss=0.1744, simple_loss=0.2631, pruned_loss=0.04286, over 12160.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2626, pruned_loss=0.044, over 2371902.47 frames. ], batch size: 36, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:05:33,376 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152286.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:05:48,173 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4475, 5.2662, 5.3395, 5.3777, 5.0082, 5.0858, 4.8262, 5.3514], device='cuda:1'), covar=tensor([0.0608, 0.0545, 0.0743, 0.0582, 0.1937, 0.1180, 0.0546, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0635, 0.0556, 0.0594, 0.0790, 0.0704, 0.0520, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:05:54,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-16 02:06:06,646 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 2.793e+02 3.433e+02 4.351e+02 2.420e+03, threshold=6.867e+02, percent-clipped=7.0 2023-05-16 02:06:08,260 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152334.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:06:08,750 INFO [finetune.py:992] (1/2) Epoch 5, batch 3400, loss[loss=0.1564, simple_loss=0.2485, pruned_loss=0.03214, over 12337.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2618, pruned_loss=0.04345, over 2376657.92 frames. ], batch size: 36, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:06:14,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 02:06:15,284 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152344.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:06:35,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-05-16 02:06:45,193 INFO [finetune.py:992] (1/2) Epoch 5, batch 3450, loss[loss=0.1638, simple_loss=0.2581, pruned_loss=0.03471, over 12008.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2622, pruned_loss=0.04339, over 2375516.22 frames. ], batch size: 42, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:06:50,196 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152392.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:07:18,641 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.749e+02 3.256e+02 3.945e+02 8.450e+02, threshold=6.513e+02, percent-clipped=1.0 2023-05-16 02:07:20,735 INFO [finetune.py:992] (1/2) Epoch 5, batch 3500, loss[loss=0.1633, simple_loss=0.2551, pruned_loss=0.03578, over 12293.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2618, pruned_loss=0.04318, over 2382224.68 frames. ], batch size: 33, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:07:27,870 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152445.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:07:36,958 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152457.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:07:56,774 INFO [finetune.py:992] (1/2) Epoch 5, batch 3550, loss[loss=0.1791, simple_loss=0.2687, pruned_loss=0.04477, over 12013.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2622, pruned_loss=0.04329, over 2381337.35 frames. ], batch size: 40, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:08:11,086 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152505.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:08:11,984 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152506.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:08:31,206 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.698e+02 3.317e+02 3.795e+02 6.447e+02, threshold=6.634e+02, percent-clipped=0.0 2023-05-16 02:08:33,454 INFO [finetune.py:992] (1/2) Epoch 5, batch 3600, loss[loss=0.1824, simple_loss=0.2774, pruned_loss=0.04371, over 12204.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2619, pruned_loss=0.043, over 2380335.15 frames. ], batch size: 35, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:08:34,310 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1655, 4.6410, 5.0993, 4.3879, 4.7196, 4.5168, 5.1367, 4.8111], device='cuda:1'), covar=tensor([0.0250, 0.0405, 0.0293, 0.0277, 0.0388, 0.0303, 0.0249, 0.0318], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0242, 0.0260, 0.0235, 0.0232, 0.0232, 0.0212, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:09:06,147 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152581.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:09:06,874 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152582.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:09:08,893 INFO [finetune.py:992] (1/2) Epoch 5, batch 3650, loss[loss=0.1561, simple_loss=0.2436, pruned_loss=0.03431, over 12299.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2614, pruned_loss=0.04256, over 2377470.33 frames. ], batch size: 33, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:09:11,845 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152589.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:09:31,381 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3864, 4.1638, 4.1830, 4.4983, 3.1823, 3.9636, 2.9228, 4.1841], device='cuda:1'), covar=tensor([0.1477, 0.0631, 0.0842, 0.0681, 0.1073, 0.0575, 0.1578, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0261, 0.0292, 0.0349, 0.0237, 0.0236, 0.0255, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:09:41,293 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152629.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:09:43,340 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 2.862e+02 3.312e+02 3.947e+02 9.238e+02, threshold=6.625e+02, percent-clipped=2.0 2023-05-16 02:09:45,494 INFO [finetune.py:992] (1/2) Epoch 5, batch 3700, loss[loss=0.1495, simple_loss=0.2411, pruned_loss=0.02893, over 12259.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2612, pruned_loss=0.0425, over 2374702.19 frames. ], batch size: 32, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:09:51,393 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152643.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:09:56,384 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152650.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:10:14,904 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0342, 4.6804, 4.8115, 4.8405, 4.6789, 4.8481, 4.7861, 2.6593], device='cuda:1'), covar=tensor([0.0088, 0.0065, 0.0077, 0.0066, 0.0053, 0.0094, 0.0075, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0071, 0.0075, 0.0069, 0.0056, 0.0087, 0.0074, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:10:22,087 INFO [finetune.py:992] (1/2) Epoch 5, batch 3750, loss[loss=0.1735, simple_loss=0.2688, pruned_loss=0.03909, over 12335.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2611, pruned_loss=0.04234, over 2380494.19 frames. ], batch size: 36, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:10:51,690 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152726.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:10:55,756 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.124e+02 2.824e+02 3.202e+02 3.601e+02 7.259e+02, threshold=6.404e+02, percent-clipped=1.0 2023-05-16 02:10:57,908 INFO [finetune.py:992] (1/2) Epoch 5, batch 3800, loss[loss=0.1825, simple_loss=0.277, pruned_loss=0.04401, over 12266.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2608, pruned_loss=0.04246, over 2373402.82 frames. ], batch size: 37, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:11:33,961 INFO [finetune.py:992] (1/2) Epoch 5, batch 3850, loss[loss=0.1668, simple_loss=0.2629, pruned_loss=0.03535, over 12149.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2608, pruned_loss=0.04238, over 2368007.16 frames. ], batch size: 36, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:11:35,619 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152787.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:11:46,291 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152801.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:12:08,247 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.857e+02 3.300e+02 4.012e+02 8.037e+02, threshold=6.600e+02, percent-clipped=3.0 2023-05-16 02:12:10,400 INFO [finetune.py:992] (1/2) Epoch 5, batch 3900, loss[loss=0.183, simple_loss=0.2693, pruned_loss=0.04838, over 12162.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2617, pruned_loss=0.04282, over 2370523.88 frames. ], batch size: 36, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:12:40,703 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9497, 4.9072, 4.7349, 4.8604, 4.3974, 4.8529, 4.8050, 5.1638], device='cuda:1'), covar=tensor([0.0230, 0.0138, 0.0199, 0.0284, 0.0807, 0.0337, 0.0174, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0184, 0.0180, 0.0233, 0.0233, 0.0201, 0.0166, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 02:12:43,613 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152881.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:12:46,369 INFO [finetune.py:992] (1/2) Epoch 5, batch 3950, loss[loss=0.1909, simple_loss=0.2759, pruned_loss=0.05288, over 12380.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.262, pruned_loss=0.04288, over 2372180.66 frames. ], batch size: 38, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:13:18,184 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152929.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:13:18,236 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152929.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:13:20,243 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.843e+02 3.254e+02 3.920e+02 8.560e+02, threshold=6.509e+02, percent-clipped=1.0 2023-05-16 02:13:22,300 INFO [finetune.py:992] (1/2) Epoch 5, batch 4000, loss[loss=0.236, simple_loss=0.3172, pruned_loss=0.07741, over 8140.00 frames. ], tot_loss[loss=0.173, simple_loss=0.261, pruned_loss=0.04253, over 2367067.38 frames. ], batch size: 98, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:13:24,889 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152938.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:13:29,699 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152945.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:13:52,673 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152977.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:13:58,472 INFO [finetune.py:992] (1/2) Epoch 5, batch 4050, loss[loss=0.1452, simple_loss=0.2285, pruned_loss=0.03099, over 12358.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2613, pruned_loss=0.04287, over 2373538.18 frames. ], batch size: 30, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:14:08,422 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8792, 4.4865, 4.8333, 4.2500, 4.5161, 4.3307, 4.8711, 4.5476], device='cuda:1'), covar=tensor([0.0270, 0.0332, 0.0278, 0.0236, 0.0302, 0.0289, 0.0199, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0241, 0.0260, 0.0234, 0.0233, 0.0233, 0.0212, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:14:31,793 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.109e+02 3.060e+02 3.470e+02 4.175e+02 7.154e+02, threshold=6.940e+02, percent-clipped=2.0 2023-05-16 02:14:33,951 INFO [finetune.py:992] (1/2) Epoch 5, batch 4100, loss[loss=0.1717, simple_loss=0.2623, pruned_loss=0.04061, over 12141.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2612, pruned_loss=0.04288, over 2380358.77 frames. ], batch size: 39, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:14:44,389 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 02:14:47,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 02:14:49,143 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1088, 3.5630, 3.6679, 3.9750, 2.7691, 3.4553, 2.5035, 3.5028], device='cuda:1'), covar=tensor([0.1553, 0.0809, 0.0879, 0.0646, 0.1099, 0.0689, 0.1740, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0291, 0.0348, 0.0235, 0.0234, 0.0252, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:15:08,760 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:15:10,812 INFO [finetune.py:992] (1/2) Epoch 5, batch 4150, loss[loss=0.1637, simple_loss=0.2448, pruned_loss=0.0413, over 11863.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2615, pruned_loss=0.04303, over 2379501.16 frames. ], batch size: 26, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:15:22,195 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153101.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:15:33,552 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153117.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:15:35,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-05-16 02:15:44,065 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 3.037e+02 3.431e+02 4.052e+02 6.200e+02, threshold=6.862e+02, percent-clipped=0.0 2023-05-16 02:15:46,154 INFO [finetune.py:992] (1/2) Epoch 5, batch 4200, loss[loss=0.1853, simple_loss=0.2765, pruned_loss=0.04707, over 11188.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2614, pruned_loss=0.04329, over 2378962.79 frames. ], batch size: 55, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:15:49,143 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4851, 4.9802, 5.4203, 4.7279, 5.0929, 4.9098, 5.4521, 5.0367], device='cuda:1'), covar=tensor([0.0216, 0.0310, 0.0242, 0.0241, 0.0309, 0.0260, 0.0205, 0.0319], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0240, 0.0259, 0.0234, 0.0231, 0.0232, 0.0212, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:15:56,132 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153149.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:16:17,143 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153178.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:16:21,884 INFO [finetune.py:992] (1/2) Epoch 5, batch 4250, loss[loss=0.1711, simple_loss=0.2602, pruned_loss=0.04101, over 12079.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2629, pruned_loss=0.04394, over 2372792.11 frames. ], batch size: 32, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:16:42,012 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3581, 4.0291, 4.0296, 4.3156, 2.9837, 4.0926, 2.7046, 4.2079], device='cuda:1'), covar=tensor([0.1425, 0.0655, 0.1032, 0.0685, 0.1073, 0.0495, 0.1663, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0291, 0.0348, 0.0236, 0.0233, 0.0252, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:16:56,564 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.049e+02 2.788e+02 3.231e+02 3.915e+02 1.320e+03, threshold=6.462e+02, percent-clipped=1.0 2023-05-16 02:16:58,688 INFO [finetune.py:992] (1/2) Epoch 5, batch 4300, loss[loss=0.1489, simple_loss=0.2364, pruned_loss=0.03065, over 12351.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2632, pruned_loss=0.0439, over 2380694.49 frames. ], batch size: 31, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:17:00,990 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153238.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:17:05,996 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153245.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:17:29,168 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153278.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:17:33,850 INFO [finetune.py:992] (1/2) Epoch 5, batch 4350, loss[loss=0.1605, simple_loss=0.2385, pruned_loss=0.04127, over 12347.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.263, pruned_loss=0.044, over 2375649.35 frames. ], batch size: 30, lr: 4.70e-03, grad_scale: 16.0 2023-05-16 02:17:34,677 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153286.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:17:39,582 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153293.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:17:46,849 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7904, 5.4105, 5.0244, 4.9937, 5.5077, 4.9069, 5.0732, 5.0198], device='cuda:1'), covar=tensor([0.1218, 0.1000, 0.1138, 0.1876, 0.1161, 0.2120, 0.1632, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0462, 0.0367, 0.0411, 0.0444, 0.0419, 0.0376, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:18:07,776 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.085e+02 2.827e+02 3.260e+02 4.214e+02 1.164e+03, threshold=6.520e+02, percent-clipped=1.0 2023-05-16 02:18:09,895 INFO [finetune.py:992] (1/2) Epoch 5, batch 4400, loss[loss=0.1744, simple_loss=0.2676, pruned_loss=0.04059, over 12318.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.263, pruned_loss=0.04389, over 2376493.92 frames. ], batch size: 34, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:18:13,003 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153339.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:18:44,189 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153382.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:18:46,221 INFO [finetune.py:992] (1/2) Epoch 5, batch 4450, loss[loss=0.2344, simple_loss=0.3017, pruned_loss=0.08357, over 7578.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2631, pruned_loss=0.04388, over 2369046.01 frames. ], batch size: 98, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:19:05,847 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6584, 2.7579, 4.6302, 4.8174, 2.8086, 2.5762, 2.8226, 2.1125], device='cuda:1'), covar=tensor([0.1461, 0.3084, 0.0442, 0.0364, 0.1250, 0.2132, 0.2632, 0.3592], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0371, 0.0264, 0.0286, 0.0252, 0.0280, 0.0353, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:19:12,153 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153421.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:19:18,362 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153430.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:19:19,753 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 2.916e+02 3.318e+02 4.100e+02 6.864e+02, threshold=6.635e+02, percent-clipped=2.0 2023-05-16 02:19:21,852 INFO [finetune.py:992] (1/2) Epoch 5, batch 4500, loss[loss=0.1837, simple_loss=0.2675, pruned_loss=0.04998, over 12104.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2623, pruned_loss=0.0436, over 2372121.84 frames. ], batch size: 42, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:19:49,039 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153473.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:19:56,146 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153482.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:19:58,023 INFO [finetune.py:992] (1/2) Epoch 5, batch 4550, loss[loss=0.1716, simple_loss=0.269, pruned_loss=0.03714, over 12272.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2622, pruned_loss=0.04345, over 2365951.74 frames. ], batch size: 37, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:20:31,538 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.058e+02 2.893e+02 3.389e+02 4.281e+02 1.179e+03, threshold=6.777e+02, percent-clipped=3.0 2023-05-16 02:20:33,611 INFO [finetune.py:992] (1/2) Epoch 5, batch 4600, loss[loss=0.1889, simple_loss=0.2786, pruned_loss=0.04956, over 11173.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2634, pruned_loss=0.04416, over 2355369.55 frames. ], batch size: 55, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:20:35,227 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8376, 3.3028, 5.1761, 2.7383, 2.9334, 3.8102, 3.3153, 3.9089], device='cuda:1'), covar=tensor([0.0415, 0.1135, 0.0234, 0.1126, 0.1774, 0.1456, 0.1292, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0225, 0.0228, 0.0178, 0.0233, 0.0277, 0.0221, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:21:08,776 INFO [finetune.py:992] (1/2) Epoch 5, batch 4650, loss[loss=0.2053, simple_loss=0.2928, pruned_loss=0.05886, over 12157.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2643, pruned_loss=0.04455, over 2340729.68 frames. ], batch size: 39, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:21:15,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 02:21:16,862 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2890, 5.1580, 5.2315, 5.2706, 4.8944, 4.9527, 4.7954, 5.2674], device='cuda:1'), covar=tensor([0.0640, 0.0549, 0.0712, 0.0623, 0.1809, 0.1270, 0.0529, 0.0768], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0632, 0.0549, 0.0588, 0.0783, 0.0698, 0.0514, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:21:27,605 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4530, 2.1783, 3.1958, 4.3243, 2.0997, 4.3453, 4.4043, 4.4833], device='cuda:1'), covar=tensor([0.0105, 0.1227, 0.0445, 0.0103, 0.1276, 0.0207, 0.0154, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0200, 0.0182, 0.0115, 0.0186, 0.0175, 0.0167, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:21:29,051 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6881, 2.5876, 3.9376, 4.1244, 2.9483, 2.6547, 2.7037, 2.1520], device='cuda:1'), covar=tensor([0.1370, 0.2889, 0.0575, 0.0477, 0.1076, 0.1920, 0.2464, 0.3682], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0370, 0.0264, 0.0286, 0.0251, 0.0279, 0.0351, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:21:43,118 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.784e+02 3.285e+02 3.986e+02 8.403e+02, threshold=6.570e+02, percent-clipped=1.0 2023-05-16 02:21:44,629 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153634.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:21:45,316 INFO [finetune.py:992] (1/2) Epoch 5, batch 4700, loss[loss=0.1502, simple_loss=0.2299, pruned_loss=0.03521, over 12282.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2642, pruned_loss=0.04442, over 2341669.23 frames. ], batch size: 28, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:21:46,931 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0905, 6.0558, 5.8652, 5.4947, 5.2727, 5.9896, 5.6501, 5.3871], device='cuda:1'), covar=tensor([0.0657, 0.0797, 0.0587, 0.1286, 0.0603, 0.0652, 0.1204, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0517, 0.0492, 0.0603, 0.0392, 0.0675, 0.0739, 0.0543], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 02:22:21,128 INFO [finetune.py:992] (1/2) Epoch 5, batch 4750, loss[loss=0.1736, simple_loss=0.2632, pruned_loss=0.04198, over 11572.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2642, pruned_loss=0.04454, over 2349917.83 frames. ], batch size: 48, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:22:37,851 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3446, 4.1724, 4.3280, 4.4917, 2.9405, 4.0614, 2.8949, 4.1823], device='cuda:1'), covar=tensor([0.1592, 0.0636, 0.0758, 0.0583, 0.1160, 0.0575, 0.1627, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0262, 0.0293, 0.0350, 0.0238, 0.0236, 0.0256, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:22:54,596 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 2.771e+02 3.141e+02 3.707e+02 1.127e+03, threshold=6.281e+02, percent-clipped=2.0 2023-05-16 02:22:56,734 INFO [finetune.py:992] (1/2) Epoch 5, batch 4800, loss[loss=0.184, simple_loss=0.2779, pruned_loss=0.04498, over 12287.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.264, pruned_loss=0.04424, over 2360887.77 frames. ], batch size: 37, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:23:24,815 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153773.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:23:27,539 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153777.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:23:32,995 INFO [finetune.py:992] (1/2) Epoch 5, batch 4850, loss[loss=0.155, simple_loss=0.2466, pruned_loss=0.03168, over 12098.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.262, pruned_loss=0.0434, over 2372439.66 frames. ], batch size: 32, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:23:47,961 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 02:23:59,787 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153821.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:24:07,336 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.827e+02 3.174e+02 3.763e+02 6.877e+02, threshold=6.348e+02, percent-clipped=0.0 2023-05-16 02:24:09,572 INFO [finetune.py:992] (1/2) Epoch 5, batch 4900, loss[loss=0.1485, simple_loss=0.2279, pruned_loss=0.03449, over 11776.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2627, pruned_loss=0.04357, over 2372139.37 frames. ], batch size: 26, lr: 4.69e-03, grad_scale: 32.0 2023-05-16 02:24:16,613 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.93 vs. limit=5.0 2023-05-16 02:24:41,475 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1058, 5.9105, 5.5985, 5.5192, 6.0434, 5.3507, 5.6599, 5.6027], device='cuda:1'), covar=tensor([0.1368, 0.0970, 0.0920, 0.2028, 0.1028, 0.1916, 0.1560, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0471, 0.0374, 0.0419, 0.0448, 0.0425, 0.0382, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:24:44,826 INFO [finetune.py:992] (1/2) Epoch 5, batch 4950, loss[loss=0.1524, simple_loss=0.2407, pruned_loss=0.03205, over 12336.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2625, pruned_loss=0.04379, over 2369903.55 frames. ], batch size: 30, lr: 4.69e-03, grad_scale: 32.0 2023-05-16 02:24:50,662 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6413, 4.8695, 4.5149, 5.1749, 4.8209, 3.2906, 4.5954, 3.6045], device='cuda:1'), covar=tensor([0.0713, 0.0712, 0.1187, 0.0403, 0.0925, 0.1366, 0.0904, 0.2310], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0366, 0.0345, 0.0264, 0.0351, 0.0262, 0.0330, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:24:52,903 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 02:25:06,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 02:25:12,426 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153923.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:25:18,476 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.987e+02 3.492e+02 4.100e+02 7.921e+02, threshold=6.983e+02, percent-clipped=7.0 2023-05-16 02:25:19,987 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153934.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:25:20,616 INFO [finetune.py:992] (1/2) Epoch 5, batch 5000, loss[loss=0.1495, simple_loss=0.2288, pruned_loss=0.03511, over 12271.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2624, pruned_loss=0.04391, over 2368271.69 frames. ], batch size: 28, lr: 4.69e-03, grad_scale: 32.0 2023-05-16 02:25:54,915 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153982.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:25:56,525 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153984.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:25:57,020 INFO [finetune.py:992] (1/2) Epoch 5, batch 5050, loss[loss=0.1503, simple_loss=0.2376, pruned_loss=0.0315, over 12113.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2622, pruned_loss=0.04339, over 2376333.23 frames. ], batch size: 30, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:25:59,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 02:26:34,373 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.141e+02 2.897e+02 3.384e+02 4.051e+02 8.052e+02, threshold=6.767e+02, percent-clipped=3.0 2023-05-16 02:26:35,821 INFO [finetune.py:992] (1/2) Epoch 5, batch 5100, loss[loss=0.1848, simple_loss=0.274, pruned_loss=0.04779, over 11789.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2628, pruned_loss=0.04345, over 2371206.54 frames. ], batch size: 44, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:26:46,779 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154049.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:26:58,323 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3053, 4.7612, 2.7999, 2.8519, 4.0127, 2.5958, 4.0660, 3.3212], device='cuda:1'), covar=tensor([0.0658, 0.0518, 0.1066, 0.1398, 0.0335, 0.1332, 0.0452, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0246, 0.0170, 0.0194, 0.0137, 0.0176, 0.0189, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:27:04,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 02:27:05,954 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154076.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:27:06,604 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154077.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:27:11,554 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154084.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:27:12,096 INFO [finetune.py:992] (1/2) Epoch 5, batch 5150, loss[loss=0.1654, simple_loss=0.2476, pruned_loss=0.04167, over 12236.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2619, pruned_loss=0.04287, over 2374234.97 frames. ], batch size: 32, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:27:24,280 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4875, 4.8493, 3.0370, 3.2166, 4.1530, 2.9158, 4.1740, 3.5988], device='cuda:1'), covar=tensor([0.0668, 0.0661, 0.0973, 0.1090, 0.0257, 0.1121, 0.0427, 0.0682], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0245, 0.0170, 0.0193, 0.0137, 0.0176, 0.0189, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:27:30,764 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154110.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:27:41,116 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154125.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:27:46,558 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.960e+02 3.349e+02 3.965e+02 6.910e+02, threshold=6.698e+02, percent-clipped=1.0 2023-05-16 02:27:48,079 INFO [finetune.py:992] (1/2) Epoch 5, batch 5200, loss[loss=0.1452, simple_loss=0.2281, pruned_loss=0.03114, over 12349.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2616, pruned_loss=0.04286, over 2376169.90 frames. ], batch size: 30, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:27:49,698 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154137.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:27:50,347 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8512, 2.1953, 3.4826, 2.7978, 3.3354, 2.9024, 2.1511, 3.4139], device='cuda:1'), covar=tensor([0.0130, 0.0413, 0.0143, 0.0270, 0.0165, 0.0188, 0.0404, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0194, 0.0174, 0.0174, 0.0196, 0.0151, 0.0187, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:27:55,415 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154145.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:28:23,640 INFO [finetune.py:992] (1/2) Epoch 5, batch 5250, loss[loss=0.1655, simple_loss=0.2538, pruned_loss=0.03856, over 12343.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2616, pruned_loss=0.04307, over 2383135.89 frames. ], batch size: 31, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:28:50,144 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154221.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:28:55,819 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9032, 2.3968, 3.2876, 2.7887, 3.1922, 2.9846, 2.3488, 3.2899], device='cuda:1'), covar=tensor([0.0101, 0.0314, 0.0185, 0.0214, 0.0148, 0.0154, 0.0285, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0195, 0.0174, 0.0173, 0.0197, 0.0151, 0.0187, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:28:57,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.11 vs. limit=5.0 2023-05-16 02:28:58,488 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.850e+02 3.342e+02 3.874e+02 6.540e+02, threshold=6.683e+02, percent-clipped=0.0 2023-05-16 02:28:59,932 INFO [finetune.py:992] (1/2) Epoch 5, batch 5300, loss[loss=0.1855, simple_loss=0.2725, pruned_loss=0.04923, over 12055.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2611, pruned_loss=0.04287, over 2389443.49 frames. ], batch size: 37, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:29:31,832 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154279.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:29:34,062 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154282.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 02:29:35,960 INFO [finetune.py:992] (1/2) Epoch 5, batch 5350, loss[loss=0.1886, simple_loss=0.279, pruned_loss=0.04911, over 12110.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2611, pruned_loss=0.0431, over 2391615.69 frames. ], batch size: 38, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:29:36,896 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154286.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:29:40,351 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154291.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:30:03,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.31 vs. limit=5.0 2023-05-16 02:30:09,560 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.896e+02 3.518e+02 4.242e+02 8.242e+02, threshold=7.036e+02, percent-clipped=1.0 2023-05-16 02:30:11,026 INFO [finetune.py:992] (1/2) Epoch 5, batch 5400, loss[loss=0.1533, simple_loss=0.2371, pruned_loss=0.03478, over 12260.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2625, pruned_loss=0.0437, over 2384344.16 frames. ], batch size: 32, lr: 4.69e-03, grad_scale: 16.0 2023-05-16 02:30:12,617 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154337.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:30:20,316 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154347.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:30:23,879 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154352.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:30:37,609 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0111, 2.5064, 3.5366, 2.9952, 3.3541, 3.1765, 2.4946, 3.4867], device='cuda:1'), covar=tensor([0.0118, 0.0279, 0.0144, 0.0200, 0.0127, 0.0140, 0.0317, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0194, 0.0172, 0.0171, 0.0195, 0.0149, 0.0185, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:30:47,158 INFO [finetune.py:992] (1/2) Epoch 5, batch 5450, loss[loss=0.1759, simple_loss=0.271, pruned_loss=0.04037, over 12124.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2619, pruned_loss=0.04336, over 2386169.62 frames. ], batch size: 38, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:30:57,335 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154398.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:31:02,463 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154405.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:31:08,185 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3183, 5.0655, 5.2366, 5.2648, 4.7975, 4.8714, 4.6903, 5.2108], device='cuda:1'), covar=tensor([0.0624, 0.0692, 0.0717, 0.0626, 0.2234, 0.1350, 0.0609, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0642, 0.0556, 0.0595, 0.0796, 0.0706, 0.0519, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:31:18,215 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2981, 2.7354, 3.9342, 3.3155, 3.7522, 3.4742, 2.7953, 3.8626], device='cuda:1'), covar=tensor([0.0113, 0.0285, 0.0141, 0.0192, 0.0125, 0.0139, 0.0305, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0193, 0.0172, 0.0171, 0.0194, 0.0149, 0.0185, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:31:21,683 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154432.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:31:22,248 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.125e+02 2.884e+02 3.458e+02 3.962e+02 7.721e+02, threshold=6.916e+02, percent-clipped=3.0 2023-05-16 02:31:23,713 INFO [finetune.py:992] (1/2) Epoch 5, batch 5500, loss[loss=0.1755, simple_loss=0.2638, pruned_loss=0.04363, over 12108.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2623, pruned_loss=0.04355, over 2375654.47 frames. ], batch size: 38, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:31:27,385 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154440.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:31:51,523 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154474.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:31:59,104 INFO [finetune.py:992] (1/2) Epoch 5, batch 5550, loss[loss=0.1713, simple_loss=0.2673, pruned_loss=0.03764, over 12152.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2637, pruned_loss=0.04393, over 2373092.02 frames. ], batch size: 34, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:32:12,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 02:32:24,456 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3599, 4.9235, 5.3144, 4.6895, 4.9646, 4.7931, 5.3345, 4.9062], device='cuda:1'), covar=tensor([0.0210, 0.0293, 0.0222, 0.0212, 0.0303, 0.0251, 0.0218, 0.0284], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0241, 0.0261, 0.0235, 0.0235, 0.0235, 0.0212, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:32:27,600 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-05-16 02:32:34,220 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.833e+02 3.428e+02 4.181e+02 7.252e+02, threshold=6.856e+02, percent-clipped=1.0 2023-05-16 02:32:35,626 INFO [finetune.py:992] (1/2) Epoch 5, batch 5600, loss[loss=0.1614, simple_loss=0.2498, pruned_loss=0.03646, over 12081.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2628, pruned_loss=0.0435, over 2374203.86 frames. ], batch size: 32, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:32:35,846 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154535.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:33:05,518 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154577.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:33:07,047 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154579.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:33:11,134 INFO [finetune.py:992] (1/2) Epoch 5, batch 5650, loss[loss=0.1726, simple_loss=0.2654, pruned_loss=0.03985, over 12337.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2622, pruned_loss=0.04319, over 2373795.35 frames. ], batch size: 36, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:33:40,781 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154627.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:33:41,597 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154628.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:33:45,683 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.849e+02 3.278e+02 3.834e+02 6.939e+02, threshold=6.556e+02, percent-clipped=1.0 2023-05-16 02:33:47,071 INFO [finetune.py:992] (1/2) Epoch 5, batch 5700, loss[loss=0.195, simple_loss=0.2853, pruned_loss=0.05241, over 12279.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.263, pruned_loss=0.04383, over 2380611.28 frames. ], batch size: 37, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:33:51,998 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154642.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:33:55,598 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154647.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:34:04,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 02:34:23,037 INFO [finetune.py:992] (1/2) Epoch 5, batch 5750, loss[loss=0.1905, simple_loss=0.2862, pruned_loss=0.04742, over 10622.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2626, pruned_loss=0.044, over 2373213.29 frames. ], batch size: 68, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:34:26,113 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154689.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:34:28,937 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154693.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:34:37,240 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154705.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:34:48,101 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154720.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:34:56,559 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154732.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:34:57,131 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.814e+02 3.331e+02 3.980e+02 7.563e+02, threshold=6.661e+02, percent-clipped=4.0 2023-05-16 02:34:58,577 INFO [finetune.py:992] (1/2) Epoch 5, batch 5800, loss[loss=0.1705, simple_loss=0.2593, pruned_loss=0.04086, over 12302.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2628, pruned_loss=0.04393, over 2373765.75 frames. ], batch size: 34, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:35:02,449 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154740.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:35:11,511 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154753.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:35:11,695 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7499, 2.5852, 4.6535, 5.0325, 3.2787, 2.6121, 2.8797, 1.9321], device='cuda:1'), covar=tensor([0.1434, 0.3521, 0.0419, 0.0263, 0.0923, 0.2094, 0.2745, 0.4729], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0369, 0.0262, 0.0282, 0.0251, 0.0278, 0.0350, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:35:19,615 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2059, 5.1031, 5.1998, 5.2484, 4.8673, 4.8669, 4.7575, 5.2328], device='cuda:1'), covar=tensor([0.0730, 0.0560, 0.0673, 0.0557, 0.1863, 0.1270, 0.0489, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0639, 0.0553, 0.0589, 0.0785, 0.0703, 0.0517, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:35:24,867 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-16 02:35:29,759 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3666, 4.5481, 4.0463, 5.0465, 4.7471, 3.0909, 4.3754, 3.1686], device='cuda:1'), covar=tensor([0.0713, 0.0861, 0.1344, 0.0358, 0.0810, 0.1431, 0.0880, 0.2900], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0365, 0.0345, 0.0265, 0.0351, 0.0262, 0.0328, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:35:31,691 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154780.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:35:32,500 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154781.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:35:35,160 INFO [finetune.py:992] (1/2) Epoch 5, batch 5850, loss[loss=0.1554, simple_loss=0.2462, pruned_loss=0.03224, over 12105.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2626, pruned_loss=0.0439, over 2368150.26 frames. ], batch size: 33, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:35:37,412 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154788.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:35:59,375 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4045, 4.8164, 3.0119, 2.6548, 4.1163, 2.6418, 4.0645, 3.4399], device='cuda:1'), covar=tensor([0.0651, 0.0563, 0.0902, 0.1379, 0.0231, 0.1230, 0.0412, 0.0699], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0248, 0.0174, 0.0196, 0.0139, 0.0180, 0.0192, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:36:01,058 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 02:36:08,423 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154830.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:36:10,427 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 2.903e+02 3.495e+02 4.253e+02 2.305e+03, threshold=6.990e+02, percent-clipped=5.0 2023-05-16 02:36:11,791 INFO [finetune.py:992] (1/2) Epoch 5, batch 5900, loss[loss=0.1909, simple_loss=0.2858, pruned_loss=0.04798, over 12084.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2624, pruned_loss=0.04379, over 2372585.29 frames. ], batch size: 42, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:36:41,922 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154877.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 02:36:47,283 INFO [finetune.py:992] (1/2) Epoch 5, batch 5950, loss[loss=0.1626, simple_loss=0.2558, pruned_loss=0.03468, over 12038.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2621, pruned_loss=0.04361, over 2379883.06 frames. ], batch size: 40, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:37:16,055 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154925.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:37:21,774 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.820e+02 3.339e+02 4.158e+02 6.543e+02, threshold=6.678e+02, percent-clipped=0.0 2023-05-16 02:37:23,224 INFO [finetune.py:992] (1/2) Epoch 5, batch 6000, loss[loss=0.1703, simple_loss=0.2611, pruned_loss=0.03972, over 12039.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2625, pruned_loss=0.04342, over 2383370.69 frames. ], batch size: 40, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:37:23,224 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 02:37:35,369 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8229, 5.7957, 5.7252, 5.0875, 5.1224, 5.7403, 5.2046, 5.2107], device='cuda:1'), covar=tensor([0.0522, 0.0770, 0.0453, 0.1445, 0.0436, 0.0611, 0.1293, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0513, 0.0488, 0.0603, 0.0393, 0.0671, 0.0734, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 02:37:41,095 INFO [finetune.py:1026] (1/2) Epoch 5, validation: loss=0.3218, simple_loss=0.3996, pruned_loss=0.122, over 1020973.00 frames. 2023-05-16 02:37:41,096 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 02:37:46,195 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154942.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:37:49,585 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154947.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:38:09,978 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154976.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:38:15,639 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154984.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:38:16,236 INFO [finetune.py:992] (1/2) Epoch 5, batch 6050, loss[loss=0.1786, simple_loss=0.2625, pruned_loss=0.04734, over 12256.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2623, pruned_loss=0.04354, over 2376539.65 frames. ], batch size: 32, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:38:19,793 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:38:22,020 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154993.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:38:23,336 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154995.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:38:51,370 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.886e+02 3.414e+02 4.219e+02 1.110e+03, threshold=6.829e+02, percent-clipped=1.0 2023-05-16 02:38:52,864 INFO [finetune.py:992] (1/2) Epoch 5, batch 6100, loss[loss=0.1517, simple_loss=0.2382, pruned_loss=0.03262, over 12280.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2621, pruned_loss=0.04344, over 2382940.79 frames. ], batch size: 28, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:38:54,483 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155037.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:38:57,120 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155041.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:38:57,892 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6481, 3.6323, 3.5013, 3.2592, 3.0587, 2.7756, 3.7640, 2.3000], device='cuda:1'), covar=tensor([0.0291, 0.0127, 0.0103, 0.0148, 0.0277, 0.0296, 0.0075, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0157, 0.0149, 0.0180, 0.0198, 0.0193, 0.0157, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:39:10,660 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0161, 3.4400, 5.3461, 2.7581, 2.8092, 4.0203, 3.2573, 4.0395], device='cuda:1'), covar=tensor([0.0376, 0.1045, 0.0251, 0.1207, 0.1767, 0.1225, 0.1365, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0229, 0.0233, 0.0181, 0.0236, 0.0280, 0.0226, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:39:17,515 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155069.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:39:22,590 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155076.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:39:29,023 INFO [finetune.py:992] (1/2) Epoch 5, batch 6150, loss[loss=0.1536, simple_loss=0.2399, pruned_loss=0.03365, over 12128.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2613, pruned_loss=0.04307, over 2389867.50 frames. ], batch size: 30, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:39:53,437 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7791, 2.8752, 4.6439, 4.9282, 2.9112, 2.7185, 2.9629, 2.0762], device='cuda:1'), covar=tensor([0.1342, 0.2951, 0.0426, 0.0302, 0.1113, 0.2100, 0.2451, 0.3888], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0372, 0.0263, 0.0284, 0.0253, 0.0282, 0.0352, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:40:01,096 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155130.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:40:01,220 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155130.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:40:03,124 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.118e+02 2.887e+02 3.492e+02 4.080e+02 6.055e+02, threshold=6.984e+02, percent-clipped=0.0 2023-05-16 02:40:04,463 INFO [finetune.py:992] (1/2) Epoch 5, batch 6200, loss[loss=0.3012, simple_loss=0.3538, pruned_loss=0.1244, over 8231.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.262, pruned_loss=0.04355, over 2382365.00 frames. ], batch size: 99, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:40:35,838 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155178.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:40:36,618 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9353, 3.9171, 3.8595, 4.0184, 3.7718, 3.7727, 3.6965, 3.9635], device='cuda:1'), covar=tensor([0.1039, 0.0706, 0.1420, 0.0755, 0.1784, 0.1437, 0.0610, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0635, 0.0549, 0.0587, 0.0784, 0.0703, 0.0513, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:40:40,794 INFO [finetune.py:992] (1/2) Epoch 5, batch 6250, loss[loss=0.1628, simple_loss=0.2599, pruned_loss=0.03281, over 12127.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2614, pruned_loss=0.04329, over 2374182.05 frames. ], batch size: 30, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:41:15,324 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.915e+02 3.294e+02 3.991e+02 1.581e+03, threshold=6.587e+02, percent-clipped=1.0 2023-05-16 02:41:16,784 INFO [finetune.py:992] (1/2) Epoch 5, batch 6300, loss[loss=0.1682, simple_loss=0.2568, pruned_loss=0.03976, over 12106.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2625, pruned_loss=0.04371, over 2374376.51 frames. ], batch size: 32, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:41:51,797 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155284.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:41:52,393 INFO [finetune.py:992] (1/2) Epoch 5, batch 6350, loss[loss=0.1776, simple_loss=0.2709, pruned_loss=0.04217, over 12363.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2623, pruned_loss=0.04368, over 2370500.37 frames. ], batch size: 36, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:41:55,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 02:42:25,965 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155332.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:42:25,982 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155332.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:42:26,607 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 2.968e+02 3.532e+02 4.284e+02 2.270e+03, threshold=7.063e+02, percent-clipped=5.0 2023-05-16 02:42:28,037 INFO [finetune.py:992] (1/2) Epoch 5, batch 6400, loss[loss=0.1656, simple_loss=0.2485, pruned_loss=0.0414, over 12310.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04376, over 2371817.93 frames. ], batch size: 33, lr: 4.68e-03, grad_scale: 16.0 2023-05-16 02:42:39,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 02:42:57,622 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155376.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:43:03,960 INFO [finetune.py:992] (1/2) Epoch 5, batch 6450, loss[loss=0.1809, simple_loss=0.2754, pruned_loss=0.0432, over 12137.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2629, pruned_loss=0.04378, over 2372403.61 frames. ], batch size: 36, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:43:31,763 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155424.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:43:32,540 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155425.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:43:38,064 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.153e+02 3.057e+02 3.788e+02 4.404e+02 2.061e+03, threshold=7.575e+02, percent-clipped=3.0 2023-05-16 02:43:39,500 INFO [finetune.py:992] (1/2) Epoch 5, batch 6500, loss[loss=0.2016, simple_loss=0.2919, pruned_loss=0.05565, over 10470.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2637, pruned_loss=0.04405, over 2373333.13 frames. ], batch size: 68, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:43:55,964 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155457.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:43:56,528 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2369, 6.1579, 5.9688, 5.5200, 5.3613, 6.1476, 5.7408, 5.5247], device='cuda:1'), covar=tensor([0.0557, 0.0802, 0.0525, 0.1382, 0.0459, 0.0639, 0.1234, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0504, 0.0481, 0.0593, 0.0389, 0.0664, 0.0726, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 02:44:04,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-05-16 02:44:09,819 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0556, 2.2581, 2.4378, 2.3379, 2.1836, 1.9712, 2.3095, 1.7770], device='cuda:1'), covar=tensor([0.0267, 0.0184, 0.0154, 0.0160, 0.0266, 0.0244, 0.0153, 0.0348], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0159, 0.0151, 0.0184, 0.0201, 0.0197, 0.0161, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:44:15,822 INFO [finetune.py:992] (1/2) Epoch 5, batch 6550, loss[loss=0.1646, simple_loss=0.2391, pruned_loss=0.04504, over 12273.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2635, pruned_loss=0.04438, over 2367134.59 frames. ], batch size: 28, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:44:39,931 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155518.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:44:46,798 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-05-16 02:44:50,329 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.898e+02 3.292e+02 3.853e+02 7.914e+02, threshold=6.584e+02, percent-clipped=1.0 2023-05-16 02:44:51,716 INFO [finetune.py:992] (1/2) Epoch 5, batch 6600, loss[loss=0.1578, simple_loss=0.2374, pruned_loss=0.03911, over 12195.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2645, pruned_loss=0.04477, over 2363130.77 frames. ], batch size: 29, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:45:10,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 02:45:12,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 02:45:27,002 INFO [finetune.py:992] (1/2) Epoch 5, batch 6650, loss[loss=0.181, simple_loss=0.2576, pruned_loss=0.05222, over 12102.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2648, pruned_loss=0.04477, over 2366783.79 frames. ], batch size: 33, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:45:42,504 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155605.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:46:01,636 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155632.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:46:02,201 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 2.948e+02 3.456e+02 4.032e+02 7.773e+02, threshold=6.912e+02, percent-clipped=4.0 2023-05-16 02:46:03,599 INFO [finetune.py:992] (1/2) Epoch 5, batch 6700, loss[loss=0.181, simple_loss=0.2703, pruned_loss=0.04583, over 12274.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.264, pruned_loss=0.04461, over 2367729.76 frames. ], batch size: 37, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:46:14,969 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4584, 4.9880, 5.4443, 4.7587, 5.0909, 4.8862, 5.4975, 5.1022], device='cuda:1'), covar=tensor([0.0204, 0.0289, 0.0208, 0.0232, 0.0275, 0.0278, 0.0169, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0240, 0.0259, 0.0236, 0.0232, 0.0235, 0.0212, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:46:20,017 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155657.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:46:26,300 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155666.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:46:36,193 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155680.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:46:39,691 INFO [finetune.py:992] (1/2) Epoch 5, batch 6750, loss[loss=0.1562, simple_loss=0.2405, pruned_loss=0.03593, over 11982.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2627, pruned_loss=0.0439, over 2373329.61 frames. ], batch size: 28, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:47:03,094 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155718.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:47:08,231 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155725.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:47:10,453 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2062, 4.2927, 4.2281, 4.5769, 3.4062, 4.0743, 2.9614, 4.2368], device='cuda:1'), covar=tensor([0.1559, 0.0560, 0.0786, 0.0459, 0.0931, 0.0531, 0.1483, 0.1217], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0295, 0.0349, 0.0236, 0.0235, 0.0254, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:47:13,778 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.812e+02 3.211e+02 3.815e+02 1.253e+03, threshold=6.422e+02, percent-clipped=3.0 2023-05-16 02:47:15,216 INFO [finetune.py:992] (1/2) Epoch 5, batch 6800, loss[loss=0.1726, simple_loss=0.2565, pruned_loss=0.04435, over 12355.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2625, pruned_loss=0.04391, over 2370792.67 frames. ], batch size: 30, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:47:42,950 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155773.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:47:49,659 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1583, 4.2380, 3.8399, 4.7144, 4.2076, 2.7908, 4.1177, 2.8576], device='cuda:1'), covar=tensor([0.0778, 0.0983, 0.1455, 0.0477, 0.1329, 0.1604, 0.0935, 0.3197], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0365, 0.0345, 0.0264, 0.0352, 0.0260, 0.0328, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:47:51,074 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155784.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:47:51,561 INFO [finetune.py:992] (1/2) Epoch 5, batch 6850, loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.0444, over 12021.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2634, pruned_loss=0.044, over 2374032.06 frames. ], batch size: 28, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:48:12,467 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155813.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:48:26,679 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.816e+02 3.262e+02 4.172e+02 9.329e+02, threshold=6.524e+02, percent-clipped=4.0 2023-05-16 02:48:27,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 02:48:28,089 INFO [finetune.py:992] (1/2) Epoch 5, batch 6900, loss[loss=0.1637, simple_loss=0.2519, pruned_loss=0.03778, over 12133.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2636, pruned_loss=0.04407, over 2377573.38 frames. ], batch size: 30, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:48:35,329 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155845.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 02:48:45,317 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5895, 3.6579, 3.4985, 3.2523, 3.0256, 2.9108, 3.7212, 2.4229], device='cuda:1'), covar=tensor([0.0333, 0.0096, 0.0122, 0.0150, 0.0311, 0.0331, 0.0093, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0160, 0.0152, 0.0183, 0.0203, 0.0196, 0.0161, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:49:03,614 INFO [finetune.py:992] (1/2) Epoch 5, batch 6950, loss[loss=0.1606, simple_loss=0.2437, pruned_loss=0.03873, over 12241.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2638, pruned_loss=0.04383, over 2384604.46 frames. ], batch size: 32, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:49:08,775 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1198, 5.9146, 5.5911, 5.4433, 6.0521, 5.3657, 5.6670, 5.5154], device='cuda:1'), covar=tensor([0.1281, 0.0944, 0.0856, 0.1877, 0.0789, 0.1982, 0.1386, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0472, 0.0370, 0.0420, 0.0448, 0.0431, 0.0384, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:49:26,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 02:49:37,946 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 2.826e+02 3.292e+02 4.057e+02 7.257e+02, threshold=6.584e+02, percent-clipped=2.0 2023-05-16 02:49:39,399 INFO [finetune.py:992] (1/2) Epoch 5, batch 7000, loss[loss=0.1785, simple_loss=0.2692, pruned_loss=0.04387, over 12111.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2644, pruned_loss=0.04415, over 2379515.88 frames. ], batch size: 33, lr: 4.67e-03, grad_scale: 16.0 2023-05-16 02:49:49,080 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8701, 3.0388, 4.8267, 5.0142, 2.9048, 2.7763, 3.0341, 2.3077], device='cuda:1'), covar=tensor([0.1312, 0.2832, 0.0412, 0.0341, 0.1119, 0.2022, 0.2476, 0.3497], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0370, 0.0262, 0.0284, 0.0251, 0.0280, 0.0348, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:49:57,185 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9454, 5.7725, 5.4080, 5.3303, 5.8799, 5.1414, 5.4809, 5.3232], device='cuda:1'), covar=tensor([0.1491, 0.1075, 0.0985, 0.2107, 0.1054, 0.2227, 0.1858, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0475, 0.0372, 0.0425, 0.0452, 0.0436, 0.0388, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:49:58,615 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155961.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:50:00,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-05-16 02:50:15,792 INFO [finetune.py:992] (1/2) Epoch 5, batch 7050, loss[loss=0.1587, simple_loss=0.2508, pruned_loss=0.03324, over 12018.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2639, pruned_loss=0.0438, over 2375521.65 frames. ], batch size: 40, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:50:39,279 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156013.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:50:52,186 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5110, 3.1289, 5.0000, 2.6313, 2.6312, 3.6833, 3.2317, 3.7186], device='cuda:1'), covar=tensor([0.0480, 0.1158, 0.0267, 0.1145, 0.1974, 0.1403, 0.1298, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0231, 0.0234, 0.0182, 0.0239, 0.0286, 0.0227, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:50:53,333 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.045e+02 2.953e+02 3.641e+02 4.488e+02 8.703e+02, threshold=7.282e+02, percent-clipped=6.0 2023-05-16 02:50:54,730 INFO [finetune.py:992] (1/2) Epoch 5, batch 7100, loss[loss=0.1672, simple_loss=0.2617, pruned_loss=0.03636, over 12111.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2644, pruned_loss=0.04409, over 2375697.08 frames. ], batch size: 39, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:51:20,365 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156070.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:51:31,373 INFO [finetune.py:992] (1/2) Epoch 5, batch 7150, loss[loss=0.2147, simple_loss=0.2997, pruned_loss=0.06491, over 8146.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2633, pruned_loss=0.04413, over 2371585.68 frames. ], batch size: 98, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:51:51,357 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156113.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:52:04,277 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156131.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 02:52:05,481 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 2.951e+02 3.337e+02 3.935e+02 8.219e+02, threshold=6.675e+02, percent-clipped=0.0 2023-05-16 02:52:06,923 INFO [finetune.py:992] (1/2) Epoch 5, batch 7200, loss[loss=0.1969, simple_loss=0.2943, pruned_loss=0.04979, over 11848.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2636, pruned_loss=0.04428, over 2363353.61 frames. ], batch size: 44, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:52:10,476 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156140.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:52:25,183 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156161.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:52:42,320 INFO [finetune.py:992] (1/2) Epoch 5, batch 7250, loss[loss=0.1641, simple_loss=0.242, pruned_loss=0.04307, over 12179.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2635, pruned_loss=0.04442, over 2368885.74 frames. ], batch size: 29, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:52:58,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-16 02:53:05,276 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156216.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:53:17,791 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.880e+02 2.791e+02 3.354e+02 4.246e+02 6.489e+02, threshold=6.708e+02, percent-clipped=1.0 2023-05-16 02:53:19,266 INFO [finetune.py:992] (1/2) Epoch 5, batch 7300, loss[loss=0.1701, simple_loss=0.2625, pruned_loss=0.03891, over 12095.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.0444, over 2364794.97 frames. ], batch size: 32, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:53:37,824 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156261.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:53:49,244 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156277.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:53:51,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 02:53:54,753 INFO [finetune.py:992] (1/2) Epoch 5, batch 7350, loss[loss=0.1459, simple_loss=0.236, pruned_loss=0.02786, over 12343.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2632, pruned_loss=0.04448, over 2357142.79 frames. ], batch size: 31, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:53:58,571 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156290.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:54:11,927 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156309.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:54:14,914 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156313.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:54:29,421 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.962e+02 3.365e+02 3.856e+02 1.085e+03, threshold=6.730e+02, percent-clipped=2.0 2023-05-16 02:54:30,884 INFO [finetune.py:992] (1/2) Epoch 5, batch 7400, loss[loss=0.2279, simple_loss=0.3147, pruned_loss=0.07058, over 11802.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2633, pruned_loss=0.04454, over 2362378.91 frames. ], batch size: 44, lr: 4.67e-03, grad_scale: 32.0 2023-05-16 02:54:39,738 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6529, 4.5591, 4.2247, 5.1203, 4.7773, 3.2258, 4.3774, 3.1879], device='cuda:1'), covar=tensor([0.0604, 0.0922, 0.1253, 0.0361, 0.0979, 0.1284, 0.0904, 0.2840], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0367, 0.0345, 0.0266, 0.0353, 0.0259, 0.0329, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:54:42,519 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156351.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:54:49,608 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156361.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:55:06,911 INFO [finetune.py:992] (1/2) Epoch 5, batch 7450, loss[loss=0.1562, simple_loss=0.2365, pruned_loss=0.03793, over 12321.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.264, pruned_loss=0.04463, over 2364631.23 frames. ], batch size: 30, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:55:36,583 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156426.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:55:41,211 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.806e+02 3.421e+02 3.972e+02 8.442e+02, threshold=6.841e+02, percent-clipped=3.0 2023-05-16 02:55:42,624 INFO [finetune.py:992] (1/2) Epoch 5, batch 7500, loss[loss=0.1601, simple_loss=0.234, pruned_loss=0.04309, over 12348.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2634, pruned_loss=0.04454, over 2360675.74 frames. ], batch size: 31, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:55:46,276 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156440.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 02:55:54,854 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6212, 3.4823, 3.4058, 3.0701, 2.9876, 2.8136, 3.7006, 2.4659], device='cuda:1'), covar=tensor([0.0306, 0.0193, 0.0127, 0.0186, 0.0329, 0.0287, 0.0089, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0159, 0.0152, 0.0183, 0.0201, 0.0194, 0.0160, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:56:18,458 INFO [finetune.py:992] (1/2) Epoch 5, batch 7550, loss[loss=0.2086, simple_loss=0.2931, pruned_loss=0.06201, over 12366.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2637, pruned_loss=0.04469, over 2362298.24 frames. ], batch size: 38, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:56:20,477 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156488.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:56:52,755 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.893e+02 3.396e+02 4.183e+02 9.129e+02, threshold=6.793e+02, percent-clipped=2.0 2023-05-16 02:56:54,235 INFO [finetune.py:992] (1/2) Epoch 5, batch 7600, loss[loss=0.1494, simple_loss=0.2329, pruned_loss=0.03297, over 11812.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2634, pruned_loss=0.04432, over 2370282.27 frames. ], batch size: 26, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:56:59,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 02:57:10,433 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3892, 5.2365, 5.3346, 5.3998, 4.9877, 5.1312, 4.8610, 5.3507], device='cuda:1'), covar=tensor([0.0653, 0.0494, 0.0745, 0.0499, 0.1551, 0.0990, 0.0426, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0644, 0.0548, 0.0591, 0.0787, 0.0699, 0.0516, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:57:20,281 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156572.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:57:29,558 INFO [finetune.py:992] (1/2) Epoch 5, batch 7650, loss[loss=0.1601, simple_loss=0.2416, pruned_loss=0.03932, over 12116.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2631, pruned_loss=0.04411, over 2376654.81 frames. ], batch size: 30, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:57:36,037 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6397, 2.7574, 3.9554, 4.1307, 2.9646, 2.7373, 2.7872, 2.2810], device='cuda:1'), covar=tensor([0.1305, 0.2693, 0.0529, 0.0414, 0.1009, 0.1829, 0.2404, 0.3346], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0370, 0.0263, 0.0285, 0.0253, 0.0282, 0.0350, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:58:03,661 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3961, 5.2504, 5.2742, 5.3914, 5.0146, 5.0888, 4.8206, 5.3602], device='cuda:1'), covar=tensor([0.0630, 0.0514, 0.0879, 0.0501, 0.1686, 0.1044, 0.0481, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0652, 0.0556, 0.0598, 0.0799, 0.0709, 0.0522, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 02:58:04,221 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.831e+02 3.260e+02 4.176e+02 6.963e+02, threshold=6.520e+02, percent-clipped=1.0 2023-05-16 02:58:05,710 INFO [finetune.py:992] (1/2) Epoch 5, batch 7700, loss[loss=0.1627, simple_loss=0.2524, pruned_loss=0.03655, over 12358.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2624, pruned_loss=0.04374, over 2382799.32 frames. ], batch size: 31, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:58:13,620 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156646.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:58:41,771 INFO [finetune.py:992] (1/2) Epoch 5, batch 7750, loss[loss=0.1964, simple_loss=0.2779, pruned_loss=0.05742, over 12122.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2621, pruned_loss=0.04344, over 2390221.31 frames. ], batch size: 38, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:59:06,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7894, 2.9032, 4.7086, 4.8532, 2.8442, 2.6876, 2.9654, 2.2583], device='cuda:1'), covar=tensor([0.1334, 0.3042, 0.0422, 0.0351, 0.1244, 0.2076, 0.2612, 0.3751], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0367, 0.0261, 0.0282, 0.0251, 0.0279, 0.0348, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 02:59:10,782 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156726.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 02:59:15,575 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 2.973e+02 3.437e+02 3.966e+02 7.070e+02, threshold=6.873e+02, percent-clipped=1.0 2023-05-16 02:59:17,010 INFO [finetune.py:992] (1/2) Epoch 5, batch 7800, loss[loss=0.1799, simple_loss=0.2736, pruned_loss=0.04316, over 11079.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2613, pruned_loss=0.04292, over 2389048.54 frames. ], batch size: 55, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 02:59:27,259 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156749.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:59:33,033 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3388, 2.4562, 3.5116, 4.3119, 3.7731, 4.2378, 3.7957, 3.0493], device='cuda:1'), covar=tensor([0.0033, 0.0365, 0.0136, 0.0038, 0.0118, 0.0059, 0.0086, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0118, 0.0098, 0.0072, 0.0097, 0.0107, 0.0088, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:59:33,775 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6079, 2.5903, 3.6443, 4.5428, 3.9334, 4.4923, 3.9552, 3.3224], device='cuda:1'), covar=tensor([0.0026, 0.0344, 0.0119, 0.0036, 0.0109, 0.0057, 0.0097, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0118, 0.0098, 0.0072, 0.0097, 0.0107, 0.0088, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 02:59:45,717 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156774.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 02:59:50,691 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2287, 4.6990, 3.0427, 2.6413, 4.1115, 2.4785, 4.0029, 3.2434], device='cuda:1'), covar=tensor([0.0768, 0.0485, 0.0976, 0.1420, 0.0220, 0.1309, 0.0393, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0249, 0.0175, 0.0196, 0.0138, 0.0180, 0.0194, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 02:59:53,417 INFO [finetune.py:992] (1/2) Epoch 5, batch 7850, loss[loss=0.1727, simple_loss=0.2472, pruned_loss=0.04905, over 12179.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2622, pruned_loss=0.04363, over 2371981.20 frames. ], batch size: 29, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:00:12,412 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156810.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:00:28,454 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.049e+02 2.736e+02 3.172e+02 3.920e+02 5.453e+02, threshold=6.343e+02, percent-clipped=0.0 2023-05-16 03:00:29,800 INFO [finetune.py:992] (1/2) Epoch 5, batch 7900, loss[loss=0.1911, simple_loss=0.2853, pruned_loss=0.0484, over 12290.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2631, pruned_loss=0.04404, over 2368736.12 frames. ], batch size: 37, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:00:47,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.05 vs. limit=5.0 2023-05-16 03:00:55,908 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156872.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:01:05,102 INFO [finetune.py:992] (1/2) Epoch 5, batch 7950, loss[loss=0.1712, simple_loss=0.2553, pruned_loss=0.04354, over 11437.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2631, pruned_loss=0.04431, over 2361318.17 frames. ], batch size: 25, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:01:05,282 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156885.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:01:06,742 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8446, 4.8140, 4.6384, 4.7187, 4.3226, 4.8228, 4.8683, 5.0786], device='cuda:1'), covar=tensor([0.0251, 0.0171, 0.0220, 0.0360, 0.0790, 0.0429, 0.0172, 0.0186], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0184, 0.0181, 0.0229, 0.0232, 0.0200, 0.0165, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1') 2023-05-16 03:01:31,030 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156920.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:01:40,151 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 2.948e+02 3.353e+02 3.956e+02 7.512e+02, threshold=6.705e+02, percent-clipped=2.0 2023-05-16 03:01:41,646 INFO [finetune.py:992] (1/2) Epoch 5, batch 8000, loss[loss=0.1669, simple_loss=0.2448, pruned_loss=0.04455, over 11814.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.04442, over 2359933.47 frames. ], batch size: 26, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:01:49,399 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156946.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:01:49,468 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156946.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:02:11,931 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6914, 3.7649, 3.5787, 3.1894, 3.0571, 3.0804, 3.7221, 2.4583], device='cuda:1'), covar=tensor([0.0291, 0.0097, 0.0119, 0.0163, 0.0306, 0.0253, 0.0095, 0.0401], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0160, 0.0153, 0.0184, 0.0204, 0.0196, 0.0162, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:02:17,226 INFO [finetune.py:992] (1/2) Epoch 5, batch 8050, loss[loss=0.2491, simple_loss=0.31, pruned_loss=0.09413, over 6987.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2646, pruned_loss=0.04484, over 2360330.47 frames. ], batch size: 99, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:02:23,657 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156994.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:02:35,368 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157010.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:02:51,176 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.775e+02 3.369e+02 4.080e+02 5.929e+02, threshold=6.738e+02, percent-clipped=0.0 2023-05-16 03:02:52,684 INFO [finetune.py:992] (1/2) Epoch 5, batch 8100, loss[loss=0.1901, simple_loss=0.2801, pruned_loss=0.05004, over 10820.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2652, pruned_loss=0.04513, over 2370008.91 frames. ], batch size: 69, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:03:19,215 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157071.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:03:27,786 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2784, 4.6808, 2.9323, 2.5845, 4.0371, 2.6013, 3.8859, 3.0667], device='cuda:1'), covar=tensor([0.0745, 0.0606, 0.1145, 0.1624, 0.0236, 0.1359, 0.0529, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0251, 0.0177, 0.0198, 0.0139, 0.0181, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:03:29,023 INFO [finetune.py:992] (1/2) Epoch 5, batch 8150, loss[loss=0.1612, simple_loss=0.2496, pruned_loss=0.0364, over 12253.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2653, pruned_loss=0.04525, over 2368391.59 frames. ], batch size: 32, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:03:35,452 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4678, 4.9565, 5.3938, 4.7345, 5.0539, 4.9062, 5.5028, 5.0992], device='cuda:1'), covar=tensor([0.0212, 0.0334, 0.0257, 0.0234, 0.0257, 0.0262, 0.0155, 0.0234], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0246, 0.0268, 0.0239, 0.0235, 0.0238, 0.0214, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:03:41,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 03:03:43,849 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157105.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:04:02,345 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0446, 4.7735, 4.8026, 4.9683, 4.8030, 5.0515, 4.8377, 2.5623], device='cuda:1'), covar=tensor([0.0099, 0.0052, 0.0077, 0.0053, 0.0046, 0.0072, 0.0064, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0070, 0.0074, 0.0068, 0.0056, 0.0085, 0.0074, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:04:02,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 03:04:03,573 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.287e+02 2.872e+02 3.313e+02 3.901e+02 6.271e+02, threshold=6.625e+02, percent-clipped=0.0 2023-05-16 03:04:05,034 INFO [finetune.py:992] (1/2) Epoch 5, batch 8200, loss[loss=0.1857, simple_loss=0.2812, pruned_loss=0.04508, over 12154.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2654, pruned_loss=0.04518, over 2371007.22 frames. ], batch size: 36, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:04:10,192 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157142.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:04:29,688 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-05-16 03:04:37,923 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8699, 3.3820, 5.0966, 2.8284, 2.6672, 3.9433, 3.3334, 4.0118], device='cuda:1'), covar=tensor([0.0365, 0.1081, 0.0361, 0.1063, 0.1950, 0.1260, 0.1266, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0228, 0.0233, 0.0178, 0.0236, 0.0282, 0.0224, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:04:40,634 INFO [finetune.py:992] (1/2) Epoch 5, batch 8250, loss[loss=0.1484, simple_loss=0.2315, pruned_loss=0.03265, over 11778.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2651, pruned_loss=0.04519, over 2360120.02 frames. ], batch size: 26, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:04:54,465 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157203.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:04:56,615 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1437, 3.9834, 4.1208, 4.3797, 3.1332, 3.7200, 2.5699, 4.0880], device='cuda:1'), covar=tensor([0.1587, 0.0764, 0.0873, 0.0627, 0.0988, 0.0663, 0.1840, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0257, 0.0291, 0.0344, 0.0233, 0.0234, 0.0252, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:05:15,221 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.813e+02 3.377e+02 4.175e+02 1.079e+03, threshold=6.754e+02, percent-clipped=4.0 2023-05-16 03:05:16,764 INFO [finetune.py:992] (1/2) Epoch 5, batch 8300, loss[loss=0.1488, simple_loss=0.2359, pruned_loss=0.03089, over 12274.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2641, pruned_loss=0.0444, over 2368953.17 frames. ], batch size: 28, lr: 4.66e-03, grad_scale: 32.0 2023-05-16 03:05:21,673 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157241.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:05:43,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 03:05:52,681 INFO [finetune.py:992] (1/2) Epoch 5, batch 8350, loss[loss=0.2098, simple_loss=0.2969, pruned_loss=0.06134, over 11767.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2638, pruned_loss=0.04441, over 2370524.42 frames. ], batch size: 44, lr: 4.66e-03, grad_scale: 16.0 2023-05-16 03:05:57,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 03:06:27,091 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.169e+02 2.785e+02 3.372e+02 4.179e+02 8.157e+02, threshold=6.744e+02, percent-clipped=2.0 2023-05-16 03:06:27,765 INFO [finetune.py:992] (1/2) Epoch 5, batch 8400, loss[loss=0.1863, simple_loss=0.2745, pruned_loss=0.04904, over 12134.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2636, pruned_loss=0.04434, over 2374253.98 frames. ], batch size: 39, lr: 4.66e-03, grad_scale: 16.0 2023-05-16 03:06:33,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.76 vs. limit=5.0 2023-05-16 03:06:37,783 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0306, 4.6368, 4.8162, 4.9046, 4.6309, 4.9458, 4.7948, 2.9818], device='cuda:1'), covar=tensor([0.0078, 0.0084, 0.0090, 0.0067, 0.0059, 0.0086, 0.0117, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0072, 0.0075, 0.0069, 0.0057, 0.0086, 0.0075, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:06:43,581 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157356.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:06:50,574 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157366.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:07:01,824 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157382.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:07:03,856 INFO [finetune.py:992] (1/2) Epoch 5, batch 8450, loss[loss=0.164, simple_loss=0.2443, pruned_loss=0.04184, over 12162.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2633, pruned_loss=0.0446, over 2365004.85 frames. ], batch size: 29, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:07:18,220 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157404.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 03:07:18,964 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157405.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:07:27,431 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157417.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:07:39,350 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.701e+02 3.200e+02 3.996e+02 1.013e+03, threshold=6.400e+02, percent-clipped=3.0 2023-05-16 03:07:40,016 INFO [finetune.py:992] (1/2) Epoch 5, batch 8500, loss[loss=0.1886, simple_loss=0.2796, pruned_loss=0.0488, over 12043.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2626, pruned_loss=0.04415, over 2364772.39 frames. ], batch size: 42, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:07:45,774 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157443.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:07:52,619 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157453.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:07:57,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 03:08:01,249 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157465.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 03:08:15,980 INFO [finetune.py:992] (1/2) Epoch 5, batch 8550, loss[loss=0.1794, simple_loss=0.2599, pruned_loss=0.04942, over 12293.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2629, pruned_loss=0.04434, over 2367076.84 frames. ], batch size: 33, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:08:25,356 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157498.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:08:31,262 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157506.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:08:43,446 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5577, 4.0185, 3.7219, 4.2750, 3.9833, 2.4397, 3.6453, 2.7957], device='cuda:1'), covar=tensor([0.0912, 0.0851, 0.1220, 0.0573, 0.0964, 0.1726, 0.1066, 0.3119], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0367, 0.0343, 0.0266, 0.0351, 0.0261, 0.0329, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:08:51,041 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 2.972e+02 3.489e+02 4.078e+02 7.596e+02, threshold=6.977e+02, percent-clipped=3.0 2023-05-16 03:08:51,789 INFO [finetune.py:992] (1/2) Epoch 5, batch 8600, loss[loss=0.1583, simple_loss=0.2589, pruned_loss=0.02886, over 12192.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2636, pruned_loss=0.04455, over 2360863.54 frames. ], batch size: 35, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:08:56,975 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157541.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:08:59,208 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4161, 3.5044, 3.4444, 3.1366, 2.9001, 2.7180, 3.6170, 2.2851], device='cuda:1'), covar=tensor([0.0353, 0.0122, 0.0115, 0.0149, 0.0345, 0.0333, 0.0100, 0.0426], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0159, 0.0153, 0.0183, 0.0203, 0.0197, 0.0161, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:09:04,307 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-16 03:09:15,447 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157567.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 03:09:16,136 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157568.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:09:21,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 03:09:28,009 INFO [finetune.py:992] (1/2) Epoch 5, batch 8650, loss[loss=0.1876, simple_loss=0.275, pruned_loss=0.05007, over 12292.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2648, pruned_loss=0.04523, over 2354148.78 frames. ], batch size: 37, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:09:30,951 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157589.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:09:47,422 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4203, 4.7559, 2.8301, 2.3982, 4.1568, 2.4721, 3.9529, 3.2162], device='cuda:1'), covar=tensor([0.0573, 0.0503, 0.1042, 0.1695, 0.0267, 0.1494, 0.0530, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0251, 0.0175, 0.0196, 0.0139, 0.0181, 0.0195, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:09:59,781 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157629.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:10:02,964 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 3.050e+02 3.438e+02 4.172e+02 7.635e+02, threshold=6.877e+02, percent-clipped=1.0 2023-05-16 03:10:03,665 INFO [finetune.py:992] (1/2) Epoch 5, batch 8700, loss[loss=0.1673, simple_loss=0.2511, pruned_loss=0.04176, over 12342.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2647, pruned_loss=0.04504, over 2362701.03 frames. ], batch size: 31, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:10:09,274 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1873, 6.1204, 5.9688, 5.4914, 5.3058, 6.0611, 5.6501, 5.4011], device='cuda:1'), covar=tensor([0.0656, 0.0852, 0.0615, 0.1597, 0.0608, 0.0717, 0.1505, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0570, 0.0506, 0.0485, 0.0595, 0.0389, 0.0669, 0.0729, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 03:10:23,403 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4010, 4.1957, 3.8512, 4.3037, 3.3771, 3.8742, 2.7117, 4.4019], device='cuda:1'), covar=tensor([0.1312, 0.0617, 0.1277, 0.0889, 0.0813, 0.0543, 0.1493, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0257, 0.0292, 0.0346, 0.0232, 0.0234, 0.0254, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:10:25,433 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157666.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:10:34,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 03:10:39,264 INFO [finetune.py:992] (1/2) Epoch 5, batch 8750, loss[loss=0.1929, simple_loss=0.2875, pruned_loss=0.04914, over 11813.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2662, pruned_loss=0.04605, over 2351144.45 frames. ], batch size: 44, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:10:58,717 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157712.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:11:00,075 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157714.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:11:10,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-05-16 03:11:14,024 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.074e+02 2.815e+02 3.326e+02 4.142e+02 6.744e+02, threshold=6.652e+02, percent-clipped=0.0 2023-05-16 03:11:14,764 INFO [finetune.py:992] (1/2) Epoch 5, batch 8800, loss[loss=0.1969, simple_loss=0.2811, pruned_loss=0.05639, over 12122.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2649, pruned_loss=0.0451, over 2363135.51 frames. ], batch size: 39, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:11:17,035 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157738.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:11:18,157 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-16 03:11:28,466 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2861, 4.0427, 4.0377, 4.3735, 2.8233, 3.8792, 2.5844, 4.0677], device='cuda:1'), covar=tensor([0.1437, 0.0703, 0.0908, 0.0606, 0.1141, 0.0577, 0.1755, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0258, 0.0292, 0.0347, 0.0233, 0.0234, 0.0253, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:11:32,433 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157760.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 03:11:50,852 INFO [finetune.py:992] (1/2) Epoch 5, batch 8850, loss[loss=0.2784, simple_loss=0.3473, pruned_loss=0.1048, over 8005.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2642, pruned_loss=0.04479, over 2360964.23 frames. ], batch size: 99, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:12:00,169 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157798.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:12:10,173 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4615, 5.2648, 5.3497, 5.3854, 5.0485, 5.1053, 4.8896, 5.3730], device='cuda:1'), covar=tensor([0.0612, 0.0555, 0.0609, 0.0628, 0.1936, 0.1244, 0.0527, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0649, 0.0555, 0.0594, 0.0795, 0.0710, 0.0519, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 03:12:26,200 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.157e+02 3.014e+02 3.460e+02 4.025e+02 6.937e+02, threshold=6.919e+02, percent-clipped=1.0 2023-05-16 03:12:26,953 INFO [finetune.py:992] (1/2) Epoch 5, batch 8900, loss[loss=0.1823, simple_loss=0.2697, pruned_loss=0.04744, over 12372.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2639, pruned_loss=0.045, over 2364657.90 frames. ], batch size: 38, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:12:34,727 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157846.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:12:40,011 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157853.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:12:46,255 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157862.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 03:13:02,492 INFO [finetune.py:992] (1/2) Epoch 5, batch 8950, loss[loss=0.1858, simple_loss=0.2818, pruned_loss=0.04497, over 12344.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2644, pruned_loss=0.04508, over 2371808.81 frames. ], batch size: 36, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:13:24,048 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157914.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:13:26,191 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4567, 3.5442, 3.3450, 3.0383, 2.8875, 2.8364, 3.6065, 2.1659], device='cuda:1'), covar=tensor([0.0331, 0.0126, 0.0131, 0.0177, 0.0323, 0.0280, 0.0109, 0.0424], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0158, 0.0153, 0.0183, 0.0203, 0.0195, 0.0161, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:13:29,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-05-16 03:13:30,678 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.14 vs. limit=5.0 2023-05-16 03:13:31,086 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157924.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:13:37,962 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.977e+02 3.480e+02 4.159e+02 6.142e+02, threshold=6.960e+02, percent-clipped=0.0 2023-05-16 03:13:38,629 INFO [finetune.py:992] (1/2) Epoch 5, batch 9000, loss[loss=0.1847, simple_loss=0.2774, pruned_loss=0.04601, over 12150.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2646, pruned_loss=0.04544, over 2367842.09 frames. ], batch size: 34, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:13:38,629 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 03:13:57,466 INFO [finetune.py:1026] (1/2) Epoch 5, validation: loss=0.3387, simple_loss=0.4077, pruned_loss=0.1348, over 1020973.00 frames. 2023-05-16 03:13:57,467 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 03:14:05,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 03:14:32,904 INFO [finetune.py:992] (1/2) Epoch 5, batch 9050, loss[loss=0.2128, simple_loss=0.3057, pruned_loss=0.05991, over 10301.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2649, pruned_loss=0.04589, over 2363398.77 frames. ], batch size: 68, lr: 4.65e-03, grad_scale: 16.0 2023-05-16 03:14:55,856 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158012.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:15:12,121 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.887e+02 3.613e+02 4.458e+02 1.304e+03, threshold=7.226e+02, percent-clipped=2.0 2023-05-16 03:15:12,142 INFO [finetune.py:992] (1/2) Epoch 5, batch 9100, loss[loss=0.1935, simple_loss=0.2833, pruned_loss=0.05178, over 11756.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.265, pruned_loss=0.04605, over 2360790.20 frames. ], batch size: 44, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:15:14,463 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158038.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:15:27,319 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1609, 5.9843, 5.5427, 5.4691, 6.0534, 5.4092, 5.5941, 5.5826], device='cuda:1'), covar=tensor([0.1333, 0.0884, 0.0957, 0.1696, 0.0833, 0.1748, 0.1604, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0460, 0.0365, 0.0410, 0.0442, 0.0421, 0.0377, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:15:30,072 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158060.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:15:30,177 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158060.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 03:15:47,993 INFO [finetune.py:992] (1/2) Epoch 5, batch 9150, loss[loss=0.1586, simple_loss=0.2421, pruned_loss=0.03754, over 12136.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2656, pruned_loss=0.04608, over 2351473.34 frames. ], batch size: 30, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:15:48,785 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158086.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:16:01,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1513, 2.1490, 2.7509, 3.0524, 3.0005, 3.0701, 2.8201, 2.5060], device='cuda:1'), covar=tensor([0.0059, 0.0341, 0.0165, 0.0054, 0.0095, 0.0097, 0.0107, 0.0283], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0120, 0.0101, 0.0073, 0.0097, 0.0111, 0.0089, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:16:04,405 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158108.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 03:16:14,078 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158122.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:16:23,215 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 3.039e+02 3.389e+02 4.190e+02 1.167e+03, threshold=6.777e+02, percent-clipped=1.0 2023-05-16 03:16:23,235 INFO [finetune.py:992] (1/2) Epoch 5, batch 9200, loss[loss=0.1922, simple_loss=0.2788, pruned_loss=0.05281, over 12042.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2658, pruned_loss=0.04644, over 2347950.99 frames. ], batch size: 42, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:16:24,188 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5368, 2.5397, 3.6239, 4.4433, 3.9942, 4.4230, 3.8124, 3.1474], device='cuda:1'), covar=tensor([0.0024, 0.0353, 0.0120, 0.0035, 0.0088, 0.0064, 0.0090, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0120, 0.0101, 0.0073, 0.0097, 0.0110, 0.0089, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:16:29,005 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6006, 4.5314, 4.3348, 4.7462, 3.5181, 4.2798, 3.0266, 4.3935], device='cuda:1'), covar=tensor([0.1327, 0.0545, 0.0716, 0.0572, 0.0848, 0.0453, 0.1423, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0254, 0.0289, 0.0345, 0.0230, 0.0232, 0.0250, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:16:36,225 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6356, 2.6952, 3.6610, 4.5669, 4.0310, 4.5704, 3.8730, 3.2717], device='cuda:1'), covar=tensor([0.0028, 0.0345, 0.0124, 0.0037, 0.0109, 0.0054, 0.0084, 0.0291], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0120, 0.0101, 0.0073, 0.0097, 0.0111, 0.0089, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:16:43,448 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158162.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 03:16:58,170 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158183.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:16:59,346 INFO [finetune.py:992] (1/2) Epoch 5, batch 9250, loss[loss=0.1847, simple_loss=0.2709, pruned_loss=0.04921, over 12132.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2656, pruned_loss=0.04614, over 2347270.26 frames. ], batch size: 39, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:17:12,578 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6320, 2.7616, 3.6417, 4.5404, 4.0328, 4.5093, 3.8449, 3.0855], device='cuda:1'), covar=tensor([0.0036, 0.0342, 0.0132, 0.0050, 0.0098, 0.0073, 0.0096, 0.0355], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0121, 0.0102, 0.0074, 0.0098, 0.0112, 0.0090, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:17:16,816 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158209.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:17:17,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 03:17:17,576 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158210.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:17:23,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4197, 5.2127, 5.2898, 5.3951, 5.0593, 5.0372, 4.8702, 5.3213], device='cuda:1'), covar=tensor([0.0653, 0.0550, 0.0712, 0.0544, 0.1719, 0.1166, 0.0502, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0641, 0.0547, 0.0586, 0.0781, 0.0698, 0.0513, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 03:17:25,119 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6331, 4.6021, 4.5000, 4.1618, 4.2824, 4.5874, 4.3098, 4.1376], device='cuda:1'), covar=tensor([0.0703, 0.0838, 0.0691, 0.1314, 0.1632, 0.0862, 0.1423, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0513, 0.0490, 0.0601, 0.0391, 0.0676, 0.0740, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 03:17:27,979 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158224.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:17:30,806 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9189, 5.8916, 5.6440, 5.1840, 5.0771, 5.7713, 5.4576, 5.1662], device='cuda:1'), covar=tensor([0.0590, 0.0704, 0.0633, 0.1403, 0.0673, 0.0655, 0.1396, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0512, 0.0489, 0.0600, 0.0390, 0.0675, 0.0738, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 03:17:35,584 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.936e+02 3.408e+02 4.274e+02 9.884e+02, threshold=6.817e+02, percent-clipped=6.0 2023-05-16 03:17:35,603 INFO [finetune.py:992] (1/2) Epoch 5, batch 9300, loss[loss=0.1621, simple_loss=0.244, pruned_loss=0.04013, over 12164.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2659, pruned_loss=0.04598, over 2352464.05 frames. ], batch size: 31, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:18:02,085 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158272.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:18:11,518 INFO [finetune.py:992] (1/2) Epoch 5, batch 9350, loss[loss=0.1691, simple_loss=0.2474, pruned_loss=0.04546, over 12093.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.2663, pruned_loss=0.04628, over 2345139.90 frames. ], batch size: 32, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:18:41,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 03:18:46,868 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.787e+02 3.259e+02 4.117e+02 8.884e+02, threshold=6.518e+02, percent-clipped=2.0 2023-05-16 03:18:46,887 INFO [finetune.py:992] (1/2) Epoch 5, batch 9400, loss[loss=0.1726, simple_loss=0.2662, pruned_loss=0.03953, over 12357.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2643, pruned_loss=0.04533, over 2352908.24 frames. ], batch size: 35, lr: 4.65e-03, grad_scale: 8.0 2023-05-16 03:19:22,498 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158384.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:19:23,016 INFO [finetune.py:992] (1/2) Epoch 5, batch 9450, loss[loss=0.1892, simple_loss=0.2807, pruned_loss=0.04888, over 11202.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2644, pruned_loss=0.04486, over 2360008.67 frames. ], batch size: 55, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:19:48,426 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-16 03:19:51,546 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2547, 4.8080, 5.2238, 4.5431, 4.8898, 4.6763, 5.2830, 4.9121], device='cuda:1'), covar=tensor([0.0227, 0.0345, 0.0242, 0.0232, 0.0285, 0.0252, 0.0179, 0.0292], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0248, 0.0267, 0.0238, 0.0234, 0.0238, 0.0215, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:19:59,143 INFO [finetune.py:992] (1/2) Epoch 5, batch 9500, loss[loss=0.1659, simple_loss=0.2604, pruned_loss=0.03568, over 12104.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2636, pruned_loss=0.04434, over 2362289.21 frames. ], batch size: 33, lr: 4.64e-03, grad_scale: 4.0 2023-05-16 03:19:59,875 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 2.755e+02 3.329e+02 4.230e+02 9.398e+02, threshold=6.658e+02, percent-clipped=3.0 2023-05-16 03:20:05,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 03:20:06,415 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158445.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:20:29,333 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158478.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:20:34,331 INFO [finetune.py:992] (1/2) Epoch 5, batch 9550, loss[loss=0.1603, simple_loss=0.2465, pruned_loss=0.03708, over 12344.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2629, pruned_loss=0.04434, over 2365635.10 frames. ], batch size: 30, lr: 4.64e-03, grad_scale: 4.0 2023-05-16 03:20:51,536 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158509.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:21:10,490 INFO [finetune.py:992] (1/2) Epoch 5, batch 9600, loss[loss=0.1897, simple_loss=0.2741, pruned_loss=0.05269, over 12099.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2635, pruned_loss=0.04486, over 2354400.09 frames. ], batch size: 38, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:21:11,202 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 3.116e+02 3.732e+02 4.442e+02 1.315e+03, threshold=7.464e+02, percent-clipped=4.0 2023-05-16 03:21:26,404 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158557.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:21:45,279 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0361, 2.9218, 4.3905, 2.4344, 2.5784, 3.3826, 3.0601, 3.5252], device='cuda:1'), covar=tensor([0.0533, 0.1264, 0.0375, 0.1174, 0.1827, 0.1305, 0.1269, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0226, 0.0233, 0.0177, 0.0235, 0.0281, 0.0222, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:21:46,381 INFO [finetune.py:992] (1/2) Epoch 5, batch 9650, loss[loss=0.1675, simple_loss=0.2571, pruned_loss=0.03893, over 12139.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2641, pruned_loss=0.0448, over 2361108.33 frames. ], batch size: 30, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:22:22,042 INFO [finetune.py:992] (1/2) Epoch 5, batch 9700, loss[loss=0.1621, simple_loss=0.2482, pruned_loss=0.03802, over 12290.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2631, pruned_loss=0.04437, over 2368471.17 frames. ], batch size: 28, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:22:22,702 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.939e+02 3.389e+02 3.950e+02 7.191e+02, threshold=6.777e+02, percent-clipped=0.0 2023-05-16 03:22:45,737 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2120, 3.8014, 4.1394, 4.4247, 3.0020, 4.0070, 2.4987, 4.0617], device='cuda:1'), covar=tensor([0.1554, 0.0820, 0.0858, 0.0580, 0.1060, 0.0566, 0.1855, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0254, 0.0288, 0.0343, 0.0229, 0.0232, 0.0248, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:22:58,191 INFO [finetune.py:992] (1/2) Epoch 5, batch 9750, loss[loss=0.1723, simple_loss=0.2609, pruned_loss=0.04188, over 12043.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2632, pruned_loss=0.04458, over 2368375.82 frames. ], batch size: 40, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:23:24,787 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4230, 2.7646, 3.9057, 3.3896, 3.6960, 3.4469, 2.8182, 3.6988], device='cuda:1'), covar=tensor([0.0099, 0.0268, 0.0110, 0.0163, 0.0128, 0.0144, 0.0267, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0196, 0.0177, 0.0176, 0.0201, 0.0153, 0.0187, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:23:34,532 INFO [finetune.py:992] (1/2) Epoch 5, batch 9800, loss[loss=0.1872, simple_loss=0.2858, pruned_loss=0.04433, over 12083.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2627, pruned_loss=0.04393, over 2376537.27 frames. ], batch size: 42, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:23:35,226 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.850e+02 3.458e+02 4.279e+02 8.665e+02, threshold=6.916e+02, percent-clipped=3.0 2023-05-16 03:23:38,184 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158740.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:23:49,577 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4008, 4.8249, 3.1134, 2.7373, 4.1571, 2.5560, 4.1202, 3.5154], device='cuda:1'), covar=tensor([0.0545, 0.0445, 0.0880, 0.1289, 0.0243, 0.1293, 0.0438, 0.0635], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0247, 0.0173, 0.0192, 0.0137, 0.0177, 0.0191, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:24:04,289 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158777.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:24:04,901 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158778.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:24:08,558 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158783.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:24:09,799 INFO [finetune.py:992] (1/2) Epoch 5, batch 9850, loss[loss=0.1642, simple_loss=0.2467, pruned_loss=0.04084, over 12032.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2614, pruned_loss=0.04344, over 2374185.71 frames. ], batch size: 31, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:24:22,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 03:24:29,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 03:24:30,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 03:24:39,906 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158826.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:24:46,324 INFO [finetune.py:992] (1/2) Epoch 5, batch 9900, loss[loss=0.1561, simple_loss=0.2416, pruned_loss=0.03532, over 12349.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2619, pruned_loss=0.04381, over 2374541.08 frames. ], batch size: 31, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:24:46,528 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158835.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:24:47,058 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.906e+02 3.304e+02 4.242e+02 8.853e+02, threshold=6.608e+02, percent-clipped=3.0 2023-05-16 03:24:48,638 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158838.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:24:52,950 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158844.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:24:53,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 03:25:13,633 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 03:25:22,457 INFO [finetune.py:992] (1/2) Epoch 5, batch 9950, loss[loss=0.1458, simple_loss=0.2335, pruned_loss=0.02907, over 12336.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2614, pruned_loss=0.04323, over 2382839.14 frames. ], batch size: 31, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:25:30,705 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158896.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:25:58,018 INFO [finetune.py:992] (1/2) Epoch 5, batch 10000, loss[loss=0.1813, simple_loss=0.2768, pruned_loss=0.04288, over 11365.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2619, pruned_loss=0.0434, over 2381548.78 frames. ], batch size: 55, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:25:58,701 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 2.839e+02 3.264e+02 3.961e+02 6.729e+02, threshold=6.527e+02, percent-clipped=1.0 2023-05-16 03:26:33,998 INFO [finetune.py:992] (1/2) Epoch 5, batch 10050, loss[loss=0.1559, simple_loss=0.2401, pruned_loss=0.0358, over 12174.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2618, pruned_loss=0.04322, over 2383934.76 frames. ], batch size: 29, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:26:46,266 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 03:27:10,572 INFO [finetune.py:992] (1/2) Epoch 5, batch 10100, loss[loss=0.1807, simple_loss=0.275, pruned_loss=0.04319, over 12007.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2623, pruned_loss=0.04343, over 2387493.33 frames. ], batch size: 40, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:27:11,289 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 3.065e+02 3.689e+02 4.338e+02 1.683e+03, threshold=7.377e+02, percent-clipped=4.0 2023-05-16 03:27:12,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 03:27:14,258 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159040.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:27:17,711 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1686, 5.9665, 5.6372, 5.5998, 6.0649, 5.5738, 5.5170, 5.6331], device='cuda:1'), covar=tensor([0.1543, 0.0937, 0.1078, 0.1791, 0.0987, 0.1880, 0.1810, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0465, 0.0366, 0.0418, 0.0446, 0.0428, 0.0382, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:27:35,270 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159069.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:27:46,475 INFO [finetune.py:992] (1/2) Epoch 5, batch 10150, loss[loss=0.1665, simple_loss=0.2582, pruned_loss=0.0374, over 12157.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2623, pruned_loss=0.04303, over 2388595.23 frames. ], batch size: 34, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:27:48,669 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159088.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:27:58,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-05-16 03:28:19,026 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159130.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:28:21,087 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159133.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:28:22,021 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 03:28:22,438 INFO [finetune.py:992] (1/2) Epoch 5, batch 10200, loss[loss=0.2008, simple_loss=0.292, pruned_loss=0.05478, over 11704.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2634, pruned_loss=0.0437, over 2376582.04 frames. ], batch size: 48, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:28:23,160 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.159e+02 2.738e+02 3.327e+02 4.064e+02 6.892e+02, threshold=6.654e+02, percent-clipped=0.0 2023-05-16 03:28:25,382 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159139.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:28:40,240 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159159.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:28:58,530 INFO [finetune.py:992] (1/2) Epoch 5, batch 10250, loss[loss=0.146, simple_loss=0.2379, pruned_loss=0.02703, over 12180.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2633, pruned_loss=0.04375, over 2380375.31 frames. ], batch size: 31, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:29:02,733 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159191.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 03:29:02,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2376, 4.1816, 4.1918, 4.5805, 3.0182, 4.0291, 2.6644, 4.2544], device='cuda:1'), covar=tensor([0.1637, 0.0679, 0.0805, 0.0578, 0.1128, 0.0588, 0.1760, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0257, 0.0292, 0.0347, 0.0233, 0.0233, 0.0252, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:29:23,469 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159220.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:29:33,798 INFO [finetune.py:992] (1/2) Epoch 5, batch 10300, loss[loss=0.1845, simple_loss=0.2781, pruned_loss=0.0455, over 12192.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2638, pruned_loss=0.04385, over 2379598.36 frames. ], batch size: 35, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:29:34,471 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.968e+02 3.473e+02 4.181e+02 8.488e+02, threshold=6.946e+02, percent-clipped=6.0 2023-05-16 03:29:46,913 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7320, 2.6314, 3.4358, 4.5655, 2.4172, 4.5401, 4.6417, 4.8113], device='cuda:1'), covar=tensor([0.0118, 0.1228, 0.0408, 0.0178, 0.1317, 0.0226, 0.0139, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0200, 0.0184, 0.0113, 0.0188, 0.0176, 0.0169, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:30:01,575 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1920, 3.1631, 4.6052, 2.4985, 2.5715, 3.5324, 3.1592, 3.5855], device='cuda:1'), covar=tensor([0.0540, 0.1100, 0.0348, 0.1176, 0.1993, 0.1307, 0.1228, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0226, 0.0233, 0.0178, 0.0234, 0.0281, 0.0223, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:30:09,732 INFO [finetune.py:992] (1/2) Epoch 5, batch 10350, loss[loss=0.1503, simple_loss=0.233, pruned_loss=0.03377, over 12175.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.263, pruned_loss=0.0436, over 2378788.82 frames. ], batch size: 29, lr: 4.64e-03, grad_scale: 8.0 2023-05-16 03:30:14,207 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159290.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:30:19,299 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6197, 3.2920, 5.0419, 2.4655, 2.6045, 3.7237, 3.3261, 3.7643], device='cuda:1'), covar=tensor([0.0430, 0.1133, 0.0309, 0.1224, 0.2058, 0.1255, 0.1282, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0227, 0.0234, 0.0178, 0.0235, 0.0282, 0.0223, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:30:45,686 INFO [finetune.py:992] (1/2) Epoch 5, batch 10400, loss[loss=0.1814, simple_loss=0.2727, pruned_loss=0.04504, over 12289.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2639, pruned_loss=0.04419, over 2376246.15 frames. ], batch size: 37, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:30:46,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.159e+02 2.730e+02 3.340e+02 3.902e+02 7.144e+02, threshold=6.680e+02, percent-clipped=1.0 2023-05-16 03:30:50,917 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3763, 4.6766, 2.7320, 2.6948, 3.9880, 2.5450, 4.0875, 3.4077], device='cuda:1'), covar=tensor([0.0539, 0.0543, 0.1083, 0.1279, 0.0252, 0.1221, 0.0363, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0251, 0.0175, 0.0195, 0.0138, 0.0179, 0.0194, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:30:57,186 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159351.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:31:01,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 03:31:20,671 INFO [finetune.py:992] (1/2) Epoch 5, batch 10450, loss[loss=0.1498, simple_loss=0.2344, pruned_loss=0.03257, over 12341.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2627, pruned_loss=0.0437, over 2381289.03 frames. ], batch size: 31, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:31:49,887 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159425.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:31:55,870 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159433.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:31:57,739 INFO [finetune.py:992] (1/2) Epoch 5, batch 10500, loss[loss=0.2429, simple_loss=0.3179, pruned_loss=0.08393, over 8034.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2628, pruned_loss=0.04379, over 2382667.21 frames. ], batch size: 98, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:31:58,430 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.917e+02 3.389e+02 4.133e+02 8.328e+02, threshold=6.779e+02, percent-clipped=3.0 2023-05-16 03:32:00,777 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159439.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:32:22,979 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-05-16 03:32:30,262 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159481.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:32:33,173 INFO [finetune.py:992] (1/2) Epoch 5, batch 10550, loss[loss=0.1739, simple_loss=0.2641, pruned_loss=0.04185, over 12273.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2629, pruned_loss=0.04398, over 2377176.48 frames. ], batch size: 37, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:32:34,640 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159487.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:32:37,531 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159491.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 03:32:54,604 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159515.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:32:58,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 03:33:08,619 INFO [finetune.py:992] (1/2) Epoch 5, batch 10600, loss[loss=0.158, simple_loss=0.2383, pruned_loss=0.03881, over 12355.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2641, pruned_loss=0.04476, over 2356950.47 frames. ], batch size: 30, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:33:09,308 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.825e+02 2.829e+02 3.235e+02 4.043e+02 7.652e+02, threshold=6.471e+02, percent-clipped=2.0 2023-05-16 03:33:11,570 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159539.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:33:11,770 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6656, 2.9423, 4.6093, 4.8659, 2.9376, 2.6508, 2.9380, 2.1227], device='cuda:1'), covar=tensor([0.1397, 0.2672, 0.0421, 0.0340, 0.1103, 0.2042, 0.2503, 0.3580], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0368, 0.0264, 0.0287, 0.0252, 0.0281, 0.0353, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:33:22,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 03:33:32,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-05-16 03:33:41,357 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4568, 5.0439, 5.4573, 4.7707, 5.0732, 4.8827, 5.5134, 5.0608], device='cuda:1'), covar=tensor([0.0221, 0.0303, 0.0196, 0.0211, 0.0302, 0.0235, 0.0166, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0247, 0.0263, 0.0238, 0.0234, 0.0237, 0.0214, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:33:45,382 INFO [finetune.py:992] (1/2) Epoch 5, batch 10650, loss[loss=0.2034, simple_loss=0.2948, pruned_loss=0.05604, over 10535.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2636, pruned_loss=0.04465, over 2352410.02 frames. ], batch size: 69, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:34:11,844 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0163, 3.8574, 4.0080, 3.7046, 3.8575, 3.6902, 4.0110, 3.5854], device='cuda:1'), covar=tensor([0.0317, 0.0331, 0.0303, 0.0239, 0.0308, 0.0294, 0.0270, 0.1099], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0248, 0.0264, 0.0239, 0.0234, 0.0238, 0.0216, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:34:18,253 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159631.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:34:20,998 INFO [finetune.py:992] (1/2) Epoch 5, batch 10700, loss[loss=0.1478, simple_loss=0.233, pruned_loss=0.03128, over 12167.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2639, pruned_loss=0.0446, over 2352160.01 frames. ], batch size: 29, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:34:21,650 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 2.820e+02 3.467e+02 4.069e+02 7.983e+02, threshold=6.934e+02, percent-clipped=3.0 2023-05-16 03:34:28,989 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159646.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:34:54,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.27 vs. limit=5.0 2023-05-16 03:34:56,620 INFO [finetune.py:992] (1/2) Epoch 5, batch 10750, loss[loss=0.1878, simple_loss=0.2739, pruned_loss=0.05089, over 12179.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.263, pruned_loss=0.04396, over 2359839.62 frames. ], batch size: 35, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:35:01,539 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159692.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:35:26,135 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159725.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:35:33,091 INFO [finetune.py:992] (1/2) Epoch 5, batch 10800, loss[loss=0.1471, simple_loss=0.2422, pruned_loss=0.026, over 12330.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2616, pruned_loss=0.04319, over 2370958.39 frames. ], batch size: 31, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:35:33,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.840e+02 3.248e+02 3.728e+02 9.603e+02, threshold=6.496e+02, percent-clipped=2.0 2023-05-16 03:36:00,165 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159773.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:36:08,599 INFO [finetune.py:992] (1/2) Epoch 5, batch 10850, loss[loss=0.1573, simple_loss=0.2605, pruned_loss=0.02711, over 12300.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.263, pruned_loss=0.04407, over 2362467.52 frames. ], batch size: 34, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:36:09,562 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1730, 2.3090, 3.6839, 3.1265, 3.5413, 3.2185, 2.4761, 3.6132], device='cuda:1'), covar=tensor([0.0121, 0.0345, 0.0121, 0.0189, 0.0107, 0.0161, 0.0340, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0194, 0.0175, 0.0174, 0.0200, 0.0152, 0.0186, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:36:31,057 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159815.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:36:45,705 INFO [finetune.py:992] (1/2) Epoch 5, batch 10900, loss[loss=0.1703, simple_loss=0.2689, pruned_loss=0.03586, over 12301.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2633, pruned_loss=0.04405, over 2368129.25 frames. ], batch size: 34, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:36:46,403 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 2.789e+02 3.278e+02 4.088e+02 6.595e+02, threshold=6.557e+02, percent-clipped=3.0 2023-05-16 03:37:00,787 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159855.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:37:06,418 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159863.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:37:16,537 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5245, 2.8308, 3.8243, 2.3748, 2.5699, 3.1713, 2.8965, 3.2089], device='cuda:1'), covar=tensor([0.0513, 0.1138, 0.0411, 0.1221, 0.1693, 0.1276, 0.1147, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0230, 0.0234, 0.0179, 0.0235, 0.0283, 0.0223, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:37:22,635 INFO [finetune.py:992] (1/2) Epoch 5, batch 10950, loss[loss=0.2799, simple_loss=0.3452, pruned_loss=0.1073, over 8622.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.264, pruned_loss=0.04459, over 2362571.29 frames. ], batch size: 98, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:37:27,722 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2535, 4.8649, 5.2332, 4.5318, 4.8967, 4.5934, 5.2419, 4.8964], device='cuda:1'), covar=tensor([0.0261, 0.0338, 0.0285, 0.0241, 0.0344, 0.0308, 0.0230, 0.0316], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0247, 0.0264, 0.0239, 0.0234, 0.0237, 0.0214, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:37:40,261 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159910.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:37:44,425 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159916.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:37:57,637 INFO [finetune.py:992] (1/2) Epoch 5, batch 11000, loss[loss=0.2557, simple_loss=0.336, pruned_loss=0.08766, over 10488.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2684, pruned_loss=0.04718, over 2330748.68 frames. ], batch size: 68, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:37:58,345 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 3.202e+02 3.873e+02 4.673e+02 7.110e+02, threshold=7.746e+02, percent-clipped=2.0 2023-05-16 03:38:05,619 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159946.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:38:06,374 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0462, 4.6303, 5.0238, 4.3801, 4.7156, 4.4732, 5.0598, 4.6950], device='cuda:1'), covar=tensor([0.0254, 0.0332, 0.0264, 0.0236, 0.0302, 0.0263, 0.0192, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0246, 0.0262, 0.0237, 0.0233, 0.0235, 0.0213, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:38:23,992 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159971.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:38:25,437 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2098, 1.9573, 2.3462, 2.1269, 2.2751, 2.4101, 1.8867, 2.3237], device='cuda:1'), covar=tensor([0.0095, 0.0231, 0.0143, 0.0146, 0.0122, 0.0133, 0.0220, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0192, 0.0174, 0.0173, 0.0199, 0.0151, 0.0186, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:38:34,241 INFO [finetune.py:992] (1/2) Epoch 5, batch 11050, loss[loss=0.1533, simple_loss=0.2357, pruned_loss=0.03551, over 12330.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2708, pruned_loss=0.04876, over 2305697.47 frames. ], batch size: 30, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:38:35,730 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159987.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:38:40,680 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159994.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:39:04,908 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1985, 4.3888, 4.1093, 4.8743, 4.5695, 2.7574, 4.2046, 3.0974], device='cuda:1'), covar=tensor([0.0785, 0.0912, 0.1197, 0.0436, 0.0983, 0.1603, 0.0960, 0.3106], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0366, 0.0341, 0.0264, 0.0350, 0.0257, 0.0326, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:39:10,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 03:39:12,049 INFO [finetune.py:992] (1/2) Epoch 5, batch 11100, loss[loss=0.2041, simple_loss=0.2988, pruned_loss=0.0547, over 11474.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2752, pruned_loss=0.05152, over 2263044.57 frames. ], batch size: 48, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:39:12,760 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.336e+02 3.352e+02 3.764e+02 4.543e+02 1.015e+03, threshold=7.528e+02, percent-clipped=3.0 2023-05-16 03:39:14,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 03:39:34,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-05-16 03:39:48,067 INFO [finetune.py:992] (1/2) Epoch 5, batch 11150, loss[loss=0.2003, simple_loss=0.2915, pruned_loss=0.05456, over 11605.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2804, pruned_loss=0.05537, over 2205407.13 frames. ], batch size: 48, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:40:22,744 INFO [finetune.py:992] (1/2) Epoch 5, batch 11200, loss[loss=0.2108, simple_loss=0.3, pruned_loss=0.06083, over 12008.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2864, pruned_loss=0.05941, over 2141537.83 frames. ], batch size: 40, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:40:23,515 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.431e+02 3.491e+02 4.351e+02 5.198e+02 9.612e+02, threshold=8.701e+02, percent-clipped=4.0 2023-05-16 03:40:51,990 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5205, 4.3881, 4.4189, 4.0553, 4.1658, 4.4818, 4.1828, 4.0867], device='cuda:1'), covar=tensor([0.0797, 0.1230, 0.0673, 0.1468, 0.1782, 0.0880, 0.1555, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0506, 0.0482, 0.0587, 0.0383, 0.0661, 0.0717, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 03:40:54,138 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6341, 4.3482, 4.6113, 4.1128, 4.3879, 4.1823, 4.5945, 4.1850], device='cuda:1'), covar=tensor([0.0235, 0.0315, 0.0246, 0.0260, 0.0302, 0.0285, 0.0230, 0.0533], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0235, 0.0250, 0.0228, 0.0224, 0.0226, 0.0205, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:40:58,857 INFO [finetune.py:992] (1/2) Epoch 5, batch 11250, loss[loss=0.3016, simple_loss=0.3602, pruned_loss=0.1216, over 6738.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2938, pruned_loss=0.06475, over 2081585.46 frames. ], batch size: 99, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:41:05,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 03:41:16,637 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:41:34,364 INFO [finetune.py:992] (1/2) Epoch 5, batch 11300, loss[loss=0.2223, simple_loss=0.3085, pruned_loss=0.06804, over 11114.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.3001, pruned_loss=0.06904, over 2032938.96 frames. ], batch size: 55, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:41:35,060 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.451e+02 3.864e+02 4.459e+02 5.256e+02 1.641e+03, threshold=8.918e+02, percent-clipped=3.0 2023-05-16 03:41:55,497 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160266.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:42:08,363 INFO [finetune.py:992] (1/2) Epoch 5, batch 11350, loss[loss=0.2382, simple_loss=0.3136, pruned_loss=0.08142, over 10989.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3056, pruned_loss=0.07293, over 1954284.74 frames. ], batch size: 55, lr: 4.63e-03, grad_scale: 8.0 2023-05-16 03:42:10,618 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160287.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:42:14,631 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4045, 4.9817, 5.3864, 4.7658, 5.0076, 4.8574, 5.4086, 5.0913], device='cuda:1'), covar=tensor([0.0232, 0.0300, 0.0236, 0.0232, 0.0338, 0.0267, 0.0193, 0.0230], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0233, 0.0248, 0.0227, 0.0223, 0.0224, 0.0203, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:42:24,916 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160307.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 03:42:43,616 INFO [finetune.py:992] (1/2) Epoch 5, batch 11400, loss[loss=0.2703, simple_loss=0.3465, pruned_loss=0.09703, over 6792.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3096, pruned_loss=0.07549, over 1925609.43 frames. ], batch size: 98, lr: 4.62e-03, grad_scale: 8.0 2023-05-16 03:42:43,723 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160335.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:42:44,360 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.326e+02 3.579e+02 4.391e+02 4.990e+02 9.272e+02, threshold=8.781e+02, percent-clipped=0.0 2023-05-16 03:42:52,696 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1463, 4.2171, 2.5280, 2.1713, 3.7430, 2.2183, 3.8247, 2.9137], device='cuda:1'), covar=tensor([0.0606, 0.0334, 0.1139, 0.1786, 0.0248, 0.1524, 0.0348, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0243, 0.0171, 0.0193, 0.0135, 0.0178, 0.0188, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:43:06,741 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5190, 3.1437, 5.0550, 2.6243, 2.6928, 4.0141, 3.1087, 4.0407], device='cuda:1'), covar=tensor([0.0467, 0.1182, 0.0174, 0.1250, 0.1987, 0.1050, 0.1469, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0223, 0.0223, 0.0173, 0.0227, 0.0272, 0.0215, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:43:07,368 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160368.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 03:43:19,020 INFO [finetune.py:992] (1/2) Epoch 5, batch 11450, loss[loss=0.1995, simple_loss=0.2878, pruned_loss=0.05565, over 12258.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3126, pruned_loss=0.07745, over 1910582.31 frames. ], batch size: 32, lr: 4.62e-03, grad_scale: 8.0 2023-05-16 03:43:31,125 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2352, 1.9333, 2.2619, 2.0331, 2.1679, 2.3142, 1.7589, 2.2184], device='cuda:1'), covar=tensor([0.0082, 0.0251, 0.0107, 0.0186, 0.0128, 0.0136, 0.0272, 0.0097], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0187, 0.0166, 0.0166, 0.0190, 0.0146, 0.0179, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:43:54,192 INFO [finetune.py:992] (1/2) Epoch 5, batch 11500, loss[loss=0.2246, simple_loss=0.3094, pruned_loss=0.06993, over 11213.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3152, pruned_loss=0.08011, over 1859654.39 frames. ], batch size: 55, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:43:54,823 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.655e+02 3.566e+02 4.414e+02 5.138e+02 9.853e+02, threshold=8.829e+02, percent-clipped=2.0 2023-05-16 03:44:30,037 INFO [finetune.py:992] (1/2) Epoch 5, batch 11550, loss[loss=0.3028, simple_loss=0.3544, pruned_loss=0.1256, over 7165.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3168, pruned_loss=0.08174, over 1842589.66 frames. ], batch size: 99, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:44:47,744 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160511.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:45:03,835 INFO [finetune.py:992] (1/2) Epoch 5, batch 11600, loss[loss=0.2258, simple_loss=0.3052, pruned_loss=0.0732, over 10782.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3182, pruned_loss=0.08325, over 1826942.69 frames. ], batch size: 69, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:45:04,463 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.282e+02 3.465e+02 4.046e+02 4.621e+02 7.818e+02, threshold=8.092e+02, percent-clipped=0.0 2023-05-16 03:45:21,378 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160559.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:45:26,657 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160566.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:45:40,419 INFO [finetune.py:992] (1/2) Epoch 5, batch 11650, loss[loss=0.2876, simple_loss=0.3405, pruned_loss=0.1174, over 6726.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3168, pruned_loss=0.08299, over 1808353.91 frames. ], batch size: 99, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:46:02,215 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160614.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:46:16,235 INFO [finetune.py:992] (1/2) Epoch 5, batch 11700, loss[loss=0.2341, simple_loss=0.3191, pruned_loss=0.07449, over 10343.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3167, pruned_loss=0.08384, over 1777172.49 frames. ], batch size: 68, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:46:16,863 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.596e+02 3.486e+02 4.193e+02 4.820e+02 1.126e+03, threshold=8.386e+02, percent-clipped=4.0 2023-05-16 03:46:18,701 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 03:46:35,704 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160663.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 03:46:51,084 INFO [finetune.py:992] (1/2) Epoch 5, batch 11750, loss[loss=0.2316, simple_loss=0.3128, pruned_loss=0.07522, over 12296.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3182, pruned_loss=0.08614, over 1731017.91 frames. ], batch size: 34, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:47:25,814 INFO [finetune.py:992] (1/2) Epoch 5, batch 11800, loss[loss=0.2723, simple_loss=0.3312, pruned_loss=0.1066, over 6500.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3209, pruned_loss=0.08778, over 1711260.02 frames. ], batch size: 98, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:47:26,496 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.148e+02 3.837e+02 4.388e+02 5.262e+02 1.354e+03, threshold=8.776e+02, percent-clipped=6.0 2023-05-16 03:48:00,547 INFO [finetune.py:992] (1/2) Epoch 5, batch 11850, loss[loss=0.2597, simple_loss=0.3243, pruned_loss=0.0976, over 7342.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3227, pruned_loss=0.08835, over 1702989.95 frames. ], batch size: 98, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:48:34,801 INFO [finetune.py:992] (1/2) Epoch 5, batch 11900, loss[loss=0.2197, simple_loss=0.3067, pruned_loss=0.06632, over 6912.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3224, pruned_loss=0.0871, over 1696463.33 frames. ], batch size: 98, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:48:35,452 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.711e+02 3.579e+02 4.133e+02 4.663e+02 8.316e+02, threshold=8.266e+02, percent-clipped=0.0 2023-05-16 03:48:53,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 03:49:09,828 INFO [finetune.py:992] (1/2) Epoch 5, batch 11950, loss[loss=0.2432, simple_loss=0.3144, pruned_loss=0.08598, over 7159.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3181, pruned_loss=0.08313, over 1697786.97 frames. ], batch size: 99, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:49:21,564 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160901.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:49:23,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 2023-05-16 03:49:45,477 INFO [finetune.py:992] (1/2) Epoch 5, batch 12000, loss[loss=0.1838, simple_loss=0.2821, pruned_loss=0.04272, over 10122.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3129, pruned_loss=0.07891, over 1702631.86 frames. ], batch size: 69, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:49:45,477 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 03:50:03,324 INFO [finetune.py:1026] (1/2) Epoch 5, validation: loss=0.2936, simple_loss=0.368, pruned_loss=0.1096, over 1020973.00 frames. 2023-05-16 03:50:03,325 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 03:50:03,983 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 3.113e+02 3.392e+02 4.101e+02 1.232e+03, threshold=6.784e+02, percent-clipped=2.0 2023-05-16 03:50:04,128 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8493, 4.5264, 4.2065, 4.2313, 4.5897, 4.0271, 4.2681, 4.0065], device='cuda:1'), covar=tensor([0.1549, 0.0913, 0.0965, 0.1823, 0.0878, 0.2027, 0.1370, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0428, 0.0346, 0.0392, 0.0419, 0.0397, 0.0354, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 03:50:23,038 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160962.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:50:23,612 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160963.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:50:24,623 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 03:50:24,987 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:50:38,411 INFO [finetune.py:992] (1/2) Epoch 5, batch 12050, loss[loss=0.1883, simple_loss=0.2859, pruned_loss=0.04533, over 10061.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3081, pruned_loss=0.07511, over 1710280.71 frames. ], batch size: 68, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:50:56,232 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161011.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 03:51:05,953 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161026.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:51:12,206 INFO [finetune.py:992] (1/2) Epoch 5, batch 12100, loss[loss=0.2286, simple_loss=0.3148, pruned_loss=0.07119, over 12109.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3066, pruned_loss=0.07348, over 1727017.13 frames. ], batch size: 39, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:51:12,854 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.086e+02 2.989e+02 3.607e+02 4.106e+02 1.013e+03, threshold=7.215e+02, percent-clipped=2.0 2023-05-16 03:51:35,731 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3730, 3.2096, 3.2381, 3.4382, 2.6288, 3.2314, 2.6203, 2.9322], device='cuda:1'), covar=tensor([0.1467, 0.0724, 0.0809, 0.0530, 0.0945, 0.0644, 0.1479, 0.0706], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0251, 0.0283, 0.0330, 0.0226, 0.0229, 0.0246, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:51:38,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9654, 2.2246, 2.6476, 3.0650, 2.2063, 3.1229, 3.0703, 3.1565], device='cuda:1'), covar=tensor([0.0174, 0.1082, 0.0427, 0.0169, 0.1159, 0.0219, 0.0292, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0190, 0.0173, 0.0104, 0.0177, 0.0159, 0.0156, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:51:43,670 INFO [finetune.py:992] (1/2) Epoch 5, batch 12150, loss[loss=0.2015, simple_loss=0.2925, pruned_loss=0.05522, over 10393.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3075, pruned_loss=0.07405, over 1727572.03 frames. ], batch size: 68, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:52:14,801 INFO [finetune.py:992] (1/2) Epoch 5, batch 12200, loss[loss=0.2527, simple_loss=0.3178, pruned_loss=0.09381, over 6718.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3087, pruned_loss=0.07555, over 1695702.58 frames. ], batch size: 98, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:52:15,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.269e+02 3.485e+02 4.182e+02 5.020e+02 1.258e+03, threshold=8.364e+02, percent-clipped=6.0 2023-05-16 03:53:00,863 INFO [finetune.py:992] (1/2) Epoch 6, batch 0, loss[loss=0.2562, simple_loss=0.3293, pruned_loss=0.09149, over 12113.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3293, pruned_loss=0.09149, over 12113.00 frames. ], batch size: 38, lr: 4.62e-03, grad_scale: 16.0 2023-05-16 03:53:00,863 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 03:53:08,562 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1428, 2.0359, 3.5004, 4.1289, 3.7962, 4.0608, 3.6862, 2.7635], device='cuda:1'), covar=tensor([0.0036, 0.0450, 0.0132, 0.0031, 0.0078, 0.0065, 0.0091, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0115, 0.0095, 0.0070, 0.0090, 0.0105, 0.0083, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:53:18,207 INFO [finetune.py:1026] (1/2) Epoch 6, validation: loss=0.2892, simple_loss=0.365, pruned_loss=0.1067, over 1020973.00 frames. 2023-05-16 03:53:18,208 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 03:53:40,712 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4862, 4.3590, 3.9301, 4.4559, 3.3413, 4.2221, 2.6036, 4.4546], device='cuda:1'), covar=tensor([0.1457, 0.0567, 0.1352, 0.0836, 0.0958, 0.0479, 0.1782, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0251, 0.0281, 0.0329, 0.0225, 0.0229, 0.0245, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:53:46,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 03:53:52,207 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161216.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:53:54,140 INFO [finetune.py:992] (1/2) Epoch 6, batch 50, loss[loss=0.1975, simple_loss=0.2866, pruned_loss=0.05419, over 12152.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2752, pruned_loss=0.05023, over 537133.53 frames. ], batch size: 34, lr: 4.61e-03, grad_scale: 16.0 2023-05-16 03:54:02,594 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161231.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:54:06,013 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 3.170e+02 3.764e+02 4.622e+02 1.120e+03, threshold=7.528e+02, percent-clipped=1.0 2023-05-16 03:54:16,986 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8521, 3.4411, 5.1236, 2.6288, 2.8355, 3.7791, 3.2725, 3.7812], device='cuda:1'), covar=tensor([0.1028, 0.1163, 0.0361, 0.1378, 0.2047, 0.1676, 0.1449, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0218, 0.0213, 0.0173, 0.0222, 0.0263, 0.0211, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:54:21,719 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161257.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:54:30,231 INFO [finetune.py:992] (1/2) Epoch 6, batch 100, loss[loss=0.1814, simple_loss=0.275, pruned_loss=0.04394, over 12160.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2724, pruned_loss=0.04818, over 942835.08 frames. ], batch size: 34, lr: 4.61e-03, grad_scale: 16.0 2023-05-16 03:54:36,739 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161277.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:54:41,664 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9801, 4.8935, 4.6999, 4.8153, 4.5390, 4.9283, 4.9081, 5.2168], device='cuda:1'), covar=tensor([0.0220, 0.0136, 0.0200, 0.0313, 0.0681, 0.0240, 0.0156, 0.0161], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0164, 0.0163, 0.0204, 0.0206, 0.0179, 0.0148, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 03:54:47,482 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161292.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:54:59,159 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 03:55:05,938 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1068, 5.0201, 4.8280, 4.9151, 4.6514, 5.0099, 4.9885, 5.3323], device='cuda:1'), covar=tensor([0.0252, 0.0141, 0.0195, 0.0318, 0.0754, 0.0303, 0.0168, 0.0143], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0166, 0.0165, 0.0207, 0.0209, 0.0181, 0.0149, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 03:55:06,491 INFO [finetune.py:992] (1/2) Epoch 6, batch 150, loss[loss=0.1527, simple_loss=0.2401, pruned_loss=0.03262, over 12242.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2708, pruned_loss=0.04748, over 1265382.42 frames. ], batch size: 32, lr: 4.61e-03, grad_scale: 16.0 2023-05-16 03:55:08,010 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161321.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:55:08,856 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7471, 2.6633, 3.6164, 4.6690, 3.9644, 4.6704, 3.8835, 3.3241], device='cuda:1'), covar=tensor([0.0021, 0.0380, 0.0134, 0.0031, 0.0117, 0.0054, 0.0120, 0.0303], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0118, 0.0097, 0.0072, 0.0093, 0.0109, 0.0086, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:55:13,296 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9212, 2.3550, 3.5296, 3.0531, 3.4570, 3.0741, 2.2768, 3.4734], device='cuda:1'), covar=tensor([0.0151, 0.0376, 0.0156, 0.0247, 0.0174, 0.0180, 0.0409, 0.0166], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0183, 0.0161, 0.0161, 0.0185, 0.0142, 0.0177, 0.0155], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 03:55:18,798 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.699e+02 3.068e+02 3.800e+02 6.191e+02, threshold=6.136e+02, percent-clipped=0.0 2023-05-16 03:55:42,308 INFO [finetune.py:992] (1/2) Epoch 6, batch 200, loss[loss=0.1688, simple_loss=0.2488, pruned_loss=0.04444, over 12192.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2672, pruned_loss=0.046, over 1517929.75 frames. ], batch size: 31, lr: 4.61e-03, grad_scale: 16.0 2023-05-16 03:55:48,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 03:56:05,485 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161401.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:56:18,788 INFO [finetune.py:992] (1/2) Epoch 6, batch 250, loss[loss=0.1681, simple_loss=0.2618, pruned_loss=0.03719, over 12298.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2674, pruned_loss=0.04665, over 1701525.54 frames. ], batch size: 34, lr: 4.61e-03, grad_scale: 16.0 2023-05-16 03:56:32,102 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 3.211e+02 3.699e+02 4.349e+02 1.199e+03, threshold=7.397e+02, percent-clipped=9.0 2023-05-16 03:56:48,243 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161460.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:56:49,768 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161462.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:56:54,330 INFO [finetune.py:992] (1/2) Epoch 6, batch 300, loss[loss=0.1795, simple_loss=0.2681, pruned_loss=0.04541, over 12265.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2662, pruned_loss=0.04647, over 1846827.37 frames. ], batch size: 37, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 03:57:22,167 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5041, 5.3543, 5.4496, 5.4832, 5.0952, 5.1447, 4.8919, 5.4057], device='cuda:1'), covar=tensor([0.0795, 0.0572, 0.0766, 0.0733, 0.2052, 0.1383, 0.0559, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0605, 0.0525, 0.0551, 0.0735, 0.0658, 0.0492, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:57:29,867 INFO [finetune.py:992] (1/2) Epoch 6, batch 350, loss[loss=0.1802, simple_loss=0.267, pruned_loss=0.04672, over 12311.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.266, pruned_loss=0.04592, over 1969417.13 frames. ], batch size: 34, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 03:57:31,501 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161521.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:57:42,284 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.107e+02 2.879e+02 3.397e+02 4.006e+02 7.575e+02, threshold=6.795e+02, percent-clipped=1.0 2023-05-16 03:57:52,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 03:57:57,073 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161557.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:58:05,464 INFO [finetune.py:992] (1/2) Epoch 6, batch 400, loss[loss=0.168, simple_loss=0.2636, pruned_loss=0.03618, over 12185.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2648, pruned_loss=0.04504, over 2063872.22 frames. ], batch size: 35, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 03:58:08,312 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161572.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:58:18,839 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161587.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:58:31,852 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161605.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:58:41,630 INFO [finetune.py:992] (1/2) Epoch 6, batch 450, loss[loss=0.1524, simple_loss=0.242, pruned_loss=0.03139, over 12094.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2645, pruned_loss=0.04484, over 2127198.57 frames. ], batch size: 32, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 03:58:43,118 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:58:54,246 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.271e+02 2.709e+02 3.206e+02 3.955e+02 8.153e+02, threshold=6.411e+02, percent-clipped=2.0 2023-05-16 03:59:00,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 03:59:14,945 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0850, 5.9235, 5.5978, 5.4082, 6.0508, 5.4313, 5.6487, 5.4914], device='cuda:1'), covar=tensor([0.1663, 0.1052, 0.0978, 0.2012, 0.1025, 0.2259, 0.1611, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0449, 0.0362, 0.0411, 0.0435, 0.0416, 0.0369, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 03:59:17,020 INFO [finetune.py:992] (1/2) Epoch 6, batch 500, loss[loss=0.1932, simple_loss=0.2773, pruned_loss=0.05462, over 10520.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2643, pruned_loss=0.04484, over 2181695.52 frames. ], batch size: 68, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 03:59:17,090 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161669.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 03:59:27,398 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7383, 2.9382, 3.6527, 4.7049, 4.0841, 4.6672, 3.9944, 3.2969], device='cuda:1'), covar=tensor([0.0022, 0.0317, 0.0130, 0.0026, 0.0094, 0.0045, 0.0113, 0.0315], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0119, 0.0098, 0.0073, 0.0094, 0.0110, 0.0087, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 03:59:36,992 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.87 vs. limit=5.0 2023-05-16 03:59:54,166 INFO [finetune.py:992] (1/2) Epoch 6, batch 550, loss[loss=0.2035, simple_loss=0.2953, pruned_loss=0.05589, over 12359.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2648, pruned_loss=0.04469, over 2220176.13 frames. ], batch size: 35, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:00:06,815 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 2.917e+02 3.411e+02 3.994e+02 7.722e+02, threshold=6.823e+02, percent-clipped=2.0 2023-05-16 04:00:20,663 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161757.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:00:24,323 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0983, 5.9966, 5.8060, 5.3746, 5.1357, 5.9662, 5.5939, 5.2475], device='cuda:1'), covar=tensor([0.0672, 0.0970, 0.0668, 0.1538, 0.0644, 0.0728, 0.1546, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0494, 0.0464, 0.0578, 0.0373, 0.0647, 0.0700, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:00:29,156 INFO [finetune.py:992] (1/2) Epoch 6, batch 600, loss[loss=0.1755, simple_loss=0.2636, pruned_loss=0.04367, over 12268.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2646, pruned_loss=0.0445, over 2258971.49 frames. ], batch size: 32, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:00:40,640 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2394, 6.1328, 5.9770, 5.4139, 5.1900, 6.0896, 5.7562, 5.4400], device='cuda:1'), covar=tensor([0.0554, 0.0912, 0.0600, 0.1404, 0.0600, 0.0647, 0.1398, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0493, 0.0464, 0.0577, 0.0373, 0.0646, 0.0700, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:00:59,405 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9651, 4.7428, 5.0000, 5.0613, 4.6989, 5.0012, 4.8180, 2.7704], device='cuda:1'), covar=tensor([0.0102, 0.0084, 0.0090, 0.0058, 0.0057, 0.0086, 0.0139, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0070, 0.0073, 0.0067, 0.0056, 0.0084, 0.0072, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:01:02,228 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161815.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:01:02,846 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161816.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:01:04,943 INFO [finetune.py:992] (1/2) Epoch 6, batch 650, loss[loss=0.1581, simple_loss=0.2548, pruned_loss=0.03065, over 12281.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2639, pruned_loss=0.04432, over 2280020.31 frames. ], batch size: 37, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:01:17,789 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.853e+02 3.256e+02 3.784e+02 6.036e+02, threshold=6.512e+02, percent-clipped=0.0 2023-05-16 04:01:40,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4924, 2.8443, 3.5417, 4.4456, 3.8828, 4.3391, 3.7728, 3.2611], device='cuda:1'), covar=tensor([0.0022, 0.0287, 0.0128, 0.0030, 0.0103, 0.0067, 0.0104, 0.0261], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0119, 0.0099, 0.0073, 0.0094, 0.0110, 0.0087, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:01:41,722 INFO [finetune.py:992] (1/2) Epoch 6, batch 700, loss[loss=0.2165, simple_loss=0.2979, pruned_loss=0.06756, over 12073.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2644, pruned_loss=0.0442, over 2309532.93 frames. ], batch size: 42, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:01:43,932 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161872.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:01:46,847 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161876.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:01:54,766 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161887.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:02:17,230 INFO [finetune.py:992] (1/2) Epoch 6, batch 750, loss[loss=0.1697, simple_loss=0.2593, pruned_loss=0.04003, over 12105.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2631, pruned_loss=0.04367, over 2328049.25 frames. ], batch size: 33, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:02:18,023 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161920.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:02:28,405 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161935.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:02:29,710 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.912e+02 3.415e+02 4.088e+02 6.399e+02, threshold=6.830e+02, percent-clipped=0.0 2023-05-16 04:02:34,248 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9788, 3.6091, 5.2090, 2.7231, 2.8925, 3.7999, 3.4163, 3.7858], device='cuda:1'), covar=tensor([0.0351, 0.1016, 0.0262, 0.1182, 0.1999, 0.1492, 0.1234, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0227, 0.0227, 0.0179, 0.0232, 0.0276, 0.0222, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:02:52,643 INFO [finetune.py:992] (1/2) Epoch 6, batch 800, loss[loss=0.1683, simple_loss=0.2581, pruned_loss=0.03925, over 11582.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2638, pruned_loss=0.04378, over 2338187.53 frames. ], batch size: 48, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:03:12,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1876, 4.5027, 4.1541, 4.8815, 4.5349, 2.9191, 4.2192, 3.0893], device='cuda:1'), covar=tensor([0.0830, 0.0885, 0.1186, 0.0432, 0.0975, 0.1561, 0.0955, 0.3050], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0365, 0.0341, 0.0261, 0.0349, 0.0262, 0.0329, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:03:19,402 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2633, 4.5382, 2.7142, 2.5064, 3.8160, 2.5107, 3.9366, 3.2190], device='cuda:1'), covar=tensor([0.0670, 0.0470, 0.1221, 0.1579, 0.0279, 0.1334, 0.0468, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0242, 0.0171, 0.0195, 0.0134, 0.0179, 0.0188, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:03:32,643 INFO [finetune.py:992] (1/2) Epoch 6, batch 850, loss[loss=0.1913, simple_loss=0.2775, pruned_loss=0.05256, over 12098.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2639, pruned_loss=0.04395, over 2336633.03 frames. ], batch size: 33, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:03:45,569 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 2.955e+02 3.400e+02 4.002e+02 7.780e+02, threshold=6.800e+02, percent-clipped=2.0 2023-05-16 04:03:59,761 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162057.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:04:08,149 INFO [finetune.py:992] (1/2) Epoch 6, batch 900, loss[loss=0.205, simple_loss=0.2813, pruned_loss=0.06432, over 12139.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2633, pruned_loss=0.04369, over 2342216.10 frames. ], batch size: 34, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:04:14,195 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162077.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:04:16,954 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6140, 4.3064, 4.6037, 4.1054, 4.3930, 4.1592, 4.6035, 4.2217], device='cuda:1'), covar=tensor([0.0261, 0.0343, 0.0278, 0.0277, 0.0313, 0.0307, 0.0220, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0238, 0.0253, 0.0234, 0.0228, 0.0230, 0.0209, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:04:33,885 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162105.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:04:41,593 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162116.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:04:43,599 INFO [finetune.py:992] (1/2) Epoch 6, batch 950, loss[loss=0.1865, simple_loss=0.2803, pruned_loss=0.04633, over 12121.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2639, pruned_loss=0.04416, over 2354566.91 frames. ], batch size: 39, lr: 4.61e-03, grad_scale: 8.0 2023-05-16 04:04:54,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 04:04:56,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 2.818e+02 3.634e+02 4.529e+02 2.030e+03, threshold=7.268e+02, percent-clipped=4.0 2023-05-16 04:04:57,971 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162138.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:05:08,619 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2756, 2.9102, 2.8575, 2.7172, 2.4827, 2.3443, 2.8419, 2.0381], device='cuda:1'), covar=tensor([0.0337, 0.0160, 0.0170, 0.0203, 0.0363, 0.0295, 0.0136, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0153, 0.0149, 0.0177, 0.0197, 0.0190, 0.0155, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 04:05:17,098 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162164.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:05:20,603 INFO [finetune.py:992] (1/2) Epoch 6, batch 1000, loss[loss=0.2003, simple_loss=0.2962, pruned_loss=0.05222, over 12066.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.264, pruned_loss=0.04391, over 2361193.89 frames. ], batch size: 42, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:05:22,161 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162171.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:05:29,775 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7758, 4.7303, 4.5895, 4.6686, 4.2842, 4.8285, 4.7650, 5.0253], device='cuda:1'), covar=tensor([0.0221, 0.0154, 0.0217, 0.0355, 0.0826, 0.0367, 0.0167, 0.0176], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0180, 0.0178, 0.0224, 0.0227, 0.0197, 0.0161, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 04:05:29,842 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9158, 2.4836, 3.5568, 2.9001, 3.4222, 3.0489, 2.4098, 3.4419], device='cuda:1'), covar=tensor([0.0118, 0.0312, 0.0135, 0.0237, 0.0112, 0.0161, 0.0328, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0195, 0.0173, 0.0173, 0.0198, 0.0150, 0.0186, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:05:40,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-05-16 04:05:50,624 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7832, 2.3501, 2.9809, 3.6830, 2.1926, 3.8465, 3.6607, 3.8940], device='cuda:1'), covar=tensor([0.0149, 0.1088, 0.0462, 0.0176, 0.1265, 0.0335, 0.0239, 0.0103], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0199, 0.0180, 0.0110, 0.0185, 0.0170, 0.0166, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:05:56,043 INFO [finetune.py:992] (1/2) Epoch 6, batch 1050, loss[loss=0.1567, simple_loss=0.2508, pruned_loss=0.03133, over 12108.00 frames. ], tot_loss[loss=0.175, simple_loss=0.263, pruned_loss=0.04348, over 2370810.54 frames. ], batch size: 33, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:06:04,675 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162231.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:06:08,884 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.802e+02 3.110e+02 3.560e+02 6.634e+02, threshold=6.220e+02, percent-clipped=0.0 2023-05-16 04:06:14,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 04:06:31,900 INFO [finetune.py:992] (1/2) Epoch 6, batch 1100, loss[loss=0.1587, simple_loss=0.25, pruned_loss=0.03374, over 12338.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2624, pruned_loss=0.04328, over 2375484.03 frames. ], batch size: 31, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:06:38,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 04:06:49,915 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162292.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:07:08,838 INFO [finetune.py:992] (1/2) Epoch 6, batch 1150, loss[loss=0.1667, simple_loss=0.2565, pruned_loss=0.0384, over 12191.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2625, pruned_loss=0.04337, over 2374952.26 frames. ], batch size: 31, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:07:22,447 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.089e+02 3.128e+02 3.626e+02 4.185e+02 7.642e+02, threshold=7.251e+02, percent-clipped=5.0 2023-05-16 04:07:44,452 INFO [finetune.py:992] (1/2) Epoch 6, batch 1200, loss[loss=0.1485, simple_loss=0.2333, pruned_loss=0.03187, over 12124.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2629, pruned_loss=0.04327, over 2374335.89 frames. ], batch size: 30, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:08:20,140 INFO [finetune.py:992] (1/2) Epoch 6, batch 1250, loss[loss=0.173, simple_loss=0.2629, pruned_loss=0.04155, over 12116.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2643, pruned_loss=0.04377, over 2380280.50 frames. ], batch size: 45, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:08:21,711 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5539, 2.7295, 3.6695, 4.5297, 3.9803, 4.2941, 3.7248, 3.1248], device='cuda:1'), covar=tensor([0.0025, 0.0321, 0.0124, 0.0025, 0.0100, 0.0078, 0.0116, 0.0307], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0122, 0.0101, 0.0074, 0.0096, 0.0112, 0.0091, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:08:30,653 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162433.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:08:34,119 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.032e+02 2.846e+02 3.349e+02 4.161e+02 7.273e+02, threshold=6.698e+02, percent-clipped=1.0 2023-05-16 04:08:56,533 INFO [finetune.py:992] (1/2) Epoch 6, batch 1300, loss[loss=0.1674, simple_loss=0.2674, pruned_loss=0.03367, over 12152.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.264, pruned_loss=0.04337, over 2384129.58 frames. ], batch size: 36, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:08:58,069 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162471.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:09:00,935 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2823, 4.8853, 5.2476, 4.5668, 4.9466, 4.6679, 5.2856, 5.0126], device='cuda:1'), covar=tensor([0.0222, 0.0313, 0.0260, 0.0269, 0.0336, 0.0293, 0.0228, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0241, 0.0256, 0.0236, 0.0232, 0.0234, 0.0211, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:09:21,817 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4662, 2.3083, 3.0930, 4.2874, 1.9665, 4.4478, 4.3415, 4.5563], device='cuda:1'), covar=tensor([0.0114, 0.1231, 0.0459, 0.0152, 0.1478, 0.0171, 0.0181, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0200, 0.0182, 0.0112, 0.0188, 0.0172, 0.0168, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:09:26,821 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162511.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:09:32,396 INFO [finetune.py:992] (1/2) Epoch 6, batch 1350, loss[loss=0.1434, simple_loss=0.2257, pruned_loss=0.03052, over 12186.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2645, pruned_loss=0.04325, over 2381496.14 frames. ], batch size: 29, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:09:32,462 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162519.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:09:46,039 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 2.606e+02 2.970e+02 3.970e+02 1.108e+03, threshold=5.939e+02, percent-clipped=2.0 2023-05-16 04:10:00,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 04:10:08,428 INFO [finetune.py:992] (1/2) Epoch 6, batch 1400, loss[loss=0.1788, simple_loss=0.273, pruned_loss=0.04228, over 12036.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2637, pruned_loss=0.04297, over 2379002.58 frames. ], batch size: 42, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:10:10,767 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162572.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 04:10:22,056 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162587.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:10:45,565 INFO [finetune.py:992] (1/2) Epoch 6, batch 1450, loss[loss=0.1537, simple_loss=0.2373, pruned_loss=0.03504, over 12350.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2629, pruned_loss=0.04288, over 2385797.85 frames. ], batch size: 31, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:10:59,151 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.825e+02 3.375e+02 3.791e+02 8.443e+02, threshold=6.750e+02, percent-clipped=3.0 2023-05-16 04:11:01,419 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0929, 4.9571, 4.8631, 4.8872, 4.5876, 5.0123, 5.0447, 5.2907], device='cuda:1'), covar=tensor([0.0240, 0.0139, 0.0180, 0.0304, 0.0707, 0.0279, 0.0130, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0183, 0.0181, 0.0228, 0.0229, 0.0200, 0.0163, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 04:11:20,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2023-05-16 04:11:21,162 INFO [finetune.py:992] (1/2) Epoch 6, batch 1500, loss[loss=0.1821, simple_loss=0.2677, pruned_loss=0.0483, over 12045.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2638, pruned_loss=0.04348, over 2385610.68 frames. ], batch size: 42, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:11:39,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3176, 3.3348, 3.2450, 3.0373, 2.7997, 2.6074, 3.4089, 2.1567], device='cuda:1'), covar=tensor([0.0372, 0.0133, 0.0129, 0.0160, 0.0329, 0.0301, 0.0126, 0.0426], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0153, 0.0149, 0.0177, 0.0195, 0.0189, 0.0157, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:11:56,327 INFO [finetune.py:992] (1/2) Epoch 6, batch 1550, loss[loss=0.1528, simple_loss=0.2368, pruned_loss=0.03438, over 12348.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.264, pruned_loss=0.04375, over 2382218.44 frames. ], batch size: 30, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:11:56,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 04:12:00,713 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3806, 3.3369, 3.2430, 3.0108, 2.7847, 2.5703, 3.3742, 2.1018], device='cuda:1'), covar=tensor([0.0351, 0.0145, 0.0131, 0.0172, 0.0343, 0.0325, 0.0126, 0.0458], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0153, 0.0149, 0.0177, 0.0195, 0.0189, 0.0157, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:12:07,112 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162733.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:12:10,538 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.853e+02 3.212e+02 3.842e+02 6.930e+02, threshold=6.425e+02, percent-clipped=1.0 2023-05-16 04:12:33,553 INFO [finetune.py:992] (1/2) Epoch 6, batch 1600, loss[loss=0.1996, simple_loss=0.2854, pruned_loss=0.05697, over 11653.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.263, pruned_loss=0.04364, over 2383466.09 frames. ], batch size: 48, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:12:42,054 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162781.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:12:57,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 04:13:06,785 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1451, 6.0072, 5.5522, 5.4470, 6.0830, 5.1900, 5.7109, 5.5541], device='cuda:1'), covar=tensor([0.1478, 0.0783, 0.0940, 0.1873, 0.0908, 0.2411, 0.1408, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0458, 0.0366, 0.0415, 0.0441, 0.0424, 0.0376, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:13:09,569 INFO [finetune.py:992] (1/2) Epoch 6, batch 1650, loss[loss=0.2073, simple_loss=0.2864, pruned_loss=0.06409, over 12313.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2636, pruned_loss=0.04384, over 2384057.67 frames. ], batch size: 34, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:13:15,389 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162827.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:13:18,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 04:13:23,017 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.994e+02 3.423e+02 4.213e+02 1.099e+03, threshold=6.846e+02, percent-clipped=1.0 2023-05-16 04:13:41,449 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2295, 2.3572, 3.4378, 4.2197, 3.7141, 4.0077, 3.6020, 2.9765], device='cuda:1'), covar=tensor([0.0035, 0.0391, 0.0142, 0.0039, 0.0103, 0.0082, 0.0119, 0.0320], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0122, 0.0101, 0.0075, 0.0097, 0.0112, 0.0090, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:13:44,207 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162867.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 04:13:45,498 INFO [finetune.py:992] (1/2) Epoch 6, batch 1700, loss[loss=0.2441, simple_loss=0.3098, pruned_loss=0.08919, over 7934.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2627, pruned_loss=0.0434, over 2378780.97 frames. ], batch size: 98, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:13:59,067 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162887.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:13:59,884 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162888.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:14:10,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-05-16 04:14:13,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 04:14:17,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 04:14:22,115 INFO [finetune.py:992] (1/2) Epoch 6, batch 1750, loss[loss=0.1711, simple_loss=0.2652, pruned_loss=0.03851, over 12149.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2633, pruned_loss=0.04385, over 2375350.06 frames. ], batch size: 34, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:14:33,501 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162935.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:14:35,534 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.908e+02 3.413e+02 4.052e+02 1.694e+03, threshold=6.827e+02, percent-clipped=2.0 2023-05-16 04:14:57,147 INFO [finetune.py:992] (1/2) Epoch 6, batch 1800, loss[loss=0.1818, simple_loss=0.2782, pruned_loss=0.04267, over 12135.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2642, pruned_loss=0.04419, over 2382648.22 frames. ], batch size: 42, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:15:30,647 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8934, 2.2883, 3.1815, 2.7461, 3.0927, 2.9938, 2.3193, 3.1637], device='cuda:1'), covar=tensor([0.0116, 0.0316, 0.0157, 0.0218, 0.0145, 0.0154, 0.0318, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0195, 0.0174, 0.0175, 0.0198, 0.0152, 0.0186, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:15:32,557 INFO [finetune.py:992] (1/2) Epoch 6, batch 1850, loss[loss=0.1762, simple_loss=0.2689, pruned_loss=0.04179, over 11246.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.264, pruned_loss=0.04379, over 2382760.78 frames. ], batch size: 55, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:15:46,652 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.870e+02 3.284e+02 4.212e+02 1.107e+03, threshold=6.567e+02, percent-clipped=4.0 2023-05-16 04:16:09,552 INFO [finetune.py:992] (1/2) Epoch 6, batch 1900, loss[loss=0.1345, simple_loss=0.2154, pruned_loss=0.02677, over 12289.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2631, pruned_loss=0.04321, over 2392455.23 frames. ], batch size: 28, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:16:24,708 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163090.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:16:45,143 INFO [finetune.py:992] (1/2) Epoch 6, batch 1950, loss[loss=0.1802, simple_loss=0.2582, pruned_loss=0.05107, over 12170.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2629, pruned_loss=0.04318, over 2394731.79 frames. ], batch size: 31, lr: 4.60e-03, grad_scale: 8.0 2023-05-16 04:16:46,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5090, 2.5088, 3.6310, 4.5445, 3.8523, 4.3745, 3.7957, 3.1873], device='cuda:1'), covar=tensor([0.0035, 0.0401, 0.0120, 0.0031, 0.0125, 0.0064, 0.0104, 0.0352], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0122, 0.0102, 0.0076, 0.0099, 0.0114, 0.0091, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:16:58,925 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.301e+02 2.853e+02 3.305e+02 3.872e+02 8.317e+02, threshold=6.610e+02, percent-clipped=2.0 2023-05-16 04:17:08,556 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163151.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:17:19,991 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163167.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 04:17:21,169 INFO [finetune.py:992] (1/2) Epoch 6, batch 2000, loss[loss=0.1512, simple_loss=0.2459, pruned_loss=0.02825, over 12035.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2632, pruned_loss=0.04351, over 2388489.62 frames. ], batch size: 31, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:17:29,876 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9568, 4.3811, 3.9014, 4.8069, 4.4292, 2.7076, 4.1058, 2.9532], device='cuda:1'), covar=tensor([0.0895, 0.0904, 0.1369, 0.0464, 0.1018, 0.1720, 0.0914, 0.3245], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0365, 0.0340, 0.0262, 0.0349, 0.0259, 0.0325, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:17:31,892 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163183.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:17:55,693 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163215.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:17:58,417 INFO [finetune.py:992] (1/2) Epoch 6, batch 2050, loss[loss=0.1919, simple_loss=0.279, pruned_loss=0.05239, over 12189.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2622, pruned_loss=0.04313, over 2387041.45 frames. ], batch size: 35, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:18:11,937 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.203e+02 2.798e+02 3.197e+02 3.643e+02 5.178e+02, threshold=6.394e+02, percent-clipped=0.0 2023-05-16 04:18:12,960 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3667, 2.7218, 3.9524, 3.3219, 3.7669, 3.5234, 2.7457, 3.7185], device='cuda:1'), covar=tensor([0.0093, 0.0275, 0.0085, 0.0166, 0.0090, 0.0115, 0.0280, 0.0098], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0196, 0.0175, 0.0176, 0.0200, 0.0152, 0.0187, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:18:13,680 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163240.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:18:33,534 INFO [finetune.py:992] (1/2) Epoch 6, batch 2100, loss[loss=0.1656, simple_loss=0.2381, pruned_loss=0.04654, over 12353.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2632, pruned_loss=0.04381, over 2381262.27 frames. ], batch size: 31, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:18:56,499 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163301.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:19:08,984 INFO [finetune.py:992] (1/2) Epoch 6, batch 2150, loss[loss=0.1641, simple_loss=0.2464, pruned_loss=0.04093, over 12357.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2623, pruned_loss=0.04348, over 2386249.20 frames. ], batch size: 30, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:19:15,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.25 vs. limit=5.0 2023-05-16 04:19:23,950 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.121e+02 2.741e+02 3.315e+02 4.107e+02 8.163e+02, threshold=6.630e+02, percent-clipped=2.0 2023-05-16 04:19:38,983 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0361, 5.8016, 5.4024, 5.2934, 5.8879, 5.1442, 5.4290, 5.3604], device='cuda:1'), covar=tensor([0.1418, 0.0886, 0.1101, 0.2011, 0.0865, 0.2208, 0.1715, 0.1000], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0460, 0.0363, 0.0416, 0.0440, 0.0420, 0.0377, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:19:46,058 INFO [finetune.py:992] (1/2) Epoch 6, batch 2200, loss[loss=0.1305, simple_loss=0.2157, pruned_loss=0.02258, over 12359.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2612, pruned_loss=0.04278, over 2388066.31 frames. ], batch size: 30, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:20:21,676 INFO [finetune.py:992] (1/2) Epoch 6, batch 2250, loss[loss=0.1808, simple_loss=0.2708, pruned_loss=0.04538, over 12008.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2619, pruned_loss=0.04273, over 2382537.11 frames. ], batch size: 40, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:20:35,266 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.927e+02 3.259e+02 3.882e+02 7.382e+02, threshold=6.518e+02, percent-clipped=2.0 2023-05-16 04:20:41,100 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163446.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:20:41,166 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4368, 5.2599, 5.3713, 5.4531, 5.0232, 5.0618, 4.9208, 5.3559], device='cuda:1'), covar=tensor([0.0589, 0.0520, 0.0687, 0.0540, 0.1652, 0.1258, 0.0499, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0635, 0.0550, 0.0576, 0.0779, 0.0691, 0.0516, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 04:20:57,247 INFO [finetune.py:992] (1/2) Epoch 6, batch 2300, loss[loss=0.1821, simple_loss=0.2629, pruned_loss=0.05066, over 12034.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04255, over 2379256.77 frames. ], batch size: 31, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:21:08,193 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163483.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:21:28,096 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3109, 4.3226, 2.5836, 2.3181, 3.7054, 2.3478, 3.7174, 2.9556], device='cuda:1'), covar=tensor([0.0629, 0.0557, 0.1203, 0.1632, 0.0312, 0.1433, 0.0526, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0246, 0.0173, 0.0196, 0.0136, 0.0179, 0.0192, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:21:34,426 INFO [finetune.py:992] (1/2) Epoch 6, batch 2350, loss[loss=0.1684, simple_loss=0.2524, pruned_loss=0.04216, over 12253.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2628, pruned_loss=0.04343, over 2365793.92 frames. ], batch size: 32, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:21:38,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 04:21:43,133 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163531.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:21:47,863 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.136e+02 2.894e+02 3.393e+02 4.368e+02 1.170e+03, threshold=6.786e+02, percent-clipped=2.0 2023-05-16 04:22:01,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-05-16 04:22:09,893 INFO [finetune.py:992] (1/2) Epoch 6, batch 2400, loss[loss=0.1491, simple_loss=0.2308, pruned_loss=0.03371, over 12294.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2618, pruned_loss=0.04304, over 2373333.09 frames. ], batch size: 28, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:22:29,242 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163596.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:22:46,561 INFO [finetune.py:992] (1/2) Epoch 6, batch 2450, loss[loss=0.1719, simple_loss=0.2629, pruned_loss=0.04044, over 10478.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2611, pruned_loss=0.0427, over 2373481.13 frames. ], batch size: 68, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:23:00,672 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.696e+02 3.263e+02 3.997e+02 1.017e+03, threshold=6.526e+02, percent-clipped=2.0 2023-05-16 04:23:17,714 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5792, 3.6638, 3.4560, 3.2360, 3.0347, 2.7659, 3.8039, 2.3862], device='cuda:1'), covar=tensor([0.0297, 0.0118, 0.0149, 0.0158, 0.0338, 0.0326, 0.0096, 0.0374], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0154, 0.0151, 0.0179, 0.0197, 0.0191, 0.0157, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:23:22,556 INFO [finetune.py:992] (1/2) Epoch 6, batch 2500, loss[loss=0.1858, simple_loss=0.28, pruned_loss=0.04577, over 12141.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2612, pruned_loss=0.04288, over 2377011.78 frames. ], batch size: 36, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:23:58,065 INFO [finetune.py:992] (1/2) Epoch 6, batch 2550, loss[loss=0.1615, simple_loss=0.2496, pruned_loss=0.03671, over 12014.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2613, pruned_loss=0.0432, over 2375991.08 frames. ], batch size: 31, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:24:12,126 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.990e+02 3.429e+02 4.121e+02 9.520e+02, threshold=6.858e+02, percent-clipped=5.0 2023-05-16 04:24:17,290 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163746.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:24:33,964 INFO [finetune.py:992] (1/2) Epoch 6, batch 2600, loss[loss=0.1705, simple_loss=0.2502, pruned_loss=0.04539, over 12367.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2601, pruned_loss=0.04301, over 2380116.86 frames. ], batch size: 30, lr: 4.59e-03, grad_scale: 4.0 2023-05-16 04:24:37,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 04:24:52,472 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163794.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:25:10,841 INFO [finetune.py:992] (1/2) Epoch 6, batch 2650, loss[loss=0.1623, simple_loss=0.2451, pruned_loss=0.0398, over 11805.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2594, pruned_loss=0.04233, over 2386965.66 frames. ], batch size: 26, lr: 4.59e-03, grad_scale: 4.0 2023-05-16 04:25:24,958 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.806e+02 3.186e+02 3.917e+02 6.948e+02, threshold=6.372e+02, percent-clipped=1.0 2023-05-16 04:25:29,594 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3202, 3.2998, 3.2298, 3.5911, 2.6211, 3.2047, 2.6230, 3.0051], device='cuda:1'), covar=tensor([0.1358, 0.0718, 0.0921, 0.0691, 0.0891, 0.0644, 0.1377, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0264, 0.0298, 0.0352, 0.0235, 0.0238, 0.0256, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:25:29,639 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7539, 2.4895, 3.5852, 3.6632, 2.7407, 2.6415, 2.6322, 2.2997], device='cuda:1'), covar=tensor([0.1214, 0.2565, 0.0586, 0.0500, 0.0962, 0.1974, 0.2203, 0.3389], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0370, 0.0258, 0.0288, 0.0252, 0.0284, 0.0354, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:25:46,445 INFO [finetune.py:992] (1/2) Epoch 6, batch 2700, loss[loss=0.1565, simple_loss=0.2433, pruned_loss=0.03491, over 12164.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2603, pruned_loss=0.04295, over 2367209.51 frames. ], batch size: 29, lr: 4.59e-03, grad_scale: 4.0 2023-05-16 04:26:05,308 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163896.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:26:11,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 04:26:22,186 INFO [finetune.py:992] (1/2) Epoch 6, batch 2750, loss[loss=0.185, simple_loss=0.2844, pruned_loss=0.04284, over 12026.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2608, pruned_loss=0.0427, over 2368066.26 frames. ], batch size: 40, lr: 4.59e-03, grad_scale: 4.0 2023-05-16 04:26:28,859 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9373, 4.6859, 4.7351, 4.8595, 4.6736, 4.9148, 4.7330, 2.6583], device='cuda:1'), covar=tensor([0.0088, 0.0058, 0.0090, 0.0055, 0.0049, 0.0080, 0.0077, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0076, 0.0070, 0.0058, 0.0088, 0.0076, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:26:37,201 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.057e+02 2.839e+02 3.207e+02 3.708e+02 7.990e+02, threshold=6.415e+02, percent-clipped=1.0 2023-05-16 04:26:40,776 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163944.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:26:58,611 INFO [finetune.py:992] (1/2) Epoch 6, batch 2800, loss[loss=0.1625, simple_loss=0.2389, pruned_loss=0.04309, over 12285.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.26, pruned_loss=0.04261, over 2357373.08 frames. ], batch size: 28, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:27:24,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 04:27:36,863 INFO [finetune.py:992] (1/2) Epoch 6, batch 2850, loss[loss=0.1782, simple_loss=0.2758, pruned_loss=0.04032, over 12154.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.26, pruned_loss=0.04272, over 2363323.16 frames. ], batch size: 34, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:27:45,538 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5358, 2.2542, 3.1815, 4.3010, 2.2143, 4.3763, 4.4039, 4.6167], device='cuda:1'), covar=tensor([0.0114, 0.1434, 0.0458, 0.0148, 0.1335, 0.0241, 0.0161, 0.0069], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0201, 0.0182, 0.0113, 0.0186, 0.0173, 0.0169, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:27:50,947 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 2.814e+02 3.240e+02 3.538e+02 6.135e+02, threshold=6.480e+02, percent-clipped=0.0 2023-05-16 04:28:12,665 INFO [finetune.py:992] (1/2) Epoch 6, batch 2900, loss[loss=0.1863, simple_loss=0.2827, pruned_loss=0.04491, over 10463.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2599, pruned_loss=0.04244, over 2362232.85 frames. ], batch size: 69, lr: 4.59e-03, grad_scale: 8.0 2023-05-16 04:28:22,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-16 04:28:27,797 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6517, 2.2850, 2.8898, 2.5910, 2.8638, 2.8263, 2.2000, 2.9103], device='cuda:1'), covar=tensor([0.0105, 0.0234, 0.0143, 0.0196, 0.0139, 0.0140, 0.0273, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0193, 0.0177, 0.0174, 0.0199, 0.0151, 0.0185, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:28:28,442 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164090.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:28:39,205 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7114, 2.6659, 4.1134, 4.2972, 2.8631, 2.6226, 2.7759, 2.1309], device='cuda:1'), covar=tensor([0.1333, 0.2816, 0.0564, 0.0436, 0.1128, 0.2008, 0.2475, 0.3824], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0371, 0.0259, 0.0290, 0.0254, 0.0284, 0.0354, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:28:46,383 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3257, 2.6937, 3.9269, 3.2075, 3.7105, 3.3969, 2.6895, 3.6894], device='cuda:1'), covar=tensor([0.0144, 0.0283, 0.0086, 0.0213, 0.0130, 0.0149, 0.0340, 0.0113], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0192, 0.0175, 0.0173, 0.0198, 0.0150, 0.0184, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:28:48,993 INFO [finetune.py:992] (1/2) Epoch 6, batch 2950, loss[loss=0.1739, simple_loss=0.2643, pruned_loss=0.04175, over 12110.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2595, pruned_loss=0.0424, over 2364406.96 frames. ], batch size: 38, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:29:03,528 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.731e+02 3.279e+02 3.824e+02 1.716e+03, threshold=6.558e+02, percent-clipped=4.0 2023-05-16 04:29:12,404 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164151.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:29:15,174 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164155.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:29:24,948 INFO [finetune.py:992] (1/2) Epoch 6, batch 3000, loss[loss=0.1609, simple_loss=0.2538, pruned_loss=0.03396, over 12252.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2593, pruned_loss=0.04232, over 2369473.17 frames. ], batch size: 37, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:29:24,948 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 04:29:43,358 INFO [finetune.py:1026] (1/2) Epoch 6, validation: loss=0.3222, simple_loss=0.3994, pruned_loss=0.1225, over 1020973.00 frames. 2023-05-16 04:29:43,359 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 04:29:45,080 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7098, 2.7321, 4.5910, 4.7710, 2.6766, 2.6784, 2.8725, 2.1442], device='cuda:1'), covar=tensor([0.1473, 0.3076, 0.0427, 0.0373, 0.1356, 0.2191, 0.2573, 0.3832], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0370, 0.0259, 0.0289, 0.0253, 0.0283, 0.0352, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:29:52,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 04:30:08,606 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6253, 2.5300, 4.4484, 4.7115, 3.0797, 2.5208, 2.6242, 1.9394], device='cuda:1'), covar=tensor([0.1354, 0.3288, 0.0422, 0.0315, 0.0948, 0.2058, 0.2850, 0.4125], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0371, 0.0259, 0.0288, 0.0253, 0.0283, 0.0352, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:30:17,637 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164216.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:30:19,520 INFO [finetune.py:992] (1/2) Epoch 6, batch 3050, loss[loss=0.1795, simple_loss=0.266, pruned_loss=0.04655, over 12150.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2601, pruned_loss=0.04254, over 2371197.51 frames. ], batch size: 39, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:30:25,621 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0976, 4.7427, 4.8459, 4.9597, 4.7035, 5.0561, 4.8612, 2.7802], device='cuda:1'), covar=tensor([0.0081, 0.0059, 0.0080, 0.0066, 0.0049, 0.0069, 0.0063, 0.0645], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0073, 0.0076, 0.0070, 0.0058, 0.0087, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:30:32,649 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0323, 4.7250, 4.7982, 4.8881, 4.6644, 5.0190, 4.7805, 2.5108], device='cuda:1'), covar=tensor([0.0101, 0.0056, 0.0084, 0.0067, 0.0048, 0.0072, 0.0062, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0073, 0.0076, 0.0070, 0.0058, 0.0087, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:30:33,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.794e+02 3.284e+02 3.936e+02 7.141e+02, threshold=6.567e+02, percent-clipped=3.0 2023-05-16 04:30:55,522 INFO [finetune.py:992] (1/2) Epoch 6, batch 3100, loss[loss=0.1687, simple_loss=0.2563, pruned_loss=0.04051, over 12305.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2592, pruned_loss=0.04209, over 2377006.14 frames. ], batch size: 34, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:31:31,734 INFO [finetune.py:992] (1/2) Epoch 6, batch 3150, loss[loss=0.1927, simple_loss=0.2705, pruned_loss=0.05741, over 12120.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2596, pruned_loss=0.04271, over 2375932.41 frames. ], batch size: 38, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:31:46,570 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.672e+02 3.320e+02 3.847e+02 1.508e+03, threshold=6.640e+02, percent-clipped=5.0 2023-05-16 04:32:01,539 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164360.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:32:07,788 INFO [finetune.py:992] (1/2) Epoch 6, batch 3200, loss[loss=0.1494, simple_loss=0.2384, pruned_loss=0.0302, over 12353.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2596, pruned_loss=0.04259, over 2376974.98 frames. ], batch size: 31, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:32:43,314 INFO [finetune.py:992] (1/2) Epoch 6, batch 3250, loss[loss=0.1319, simple_loss=0.2165, pruned_loss=0.02365, over 12271.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2598, pruned_loss=0.04235, over 2378294.76 frames. ], batch size: 28, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:32:45,022 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164421.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:32:58,252 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.220e+02 2.909e+02 3.393e+02 4.090e+02 1.179e+03, threshold=6.786e+02, percent-clipped=4.0 2023-05-16 04:33:03,373 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164446.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:33:09,634 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5764, 2.5411, 3.1304, 4.3272, 2.3186, 4.4459, 4.4975, 4.6701], device='cuda:1'), covar=tensor([0.0106, 0.1229, 0.0476, 0.0172, 0.1310, 0.0193, 0.0140, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0203, 0.0184, 0.0114, 0.0188, 0.0175, 0.0171, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:33:11,065 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9697, 4.6656, 4.7374, 4.9034, 4.6695, 4.9258, 4.7129, 2.7116], device='cuda:1'), covar=tensor([0.0082, 0.0067, 0.0090, 0.0056, 0.0051, 0.0076, 0.0121, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0073, 0.0076, 0.0069, 0.0058, 0.0087, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:33:20,169 INFO [finetune.py:992] (1/2) Epoch 6, batch 3300, loss[loss=0.188, simple_loss=0.2794, pruned_loss=0.04831, over 12054.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2603, pruned_loss=0.04257, over 2377990.00 frames. ], batch size: 42, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:33:50,144 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164511.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:33:55,505 INFO [finetune.py:992] (1/2) Epoch 6, batch 3350, loss[loss=0.2014, simple_loss=0.2731, pruned_loss=0.06491, over 7864.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2601, pruned_loss=0.04253, over 2372986.03 frames. ], batch size: 98, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:34:09,192 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.787e+02 3.205e+02 4.035e+02 6.481e+02, threshold=6.410e+02, percent-clipped=0.0 2023-05-16 04:34:30,149 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164568.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:34:30,649 INFO [finetune.py:992] (1/2) Epoch 6, batch 3400, loss[loss=0.1915, simple_loss=0.2779, pruned_loss=0.05262, over 12145.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2612, pruned_loss=0.0431, over 2377693.55 frames. ], batch size: 38, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:34:37,184 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164578.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:35:07,978 INFO [finetune.py:992] (1/2) Epoch 6, batch 3450, loss[loss=0.1598, simple_loss=0.2405, pruned_loss=0.0395, over 11351.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2609, pruned_loss=0.04321, over 2375238.12 frames. ], batch size: 25, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:35:15,281 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164629.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:35:22,009 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.736e+02 3.192e+02 3.839e+02 9.389e+02, threshold=6.385e+02, percent-clipped=1.0 2023-05-16 04:35:22,233 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164639.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 04:35:35,919 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164658.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:35:41,636 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4345, 2.3338, 3.1764, 4.2624, 2.2549, 4.3150, 4.3525, 4.5679], device='cuda:1'), covar=tensor([0.0131, 0.1332, 0.0468, 0.0147, 0.1284, 0.0216, 0.0144, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0204, 0.0185, 0.0114, 0.0188, 0.0176, 0.0172, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:35:43,532 INFO [finetune.py:992] (1/2) Epoch 6, batch 3500, loss[loss=0.1879, simple_loss=0.2753, pruned_loss=0.05025, over 12146.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2607, pruned_loss=0.04284, over 2375826.65 frames. ], batch size: 34, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:35:51,632 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2136, 3.2699, 3.1612, 3.0375, 2.7266, 2.5704, 3.3876, 2.1192], device='cuda:1'), covar=tensor([0.0379, 0.0132, 0.0169, 0.0182, 0.0343, 0.0344, 0.0140, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0153, 0.0151, 0.0178, 0.0198, 0.0190, 0.0158, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:36:01,006 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4929, 4.8902, 3.0950, 2.5615, 4.2887, 2.4711, 4.1871, 3.5688], device='cuda:1'), covar=tensor([0.0557, 0.0379, 0.0910, 0.1509, 0.0238, 0.1395, 0.0378, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0243, 0.0171, 0.0193, 0.0133, 0.0175, 0.0189, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:36:16,920 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164716.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:36:18,932 INFO [finetune.py:992] (1/2) Epoch 6, batch 3550, loss[loss=0.1601, simple_loss=0.2367, pruned_loss=0.04178, over 12337.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2598, pruned_loss=0.04245, over 2382447.63 frames. ], batch size: 30, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:36:19,126 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164719.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:36:24,860 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 04:36:29,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 04:36:33,856 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 2.920e+02 3.415e+02 3.980e+02 6.979e+02, threshold=6.831e+02, percent-clipped=1.0 2023-05-16 04:36:36,807 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2156, 5.0308, 5.1692, 5.2191, 4.8188, 4.8941, 4.6537, 5.1780], device='cuda:1'), covar=tensor([0.0603, 0.0564, 0.0801, 0.0482, 0.1749, 0.1198, 0.0574, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0648, 0.0556, 0.0587, 0.0797, 0.0702, 0.0525, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 04:36:38,979 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164746.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:36:39,000 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164746.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:36:42,167 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 04:36:52,590 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2311, 5.0793, 5.1764, 5.2287, 4.8658, 4.9375, 4.7484, 5.1619], device='cuda:1'), covar=tensor([0.0579, 0.0534, 0.0627, 0.0476, 0.1511, 0.1052, 0.0503, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0649, 0.0556, 0.0588, 0.0798, 0.0702, 0.0525, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 04:36:56,022 INFO [finetune.py:992] (1/2) Epoch 6, batch 3600, loss[loss=0.1636, simple_loss=0.2493, pruned_loss=0.03894, over 12036.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2591, pruned_loss=0.04219, over 2378570.12 frames. ], batch size: 31, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:37:02,572 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5614, 4.2744, 4.4876, 4.0160, 4.3216, 3.9735, 4.4582, 4.3228], device='cuda:1'), covar=tensor([0.0375, 0.0450, 0.0542, 0.0309, 0.0347, 0.0425, 0.0411, 0.0635], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0251, 0.0268, 0.0245, 0.0243, 0.0244, 0.0220, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:37:13,970 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164794.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:37:17,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.32 vs. limit=5.0 2023-05-16 04:37:23,582 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164807.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:37:26,292 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164811.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:37:31,884 INFO [finetune.py:992] (1/2) Epoch 6, batch 3650, loss[loss=0.1511, simple_loss=0.2334, pruned_loss=0.03446, over 12353.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.26, pruned_loss=0.04218, over 2382722.92 frames. ], batch size: 31, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:37:45,910 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.801e+02 3.356e+02 4.004e+02 7.977e+02, threshold=6.711e+02, percent-clipped=2.0 2023-05-16 04:37:59,681 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3815, 4.9894, 5.3691, 4.6818, 4.9556, 4.7615, 5.3770, 5.0589], device='cuda:1'), covar=tensor([0.0276, 0.0351, 0.0300, 0.0246, 0.0344, 0.0322, 0.0252, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0252, 0.0269, 0.0246, 0.0244, 0.0245, 0.0220, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:38:00,265 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164859.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:38:07,856 INFO [finetune.py:992] (1/2) Epoch 6, batch 3700, loss[loss=0.1662, simple_loss=0.2538, pruned_loss=0.03931, over 12195.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2606, pruned_loss=0.04232, over 2378908.27 frames. ], batch size: 31, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:38:44,103 INFO [finetune.py:992] (1/2) Epoch 6, batch 3750, loss[loss=0.1905, simple_loss=0.2829, pruned_loss=0.04907, over 10488.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2619, pruned_loss=0.04288, over 2375695.95 frames. ], batch size: 68, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:38:47,739 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164924.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:38:54,681 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164934.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 04:38:58,188 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.737e+02 3.424e+02 3.917e+02 2.058e+03, threshold=6.847e+02, percent-clipped=1.0 2023-05-16 04:39:19,911 INFO [finetune.py:992] (1/2) Epoch 6, batch 3800, loss[loss=0.1831, simple_loss=0.2721, pruned_loss=0.04707, over 10503.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.261, pruned_loss=0.04278, over 2376048.58 frames. ], batch size: 68, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:39:37,398 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7592, 5.8140, 5.5222, 5.0794, 5.0700, 5.6984, 5.2619, 5.0169], device='cuda:1'), covar=tensor([0.0824, 0.0736, 0.0648, 0.1763, 0.0692, 0.0765, 0.1730, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0583, 0.0515, 0.0482, 0.0599, 0.0389, 0.0678, 0.0736, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:39:41,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9786, 4.1621, 3.8104, 4.4527, 4.1876, 2.8188, 3.9375, 3.0636], device='cuda:1'), covar=tensor([0.0783, 0.0839, 0.1422, 0.0587, 0.1004, 0.1480, 0.1014, 0.2797], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0369, 0.0348, 0.0268, 0.0356, 0.0263, 0.0331, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:39:52,518 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165014.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:39:53,788 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165016.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:39:55,859 INFO [finetune.py:992] (1/2) Epoch 6, batch 3850, loss[loss=0.1666, simple_loss=0.2461, pruned_loss=0.04358, over 12015.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.261, pruned_loss=0.04282, over 2380043.55 frames. ], batch size: 31, lr: 4.58e-03, grad_scale: 8.0 2023-05-16 04:40:09,948 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 2.707e+02 3.274e+02 3.886e+02 1.094e+03, threshold=6.548e+02, percent-clipped=4.0 2023-05-16 04:40:19,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.86 vs. limit=5.0 2023-05-16 04:40:28,678 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165064.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:40:32,077 INFO [finetune.py:992] (1/2) Epoch 6, batch 3900, loss[loss=0.1513, simple_loss=0.2369, pruned_loss=0.03284, over 12358.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2619, pruned_loss=0.04318, over 2367485.77 frames. ], batch size: 31, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:40:55,686 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165102.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:41:07,760 INFO [finetune.py:992] (1/2) Epoch 6, batch 3950, loss[loss=0.1453, simple_loss=0.2322, pruned_loss=0.02919, over 12366.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2623, pruned_loss=0.04341, over 2366295.21 frames. ], batch size: 31, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:41:10,841 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165123.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:41:21,831 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.843e+02 2.809e+02 3.386e+02 4.084e+02 8.406e+02, threshold=6.772e+02, percent-clipped=2.0 2023-05-16 04:41:43,513 INFO [finetune.py:992] (1/2) Epoch 6, batch 4000, loss[loss=0.1632, simple_loss=0.2551, pruned_loss=0.03561, over 12357.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2617, pruned_loss=0.04295, over 2368811.04 frames. ], batch size: 36, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:41:54,472 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165184.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:42:04,604 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6558, 2.9495, 4.6911, 4.8179, 2.8766, 2.6548, 2.8707, 2.0874], device='cuda:1'), covar=tensor([0.1401, 0.2500, 0.0355, 0.0341, 0.1192, 0.2021, 0.2602, 0.3665], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0373, 0.0261, 0.0291, 0.0255, 0.0285, 0.0355, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:42:15,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-05-16 04:42:20,389 INFO [finetune.py:992] (1/2) Epoch 6, batch 4050, loss[loss=0.1665, simple_loss=0.2541, pruned_loss=0.03943, over 12092.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2615, pruned_loss=0.04269, over 2374671.45 frames. ], batch size: 32, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:42:22,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2186, 5.1782, 5.0747, 5.0896, 4.6969, 5.1608, 5.0997, 5.3699], device='cuda:1'), covar=tensor([0.0207, 0.0140, 0.0174, 0.0310, 0.0729, 0.0333, 0.0158, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0184, 0.0186, 0.0232, 0.0234, 0.0204, 0.0164, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 04:42:24,046 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165224.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:42:31,072 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165234.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 04:42:32,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2147, 2.4488, 3.0856, 4.0954, 2.1877, 4.1697, 4.0534, 4.3175], device='cuda:1'), covar=tensor([0.0129, 0.1183, 0.0439, 0.0123, 0.1243, 0.0225, 0.0180, 0.0090], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0201, 0.0183, 0.0113, 0.0187, 0.0175, 0.0170, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:42:34,443 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 2.876e+02 3.378e+02 4.022e+02 7.072e+02, threshold=6.755e+02, percent-clipped=1.0 2023-05-16 04:42:41,130 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4842, 4.8094, 4.2856, 4.9939, 4.7739, 3.0221, 4.5146, 3.1322], device='cuda:1'), covar=tensor([0.0649, 0.0631, 0.1303, 0.0436, 0.0867, 0.1438, 0.0867, 0.3217], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0369, 0.0348, 0.0269, 0.0357, 0.0265, 0.0330, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:42:55,895 INFO [finetune.py:992] (1/2) Epoch 6, batch 4100, loss[loss=0.1603, simple_loss=0.2531, pruned_loss=0.03377, over 12284.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2618, pruned_loss=0.04256, over 2366981.49 frames. ], batch size: 33, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:42:58,191 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165272.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:43:05,113 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:43:19,425 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7027, 2.3891, 3.7279, 4.7049, 4.1503, 4.5708, 3.9831, 3.4172], device='cuda:1'), covar=tensor([0.0036, 0.0459, 0.0138, 0.0027, 0.0100, 0.0079, 0.0111, 0.0297], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0123, 0.0102, 0.0077, 0.0098, 0.0114, 0.0091, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:43:28,568 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165314.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:43:32,087 INFO [finetune.py:992] (1/2) Epoch 6, batch 4150, loss[loss=0.2321, simple_loss=0.3075, pruned_loss=0.07829, over 8301.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2621, pruned_loss=0.04295, over 2367972.03 frames. ], batch size: 98, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:43:32,926 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0552, 6.0716, 5.8165, 5.2935, 5.1002, 5.9433, 5.5197, 5.3182], device='cuda:1'), covar=tensor([0.0582, 0.0703, 0.0635, 0.1425, 0.0630, 0.0682, 0.1356, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0579, 0.0512, 0.0482, 0.0598, 0.0388, 0.0675, 0.0727, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:43:45,794 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.822e+02 3.342e+02 4.030e+02 7.682e+02, threshold=6.684e+02, percent-clipped=1.0 2023-05-16 04:44:02,661 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165362.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:44:07,638 INFO [finetune.py:992] (1/2) Epoch 6, batch 4200, loss[loss=0.1531, simple_loss=0.2278, pruned_loss=0.03921, over 12010.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2613, pruned_loss=0.04269, over 2370045.87 frames. ], batch size: 28, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:44:19,994 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0175, 4.6956, 4.8124, 4.9025, 4.6896, 4.9829, 4.8501, 2.7360], device='cuda:1'), covar=tensor([0.0089, 0.0060, 0.0084, 0.0056, 0.0043, 0.0069, 0.0062, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0077, 0.0071, 0.0058, 0.0089, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:44:31,802 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165402.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:44:42,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.08 vs. limit=5.0 2023-05-16 04:44:43,708 INFO [finetune.py:992] (1/2) Epoch 6, batch 4250, loss[loss=0.176, simple_loss=0.2658, pruned_loss=0.04316, over 12108.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2615, pruned_loss=0.04294, over 2370484.05 frames. ], batch size: 33, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:44:58,495 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.871e+02 3.395e+02 3.960e+02 1.167e+03, threshold=6.791e+02, percent-clipped=2.0 2023-05-16 04:45:06,276 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165450.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:45:08,645 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6142, 4.0874, 4.1435, 4.4850, 3.1932, 3.9048, 2.4888, 4.1921], device='cuda:1'), covar=tensor([0.1483, 0.0851, 0.1119, 0.0935, 0.1141, 0.0715, 0.2085, 0.1351], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0264, 0.0296, 0.0356, 0.0236, 0.0239, 0.0257, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:45:12,297 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8031, 3.4037, 5.1849, 2.6286, 2.8788, 3.9519, 3.3224, 4.0205], device='cuda:1'), covar=tensor([0.0505, 0.1097, 0.0256, 0.1250, 0.1882, 0.1279, 0.1397, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0230, 0.0234, 0.0180, 0.0236, 0.0283, 0.0225, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:45:19,592 INFO [finetune.py:992] (1/2) Epoch 6, batch 4300, loss[loss=0.1835, simple_loss=0.2822, pruned_loss=0.04244, over 11101.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2618, pruned_loss=0.04304, over 2377386.69 frames. ], batch size: 55, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:45:26,824 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:45:55,520 INFO [finetune.py:992] (1/2) Epoch 6, batch 4350, loss[loss=0.1336, simple_loss=0.2137, pruned_loss=0.02671, over 11758.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2617, pruned_loss=0.04295, over 2369866.06 frames. ], batch size: 26, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:46:06,295 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3170, 4.5865, 4.0228, 4.9148, 4.5307, 2.4443, 4.0272, 2.9750], device='cuda:1'), covar=tensor([0.0663, 0.0724, 0.1268, 0.0440, 0.1004, 0.1868, 0.1142, 0.3241], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0371, 0.0350, 0.0271, 0.0360, 0.0267, 0.0333, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:46:07,721 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2704, 4.7345, 3.0741, 2.4827, 4.0958, 2.5013, 4.0734, 3.4863], device='cuda:1'), covar=tensor([0.0712, 0.0654, 0.0975, 0.1657, 0.0320, 0.1348, 0.0453, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0249, 0.0173, 0.0198, 0.0136, 0.0179, 0.0194, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:46:09,535 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.872e+02 3.335e+02 4.070e+02 1.129e+03, threshold=6.671e+02, percent-clipped=3.0 2023-05-16 04:46:18,313 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165551.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:46:31,188 INFO [finetune.py:992] (1/2) Epoch 6, batch 4400, loss[loss=0.174, simple_loss=0.2694, pruned_loss=0.03927, over 12148.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2616, pruned_loss=0.0431, over 2371482.88 frames. ], batch size: 34, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:47:02,789 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165612.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:47:07,534 INFO [finetune.py:992] (1/2) Epoch 6, batch 4450, loss[loss=0.194, simple_loss=0.2836, pruned_loss=0.0522, over 11894.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2622, pruned_loss=0.04327, over 2372584.40 frames. ], batch size: 44, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:47:08,402 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5472, 5.0158, 5.4829, 4.7890, 5.0479, 4.9002, 5.5259, 5.1006], device='cuda:1'), covar=tensor([0.0211, 0.0360, 0.0219, 0.0248, 0.0322, 0.0284, 0.0188, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0250, 0.0269, 0.0245, 0.0243, 0.0242, 0.0219, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:47:16,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4912, 3.5442, 3.2889, 3.2208, 2.8935, 2.6799, 3.6830, 2.2382], device='cuda:1'), covar=tensor([0.0344, 0.0137, 0.0175, 0.0167, 0.0383, 0.0337, 0.0134, 0.0462], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0154, 0.0152, 0.0180, 0.0199, 0.0192, 0.0159, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:47:21,527 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.094e+02 2.699e+02 3.249e+02 3.797e+02 5.988e+02, threshold=6.497e+02, percent-clipped=0.0 2023-05-16 04:47:43,179 INFO [finetune.py:992] (1/2) Epoch 6, batch 4500, loss[loss=0.1727, simple_loss=0.2625, pruned_loss=0.04145, over 10551.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.262, pruned_loss=0.04336, over 2375531.79 frames. ], batch size: 69, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:47:47,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 04:48:06,970 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0037, 5.9480, 5.7801, 5.3068, 5.1491, 5.9181, 5.4862, 5.3191], device='cuda:1'), covar=tensor([0.0660, 0.0840, 0.0620, 0.1443, 0.0688, 0.0711, 0.1508, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0515, 0.0488, 0.0600, 0.0392, 0.0685, 0.0737, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:48:18,249 INFO [finetune.py:992] (1/2) Epoch 6, batch 4550, loss[loss=0.2136, simple_loss=0.297, pruned_loss=0.06514, over 12284.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2617, pruned_loss=0.04326, over 2378058.20 frames. ], batch size: 37, lr: 4.57e-03, grad_scale: 8.0 2023-05-16 04:48:33,098 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.258e+02 2.965e+02 3.483e+02 4.253e+02 9.406e+02, threshold=6.967e+02, percent-clipped=5.0 2023-05-16 04:48:36,241 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6643, 2.8994, 4.3555, 4.6184, 2.8926, 2.6961, 2.8765, 2.0299], device='cuda:1'), covar=tensor([0.1352, 0.2671, 0.0497, 0.0356, 0.1125, 0.1960, 0.2492, 0.3910], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0373, 0.0262, 0.0290, 0.0255, 0.0285, 0.0356, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:48:41,190 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165750.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:48:54,410 INFO [finetune.py:992] (1/2) Epoch 6, batch 4600, loss[loss=0.2187, simple_loss=0.2976, pruned_loss=0.06986, over 12069.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2626, pruned_loss=0.0434, over 2380454.17 frames. ], batch size: 42, lr: 4.57e-03, grad_scale: 16.0 2023-05-16 04:49:00,165 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3567, 4.1147, 4.2390, 4.2048, 4.0930, 4.3613, 4.1786, 2.4910], device='cuda:1'), covar=tensor([0.0096, 0.0076, 0.0090, 0.0076, 0.0060, 0.0091, 0.0097, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0078, 0.0071, 0.0059, 0.0088, 0.0077, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:49:01,545 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165779.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:49:19,466 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165803.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:49:25,294 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165811.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:49:30,745 INFO [finetune.py:992] (1/2) Epoch 6, batch 4650, loss[loss=0.1938, simple_loss=0.2771, pruned_loss=0.05527, over 11086.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2625, pruned_loss=0.04355, over 2370858.66 frames. ], batch size: 55, lr: 4.57e-03, grad_scale: 16.0 2023-05-16 04:49:31,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 04:49:36,496 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165827.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:49:36,572 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1398, 6.1251, 5.9366, 5.4023, 5.2891, 6.0360, 5.6196, 5.3550], device='cuda:1'), covar=tensor([0.0612, 0.0716, 0.0542, 0.1393, 0.0595, 0.0632, 0.1340, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0513, 0.0487, 0.0599, 0.0391, 0.0682, 0.0736, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:49:44,550 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.807e+02 3.354e+02 4.024e+02 1.121e+03, threshold=6.709e+02, percent-clipped=1.0 2023-05-16 04:50:02,746 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165864.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:50:06,664 INFO [finetune.py:992] (1/2) Epoch 6, batch 4700, loss[loss=0.1572, simple_loss=0.2393, pruned_loss=0.03758, over 12240.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.262, pruned_loss=0.04321, over 2373897.74 frames. ], batch size: 28, lr: 4.57e-03, grad_scale: 16.0 2023-05-16 04:50:06,829 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8496, 4.4391, 4.8197, 4.2025, 4.4862, 4.1947, 4.8250, 4.4642], device='cuda:1'), covar=tensor([0.0288, 0.0442, 0.0308, 0.0271, 0.0343, 0.0379, 0.0247, 0.0414], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0251, 0.0268, 0.0246, 0.0243, 0.0244, 0.0220, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:50:33,713 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165907.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:50:42,177 INFO [finetune.py:992] (1/2) Epoch 6, batch 4750, loss[loss=0.1745, simple_loss=0.2651, pruned_loss=0.04197, over 12299.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2621, pruned_loss=0.04341, over 2368712.19 frames. ], batch size: 34, lr: 4.57e-03, grad_scale: 16.0 2023-05-16 04:50:56,972 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.971e+02 3.330e+02 3.952e+02 9.689e+02, threshold=6.660e+02, percent-clipped=4.0 2023-05-16 04:51:11,500 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4217, 4.7449, 2.8975, 2.8233, 4.0986, 2.5094, 3.9461, 3.5835], device='cuda:1'), covar=tensor([0.0548, 0.0469, 0.0966, 0.1226, 0.0260, 0.1254, 0.0402, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0250, 0.0175, 0.0198, 0.0137, 0.0180, 0.0194, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:51:15,709 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4165, 3.5772, 3.2226, 3.1036, 2.9178, 2.6436, 3.6314, 2.2340], device='cuda:1'), covar=tensor([0.0356, 0.0116, 0.0166, 0.0180, 0.0346, 0.0350, 0.0120, 0.0451], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0154, 0.0153, 0.0181, 0.0198, 0.0191, 0.0159, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:51:18,402 INFO [finetune.py:992] (1/2) Epoch 6, batch 4800, loss[loss=0.1862, simple_loss=0.2753, pruned_loss=0.0485, over 12112.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2624, pruned_loss=0.04367, over 2372000.68 frames. ], batch size: 38, lr: 4.57e-03, grad_scale: 16.0 2023-05-16 04:51:56,977 INFO [finetune.py:992] (1/2) Epoch 6, batch 4850, loss[loss=0.1902, simple_loss=0.2757, pruned_loss=0.05232, over 12140.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2632, pruned_loss=0.04383, over 2375698.05 frames. ], batch size: 39, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:52:05,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 04:52:11,249 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.971e+02 3.475e+02 4.113e+02 7.039e+02, threshold=6.950e+02, percent-clipped=1.0 2023-05-16 04:52:18,032 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4977, 3.3004, 4.8268, 2.5594, 2.7063, 3.6774, 3.0549, 3.7682], device='cuda:1'), covar=tensor([0.0460, 0.1093, 0.0412, 0.1214, 0.1894, 0.1377, 0.1354, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0229, 0.0234, 0.0179, 0.0234, 0.0282, 0.0223, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:52:22,075 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166054.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 04:52:32,632 INFO [finetune.py:992] (1/2) Epoch 6, batch 4900, loss[loss=0.1745, simple_loss=0.2664, pruned_loss=0.0413, over 12271.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2621, pruned_loss=0.04332, over 2378121.15 frames. ], batch size: 37, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:52:59,861 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166106.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:53:06,390 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166115.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 04:53:08,950 INFO [finetune.py:992] (1/2) Epoch 6, batch 4950, loss[loss=0.1578, simple_loss=0.2371, pruned_loss=0.03922, over 12136.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.262, pruned_loss=0.04319, over 2379439.41 frames. ], batch size: 30, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:53:19,855 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0188, 5.9180, 5.9388, 5.2475, 5.2521, 6.0769, 5.2383, 5.5901], device='cuda:1'), covar=tensor([0.1090, 0.1414, 0.1040, 0.2815, 0.1155, 0.1320, 0.3241, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.0588, 0.0520, 0.0495, 0.0609, 0.0396, 0.0691, 0.0745, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 04:53:23,232 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.719e+02 3.145e+02 3.634e+02 5.945e+02, threshold=6.289e+02, percent-clipped=0.0 2023-05-16 04:53:28,851 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166147.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:53:37,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 04:53:37,394 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166159.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:53:45,061 INFO [finetune.py:992] (1/2) Epoch 6, batch 5000, loss[loss=0.2209, simple_loss=0.2975, pruned_loss=0.07218, over 7761.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2605, pruned_loss=0.04278, over 2373168.26 frames. ], batch size: 98, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:53:59,721 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0416, 4.6559, 4.8318, 4.8567, 4.6540, 4.8694, 4.7881, 2.6453], device='cuda:1'), covar=tensor([0.0087, 0.0066, 0.0083, 0.0064, 0.0048, 0.0087, 0.0078, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0077, 0.0071, 0.0058, 0.0088, 0.0077, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:54:00,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-16 04:54:01,238 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4377, 3.6201, 3.2431, 3.2837, 3.1012, 2.8345, 3.7400, 2.4564], device='cuda:1'), covar=tensor([0.0402, 0.0137, 0.0182, 0.0188, 0.0334, 0.0363, 0.0121, 0.0410], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0154, 0.0152, 0.0180, 0.0198, 0.0192, 0.0159, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:54:05,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 04:54:12,746 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166207.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:54:13,513 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166208.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:54:21,065 INFO [finetune.py:992] (1/2) Epoch 6, batch 5050, loss[loss=0.1697, simple_loss=0.2689, pruned_loss=0.03529, over 12195.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2606, pruned_loss=0.04272, over 2380090.27 frames. ], batch size: 35, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:54:35,720 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 2.948e+02 3.357e+02 4.167e+02 1.413e+03, threshold=6.713e+02, percent-clipped=3.0 2023-05-16 04:54:47,223 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166255.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:54:56,978 INFO [finetune.py:992] (1/2) Epoch 6, batch 5100, loss[loss=0.1975, simple_loss=0.2759, pruned_loss=0.05956, over 12034.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2612, pruned_loss=0.04289, over 2380646.85 frames. ], batch size: 31, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:55:10,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 04:55:23,832 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166307.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:55:23,911 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2054, 3.8179, 3.8311, 4.1625, 3.1840, 3.6840, 2.7447, 3.6228], device='cuda:1'), covar=tensor([0.1490, 0.0650, 0.0810, 0.0644, 0.0889, 0.0586, 0.1521, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0260, 0.0294, 0.0350, 0.0233, 0.0236, 0.0253, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:55:32,840 INFO [finetune.py:992] (1/2) Epoch 6, batch 5150, loss[loss=0.1516, simple_loss=0.2355, pruned_loss=0.0338, over 12332.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2614, pruned_loss=0.04309, over 2376581.82 frames. ], batch size: 30, lr: 4.56e-03, grad_scale: 16.0 2023-05-16 04:55:46,813 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.176e+02 2.917e+02 3.425e+02 4.057e+02 6.546e+02, threshold=6.849e+02, percent-clipped=0.0 2023-05-16 04:55:51,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3332, 2.3357, 3.1822, 4.1595, 2.2518, 4.3030, 4.2514, 4.3672], device='cuda:1'), covar=tensor([0.0106, 0.1325, 0.0440, 0.0142, 0.1268, 0.0207, 0.0163, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0202, 0.0184, 0.0115, 0.0187, 0.0176, 0.0170, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:56:02,334 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3002, 4.9590, 5.1864, 5.2188, 5.0283, 5.2522, 5.0821, 3.0619], device='cuda:1'), covar=tensor([0.0083, 0.0055, 0.0075, 0.0047, 0.0037, 0.0076, 0.0084, 0.0557], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0077, 0.0070, 0.0058, 0.0088, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:56:08,098 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166368.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:56:08,609 INFO [finetune.py:992] (1/2) Epoch 6, batch 5200, loss[loss=0.1826, simple_loss=0.2598, pruned_loss=0.05267, over 12111.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2619, pruned_loss=0.04353, over 2376934.61 frames. ], batch size: 39, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:56:35,393 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166406.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:56:38,156 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166410.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 04:56:44,238 INFO [finetune.py:992] (1/2) Epoch 6, batch 5250, loss[loss=0.2118, simple_loss=0.2969, pruned_loss=0.06332, over 11272.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2626, pruned_loss=0.04368, over 2375948.68 frames. ], batch size: 55, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:56:59,458 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.900e+02 3.310e+02 3.936e+02 6.741e+02, threshold=6.620e+02, percent-clipped=0.0 2023-05-16 04:57:07,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 04:57:10,145 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166454.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:57:13,821 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166459.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:57:20,636 INFO [finetune.py:992] (1/2) Epoch 6, batch 5300, loss[loss=0.1616, simple_loss=0.2353, pruned_loss=0.04392, over 11993.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2633, pruned_loss=0.044, over 2370793.96 frames. ], batch size: 28, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:57:29,478 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166481.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:57:44,823 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166503.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:57:47,432 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166507.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:57:48,327 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8448, 3.0094, 3.7692, 4.8245, 4.1809, 4.7519, 4.0638, 3.4770], device='cuda:1'), covar=tensor([0.0022, 0.0334, 0.0125, 0.0039, 0.0111, 0.0056, 0.0091, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0121, 0.0101, 0.0076, 0.0098, 0.0111, 0.0090, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 04:57:49,059 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166509.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:57:56,710 INFO [finetune.py:992] (1/2) Epoch 6, batch 5350, loss[loss=0.2079, simple_loss=0.2988, pruned_loss=0.05853, over 12069.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2632, pruned_loss=0.04393, over 2364177.41 frames. ], batch size: 40, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:58:05,547 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166531.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:58:11,565 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.816e+02 3.217e+02 3.761e+02 6.495e+02, threshold=6.434e+02, percent-clipped=0.0 2023-05-16 04:58:13,152 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166542.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:58:32,284 INFO [finetune.py:992] (1/2) Epoch 6, batch 5400, loss[loss=0.1498, simple_loss=0.2348, pruned_loss=0.03238, over 12178.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2622, pruned_loss=0.04332, over 2370115.83 frames. ], batch size: 31, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:58:33,157 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166570.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:58:49,456 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166592.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:59:06,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2280, 4.6872, 2.8963, 2.4745, 3.8994, 2.5061, 3.9745, 3.3549], device='cuda:1'), covar=tensor([0.0709, 0.0462, 0.1100, 0.1583, 0.0293, 0.1257, 0.0488, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0250, 0.0174, 0.0198, 0.0137, 0.0180, 0.0194, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 04:59:08,576 INFO [finetune.py:992] (1/2) Epoch 6, batch 5450, loss[loss=0.1896, simple_loss=0.2766, pruned_loss=0.05133, over 11343.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2629, pruned_loss=0.04353, over 2367428.77 frames. ], batch size: 55, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:59:23,339 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.821e+02 3.503e+02 4.549e+02 8.622e+02, threshold=7.005e+02, percent-clipped=5.0 2023-05-16 04:59:25,716 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4233, 4.6238, 4.1718, 4.9131, 4.5145, 2.8876, 4.3324, 3.0536], device='cuda:1'), covar=tensor([0.0679, 0.0782, 0.1199, 0.0490, 0.1029, 0.1577, 0.0941, 0.3056], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0367, 0.0346, 0.0270, 0.0357, 0.0263, 0.0330, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:59:34,241 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4649, 4.7533, 4.1163, 5.0487, 4.4989, 2.8968, 4.4110, 3.0833], device='cuda:1'), covar=tensor([0.0699, 0.0737, 0.1280, 0.0412, 0.1181, 0.1661, 0.0879, 0.3022], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0367, 0.0346, 0.0270, 0.0357, 0.0263, 0.0330, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 04:59:40,648 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166663.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 04:59:44,772 INFO [finetune.py:992] (1/2) Epoch 6, batch 5500, loss[loss=0.1539, simple_loss=0.2343, pruned_loss=0.03671, over 12261.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2618, pruned_loss=0.0429, over 2374199.37 frames. ], batch size: 28, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 04:59:44,979 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3107, 2.7126, 3.8528, 3.2787, 3.7288, 3.3957, 2.6787, 3.6439], device='cuda:1'), covar=tensor([0.0112, 0.0322, 0.0144, 0.0214, 0.0137, 0.0170, 0.0336, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0199, 0.0181, 0.0177, 0.0204, 0.0155, 0.0190, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:00:05,298 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1643, 2.4751, 3.6727, 3.1452, 3.5727, 3.2443, 2.5364, 3.5025], device='cuda:1'), covar=tensor([0.0117, 0.0315, 0.0128, 0.0204, 0.0127, 0.0158, 0.0332, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0198, 0.0181, 0.0177, 0.0203, 0.0154, 0.0189, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:00:07,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 05:00:13,937 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166710.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 05:00:20,105 INFO [finetune.py:992] (1/2) Epoch 6, batch 5550, loss[loss=0.1443, simple_loss=0.2192, pruned_loss=0.03471, over 11858.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2615, pruned_loss=0.04293, over 2371290.99 frames. ], batch size: 26, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 05:00:26,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4013, 4.8430, 3.0452, 2.5259, 4.0537, 2.4350, 4.0869, 3.4754], device='cuda:1'), covar=tensor([0.0571, 0.0369, 0.0997, 0.1527, 0.0251, 0.1350, 0.0402, 0.0680], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0246, 0.0172, 0.0195, 0.0136, 0.0178, 0.0191, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:00:32,369 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166735.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:00:35,573 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.867e+02 3.413e+02 3.972e+02 6.921e+02, threshold=6.827e+02, percent-clipped=0.0 2023-05-16 05:00:48,529 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166758.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 05:00:52,895 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166764.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:00:56,225 INFO [finetune.py:992] (1/2) Epoch 6, batch 5600, loss[loss=0.1825, simple_loss=0.2777, pruned_loss=0.04368, over 11376.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2624, pruned_loss=0.04325, over 2372845.98 frames. ], batch size: 55, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 05:00:57,923 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2307, 3.6234, 3.6868, 4.0173, 2.7265, 3.3942, 2.6219, 3.4483], device='cuda:1'), covar=tensor([0.1552, 0.0691, 0.0844, 0.0571, 0.1066, 0.0706, 0.1623, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0260, 0.0291, 0.0348, 0.0232, 0.0235, 0.0252, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:01:15,831 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166796.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:01:21,714 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166803.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:01:32,930 INFO [finetune.py:992] (1/2) Epoch 6, batch 5650, loss[loss=0.1493, simple_loss=0.2322, pruned_loss=0.03325, over 12179.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2616, pruned_loss=0.04292, over 2371341.95 frames. ], batch size: 29, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 05:01:34,629 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5129, 2.6123, 3.6579, 4.4281, 3.9089, 4.5152, 3.8563, 3.2578], device='cuda:1'), covar=tensor([0.0028, 0.0397, 0.0131, 0.0049, 0.0114, 0.0056, 0.0108, 0.0315], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0120, 0.0100, 0.0075, 0.0097, 0.0110, 0.0089, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:01:35,329 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3965, 3.1967, 3.1680, 3.4818, 2.6310, 3.1372, 2.5272, 2.8568], device='cuda:1'), covar=tensor([0.1341, 0.0789, 0.0972, 0.0674, 0.0996, 0.0688, 0.1496, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0291, 0.0348, 0.0231, 0.0234, 0.0251, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:01:37,528 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166825.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:01:45,984 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166837.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:01:48,043 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.719e+02 3.399e+02 4.144e+02 8.049e+02, threshold=6.798e+02, percent-clipped=1.0 2023-05-16 05:01:55,883 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166851.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:02:05,833 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166865.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:02:08,606 INFO [finetune.py:992] (1/2) Epoch 6, batch 5700, loss[loss=0.1799, simple_loss=0.2698, pruned_loss=0.04497, over 12375.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2617, pruned_loss=0.04327, over 2368914.33 frames. ], batch size: 38, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 05:02:21,951 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166887.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:02:44,682 INFO [finetune.py:992] (1/2) Epoch 6, batch 5750, loss[loss=0.1614, simple_loss=0.2555, pruned_loss=0.03368, over 12164.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2615, pruned_loss=0.04315, over 2377610.05 frames. ], batch size: 34, lr: 4.56e-03, grad_scale: 8.0 2023-05-16 05:02:51,872 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166929.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:02:59,396 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.780e+02 3.132e+02 3.989e+02 8.117e+02, threshold=6.265e+02, percent-clipped=1.0 2023-05-16 05:03:07,503 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3739, 4.4861, 4.2418, 4.9018, 4.6677, 2.8320, 4.2125, 3.0435], device='cuda:1'), covar=tensor([0.0718, 0.0827, 0.1148, 0.0449, 0.0893, 0.1590, 0.0966, 0.3214], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0365, 0.0345, 0.0268, 0.0355, 0.0262, 0.0330, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:03:16,807 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166963.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:03:20,778 INFO [finetune.py:992] (1/2) Epoch 6, batch 5800, loss[loss=0.1808, simple_loss=0.263, pruned_loss=0.04935, over 12141.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2612, pruned_loss=0.04301, over 2379454.92 frames. ], batch size: 30, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:03:35,773 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166990.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:03:50,912 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167011.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:03:56,463 INFO [finetune.py:992] (1/2) Epoch 6, batch 5850, loss[loss=0.2039, simple_loss=0.2886, pruned_loss=0.0596, over 12288.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2621, pruned_loss=0.04371, over 2378031.47 frames. ], batch size: 33, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:04:07,975 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8359, 2.5514, 3.5788, 3.6791, 2.9070, 2.7618, 2.7376, 2.4360], device='cuda:1'), covar=tensor([0.1122, 0.2709, 0.0599, 0.0492, 0.0922, 0.1838, 0.2126, 0.3060], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0373, 0.0261, 0.0290, 0.0256, 0.0287, 0.0356, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:04:11,948 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.923e+02 3.385e+02 4.055e+02 7.861e+02, threshold=6.771e+02, percent-clipped=2.0 2023-05-16 05:04:32,513 INFO [finetune.py:992] (1/2) Epoch 6, batch 5900, loss[loss=0.1811, simple_loss=0.2668, pruned_loss=0.04767, over 11850.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2623, pruned_loss=0.04384, over 2377560.98 frames. ], batch size: 44, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:04:48,344 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167091.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:04:55,436 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167100.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:04:58,297 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167104.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:05:05,158 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0675, 6.0879, 5.8930, 5.5012, 5.1621, 6.0019, 5.5955, 5.3363], device='cuda:1'), covar=tensor([0.0595, 0.0704, 0.0611, 0.1244, 0.0667, 0.0650, 0.1318, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0587, 0.0513, 0.0491, 0.0604, 0.0393, 0.0689, 0.0740, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 05:05:08,549 INFO [finetune.py:992] (1/2) Epoch 6, batch 5950, loss[loss=0.1859, simple_loss=0.2753, pruned_loss=0.04827, over 12369.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.262, pruned_loss=0.04362, over 2378726.15 frames. ], batch size: 38, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:05:09,284 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167120.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:05:21,614 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167137.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:05:23,598 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.888e+02 3.544e+02 4.208e+02 1.110e+03, threshold=7.088e+02, percent-clipped=2.0 2023-05-16 05:05:38,579 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167161.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:05:41,920 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167165.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:05:42,050 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167165.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:05:44,723 INFO [finetune.py:992] (1/2) Epoch 6, batch 6000, loss[loss=0.2112, simple_loss=0.2982, pruned_loss=0.06213, over 12094.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2631, pruned_loss=0.04374, over 2374509.52 frames. ], batch size: 42, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:05:44,723 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 05:05:55,396 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0960, 5.4579, 5.2583, 5.2067, 5.6527, 4.8776, 4.9945, 5.1692], device='cuda:1'), covar=tensor([0.1154, 0.0903, 0.0982, 0.1441, 0.0828, 0.1990, 0.2276, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0469, 0.0369, 0.0422, 0.0447, 0.0426, 0.0385, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:06:03,175 INFO [finetune.py:1026] (1/2) Epoch 6, validation: loss=0.3235, simple_loss=0.4002, pruned_loss=0.1234, over 1020973.00 frames. 2023-05-16 05:06:03,176 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 05:06:14,360 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167185.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:06:14,563 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1676, 4.4963, 3.9691, 4.8305, 4.3942, 2.7467, 4.1062, 2.9413], device='cuda:1'), covar=tensor([0.0691, 0.0745, 0.1376, 0.0373, 0.1064, 0.1624, 0.1011, 0.3129], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0366, 0.0345, 0.0269, 0.0356, 0.0263, 0.0330, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:06:15,863 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167187.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:06:19,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 05:06:35,289 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167213.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:06:39,499 INFO [finetune.py:992] (1/2) Epoch 6, batch 6050, loss[loss=0.1705, simple_loss=0.2587, pruned_loss=0.04111, over 12082.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2631, pruned_loss=0.04416, over 2374148.37 frames. ], batch size: 42, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:06:50,748 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167235.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:06:52,353 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7840, 3.3036, 5.1015, 2.7366, 2.8613, 3.7983, 3.3264, 3.8012], device='cuda:1'), covar=tensor([0.0465, 0.1210, 0.0414, 0.1203, 0.1860, 0.1501, 0.1328, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0229, 0.0232, 0.0177, 0.0232, 0.0278, 0.0220, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:06:54,240 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.751e+02 3.329e+02 3.847e+02 8.710e+02, threshold=6.657e+02, percent-clipped=1.0 2023-05-16 05:07:12,259 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3469, 2.3863, 3.1034, 4.2344, 2.1757, 4.3035, 4.2904, 4.4672], device='cuda:1'), covar=tensor([0.0124, 0.1308, 0.0499, 0.0123, 0.1301, 0.0174, 0.0152, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0203, 0.0186, 0.0116, 0.0188, 0.0177, 0.0171, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:07:15,450 INFO [finetune.py:992] (1/2) Epoch 6, batch 6100, loss[loss=0.19, simple_loss=0.2825, pruned_loss=0.04873, over 12122.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2637, pruned_loss=0.04433, over 2374380.63 frames. ], batch size: 38, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:07:26,566 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167285.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:07:50,724 INFO [finetune.py:992] (1/2) Epoch 6, batch 6150, loss[loss=0.171, simple_loss=0.2706, pruned_loss=0.03568, over 12157.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2636, pruned_loss=0.04426, over 2371441.61 frames. ], batch size: 34, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:07:54,504 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5634, 2.6824, 3.1121, 4.3606, 2.0762, 4.3811, 4.4327, 4.5723], device='cuda:1'), covar=tensor([0.0122, 0.1108, 0.0480, 0.0164, 0.1350, 0.0202, 0.0123, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0201, 0.0186, 0.0116, 0.0188, 0.0176, 0.0171, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:08:06,377 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 2.796e+02 3.234e+02 3.972e+02 7.631e+02, threshold=6.469e+02, percent-clipped=1.0 2023-05-16 05:08:26,729 INFO [finetune.py:992] (1/2) Epoch 6, batch 6200, loss[loss=0.1652, simple_loss=0.2495, pruned_loss=0.0404, over 12187.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.264, pruned_loss=0.04444, over 2373118.04 frames. ], batch size: 31, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:08:42,496 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167391.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:03,327 INFO [finetune.py:992] (1/2) Epoch 6, batch 6250, loss[loss=0.1731, simple_loss=0.2729, pruned_loss=0.03668, over 12352.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2638, pruned_loss=0.04428, over 2371443.90 frames. ], batch size: 35, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:09:04,152 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:09,177 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4577, 5.2673, 5.4099, 5.4105, 5.0333, 5.0389, 4.8564, 5.3903], device='cuda:1'), covar=tensor([0.0614, 0.0575, 0.0644, 0.0586, 0.1990, 0.1394, 0.0574, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0662, 0.0572, 0.0598, 0.0818, 0.0724, 0.0534, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 05:09:12,882 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167432.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:17,741 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167439.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:18,321 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 2.911e+02 3.395e+02 4.123e+02 7.089e+02, threshold=6.790e+02, percent-clipped=2.0 2023-05-16 05:09:29,671 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167456.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:31,925 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6088, 2.4516, 3.1054, 4.4337, 2.2724, 4.4289, 4.5035, 4.6674], device='cuda:1'), covar=tensor([0.0101, 0.1276, 0.0501, 0.0147, 0.1297, 0.0271, 0.0149, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0200, 0.0185, 0.0115, 0.0186, 0.0175, 0.0169, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:09:32,535 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167460.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:38,151 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167468.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:09:38,832 INFO [finetune.py:992] (1/2) Epoch 6, batch 6300, loss[loss=0.137, simple_loss=0.2214, pruned_loss=0.02629, over 12257.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2639, pruned_loss=0.04411, over 2362077.27 frames. ], batch size: 28, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:09:41,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 05:09:56,759 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167493.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:10:01,032 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4959, 5.5036, 5.2881, 4.9382, 4.9174, 5.4403, 5.0463, 4.8218], device='cuda:1'), covar=tensor([0.0647, 0.0745, 0.0651, 0.1482, 0.0860, 0.0710, 0.1355, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0505, 0.0487, 0.0600, 0.0390, 0.0681, 0.0734, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 05:10:04,754 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6748, 2.6380, 3.3466, 4.5267, 2.4358, 4.5092, 4.5316, 4.7501], device='cuda:1'), covar=tensor([0.0100, 0.1153, 0.0417, 0.0126, 0.1258, 0.0203, 0.0148, 0.0073], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0201, 0.0186, 0.0116, 0.0187, 0.0176, 0.0170, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:10:12,536 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0309, 4.9686, 4.8597, 4.8937, 4.4845, 4.9884, 4.9697, 5.1751], device='cuda:1'), covar=tensor([0.0215, 0.0136, 0.0181, 0.0276, 0.0749, 0.0274, 0.0152, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0184, 0.0185, 0.0236, 0.0234, 0.0205, 0.0168, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 05:10:15,219 INFO [finetune.py:992] (1/2) Epoch 6, batch 6350, loss[loss=0.1578, simple_loss=0.2463, pruned_loss=0.03463, over 12264.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2635, pruned_loss=0.04383, over 2366019.68 frames. ], batch size: 32, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:10:17,747 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 05:10:29,929 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.792e+02 3.317e+02 4.286e+02 2.773e+03, threshold=6.635e+02, percent-clipped=3.0 2023-05-16 05:10:40,987 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1305, 3.4528, 3.5725, 4.0360, 2.8289, 3.3674, 2.4504, 3.4831], device='cuda:1'), covar=tensor([0.1591, 0.0825, 0.1009, 0.0762, 0.1069, 0.0709, 0.1873, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0261, 0.0292, 0.0350, 0.0234, 0.0235, 0.0255, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:10:51,366 INFO [finetune.py:992] (1/2) Epoch 6, batch 6400, loss[loss=0.1569, simple_loss=0.2382, pruned_loss=0.03777, over 12153.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2644, pruned_loss=0.04419, over 2360074.53 frames. ], batch size: 29, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:11:03,046 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0843, 5.7968, 5.5134, 5.3741, 5.9100, 5.2398, 5.5240, 5.4242], device='cuda:1'), covar=tensor([0.1345, 0.0955, 0.0942, 0.1844, 0.0866, 0.1997, 0.1610, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0458, 0.0361, 0.0412, 0.0436, 0.0417, 0.0379, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:11:03,086 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167585.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:11:27,445 INFO [finetune.py:992] (1/2) Epoch 6, batch 6450, loss[loss=0.1706, simple_loss=0.2537, pruned_loss=0.04372, over 12350.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2634, pruned_loss=0.04394, over 2370125.62 frames. ], batch size: 31, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:11:37,439 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167633.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:11:38,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 2023-05-16 05:11:43,113 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.692e+02 3.176e+02 3.788e+02 6.386e+02, threshold=6.352e+02, percent-clipped=0.0 2023-05-16 05:12:00,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.48 vs. limit=5.0 2023-05-16 05:12:03,561 INFO [finetune.py:992] (1/2) Epoch 6, batch 6500, loss[loss=0.1549, simple_loss=0.2366, pruned_loss=0.03663, over 12302.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2637, pruned_loss=0.04403, over 2363031.91 frames. ], batch size: 28, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:12:20,474 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 05:12:32,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 05:12:39,435 INFO [finetune.py:992] (1/2) Epoch 6, batch 6550, loss[loss=0.1601, simple_loss=0.2555, pruned_loss=0.03234, over 12202.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2638, pruned_loss=0.04385, over 2372053.73 frames. ], batch size: 35, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:12:39,673 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167719.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:12:54,561 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.847e+02 3.210e+02 3.816e+02 6.751e+02, threshold=6.421e+02, percent-clipped=3.0 2023-05-16 05:13:05,932 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167756.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:08,652 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167760.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:13,782 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167767.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:15,086 INFO [finetune.py:992] (1/2) Epoch 6, batch 6600, loss[loss=0.1902, simple_loss=0.2841, pruned_loss=0.04813, over 12160.00 frames. ], tot_loss[loss=0.176, simple_loss=0.264, pruned_loss=0.04397, over 2364779.72 frames. ], batch size: 36, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:13:23,864 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167780.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:29,501 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167788.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:41,097 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167804.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:43,958 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167808.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:13:51,674 INFO [finetune.py:992] (1/2) Epoch 6, batch 6650, loss[loss=0.1589, simple_loss=0.2502, pruned_loss=0.03379, over 12105.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2637, pruned_loss=0.04391, over 2356196.68 frames. ], batch size: 33, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:13:58,785 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167828.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:14:07,254 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.793e+02 3.244e+02 3.782e+02 8.020e+02, threshold=6.487e+02, percent-clipped=1.0 2023-05-16 05:14:09,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2023-05-16 05:14:18,881 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0011, 3.4182, 5.2420, 2.5656, 2.8278, 3.9454, 3.3247, 3.7973], device='cuda:1'), covar=tensor([0.0282, 0.1012, 0.0213, 0.1132, 0.1807, 0.1199, 0.1187, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0225, 0.0228, 0.0176, 0.0229, 0.0275, 0.0217, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:14:27,890 INFO [finetune.py:992] (1/2) Epoch 6, batch 6700, loss[loss=0.1935, simple_loss=0.2832, pruned_loss=0.05184, over 11782.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2644, pruned_loss=0.0446, over 2353308.49 frames. ], batch size: 44, lr: 4.55e-03, grad_scale: 8.0 2023-05-16 05:14:30,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 05:14:31,669 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.8607, 3.9411, 3.9648, 4.4204, 2.8644, 3.7197, 2.4549, 3.9636], device='cuda:1'), covar=tensor([0.1807, 0.0806, 0.0890, 0.0629, 0.1239, 0.0719, 0.1980, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0260, 0.0290, 0.0348, 0.0233, 0.0235, 0.0253, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:14:33,721 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5395, 2.7462, 3.2188, 4.3881, 2.1858, 4.4894, 4.4581, 4.6480], device='cuda:1'), covar=tensor([0.0122, 0.1077, 0.0461, 0.0124, 0.1330, 0.0187, 0.0136, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0200, 0.0184, 0.0115, 0.0187, 0.0175, 0.0169, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:15:03,834 INFO [finetune.py:992] (1/2) Epoch 6, batch 6750, loss[loss=0.1769, simple_loss=0.2729, pruned_loss=0.04043, over 10524.00 frames. ], tot_loss[loss=0.177, simple_loss=0.265, pruned_loss=0.04451, over 2361512.93 frames. ], batch size: 68, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:15:18,836 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.940e+02 3.475e+02 4.006e+02 8.248e+02, threshold=6.951e+02, percent-clipped=1.0 2023-05-16 05:15:39,987 INFO [finetune.py:992] (1/2) Epoch 6, batch 6800, loss[loss=0.1684, simple_loss=0.2629, pruned_loss=0.0369, over 12274.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2639, pruned_loss=0.04401, over 2362842.89 frames. ], batch size: 37, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:15:52,983 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4760, 3.3835, 3.1658, 3.1020, 2.8573, 2.6280, 3.3509, 2.1610], device='cuda:1'), covar=tensor([0.0305, 0.0127, 0.0153, 0.0162, 0.0340, 0.0293, 0.0126, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0152, 0.0150, 0.0177, 0.0193, 0.0188, 0.0156, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 05:15:54,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 05:16:08,009 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8075, 4.7090, 4.6518, 4.6472, 4.2924, 4.7117, 4.7166, 5.0101], device='cuda:1'), covar=tensor([0.0213, 0.0154, 0.0190, 0.0319, 0.0718, 0.0291, 0.0173, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0183, 0.0183, 0.0232, 0.0232, 0.0204, 0.0167, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 05:16:18,600 INFO [finetune.py:992] (1/2) Epoch 6, batch 6850, loss[loss=0.1772, simple_loss=0.2673, pruned_loss=0.04353, over 10637.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2633, pruned_loss=0.04353, over 2364438.04 frames. ], batch size: 68, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:16:30,184 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168035.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:16:33,494 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.783e+02 3.526e+02 3.993e+02 7.031e+02, threshold=7.052e+02, percent-clipped=1.0 2023-05-16 05:16:37,294 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0997, 4.2304, 4.1691, 4.5479, 2.8665, 3.9956, 2.7473, 4.1674], device='cuda:1'), covar=tensor([0.1589, 0.0561, 0.0802, 0.0545, 0.1115, 0.0535, 0.1565, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0291, 0.0348, 0.0232, 0.0235, 0.0254, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:16:39,535 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2157, 4.2415, 4.2280, 4.6289, 2.9730, 3.9754, 2.7666, 4.2519], device='cuda:1'), covar=tensor([0.1505, 0.0585, 0.0745, 0.0523, 0.1085, 0.0558, 0.1614, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0291, 0.0348, 0.0232, 0.0235, 0.0254, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:16:54,696 INFO [finetune.py:992] (1/2) Epoch 6, batch 6900, loss[loss=0.201, simple_loss=0.289, pruned_loss=0.05654, over 12117.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.263, pruned_loss=0.04337, over 2366531.75 frames. ], batch size: 39, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:16:59,106 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168075.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:17:08,316 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168088.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:17:13,989 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168096.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 05:17:30,854 INFO [finetune.py:992] (1/2) Epoch 6, batch 6950, loss[loss=0.2004, simple_loss=0.2793, pruned_loss=0.06074, over 8439.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2629, pruned_loss=0.04296, over 2373515.28 frames. ], batch size: 101, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:17:33,791 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168123.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:17:43,187 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168136.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:17:45,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 05:17:46,043 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.788e+02 3.289e+02 3.840e+02 6.856e+02, threshold=6.579e+02, percent-clipped=0.0 2023-05-16 05:18:06,447 INFO [finetune.py:992] (1/2) Epoch 6, batch 7000, loss[loss=0.1764, simple_loss=0.27, pruned_loss=0.04139, over 12184.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2619, pruned_loss=0.04252, over 2381108.70 frames. ], batch size: 35, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:18:12,515 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9414, 4.8249, 4.7528, 4.7842, 4.3792, 4.8794, 4.8987, 5.0859], device='cuda:1'), covar=tensor([0.0222, 0.0169, 0.0207, 0.0323, 0.0774, 0.0369, 0.0163, 0.0186], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0185, 0.0184, 0.0233, 0.0235, 0.0205, 0.0167, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 05:18:31,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9109, 2.3420, 3.3520, 2.8124, 3.2135, 3.0238, 2.2978, 3.2830], device='cuda:1'), covar=tensor([0.0145, 0.0342, 0.0125, 0.0235, 0.0131, 0.0162, 0.0345, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0198, 0.0180, 0.0177, 0.0203, 0.0152, 0.0190, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:18:42,877 INFO [finetune.py:992] (1/2) Epoch 6, batch 7050, loss[loss=0.2001, simple_loss=0.2887, pruned_loss=0.0558, over 12029.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2623, pruned_loss=0.04263, over 2387569.86 frames. ], batch size: 40, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:18:44,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 05:18:52,752 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168233.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:18:57,556 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.650e+02 3.194e+02 3.802e+02 6.150e+02, threshold=6.388e+02, percent-clipped=0.0 2023-05-16 05:19:09,600 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3264, 2.4425, 3.0906, 4.1308, 2.2407, 4.3143, 4.1944, 4.4169], device='cuda:1'), covar=tensor([0.0110, 0.1096, 0.0476, 0.0169, 0.1224, 0.0185, 0.0145, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0201, 0.0185, 0.0117, 0.0188, 0.0176, 0.0170, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:19:18,616 INFO [finetune.py:992] (1/2) Epoch 6, batch 7100, loss[loss=0.1935, simple_loss=0.2833, pruned_loss=0.05182, over 10665.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2636, pruned_loss=0.04339, over 2383959.68 frames. ], batch size: 68, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:19:25,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1224, 6.0208, 5.9370, 5.2825, 5.2370, 6.0267, 5.6359, 5.4514], device='cuda:1'), covar=tensor([0.0595, 0.0989, 0.0616, 0.1506, 0.0602, 0.0711, 0.1435, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0508, 0.0490, 0.0603, 0.0392, 0.0682, 0.0737, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 05:19:36,372 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168294.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:19:38,475 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4648, 4.9100, 3.0800, 2.8584, 4.0864, 2.8376, 4.0917, 3.3960], device='cuda:1'), covar=tensor([0.0625, 0.0488, 0.1051, 0.1389, 0.0272, 0.1200, 0.0457, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0250, 0.0175, 0.0196, 0.0137, 0.0179, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:19:43,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 05:19:53,532 INFO [finetune.py:992] (1/2) Epoch 6, batch 7150, loss[loss=0.1763, simple_loss=0.2534, pruned_loss=0.04962, over 12097.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2636, pruned_loss=0.04347, over 2389047.53 frames. ], batch size: 32, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:20:04,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 05:20:07,115 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-05-16 05:20:08,625 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 2.955e+02 3.468e+02 4.300e+02 9.458e+02, threshold=6.936e+02, percent-clipped=4.0 2023-05-16 05:20:29,782 INFO [finetune.py:992] (1/2) Epoch 6, batch 7200, loss[loss=0.1668, simple_loss=0.2614, pruned_loss=0.03611, over 12097.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2624, pruned_loss=0.04303, over 2385118.15 frames. ], batch size: 32, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:20:34,003 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168375.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:20:36,425 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-05-16 05:20:45,202 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168391.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 05:20:53,091 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 05:21:06,130 INFO [finetune.py:992] (1/2) Epoch 6, batch 7250, loss[loss=0.2654, simple_loss=0.3342, pruned_loss=0.0983, over 7866.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2628, pruned_loss=0.04314, over 2378173.53 frames. ], batch size: 98, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:21:08,943 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168423.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:21:08,981 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168423.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:21:18,550 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9576, 4.2162, 3.7655, 4.6124, 4.2002, 2.8206, 3.9713, 2.9816], device='cuda:1'), covar=tensor([0.0865, 0.0968, 0.1493, 0.0501, 0.1151, 0.1532, 0.0977, 0.2914], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0373, 0.0348, 0.0271, 0.0359, 0.0262, 0.0334, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:21:21,088 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.057e+02 2.870e+02 3.424e+02 3.941e+02 5.479e+02, threshold=6.848e+02, percent-clipped=0.0 2023-05-16 05:21:37,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2023-05-16 05:21:41,642 INFO [finetune.py:992] (1/2) Epoch 6, batch 7300, loss[loss=0.1795, simple_loss=0.2584, pruned_loss=0.05029, over 12309.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2616, pruned_loss=0.04286, over 2381944.46 frames. ], batch size: 33, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:21:43,211 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168471.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:22:13,648 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2368, 2.7220, 3.7963, 3.1361, 3.5709, 3.4097, 2.5858, 3.6785], device='cuda:1'), covar=tensor([0.0101, 0.0257, 0.0124, 0.0205, 0.0123, 0.0120, 0.0284, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0197, 0.0179, 0.0176, 0.0202, 0.0151, 0.0187, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:22:17,718 INFO [finetune.py:992] (1/2) Epoch 6, batch 7350, loss[loss=0.2459, simple_loss=0.3125, pruned_loss=0.08961, over 7988.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2618, pruned_loss=0.04303, over 2377455.10 frames. ], batch size: 98, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:22:32,507 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.876e+02 3.392e+02 4.190e+02 6.520e+02, threshold=6.783e+02, percent-clipped=0.0 2023-05-16 05:22:53,788 INFO [finetune.py:992] (1/2) Epoch 6, batch 7400, loss[loss=0.2014, simple_loss=0.291, pruned_loss=0.05585, over 12040.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.263, pruned_loss=0.04362, over 2373129.52 frames. ], batch size: 40, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:23:00,305 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9202, 3.3538, 5.1972, 2.6171, 2.8592, 3.8569, 3.3424, 3.9028], device='cuda:1'), covar=tensor([0.0347, 0.1028, 0.0221, 0.1367, 0.1873, 0.1447, 0.1241, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0226, 0.0231, 0.0178, 0.0231, 0.0278, 0.0219, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:23:05,871 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6820, 3.6834, 3.4026, 3.1727, 3.0508, 2.9572, 3.7704, 2.4277], device='cuda:1'), covar=tensor([0.0351, 0.0133, 0.0168, 0.0190, 0.0334, 0.0309, 0.0105, 0.0445], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0155, 0.0151, 0.0183, 0.0197, 0.0193, 0.0160, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:23:07,847 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168589.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:23:19,005 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7938, 4.8066, 4.7562, 4.8559, 3.6368, 4.9196, 4.9152, 4.9913], device='cuda:1'), covar=tensor([0.0270, 0.0204, 0.0209, 0.0313, 0.1233, 0.0336, 0.0181, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0185, 0.0182, 0.0233, 0.0234, 0.0205, 0.0166, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 05:23:29,677 INFO [finetune.py:992] (1/2) Epoch 6, batch 7450, loss[loss=0.1844, simple_loss=0.2701, pruned_loss=0.04934, over 12036.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2623, pruned_loss=0.04303, over 2379994.24 frames. ], batch size: 42, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:23:44,425 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.051e+02 2.861e+02 3.391e+02 3.964e+02 1.281e+03, threshold=6.782e+02, percent-clipped=3.0 2023-05-16 05:24:05,610 INFO [finetune.py:992] (1/2) Epoch 6, batch 7500, loss[loss=0.1641, simple_loss=0.259, pruned_loss=0.03458, over 12145.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2626, pruned_loss=0.04308, over 2378430.77 frames. ], batch size: 39, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:24:06,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 05:24:21,194 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168691.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:24:41,654 INFO [finetune.py:992] (1/2) Epoch 6, batch 7550, loss[loss=0.1775, simple_loss=0.259, pruned_loss=0.048, over 12343.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2632, pruned_loss=0.04345, over 2377043.55 frames. ], batch size: 31, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:24:49,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-16 05:24:55,705 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168739.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:24:56,317 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.882e+02 3.472e+02 4.339e+02 9.651e+02, threshold=6.943e+02, percent-clipped=3.0 2023-05-16 05:25:16,599 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5371, 2.2942, 3.1695, 4.3887, 2.2738, 4.5390, 4.3596, 4.6590], device='cuda:1'), covar=tensor([0.0123, 0.1273, 0.0495, 0.0152, 0.1317, 0.0175, 0.0153, 0.0074], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0203, 0.0186, 0.0116, 0.0189, 0.0176, 0.0171, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:25:17,127 INFO [finetune.py:992] (1/2) Epoch 6, batch 7600, loss[loss=0.1737, simple_loss=0.2654, pruned_loss=0.04099, over 12148.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2633, pruned_loss=0.04394, over 2366899.46 frames. ], batch size: 34, lr: 4.54e-03, grad_scale: 16.0 2023-05-16 05:25:54,177 INFO [finetune.py:992] (1/2) Epoch 6, batch 7650, loss[loss=0.1978, simple_loss=0.2843, pruned_loss=0.05561, over 11827.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2631, pruned_loss=0.04386, over 2375856.12 frames. ], batch size: 44, lr: 4.54e-03, grad_scale: 8.0 2023-05-16 05:26:03,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 05:26:10,537 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 3.049e+02 3.497e+02 4.286e+02 7.262e+02, threshold=6.993e+02, percent-clipped=1.0 2023-05-16 05:26:30,293 INFO [finetune.py:992] (1/2) Epoch 6, batch 7700, loss[loss=0.1399, simple_loss=0.2283, pruned_loss=0.02579, over 12182.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2631, pruned_loss=0.04384, over 2378457.46 frames. ], batch size: 31, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:26:41,274 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 05:26:44,569 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168889.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:27:05,781 INFO [finetune.py:992] (1/2) Epoch 6, batch 7750, loss[loss=0.1859, simple_loss=0.2728, pruned_loss=0.04954, over 12144.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2631, pruned_loss=0.04422, over 2364230.30 frames. ], batch size: 39, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:27:19,094 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168937.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:27:21,701 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.889e+02 3.508e+02 4.779e+02 7.003e+02, threshold=7.017e+02, percent-clipped=1.0 2023-05-16 05:27:40,265 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168967.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:27:41,492 INFO [finetune.py:992] (1/2) Epoch 6, batch 7800, loss[loss=0.1891, simple_loss=0.2752, pruned_loss=0.05149, over 12047.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.264, pruned_loss=0.04443, over 2363270.72 frames. ], batch size: 42, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:27:42,336 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168969.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 05:28:18,032 INFO [finetune.py:992] (1/2) Epoch 6, batch 7850, loss[loss=0.1716, simple_loss=0.2617, pruned_loss=0.0407, over 11222.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.264, pruned_loss=0.04435, over 2367069.19 frames. ], batch size: 55, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:28:20,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 05:28:20,334 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6964, 5.5677, 5.5797, 4.8968, 4.9540, 5.7630, 4.7961, 5.1587], device='cuda:1'), covar=tensor([0.1138, 0.1568, 0.1026, 0.2310, 0.0920, 0.1136, 0.3120, 0.1729], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0504, 0.0485, 0.0592, 0.0388, 0.0674, 0.0727, 0.0534], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 05:28:24,422 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169028.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:28:25,730 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169030.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 05:28:33,235 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.923e+02 3.545e+02 4.610e+02 1.978e+03, threshold=7.090e+02, percent-clipped=5.0 2023-05-16 05:28:40,604 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3226, 4.6176, 2.7673, 2.6971, 3.8707, 2.3794, 3.8613, 3.0887], device='cuda:1'), covar=tensor([0.0576, 0.0387, 0.1051, 0.1249, 0.0258, 0.1216, 0.0405, 0.0772], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0249, 0.0174, 0.0195, 0.0136, 0.0179, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:28:54,141 INFO [finetune.py:992] (1/2) Epoch 6, batch 7900, loss[loss=0.1706, simple_loss=0.2597, pruned_loss=0.04077, over 12112.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2645, pruned_loss=0.04475, over 2367869.38 frames. ], batch size: 33, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:29:03,300 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:29:11,953 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6337, 3.6876, 3.4219, 3.2415, 2.9825, 2.8081, 3.7917, 2.2984], device='cuda:1'), covar=tensor([0.0321, 0.0134, 0.0131, 0.0167, 0.0316, 0.0327, 0.0122, 0.0412], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0159, 0.0155, 0.0186, 0.0201, 0.0196, 0.0164, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:29:28,378 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169116.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:29:30,253 INFO [finetune.py:992] (1/2) Epoch 6, batch 7950, loss[loss=0.1852, simple_loss=0.2735, pruned_loss=0.04843, over 12346.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2639, pruned_loss=0.04423, over 2373561.02 frames. ], batch size: 35, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:29:32,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 05:29:34,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 05:29:45,742 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.769e+02 3.316e+02 3.894e+02 6.900e+02, threshold=6.632e+02, percent-clipped=0.0 2023-05-16 05:29:47,419 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169143.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:29:58,800 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169159.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:30:05,677 INFO [finetune.py:992] (1/2) Epoch 6, batch 8000, loss[loss=0.1403, simple_loss=0.2246, pruned_loss=0.02804, over 12028.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2638, pruned_loss=0.04404, over 2373227.36 frames. ], batch size: 28, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:30:11,599 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169177.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:30:38,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 05:30:42,274 INFO [finetune.py:992] (1/2) Epoch 6, batch 8050, loss[loss=0.1818, simple_loss=0.269, pruned_loss=0.04729, over 11815.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2635, pruned_loss=0.04419, over 2367352.01 frames. ], batch size: 44, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:30:43,162 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169220.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:30:48,695 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5225, 4.9814, 5.4837, 4.8278, 5.0288, 4.8749, 5.5099, 5.1514], device='cuda:1'), covar=tensor([0.0204, 0.0347, 0.0228, 0.0201, 0.0325, 0.0227, 0.0194, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0243, 0.0262, 0.0239, 0.0237, 0.0236, 0.0215, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:30:57,585 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.811e+02 3.316e+02 3.937e+02 6.960e+02, threshold=6.632e+02, percent-clipped=1.0 2023-05-16 05:31:18,472 INFO [finetune.py:992] (1/2) Epoch 6, batch 8100, loss[loss=0.1483, simple_loss=0.2345, pruned_loss=0.03107, over 12194.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2638, pruned_loss=0.04455, over 2367229.34 frames. ], batch size: 29, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:31:54,302 INFO [finetune.py:992] (1/2) Epoch 6, batch 8150, loss[loss=0.2035, simple_loss=0.2895, pruned_loss=0.05879, over 11213.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2642, pruned_loss=0.04515, over 2352542.23 frames. ], batch size: 55, lr: 4.53e-03, grad_scale: 4.0 2023-05-16 05:31:57,268 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169323.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:31:58,758 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169325.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 05:32:10,277 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 2.928e+02 3.402e+02 4.619e+02 3.270e+03, threshold=6.805e+02, percent-clipped=9.0 2023-05-16 05:32:10,574 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3267, 2.9762, 4.6402, 2.3799, 2.6699, 3.6181, 3.0414, 3.6512], device='cuda:1'), covar=tensor([0.0449, 0.1253, 0.0446, 0.1279, 0.1819, 0.1342, 0.1346, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0226, 0.0231, 0.0178, 0.0232, 0.0278, 0.0220, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:32:30,110 INFO [finetune.py:992] (1/2) Epoch 6, batch 8200, loss[loss=0.1888, simple_loss=0.2912, pruned_loss=0.04319, over 12360.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2644, pruned_loss=0.04513, over 2357519.43 frames. ], batch size: 35, lr: 4.53e-03, grad_scale: 4.0 2023-05-16 05:33:06,078 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2899, 3.1991, 3.3534, 3.5780, 2.7078, 3.2380, 2.6787, 3.1017], device='cuda:1'), covar=tensor([0.1283, 0.0670, 0.0680, 0.0675, 0.0817, 0.0586, 0.1240, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0255, 0.0289, 0.0345, 0.0230, 0.0233, 0.0250, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:33:06,642 INFO [finetune.py:992] (1/2) Epoch 6, batch 8250, loss[loss=0.2873, simple_loss=0.336, pruned_loss=0.1193, over 7812.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2653, pruned_loss=0.04535, over 2355379.93 frames. ], batch size: 97, lr: 4.53e-03, grad_scale: 4.0 2023-05-16 05:33:19,843 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8735, 3.2988, 2.3701, 2.1537, 2.9287, 2.3202, 3.1164, 2.6233], device='cuda:1'), covar=tensor([0.0531, 0.0666, 0.0899, 0.1266, 0.0344, 0.1078, 0.0525, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0248, 0.0174, 0.0194, 0.0135, 0.0177, 0.0193, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:33:20,412 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169438.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:33:23,185 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.299e+02 2.930e+02 3.366e+02 4.211e+02 1.512e+03, threshold=6.732e+02, percent-clipped=5.0 2023-05-16 05:33:42,697 INFO [finetune.py:992] (1/2) Epoch 6, batch 8300, loss[loss=0.1769, simple_loss=0.2664, pruned_loss=0.0437, over 12107.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.265, pruned_loss=0.04535, over 2349398.19 frames. ], batch size: 33, lr: 4.53e-03, grad_scale: 4.0 2023-05-16 05:33:44,820 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169472.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:34:16,010 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169515.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:34:18,743 INFO [finetune.py:992] (1/2) Epoch 6, batch 8350, loss[loss=0.174, simple_loss=0.2588, pruned_loss=0.04455, over 12368.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2641, pruned_loss=0.04468, over 2359596.15 frames. ], batch size: 35, lr: 4.53e-03, grad_scale: 4.0 2023-05-16 05:34:35,832 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.672e+02 3.314e+02 4.024e+02 1.331e+03, threshold=6.629e+02, percent-clipped=5.0 2023-05-16 05:34:39,075 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5312, 4.7544, 4.2486, 5.0583, 4.6326, 2.9347, 4.4885, 3.1022], device='cuda:1'), covar=tensor([0.0628, 0.0682, 0.1240, 0.0424, 0.1024, 0.1470, 0.0828, 0.3077], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0371, 0.0347, 0.0270, 0.0357, 0.0261, 0.0331, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:34:39,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 05:34:49,127 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3807, 4.8378, 2.9975, 2.7123, 4.1400, 2.6038, 4.1335, 3.3782], device='cuda:1'), covar=tensor([0.0627, 0.0440, 0.0992, 0.1428, 0.0249, 0.1255, 0.0411, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0246, 0.0172, 0.0193, 0.0134, 0.0176, 0.0192, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:34:55,377 INFO [finetune.py:992] (1/2) Epoch 6, batch 8400, loss[loss=0.2649, simple_loss=0.3219, pruned_loss=0.104, over 7857.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2642, pruned_loss=0.04468, over 2345724.00 frames. ], batch size: 98, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:35:08,853 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1626, 5.1251, 5.0379, 5.1161, 4.7127, 5.2270, 5.1854, 5.4316], device='cuda:1'), covar=tensor([0.0182, 0.0133, 0.0148, 0.0251, 0.0667, 0.0255, 0.0123, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0188, 0.0184, 0.0236, 0.0236, 0.0207, 0.0169, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 05:35:31,416 INFO [finetune.py:992] (1/2) Epoch 6, batch 8450, loss[loss=0.1812, simple_loss=0.2644, pruned_loss=0.04895, over 12096.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2635, pruned_loss=0.0445, over 2346787.20 frames. ], batch size: 32, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:35:33,033 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:35:34,524 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169623.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:35:35,879 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169625.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 05:35:41,026 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 05:35:48,306 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.765e+02 3.281e+02 4.006e+02 6.222e+02, threshold=6.562e+02, percent-clipped=0.0 2023-05-16 05:36:00,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-05-16 05:36:00,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 05:36:08,345 INFO [finetune.py:992] (1/2) Epoch 6, batch 8500, loss[loss=0.1638, simple_loss=0.251, pruned_loss=0.03835, over 12124.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2641, pruned_loss=0.04451, over 2352943.48 frames. ], batch size: 30, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:36:09,846 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169671.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:36:10,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 05:36:11,328 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169673.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 05:36:17,593 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1944, 6.1495, 5.9072, 5.5695, 5.2429, 6.0996, 5.7005, 5.4719], device='cuda:1'), covar=tensor([0.0666, 0.0931, 0.0616, 0.1461, 0.0589, 0.0675, 0.1430, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0583, 0.0511, 0.0491, 0.0600, 0.0394, 0.0683, 0.0741, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 05:36:17,675 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169682.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:36:30,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2023-05-16 05:36:43,591 INFO [finetune.py:992] (1/2) Epoch 6, batch 8550, loss[loss=0.1729, simple_loss=0.2727, pruned_loss=0.03655, over 12120.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.264, pruned_loss=0.04413, over 2355573.78 frames. ], batch size: 39, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:36:50,962 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169729.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:36:54,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2023-05-16 05:36:57,386 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169738.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:00,184 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.624e+02 3.312e+02 3.872e+02 8.850e+02, threshold=6.624e+02, percent-clipped=2.0 2023-05-16 05:37:04,020 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169747.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:19,693 INFO [finetune.py:992] (1/2) Epoch 6, batch 8600, loss[loss=0.1997, simple_loss=0.285, pruned_loss=0.05718, over 10684.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2633, pruned_loss=0.04381, over 2359843.09 frames. ], batch size: 69, lr: 4.53e-03, grad_scale: 8.0 2023-05-16 05:37:21,997 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169772.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:32,576 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169786.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:32,738 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169786.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:35,492 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169790.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:40,541 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7787, 2.6014, 3.7089, 4.7420, 4.1926, 4.6401, 3.9511, 3.6662], device='cuda:1'), covar=tensor([0.0030, 0.0398, 0.0164, 0.0035, 0.0094, 0.0064, 0.0111, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0121, 0.0103, 0.0076, 0.0100, 0.0111, 0.0092, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:37:47,498 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 05:37:49,377 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169808.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:54,421 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169815.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:37:57,090 INFO [finetune.py:992] (1/2) Epoch 6, batch 8650, loss[loss=0.1999, simple_loss=0.2874, pruned_loss=0.05617, over 10445.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2635, pruned_loss=0.04365, over 2365212.12 frames. ], batch size: 68, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:37:57,872 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169820.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:38:13,691 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.733e+02 3.214e+02 3.902e+02 8.674e+02, threshold=6.428e+02, percent-clipped=2.0 2023-05-16 05:38:17,536 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169847.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:38:28,946 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169863.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:38:33,159 INFO [finetune.py:992] (1/2) Epoch 6, batch 8700, loss[loss=0.1488, simple_loss=0.2305, pruned_loss=0.03355, over 12299.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2629, pruned_loss=0.04346, over 2364230.61 frames. ], batch size: 28, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:38:39,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2094, 4.2278, 2.6623, 2.4047, 3.6363, 2.4199, 3.6742, 2.8528], device='cuda:1'), covar=tensor([0.0615, 0.0442, 0.1093, 0.1426, 0.0240, 0.1272, 0.0466, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0249, 0.0175, 0.0195, 0.0137, 0.0178, 0.0194, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:39:08,887 INFO [finetune.py:992] (1/2) Epoch 6, batch 8750, loss[loss=0.172, simple_loss=0.2676, pruned_loss=0.03821, over 12354.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2631, pruned_loss=0.04337, over 2368504.29 frames. ], batch size: 35, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:39:26,000 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.814e+02 3.203e+02 3.882e+02 8.670e+02, threshold=6.406e+02, percent-clipped=2.0 2023-05-16 05:39:32,712 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6547, 2.9823, 4.5607, 4.7405, 2.9916, 2.7664, 2.9792, 2.0948], device='cuda:1'), covar=tensor([0.1369, 0.2475, 0.0389, 0.0364, 0.1050, 0.1817, 0.2436, 0.3577], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0373, 0.0265, 0.0291, 0.0258, 0.0287, 0.0357, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:39:43,885 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6838, 2.6060, 3.6343, 4.6304, 4.0736, 4.5516, 3.9724, 3.5750], device='cuda:1'), covar=tensor([0.0031, 0.0383, 0.0146, 0.0043, 0.0110, 0.0068, 0.0108, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0122, 0.0103, 0.0076, 0.0101, 0.0112, 0.0093, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:39:45,838 INFO [finetune.py:992] (1/2) Epoch 6, batch 8800, loss[loss=0.168, simple_loss=0.2552, pruned_loss=0.04039, over 12143.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2627, pruned_loss=0.04286, over 2379262.72 frames. ], batch size: 36, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:39:48,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 05:39:51,492 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169977.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:40:17,487 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0766, 4.6630, 4.8427, 5.0004, 4.8540, 4.9473, 4.9322, 2.4070], device='cuda:1'), covar=tensor([0.0102, 0.0062, 0.0077, 0.0053, 0.0047, 0.0081, 0.0066, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0073, 0.0076, 0.0069, 0.0058, 0.0088, 0.0076, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:40:24,645 INFO [finetune.py:992] (1/2) Epoch 6, batch 8850, loss[loss=0.155, simple_loss=0.238, pruned_loss=0.03601, over 12117.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2619, pruned_loss=0.04265, over 2373249.76 frames. ], batch size: 30, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:40:40,962 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.125e+02 2.852e+02 3.342e+02 4.079e+02 1.347e+03, threshold=6.684e+02, percent-clipped=1.0 2023-05-16 05:40:43,317 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170045.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:40:50,371 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0164, 4.6545, 4.8710, 4.9524, 4.6465, 4.9917, 4.8869, 2.7649], device='cuda:1'), covar=tensor([0.0116, 0.0068, 0.0085, 0.0054, 0.0063, 0.0073, 0.0077, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0077, 0.0070, 0.0058, 0.0088, 0.0076, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:41:00,651 INFO [finetune.py:992] (1/2) Epoch 6, batch 8900, loss[loss=0.1835, simple_loss=0.2727, pruned_loss=0.04712, over 12042.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2623, pruned_loss=0.04291, over 2370231.69 frames. ], batch size: 40, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:41:12,247 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170085.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:41:25,966 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170103.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:41:28,209 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170106.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:41:37,257 INFO [finetune.py:992] (1/2) Epoch 6, batch 8950, loss[loss=0.1683, simple_loss=0.2607, pruned_loss=0.03799, over 12114.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2619, pruned_loss=0.04319, over 2370075.75 frames. ], batch size: 33, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:41:41,904 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0700, 4.6939, 4.9623, 5.1056, 4.8125, 5.0091, 4.9586, 2.7512], device='cuda:1'), covar=tensor([0.0103, 0.0080, 0.0075, 0.0060, 0.0051, 0.0077, 0.0108, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0077, 0.0070, 0.0058, 0.0088, 0.0076, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:41:53,764 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.640e+02 3.012e+02 3.913e+02 7.653e+02, threshold=6.023e+02, percent-clipped=3.0 2023-05-16 05:41:53,866 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170142.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:41:59,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.51 vs. limit=5.0 2023-05-16 05:42:13,132 INFO [finetune.py:992] (1/2) Epoch 6, batch 9000, loss[loss=0.1659, simple_loss=0.255, pruned_loss=0.03838, over 12016.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2614, pruned_loss=0.04294, over 2375974.59 frames. ], batch size: 31, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:42:13,132 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 05:42:28,419 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8063, 4.5195, 4.7996, 4.7867, 4.4867, 4.6020, 4.7727, 2.5870], device='cuda:1'), covar=tensor([0.0086, 0.0068, 0.0063, 0.0064, 0.0057, 0.0101, 0.0061, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0073, 0.0077, 0.0070, 0.0059, 0.0088, 0.0076, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:42:30,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3036, 5.1927, 5.1204, 4.6050, 4.8776, 5.2020, 4.6822, 4.6709], device='cuda:1'), covar=tensor([0.0638, 0.0929, 0.0628, 0.1690, 0.0584, 0.0774, 0.1970, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0512, 0.0492, 0.0601, 0.0395, 0.0686, 0.0745, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 05:42:31,143 INFO [finetune.py:1026] (1/2) Epoch 6, validation: loss=0.3329, simple_loss=0.4046, pruned_loss=0.1306, over 1020973.00 frames. 2023-05-16 05:42:31,143 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 05:43:08,309 INFO [finetune.py:992] (1/2) Epoch 6, batch 9050, loss[loss=0.1574, simple_loss=0.2392, pruned_loss=0.03775, over 12178.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.0436, over 2367744.35 frames. ], batch size: 29, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:43:24,845 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.254e+02 2.871e+02 3.359e+02 3.998e+02 1.069e+03, threshold=6.719e+02, percent-clipped=4.0 2023-05-16 05:43:43,968 INFO [finetune.py:992] (1/2) Epoch 6, batch 9100, loss[loss=0.1588, simple_loss=0.2458, pruned_loss=0.03591, over 12348.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2626, pruned_loss=0.04367, over 2367692.53 frames. ], batch size: 31, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:43:49,829 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170277.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:44:19,917 INFO [finetune.py:992] (1/2) Epoch 6, batch 9150, loss[loss=0.1917, simple_loss=0.2727, pruned_loss=0.05528, over 12026.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2631, pruned_loss=0.04389, over 2367033.83 frames. ], batch size: 31, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:44:24,184 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170325.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:44:36,792 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.818e+02 3.271e+02 4.074e+02 1.050e+03, threshold=6.541e+02, percent-clipped=5.0 2023-05-16 05:44:56,295 INFO [finetune.py:992] (1/2) Epoch 6, batch 9200, loss[loss=0.1889, simple_loss=0.2785, pruned_loss=0.04968, over 10375.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2619, pruned_loss=0.04314, over 2371907.55 frames. ], batch size: 68, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:45:07,756 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170385.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:45:19,479 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170401.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:45:20,828 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170403.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:45:32,036 INFO [finetune.py:992] (1/2) Epoch 6, batch 9250, loss[loss=0.2561, simple_loss=0.3261, pruned_loss=0.09305, over 8093.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2623, pruned_loss=0.04319, over 2371556.62 frames. ], batch size: 97, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:45:42,038 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170433.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:45:48,292 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.833e+02 3.194e+02 3.964e+02 6.931e+02, threshold=6.389e+02, percent-clipped=0.0 2023-05-16 05:45:48,482 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170442.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:45:50,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 05:45:52,102 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5807, 4.2679, 4.4442, 4.4486, 4.2451, 4.5112, 4.3724, 2.4327], device='cuda:1'), covar=tensor([0.0119, 0.0077, 0.0091, 0.0074, 0.0073, 0.0098, 0.0099, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0077, 0.0070, 0.0059, 0.0089, 0.0077, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:45:54,899 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170451.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:45:56,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7149, 4.0129, 3.6136, 4.2908, 3.8769, 2.6347, 3.7237, 3.0258], device='cuda:1'), covar=tensor([0.0893, 0.0841, 0.1482, 0.0497, 0.1143, 0.1660, 0.0990, 0.2733], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0370, 0.0348, 0.0270, 0.0356, 0.0260, 0.0332, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:46:02,812 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170461.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:46:08,217 INFO [finetune.py:992] (1/2) Epoch 6, batch 9300, loss[loss=0.1763, simple_loss=0.2685, pruned_loss=0.04208, over 11906.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2635, pruned_loss=0.04391, over 2370518.94 frames. ], batch size: 44, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:46:23,805 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170490.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:46:44,296 INFO [finetune.py:992] (1/2) Epoch 6, batch 9350, loss[loss=0.1423, simple_loss=0.2212, pruned_loss=0.03164, over 11822.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2637, pruned_loss=0.04396, over 2365410.70 frames. ], batch size: 26, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:46:46,721 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170522.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 05:47:00,814 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 2.784e+02 3.192e+02 3.869e+02 5.304e+02, threshold=6.383e+02, percent-clipped=1.0 2023-05-16 05:47:09,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 05:47:19,836 INFO [finetune.py:992] (1/2) Epoch 6, batch 9400, loss[loss=0.1399, simple_loss=0.2184, pruned_loss=0.0307, over 12022.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2628, pruned_loss=0.04351, over 2370974.63 frames. ], batch size: 28, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:47:20,007 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170569.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:47:23,950 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 05:47:27,792 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2791, 4.5493, 4.0363, 4.9234, 4.5079, 2.8630, 4.3138, 3.0776], device='cuda:1'), covar=tensor([0.0764, 0.0798, 0.1345, 0.0416, 0.1061, 0.1521, 0.0904, 0.3105], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0373, 0.0350, 0.0274, 0.0358, 0.0261, 0.0334, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:47:56,152 INFO [finetune.py:992] (1/2) Epoch 6, batch 9450, loss[loss=0.1557, simple_loss=0.2429, pruned_loss=0.0342, over 12346.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2633, pruned_loss=0.04392, over 2368691.96 frames. ], batch size: 31, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:48:04,949 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170630.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:48:13,258 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 2.989e+02 3.445e+02 4.059e+02 9.553e+02, threshold=6.891e+02, percent-clipped=4.0 2023-05-16 05:48:32,556 INFO [finetune.py:992] (1/2) Epoch 6, batch 9500, loss[loss=0.1903, simple_loss=0.2754, pruned_loss=0.05259, over 11560.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2628, pruned_loss=0.04387, over 2369589.76 frames. ], batch size: 48, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:48:55,535 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170701.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:49:08,205 INFO [finetune.py:992] (1/2) Epoch 6, batch 9550, loss[loss=0.1883, simple_loss=0.2784, pruned_loss=0.04907, over 10608.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2625, pruned_loss=0.04402, over 2363855.80 frames. ], batch size: 68, lr: 4.52e-03, grad_scale: 8.0 2023-05-16 05:49:22,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2023-05-16 05:49:24,654 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.831e+02 3.344e+02 4.153e+02 1.450e+03, threshold=6.688e+02, percent-clipped=3.0 2023-05-16 05:49:29,550 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170749.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:49:45,084 INFO [finetune.py:992] (1/2) Epoch 6, batch 9600, loss[loss=0.2351, simple_loss=0.3065, pruned_loss=0.08184, over 8325.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2623, pruned_loss=0.04367, over 2367692.12 frames. ], batch size: 98, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:50:19,711 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170817.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 05:50:20,995 INFO [finetune.py:992] (1/2) Epoch 6, batch 9650, loss[loss=0.1508, simple_loss=0.2341, pruned_loss=0.0338, over 12249.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2618, pruned_loss=0.04384, over 2369177.63 frames. ], batch size: 32, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:50:32,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 05:50:37,412 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.759e+02 3.162e+02 3.723e+02 7.931e+02, threshold=6.323e+02, percent-clipped=2.0 2023-05-16 05:50:56,767 INFO [finetune.py:992] (1/2) Epoch 6, batch 9700, loss[loss=0.1745, simple_loss=0.2676, pruned_loss=0.04069, over 11694.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2624, pruned_loss=0.04408, over 2366331.43 frames. ], batch size: 48, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:51:07,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4977, 4.7964, 4.1548, 5.1496, 4.6062, 2.7509, 4.1615, 3.0763], device='cuda:1'), covar=tensor([0.0626, 0.0613, 0.1262, 0.0380, 0.0976, 0.1529, 0.1042, 0.2872], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0371, 0.0349, 0.0273, 0.0359, 0.0261, 0.0335, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:51:33,889 INFO [finetune.py:992] (1/2) Epoch 6, batch 9750, loss[loss=0.1561, simple_loss=0.2379, pruned_loss=0.03719, over 12100.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2621, pruned_loss=0.0439, over 2376828.50 frames. ], batch size: 32, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:51:36,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-05-16 05:51:38,248 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170925.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:51:40,553 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170928.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:51:50,277 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.889e+02 3.336e+02 3.908e+02 8.110e+02, threshold=6.672e+02, percent-clipped=2.0 2023-05-16 05:51:50,463 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4459, 4.9601, 5.4413, 4.7781, 5.0090, 4.8336, 5.4794, 5.1397], device='cuda:1'), covar=tensor([0.0258, 0.0377, 0.0234, 0.0216, 0.0322, 0.0281, 0.0185, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0246, 0.0270, 0.0247, 0.0242, 0.0243, 0.0220, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 05:52:09,564 INFO [finetune.py:992] (1/2) Epoch 6, batch 9800, loss[loss=0.1583, simple_loss=0.2474, pruned_loss=0.03464, over 12090.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.262, pruned_loss=0.04394, over 2374295.25 frames. ], batch size: 32, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:52:24,073 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170989.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:52:42,063 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2124, 5.1454, 5.1085, 5.0946, 4.7984, 5.1929, 5.2271, 5.4003], device='cuda:1'), covar=tensor([0.0191, 0.0142, 0.0141, 0.0250, 0.0616, 0.0267, 0.0133, 0.0138], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0186, 0.0182, 0.0232, 0.0232, 0.0205, 0.0166, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 05:52:43,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 05:52:45,521 INFO [finetune.py:992] (1/2) Epoch 6, batch 9850, loss[loss=0.1606, simple_loss=0.2515, pruned_loss=0.0348, over 12295.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2624, pruned_loss=0.04402, over 2365746.80 frames. ], batch size: 33, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:53:03,170 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.930e+02 3.532e+02 4.214e+02 7.457e+02, threshold=7.065e+02, percent-clipped=2.0 2023-05-16 05:53:07,343 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2287, 5.0630, 5.1453, 5.1986, 4.8222, 4.9129, 4.6938, 5.1710], device='cuda:1'), covar=tensor([0.0595, 0.0529, 0.0749, 0.0527, 0.1686, 0.1141, 0.0550, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0659, 0.0567, 0.0594, 0.0802, 0.0710, 0.0533, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 05:53:21,977 INFO [finetune.py:992] (1/2) Epoch 6, batch 9900, loss[loss=0.1475, simple_loss=0.2363, pruned_loss=0.0294, over 12266.00 frames. ], tot_loss[loss=0.175, simple_loss=0.262, pruned_loss=0.04394, over 2366018.49 frames. ], batch size: 28, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:53:38,425 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171092.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:53:38,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.75 vs. limit=5.0 2023-05-16 05:53:56,053 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171117.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 05:53:57,355 INFO [finetune.py:992] (1/2) Epoch 6, batch 9950, loss[loss=0.1767, simple_loss=0.2697, pruned_loss=0.04186, over 10479.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2622, pruned_loss=0.04378, over 2364483.65 frames. ], batch size: 68, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:54:13,383 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.846e+02 3.389e+02 4.113e+02 8.851e+02, threshold=6.779e+02, percent-clipped=1.0 2023-05-16 05:54:21,288 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171153.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:54:29,801 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171165.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:54:32,654 INFO [finetune.py:992] (1/2) Epoch 6, batch 10000, loss[loss=0.1837, simple_loss=0.2754, pruned_loss=0.04604, over 11262.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2616, pruned_loss=0.04339, over 2369908.94 frames. ], batch size: 55, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:54:43,365 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6346, 4.7230, 4.3594, 5.1497, 4.8882, 3.0475, 4.4110, 3.1479], device='cuda:1'), covar=tensor([0.0622, 0.0805, 0.1170, 0.0427, 0.0793, 0.1471, 0.0969, 0.3255], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0372, 0.0348, 0.0273, 0.0359, 0.0262, 0.0335, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:54:53,187 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171196.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:55:09,436 INFO [finetune.py:992] (1/2) Epoch 6, batch 10050, loss[loss=0.146, simple_loss=0.2333, pruned_loss=0.02934, over 12032.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2618, pruned_loss=0.04319, over 2371100.86 frames. ], batch size: 31, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:55:13,652 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171225.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:55:20,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.87 vs. limit=5.0 2023-05-16 05:55:25,460 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 2.682e+02 3.254e+02 3.835e+02 7.863e+02, threshold=6.507e+02, percent-clipped=3.0 2023-05-16 05:55:36,169 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171257.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:55:44,425 INFO [finetune.py:992] (1/2) Epoch 6, batch 10100, loss[loss=0.1667, simple_loss=0.2678, pruned_loss=0.03276, over 12189.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2621, pruned_loss=0.04301, over 2378507.01 frames. ], batch size: 35, lr: 4.51e-03, grad_scale: 8.0 2023-05-16 05:55:47,307 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171273.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:55:55,319 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171284.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:55:57,579 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6748, 4.5557, 4.5018, 4.5480, 4.2364, 4.7082, 4.6630, 4.7883], device='cuda:1'), covar=tensor([0.0190, 0.0161, 0.0167, 0.0274, 0.0660, 0.0281, 0.0160, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0186, 0.0182, 0.0232, 0.0232, 0.0204, 0.0166, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 05:56:20,083 INFO [finetune.py:992] (1/2) Epoch 6, batch 10150, loss[loss=0.1577, simple_loss=0.2349, pruned_loss=0.0403, over 12281.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2625, pruned_loss=0.04313, over 2376113.50 frames. ], batch size: 28, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 05:56:37,678 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.599e+02 3.165e+02 3.723e+02 8.067e+02, threshold=6.331e+02, percent-clipped=2.0 2023-05-16 05:56:40,011 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171345.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:56:55,022 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2440, 4.3205, 4.2081, 4.5773, 3.0107, 4.1595, 2.6441, 4.2644], device='cuda:1'), covar=tensor([0.1510, 0.0542, 0.0746, 0.0589, 0.1086, 0.0506, 0.1630, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0258, 0.0295, 0.0348, 0.0233, 0.0237, 0.0252, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 05:56:56,938 INFO [finetune.py:992] (1/2) Epoch 6, batch 10200, loss[loss=0.1834, simple_loss=0.2727, pruned_loss=0.04704, over 12357.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2618, pruned_loss=0.0427, over 2376688.22 frames. ], batch size: 36, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 05:57:04,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 05:57:24,014 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171406.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:57:31,898 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7325, 3.6020, 3.3412, 3.2265, 2.9933, 2.8012, 3.6181, 2.3264], device='cuda:1'), covar=tensor([0.0296, 0.0113, 0.0138, 0.0153, 0.0348, 0.0298, 0.0127, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0155, 0.0151, 0.0180, 0.0196, 0.0190, 0.0159, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:57:33,085 INFO [finetune.py:992] (1/2) Epoch 6, batch 10250, loss[loss=0.1545, simple_loss=0.2559, pruned_loss=0.02653, over 12107.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2602, pruned_loss=0.04197, over 2378394.09 frames. ], batch size: 33, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 05:57:40,607 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5311, 2.7384, 4.6092, 4.7534, 2.8235, 2.5118, 2.6937, 2.0782], device='cuda:1'), covar=tensor([0.1579, 0.3015, 0.0401, 0.0372, 0.1213, 0.2215, 0.2885, 0.3685], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0378, 0.0268, 0.0295, 0.0261, 0.0291, 0.0363, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 05:57:49,595 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.739e+02 3.296e+02 4.036e+02 9.177e+02, threshold=6.592e+02, percent-clipped=3.0 2023-05-16 05:57:53,899 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171448.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:58:08,826 INFO [finetune.py:992] (1/2) Epoch 6, batch 10300, loss[loss=0.1895, simple_loss=0.2753, pruned_loss=0.05181, over 12298.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2612, pruned_loss=0.04261, over 2380303.79 frames. ], batch size: 34, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 05:58:30,835 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171498.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:58:45,643 INFO [finetune.py:992] (1/2) Epoch 6, batch 10350, loss[loss=0.1628, simple_loss=0.2626, pruned_loss=0.03151, over 12276.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.262, pruned_loss=0.0428, over 2379650.84 frames. ], batch size: 37, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 05:59:01,572 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.760e+02 3.243e+02 4.133e+02 1.006e+03, threshold=6.485e+02, percent-clipped=2.0 2023-05-16 05:59:08,827 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171552.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:59:13,989 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171559.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:59:20,900 INFO [finetune.py:992] (1/2) Epoch 6, batch 10400, loss[loss=0.1858, simple_loss=0.2752, pruned_loss=0.04818, over 12113.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2621, pruned_loss=0.04303, over 2382377.65 frames. ], batch size: 38, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 05:59:31,486 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171584.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 05:59:56,467 INFO [finetune.py:992] (1/2) Epoch 6, batch 10450, loss[loss=0.1653, simple_loss=0.2559, pruned_loss=0.03733, over 11818.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2606, pruned_loss=0.04249, over 2380645.70 frames. ], batch size: 44, lr: 4.51e-03, grad_scale: 16.0 2023-05-16 06:00:07,047 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171632.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:00:14,128 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.168e+02 2.841e+02 3.503e+02 4.038e+02 1.380e+03, threshold=7.007e+02, percent-clipped=4.0 2023-05-16 06:00:33,295 INFO [finetune.py:992] (1/2) Epoch 6, batch 10500, loss[loss=0.1753, simple_loss=0.2588, pruned_loss=0.04593, over 12259.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2618, pruned_loss=0.04328, over 2373423.48 frames. ], batch size: 32, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:00:56,122 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171701.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:01:02,858 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9217, 3.4740, 5.3185, 2.8314, 2.9955, 4.0165, 3.2826, 4.0488], device='cuda:1'), covar=tensor([0.0389, 0.1026, 0.0247, 0.1092, 0.1716, 0.1343, 0.1286, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0227, 0.0233, 0.0178, 0.0233, 0.0279, 0.0220, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:01:09,085 INFO [finetune.py:992] (1/2) Epoch 6, batch 10550, loss[loss=0.262, simple_loss=0.3298, pruned_loss=0.09713, over 7827.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.261, pruned_loss=0.04268, over 2379383.04 frames. ], batch size: 99, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:01:25,602 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.657e+02 3.113e+02 3.809e+02 6.399e+02, threshold=6.226e+02, percent-clipped=0.0 2023-05-16 06:01:29,959 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171748.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:01:44,989 INFO [finetune.py:992] (1/2) Epoch 6, batch 10600, loss[loss=0.1582, simple_loss=0.2335, pruned_loss=0.04143, over 11787.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2617, pruned_loss=0.04305, over 2367809.73 frames. ], batch size: 26, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:02:05,089 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171796.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:02:21,639 INFO [finetune.py:992] (1/2) Epoch 6, batch 10650, loss[loss=0.1381, simple_loss=0.2233, pruned_loss=0.02648, over 12016.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.261, pruned_loss=0.0428, over 2362947.13 frames. ], batch size: 28, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:02:31,367 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2305, 4.5596, 4.1775, 4.8965, 4.5896, 2.9062, 4.2063, 3.0058], device='cuda:1'), covar=tensor([0.0755, 0.0765, 0.1158, 0.0461, 0.1004, 0.1551, 0.0992, 0.3129], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0374, 0.0351, 0.0276, 0.0360, 0.0265, 0.0335, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:02:38,042 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.731e+02 3.265e+02 4.072e+02 6.938e+02, threshold=6.531e+02, percent-clipped=2.0 2023-05-16 06:02:45,273 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171852.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:02:46,591 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:02:56,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4428, 5.0152, 5.3191, 4.6660, 5.0579, 4.6357, 5.3198, 5.0629], device='cuda:1'), covar=tensor([0.0333, 0.0430, 0.0500, 0.0288, 0.0329, 0.0355, 0.0342, 0.0282], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0246, 0.0268, 0.0246, 0.0240, 0.0241, 0.0220, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:02:57,138 INFO [finetune.py:992] (1/2) Epoch 6, batch 10700, loss[loss=0.1507, simple_loss=0.2393, pruned_loss=0.03107, over 12340.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2611, pruned_loss=0.04249, over 2361638.22 frames. ], batch size: 31, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:03:18,895 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171900.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:03:33,720 INFO [finetune.py:992] (1/2) Epoch 6, batch 10750, loss[loss=0.161, simple_loss=0.2476, pruned_loss=0.03717, over 12353.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2618, pruned_loss=0.04254, over 2368374.63 frames. ], batch size: 31, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:03:46,702 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171937.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:03:50,064 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.851e+02 2.826e+02 3.370e+02 4.038e+02 8.088e+02, threshold=6.740e+02, percent-clipped=7.0 2023-05-16 06:03:54,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-05-16 06:04:09,490 INFO [finetune.py:992] (1/2) Epoch 6, batch 10800, loss[loss=0.1583, simple_loss=0.242, pruned_loss=0.03735, over 11812.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2623, pruned_loss=0.04315, over 2363704.87 frames. ], batch size: 26, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:04:11,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 06:04:27,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 06:04:30,083 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171998.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:04:35,444 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172001.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:04:44,716 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0093, 6.0241, 5.7857, 5.3262, 5.1093, 5.9246, 5.5315, 5.2995], device='cuda:1'), covar=tensor([0.0662, 0.0736, 0.0607, 0.1238, 0.0650, 0.0657, 0.1323, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.0590, 0.0515, 0.0494, 0.0599, 0.0401, 0.0691, 0.0744, 0.0546], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:04:48,141 INFO [finetune.py:992] (1/2) Epoch 6, batch 10850, loss[loss=0.1686, simple_loss=0.2596, pruned_loss=0.03886, over 12318.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2623, pruned_loss=0.04335, over 2367708.83 frames. ], batch size: 34, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:05:05,514 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.817e+02 3.464e+02 4.128e+02 6.888e+02, threshold=6.928e+02, percent-clipped=1.0 2023-05-16 06:05:10,648 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172049.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:05:25,733 INFO [finetune.py:992] (1/2) Epoch 6, batch 10900, loss[loss=0.1883, simple_loss=0.2569, pruned_loss=0.05987, over 12326.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2624, pruned_loss=0.04349, over 2366729.37 frames. ], batch size: 30, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:05:44,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.30 vs. limit=5.0 2023-05-16 06:06:00,461 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172118.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:06:00,952 INFO [finetune.py:992] (1/2) Epoch 6, batch 10950, loss[loss=0.1906, simple_loss=0.2779, pruned_loss=0.05164, over 12035.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2623, pruned_loss=0.0434, over 2366366.54 frames. ], batch size: 37, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:06:17,107 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 2.767e+02 3.256e+02 4.075e+02 8.444e+02, threshold=6.513e+02, percent-clipped=1.0 2023-05-16 06:06:25,656 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172154.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:06:26,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-05-16 06:06:32,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 06:06:36,280 INFO [finetune.py:992] (1/2) Epoch 6, batch 11000, loss[loss=0.1857, simple_loss=0.2807, pruned_loss=0.04533, over 12027.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2642, pruned_loss=0.04408, over 2362230.22 frames. ], batch size: 42, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:06:43,979 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172179.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:06:49,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.49 vs. limit=5.0 2023-05-16 06:07:01,137 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172202.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:07:12,987 INFO [finetune.py:992] (1/2) Epoch 6, batch 11050, loss[loss=0.2315, simple_loss=0.3001, pruned_loss=0.08151, over 7829.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2673, pruned_loss=0.0463, over 2323703.72 frames. ], batch size: 98, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:07:28,817 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.158e+02 3.130e+02 3.953e+02 4.943e+02 1.066e+03, threshold=7.907e+02, percent-clipped=6.0 2023-05-16 06:07:47,798 INFO [finetune.py:992] (1/2) Epoch 6, batch 11100, loss[loss=0.2309, simple_loss=0.32, pruned_loss=0.07085, over 11545.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2721, pruned_loss=0.04956, over 2274641.84 frames. ], batch size: 48, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:07:58,345 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5256, 2.5415, 4.5328, 4.8517, 3.0237, 2.5297, 2.9619, 1.8749], device='cuda:1'), covar=tensor([0.1537, 0.3336, 0.0434, 0.0328, 0.1065, 0.2326, 0.2645, 0.4705], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0368, 0.0261, 0.0288, 0.0254, 0.0283, 0.0354, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:08:05,795 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172293.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:08:24,084 INFO [finetune.py:992] (1/2) Epoch 6, batch 11150, loss[loss=0.2625, simple_loss=0.3355, pruned_loss=0.09472, over 10466.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2785, pruned_loss=0.05385, over 2204638.97 frames. ], batch size: 68, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:08:26,250 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0525, 3.8522, 4.0113, 4.4235, 2.8149, 3.9507, 2.5431, 3.8859], device='cuda:1'), covar=tensor([0.1756, 0.0782, 0.0818, 0.0485, 0.1191, 0.0595, 0.1793, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0293, 0.0346, 0.0234, 0.0236, 0.0252, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:08:32,700 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4221, 4.3002, 4.3022, 4.3527, 3.9518, 4.4352, 4.4150, 4.5418], device='cuda:1'), covar=tensor([0.0207, 0.0172, 0.0170, 0.0304, 0.0709, 0.0313, 0.0164, 0.0206], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0181, 0.0178, 0.0227, 0.0226, 0.0201, 0.0163, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 06:08:40,177 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 3.335e+02 4.017e+02 5.044e+02 1.045e+03, threshold=8.034e+02, percent-clipped=3.0 2023-05-16 06:09:00,370 INFO [finetune.py:992] (1/2) Epoch 6, batch 11200, loss[loss=0.3095, simple_loss=0.3596, pruned_loss=0.1298, over 7148.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2853, pruned_loss=0.05811, over 2151927.44 frames. ], batch size: 100, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:09:04,666 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7419, 3.4006, 3.6175, 3.6957, 3.4111, 3.7430, 3.7872, 3.8406], device='cuda:1'), covar=tensor([0.0210, 0.0238, 0.0186, 0.0385, 0.0528, 0.0314, 0.0170, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0180, 0.0176, 0.0225, 0.0224, 0.0199, 0.0161, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 06:09:06,118 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3061, 3.4645, 3.2305, 3.5446, 3.3472, 2.5950, 3.2410, 2.8663], device='cuda:1'), covar=tensor([0.0716, 0.0902, 0.1187, 0.0563, 0.1222, 0.1420, 0.1013, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0364, 0.0342, 0.0266, 0.0351, 0.0258, 0.0328, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:09:25,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2023-05-16 06:09:27,818 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3000, 5.2428, 5.1066, 4.6586, 4.7150, 5.2390, 4.8999, 4.7736], device='cuda:1'), covar=tensor([0.0650, 0.0828, 0.0602, 0.1509, 0.1081, 0.0688, 0.1355, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0570, 0.0501, 0.0479, 0.0582, 0.0389, 0.0668, 0.0719, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:09:35,105 INFO [finetune.py:992] (1/2) Epoch 6, batch 11250, loss[loss=0.3199, simple_loss=0.3676, pruned_loss=0.1361, over 6764.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2928, pruned_loss=0.06337, over 2090598.52 frames. ], batch size: 100, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:09:51,987 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.255e+02 3.510e+02 4.248e+02 5.135e+02 1.202e+03, threshold=8.497e+02, percent-clipped=3.0 2023-05-16 06:10:10,373 INFO [finetune.py:992] (1/2) Epoch 6, batch 11300, loss[loss=0.3048, simple_loss=0.3649, pruned_loss=0.1223, over 6845.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.299, pruned_loss=0.06743, over 2035918.26 frames. ], batch size: 98, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:10:13,965 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172474.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:10:17,295 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3285, 3.5132, 3.2357, 3.5928, 3.4175, 2.5466, 3.2486, 2.8491], device='cuda:1'), covar=tensor([0.0807, 0.0946, 0.1308, 0.0670, 0.1320, 0.1594, 0.1114, 0.2715], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0362, 0.0340, 0.0264, 0.0349, 0.0257, 0.0326, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:10:17,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 06:10:18,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-05-16 06:10:45,719 INFO [finetune.py:992] (1/2) Epoch 6, batch 11350, loss[loss=0.2012, simple_loss=0.2889, pruned_loss=0.05676, over 12117.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3051, pruned_loss=0.07164, over 1978128.95 frames. ], batch size: 38, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:10:50,646 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7209, 2.0724, 2.8735, 3.6942, 2.2282, 3.8208, 3.6953, 3.8728], device='cuda:1'), covar=tensor([0.0152, 0.1324, 0.0422, 0.0116, 0.1165, 0.0199, 0.0260, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0199, 0.0182, 0.0113, 0.0185, 0.0172, 0.0166, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:11:00,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2023-05-16 06:11:01,354 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.533e+02 3.462e+02 4.203e+02 5.363e+02 9.829e+02, threshold=8.405e+02, percent-clipped=2.0 2023-05-16 06:11:18,477 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7803, 3.0093, 3.8448, 4.7602, 4.1992, 4.7106, 4.2652, 3.4415], device='cuda:1'), covar=tensor([0.0022, 0.0337, 0.0112, 0.0022, 0.0075, 0.0055, 0.0087, 0.0316], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0116, 0.0099, 0.0072, 0.0097, 0.0108, 0.0089, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:11:20,366 INFO [finetune.py:992] (1/2) Epoch 6, batch 11400, loss[loss=0.2975, simple_loss=0.3613, pruned_loss=0.1169, over 7381.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3081, pruned_loss=0.07382, over 1942999.19 frames. ], batch size: 98, lr: 4.50e-03, grad_scale: 16.0 2023-05-16 06:11:37,352 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172593.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:11:40,870 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172598.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:11:51,527 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2382, 3.4673, 3.2277, 3.5627, 3.3830, 2.5099, 3.2358, 2.8644], device='cuda:1'), covar=tensor([0.0826, 0.0927, 0.1217, 0.0608, 0.1140, 0.1562, 0.1093, 0.2607], device='cuda:1'), in_proj_covar=tensor([0.0288, 0.0358, 0.0336, 0.0260, 0.0346, 0.0255, 0.0323, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:11:55,218 INFO [finetune.py:992] (1/2) Epoch 6, batch 11450, loss[loss=0.2359, simple_loss=0.3151, pruned_loss=0.07833, over 10162.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3111, pruned_loss=0.07651, over 1909289.18 frames. ], batch size: 68, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:11:59,395 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172625.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:12:11,420 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172641.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:12:11,964 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.535e+02 3.484e+02 4.035e+02 4.703e+02 1.176e+03, threshold=8.069e+02, percent-clipped=1.0 2023-05-16 06:12:23,805 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172659.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:12:30,357 INFO [finetune.py:992] (1/2) Epoch 6, batch 11500, loss[loss=0.2848, simple_loss=0.3445, pruned_loss=0.1126, over 7183.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3136, pruned_loss=0.07899, over 1859575.61 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:12:35,230 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7994, 3.7861, 3.7752, 3.8512, 3.6320, 3.6843, 3.6467, 3.7464], device='cuda:1'), covar=tensor([0.0881, 0.0746, 0.1171, 0.0717, 0.1582, 0.1257, 0.0528, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0619, 0.0527, 0.0554, 0.0740, 0.0670, 0.0496, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:12:41,997 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172686.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:13:05,365 INFO [finetune.py:992] (1/2) Epoch 6, batch 11550, loss[loss=0.2565, simple_loss=0.3367, pruned_loss=0.08815, over 12061.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3163, pruned_loss=0.0813, over 1825149.58 frames. ], batch size: 42, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:13:14,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-05-16 06:13:15,591 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172734.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:13:20,750 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.324e+02 3.468e+02 4.001e+02 4.655e+02 7.886e+02, threshold=8.001e+02, percent-clipped=0.0 2023-05-16 06:13:24,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 06:13:40,067 INFO [finetune.py:992] (1/2) Epoch 6, batch 11600, loss[loss=0.3155, simple_loss=0.3736, pruned_loss=0.1287, over 6765.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3178, pruned_loss=0.0826, over 1806499.07 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:13:42,892 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8044, 3.6947, 3.8053, 3.5524, 3.7014, 3.5654, 3.7723, 3.4907], device='cuda:1'), covar=tensor([0.0351, 0.0320, 0.0320, 0.0247, 0.0318, 0.0288, 0.0283, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0227, 0.0249, 0.0228, 0.0223, 0.0223, 0.0205, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:13:43,522 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172774.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:13:55,615 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7944, 2.5174, 3.4813, 3.5853, 2.9082, 2.7356, 2.5949, 2.3893], device='cuda:1'), covar=tensor([0.1031, 0.2205, 0.0531, 0.0438, 0.0823, 0.1676, 0.2362, 0.3347], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0360, 0.0256, 0.0282, 0.0248, 0.0278, 0.0348, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:13:57,751 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172795.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:14:16,307 INFO [finetune.py:992] (1/2) Epoch 6, batch 11650, loss[loss=0.2318, simple_loss=0.3142, pruned_loss=0.07466, over 6618.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3179, pruned_loss=0.08389, over 1776567.92 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:14:18,727 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172822.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:14:32,715 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.738e+02 3.666e+02 4.160e+02 4.980e+02 7.102e+02, threshold=8.321e+02, percent-clipped=0.0 2023-05-16 06:14:34,957 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172845.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:14:43,601 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2310, 4.8986, 5.2176, 4.6699, 4.9045, 4.6966, 5.2211, 4.8978], device='cuda:1'), covar=tensor([0.0276, 0.0285, 0.0250, 0.0245, 0.0372, 0.0295, 0.0223, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0227, 0.0249, 0.0227, 0.0222, 0.0223, 0.0204, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:14:50,850 INFO [finetune.py:992] (1/2) Epoch 6, batch 11700, loss[loss=0.3048, simple_loss=0.3522, pruned_loss=0.1288, over 6271.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3175, pruned_loss=0.08444, over 1741196.22 frames. ], batch size: 101, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:15:05,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-05-16 06:15:06,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-05-16 06:15:12,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-05-16 06:15:17,194 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172906.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:15:25,695 INFO [finetune.py:992] (1/2) Epoch 6, batch 11750, loss[loss=0.2676, simple_loss=0.3316, pruned_loss=0.1018, over 6956.00 frames. ], tot_loss[loss=0.243, simple_loss=0.317, pruned_loss=0.08446, over 1732862.86 frames. ], batch size: 102, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:15:42,077 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.367e+02 3.625e+02 4.400e+02 5.167e+02 1.076e+03, threshold=8.801e+02, percent-clipped=4.0 2023-05-16 06:15:50,907 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172954.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:16:00,830 INFO [finetune.py:992] (1/2) Epoch 6, batch 11800, loss[loss=0.2105, simple_loss=0.2979, pruned_loss=0.06152, over 11062.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3194, pruned_loss=0.08648, over 1704628.87 frames. ], batch size: 55, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:16:09,219 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172981.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:16:36,310 INFO [finetune.py:992] (1/2) Epoch 6, batch 11850, loss[loss=0.2685, simple_loss=0.3311, pruned_loss=0.103, over 6977.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3206, pruned_loss=0.08692, over 1682302.83 frames. ], batch size: 99, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:16:48,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.45 vs. limit=5.0 2023-05-16 06:16:51,797 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 3.587e+02 4.059e+02 5.000e+02 7.524e+02, threshold=8.117e+02, percent-clipped=0.0 2023-05-16 06:17:05,664 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2736, 3.4910, 3.2351, 3.5132, 3.4198, 2.5541, 3.2842, 2.8996], device='cuda:1'), covar=tensor([0.0875, 0.0942, 0.1582, 0.0607, 0.1229, 0.1612, 0.1091, 0.2576], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0345, 0.0323, 0.0248, 0.0333, 0.0248, 0.0313, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:17:10,790 INFO [finetune.py:992] (1/2) Epoch 6, batch 11900, loss[loss=0.2223, simple_loss=0.2969, pruned_loss=0.07383, over 7402.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3194, pruned_loss=0.08487, over 1697210.88 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:17:25,906 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173090.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:17:45,687 INFO [finetune.py:992] (1/2) Epoch 6, batch 11950, loss[loss=0.207, simple_loss=0.2877, pruned_loss=0.06313, over 7273.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3151, pruned_loss=0.08086, over 1700324.28 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:18:01,530 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 06:18:01,876 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.108e+02 3.189e+02 3.702e+02 4.307e+02 7.149e+02, threshold=7.404e+02, percent-clipped=0.0 2023-05-16 06:18:07,446 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173150.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:18:19,543 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8733, 4.5161, 4.2485, 4.1553, 4.5616, 4.0877, 4.2466, 4.0770], device='cuda:1'), covar=tensor([0.1558, 0.1024, 0.0984, 0.1812, 0.1005, 0.1714, 0.1650, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0434, 0.0346, 0.0390, 0.0417, 0.0396, 0.0353, 0.0338], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 06:18:20,086 INFO [finetune.py:992] (1/2) Epoch 6, batch 12000, loss[loss=0.2087, simple_loss=0.2868, pruned_loss=0.06527, over 7243.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3109, pruned_loss=0.07741, over 1690461.29 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:18:20,087 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 06:18:38,371 INFO [finetune.py:1026] (1/2) Epoch 6, validation: loss=0.2919, simple_loss=0.3683, pruned_loss=0.1078, over 1020973.00 frames. 2023-05-16 06:18:38,372 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 06:19:00,963 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173201.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:19:04,373 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3448, 2.4930, 3.9583, 3.3265, 3.5972, 3.5006, 2.5249, 3.6837], device='cuda:1'), covar=tensor([0.0109, 0.0332, 0.0065, 0.0186, 0.0128, 0.0128, 0.0326, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0185, 0.0162, 0.0161, 0.0185, 0.0141, 0.0175, 0.0156], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:19:08,030 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:19:13,103 INFO [finetune.py:992] (1/2) Epoch 6, batch 12050, loss[loss=0.1974, simple_loss=0.28, pruned_loss=0.05741, over 7278.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3062, pruned_loss=0.0739, over 1697994.96 frames. ], batch size: 99, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:19:29,349 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.293e+02 2.981e+02 3.503e+02 4.122e+02 6.105e+02, threshold=7.005e+02, percent-clipped=0.0 2023-05-16 06:19:36,799 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173254.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:19:38,161 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3549, 4.9571, 5.2947, 4.7650, 5.0306, 4.8307, 5.2896, 5.0167], device='cuda:1'), covar=tensor([0.0270, 0.0371, 0.0287, 0.0237, 0.0372, 0.0303, 0.0276, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0222, 0.0242, 0.0221, 0.0216, 0.0217, 0.0198, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:19:41,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-05-16 06:19:46,390 INFO [finetune.py:992] (1/2) Epoch 6, batch 12100, loss[loss=0.2168, simple_loss=0.2874, pruned_loss=0.07311, over 7237.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3057, pruned_loss=0.07321, over 1696588.65 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 16.0 2023-05-16 06:19:50,515 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4444, 2.9491, 3.6970, 2.3351, 2.5955, 3.0468, 2.8390, 3.1316], device='cuda:1'), covar=tensor([0.0445, 0.1018, 0.0259, 0.1285, 0.1695, 0.1221, 0.1176, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0215, 0.0211, 0.0170, 0.0219, 0.0257, 0.0208, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:19:53,608 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0381, 2.0706, 2.7449, 3.0924, 2.1147, 3.1928, 3.1178, 3.1423], device='cuda:1'), covar=tensor([0.0154, 0.0928, 0.0315, 0.0141, 0.0978, 0.0170, 0.0205, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0196, 0.0176, 0.0109, 0.0181, 0.0165, 0.0160, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:19:54,149 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173281.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:20:07,476 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173302.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:20:12,075 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173309.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:20:18,240 INFO [finetune.py:992] (1/2) Epoch 6, batch 12150, loss[loss=0.2139, simple_loss=0.3088, pruned_loss=0.05945, over 10189.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3067, pruned_loss=0.07399, over 1694687.37 frames. ], batch size: 68, lr: 4.49e-03, grad_scale: 32.0 2023-05-16 06:20:24,583 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173329.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:20:26,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 06:20:32,636 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 3.251e+02 3.790e+02 4.347e+02 7.490e+02, threshold=7.581e+02, percent-clipped=1.0 2023-05-16 06:20:41,697 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3200, 3.1615, 3.2091, 3.4497, 2.6663, 3.1596, 2.6362, 2.9175], device='cuda:1'), covar=tensor([0.1784, 0.0887, 0.0941, 0.0614, 0.1081, 0.0831, 0.1719, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0253, 0.0282, 0.0331, 0.0227, 0.0231, 0.0249, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:20:49,382 INFO [finetune.py:992] (1/2) Epoch 6, batch 12200, loss[loss=0.2618, simple_loss=0.3221, pruned_loss=0.1007, over 6795.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3083, pruned_loss=0.07592, over 1650836.98 frames. ], batch size: 98, lr: 4.49e-03, grad_scale: 32.0 2023-05-16 06:20:50,671 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173370.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:20:52,450 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5416, 4.4958, 4.4083, 4.1086, 4.1234, 4.5135, 4.2556, 4.1452], device='cuda:1'), covar=tensor([0.0750, 0.0863, 0.0699, 0.1195, 0.1828, 0.0771, 0.1432, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0477, 0.0452, 0.0546, 0.0363, 0.0620, 0.0667, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 06:21:01,130 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9754, 2.1450, 2.6381, 3.0438, 2.2859, 3.1030, 3.0443, 3.1658], device='cuda:1'), covar=tensor([0.0177, 0.1109, 0.0437, 0.0155, 0.1042, 0.0256, 0.0260, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0195, 0.0176, 0.0108, 0.0181, 0.0164, 0.0159, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:21:02,921 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173390.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:21:37,540 INFO [finetune.py:992] (1/2) Epoch 7, batch 0, loss[loss=0.1862, simple_loss=0.2671, pruned_loss=0.05271, over 12273.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2671, pruned_loss=0.05271, over 12273.00 frames. ], batch size: 33, lr: 4.49e-03, grad_scale: 32.0 2023-05-16 06:21:37,540 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 06:21:55,240 INFO [finetune.py:1026] (1/2) Epoch 7, validation: loss=0.2955, simple_loss=0.3679, pruned_loss=0.1116, over 1020973.00 frames. 2023-05-16 06:21:55,241 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 06:22:20,801 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173438.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:22:23,613 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.092e+02 3.197e+02 3.790e+02 4.651e+02 1.246e+03, threshold=7.581e+02, percent-clipped=2.0 2023-05-16 06:22:30,415 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-16 06:22:31,473 INFO [finetune.py:992] (1/2) Epoch 7, batch 50, loss[loss=0.1673, simple_loss=0.2507, pruned_loss=0.04196, over 12412.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2684, pruned_loss=0.04679, over 542359.04 frames. ], batch size: 32, lr: 4.48e-03, grad_scale: 32.0 2023-05-16 06:22:59,459 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173491.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:23:01,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-05-16 06:23:06,457 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173501.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:23:07,733 INFO [finetune.py:992] (1/2) Epoch 7, batch 100, loss[loss=0.1746, simple_loss=0.2667, pruned_loss=0.04126, over 12074.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.269, pruned_loss=0.04677, over 950451.98 frames. ], batch size: 32, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:23:09,933 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173506.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:23:12,980 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0020, 3.8562, 3.9959, 4.4032, 2.9061, 3.6179, 2.3361, 4.0202], device='cuda:1'), covar=tensor([0.2095, 0.0909, 0.0984, 0.0583, 0.1224, 0.0822, 0.2265, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0257, 0.0287, 0.0337, 0.0231, 0.0234, 0.0252, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:23:35,920 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.918e+02 3.480e+02 4.159e+02 1.047e+03, threshold=6.960e+02, percent-clipped=3.0 2023-05-16 06:23:39,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 06:23:40,271 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173549.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:23:40,400 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1511, 4.7767, 4.9200, 5.0251, 4.9392, 4.9805, 4.9547, 2.9265], device='cuda:1'), covar=tensor([0.0076, 0.0050, 0.0071, 0.0050, 0.0039, 0.0079, 0.0064, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0070, 0.0073, 0.0065, 0.0055, 0.0083, 0.0071, 0.0088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:23:42,553 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173552.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 06:23:42,993 INFO [finetune.py:992] (1/2) Epoch 7, batch 150, loss[loss=0.1691, simple_loss=0.2564, pruned_loss=0.04094, over 12254.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2696, pruned_loss=0.04631, over 1262767.04 frames. ], batch size: 32, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:23:55,151 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-16 06:24:13,296 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4208, 4.9628, 5.3626, 4.7608, 5.0519, 4.8017, 5.4110, 5.0436], device='cuda:1'), covar=tensor([0.0220, 0.0328, 0.0245, 0.0240, 0.0332, 0.0320, 0.0211, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0228, 0.0248, 0.0227, 0.0222, 0.0222, 0.0203, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:24:19,662 INFO [finetune.py:992] (1/2) Epoch 7, batch 200, loss[loss=0.1867, simple_loss=0.2742, pruned_loss=0.04962, over 12161.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2679, pruned_loss=0.04534, over 1517391.79 frames. ], batch size: 39, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:24:27,544 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173614.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:24:27,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.64 vs. limit=5.0 2023-05-16 06:24:48,600 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.259e+02 2.761e+02 3.100e+02 3.797e+02 5.593e+02, threshold=6.201e+02, percent-clipped=0.0 2023-05-16 06:24:50,288 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9593, 4.9277, 4.7943, 4.7680, 4.5173, 4.9149, 4.9595, 5.1485], device='cuda:1'), covar=tensor([0.0251, 0.0160, 0.0204, 0.0387, 0.0770, 0.0402, 0.0148, 0.0191], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0169, 0.0167, 0.0213, 0.0211, 0.0188, 0.0152, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 06:24:55,729 INFO [finetune.py:992] (1/2) Epoch 7, batch 250, loss[loss=0.1578, simple_loss=0.2416, pruned_loss=0.03699, over 12330.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2668, pruned_loss=0.04517, over 1702294.12 frames. ], batch size: 31, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:25:04,503 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173665.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:25:10,257 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6255, 2.8337, 5.1327, 2.4417, 2.2540, 3.9243, 2.8123, 3.8278], device='cuda:1'), covar=tensor([0.0474, 0.1506, 0.0297, 0.1476, 0.2362, 0.1334, 0.1683, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0221, 0.0219, 0.0175, 0.0226, 0.0266, 0.0214, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:25:11,679 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173675.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 06:25:31,214 INFO [finetune.py:992] (1/2) Epoch 7, batch 300, loss[loss=0.2109, simple_loss=0.29, pruned_loss=0.06585, over 11848.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2669, pruned_loss=0.04554, over 1840955.52 frames. ], batch size: 44, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:25:32,159 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2929, 4.6007, 4.1481, 4.9110, 4.5598, 2.7840, 4.4044, 3.0227], device='cuda:1'), covar=tensor([0.0769, 0.0862, 0.1326, 0.0541, 0.1045, 0.1703, 0.0890, 0.3463], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0360, 0.0338, 0.0259, 0.0349, 0.0258, 0.0327, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:25:46,332 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4222, 4.9528, 5.3511, 4.7431, 4.9711, 4.7960, 5.4376, 5.0310], device='cuda:1'), covar=tensor([0.0236, 0.0359, 0.0271, 0.0231, 0.0388, 0.0291, 0.0209, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0233, 0.0253, 0.0231, 0.0227, 0.0227, 0.0207, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:26:00,435 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.313e+02 2.879e+02 3.329e+02 4.149e+02 7.087e+02, threshold=6.658e+02, percent-clipped=4.0 2023-05-16 06:26:05,714 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0053, 3.4471, 5.2677, 2.6640, 2.9522, 3.8949, 3.1742, 3.8773], device='cuda:1'), covar=tensor([0.0345, 0.1097, 0.0251, 0.1302, 0.1843, 0.1450, 0.1355, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0222, 0.0222, 0.0176, 0.0227, 0.0268, 0.0215, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:26:07,583 INFO [finetune.py:992] (1/2) Epoch 7, batch 350, loss[loss=0.1703, simple_loss=0.2592, pruned_loss=0.04073, over 12038.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2654, pruned_loss=0.04526, over 1963823.35 frames. ], batch size: 31, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:26:26,765 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8412, 4.2768, 3.7951, 4.5907, 4.1830, 2.7307, 4.0592, 2.8674], device='cuda:1'), covar=tensor([0.0956, 0.0866, 0.1393, 0.0502, 0.1056, 0.1669, 0.0849, 0.3480], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0361, 0.0340, 0.0261, 0.0351, 0.0260, 0.0329, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:26:43,878 INFO [finetune.py:992] (1/2) Epoch 7, batch 400, loss[loss=0.1907, simple_loss=0.2702, pruned_loss=0.05563, over 12277.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2651, pruned_loss=0.04522, over 2050776.22 frames. ], batch size: 33, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:26:46,091 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173806.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:27:09,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6926, 2.7084, 4.3292, 4.5923, 2.9262, 2.6474, 2.8339, 2.1228], device='cuda:1'), covar=tensor([0.1502, 0.2942, 0.0542, 0.0428, 0.1187, 0.2161, 0.2756, 0.4022], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0365, 0.0258, 0.0284, 0.0250, 0.0284, 0.0356, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:27:12,085 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 2.932e+02 3.449e+02 4.062e+02 1.100e+03, threshold=6.898e+02, percent-clipped=4.0 2023-05-16 06:27:15,122 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173847.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 06:27:19,200 INFO [finetune.py:992] (1/2) Epoch 7, batch 450, loss[loss=0.1816, simple_loss=0.2741, pruned_loss=0.04458, over 12182.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2653, pruned_loss=0.04476, over 2128318.13 frames. ], batch size: 35, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:27:19,913 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:27:54,958 INFO [finetune.py:992] (1/2) Epoch 7, batch 500, loss[loss=0.1867, simple_loss=0.2701, pruned_loss=0.05163, over 12276.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2647, pruned_loss=0.04454, over 2182027.04 frames. ], batch size: 37, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:28:23,982 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.971e+02 3.357e+02 3.784e+02 6.011e+02, threshold=6.715e+02, percent-clipped=0.0 2023-05-16 06:28:31,212 INFO [finetune.py:992] (1/2) Epoch 7, batch 550, loss[loss=0.162, simple_loss=0.2568, pruned_loss=0.03354, over 12166.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2631, pruned_loss=0.04356, over 2232581.23 frames. ], batch size: 36, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:28:39,902 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:28:43,470 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173970.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:29:09,943 INFO [finetune.py:992] (1/2) Epoch 7, batch 600, loss[loss=0.1883, simple_loss=0.2842, pruned_loss=0.0462, over 11649.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2628, pruned_loss=0.04365, over 2261590.50 frames. ], batch size: 48, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:29:12,248 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6470, 5.6302, 5.3495, 4.8546, 4.8409, 5.5447, 5.2170, 4.9628], device='cuda:1'), covar=tensor([0.0720, 0.0836, 0.0633, 0.1705, 0.0824, 0.0736, 0.1437, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0501, 0.0477, 0.0580, 0.0383, 0.0659, 0.0711, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:29:17,303 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174013.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:29:18,869 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174015.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:29:20,181 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174017.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:29:27,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0930, 4.1722, 2.7012, 2.4394, 3.7279, 2.1414, 3.6991, 2.9439], device='cuda:1'), covar=tensor([0.0730, 0.0565, 0.1165, 0.1596, 0.0250, 0.1557, 0.0480, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0238, 0.0172, 0.0193, 0.0131, 0.0176, 0.0188, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:29:39,004 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.752e+02 3.364e+02 3.945e+02 6.201e+02, threshold=6.728e+02, percent-clipped=0.0 2023-05-16 06:29:46,139 INFO [finetune.py:992] (1/2) Epoch 7, batch 650, loss[loss=0.197, simple_loss=0.2918, pruned_loss=0.05107, over 12011.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2632, pruned_loss=0.04348, over 2289695.47 frames. ], batch size: 42, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:30:03,458 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174076.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 06:30:04,728 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174078.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:30:22,238 INFO [finetune.py:992] (1/2) Epoch 7, batch 700, loss[loss=0.1792, simple_loss=0.2707, pruned_loss=0.04384, over 10347.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2627, pruned_loss=0.04312, over 2309927.28 frames. ], batch size: 68, lr: 4.48e-03, grad_scale: 16.0 2023-05-16 06:30:47,769 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174139.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:30:50,940 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 2.951e+02 3.360e+02 4.029e+02 6.160e+02, threshold=6.721e+02, percent-clipped=0.0 2023-05-16 06:30:52,955 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174147.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:30:57,125 INFO [finetune.py:992] (1/2) Epoch 7, batch 750, loss[loss=0.1758, simple_loss=0.2696, pruned_loss=0.04098, over 12277.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2627, pruned_loss=0.04318, over 2330622.86 frames. ], batch size: 37, lr: 4.48e-03, grad_scale: 8.0 2023-05-16 06:31:27,041 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7383, 2.8401, 4.5074, 4.6469, 2.6633, 2.6606, 2.9326, 2.1176], device='cuda:1'), covar=tensor([0.1424, 0.3040, 0.0476, 0.0476, 0.1363, 0.2185, 0.2622, 0.4051], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0370, 0.0260, 0.0287, 0.0254, 0.0287, 0.0359, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:31:27,574 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174195.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:31:31,562 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174200.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:31:33,574 INFO [finetune.py:992] (1/2) Epoch 7, batch 800, loss[loss=0.1633, simple_loss=0.2528, pruned_loss=0.03695, over 12349.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2626, pruned_loss=0.04315, over 2338525.83 frames. ], batch size: 31, lr: 4.48e-03, grad_scale: 8.0 2023-05-16 06:32:03,472 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.898e+02 3.453e+02 4.336e+02 9.975e+02, threshold=6.906e+02, percent-clipped=2.0 2023-05-16 06:32:09,685 INFO [finetune.py:992] (1/2) Epoch 7, batch 850, loss[loss=0.1761, simple_loss=0.276, pruned_loss=0.03812, over 12354.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2628, pruned_loss=0.04326, over 2340716.02 frames. ], batch size: 36, lr: 4.48e-03, grad_scale: 8.0 2023-05-16 06:32:10,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 06:32:21,762 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174270.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 06:32:25,374 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0776, 5.0245, 4.8815, 4.9343, 4.5744, 5.0526, 5.0613, 5.2924], device='cuda:1'), covar=tensor([0.0266, 0.0143, 0.0205, 0.0300, 0.0759, 0.0263, 0.0167, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0179, 0.0177, 0.0224, 0.0223, 0.0199, 0.0162, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 06:32:31,159 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174283.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:32:36,731 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174291.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:32:45,184 INFO [finetune.py:992] (1/2) Epoch 7, batch 900, loss[loss=0.1846, simple_loss=0.2729, pruned_loss=0.04812, over 12131.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2622, pruned_loss=0.04297, over 2350811.10 frames. ], batch size: 38, lr: 4.48e-03, grad_scale: 8.0 2023-05-16 06:32:47,471 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8344, 3.5873, 3.2761, 3.3192, 2.9579, 3.0108, 3.7378, 2.3999], device='cuda:1'), covar=tensor([0.0301, 0.0143, 0.0186, 0.0177, 0.0375, 0.0310, 0.0123, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0150, 0.0146, 0.0177, 0.0194, 0.0186, 0.0154, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 06:32:56,723 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174318.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:33:08,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 06:33:14,759 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.707e+02 3.185e+02 3.751e+02 7.083e+02, threshold=6.370e+02, percent-clipped=1.0 2023-05-16 06:33:14,985 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174344.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:33:20,663 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174352.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 06:33:21,149 INFO [finetune.py:992] (1/2) Epoch 7, batch 950, loss[loss=0.2242, simple_loss=0.2997, pruned_loss=0.07432, over 8120.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2622, pruned_loss=0.04287, over 2356347.69 frames. ], batch size: 98, lr: 4.48e-03, grad_scale: 8.0 2023-05-16 06:33:24,334 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1193, 4.5363, 4.0082, 4.8914, 4.5156, 2.4796, 4.0265, 2.9659], device='cuda:1'), covar=tensor([0.0857, 0.0762, 0.1338, 0.0474, 0.1035, 0.1909, 0.1002, 0.3278], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0365, 0.0344, 0.0266, 0.0355, 0.0261, 0.0333, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:33:34,780 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174371.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 06:33:36,178 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174373.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:33:47,017 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174388.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:33:57,759 INFO [finetune.py:992] (1/2) Epoch 7, batch 1000, loss[loss=0.188, simple_loss=0.2752, pruned_loss=0.05039, over 10640.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2623, pruned_loss=0.0428, over 2366216.46 frames. ], batch size: 68, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:34:27,212 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.792e+02 3.331e+02 4.189e+02 8.922e+02, threshold=6.662e+02, percent-clipped=2.0 2023-05-16 06:34:30,960 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174449.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:34:33,401 INFO [finetune.py:992] (1/2) Epoch 7, batch 1050, loss[loss=0.1723, simple_loss=0.2528, pruned_loss=0.04594, over 12352.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2618, pruned_loss=0.04246, over 2372590.38 frames. ], batch size: 31, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:34:47,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 06:35:03,858 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174495.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:35:09,341 INFO [finetune.py:992] (1/2) Epoch 7, batch 1100, loss[loss=0.1814, simple_loss=0.2717, pruned_loss=0.04556, over 12146.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2608, pruned_loss=0.04229, over 2377252.63 frames. ], batch size: 36, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:35:26,498 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2753, 4.7920, 5.1940, 4.6024, 4.8625, 4.7066, 5.2582, 4.9777], device='cuda:1'), covar=tensor([0.0265, 0.0416, 0.0374, 0.0278, 0.0365, 0.0301, 0.0242, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0243, 0.0267, 0.0240, 0.0236, 0.0235, 0.0216, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:35:39,032 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.923e+02 3.347e+02 4.160e+02 1.417e+03, threshold=6.693e+02, percent-clipped=6.0 2023-05-16 06:35:45,194 INFO [finetune.py:992] (1/2) Epoch 7, batch 1150, loss[loss=0.2052, simple_loss=0.2918, pruned_loss=0.0593, over 12027.00 frames. ], tot_loss[loss=0.172, simple_loss=0.26, pruned_loss=0.04197, over 2381677.38 frames. ], batch size: 40, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:35:58,846 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3930, 3.2438, 4.7183, 2.2764, 2.6156, 3.5555, 3.1642, 3.6575], device='cuda:1'), covar=tensor([0.0437, 0.1083, 0.0394, 0.1454, 0.1990, 0.1380, 0.1243, 0.1076], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0226, 0.0229, 0.0179, 0.0230, 0.0275, 0.0218, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:36:12,724 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0922, 5.9020, 5.4231, 5.5274, 5.9474, 5.3081, 5.4820, 5.5110], device='cuda:1'), covar=tensor([0.1479, 0.0847, 0.1026, 0.1700, 0.0942, 0.1967, 0.1866, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0462, 0.0372, 0.0423, 0.0449, 0.0420, 0.0381, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:36:21,264 INFO [finetune.py:992] (1/2) Epoch 7, batch 1200, loss[loss=0.1708, simple_loss=0.2569, pruned_loss=0.04239, over 11191.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2609, pruned_loss=0.04264, over 2378265.57 frames. ], batch size: 55, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:36:47,038 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174639.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:36:50,613 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.797e+02 3.322e+02 3.820e+02 7.213e+02, threshold=6.643e+02, percent-clipped=1.0 2023-05-16 06:36:52,773 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174647.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:36:56,939 INFO [finetune.py:992] (1/2) Epoch 7, batch 1250, loss[loss=0.1784, simple_loss=0.2705, pruned_loss=0.04312, over 12153.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2602, pruned_loss=0.04233, over 2369623.20 frames. ], batch size: 36, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:37:10,634 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174671.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:37:12,088 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174673.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:37:33,265 INFO [finetune.py:992] (1/2) Epoch 7, batch 1300, loss[loss=0.134, simple_loss=0.2184, pruned_loss=0.02479, over 12296.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2597, pruned_loss=0.04204, over 2371015.44 frames. ], batch size: 28, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:37:34,939 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174705.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:37:45,132 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174719.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:37:46,527 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174721.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:38:02,924 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.824e+02 3.276e+02 4.021e+02 7.300e+02, threshold=6.552e+02, percent-clipped=2.0 2023-05-16 06:38:03,022 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174744.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:38:10,032 INFO [finetune.py:992] (1/2) Epoch 7, batch 1350, loss[loss=0.1911, simple_loss=0.2814, pruned_loss=0.05041, over 11770.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2594, pruned_loss=0.04177, over 2382868.34 frames. ], batch size: 44, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:38:19,650 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174766.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:38:40,200 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174795.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:38:46,040 INFO [finetune.py:992] (1/2) Epoch 7, batch 1400, loss[loss=0.168, simple_loss=0.2481, pruned_loss=0.04398, over 12175.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2601, pruned_loss=0.04199, over 2383265.84 frames. ], batch size: 31, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:38:52,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7709, 2.8197, 4.3342, 4.5508, 2.8175, 2.5248, 2.9875, 2.0470], device='cuda:1'), covar=tensor([0.1383, 0.2892, 0.0534, 0.0403, 0.1192, 0.2174, 0.2534, 0.4037], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0370, 0.0260, 0.0286, 0.0254, 0.0286, 0.0359, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:38:55,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 06:38:59,933 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9673, 4.6409, 4.5528, 4.8084, 4.5883, 4.8874, 4.8302, 2.5031], device='cuda:1'), covar=tensor([0.0115, 0.0066, 0.0114, 0.0063, 0.0056, 0.0085, 0.0075, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0077, 0.0069, 0.0058, 0.0087, 0.0075, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:39:15,459 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174843.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:39:16,075 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.886e+02 3.412e+02 3.861e+02 6.454e+02, threshold=6.823e+02, percent-clipped=0.0 2023-05-16 06:39:22,177 INFO [finetune.py:992] (1/2) Epoch 7, batch 1450, loss[loss=0.1636, simple_loss=0.2503, pruned_loss=0.03843, over 12246.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2607, pruned_loss=0.04213, over 2384642.26 frames. ], batch size: 32, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:39:34,497 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4583, 5.2741, 5.4240, 5.4250, 5.0280, 5.1356, 4.8806, 5.4169], device='cuda:1'), covar=tensor([0.0676, 0.0584, 0.0697, 0.0523, 0.2045, 0.1200, 0.0510, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0643, 0.0552, 0.0582, 0.0780, 0.0692, 0.0518, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:39:57,717 INFO [finetune.py:992] (1/2) Epoch 7, batch 1500, loss[loss=0.1555, simple_loss=0.2419, pruned_loss=0.03456, over 12296.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2621, pruned_loss=0.04285, over 2379380.23 frames. ], batch size: 33, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:40:23,176 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174939.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:40:26,543 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.872e+02 3.588e+02 4.496e+02 1.019e+03, threshold=7.175e+02, percent-clipped=4.0 2023-05-16 06:40:28,771 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174947.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 06:40:32,850 INFO [finetune.py:992] (1/2) Epoch 7, batch 1550, loss[loss=0.1567, simple_loss=0.2373, pruned_loss=0.03802, over 12034.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2618, pruned_loss=0.04292, over 2380915.55 frames. ], batch size: 31, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:40:57,953 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174987.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:41:03,693 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174995.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:41:09,613 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-05-16 06:41:09,684 INFO [finetune.py:992] (1/2) Epoch 7, batch 1600, loss[loss=0.1615, simple_loss=0.2417, pruned_loss=0.04068, over 12180.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2611, pruned_loss=0.04267, over 2389291.15 frames. ], batch size: 29, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:41:39,461 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.171e+02 2.821e+02 3.320e+02 3.769e+02 7.996e+02, threshold=6.639e+02, percent-clipped=1.0 2023-05-16 06:41:39,590 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175044.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:41:45,897 INFO [finetune.py:992] (1/2) Epoch 7, batch 1650, loss[loss=0.1677, simple_loss=0.2422, pruned_loss=0.0466, over 12333.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2619, pruned_loss=0.04297, over 2381804.46 frames. ], batch size: 30, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:41:51,863 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175061.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:42:11,150 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 06:42:13,608 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175092.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:42:22,025 INFO [finetune.py:992] (1/2) Epoch 7, batch 1700, loss[loss=0.172, simple_loss=0.2592, pruned_loss=0.04242, over 12246.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2626, pruned_loss=0.04345, over 2377766.32 frames. ], batch size: 32, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:42:47,881 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175139.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:42:51,243 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.856e+02 3.271e+02 4.086e+02 1.198e+03, threshold=6.542e+02, percent-clipped=2.0 2023-05-16 06:42:57,746 INFO [finetune.py:992] (1/2) Epoch 7, batch 1750, loss[loss=0.1554, simple_loss=0.2479, pruned_loss=0.03143, over 12200.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2628, pruned_loss=0.04298, over 2376633.71 frames. ], batch size: 35, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:43:03,609 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6236, 2.4125, 3.1841, 4.5167, 2.2726, 4.5084, 4.5498, 4.7970], device='cuda:1'), covar=tensor([0.0082, 0.1246, 0.0427, 0.0097, 0.1299, 0.0187, 0.0110, 0.0047], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0201, 0.0183, 0.0113, 0.0188, 0.0174, 0.0171, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:43:14,365 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1586, 5.0354, 4.9570, 5.0129, 4.5971, 5.0955, 5.1432, 5.3131], device='cuda:1'), covar=tensor([0.0198, 0.0146, 0.0189, 0.0265, 0.0701, 0.0250, 0.0143, 0.0149], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0184, 0.0181, 0.0232, 0.0231, 0.0205, 0.0167, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 06:43:32,357 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175200.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:43:34,295 INFO [finetune.py:992] (1/2) Epoch 7, batch 1800, loss[loss=0.1922, simple_loss=0.28, pruned_loss=0.05218, over 12182.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2625, pruned_loss=0.04293, over 2377065.61 frames. ], batch size: 35, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:43:35,205 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6677, 2.4202, 3.2380, 4.5836, 2.4161, 4.5338, 4.4874, 4.8128], device='cuda:1'), covar=tensor([0.0082, 0.1272, 0.0431, 0.0090, 0.1208, 0.0187, 0.0147, 0.0051], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0202, 0.0184, 0.0113, 0.0188, 0.0174, 0.0171, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:43:37,339 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175207.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:43:37,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 06:44:03,190 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.769e+02 3.289e+02 3.861e+02 5.923e+02, threshold=6.578e+02, percent-clipped=0.0 2023-05-16 06:44:09,691 INFO [finetune.py:992] (1/2) Epoch 7, batch 1850, loss[loss=0.1711, simple_loss=0.2638, pruned_loss=0.03921, over 12081.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2617, pruned_loss=0.04269, over 2389767.23 frames. ], batch size: 32, lr: 4.47e-03, grad_scale: 8.0 2023-05-16 06:44:15,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 06:44:21,066 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175268.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:44:45,932 INFO [finetune.py:992] (1/2) Epoch 7, batch 1900, loss[loss=0.1838, simple_loss=0.2729, pruned_loss=0.04738, over 12308.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2614, pruned_loss=0.04238, over 2393193.60 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:45:15,874 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 2.899e+02 3.482e+02 3.970e+02 5.767e+02, threshold=6.964e+02, percent-clipped=0.0 2023-05-16 06:45:22,366 INFO [finetune.py:992] (1/2) Epoch 7, batch 1950, loss[loss=0.16, simple_loss=0.2459, pruned_loss=0.03707, over 12098.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.261, pruned_loss=0.04208, over 2398062.86 frames. ], batch size: 32, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:45:28,247 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175361.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:45:34,551 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175370.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:45:34,621 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2093, 3.1843, 3.1914, 3.5568, 2.6993, 3.1740, 2.4027, 3.0694], device='cuda:1'), covar=tensor([0.1636, 0.0779, 0.0907, 0.0530, 0.1008, 0.0703, 0.1660, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0261, 0.0290, 0.0350, 0.0234, 0.0238, 0.0255, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:45:51,450 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175394.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:45:58,659 INFO [finetune.py:992] (1/2) Epoch 7, batch 2000, loss[loss=0.1454, simple_loss=0.228, pruned_loss=0.0314, over 12328.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2603, pruned_loss=0.0421, over 2389386.45 frames. ], batch size: 30, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:45:59,670 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7422, 2.5444, 3.5877, 3.6826, 2.7640, 2.6136, 2.5991, 2.3305], device='cuda:1'), covar=tensor([0.1246, 0.2539, 0.0611, 0.0538, 0.1078, 0.2050, 0.2504, 0.3552], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0372, 0.0260, 0.0287, 0.0254, 0.0286, 0.0360, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:46:02,917 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175409.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:46:18,649 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175431.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:46:25,878 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7398, 2.9363, 3.8205, 4.6554, 4.0967, 4.7044, 3.9620, 3.4969], device='cuda:1'), covar=tensor([0.0030, 0.0323, 0.0133, 0.0046, 0.0102, 0.0062, 0.0087, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0122, 0.0103, 0.0075, 0.0102, 0.0113, 0.0095, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:46:27,811 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.722e+02 3.209e+02 4.046e+02 9.248e+02, threshold=6.418e+02, percent-clipped=0.0 2023-05-16 06:46:34,319 INFO [finetune.py:992] (1/2) Epoch 7, batch 2050, loss[loss=0.1601, simple_loss=0.2475, pruned_loss=0.03633, over 12181.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2596, pruned_loss=0.04175, over 2379642.58 frames. ], batch size: 31, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:46:35,175 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0494, 6.0345, 5.8526, 5.3700, 5.1590, 5.9683, 5.5844, 5.3993], device='cuda:1'), covar=tensor([0.0706, 0.0826, 0.0613, 0.1438, 0.0587, 0.0621, 0.1323, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0579, 0.0510, 0.0488, 0.0591, 0.0390, 0.0678, 0.0729, 0.0538], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:46:35,951 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175455.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:46:55,073 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175481.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:47:04,914 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175495.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:47:10,608 INFO [finetune.py:992] (1/2) Epoch 7, batch 2100, loss[loss=0.1699, simple_loss=0.2633, pruned_loss=0.03826, over 12303.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2603, pruned_loss=0.04199, over 2377725.45 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:47:38,353 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175542.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:47:39,582 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.017e+02 2.816e+02 3.398e+02 4.015e+02 1.115e+03, threshold=6.795e+02, percent-clipped=4.0 2023-05-16 06:47:45,464 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3594, 5.1732, 5.3249, 5.3543, 4.9273, 4.9495, 4.7902, 5.2809], device='cuda:1'), covar=tensor([0.0640, 0.0514, 0.0690, 0.0466, 0.1886, 0.1280, 0.0499, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0646, 0.0556, 0.0586, 0.0787, 0.0697, 0.0520, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:47:46,735 INFO [finetune.py:992] (1/2) Epoch 7, batch 2150, loss[loss=0.1639, simple_loss=0.2588, pruned_loss=0.0345, over 12329.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2611, pruned_loss=0.04192, over 2382916.08 frames. ], batch size: 36, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:47:53,949 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175563.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:48:11,720 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8968, 3.7267, 3.3029, 3.3933, 3.0995, 3.0365, 3.8374, 2.4743], device='cuda:1'), covar=tensor([0.0276, 0.0114, 0.0156, 0.0151, 0.0327, 0.0273, 0.0087, 0.0394], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0154, 0.0150, 0.0180, 0.0196, 0.0190, 0.0156, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 06:48:14,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 06:48:22,512 INFO [finetune.py:992] (1/2) Epoch 7, batch 2200, loss[loss=0.1562, simple_loss=0.2343, pruned_loss=0.03911, over 11998.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2612, pruned_loss=0.04183, over 2383074.50 frames. ], batch size: 28, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:48:24,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.28 vs. limit=5.0 2023-05-16 06:48:31,857 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3065, 4.9199, 5.2526, 4.7181, 4.8887, 4.7480, 5.3080, 4.9733], device='cuda:1'), covar=tensor([0.0226, 0.0309, 0.0281, 0.0212, 0.0350, 0.0278, 0.0198, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0248, 0.0275, 0.0245, 0.0242, 0.0243, 0.0222, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:48:44,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 06:48:52,282 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.766e+02 3.230e+02 3.814e+02 1.083e+03, threshold=6.460e+02, percent-clipped=3.0 2023-05-16 06:48:58,844 INFO [finetune.py:992] (1/2) Epoch 7, batch 2250, loss[loss=0.1787, simple_loss=0.2747, pruned_loss=0.04137, over 12309.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2607, pruned_loss=0.04166, over 2383796.04 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:49:04,628 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2623, 4.8690, 5.0095, 5.0530, 4.8714, 5.1165, 4.9014, 2.6987], device='cuda:1'), covar=tensor([0.0089, 0.0058, 0.0074, 0.0060, 0.0044, 0.0066, 0.0077, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0077, 0.0069, 0.0058, 0.0088, 0.0075, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:49:34,092 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 06:49:35,056 INFO [finetune.py:992] (1/2) Epoch 7, batch 2300, loss[loss=0.1658, simple_loss=0.261, pruned_loss=0.03526, over 12171.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2607, pruned_loss=0.04176, over 2371265.18 frames. ], batch size: 36, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:49:40,073 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175710.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:49:51,230 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175726.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:50:04,147 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.742e+02 3.217e+02 3.891e+02 8.665e+02, threshold=6.435e+02, percent-clipped=4.0 2023-05-16 06:50:09,179 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175750.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:50:11,144 INFO [finetune.py:992] (1/2) Epoch 7, batch 2350, loss[loss=0.1845, simple_loss=0.2816, pruned_loss=0.0437, over 12156.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2607, pruned_loss=0.04198, over 2371703.09 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:50:22,725 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8248, 3.5296, 3.1611, 3.1940, 2.9388, 2.8289, 3.7091, 2.3393], device='cuda:1'), covar=tensor([0.0265, 0.0139, 0.0166, 0.0167, 0.0331, 0.0285, 0.0102, 0.0379], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0153, 0.0151, 0.0181, 0.0196, 0.0191, 0.0156, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 06:50:23,992 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175771.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:50:41,127 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175795.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:50:46,820 INFO [finetune.py:992] (1/2) Epoch 7, batch 2400, loss[loss=0.1653, simple_loss=0.2502, pruned_loss=0.04019, over 12113.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.26, pruned_loss=0.04185, over 2365053.93 frames. ], batch size: 33, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:50:58,953 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9871, 4.5502, 4.6634, 4.6927, 4.6228, 4.7919, 4.6196, 2.3888], device='cuda:1'), covar=tensor([0.0146, 0.0084, 0.0126, 0.0083, 0.0062, 0.0110, 0.0096, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0077, 0.0069, 0.0058, 0.0088, 0.0075, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:51:10,905 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175837.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:51:15,170 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175843.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:51:15,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.658e+02 2.949e+02 3.599e+02 5.995e+02, threshold=5.899e+02, percent-clipped=0.0 2023-05-16 06:51:22,882 INFO [finetune.py:992] (1/2) Epoch 7, batch 2450, loss[loss=0.1543, simple_loss=0.2332, pruned_loss=0.03771, over 12172.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2601, pruned_loss=0.04191, over 2365837.12 frames. ], batch size: 29, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:51:30,064 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175863.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:51:42,657 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 06:51:43,153 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175881.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:51:59,036 INFO [finetune.py:992] (1/2) Epoch 7, batch 2500, loss[loss=0.1883, simple_loss=0.2819, pruned_loss=0.04735, over 12151.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2604, pruned_loss=0.04229, over 2360481.57 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:52:01,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 06:52:02,096 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3021, 4.4869, 4.0394, 4.8680, 4.4847, 2.8311, 4.2405, 3.0306], device='cuda:1'), covar=tensor([0.0785, 0.0862, 0.1421, 0.0409, 0.1116, 0.1739, 0.0984, 0.3184], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0370, 0.0347, 0.0270, 0.0356, 0.0262, 0.0334, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:52:03,355 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6851, 2.7550, 3.3264, 4.6161, 2.5167, 4.6369, 4.6094, 4.8304], device='cuda:1'), covar=tensor([0.0121, 0.1099, 0.0409, 0.0118, 0.1156, 0.0169, 0.0121, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0204, 0.0184, 0.0114, 0.0188, 0.0176, 0.0172, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:52:04,629 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175911.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:52:22,072 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6079, 4.7702, 4.2903, 4.9666, 4.7279, 3.1550, 4.3893, 3.2154], device='cuda:1'), covar=tensor([0.0680, 0.0717, 0.1282, 0.0465, 0.0978, 0.1467, 0.1032, 0.3024], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0370, 0.0347, 0.0270, 0.0355, 0.0262, 0.0333, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:52:26,804 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175942.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:52:28,044 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.858e+02 3.344e+02 3.806e+02 7.648e+02, threshold=6.688e+02, percent-clipped=5.0 2023-05-16 06:52:28,967 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175945.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:52:34,496 INFO [finetune.py:992] (1/2) Epoch 7, batch 2550, loss[loss=0.1634, simple_loss=0.2549, pruned_loss=0.03589, over 11651.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2607, pruned_loss=0.04243, over 2356048.59 frames. ], batch size: 48, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:52:48,825 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9716, 5.9706, 5.7126, 5.2966, 5.1524, 5.8479, 5.4595, 5.2644], device='cuda:1'), covar=tensor([0.0633, 0.0767, 0.0664, 0.1331, 0.0710, 0.0734, 0.1537, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0576, 0.0508, 0.0489, 0.0589, 0.0388, 0.0675, 0.0727, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 06:53:14,411 INFO [finetune.py:992] (1/2) Epoch 7, batch 2600, loss[loss=0.1875, simple_loss=0.2778, pruned_loss=0.04861, over 12055.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2613, pruned_loss=0.04235, over 2368040.27 frames. ], batch size: 42, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:53:16,736 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176006.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:53:22,358 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5406, 5.0582, 5.4746, 4.8789, 5.1141, 4.9213, 5.5656, 5.0633], device='cuda:1'), covar=tensor([0.0207, 0.0341, 0.0267, 0.0251, 0.0342, 0.0336, 0.0175, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0248, 0.0273, 0.0245, 0.0242, 0.0243, 0.0222, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:53:23,867 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7308, 3.1406, 3.8034, 4.6700, 4.0483, 4.6959, 3.8350, 3.5058], device='cuda:1'), covar=tensor([0.0025, 0.0278, 0.0114, 0.0037, 0.0107, 0.0045, 0.0121, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0120, 0.0101, 0.0075, 0.0102, 0.0112, 0.0094, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 06:53:30,767 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176026.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:53:44,041 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 2.756e+02 3.186e+02 3.876e+02 8.146e+02, threshold=6.371e+02, percent-clipped=1.0 2023-05-16 06:53:48,529 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176050.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:53:50,360 INFO [finetune.py:992] (1/2) Epoch 7, batch 2650, loss[loss=0.2086, simple_loss=0.2969, pruned_loss=0.06017, over 11817.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2611, pruned_loss=0.04283, over 2365142.15 frames. ], batch size: 44, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:53:59,677 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176066.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:54:05,307 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176074.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:54:09,110 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4113, 2.4527, 3.0618, 4.4016, 2.1666, 4.4304, 4.3117, 4.5748], device='cuda:1'), covar=tensor([0.0138, 0.1220, 0.0569, 0.0130, 0.1451, 0.0185, 0.0160, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0205, 0.0185, 0.0115, 0.0190, 0.0177, 0.0173, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:54:22,508 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176098.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:54:25,941 INFO [finetune.py:992] (1/2) Epoch 7, batch 2700, loss[loss=0.1893, simple_loss=0.2747, pruned_loss=0.05196, over 12147.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2606, pruned_loss=0.04243, over 2373600.56 frames. ], batch size: 39, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:54:29,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 06:54:30,357 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5193, 3.3935, 3.0931, 3.1006, 2.7406, 2.6638, 3.4183, 2.1591], device='cuda:1'), covar=tensor([0.0328, 0.0133, 0.0158, 0.0189, 0.0415, 0.0324, 0.0127, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0154, 0.0150, 0.0181, 0.0197, 0.0191, 0.0158, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:54:50,724 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176137.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:54:55,647 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.627e+02 3.045e+02 3.681e+02 6.868e+02, threshold=6.091e+02, percent-clipped=1.0 2023-05-16 06:55:01,878 INFO [finetune.py:992] (1/2) Epoch 7, batch 2750, loss[loss=0.1659, simple_loss=0.2557, pruned_loss=0.038, over 12182.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.26, pruned_loss=0.04219, over 2375515.66 frames. ], batch size: 35, lr: 4.46e-03, grad_scale: 16.0 2023-05-16 06:55:25,358 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176185.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:55:36,568 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1882, 4.9014, 5.2205, 4.5586, 4.8720, 4.5580, 5.1545, 4.8884], device='cuda:1'), covar=tensor([0.0304, 0.0373, 0.0348, 0.0281, 0.0348, 0.0384, 0.0340, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0249, 0.0273, 0.0245, 0.0241, 0.0243, 0.0222, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 06:55:38,585 INFO [finetune.py:992] (1/2) Epoch 7, batch 2800, loss[loss=0.1447, simple_loss=0.227, pruned_loss=0.03122, over 12204.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2596, pruned_loss=0.04229, over 2376641.81 frames. ], batch size: 29, lr: 4.46e-03, grad_scale: 8.0 2023-05-16 06:56:02,883 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176237.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:56:03,006 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5072, 2.9375, 3.8260, 2.3758, 2.6470, 3.0817, 2.8709, 3.1689], device='cuda:1'), covar=tensor([0.0616, 0.1057, 0.0423, 0.1132, 0.1618, 0.1347, 0.1197, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0229, 0.0234, 0.0180, 0.0233, 0.0280, 0.0222, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:56:06,386 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176242.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:56:07,746 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4334, 2.3248, 2.9849, 4.4212, 2.1011, 4.4774, 4.4215, 4.6190], device='cuda:1'), covar=tensor([0.0130, 0.1355, 0.0590, 0.0140, 0.1462, 0.0206, 0.0154, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0204, 0.0184, 0.0114, 0.0189, 0.0176, 0.0172, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:56:08,233 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.792e+02 3.403e+02 4.133e+02 6.843e+02, threshold=6.805e+02, percent-clipped=5.0 2023-05-16 06:56:14,089 INFO [finetune.py:992] (1/2) Epoch 7, batch 2850, loss[loss=0.1785, simple_loss=0.2674, pruned_loss=0.04485, over 12358.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2604, pruned_loss=0.04278, over 2362644.48 frames. ], batch size: 35, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 06:56:17,157 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176257.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:56:22,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-16 06:56:41,200 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0684, 5.0053, 4.8845, 4.9518, 4.5270, 5.1111, 5.1336, 5.2121], device='cuda:1'), covar=tensor([0.0205, 0.0140, 0.0178, 0.0266, 0.0722, 0.0214, 0.0134, 0.0176], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0189, 0.0186, 0.0236, 0.0236, 0.0210, 0.0171, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 06:56:48,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 06:56:48,888 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176301.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:56:50,162 INFO [finetune.py:992] (1/2) Epoch 7, batch 2900, loss[loss=0.2016, simple_loss=0.2861, pruned_loss=0.05853, over 12108.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2607, pruned_loss=0.04248, over 2370210.73 frames. ], batch size: 38, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 06:56:50,370 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176303.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:57:01,817 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176318.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:57:20,291 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.685e+02 3.197e+02 3.783e+02 7.095e+02, threshold=6.394e+02, percent-clipped=1.0 2023-05-16 06:57:25,838 INFO [finetune.py:992] (1/2) Epoch 7, batch 2950, loss[loss=0.1728, simple_loss=0.2663, pruned_loss=0.03962, over 12024.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2615, pruned_loss=0.043, over 2372961.34 frames. ], batch size: 42, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 06:57:35,314 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176366.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:57:55,075 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4449, 3.4380, 3.1363, 3.0318, 2.7516, 2.5086, 3.4815, 2.2274], device='cuda:1'), covar=tensor([0.0331, 0.0122, 0.0167, 0.0198, 0.0335, 0.0351, 0.0113, 0.0416], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0155, 0.0151, 0.0181, 0.0196, 0.0191, 0.0158, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:58:01,733 INFO [finetune.py:992] (1/2) Epoch 7, batch 3000, loss[loss=0.1434, simple_loss=0.2287, pruned_loss=0.02909, over 12331.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2608, pruned_loss=0.04242, over 2371733.32 frames. ], batch size: 30, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 06:58:01,733 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 06:58:19,719 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2795, 2.2431, 3.5398, 2.9929, 3.3968, 3.2780, 2.5897, 3.4278], device='cuda:1'), covar=tensor([0.0072, 0.0290, 0.0052, 0.0182, 0.0105, 0.0100, 0.0241, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0196, 0.0177, 0.0171, 0.0198, 0.0151, 0.0186, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:58:21,072 INFO [finetune.py:1026] (1/2) Epoch 7, validation: loss=0.325, simple_loss=0.3994, pruned_loss=0.1253, over 1020973.00 frames. 2023-05-16 06:58:21,073 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 06:58:29,051 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176414.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:58:51,835 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 2.842e+02 3.241e+02 3.824e+02 6.982e+02, threshold=6.482e+02, percent-clipped=1.0 2023-05-16 06:58:57,480 INFO [finetune.py:992] (1/2) Epoch 7, batch 3050, loss[loss=0.1879, simple_loss=0.2725, pruned_loss=0.05162, over 12106.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2602, pruned_loss=0.04223, over 2372251.25 frames. ], batch size: 39, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 06:59:06,783 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176466.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:59:31,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 06:59:32,999 INFO [finetune.py:992] (1/2) Epoch 7, batch 3100, loss[loss=0.1495, simple_loss=0.2322, pruned_loss=0.03344, over 12343.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2592, pruned_loss=0.04203, over 2368628.01 frames. ], batch size: 30, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 06:59:38,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3052, 3.1050, 4.6754, 2.4278, 2.5804, 3.5002, 2.9852, 3.6245], device='cuda:1'), covar=tensor([0.0466, 0.1197, 0.0258, 0.1256, 0.2087, 0.1427, 0.1442, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0230, 0.0235, 0.0180, 0.0233, 0.0282, 0.0223, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 06:59:50,478 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176527.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 06:59:57,477 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176537.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:00:03,722 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.708e+02 3.147e+02 3.794e+02 6.652e+02, threshold=6.294e+02, percent-clipped=2.0 2023-05-16 07:00:09,311 INFO [finetune.py:992] (1/2) Epoch 7, batch 3150, loss[loss=0.1509, simple_loss=0.235, pruned_loss=0.03342, over 11816.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2593, pruned_loss=0.0421, over 2364828.43 frames. ], batch size: 26, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:00:32,571 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176585.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:00:41,620 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176598.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:00:44,003 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176601.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:00:45,249 INFO [finetune.py:992] (1/2) Epoch 7, batch 3200, loss[loss=0.17, simple_loss=0.2662, pruned_loss=0.03691, over 12315.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2598, pruned_loss=0.04173, over 2373187.97 frames. ], batch size: 34, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:00:52,601 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176613.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:01:01,494 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176625.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:01:15,394 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.065e+02 2.748e+02 3.297e+02 3.969e+02 7.576e+02, threshold=6.593e+02, percent-clipped=1.0 2023-05-16 07:01:18,173 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176649.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:01:18,976 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1333, 4.7966, 4.9300, 4.9521, 4.7945, 4.9644, 4.9330, 2.5379], device='cuda:1'), covar=tensor([0.0067, 0.0054, 0.0071, 0.0059, 0.0045, 0.0079, 0.0059, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0076, 0.0069, 0.0057, 0.0087, 0.0075, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:01:20,947 INFO [finetune.py:992] (1/2) Epoch 7, batch 3250, loss[loss=0.1511, simple_loss=0.2417, pruned_loss=0.03021, over 12187.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2595, pruned_loss=0.04177, over 2365509.35 frames. ], batch size: 31, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:01:29,759 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176665.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:01:44,990 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176686.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:01:57,486 INFO [finetune.py:992] (1/2) Epoch 7, batch 3300, loss[loss=0.1867, simple_loss=0.2786, pruned_loss=0.04742, over 12139.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2593, pruned_loss=0.04167, over 2375165.95 frames. ], batch size: 39, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:02:06,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2155, 2.6333, 3.8758, 3.2649, 3.5751, 3.3484, 2.6541, 3.6470], device='cuda:1'), covar=tensor([0.0123, 0.0327, 0.0108, 0.0203, 0.0126, 0.0159, 0.0319, 0.0127], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0200, 0.0180, 0.0174, 0.0203, 0.0154, 0.0190, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:02:14,048 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176726.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:02:18,368 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6181, 4.5658, 4.4942, 4.6199, 3.8505, 4.6523, 4.6889, 4.8578], device='cuda:1'), covar=tensor([0.0294, 0.0203, 0.0266, 0.0335, 0.1221, 0.0388, 0.0210, 0.0229], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0189, 0.0187, 0.0237, 0.0237, 0.0210, 0.0171, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 07:02:27,478 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.802e+02 3.191e+02 3.872e+02 5.594e+02, threshold=6.381e+02, percent-clipped=0.0 2023-05-16 07:02:33,114 INFO [finetune.py:992] (1/2) Epoch 7, batch 3350, loss[loss=0.18, simple_loss=0.2656, pruned_loss=0.04718, over 10413.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2591, pruned_loss=0.0421, over 2368885.96 frames. ], batch size: 68, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:02:48,734 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2834, 6.1848, 5.9644, 5.4717, 5.2395, 6.1147, 5.7708, 5.4984], device='cuda:1'), covar=tensor([0.0618, 0.0993, 0.0676, 0.1777, 0.0686, 0.0821, 0.1574, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0516, 0.0497, 0.0599, 0.0394, 0.0679, 0.0744, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:03:06,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 07:03:08,914 INFO [finetune.py:992] (1/2) Epoch 7, batch 3400, loss[loss=0.1934, simple_loss=0.2872, pruned_loss=0.04984, over 12132.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2595, pruned_loss=0.04231, over 2368442.27 frames. ], batch size: 38, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:03:13,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 07:03:22,452 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176822.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:03:39,294 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.123e+02 2.870e+02 3.341e+02 4.078e+02 8.005e+02, threshold=6.681e+02, percent-clipped=1.0 2023-05-16 07:03:45,555 INFO [finetune.py:992] (1/2) Epoch 7, batch 3450, loss[loss=0.2085, simple_loss=0.288, pruned_loss=0.0645, over 12035.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.261, pruned_loss=0.04317, over 2367928.04 frames. ], batch size: 40, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:03:45,778 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176853.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:04:17,266 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176898.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:04:20,631 INFO [finetune.py:992] (1/2) Epoch 7, batch 3500, loss[loss=0.1406, simple_loss=0.2209, pruned_loss=0.03013, over 12301.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2605, pruned_loss=0.04314, over 2377180.08 frames. ], batch size: 28, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:04:27,804 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176913.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:04:28,050 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.29 vs. limit=5.0 2023-05-16 07:04:28,636 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176914.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:04:50,076 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.880e+02 3.425e+02 4.316e+02 1.226e+03, threshold=6.849e+02, percent-clipped=3.0 2023-05-16 07:04:50,919 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176946.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:04:55,835 INFO [finetune.py:992] (1/2) Epoch 7, batch 3550, loss[loss=0.1753, simple_loss=0.2599, pruned_loss=0.0454, over 10427.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2605, pruned_loss=0.04319, over 2377783.21 frames. ], batch size: 69, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:05:01,449 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176961.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:05:04,429 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3522, 5.1157, 5.2904, 5.2947, 4.8560, 4.9550, 4.6938, 5.1954], device='cuda:1'), covar=tensor([0.0577, 0.0577, 0.0710, 0.0558, 0.1859, 0.1263, 0.0624, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0659, 0.0562, 0.0606, 0.0805, 0.0706, 0.0529, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 07:05:16,449 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176981.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:05:33,147 INFO [finetune.py:992] (1/2) Epoch 7, batch 3600, loss[loss=0.1886, simple_loss=0.2803, pruned_loss=0.04844, over 10612.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2608, pruned_loss=0.0433, over 2375791.17 frames. ], batch size: 68, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:05:46,121 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177021.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:05:59,046 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2450, 4.5860, 4.0136, 4.8971, 4.5555, 2.8798, 4.2052, 3.1064], device='cuda:1'), covar=tensor([0.0808, 0.0848, 0.1425, 0.0449, 0.1105, 0.1629, 0.1030, 0.3065], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0368, 0.0347, 0.0269, 0.0355, 0.0263, 0.0332, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:06:03,232 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 2.813e+02 3.293e+02 3.837e+02 6.918e+02, threshold=6.586e+02, percent-clipped=1.0 2023-05-16 07:06:08,825 INFO [finetune.py:992] (1/2) Epoch 7, batch 3650, loss[loss=0.1692, simple_loss=0.2511, pruned_loss=0.04368, over 12305.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2606, pruned_loss=0.04302, over 2381293.99 frames. ], batch size: 33, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:06:44,368 INFO [finetune.py:992] (1/2) Epoch 7, batch 3700, loss[loss=0.1536, simple_loss=0.2353, pruned_loss=0.03597, over 11814.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2601, pruned_loss=0.04245, over 2380742.36 frames. ], batch size: 26, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:06:48,789 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177109.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:06:50,280 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177111.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:06:58,698 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177122.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:07:15,339 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.904e+02 3.352e+02 3.873e+02 6.220e+02, threshold=6.704e+02, percent-clipped=0.0 2023-05-16 07:07:21,108 INFO [finetune.py:992] (1/2) Epoch 7, batch 3750, loss[loss=0.1575, simple_loss=0.2489, pruned_loss=0.03305, over 12286.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.26, pruned_loss=0.04237, over 2386704.88 frames. ], batch size: 37, lr: 4.45e-03, grad_scale: 8.0 2023-05-16 07:07:32,686 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8456, 4.5838, 4.6196, 4.8117, 4.5477, 4.7617, 4.6289, 2.3945], device='cuda:1'), covar=tensor([0.0180, 0.0092, 0.0144, 0.0102, 0.0080, 0.0153, 0.0136, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0078, 0.0070, 0.0058, 0.0088, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:07:33,212 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177170.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:07:33,336 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177170.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:07:34,750 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177172.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:07:56,596 INFO [finetune.py:992] (1/2) Epoch 7, batch 3800, loss[loss=0.1509, simple_loss=0.2352, pruned_loss=0.03325, over 12342.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2599, pruned_loss=0.04247, over 2381084.06 frames. ], batch size: 31, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:08:00,878 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177209.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:08:26,086 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.889e+02 3.301e+02 4.159e+02 8.427e+02, threshold=6.603e+02, percent-clipped=1.0 2023-05-16 07:08:31,519 INFO [finetune.py:992] (1/2) Epoch 7, batch 3850, loss[loss=0.2084, simple_loss=0.2954, pruned_loss=0.06067, over 12139.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2617, pruned_loss=0.04324, over 2379641.49 frames. ], batch size: 39, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:08:53,060 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177281.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:09:04,582 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3980, 4.7882, 2.9952, 2.6757, 4.1110, 2.5375, 4.0291, 3.2629], device='cuda:1'), covar=tensor([0.0693, 0.0440, 0.1129, 0.1532, 0.0310, 0.1404, 0.0506, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0249, 0.0176, 0.0197, 0.0136, 0.0181, 0.0195, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:09:08,640 INFO [finetune.py:992] (1/2) Epoch 7, batch 3900, loss[loss=0.1878, simple_loss=0.2765, pruned_loss=0.04956, over 12046.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2617, pruned_loss=0.04318, over 2380588.20 frames. ], batch size: 42, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:09:21,556 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177321.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:09:27,119 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177329.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:09:28,062 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7234, 2.8438, 4.5703, 4.7456, 2.8092, 2.7313, 2.8529, 2.1424], device='cuda:1'), covar=tensor([0.1429, 0.2846, 0.0416, 0.0364, 0.1292, 0.2108, 0.2563, 0.4011], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0372, 0.0260, 0.0290, 0.0256, 0.0286, 0.0359, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:09:38,530 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 2.746e+02 3.274e+02 3.920e+02 7.586e+02, threshold=6.548e+02, percent-clipped=2.0 2023-05-16 07:09:44,180 INFO [finetune.py:992] (1/2) Epoch 7, batch 3950, loss[loss=0.1618, simple_loss=0.2537, pruned_loss=0.03499, over 12022.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2624, pruned_loss=0.04353, over 2378745.36 frames. ], batch size: 31, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:09:55,564 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177369.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:10:20,361 INFO [finetune.py:992] (1/2) Epoch 7, batch 4000, loss[loss=0.2084, simple_loss=0.291, pruned_loss=0.06284, over 12140.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2615, pruned_loss=0.04325, over 2379248.81 frames. ], batch size: 38, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:10:51,244 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.807e+02 3.267e+02 3.740e+02 6.650e+02, threshold=6.534e+02, percent-clipped=2.0 2023-05-16 07:10:56,905 INFO [finetune.py:992] (1/2) Epoch 7, batch 4050, loss[loss=0.1768, simple_loss=0.2712, pruned_loss=0.04119, over 12134.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2616, pruned_loss=0.04314, over 2377598.30 frames. ], batch size: 39, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:11:05,462 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177465.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:11:06,780 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177467.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:11:32,140 INFO [finetune.py:992] (1/2) Epoch 7, batch 4100, loss[loss=0.1667, simple_loss=0.2429, pruned_loss=0.04524, over 11840.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2618, pruned_loss=0.04328, over 2371393.11 frames. ], batch size: 26, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:11:36,293 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177509.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:11:41,949 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6858, 2.6350, 3.5017, 4.6016, 2.4229, 4.6485, 4.6283, 4.8868], device='cuda:1'), covar=tensor([0.0127, 0.1268, 0.0389, 0.0132, 0.1359, 0.0200, 0.0152, 0.0056], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0201, 0.0183, 0.0113, 0.0188, 0.0175, 0.0171, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:11:42,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3536, 3.4728, 3.2516, 3.1233, 2.8593, 2.6186, 3.5598, 2.2485], device='cuda:1'), covar=tensor([0.0392, 0.0141, 0.0169, 0.0198, 0.0394, 0.0384, 0.0128, 0.0466], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0158, 0.0153, 0.0184, 0.0202, 0.0195, 0.0161, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:12:01,218 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.947e+02 2.896e+02 3.356e+02 4.010e+02 8.247e+02, threshold=6.713e+02, percent-clipped=2.0 2023-05-16 07:12:06,838 INFO [finetune.py:992] (1/2) Epoch 7, batch 4150, loss[loss=0.1777, simple_loss=0.2725, pruned_loss=0.04148, over 10660.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04365, over 2373702.68 frames. ], batch size: 68, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:12:06,968 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177553.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:12:09,582 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177557.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:12:30,433 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7715, 2.7722, 4.3633, 4.6058, 2.8231, 2.6207, 2.9727, 2.0993], device='cuda:1'), covar=tensor([0.1351, 0.2696, 0.0459, 0.0403, 0.1155, 0.2100, 0.2344, 0.3843], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0373, 0.0263, 0.0293, 0.0258, 0.0288, 0.0362, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:12:43,644 INFO [finetune.py:992] (1/2) Epoch 7, batch 4200, loss[loss=0.1764, simple_loss=0.2697, pruned_loss=0.04154, over 11594.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2631, pruned_loss=0.04367, over 2373388.83 frames. ], batch size: 48, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:12:51,818 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177614.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:13:13,546 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 2.718e+02 3.241e+02 3.751e+02 5.947e+02, threshold=6.481e+02, percent-clipped=0.0 2023-05-16 07:13:19,391 INFO [finetune.py:992] (1/2) Epoch 7, batch 4250, loss[loss=0.1849, simple_loss=0.2746, pruned_loss=0.04761, over 11585.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2636, pruned_loss=0.04378, over 2371795.81 frames. ], batch size: 48, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:13:27,267 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177664.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:13:41,492 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3161, 4.0374, 4.2612, 4.5686, 3.0827, 4.0808, 2.8215, 4.1770], device='cuda:1'), covar=tensor([0.1555, 0.0726, 0.0745, 0.0591, 0.1073, 0.0564, 0.1625, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0263, 0.0295, 0.0354, 0.0236, 0.0240, 0.0257, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:13:55,448 INFO [finetune.py:992] (1/2) Epoch 7, batch 4300, loss[loss=0.1721, simple_loss=0.2624, pruned_loss=0.04089, over 12365.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2631, pruned_loss=0.0433, over 2366324.72 frames. ], batch size: 35, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:14:08,589 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2676, 4.0947, 4.2703, 4.5862, 3.2365, 4.1078, 2.5900, 4.1846], device='cuda:1'), covar=tensor([0.1626, 0.0718, 0.0820, 0.0597, 0.1061, 0.0616, 0.1892, 0.1520], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0263, 0.0294, 0.0353, 0.0235, 0.0240, 0.0257, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:14:09,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-05-16 07:14:12,034 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177725.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:14:25,977 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.775e+02 3.290e+02 3.803e+02 7.959e+02, threshold=6.580e+02, percent-clipped=1.0 2023-05-16 07:14:31,098 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5073, 2.2901, 3.2555, 4.4843, 2.3265, 4.4199, 4.4383, 4.6831], device='cuda:1'), covar=tensor([0.0101, 0.1287, 0.0449, 0.0120, 0.1165, 0.0193, 0.0156, 0.0059], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0204, 0.0186, 0.0115, 0.0190, 0.0177, 0.0174, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:14:31,611 INFO [finetune.py:992] (1/2) Epoch 7, batch 4350, loss[loss=0.1648, simple_loss=0.2677, pruned_loss=0.03095, over 12187.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.262, pruned_loss=0.04267, over 2368698.65 frames. ], batch size: 35, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:14:33,089 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3665, 3.1662, 3.2154, 3.4607, 2.6299, 3.1601, 2.5993, 2.9740], device='cuda:1'), covar=tensor([0.1323, 0.0759, 0.0812, 0.0648, 0.0905, 0.0717, 0.1436, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0261, 0.0293, 0.0352, 0.0234, 0.0238, 0.0255, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:14:39,982 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177765.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:14:41,441 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177767.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:14:50,735 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177780.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:15:02,910 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177797.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:15:07,193 INFO [finetune.py:992] (1/2) Epoch 7, batch 4400, loss[loss=0.1923, simple_loss=0.2786, pruned_loss=0.05302, over 10547.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2631, pruned_loss=0.04314, over 2372180.18 frames. ], batch size: 68, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:15:14,372 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177813.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:15:15,731 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177815.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:15:34,386 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177841.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:15:37,041 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.989e+02 2.909e+02 3.339e+02 4.089e+02 6.461e+02, threshold=6.677e+02, percent-clipped=0.0 2023-05-16 07:15:43,241 INFO [finetune.py:992] (1/2) Epoch 7, batch 4450, loss[loss=0.1532, simple_loss=0.2291, pruned_loss=0.03863, over 12348.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2629, pruned_loss=0.04337, over 2371996.53 frames. ], batch size: 30, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:15:46,991 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177858.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:16:03,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-05-16 07:16:19,595 INFO [finetune.py:992] (1/2) Epoch 7, batch 4500, loss[loss=0.1837, simple_loss=0.2787, pruned_loss=0.04432, over 12148.00 frames. ], tot_loss[loss=0.174, simple_loss=0.262, pruned_loss=0.043, over 2372715.79 frames. ], batch size: 34, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:16:24,115 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177909.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:16:41,217 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9904, 4.9796, 4.8216, 4.8917, 4.5224, 5.0460, 5.0294, 5.1427], device='cuda:1'), covar=tensor([0.0212, 0.0152, 0.0203, 0.0298, 0.0757, 0.0307, 0.0159, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0189, 0.0188, 0.0236, 0.0237, 0.0210, 0.0172, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 07:16:49,417 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.110e+02 2.795e+02 3.211e+02 3.845e+02 9.886e+02, threshold=6.421e+02, percent-clipped=1.0 2023-05-16 07:16:55,115 INFO [finetune.py:992] (1/2) Epoch 7, batch 4550, loss[loss=0.167, simple_loss=0.258, pruned_loss=0.03794, over 12050.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2615, pruned_loss=0.04255, over 2380442.35 frames. ], batch size: 37, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:17:34,551 INFO [finetune.py:992] (1/2) Epoch 7, batch 4600, loss[loss=0.1686, simple_loss=0.2579, pruned_loss=0.03963, over 12406.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2611, pruned_loss=0.0424, over 2380743.66 frames. ], batch size: 32, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:17:47,505 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178020.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:17:48,857 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4081, 5.1888, 5.3692, 5.4255, 5.0076, 4.9536, 4.8360, 5.2822], device='cuda:1'), covar=tensor([0.0640, 0.0646, 0.0723, 0.0488, 0.1883, 0.1512, 0.0570, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0662, 0.0564, 0.0605, 0.0807, 0.0707, 0.0528, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 07:17:55,137 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9507, 5.9689, 5.7123, 5.2799, 5.1245, 5.8525, 5.4256, 5.2326], device='cuda:1'), covar=tensor([0.0670, 0.0763, 0.0602, 0.1345, 0.0667, 0.0662, 0.1467, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0514, 0.0494, 0.0594, 0.0398, 0.0683, 0.0744, 0.0542], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:18:04,913 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 2.769e+02 3.252e+02 3.820e+02 6.652e+02, threshold=6.503e+02, percent-clipped=1.0 2023-05-16 07:18:09,448 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2326, 3.9075, 2.5306, 2.0978, 3.5515, 2.1742, 3.4975, 2.7639], device='cuda:1'), covar=tensor([0.0550, 0.0736, 0.1169, 0.1882, 0.0273, 0.1610, 0.0524, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0252, 0.0177, 0.0198, 0.0138, 0.0182, 0.0197, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:18:10,554 INFO [finetune.py:992] (1/2) Epoch 7, batch 4650, loss[loss=0.2252, simple_loss=0.3014, pruned_loss=0.07452, over 8152.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2612, pruned_loss=0.04221, over 2371483.90 frames. ], batch size: 97, lr: 4.44e-03, grad_scale: 8.0 2023-05-16 07:18:39,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 07:18:46,162 INFO [finetune.py:992] (1/2) Epoch 7, batch 4700, loss[loss=0.1797, simple_loss=0.2731, pruned_loss=0.04309, over 12163.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2606, pruned_loss=0.04192, over 2377335.13 frames. ], batch size: 36, lr: 4.43e-03, grad_scale: 8.0 2023-05-16 07:19:09,578 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178136.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:19:15,451 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2964, 3.1138, 3.1961, 3.5197, 2.7057, 3.2218, 2.5665, 3.0376], device='cuda:1'), covar=tensor([0.1367, 0.0784, 0.0827, 0.0517, 0.0855, 0.0685, 0.1510, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0257, 0.0288, 0.0344, 0.0230, 0.0235, 0.0251, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:19:16,484 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.991e+02 3.455e+02 4.205e+02 8.934e+02, threshold=6.910e+02, percent-clipped=1.0 2023-05-16 07:19:22,875 INFO [finetune.py:992] (1/2) Epoch 7, batch 4750, loss[loss=0.1697, simple_loss=0.2548, pruned_loss=0.04227, over 12076.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2611, pruned_loss=0.04239, over 2376262.53 frames. ], batch size: 32, lr: 4.43e-03, grad_scale: 8.0 2023-05-16 07:19:22,960 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178153.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:19:32,453 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2594, 2.6058, 3.7952, 3.2746, 3.6808, 3.3510, 2.6913, 3.6761], device='cuda:1'), covar=tensor([0.0124, 0.0318, 0.0149, 0.0204, 0.0114, 0.0146, 0.0325, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0197, 0.0179, 0.0173, 0.0201, 0.0153, 0.0188, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:19:58,755 INFO [finetune.py:992] (1/2) Epoch 7, batch 4800, loss[loss=0.1741, simple_loss=0.2625, pruned_loss=0.04287, over 12357.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2606, pruned_loss=0.04228, over 2380640.31 frames. ], batch size: 35, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:20:03,016 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178209.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:20:21,695 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3007, 5.0634, 5.3143, 5.2987, 4.8911, 4.9257, 4.7141, 5.1655], device='cuda:1'), covar=tensor([0.0656, 0.0606, 0.0686, 0.0532, 0.1876, 0.1313, 0.0545, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0661, 0.0561, 0.0607, 0.0809, 0.0707, 0.0529, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 07:20:28,570 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 2.827e+02 3.203e+02 3.957e+02 7.089e+02, threshold=6.407e+02, percent-clipped=1.0 2023-05-16 07:20:34,380 INFO [finetune.py:992] (1/2) Epoch 7, batch 4850, loss[loss=0.1476, simple_loss=0.2304, pruned_loss=0.0324, over 12024.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2599, pruned_loss=0.04165, over 2384460.87 frames. ], batch size: 31, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:20:37,164 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178257.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:21:11,119 INFO [finetune.py:992] (1/2) Epoch 7, batch 4900, loss[loss=0.1805, simple_loss=0.2685, pruned_loss=0.04629, over 12290.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2599, pruned_loss=0.04163, over 2385740.02 frames. ], batch size: 34, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:21:11,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 07:21:23,180 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178320.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:21:26,613 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1807, 4.9457, 5.2548, 5.1791, 4.3602, 4.5152, 4.6152, 4.9439], device='cuda:1'), covar=tensor([0.0942, 0.1345, 0.0751, 0.0976, 0.3874, 0.2334, 0.0727, 0.1794], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0655, 0.0558, 0.0600, 0.0801, 0.0701, 0.0524, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:21:40,632 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.599e+02 3.080e+02 3.833e+02 7.580e+02, threshold=6.161e+02, percent-clipped=3.0 2023-05-16 07:21:46,475 INFO [finetune.py:992] (1/2) Epoch 7, batch 4950, loss[loss=0.1437, simple_loss=0.2173, pruned_loss=0.03505, over 12015.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2595, pruned_loss=0.04169, over 2377972.41 frames. ], batch size: 28, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:21:57,228 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178368.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:21:57,995 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8152, 4.5702, 4.7603, 4.7356, 4.6222, 4.8194, 4.6006, 2.6386], device='cuda:1'), covar=tensor([0.0099, 0.0066, 0.0091, 0.0079, 0.0063, 0.0087, 0.0093, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0075, 0.0078, 0.0070, 0.0058, 0.0088, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:22:22,244 INFO [finetune.py:992] (1/2) Epoch 7, batch 5000, loss[loss=0.1584, simple_loss=0.2527, pruned_loss=0.03199, over 12193.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2598, pruned_loss=0.04177, over 2385398.56 frames. ], batch size: 35, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:22:41,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 07:22:46,596 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178436.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:22:46,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 07:22:53,244 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.791e+02 3.211e+02 3.917e+02 1.429e+03, threshold=6.422e+02, percent-clipped=4.0 2023-05-16 07:22:58,440 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178452.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:22:59,018 INFO [finetune.py:992] (1/2) Epoch 7, batch 5050, loss[loss=0.1617, simple_loss=0.2441, pruned_loss=0.0396, over 12258.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.26, pruned_loss=0.04168, over 2389191.34 frames. ], batch size: 32, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:22:59,176 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178453.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:23:04,946 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5307, 2.6686, 3.6531, 4.5103, 3.8919, 4.4620, 3.6768, 3.2306], device='cuda:1'), covar=tensor([0.0031, 0.0355, 0.0126, 0.0034, 0.0112, 0.0079, 0.0142, 0.0318], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0121, 0.0102, 0.0075, 0.0102, 0.0114, 0.0095, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:23:21,084 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178484.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:23:29,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-05-16 07:23:33,134 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178501.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:23:34,475 INFO [finetune.py:992] (1/2) Epoch 7, batch 5100, loss[loss=0.1759, simple_loss=0.2709, pruned_loss=0.04045, over 12267.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2594, pruned_loss=0.04174, over 2383385.71 frames. ], batch size: 37, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:23:40,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.09 vs. limit=5.0 2023-05-16 07:23:41,731 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178513.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:23:49,307 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-16 07:24:04,298 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.754e+02 3.289e+02 3.828e+02 9.604e+02, threshold=6.577e+02, percent-clipped=1.0 2023-05-16 07:24:09,902 INFO [finetune.py:992] (1/2) Epoch 7, batch 5150, loss[loss=0.1485, simple_loss=0.238, pruned_loss=0.02948, over 12135.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2598, pruned_loss=0.04186, over 2384074.82 frames. ], batch size: 30, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:24:44,914 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9924, 5.8347, 5.2994, 5.3554, 5.9306, 5.2507, 5.5522, 5.4015], device='cuda:1'), covar=tensor([0.1368, 0.0945, 0.1129, 0.1801, 0.0935, 0.2123, 0.1524, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0474, 0.0380, 0.0419, 0.0448, 0.0424, 0.0376, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:24:46,924 INFO [finetune.py:992] (1/2) Epoch 7, batch 5200, loss[loss=0.1968, simple_loss=0.2906, pruned_loss=0.05151, over 12142.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2594, pruned_loss=0.04195, over 2386823.86 frames. ], batch size: 36, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:24:58,518 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178619.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:25:16,555 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.134e+02 2.748e+02 3.403e+02 4.136e+02 9.752e+02, threshold=6.807e+02, percent-clipped=3.0 2023-05-16 07:25:22,345 INFO [finetune.py:992] (1/2) Epoch 7, batch 5250, loss[loss=0.2074, simple_loss=0.3032, pruned_loss=0.05575, over 12070.00 frames. ], tot_loss[loss=0.172, simple_loss=0.26, pruned_loss=0.04203, over 2391231.78 frames. ], batch size: 42, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:25:41,501 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178680.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:25:51,497 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178694.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:25:53,539 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178697.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:25:57,698 INFO [finetune.py:992] (1/2) Epoch 7, batch 5300, loss[loss=0.1775, simple_loss=0.2709, pruned_loss=0.04208, over 12193.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2612, pruned_loss=0.04256, over 2387274.05 frames. ], batch size: 35, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:26:18,810 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8363, 3.5368, 5.2324, 2.7088, 3.0559, 4.0006, 3.4027, 4.1367], device='cuda:1'), covar=tensor([0.0432, 0.0964, 0.0259, 0.1173, 0.1696, 0.1220, 0.1128, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0232, 0.0237, 0.0180, 0.0234, 0.0283, 0.0222, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:26:29,082 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.900e+02 3.366e+02 3.935e+02 8.707e+02, threshold=6.733e+02, percent-clipped=1.0 2023-05-16 07:26:34,928 INFO [finetune.py:992] (1/2) Epoch 7, batch 5350, loss[loss=0.16, simple_loss=0.2625, pruned_loss=0.02872, over 12153.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2601, pruned_loss=0.04184, over 2385463.75 frames. ], batch size: 34, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:26:36,586 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178755.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 07:26:38,625 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178758.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:26:47,102 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178770.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:26:55,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 07:27:10,867 INFO [finetune.py:992] (1/2) Epoch 7, batch 5400, loss[loss=0.1527, simple_loss=0.2363, pruned_loss=0.0345, over 12268.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2593, pruned_loss=0.0415, over 2383718.72 frames. ], batch size: 28, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:27:14,518 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178808.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:27:22,460 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9864, 4.5693, 4.7539, 4.7748, 4.6554, 4.8616, 4.6948, 2.4623], device='cuda:1'), covar=tensor([0.0087, 0.0061, 0.0082, 0.0063, 0.0042, 0.0087, 0.0078, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0077, 0.0069, 0.0057, 0.0086, 0.0075, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:27:30,735 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178831.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:27:40,560 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.862e+02 3.225e+02 3.753e+02 6.699e+02, threshold=6.449e+02, percent-clipped=0.0 2023-05-16 07:27:42,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 07:27:43,464 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0665, 5.9830, 5.7902, 5.3243, 5.2373, 5.9491, 5.5582, 5.3250], device='cuda:1'), covar=tensor([0.0698, 0.0973, 0.0648, 0.1609, 0.0674, 0.0735, 0.1419, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0583, 0.0520, 0.0494, 0.0597, 0.0394, 0.0689, 0.0742, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:27:46,222 INFO [finetune.py:992] (1/2) Epoch 7, batch 5450, loss[loss=0.165, simple_loss=0.2599, pruned_loss=0.035, over 12158.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2607, pruned_loss=0.04219, over 2379650.06 frames. ], batch size: 34, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:28:16,902 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178894.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:28:23,146 INFO [finetune.py:992] (1/2) Epoch 7, batch 5500, loss[loss=0.163, simple_loss=0.2585, pruned_loss=0.03378, over 12287.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2604, pruned_loss=0.04217, over 2376804.26 frames. ], batch size: 37, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:28:53,206 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.973e+02 3.388e+02 4.216e+02 1.873e+03, threshold=6.776e+02, percent-clipped=4.0 2023-05-16 07:28:56,237 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5123, 5.3693, 5.4900, 5.5204, 5.0865, 5.1306, 4.9300, 5.4821], device='cuda:1'), covar=tensor([0.0600, 0.0529, 0.0608, 0.0557, 0.1874, 0.1348, 0.0506, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0652, 0.0558, 0.0601, 0.0803, 0.0705, 0.0522, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:28:58,957 INFO [finetune.py:992] (1/2) Epoch 7, batch 5550, loss[loss=0.1737, simple_loss=0.2628, pruned_loss=0.04235, over 12319.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2604, pruned_loss=0.04193, over 2368121.34 frames. ], batch size: 34, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:29:00,571 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178955.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:29:14,593 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178975.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:29:34,597 INFO [finetune.py:992] (1/2) Epoch 7, batch 5600, loss[loss=0.1983, simple_loss=0.2895, pruned_loss=0.05356, over 10258.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2591, pruned_loss=0.04177, over 2370841.76 frames. ], batch size: 68, lr: 4.43e-03, grad_scale: 16.0 2023-05-16 07:30:02,938 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0065, 4.4316, 3.9706, 4.7690, 4.4439, 2.9225, 4.0887, 2.9684], device='cuda:1'), covar=tensor([0.0874, 0.0810, 0.1416, 0.0487, 0.0977, 0.1575, 0.1020, 0.3127], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0369, 0.0349, 0.0273, 0.0358, 0.0263, 0.0337, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:30:03,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.60 vs. limit=5.0 2023-05-16 07:30:05,253 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 07:30:05,384 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.735e+02 3.260e+02 3.853e+02 6.561e+02, threshold=6.520e+02, percent-clipped=0.0 2023-05-16 07:30:06,243 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2531, 5.2369, 5.0404, 4.6484, 4.7263, 5.1448, 4.8045, 4.6222], device='cuda:1'), covar=tensor([0.0800, 0.0924, 0.0705, 0.1472, 0.1103, 0.0776, 0.1673, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0585, 0.0522, 0.0495, 0.0600, 0.0396, 0.0692, 0.0745, 0.0551], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:30:08,482 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9977, 2.3494, 3.6578, 3.0815, 3.4964, 3.2208, 2.5927, 3.6211], device='cuda:1'), covar=tensor([0.0140, 0.0344, 0.0121, 0.0210, 0.0135, 0.0150, 0.0323, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0197, 0.0178, 0.0173, 0.0201, 0.0153, 0.0188, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:30:09,024 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179050.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 07:30:11,004 INFO [finetune.py:992] (1/2) Epoch 7, batch 5650, loss[loss=0.164, simple_loss=0.2372, pruned_loss=0.04542, over 12275.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2605, pruned_loss=0.04272, over 2362045.42 frames. ], batch size: 28, lr: 4.42e-03, grad_scale: 16.0 2023-05-16 07:30:11,086 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179053.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:30:46,569 INFO [finetune.py:992] (1/2) Epoch 7, batch 5700, loss[loss=0.1523, simple_loss=0.2272, pruned_loss=0.03872, over 12262.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2603, pruned_loss=0.04251, over 2366754.10 frames. ], batch size: 28, lr: 4.42e-03, grad_scale: 16.0 2023-05-16 07:30:50,321 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179108.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:30:54,542 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2149, 4.3913, 2.6302, 2.2461, 3.7812, 2.4112, 3.8553, 2.8683], device='cuda:1'), covar=tensor([0.0716, 0.0480, 0.1249, 0.1725, 0.0273, 0.1438, 0.0426, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0254, 0.0176, 0.0198, 0.0139, 0.0181, 0.0196, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:30:55,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 07:31:02,879 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179126.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:31:16,223 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.766e+02 3.420e+02 4.407e+02 1.310e+03, threshold=6.840e+02, percent-clipped=1.0 2023-05-16 07:31:21,952 INFO [finetune.py:992] (1/2) Epoch 7, batch 5750, loss[loss=0.1475, simple_loss=0.2379, pruned_loss=0.02857, over 12333.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2608, pruned_loss=0.04271, over 2372470.57 frames. ], batch size: 31, lr: 4.42e-03, grad_scale: 16.0 2023-05-16 07:31:24,723 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179156.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:31:42,929 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3194, 2.7730, 3.8508, 3.2265, 3.6611, 3.3216, 2.7828, 3.7871], device='cuda:1'), covar=tensor([0.0147, 0.0298, 0.0122, 0.0242, 0.0198, 0.0192, 0.0304, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0198, 0.0180, 0.0175, 0.0203, 0.0155, 0.0190, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:31:47,078 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3113, 5.0957, 5.2586, 5.2685, 4.9039, 4.8713, 4.6843, 5.2057], device='cuda:1'), covar=tensor([0.0629, 0.0582, 0.0685, 0.0531, 0.1632, 0.1367, 0.0576, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0651, 0.0560, 0.0598, 0.0803, 0.0704, 0.0522, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:31:59,186 INFO [finetune.py:992] (1/2) Epoch 7, batch 5800, loss[loss=0.1609, simple_loss=0.2442, pruned_loss=0.03885, over 12246.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2603, pruned_loss=0.04248, over 2374880.24 frames. ], batch size: 32, lr: 4.42e-03, grad_scale: 16.0 2023-05-16 07:32:17,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 07:32:21,303 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7316, 2.8882, 4.4669, 4.7293, 3.0113, 2.6451, 2.9375, 2.2355], device='cuda:1'), covar=tensor([0.1341, 0.2744, 0.0459, 0.0335, 0.1075, 0.2127, 0.2542, 0.3705], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0375, 0.0265, 0.0294, 0.0257, 0.0288, 0.0361, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:32:28,713 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.755e+02 3.271e+02 4.027e+02 9.280e+02, threshold=6.542e+02, percent-clipped=3.0 2023-05-16 07:32:32,194 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179250.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 07:32:34,230 INFO [finetune.py:992] (1/2) Epoch 7, batch 5850, loss[loss=0.1576, simple_loss=0.2482, pruned_loss=0.03354, over 12166.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2609, pruned_loss=0.04272, over 2379608.26 frames. ], batch size: 36, lr: 4.42e-03, grad_scale: 16.0 2023-05-16 07:32:43,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3397, 3.6701, 3.2820, 3.1586, 2.8371, 2.5584, 3.6162, 2.1802], device='cuda:1'), covar=tensor([0.0377, 0.0108, 0.0153, 0.0163, 0.0363, 0.0372, 0.0128, 0.0456], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0156, 0.0153, 0.0179, 0.0199, 0.0192, 0.0160, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:32:50,052 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179275.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:32:57,160 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3596, 4.8893, 5.2937, 4.6443, 4.8620, 4.7049, 5.3294, 4.9596], device='cuda:1'), covar=tensor([0.0235, 0.0354, 0.0226, 0.0256, 0.0417, 0.0325, 0.0212, 0.0256], device='cuda:1'), in_proj_covar=tensor([0.0245, 0.0253, 0.0273, 0.0248, 0.0246, 0.0247, 0.0223, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:33:05,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7373, 2.9508, 4.7081, 4.8511, 2.8826, 2.6535, 3.0232, 2.0833], device='cuda:1'), covar=tensor([0.1444, 0.2743, 0.0396, 0.0364, 0.1226, 0.2290, 0.2425, 0.3976], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0376, 0.0266, 0.0295, 0.0259, 0.0290, 0.0362, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:33:10,424 INFO [finetune.py:992] (1/2) Epoch 7, batch 5900, loss[loss=0.2043, simple_loss=0.2963, pruned_loss=0.05611, over 11815.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2616, pruned_loss=0.04329, over 2371049.19 frames. ], batch size: 44, lr: 4.42e-03, grad_scale: 16.0 2023-05-16 07:33:24,590 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179323.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:33:28,776 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2761, 5.1108, 5.2410, 5.2430, 4.8948, 4.9181, 4.7242, 5.1985], device='cuda:1'), covar=tensor([0.0595, 0.0537, 0.0719, 0.0522, 0.1787, 0.1251, 0.0514, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0653, 0.0561, 0.0600, 0.0803, 0.0706, 0.0523, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:33:41,073 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.863e+02 3.283e+02 4.029e+02 9.331e+02, threshold=6.565e+02, percent-clipped=3.0 2023-05-16 07:33:44,226 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179350.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:33:46,128 INFO [finetune.py:992] (1/2) Epoch 7, batch 5950, loss[loss=0.1954, simple_loss=0.2898, pruned_loss=0.05048, over 11854.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2605, pruned_loss=0.04292, over 2370118.21 frames. ], batch size: 44, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:33:46,248 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179353.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:33:50,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 07:33:51,949 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179361.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:34:17,949 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179398.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:34:20,269 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179401.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:34:21,577 INFO [finetune.py:992] (1/2) Epoch 7, batch 6000, loss[loss=0.1776, simple_loss=0.2624, pruned_loss=0.04643, over 12024.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2608, pruned_loss=0.0432, over 2371526.77 frames. ], batch size: 31, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:34:21,578 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 07:34:40,546 INFO [finetune.py:1026] (1/2) Epoch 7, validation: loss=0.3232, simple_loss=0.3987, pruned_loss=0.1238, over 1020973.00 frames. 2023-05-16 07:34:40,547 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 07:34:54,301 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179422.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:34:57,028 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179426.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:35:10,466 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179444.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:35:11,692 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.756e+02 2.747e+02 3.245e+02 4.203e+02 6.986e+02, threshold=6.491e+02, percent-clipped=1.0 2023-05-16 07:35:16,758 INFO [finetune.py:992] (1/2) Epoch 7, batch 6050, loss[loss=0.1588, simple_loss=0.2579, pruned_loss=0.02978, over 12146.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2612, pruned_loss=0.04314, over 2372494.81 frames. ], batch size: 34, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:35:19,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 07:35:31,926 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179474.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:35:52,396 INFO [finetune.py:992] (1/2) Epoch 7, batch 6100, loss[loss=0.1656, simple_loss=0.253, pruned_loss=0.03906, over 12024.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2621, pruned_loss=0.04373, over 2371044.25 frames. ], batch size: 31, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:35:52,496 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0128, 4.5757, 4.2551, 4.2009, 4.6909, 4.1017, 4.2826, 4.1160], device='cuda:1'), covar=tensor([0.1581, 0.1160, 0.1307, 0.2096, 0.1060, 0.2010, 0.1823, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0486, 0.0387, 0.0432, 0.0455, 0.0439, 0.0390, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:35:52,657 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5198, 2.9438, 3.8680, 2.3653, 2.5340, 3.1295, 2.9376, 3.2432], device='cuda:1'), covar=tensor([0.0542, 0.1037, 0.0360, 0.1138, 0.1688, 0.1251, 0.1112, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0230, 0.0234, 0.0177, 0.0231, 0.0279, 0.0220, 0.0251], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:35:54,063 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179505.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:36:23,255 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.099e+02 2.694e+02 3.293e+02 4.008e+02 9.136e+02, threshold=6.586e+02, percent-clipped=3.0 2023-05-16 07:36:26,199 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179550.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:36:28,129 INFO [finetune.py:992] (1/2) Epoch 7, batch 6150, loss[loss=0.1599, simple_loss=0.2565, pruned_loss=0.03167, over 12278.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2627, pruned_loss=0.04376, over 2377717.19 frames. ], batch size: 37, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:36:45,704 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2750, 5.1493, 5.2567, 5.2848, 4.8792, 4.9354, 4.7293, 5.1880], device='cuda:1'), covar=tensor([0.0605, 0.0555, 0.0637, 0.0544, 0.1864, 0.1159, 0.0550, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0655, 0.0561, 0.0599, 0.0803, 0.0704, 0.0523, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:37:00,491 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179598.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:37:04,105 INFO [finetune.py:992] (1/2) Epoch 7, batch 6200, loss[loss=0.1546, simple_loss=0.2501, pruned_loss=0.02959, over 12360.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2626, pruned_loss=0.04347, over 2376554.10 frames. ], batch size: 36, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:37:04,977 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9596, 4.5859, 4.7375, 4.7121, 4.5274, 4.7779, 4.7137, 2.7336], device='cuda:1'), covar=tensor([0.0096, 0.0056, 0.0074, 0.0059, 0.0049, 0.0083, 0.0062, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0074, 0.0078, 0.0070, 0.0058, 0.0088, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:37:08,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 07:37:12,284 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0219, 4.8866, 4.9943, 5.0304, 4.6332, 4.6898, 4.4955, 4.9453], device='cuda:1'), covar=tensor([0.0626, 0.0619, 0.0797, 0.0595, 0.1909, 0.1301, 0.0631, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0655, 0.0560, 0.0599, 0.0802, 0.0705, 0.0523, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:37:27,815 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179636.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:37:34,801 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.901e+02 3.290e+02 4.276e+02 6.618e+02, threshold=6.581e+02, percent-clipped=1.0 2023-05-16 07:37:39,651 INFO [finetune.py:992] (1/2) Epoch 7, batch 6250, loss[loss=0.1843, simple_loss=0.2712, pruned_loss=0.04868, over 12298.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2619, pruned_loss=0.0428, over 2379376.68 frames. ], batch size: 34, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:37:47,002 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3038, 4.9805, 5.1280, 5.0467, 4.9166, 5.1311, 5.1369, 2.9989], device='cuda:1'), covar=tensor([0.0078, 0.0055, 0.0071, 0.0059, 0.0051, 0.0078, 0.0066, 0.0595], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0075, 0.0079, 0.0071, 0.0058, 0.0089, 0.0076, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:38:07,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-05-16 07:38:11,784 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179697.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:38:15,931 INFO [finetune.py:992] (1/2) Epoch 7, batch 6300, loss[loss=0.1663, simple_loss=0.2594, pruned_loss=0.0366, over 12273.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2611, pruned_loss=0.04258, over 2382363.51 frames. ], batch size: 37, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:38:22,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-05-16 07:38:25,874 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179717.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:38:25,968 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5301, 2.7709, 3.7790, 4.3837, 4.0195, 4.4666, 3.8347, 3.1068], device='cuda:1'), covar=tensor([0.0032, 0.0323, 0.0119, 0.0036, 0.0106, 0.0056, 0.0131, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0117, 0.0099, 0.0072, 0.0098, 0.0110, 0.0091, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:38:47,010 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.650e+02 3.175e+02 4.099e+02 8.156e+02, threshold=6.350e+02, percent-clipped=2.0 2023-05-16 07:38:52,184 INFO [finetune.py:992] (1/2) Epoch 7, batch 6350, loss[loss=0.1438, simple_loss=0.2201, pruned_loss=0.03373, over 12257.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04265, over 2377260.93 frames. ], batch size: 28, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:39:08,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 07:39:13,902 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6238, 2.7197, 4.5500, 4.7515, 2.9903, 2.6145, 2.8979, 2.0540], device='cuda:1'), covar=tensor([0.1532, 0.3122, 0.0446, 0.0342, 0.1166, 0.2159, 0.2634, 0.3954], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0373, 0.0263, 0.0291, 0.0256, 0.0288, 0.0358, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:39:23,220 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179796.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:39:26,193 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179800.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:39:28,308 INFO [finetune.py:992] (1/2) Epoch 7, batch 6400, loss[loss=0.1476, simple_loss=0.2271, pruned_loss=0.03405, over 12361.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2608, pruned_loss=0.04248, over 2377367.60 frames. ], batch size: 30, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:39:39,282 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4951, 5.1356, 5.4415, 4.8348, 5.1400, 4.7004, 5.3755, 5.1527], device='cuda:1'), covar=tensor([0.0366, 0.0419, 0.0430, 0.0284, 0.0411, 0.0474, 0.0474, 0.0289], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0253, 0.0276, 0.0249, 0.0246, 0.0249, 0.0225, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:39:59,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.778e+02 3.173e+02 3.946e+02 6.917e+02, threshold=6.346e+02, percent-clipped=1.0 2023-05-16 07:40:04,648 INFO [finetune.py:992] (1/2) Epoch 7, batch 6450, loss[loss=0.1561, simple_loss=0.2389, pruned_loss=0.03668, over 12352.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2602, pruned_loss=0.04221, over 2376802.68 frames. ], batch size: 30, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:40:07,619 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179857.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:40:22,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.97 vs. limit=5.0 2023-05-16 07:40:40,225 INFO [finetune.py:992] (1/2) Epoch 7, batch 6500, loss[loss=0.1966, simple_loss=0.2981, pruned_loss=0.0476, over 12125.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2608, pruned_loss=0.04261, over 2372027.35 frames. ], batch size: 38, lr: 4.42e-03, grad_scale: 8.0 2023-05-16 07:40:42,586 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179906.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:40:55,204 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179924.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:41:10,719 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.731e+02 3.331e+02 4.021e+02 7.219e+02, threshold=6.663e+02, percent-clipped=2.0 2023-05-16 07:41:15,767 INFO [finetune.py:992] (1/2) Epoch 7, batch 6550, loss[loss=0.1782, simple_loss=0.2679, pruned_loss=0.04427, over 12100.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2604, pruned_loss=0.04261, over 2368781.57 frames. ], batch size: 38, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:41:25,913 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179967.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 07:41:36,126 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0031, 4.6332, 4.9915, 4.4141, 4.7230, 4.4053, 5.0097, 4.7674], device='cuda:1'), covar=tensor([0.0322, 0.0379, 0.0345, 0.0285, 0.0361, 0.0363, 0.0277, 0.0345], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0253, 0.0275, 0.0248, 0.0245, 0.0247, 0.0224, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:41:38,893 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179985.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:41:42,627 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5514, 2.8490, 4.7233, 4.8430, 2.9560, 2.5330, 3.0005, 2.2122], device='cuda:1'), covar=tensor([0.1582, 0.2863, 0.0394, 0.0370, 0.1164, 0.2238, 0.2570, 0.3767], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0373, 0.0264, 0.0292, 0.0256, 0.0288, 0.0359, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:41:43,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-05-16 07:41:43,848 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179992.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:41:54,842 INFO [finetune.py:992] (1/2) Epoch 7, batch 6600, loss[loss=0.2064, simple_loss=0.2905, pruned_loss=0.0612, over 10634.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2609, pruned_loss=0.04268, over 2373179.53 frames. ], batch size: 68, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:42:05,747 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180017.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:42:08,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-05-16 07:42:09,510 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5684, 2.8400, 4.5720, 4.6584, 2.7296, 2.4960, 2.9440, 2.1235], device='cuda:1'), covar=tensor([0.1487, 0.2835, 0.0422, 0.0423, 0.1301, 0.2188, 0.2523, 0.3792], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0372, 0.0263, 0.0291, 0.0255, 0.0288, 0.0358, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:42:10,776 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0717, 4.7972, 4.8625, 4.9968, 4.8326, 5.0103, 4.9727, 2.7624], device='cuda:1'), covar=tensor([0.0091, 0.0056, 0.0074, 0.0060, 0.0040, 0.0075, 0.0061, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0075, 0.0078, 0.0070, 0.0058, 0.0088, 0.0076, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:42:12,188 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180026.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:42:26,111 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.026e+02 2.656e+02 3.106e+02 3.718e+02 6.929e+02, threshold=6.212e+02, percent-clipped=1.0 2023-05-16 07:42:31,178 INFO [finetune.py:992] (1/2) Epoch 7, batch 6650, loss[loss=0.1782, simple_loss=0.273, pruned_loss=0.04168, over 12140.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2611, pruned_loss=0.04245, over 2375876.79 frames. ], batch size: 39, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:42:39,684 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180065.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:42:50,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 07:42:50,889 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9766, 4.9755, 4.8448, 4.8945, 4.4802, 5.0087, 4.9876, 5.1853], device='cuda:1'), covar=tensor([0.0180, 0.0141, 0.0179, 0.0272, 0.0795, 0.0243, 0.0140, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0190, 0.0189, 0.0241, 0.0239, 0.0212, 0.0173, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 07:42:55,173 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180087.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:43:04,156 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180100.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:43:04,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-16 07:43:06,138 INFO [finetune.py:992] (1/2) Epoch 7, batch 6700, loss[loss=0.1525, simple_loss=0.2314, pruned_loss=0.03676, over 12018.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2611, pruned_loss=0.04261, over 2369767.94 frames. ], batch size: 28, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:43:37,582 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.747e+02 3.239e+02 4.138e+02 6.736e+02, threshold=6.478e+02, percent-clipped=3.0 2023-05-16 07:43:39,126 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180148.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:43:40,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 07:43:41,898 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180152.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:43:42,560 INFO [finetune.py:992] (1/2) Epoch 7, batch 6750, loss[loss=0.156, simple_loss=0.2369, pruned_loss=0.03758, over 12339.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2603, pruned_loss=0.04219, over 2375075.61 frames. ], batch size: 31, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:44:12,583 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4568, 5.2741, 5.3838, 5.4390, 5.0865, 5.0960, 4.8215, 5.3625], device='cuda:1'), covar=tensor([0.0662, 0.0658, 0.0713, 0.0599, 0.1971, 0.1334, 0.0631, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0660, 0.0562, 0.0603, 0.0804, 0.0711, 0.0524, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 07:44:18,443 INFO [finetune.py:992] (1/2) Epoch 7, batch 6800, loss[loss=0.1571, simple_loss=0.2431, pruned_loss=0.03557, over 12126.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2611, pruned_loss=0.04274, over 2366040.15 frames. ], batch size: 30, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:44:20,047 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180205.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:44:48,597 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 2.845e+02 3.434e+02 4.080e+02 7.808e+02, threshold=6.868e+02, percent-clipped=1.0 2023-05-16 07:44:53,610 INFO [finetune.py:992] (1/2) Epoch 7, batch 6850, loss[loss=0.1793, simple_loss=0.2735, pruned_loss=0.04253, over 12106.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2619, pruned_loss=0.04315, over 2362599.22 frames. ], batch size: 42, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:45:00,563 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180262.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:45:03,535 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180266.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:45:13,502 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180280.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:45:21,961 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180292.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:45:30,416 INFO [finetune.py:992] (1/2) Epoch 7, batch 6900, loss[loss=0.1646, simple_loss=0.2602, pruned_loss=0.0345, over 12311.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2615, pruned_loss=0.04281, over 2366029.86 frames. ], batch size: 34, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:45:53,891 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180336.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:45:56,661 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180340.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:46:00,858 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.614e+02 3.096e+02 3.766e+02 7.680e+02, threshold=6.192e+02, percent-clipped=1.0 2023-05-16 07:46:05,791 INFO [finetune.py:992] (1/2) Epoch 7, batch 6950, loss[loss=0.1779, simple_loss=0.2649, pruned_loss=0.04544, over 12300.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2618, pruned_loss=0.04272, over 2372338.65 frames. ], batch size: 34, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:46:14,780 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1925, 5.9673, 5.6028, 5.4946, 6.0841, 5.3397, 5.6184, 5.6005], device='cuda:1'), covar=tensor([0.1465, 0.1021, 0.1175, 0.2200, 0.0940, 0.2149, 0.1636, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0476, 0.0380, 0.0422, 0.0447, 0.0427, 0.0385, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:46:26,753 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180382.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:46:37,548 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180397.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:46:41,837 INFO [finetune.py:992] (1/2) Epoch 7, batch 7000, loss[loss=0.1675, simple_loss=0.2601, pruned_loss=0.03745, over 11800.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2619, pruned_loss=0.04268, over 2370366.88 frames. ], batch size: 44, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:46:42,772 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1677, 2.5857, 3.7055, 3.0738, 3.6768, 3.3439, 2.4770, 3.7522], device='cuda:1'), covar=tensor([0.0102, 0.0306, 0.0124, 0.0214, 0.0109, 0.0127, 0.0355, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0200, 0.0184, 0.0177, 0.0205, 0.0156, 0.0191, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:46:43,449 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6085, 3.0368, 3.8367, 4.5770, 4.1580, 4.6080, 3.9562, 3.2471], device='cuda:1'), covar=tensor([0.0033, 0.0292, 0.0098, 0.0027, 0.0080, 0.0045, 0.0100, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0119, 0.0101, 0.0073, 0.0100, 0.0112, 0.0093, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:47:01,247 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2279, 2.6803, 3.7712, 3.1415, 3.6527, 3.3137, 2.6416, 3.6949], device='cuda:1'), covar=tensor([0.0135, 0.0318, 0.0147, 0.0227, 0.0132, 0.0177, 0.0344, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0200, 0.0184, 0.0177, 0.0206, 0.0156, 0.0192, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:47:11,958 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2468, 2.6069, 3.7694, 3.1839, 3.5569, 3.2681, 2.5535, 3.6481], device='cuda:1'), covar=tensor([0.0134, 0.0346, 0.0115, 0.0203, 0.0180, 0.0165, 0.0394, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0200, 0.0184, 0.0177, 0.0205, 0.0157, 0.0192, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:47:13,165 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.795e+02 3.201e+02 3.809e+02 8.289e+02, threshold=6.402e+02, percent-clipped=2.0 2023-05-16 07:47:18,243 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180452.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:47:18,864 INFO [finetune.py:992] (1/2) Epoch 7, batch 7050, loss[loss=0.1502, simple_loss=0.2372, pruned_loss=0.0316, over 12031.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2614, pruned_loss=0.04232, over 2375720.12 frames. ], batch size: 31, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:47:35,126 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2938, 4.5518, 2.7559, 2.5828, 3.9353, 2.3682, 3.9217, 2.9503], device='cuda:1'), covar=tensor([0.0682, 0.0491, 0.1197, 0.1496, 0.0250, 0.1477, 0.0435, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0247, 0.0172, 0.0194, 0.0136, 0.0177, 0.0192, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:47:51,878 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180500.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:47:53,957 INFO [finetune.py:992] (1/2) Epoch 7, batch 7100, loss[loss=0.1926, simple_loss=0.2816, pruned_loss=0.05181, over 12342.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2621, pruned_loss=0.04266, over 2374864.78 frames. ], batch size: 36, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:47:56,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 07:48:00,765 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4089, 2.3589, 3.3603, 4.4345, 2.3575, 4.3317, 4.4635, 4.6761], device='cuda:1'), covar=tensor([0.0117, 0.1320, 0.0407, 0.0123, 0.1227, 0.0206, 0.0142, 0.0075], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0203, 0.0183, 0.0114, 0.0188, 0.0179, 0.0175, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:48:25,050 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.035e+02 2.914e+02 3.366e+02 4.178e+02 7.262e+02, threshold=6.731e+02, percent-clipped=1.0 2023-05-16 07:48:30,032 INFO [finetune.py:992] (1/2) Epoch 7, batch 7150, loss[loss=0.1805, simple_loss=0.2744, pruned_loss=0.04334, over 11745.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2628, pruned_loss=0.04309, over 2369741.04 frames. ], batch size: 48, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:48:35,725 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180561.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:48:36,547 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180562.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 07:48:42,289 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180570.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:48:49,475 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180580.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:48:54,638 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1272, 3.8379, 5.3250, 2.6314, 2.9960, 4.0646, 3.3975, 4.0432], device='cuda:1'), covar=tensor([0.0273, 0.0863, 0.0334, 0.1078, 0.1604, 0.1110, 0.1116, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0236, 0.0240, 0.0182, 0.0235, 0.0285, 0.0225, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:49:06,842 INFO [finetune.py:992] (1/2) Epoch 7, batch 7200, loss[loss=0.1446, simple_loss=0.234, pruned_loss=0.02764, over 12097.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2613, pruned_loss=0.04262, over 2374838.58 frames. ], batch size: 32, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:49:11,719 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180610.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 07:49:24,559 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180628.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:49:26,808 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180631.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:49:37,257 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.799e+02 3.250e+02 3.741e+02 6.052e+02, threshold=6.500e+02, percent-clipped=0.0 2023-05-16 07:49:42,176 INFO [finetune.py:992] (1/2) Epoch 7, batch 7250, loss[loss=0.1673, simple_loss=0.2534, pruned_loss=0.04061, over 12257.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2618, pruned_loss=0.04274, over 2380978.38 frames. ], batch size: 32, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:49:55,346 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0131, 2.3399, 3.6754, 3.0322, 3.4830, 3.2141, 2.4392, 3.5230], device='cuda:1'), covar=tensor([0.0138, 0.0388, 0.0147, 0.0231, 0.0161, 0.0171, 0.0401, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0201, 0.0184, 0.0177, 0.0206, 0.0157, 0.0192, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:50:03,001 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:50:09,956 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180692.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:50:18,545 INFO [finetune.py:992] (1/2) Epoch 7, batch 7300, loss[loss=0.1774, simple_loss=0.2744, pruned_loss=0.0402, over 12369.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2628, pruned_loss=0.043, over 2377596.35 frames. ], batch size: 36, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:50:37,616 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180730.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:50:39,179 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180732.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:50:49,592 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.048e+02 2.837e+02 3.266e+02 3.945e+02 6.850e+02, threshold=6.533e+02, percent-clipped=1.0 2023-05-16 07:50:54,524 INFO [finetune.py:992] (1/2) Epoch 7, batch 7350, loss[loss=0.1916, simple_loss=0.2847, pruned_loss=0.04918, over 12150.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2629, pruned_loss=0.04299, over 2375976.39 frames. ], batch size: 36, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:51:16,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 07:51:16,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 07:51:23,246 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180793.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:51:24,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 07:51:30,453 INFO [finetune.py:992] (1/2) Epoch 7, batch 7400, loss[loss=0.1563, simple_loss=0.2426, pruned_loss=0.03501, over 12130.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2618, pruned_loss=0.04281, over 2379749.52 frames. ], batch size: 30, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:51:46,357 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180825.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:52:01,406 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.980e+02 2.835e+02 3.344e+02 3.894e+02 5.964e+02, threshold=6.688e+02, percent-clipped=0.0 2023-05-16 07:52:06,445 INFO [finetune.py:992] (1/2) Epoch 7, batch 7450, loss[loss=0.173, simple_loss=0.2584, pruned_loss=0.04386, over 12349.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2625, pruned_loss=0.04308, over 2370788.59 frames. ], batch size: 35, lr: 4.41e-03, grad_scale: 8.0 2023-05-16 07:52:12,327 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180861.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:52:29,770 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180886.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:52:33,423 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6769, 3.2529, 5.0574, 2.5376, 2.5861, 3.7786, 3.2117, 3.7617], device='cuda:1'), covar=tensor([0.0387, 0.1131, 0.0276, 0.1199, 0.1994, 0.1347, 0.1287, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0235, 0.0240, 0.0182, 0.0235, 0.0288, 0.0225, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:52:42,384 INFO [finetune.py:992] (1/2) Epoch 7, batch 7500, loss[loss=0.2558, simple_loss=0.3153, pruned_loss=0.09817, over 8131.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2631, pruned_loss=0.04333, over 2361381.25 frames. ], batch size: 97, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:52:46,796 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180909.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:52:58,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-05-16 07:52:59,112 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180926.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:53:13,095 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 2.813e+02 3.239e+02 3.692e+02 7.434e+02, threshold=6.478e+02, percent-clipped=1.0 2023-05-16 07:53:18,162 INFO [finetune.py:992] (1/2) Epoch 7, batch 7550, loss[loss=0.1531, simple_loss=0.2429, pruned_loss=0.03167, over 12187.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2628, pruned_loss=0.0431, over 2361077.32 frames. ], batch size: 31, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:53:40,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0735, 4.6380, 5.0591, 4.4714, 4.6895, 4.4609, 5.1098, 4.7171], device='cuda:1'), covar=tensor([0.0262, 0.0355, 0.0233, 0.0272, 0.0359, 0.0351, 0.0186, 0.0349], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0254, 0.0276, 0.0249, 0.0246, 0.0248, 0.0223, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 07:53:46,171 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180992.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:53:54,160 INFO [finetune.py:992] (1/2) Epoch 7, batch 7600, loss[loss=0.1559, simple_loss=0.2419, pruned_loss=0.03495, over 12043.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2628, pruned_loss=0.04301, over 2364180.65 frames. ], batch size: 31, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:54:21,090 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181040.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:54:24,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-05-16 07:54:25,181 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.789e+02 3.341e+02 4.058e+02 7.799e+02, threshold=6.682e+02, percent-clipped=3.0 2023-05-16 07:54:30,012 INFO [finetune.py:992] (1/2) Epoch 7, batch 7650, loss[loss=0.1733, simple_loss=0.2712, pruned_loss=0.03771, over 12107.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2623, pruned_loss=0.04311, over 2371089.79 frames. ], batch size: 33, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:54:54,938 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181088.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:55:05,455 INFO [finetune.py:992] (1/2) Epoch 7, batch 7700, loss[loss=0.1598, simple_loss=0.2488, pruned_loss=0.03538, over 12087.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2628, pruned_loss=0.04352, over 2364966.53 frames. ], batch size: 32, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:55:09,308 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7555, 2.6371, 3.9372, 4.1072, 2.9402, 2.6221, 2.7227, 2.2164], device='cuda:1'), covar=tensor([0.1308, 0.2595, 0.0540, 0.0483, 0.1064, 0.1951, 0.2527, 0.3560], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0373, 0.0265, 0.0292, 0.0256, 0.0288, 0.0360, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:55:36,614 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 3.065e+02 3.525e+02 4.565e+02 2.580e+03, threshold=7.049e+02, percent-clipped=8.0 2023-05-16 07:55:41,682 INFO [finetune.py:992] (1/2) Epoch 7, batch 7750, loss[loss=0.2442, simple_loss=0.3113, pruned_loss=0.08854, over 8270.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2622, pruned_loss=0.0434, over 2367659.09 frames. ], batch size: 98, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:55:52,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 07:55:52,489 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1607, 4.1029, 4.2341, 4.4461, 3.1315, 4.0228, 2.6828, 4.0795], device='cuda:1'), covar=tensor([0.1732, 0.0644, 0.0880, 0.0604, 0.1156, 0.0595, 0.1797, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0262, 0.0295, 0.0351, 0.0238, 0.0239, 0.0257, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:55:55,909 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5965, 4.2937, 4.4900, 4.5492, 4.3792, 4.5475, 4.4732, 2.9123], device='cuda:1'), covar=tensor([0.0096, 0.0070, 0.0092, 0.0067, 0.0052, 0.0085, 0.0112, 0.0572], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0077, 0.0070, 0.0057, 0.0087, 0.0076, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:56:02,071 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181181.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:56:17,713 INFO [finetune.py:992] (1/2) Epoch 7, batch 7800, loss[loss=0.1853, simple_loss=0.2747, pruned_loss=0.048, over 12284.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2623, pruned_loss=0.04368, over 2361972.81 frames. ], batch size: 33, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:56:33,894 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181226.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:56:47,949 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 2.871e+02 3.508e+02 4.302e+02 1.050e+03, threshold=7.015e+02, percent-clipped=3.0 2023-05-16 07:56:49,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 07:56:52,969 INFO [finetune.py:992] (1/2) Epoch 7, batch 7850, loss[loss=0.2113, simple_loss=0.2936, pruned_loss=0.06449, over 10468.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2623, pruned_loss=0.04351, over 2364789.55 frames. ], batch size: 68, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:57:08,367 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181274.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:57:24,590 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181297.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:57:28,618 INFO [finetune.py:992] (1/2) Epoch 7, batch 7900, loss[loss=0.171, simple_loss=0.2636, pruned_loss=0.03917, over 12149.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2627, pruned_loss=0.04327, over 2374742.26 frames. ], batch size: 34, lr: 4.40e-03, grad_scale: 8.0 2023-05-16 07:57:34,715 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 07:57:43,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-05-16 07:57:55,511 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9367, 5.8836, 5.6675, 5.1518, 5.1333, 5.8098, 5.4045, 5.2030], device='cuda:1'), covar=tensor([0.0710, 0.0921, 0.0664, 0.1739, 0.0658, 0.0770, 0.1536, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0587, 0.0520, 0.0491, 0.0601, 0.0394, 0.0687, 0.0743, 0.0546], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 07:57:59,796 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.838e+02 3.458e+02 4.140e+02 1.351e+03, threshold=6.917e+02, percent-clipped=2.0 2023-05-16 07:58:04,676 INFO [finetune.py:992] (1/2) Epoch 7, batch 7950, loss[loss=0.1559, simple_loss=0.2328, pruned_loss=0.03953, over 11988.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.263, pruned_loss=0.04364, over 2367913.43 frames. ], batch size: 28, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 07:58:08,411 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181358.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:58:08,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 07:58:29,424 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181388.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:58:40,225 INFO [finetune.py:992] (1/2) Epoch 7, batch 8000, loss[loss=0.1686, simple_loss=0.257, pruned_loss=0.04005, over 12146.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2628, pruned_loss=0.04347, over 2374835.27 frames. ], batch size: 36, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 07:59:00,578 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6286, 3.4578, 3.2196, 3.1901, 2.8659, 2.7319, 3.6560, 2.4068], device='cuda:1'), covar=tensor([0.0333, 0.0167, 0.0172, 0.0178, 0.0390, 0.0339, 0.0130, 0.0445], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0159, 0.0153, 0.0180, 0.0200, 0.0193, 0.0162, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 07:59:04,698 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181436.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:59:05,651 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6647, 3.3708, 5.1032, 2.8838, 2.7647, 3.8754, 3.2304, 3.9934], device='cuda:1'), covar=tensor([0.0489, 0.1215, 0.0335, 0.1117, 0.1994, 0.1489, 0.1378, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0235, 0.0240, 0.0182, 0.0236, 0.0289, 0.0224, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:59:11,696 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.051e+02 2.793e+02 3.335e+02 3.930e+02 8.631e+02, threshold=6.670e+02, percent-clipped=2.0 2023-05-16 07:59:16,533 INFO [finetune.py:992] (1/2) Epoch 7, batch 8050, loss[loss=0.1866, simple_loss=0.2727, pruned_loss=0.05028, over 11789.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2626, pruned_loss=0.04346, over 2376214.20 frames. ], batch size: 44, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 07:59:37,382 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181481.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 07:59:52,913 INFO [finetune.py:992] (1/2) Epoch 7, batch 8100, loss[loss=0.1597, simple_loss=0.2534, pruned_loss=0.03305, over 12096.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2626, pruned_loss=0.04326, over 2375139.34 frames. ], batch size: 32, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 07:59:54,569 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2016, 3.7814, 4.0003, 4.2487, 2.8557, 3.8241, 2.3869, 3.9547], device='cuda:1'), covar=tensor([0.1632, 0.0786, 0.0895, 0.0663, 0.1281, 0.0644, 0.2009, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0260, 0.0291, 0.0348, 0.0235, 0.0236, 0.0254, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 07:59:56,673 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7146, 3.4085, 5.1824, 2.7713, 2.7699, 3.8693, 3.2505, 3.9512], device='cuda:1'), covar=tensor([0.0419, 0.1123, 0.0278, 0.1175, 0.1926, 0.1385, 0.1266, 0.0994], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0235, 0.0241, 0.0182, 0.0236, 0.0289, 0.0224, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:00:00,968 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8296, 2.2911, 3.2085, 2.6874, 3.0619, 2.9061, 2.2127, 3.1117], device='cuda:1'), covar=tensor([0.0116, 0.0308, 0.0136, 0.0239, 0.0137, 0.0158, 0.0335, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0199, 0.0184, 0.0177, 0.0207, 0.0157, 0.0190, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:00:11,421 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181529.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:00:12,254 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181530.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:00:23,305 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.160e+02 2.962e+02 3.368e+02 3.994e+02 1.143e+03, threshold=6.735e+02, percent-clipped=5.0 2023-05-16 08:00:28,366 INFO [finetune.py:992] (1/2) Epoch 7, batch 8150, loss[loss=0.1468, simple_loss=0.2364, pruned_loss=0.02864, over 12104.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2629, pruned_loss=0.04386, over 2366371.41 frames. ], batch size: 32, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 08:00:29,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-16 08:00:55,880 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181591.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:01:03,937 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4472, 5.2988, 5.4551, 5.4371, 5.0762, 5.0699, 4.8968, 5.3640], device='cuda:1'), covar=tensor([0.0664, 0.0606, 0.0587, 0.0632, 0.1839, 0.1353, 0.0536, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0668, 0.0566, 0.0606, 0.0818, 0.0718, 0.0534, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 08:01:04,477 INFO [finetune.py:992] (1/2) Epoch 7, batch 8200, loss[loss=0.1646, simple_loss=0.2474, pruned_loss=0.04094, over 12186.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2628, pruned_loss=0.04399, over 2366867.78 frames. ], batch size: 31, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 08:01:34,930 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 3.045e+02 3.538e+02 4.343e+02 6.804e+02, threshold=7.077e+02, percent-clipped=1.0 2023-05-16 08:01:39,851 INFO [finetune.py:992] (1/2) Epoch 7, batch 8250, loss[loss=0.1791, simple_loss=0.2705, pruned_loss=0.04392, over 12313.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2643, pruned_loss=0.04478, over 2360833.85 frames. ], batch size: 34, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 08:01:39,934 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181653.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:01:41,548 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8443, 3.4807, 5.1747, 2.4077, 2.9012, 3.8222, 3.4402, 3.8959], device='cuda:1'), covar=tensor([0.0348, 0.1020, 0.0256, 0.1198, 0.1752, 0.1277, 0.1105, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0234, 0.0240, 0.0181, 0.0235, 0.0287, 0.0223, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:01:43,628 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3195, 3.4982, 3.6480, 4.1430, 3.2307, 3.6755, 2.5923, 3.9525], device='cuda:1'), covar=tensor([0.1305, 0.0815, 0.1156, 0.0767, 0.0829, 0.0568, 0.1589, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0261, 0.0294, 0.0350, 0.0236, 0.0237, 0.0256, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:02:15,671 INFO [finetune.py:992] (1/2) Epoch 7, batch 8300, loss[loss=0.1733, simple_loss=0.2711, pruned_loss=0.03773, over 12187.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2633, pruned_loss=0.04429, over 2364249.79 frames. ], batch size: 35, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 08:02:38,007 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2890, 5.1533, 5.3022, 5.3199, 4.9588, 4.9561, 4.8074, 5.2263], device='cuda:1'), covar=tensor([0.0652, 0.0527, 0.0634, 0.0533, 0.1708, 0.1317, 0.0509, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0672, 0.0567, 0.0610, 0.0821, 0.0721, 0.0537, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 08:02:47,016 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.829e+02 3.361e+02 4.143e+02 6.488e+02, threshold=6.723e+02, percent-clipped=0.0 2023-05-16 08:02:48,030 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7234, 3.2040, 5.0390, 2.6011, 2.6989, 3.8594, 3.2509, 3.8919], device='cuda:1'), covar=tensor([0.0397, 0.1196, 0.0278, 0.1221, 0.1971, 0.1213, 0.1294, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0237, 0.0243, 0.0183, 0.0238, 0.0291, 0.0226, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:02:51,864 INFO [finetune.py:992] (1/2) Epoch 7, batch 8350, loss[loss=0.1775, simple_loss=0.2663, pruned_loss=0.04433, over 12268.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2621, pruned_loss=0.04362, over 2369163.36 frames. ], batch size: 37, lr: 4.40e-03, grad_scale: 16.0 2023-05-16 08:02:59,946 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0869, 5.0816, 4.9184, 4.9390, 4.5988, 5.1337, 4.9637, 5.3062], device='cuda:1'), covar=tensor([0.0228, 0.0137, 0.0187, 0.0284, 0.0810, 0.0369, 0.0168, 0.0162], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0190, 0.0188, 0.0240, 0.0240, 0.0212, 0.0172, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 08:03:27,745 INFO [finetune.py:992] (1/2) Epoch 7, batch 8400, loss[loss=0.1835, simple_loss=0.274, pruned_loss=0.04655, over 12146.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2611, pruned_loss=0.04312, over 2373011.35 frames. ], batch size: 34, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:03:32,927 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4583, 4.7520, 3.1023, 3.0204, 4.0433, 2.7558, 3.9688, 3.4174], device='cuda:1'), covar=tensor([0.0664, 0.0477, 0.0976, 0.1248, 0.0262, 0.1196, 0.0495, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0250, 0.0175, 0.0196, 0.0138, 0.0180, 0.0194, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:03:47,041 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0336, 4.6698, 4.8773, 4.8431, 4.7480, 4.9554, 4.7989, 2.5987], device='cuda:1'), covar=tensor([0.0131, 0.0078, 0.0103, 0.0085, 0.0058, 0.0107, 0.0089, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0075, 0.0079, 0.0071, 0.0059, 0.0089, 0.0077, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:03:59,178 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 2.791e+02 3.242e+02 3.752e+02 6.527e+02, threshold=6.484e+02, percent-clipped=0.0 2023-05-16 08:04:03,998 INFO [finetune.py:992] (1/2) Epoch 7, batch 8450, loss[loss=0.1744, simple_loss=0.2563, pruned_loss=0.04629, over 12171.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2601, pruned_loss=0.04255, over 2379677.95 frames. ], batch size: 31, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:04:09,267 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3293, 2.3470, 3.5014, 4.1443, 3.7927, 4.2551, 3.6695, 2.8343], device='cuda:1'), covar=tensor([0.0038, 0.0380, 0.0141, 0.0048, 0.0103, 0.0075, 0.0116, 0.0393], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0120, 0.0102, 0.0075, 0.0101, 0.0112, 0.0092, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:04:19,898 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.30 vs. limit=5.0 2023-05-16 08:04:27,333 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181886.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:04:40,033 INFO [finetune.py:992] (1/2) Epoch 7, batch 8500, loss[loss=0.1594, simple_loss=0.2309, pruned_loss=0.04393, over 12294.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2605, pruned_loss=0.04278, over 2383281.18 frames. ], batch size: 28, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:05:10,403 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.826e+02 3.359e+02 3.857e+02 7.437e+02, threshold=6.718e+02, percent-clipped=3.0 2023-05-16 08:05:15,303 INFO [finetune.py:992] (1/2) Epoch 7, batch 8550, loss[loss=0.1786, simple_loss=0.2618, pruned_loss=0.04766, over 12049.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.262, pruned_loss=0.04389, over 2364570.07 frames. ], batch size: 37, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:05:15,419 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181953.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:05:33,213 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181978.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:05:53,048 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182001.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:05:54,369 INFO [finetune.py:992] (1/2) Epoch 7, batch 8600, loss[loss=0.162, simple_loss=0.2546, pruned_loss=0.03468, over 12114.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2613, pruned_loss=0.04336, over 2362509.06 frames. ], batch size: 38, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:05:59,632 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1711, 4.7839, 4.9639, 5.0857, 4.8822, 5.0129, 4.9704, 2.9871], device='cuda:1'), covar=tensor([0.0080, 0.0061, 0.0079, 0.0061, 0.0041, 0.0082, 0.0070, 0.0638], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0076, 0.0079, 0.0072, 0.0059, 0.0089, 0.0078, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:06:19,649 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5512, 4.8378, 3.0911, 3.0324, 4.1316, 2.6804, 4.1134, 3.5446], device='cuda:1'), covar=tensor([0.0600, 0.0439, 0.1063, 0.1334, 0.0290, 0.1360, 0.0405, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0251, 0.0175, 0.0197, 0.0138, 0.0180, 0.0195, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:06:21,071 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182039.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:06:25,789 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.615e+02 3.209e+02 4.105e+02 7.691e+02, threshold=6.418e+02, percent-clipped=1.0 2023-05-16 08:06:30,802 INFO [finetune.py:992] (1/2) Epoch 7, batch 8650, loss[loss=0.1726, simple_loss=0.2646, pruned_loss=0.04028, over 12341.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2611, pruned_loss=0.04307, over 2369149.19 frames. ], batch size: 35, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:07:06,557 INFO [finetune.py:992] (1/2) Epoch 7, batch 8700, loss[loss=0.1647, simple_loss=0.2608, pruned_loss=0.03428, over 12276.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2606, pruned_loss=0.04263, over 2372807.68 frames. ], batch size: 37, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:07:11,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-05-16 08:07:37,403 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 2.795e+02 3.270e+02 3.852e+02 5.826e+02, threshold=6.540e+02, percent-clipped=0.0 2023-05-16 08:07:38,271 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9491, 4.5314, 4.6907, 4.8501, 4.6863, 4.8000, 4.7220, 2.3467], device='cuda:1'), covar=tensor([0.0108, 0.0072, 0.0088, 0.0067, 0.0049, 0.0081, 0.0076, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0076, 0.0079, 0.0072, 0.0059, 0.0089, 0.0078, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:07:41,171 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2691, 4.1693, 2.6203, 2.4128, 3.6170, 2.3300, 3.6943, 2.9857], device='cuda:1'), covar=tensor([0.0630, 0.0471, 0.1112, 0.1477, 0.0312, 0.1362, 0.0434, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0252, 0.0175, 0.0197, 0.0139, 0.0180, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:07:42,288 INFO [finetune.py:992] (1/2) Epoch 7, batch 8750, loss[loss=0.1914, simple_loss=0.2818, pruned_loss=0.0505, over 12100.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2621, pruned_loss=0.04332, over 2372557.29 frames. ], batch size: 33, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:08:06,782 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182186.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:08:18,795 INFO [finetune.py:992] (1/2) Epoch 7, batch 8800, loss[loss=0.1666, simple_loss=0.2443, pruned_loss=0.04443, over 12115.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2623, pruned_loss=0.04369, over 2369924.04 frames. ], batch size: 30, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:08:40,323 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7646, 4.7644, 4.6874, 4.7046, 4.2706, 4.7813, 4.7770, 5.0350], device='cuda:1'), covar=tensor([0.0254, 0.0159, 0.0192, 0.0319, 0.0858, 0.0368, 0.0164, 0.0175], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0186, 0.0185, 0.0236, 0.0235, 0.0209, 0.0169, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 08:08:40,920 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182234.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:08:49,485 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 2.888e+02 3.251e+02 4.135e+02 1.544e+03, threshold=6.502e+02, percent-clipped=3.0 2023-05-16 08:08:54,408 INFO [finetune.py:992] (1/2) Epoch 7, batch 8850, loss[loss=0.1663, simple_loss=0.2527, pruned_loss=0.03996, over 12024.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2614, pruned_loss=0.04308, over 2377512.43 frames. ], batch size: 31, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:09:04,879 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 2023-05-16 08:09:15,485 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182281.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:09:30,510 INFO [finetune.py:992] (1/2) Epoch 7, batch 8900, loss[loss=0.1676, simple_loss=0.251, pruned_loss=0.04213, over 12094.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2624, pruned_loss=0.04327, over 2365152.31 frames. ], batch size: 32, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:09:53,210 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182334.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:09:58,968 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182342.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:10:01,588 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.160e+02 2.876e+02 3.336e+02 4.119e+02 1.220e+03, threshold=6.672e+02, percent-clipped=5.0 2023-05-16 08:10:06,507 INFO [finetune.py:992] (1/2) Epoch 7, batch 8950, loss[loss=0.1678, simple_loss=0.2583, pruned_loss=0.03869, over 12111.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2622, pruned_loss=0.0435, over 2361837.52 frames. ], batch size: 33, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:10:10,910 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3587, 5.1787, 5.3099, 5.3257, 4.9662, 4.9841, 4.7902, 5.2363], device='cuda:1'), covar=tensor([0.0570, 0.0561, 0.0682, 0.0570, 0.1737, 0.1221, 0.0522, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0518, 0.0681, 0.0575, 0.0618, 0.0830, 0.0726, 0.0540, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 08:10:42,571 INFO [finetune.py:992] (1/2) Epoch 7, batch 9000, loss[loss=0.1693, simple_loss=0.2575, pruned_loss=0.04053, over 12113.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2617, pruned_loss=0.04324, over 2365175.48 frames. ], batch size: 33, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:10:42,571 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 08:10:53,126 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9960, 2.7669, 3.4310, 2.2327, 2.4014, 2.9930, 2.7439, 3.0570], device='cuda:1'), covar=tensor([0.0554, 0.0944, 0.0299, 0.1099, 0.1489, 0.1040, 0.1059, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0234, 0.0240, 0.0182, 0.0234, 0.0288, 0.0224, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:11:01,077 INFO [finetune.py:1026] (1/2) Epoch 7, validation: loss=0.3275, simple_loss=0.4008, pruned_loss=0.1272, over 1020973.00 frames. 2023-05-16 08:11:01,078 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 08:11:32,195 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.081e+02 3.005e+02 3.281e+02 4.074e+02 8.769e+02, threshold=6.562e+02, percent-clipped=2.0 2023-05-16 08:11:36,951 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-05-16 08:11:37,073 INFO [finetune.py:992] (1/2) Epoch 7, batch 9050, loss[loss=0.1425, simple_loss=0.2237, pruned_loss=0.03067, over 12345.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2611, pruned_loss=0.0427, over 2375541.83 frames. ], batch size: 30, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:12:12,734 INFO [finetune.py:992] (1/2) Epoch 7, batch 9100, loss[loss=0.1753, simple_loss=0.2677, pruned_loss=0.04145, over 12131.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.262, pruned_loss=0.0433, over 2360261.75 frames. ], batch size: 39, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:12:21,233 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182514.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:12:43,821 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 2.864e+02 3.251e+02 4.199e+02 6.849e+02, threshold=6.503e+02, percent-clipped=1.0 2023-05-16 08:12:48,851 INFO [finetune.py:992] (1/2) Epoch 7, batch 9150, loss[loss=0.1779, simple_loss=0.2791, pruned_loss=0.03834, over 11837.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.262, pruned_loss=0.04334, over 2364628.92 frames. ], batch size: 44, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:13:05,451 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182575.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:13:25,231 INFO [finetune.py:992] (1/2) Epoch 7, batch 9200, loss[loss=0.2109, simple_loss=0.29, pruned_loss=0.06589, over 12333.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2627, pruned_loss=0.04314, over 2375573.32 frames. ], batch size: 36, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:13:47,208 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182634.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:13:49,276 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182637.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:13:55,663 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.075e+02 2.773e+02 3.268e+02 3.915e+02 8.909e+02, threshold=6.536e+02, percent-clipped=1.0 2023-05-16 08:14:01,374 INFO [finetune.py:992] (1/2) Epoch 7, batch 9250, loss[loss=0.162, simple_loss=0.255, pruned_loss=0.03447, over 12288.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2614, pruned_loss=0.04271, over 2375249.65 frames. ], batch size: 33, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:14:22,088 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182682.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:14:34,374 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8225, 2.8008, 4.4012, 4.6493, 2.9537, 2.6640, 2.8530, 2.1901], device='cuda:1'), covar=tensor([0.1311, 0.2952, 0.0486, 0.0401, 0.1109, 0.2048, 0.2630, 0.3675], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0372, 0.0266, 0.0290, 0.0256, 0.0286, 0.0360, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:14:37,025 INFO [finetune.py:992] (1/2) Epoch 7, batch 9300, loss[loss=0.1442, simple_loss=0.2367, pruned_loss=0.02587, over 12358.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.262, pruned_loss=0.04286, over 2373264.79 frames. ], batch size: 31, lr: 4.39e-03, grad_scale: 16.0 2023-05-16 08:15:08,404 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.808e+02 3.401e+02 4.148e+02 1.022e+03, threshold=6.801e+02, percent-clipped=3.0 2023-05-16 08:15:13,324 INFO [finetune.py:992] (1/2) Epoch 7, batch 9350, loss[loss=0.1662, simple_loss=0.2578, pruned_loss=0.03736, over 12136.00 frames. ], tot_loss[loss=0.173, simple_loss=0.261, pruned_loss=0.04248, over 2377056.82 frames. ], batch size: 36, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:15:17,849 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5075, 2.8605, 3.2537, 4.4414, 2.4047, 4.3299, 4.5360, 4.6269], device='cuda:1'), covar=tensor([0.0107, 0.1023, 0.0424, 0.0145, 0.1297, 0.0250, 0.0121, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0204, 0.0186, 0.0116, 0.0188, 0.0180, 0.0174, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:15:49,515 INFO [finetune.py:992] (1/2) Epoch 7, batch 9400, loss[loss=0.1666, simple_loss=0.2672, pruned_loss=0.03295, over 12289.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2615, pruned_loss=0.04232, over 2375820.01 frames. ], batch size: 37, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:16:11,376 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3338, 2.7297, 3.9502, 3.3236, 3.7364, 3.4261, 2.8578, 3.7187], device='cuda:1'), covar=tensor([0.0127, 0.0311, 0.0119, 0.0196, 0.0128, 0.0156, 0.0304, 0.0136], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0200, 0.0184, 0.0177, 0.0208, 0.0157, 0.0192, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:16:20,423 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.201e+02 2.963e+02 3.390e+02 3.941e+02 6.612e+02, threshold=6.780e+02, percent-clipped=0.0 2023-05-16 08:16:25,525 INFO [finetune.py:992] (1/2) Epoch 7, batch 9450, loss[loss=0.1497, simple_loss=0.2345, pruned_loss=0.03242, over 12351.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2604, pruned_loss=0.04191, over 2376845.21 frames. ], batch size: 31, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:16:27,955 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182856.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:16:32,874 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182862.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:16:38,370 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182870.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:16:55,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-16 08:16:56,885 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2466, 4.5225, 2.7039, 2.5507, 3.9010, 2.4765, 3.9252, 2.9951], device='cuda:1'), covar=tensor([0.0619, 0.0516, 0.1125, 0.1431, 0.0283, 0.1247, 0.0407, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0251, 0.0175, 0.0196, 0.0138, 0.0178, 0.0195, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:17:01,689 INFO [finetune.py:992] (1/2) Epoch 7, batch 9500, loss[loss=0.1703, simple_loss=0.2639, pruned_loss=0.03829, over 12364.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2597, pruned_loss=0.04198, over 2374752.75 frames. ], batch size: 36, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:17:12,016 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182917.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:17:16,148 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182923.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:17:26,737 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182937.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:17:32,936 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 2.717e+02 3.235e+02 4.149e+02 5.868e+02, threshold=6.470e+02, percent-clipped=0.0 2023-05-16 08:17:35,526 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 08:17:38,012 INFO [finetune.py:992] (1/2) Epoch 7, batch 9550, loss[loss=0.1708, simple_loss=0.2655, pruned_loss=0.03809, over 12174.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.26, pruned_loss=0.04194, over 2370017.70 frames. ], batch size: 36, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:17:38,121 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1546, 5.8878, 5.5522, 5.4482, 5.9813, 5.2470, 5.5116, 5.4517], device='cuda:1'), covar=tensor([0.1278, 0.0877, 0.0904, 0.1997, 0.0918, 0.2117, 0.1742, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0469, 0.0380, 0.0421, 0.0447, 0.0422, 0.0382, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:18:00,859 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182985.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:18:06,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-16 08:18:07,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 08:18:13,725 INFO [finetune.py:992] (1/2) Epoch 7, batch 9600, loss[loss=0.2013, simple_loss=0.2816, pruned_loss=0.06052, over 11750.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2606, pruned_loss=0.04191, over 2377790.04 frames. ], batch size: 44, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:18:19,556 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8492, 3.2733, 5.2001, 2.6456, 3.0207, 3.9133, 3.4492, 3.8895], device='cuda:1'), covar=tensor([0.0456, 0.1237, 0.0303, 0.1308, 0.1754, 0.1470, 0.1265, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0232, 0.0240, 0.0181, 0.0233, 0.0287, 0.0222, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:18:30,912 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1528, 5.1373, 4.9275, 5.0769, 4.6276, 5.1737, 5.1583, 5.4227], device='cuda:1'), covar=tensor([0.0206, 0.0138, 0.0212, 0.0256, 0.0716, 0.0308, 0.0142, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0185, 0.0184, 0.0232, 0.0233, 0.0207, 0.0167, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 08:18:35,180 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2554, 4.8015, 5.2137, 4.5811, 4.8791, 4.6194, 5.2336, 4.8840], device='cuda:1'), covar=tensor([0.0264, 0.0354, 0.0255, 0.0297, 0.0334, 0.0345, 0.0230, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0251, 0.0273, 0.0248, 0.0246, 0.0246, 0.0221, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:18:44,923 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 2.671e+02 3.193e+02 3.784e+02 7.294e+02, threshold=6.385e+02, percent-clipped=2.0 2023-05-16 08:18:49,913 INFO [finetune.py:992] (1/2) Epoch 7, batch 9650, loss[loss=0.1648, simple_loss=0.2546, pruned_loss=0.03748, over 12166.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2596, pruned_loss=0.04148, over 2383189.52 frames. ], batch size: 31, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:19:26,004 INFO [finetune.py:992] (1/2) Epoch 7, batch 9700, loss[loss=0.1786, simple_loss=0.265, pruned_loss=0.04609, over 12182.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2601, pruned_loss=0.04202, over 2375680.31 frames. ], batch size: 35, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:19:36,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 08:19:56,666 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.233e+02 2.941e+02 3.488e+02 4.384e+02 8.404e+02, threshold=6.976e+02, percent-clipped=4.0 2023-05-16 08:20:01,773 INFO [finetune.py:992] (1/2) Epoch 7, batch 9750, loss[loss=0.1682, simple_loss=0.2554, pruned_loss=0.0405, over 12141.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2599, pruned_loss=0.04212, over 2376170.17 frames. ], batch size: 36, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:20:14,272 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183170.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:20:37,914 INFO [finetune.py:992] (1/2) Epoch 7, batch 9800, loss[loss=0.1724, simple_loss=0.2569, pruned_loss=0.044, over 12297.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2607, pruned_loss=0.04255, over 2376046.87 frames. ], batch size: 33, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:20:44,451 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183212.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:20:48,528 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183218.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:20:48,544 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183218.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:20:50,020 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3257, 6.0067, 5.6521, 5.5749, 6.1110, 5.4108, 5.5891, 5.6054], device='cuda:1'), covar=tensor([0.1245, 0.0878, 0.1033, 0.1808, 0.0809, 0.1862, 0.1850, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0469, 0.0379, 0.0421, 0.0447, 0.0423, 0.0385, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:21:09,311 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.964e+02 3.446e+02 4.126e+02 1.231e+03, threshold=6.891e+02, percent-clipped=2.0 2023-05-16 08:21:14,320 INFO [finetune.py:992] (1/2) Epoch 7, batch 9850, loss[loss=0.1856, simple_loss=0.2707, pruned_loss=0.0502, over 11657.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2601, pruned_loss=0.04227, over 2376933.50 frames. ], batch size: 48, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:21:44,259 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1930, 5.9990, 5.4863, 5.4849, 6.0782, 5.4797, 5.5423, 5.5781], device='cuda:1'), covar=tensor([0.1243, 0.0846, 0.1130, 0.1786, 0.0846, 0.1943, 0.1689, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0471, 0.0381, 0.0422, 0.0450, 0.0423, 0.0385, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:21:50,517 INFO [finetune.py:992] (1/2) Epoch 7, batch 9900, loss[loss=0.1613, simple_loss=0.2549, pruned_loss=0.03387, over 12182.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2608, pruned_loss=0.04261, over 2371303.04 frames. ], batch size: 31, lr: 4.38e-03, grad_scale: 16.0 2023-05-16 08:22:20,990 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 2.943e+02 3.305e+02 4.141e+02 8.788e+02, threshold=6.610e+02, percent-clipped=4.0 2023-05-16 08:22:25,968 INFO [finetune.py:992] (1/2) Epoch 7, batch 9950, loss[loss=0.1422, simple_loss=0.2261, pruned_loss=0.0292, over 12135.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2617, pruned_loss=0.04307, over 2367816.99 frames. ], batch size: 30, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:22:30,366 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9889, 5.9976, 5.7023, 5.2044, 5.1067, 5.8809, 5.5698, 5.2450], device='cuda:1'), covar=tensor([0.0672, 0.0825, 0.0618, 0.1510, 0.0660, 0.0727, 0.1376, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0528, 0.0490, 0.0604, 0.0403, 0.0689, 0.0740, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 08:22:33,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 08:23:02,358 INFO [finetune.py:992] (1/2) Epoch 7, batch 10000, loss[loss=0.1719, simple_loss=0.2676, pruned_loss=0.0381, over 11554.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2626, pruned_loss=0.04363, over 2359617.53 frames. ], batch size: 48, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:23:06,148 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183408.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:23:32,564 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.814e+02 3.286e+02 3.856e+02 7.018e+02, threshold=6.571e+02, percent-clipped=1.0 2023-05-16 08:23:36,932 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2972, 4.4982, 2.8269, 2.5939, 3.8941, 2.5079, 3.9624, 3.1029], device='cuda:1'), covar=tensor([0.0718, 0.0505, 0.1098, 0.1431, 0.0293, 0.1273, 0.0435, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0253, 0.0176, 0.0198, 0.0140, 0.0180, 0.0197, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:23:38,169 INFO [finetune.py:992] (1/2) Epoch 7, batch 10050, loss[loss=0.1798, simple_loss=0.2632, pruned_loss=0.04817, over 12263.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2631, pruned_loss=0.04369, over 2366616.83 frames. ], batch size: 32, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:23:39,767 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9327, 4.8987, 4.8213, 4.7961, 4.3896, 4.9776, 4.9545, 5.1306], device='cuda:1'), covar=tensor([0.0225, 0.0155, 0.0194, 0.0381, 0.0813, 0.0265, 0.0142, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0187, 0.0186, 0.0235, 0.0236, 0.0209, 0.0170, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 08:23:49,676 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183469.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:23:51,077 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3498, 5.1808, 5.3037, 5.3335, 4.9501, 4.9779, 4.7520, 5.3243], device='cuda:1'), covar=tensor([0.0691, 0.0640, 0.0691, 0.0567, 0.1857, 0.1376, 0.0576, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0666, 0.0564, 0.0604, 0.0810, 0.0709, 0.0527, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 08:23:58,944 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2819, 4.7412, 5.2128, 4.6134, 4.8302, 4.6305, 5.2376, 4.8605], device='cuda:1'), covar=tensor([0.0251, 0.0376, 0.0304, 0.0262, 0.0342, 0.0322, 0.0224, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0255, 0.0278, 0.0250, 0.0249, 0.0250, 0.0224, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:24:14,324 INFO [finetune.py:992] (1/2) Epoch 7, batch 10100, loss[loss=0.1549, simple_loss=0.2449, pruned_loss=0.03249, over 12071.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2626, pruned_loss=0.04332, over 2371183.72 frames. ], batch size: 32, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:24:20,717 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183512.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:24:24,911 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183518.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:24:28,528 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1351, 5.1001, 5.0125, 5.0274, 4.6097, 5.1381, 5.1085, 5.3354], device='cuda:1'), covar=tensor([0.0228, 0.0132, 0.0168, 0.0300, 0.0736, 0.0286, 0.0128, 0.0145], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0187, 0.0186, 0.0236, 0.0236, 0.0210, 0.0170, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 08:24:35,778 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4287, 3.3739, 3.0895, 3.1077, 2.8298, 2.5793, 3.4082, 2.1727], device='cuda:1'), covar=tensor([0.0363, 0.0120, 0.0193, 0.0182, 0.0395, 0.0384, 0.0132, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0154, 0.0150, 0.0175, 0.0195, 0.0190, 0.0160, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:24:44,689 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.789e+02 3.208e+02 4.103e+02 9.250e+02, threshold=6.416e+02, percent-clipped=2.0 2023-05-16 08:24:49,698 INFO [finetune.py:992] (1/2) Epoch 7, batch 10150, loss[loss=0.1736, simple_loss=0.276, pruned_loss=0.03557, over 12269.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2625, pruned_loss=0.04324, over 2369466.72 frames. ], batch size: 37, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:24:54,787 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183560.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:24:59,114 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183566.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:25:18,556 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2558, 4.7272, 2.8527, 2.5790, 4.0330, 2.3383, 3.9911, 3.1947], device='cuda:1'), covar=tensor([0.0740, 0.0413, 0.1156, 0.1597, 0.0253, 0.1449, 0.0464, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0253, 0.0177, 0.0199, 0.0141, 0.0181, 0.0198, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:25:22,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5056, 4.9156, 3.1553, 2.8344, 4.1624, 2.7517, 4.1755, 3.5895], device='cuda:1'), covar=tensor([0.0680, 0.0407, 0.0962, 0.1344, 0.0249, 0.1150, 0.0441, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0253, 0.0177, 0.0199, 0.0141, 0.0181, 0.0198, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:25:26,269 INFO [finetune.py:992] (1/2) Epoch 7, batch 10200, loss[loss=0.1848, simple_loss=0.276, pruned_loss=0.04674, over 12126.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2614, pruned_loss=0.04288, over 2366757.84 frames. ], batch size: 39, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:25:34,272 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183614.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:25:40,087 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8365, 2.7724, 3.8500, 4.6424, 4.1315, 4.7455, 4.0253, 3.2749], device='cuda:1'), covar=tensor([0.0025, 0.0329, 0.0122, 0.0051, 0.0082, 0.0053, 0.0090, 0.0309], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0120, 0.0102, 0.0076, 0.0101, 0.0114, 0.0093, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:25:50,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1575, 6.1426, 5.9403, 5.4369, 5.3093, 6.0820, 5.7163, 5.4667], device='cuda:1'), covar=tensor([0.0655, 0.0762, 0.0657, 0.1514, 0.0599, 0.0696, 0.1504, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0598, 0.0528, 0.0493, 0.0605, 0.0402, 0.0692, 0.0744, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 08:25:57,615 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.740e+02 3.151e+02 3.893e+02 9.370e+02, threshold=6.301e+02, percent-clipped=2.0 2023-05-16 08:26:02,586 INFO [finetune.py:992] (1/2) Epoch 7, batch 10250, loss[loss=0.1626, simple_loss=0.2665, pruned_loss=0.02935, over 12276.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2617, pruned_loss=0.04297, over 2367073.85 frames. ], batch size: 37, lr: 4.38e-03, grad_scale: 32.0 2023-05-16 08:26:08,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 08:26:18,581 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183675.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:26:30,736 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0 2023-05-16 08:26:37,969 INFO [finetune.py:992] (1/2) Epoch 7, batch 10300, loss[loss=0.1611, simple_loss=0.2494, pruned_loss=0.03643, over 12099.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.261, pruned_loss=0.043, over 2364595.61 frames. ], batch size: 33, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:26:57,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.97 vs. limit=5.0 2023-05-16 08:27:08,681 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.995e+02 3.469e+02 4.013e+02 6.943e+02, threshold=6.938e+02, percent-clipped=2.0 2023-05-16 08:27:13,568 INFO [finetune.py:992] (1/2) Epoch 7, batch 10350, loss[loss=0.165, simple_loss=0.2547, pruned_loss=0.03763, over 12355.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2608, pruned_loss=0.04251, over 2370821.81 frames. ], batch size: 35, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:27:15,895 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2856, 4.9173, 5.1317, 5.1386, 4.8871, 5.1739, 5.0377, 2.8290], device='cuda:1'), covar=tensor([0.0067, 0.0048, 0.0056, 0.0046, 0.0048, 0.0069, 0.0057, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0078, 0.0070, 0.0058, 0.0088, 0.0077, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:27:21,465 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183764.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:27:24,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 08:27:45,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1975, 2.7377, 3.7464, 3.1820, 3.5962, 3.2332, 2.7002, 3.6238], device='cuda:1'), covar=tensor([0.0126, 0.0288, 0.0154, 0.0217, 0.0134, 0.0173, 0.0312, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0199, 0.0183, 0.0176, 0.0206, 0.0154, 0.0192, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:27:50,106 INFO [finetune.py:992] (1/2) Epoch 7, batch 10400, loss[loss=0.1753, simple_loss=0.2683, pruned_loss=0.04116, over 11713.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2602, pruned_loss=0.04204, over 2373855.51 frames. ], batch size: 48, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:28:08,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5619, 5.4498, 5.4858, 5.5765, 5.1863, 5.1745, 5.0146, 5.5275], device='cuda:1'), covar=tensor([0.0594, 0.0438, 0.0601, 0.0482, 0.1581, 0.1195, 0.0445, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0673, 0.0573, 0.0610, 0.0823, 0.0716, 0.0532, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 08:28:17,790 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183842.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:28:20,390 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 3.060e+02 3.451e+02 4.187e+02 8.708e+02, threshold=6.903e+02, percent-clipped=1.0 2023-05-16 08:28:25,334 INFO [finetune.py:992] (1/2) Epoch 7, batch 10450, loss[loss=0.1919, simple_loss=0.28, pruned_loss=0.05184, over 10562.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2598, pruned_loss=0.04177, over 2372378.10 frames. ], batch size: 68, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:29:01,471 INFO [finetune.py:992] (1/2) Epoch 7, batch 10500, loss[loss=0.1979, simple_loss=0.2793, pruned_loss=0.05829, over 12040.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2608, pruned_loss=0.04221, over 2374693.67 frames. ], batch size: 37, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:29:01,701 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183903.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:29:32,327 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.051e+02 2.934e+02 3.579e+02 4.359e+02 7.289e+02, threshold=7.158e+02, percent-clipped=1.0 2023-05-16 08:29:37,362 INFO [finetune.py:992] (1/2) Epoch 7, batch 10550, loss[loss=0.1655, simple_loss=0.2512, pruned_loss=0.03993, over 11826.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2609, pruned_loss=0.04239, over 2376523.21 frames. ], batch size: 26, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:29:49,487 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183970.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:30:15,905 INFO [finetune.py:992] (1/2) Epoch 7, batch 10600, loss[loss=0.1502, simple_loss=0.2389, pruned_loss=0.0307, over 12026.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2602, pruned_loss=0.04175, over 2386571.14 frames. ], batch size: 31, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:30:46,780 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.156e+02 2.875e+02 3.336e+02 4.059e+02 8.731e+02, threshold=6.672e+02, percent-clipped=2.0 2023-05-16 08:30:49,947 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184050.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:30:51,821 INFO [finetune.py:992] (1/2) Epoch 7, batch 10650, loss[loss=0.176, simple_loss=0.2687, pruned_loss=0.04168, over 12285.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2605, pruned_loss=0.0418, over 2385778.46 frames. ], batch size: 33, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:31:00,029 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184064.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:31:04,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2023-05-16 08:31:10,132 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7001, 3.2530, 5.0940, 2.7853, 2.7534, 3.7193, 3.0976, 3.8221], device='cuda:1'), covar=tensor([0.0381, 0.1122, 0.0292, 0.1144, 0.1847, 0.1435, 0.1292, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0233, 0.0245, 0.0182, 0.0236, 0.0288, 0.0225, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:31:28,359 INFO [finetune.py:992] (1/2) Epoch 7, batch 10700, loss[loss=0.1693, simple_loss=0.2631, pruned_loss=0.03769, over 11999.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2597, pruned_loss=0.04137, over 2393747.17 frames. ], batch size: 42, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:31:34,200 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184111.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:31:34,766 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184112.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:31:58,554 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.790e+02 3.201e+02 3.867e+02 6.972e+02, threshold=6.402e+02, percent-clipped=2.0 2023-05-16 08:32:03,559 INFO [finetune.py:992] (1/2) Epoch 7, batch 10750, loss[loss=0.1769, simple_loss=0.2598, pruned_loss=0.04699, over 12339.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2603, pruned_loss=0.04193, over 2384746.92 frames. ], batch size: 31, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:32:22,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 08:32:24,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-05-16 08:32:35,840 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184198.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:32:38,356 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1164, 2.6409, 3.8171, 3.1209, 3.5923, 3.1534, 2.6291, 3.5969], device='cuda:1'), covar=tensor([0.0134, 0.0281, 0.0156, 0.0195, 0.0155, 0.0173, 0.0307, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0200, 0.0182, 0.0177, 0.0207, 0.0156, 0.0193, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:32:39,516 INFO [finetune.py:992] (1/2) Epoch 7, batch 10800, loss[loss=0.1745, simple_loss=0.2663, pruned_loss=0.04138, over 12152.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2608, pruned_loss=0.0421, over 2381143.82 frames. ], batch size: 34, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:32:45,542 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8398, 2.9236, 4.6239, 4.9156, 3.0956, 2.7209, 2.9326, 2.2885], device='cuda:1'), covar=tensor([0.1314, 0.2756, 0.0421, 0.0326, 0.1097, 0.2079, 0.2639, 0.3571], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0371, 0.0265, 0.0288, 0.0254, 0.0284, 0.0358, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:32:47,584 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9837, 2.3853, 3.5387, 2.9073, 3.3746, 3.0395, 2.4330, 3.4129], device='cuda:1'), covar=tensor([0.0123, 0.0319, 0.0140, 0.0210, 0.0128, 0.0155, 0.0322, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0200, 0.0183, 0.0178, 0.0208, 0.0156, 0.0193, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:33:10,723 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.798e+02 2.797e+02 3.306e+02 4.077e+02 6.295e+02, threshold=6.611e+02, percent-clipped=0.0 2023-05-16 08:33:15,685 INFO [finetune.py:992] (1/2) Epoch 7, batch 10850, loss[loss=0.2608, simple_loss=0.328, pruned_loss=0.09683, over 8204.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2609, pruned_loss=0.0422, over 2380841.64 frames. ], batch size: 98, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:33:27,636 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184270.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:33:51,622 INFO [finetune.py:992] (1/2) Epoch 7, batch 10900, loss[loss=0.165, simple_loss=0.2611, pruned_loss=0.03446, over 12004.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2613, pruned_loss=0.04234, over 2376798.09 frames. ], batch size: 42, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:34:02,487 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184318.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:34:22,842 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.205e+02 2.977e+02 3.512e+02 4.014e+02 7.651e+02, threshold=7.023e+02, percent-clipped=3.0 2023-05-16 08:34:27,665 INFO [finetune.py:992] (1/2) Epoch 7, batch 10950, loss[loss=0.1535, simple_loss=0.2357, pruned_loss=0.03567, over 12022.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2617, pruned_loss=0.04258, over 2377534.73 frames. ], batch size: 31, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:34:32,159 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184359.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:34:41,316 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1534, 2.6161, 3.7895, 3.1571, 3.5478, 3.2984, 2.6634, 3.5656], device='cuda:1'), covar=tensor([0.0116, 0.0311, 0.0145, 0.0197, 0.0135, 0.0157, 0.0292, 0.0127], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0199, 0.0182, 0.0176, 0.0206, 0.0155, 0.0191, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:34:47,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 08:35:04,165 INFO [finetune.py:992] (1/2) Epoch 7, batch 11000, loss[loss=0.2411, simple_loss=0.3311, pruned_loss=0.07554, over 10301.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2653, pruned_loss=0.04465, over 2332242.32 frames. ], batch size: 68, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:35:06,342 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184406.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:35:16,349 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:35:34,559 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 3.155e+02 3.689e+02 4.357e+02 7.918e+02, threshold=7.378e+02, percent-clipped=3.0 2023-05-16 08:35:40,147 INFO [finetune.py:992] (1/2) Epoch 7, batch 11050, loss[loss=0.1801, simple_loss=0.2767, pruned_loss=0.04176, over 12084.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2686, pruned_loss=0.04671, over 2291873.16 frames. ], batch size: 40, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:36:03,883 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184486.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:36:12,171 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184498.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:36:15,442 INFO [finetune.py:992] (1/2) Epoch 7, batch 11100, loss[loss=0.1561, simple_loss=0.2496, pruned_loss=0.03131, over 12350.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2729, pruned_loss=0.04952, over 2252556.55 frames. ], batch size: 31, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:36:33,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 08:36:45,888 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.179e+02 3.470e+02 4.055e+02 4.748e+02 9.924e+02, threshold=8.110e+02, percent-clipped=6.0 2023-05-16 08:36:45,987 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184546.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:36:46,786 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184547.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:36:51,508 INFO [finetune.py:992] (1/2) Epoch 7, batch 11150, loss[loss=0.1588, simple_loss=0.2524, pruned_loss=0.03261, over 12346.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2777, pruned_loss=0.05251, over 2215957.51 frames. ], batch size: 30, lr: 4.37e-03, grad_scale: 32.0 2023-05-16 08:37:20,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 08:37:25,979 INFO [finetune.py:992] (1/2) Epoch 7, batch 11200, loss[loss=0.2761, simple_loss=0.3501, pruned_loss=0.1011, over 7459.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2851, pruned_loss=0.05765, over 2144060.19 frames. ], batch size: 99, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:37:33,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-16 08:37:57,914 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.468e+02 3.616e+02 4.361e+02 5.521e+02 1.127e+03, threshold=8.722e+02, percent-clipped=4.0 2023-05-16 08:38:02,375 INFO [finetune.py:992] (1/2) Epoch 7, batch 11250, loss[loss=0.2815, simple_loss=0.3408, pruned_loss=0.1111, over 7338.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2922, pruned_loss=0.06226, over 2090860.54 frames. ], batch size: 97, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:38:37,119 INFO [finetune.py:992] (1/2) Epoch 7, batch 11300, loss[loss=0.2648, simple_loss=0.3549, pruned_loss=0.08737, over 10208.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2971, pruned_loss=0.06524, over 2055663.97 frames. ], batch size: 68, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:38:40,042 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184706.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:38:46,062 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184715.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:39:08,030 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.431e+02 3.560e+02 4.219e+02 5.102e+02 7.690e+02, threshold=8.438e+02, percent-clipped=0.0 2023-05-16 08:39:12,211 INFO [finetune.py:992] (1/2) Epoch 7, batch 11350, loss[loss=0.2463, simple_loss=0.3272, pruned_loss=0.08272, over 12006.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.3018, pruned_loss=0.06781, over 2023731.84 frames. ], batch size: 40, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:39:13,613 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184754.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:39:45,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-05-16 08:39:47,662 INFO [finetune.py:992] (1/2) Epoch 7, batch 11400, loss[loss=0.217, simple_loss=0.2982, pruned_loss=0.06789, over 11498.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3059, pruned_loss=0.07075, over 1976858.43 frames. ], batch size: 48, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:39:56,988 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8624, 2.5766, 3.3899, 3.5095, 2.9388, 2.6961, 2.6799, 2.4733], device='cuda:1'), covar=tensor([0.0891, 0.1978, 0.0521, 0.0422, 0.0756, 0.1599, 0.2020, 0.2871], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0364, 0.0261, 0.0283, 0.0250, 0.0280, 0.0354, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:40:15,366 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184842.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:40:18,630 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.447e+02 3.774e+02 4.236e+02 5.048e+02 8.901e+02, threshold=8.472e+02, percent-clipped=1.0 2023-05-16 08:40:22,834 INFO [finetune.py:992] (1/2) Epoch 7, batch 11450, loss[loss=0.2186, simple_loss=0.3088, pruned_loss=0.06421, over 11588.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3103, pruned_loss=0.07437, over 1925689.46 frames. ], batch size: 25, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:40:29,273 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6734, 2.7653, 3.9541, 4.1198, 3.0272, 2.6622, 2.8470, 2.0741], device='cuda:1'), covar=tensor([0.1327, 0.2440, 0.0484, 0.0410, 0.0997, 0.1995, 0.2399, 0.3961], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0364, 0.0261, 0.0282, 0.0250, 0.0280, 0.0354, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:40:31,863 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184866.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:40:36,713 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3106, 3.5205, 3.2473, 3.6480, 3.4043, 2.5074, 3.1038, 2.8792], device='cuda:1'), covar=tensor([0.0733, 0.0808, 0.1266, 0.0492, 0.1104, 0.1543, 0.1195, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0353, 0.0335, 0.0262, 0.0343, 0.0255, 0.0320, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:40:57,131 INFO [finetune.py:992] (1/2) Epoch 7, batch 11500, loss[loss=0.2634, simple_loss=0.3347, pruned_loss=0.0961, over 7370.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3128, pruned_loss=0.07664, over 1886398.23 frames. ], batch size: 98, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:40:59,384 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9734, 2.2472, 3.5130, 2.8139, 3.2240, 3.1187, 2.4003, 3.3664], device='cuda:1'), covar=tensor([0.0107, 0.0394, 0.0064, 0.0224, 0.0112, 0.0130, 0.0313, 0.0096], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0195, 0.0174, 0.0171, 0.0198, 0.0148, 0.0185, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:41:14,063 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184927.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:41:22,274 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184939.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:41:28,236 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.564e+02 3.530e+02 3.937e+02 4.658e+02 1.050e+03, threshold=7.874e+02, percent-clipped=1.0 2023-05-16 08:41:32,249 INFO [finetune.py:992] (1/2) Epoch 7, batch 11550, loss[loss=0.2109, simple_loss=0.2936, pruned_loss=0.06403, over 12182.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3146, pruned_loss=0.07857, over 1866713.30 frames. ], batch size: 31, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:42:05,499 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185000.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:42:07,365 INFO [finetune.py:992] (1/2) Epoch 7, batch 11600, loss[loss=0.1921, simple_loss=0.2847, pruned_loss=0.04975, over 12103.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3144, pruned_loss=0.07879, over 1853370.53 frames. ], batch size: 32, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:42:16,110 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185015.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:42:39,268 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.815e+02 3.850e+02 4.381e+02 5.033e+02 1.342e+03, threshold=8.762e+02, percent-clipped=5.0 2023-05-16 08:42:43,818 INFO [finetune.py:992] (1/2) Epoch 7, batch 11650, loss[loss=0.2682, simple_loss=0.3342, pruned_loss=0.1011, over 6616.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3147, pruned_loss=0.08028, over 1817281.05 frames. ], batch size: 97, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:42:49,138 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185060.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:42:51,003 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185063.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:43:19,131 INFO [finetune.py:992] (1/2) Epoch 7, batch 11700, loss[loss=0.1963, simple_loss=0.2877, pruned_loss=0.05243, over 10330.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.314, pruned_loss=0.08027, over 1788777.60 frames. ], batch size: 68, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:43:31,383 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185121.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 08:43:37,507 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185129.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:43:45,970 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185142.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:43:49,173 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.257e+02 3.357e+02 4.006e+02 4.852e+02 1.709e+03, threshold=8.012e+02, percent-clipped=2.0 2023-05-16 08:43:53,189 INFO [finetune.py:992] (1/2) Epoch 7, batch 11750, loss[loss=0.2729, simple_loss=0.3431, pruned_loss=0.1013, over 6862.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3151, pruned_loss=0.08173, over 1761866.53 frames. ], batch size: 98, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:44:18,939 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185190.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:44:19,100 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185190.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:44:20,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-05-16 08:44:28,634 INFO [finetune.py:992] (1/2) Epoch 7, batch 11800, loss[loss=0.2184, simple_loss=0.3, pruned_loss=0.06845, over 11792.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3175, pruned_loss=0.08395, over 1725005.09 frames. ], batch size: 44, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:44:42,313 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185222.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:44:51,860 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4579, 4.4058, 4.3370, 4.0410, 4.0406, 4.4403, 4.1886, 4.0054], device='cuda:1'), covar=tensor([0.0704, 0.0942, 0.0704, 0.1185, 0.2192, 0.0785, 0.1307, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0497, 0.0468, 0.0567, 0.0375, 0.0638, 0.0688, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 08:44:59,077 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.628e+02 3.541e+02 4.087e+02 4.932e+02 1.149e+03, threshold=8.173e+02, percent-clipped=3.0 2023-05-16 08:45:03,774 INFO [finetune.py:992] (1/2) Epoch 7, batch 11850, loss[loss=0.2097, simple_loss=0.3098, pruned_loss=0.05478, over 10295.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3192, pruned_loss=0.08461, over 1705052.24 frames. ], batch size: 68, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:45:06,826 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 08:45:27,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 08:45:32,683 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185295.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:45:37,876 INFO [finetune.py:992] (1/2) Epoch 7, batch 11900, loss[loss=0.2301, simple_loss=0.3108, pruned_loss=0.07466, over 7334.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3188, pruned_loss=0.08375, over 1684322.03 frames. ], batch size: 98, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:46:09,047 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.271e+02 3.494e+02 4.092e+02 4.883e+02 7.543e+02, threshold=8.183e+02, percent-clipped=0.0 2023-05-16 08:46:13,132 INFO [finetune.py:992] (1/2) Epoch 7, batch 11950, loss[loss=0.1896, simple_loss=0.2789, pruned_loss=0.05017, over 10222.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.315, pruned_loss=0.08059, over 1678346.29 frames. ], batch size: 68, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:46:48,698 INFO [finetune.py:992] (1/2) Epoch 7, batch 12000, loss[loss=0.2133, simple_loss=0.2882, pruned_loss=0.06916, over 7275.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3104, pruned_loss=0.07696, over 1679270.07 frames. ], batch size: 102, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:46:48,698 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 08:47:06,953 INFO [finetune.py:1026] (1/2) Epoch 7, validation: loss=0.2884, simple_loss=0.3646, pruned_loss=0.106, over 1020973.00 frames. 2023-05-16 08:47:06,954 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 08:47:16,077 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185416.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 08:47:28,865 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2598, 3.4666, 3.3599, 3.9493, 2.9175, 3.4651, 2.2703, 3.0422], device='cuda:1'), covar=tensor([0.1845, 0.0994, 0.1190, 0.0628, 0.1350, 0.0823, 0.2442, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0251, 0.0280, 0.0326, 0.0227, 0.0230, 0.0248, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:47:37,286 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.158e+02 3.019e+02 3.476e+02 4.163e+02 9.429e+02, threshold=6.953e+02, percent-clipped=2.0 2023-05-16 08:47:41,311 INFO [finetune.py:992] (1/2) Epoch 7, batch 12050, loss[loss=0.203, simple_loss=0.2919, pruned_loss=0.05704, over 12020.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3061, pruned_loss=0.07365, over 1687492.64 frames. ], batch size: 42, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:47:51,357 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185468.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:47:55,758 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9980, 2.2086, 2.2106, 2.2461, 1.9983, 1.9646, 2.1459, 1.6855], device='cuda:1'), covar=tensor([0.0289, 0.0194, 0.0172, 0.0191, 0.0322, 0.0225, 0.0161, 0.0363], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0146, 0.0144, 0.0170, 0.0186, 0.0182, 0.0151, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 08:47:56,471 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7711, 2.4551, 3.5923, 3.6562, 2.8947, 2.6887, 2.6198, 2.4012], device='cuda:1'), covar=tensor([0.1179, 0.2489, 0.0559, 0.0437, 0.0915, 0.1865, 0.2491, 0.3667], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0360, 0.0255, 0.0279, 0.0247, 0.0278, 0.0352, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:48:01,860 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185485.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:48:13,818 INFO [finetune.py:992] (1/2) Epoch 7, batch 12100, loss[loss=0.2748, simple_loss=0.3468, pruned_loss=0.1014, over 7079.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.3051, pruned_loss=0.07264, over 1694266.33 frames. ], batch size: 98, lr: 4.36e-03, grad_scale: 16.0 2023-05-16 08:48:17,201 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7986, 3.6816, 3.6987, 3.7949, 3.4979, 3.7994, 3.8236, 3.9160], device='cuda:1'), covar=tensor([0.0219, 0.0157, 0.0169, 0.0237, 0.0548, 0.0297, 0.0183, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0161, 0.0157, 0.0202, 0.0202, 0.0178, 0.0145, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 08:48:26,000 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185522.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:48:28,619 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1585, 3.9445, 2.4238, 2.2423, 3.5922, 2.2546, 3.5995, 2.7919], device='cuda:1'), covar=tensor([0.0722, 0.0439, 0.1283, 0.1795, 0.0305, 0.1630, 0.0493, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0231, 0.0167, 0.0188, 0.0130, 0.0173, 0.0182, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:48:30,538 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185529.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:48:41,728 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.380e+02 3.250e+02 3.695e+02 4.693e+02 9.793e+02, threshold=7.390e+02, percent-clipped=3.0 2023-05-16 08:48:46,019 INFO [finetune.py:992] (1/2) Epoch 7, batch 12150, loss[loss=0.2178, simple_loss=0.293, pruned_loss=0.07128, over 7181.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3059, pruned_loss=0.07254, over 1715772.53 frames. ], batch size: 99, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:48:56,661 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185570.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:48:58,643 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6271, 2.7425, 4.3707, 4.4761, 2.9594, 2.6747, 2.9093, 1.9138], device='cuda:1'), covar=tensor([0.1522, 0.2853, 0.0447, 0.0392, 0.1265, 0.2420, 0.2648, 0.4730], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0363, 0.0257, 0.0280, 0.0249, 0.0281, 0.0354, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:49:06,361 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9921, 2.2651, 2.2259, 2.2615, 2.0977, 1.9900, 2.1712, 1.7429], device='cuda:1'), covar=tensor([0.0310, 0.0176, 0.0193, 0.0198, 0.0323, 0.0243, 0.0160, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0144, 0.0143, 0.0169, 0.0184, 0.0180, 0.0150, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 08:49:12,565 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185595.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:49:16,592 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3498, 2.9371, 3.6583, 2.3810, 2.5427, 2.9698, 2.8468, 3.0733], device='cuda:1'), covar=tensor([0.0536, 0.1105, 0.0311, 0.1448, 0.1942, 0.1514, 0.1250, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0217, 0.0218, 0.0172, 0.0221, 0.0263, 0.0209, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:49:17,627 INFO [finetune.py:992] (1/2) Epoch 7, batch 12200, loss[loss=0.2642, simple_loss=0.3303, pruned_loss=0.09908, over 6492.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3066, pruned_loss=0.07315, over 1709443.29 frames. ], batch size: 98, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:50:02,405 INFO [finetune.py:992] (1/2) Epoch 8, batch 0, loss[loss=0.1998, simple_loss=0.2866, pruned_loss=0.05655, over 12066.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2866, pruned_loss=0.05655, over 12066.00 frames. ], batch size: 45, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:50:02,405 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 08:50:20,129 INFO [finetune.py:1026] (1/2) Epoch 8, validation: loss=0.2881, simple_loss=0.3643, pruned_loss=0.1059, over 1020973.00 frames. 2023-05-16 08:50:20,130 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 08:50:24,972 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185643.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:50:27,582 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.234e+02 3.471e+02 4.101e+02 4.868e+02 9.966e+02, threshold=8.203e+02, percent-clipped=2.0 2023-05-16 08:50:53,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.33 vs. limit=5.0 2023-05-16 08:50:55,818 INFO [finetune.py:992] (1/2) Epoch 8, batch 50, loss[loss=0.1659, simple_loss=0.2503, pruned_loss=0.04071, over 12130.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2768, pruned_loss=0.04758, over 541692.54 frames. ], batch size: 30, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:51:16,967 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185716.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 08:51:17,076 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4491, 3.3881, 3.0346, 3.1536, 2.7579, 2.7025, 3.4424, 2.2118], device='cuda:1'), covar=tensor([0.0373, 0.0143, 0.0191, 0.0210, 0.0411, 0.0334, 0.0131, 0.0461], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0146, 0.0144, 0.0170, 0.0186, 0.0182, 0.0151, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 08:51:31,695 INFO [finetune.py:992] (1/2) Epoch 8, batch 100, loss[loss=0.2445, simple_loss=0.3229, pruned_loss=0.08303, over 8211.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2721, pruned_loss=0.04622, over 948554.77 frames. ], batch size: 98, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:51:38,813 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.687e+02 3.242e+02 3.856e+02 6.671e+02, threshold=6.484e+02, percent-clipped=0.0 2023-05-16 08:51:47,589 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185759.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:51:51,015 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185764.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:52:05,507 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185785.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:52:06,843 INFO [finetune.py:992] (1/2) Epoch 8, batch 150, loss[loss=0.1422, simple_loss=0.2214, pruned_loss=0.03147, over 12007.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2719, pruned_loss=0.04643, over 1258896.02 frames. ], batch size: 28, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:52:11,861 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4252, 4.6782, 4.2437, 4.9697, 4.6456, 2.6774, 4.2177, 2.9800], device='cuda:1'), covar=tensor([0.0716, 0.0832, 0.1319, 0.0521, 0.1048, 0.1973, 0.1145, 0.3532], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0359, 0.0342, 0.0263, 0.0347, 0.0260, 0.0325, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:52:15,897 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185799.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 08:52:31,175 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185820.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 08:52:32,463 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6348, 5.6271, 5.4365, 4.9936, 4.9305, 5.5307, 5.1837, 5.0484], device='cuda:1'), covar=tensor([0.0728, 0.0974, 0.0720, 0.1462, 0.0889, 0.0826, 0.1547, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0499, 0.0468, 0.0571, 0.0376, 0.0642, 0.0691, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:52:33,807 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185824.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:52:40,218 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185833.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:52:42,952 INFO [finetune.py:992] (1/2) Epoch 8, batch 200, loss[loss=0.163, simple_loss=0.2468, pruned_loss=0.03963, over 12285.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.27, pruned_loss=0.04617, over 1513182.71 frames. ], batch size: 28, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:52:49,902 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.956e+02 3.477e+02 4.036e+02 6.994e+02, threshold=6.953e+02, percent-clipped=1.0 2023-05-16 08:52:59,294 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185860.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 08:53:06,419 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2951, 3.5242, 3.2738, 3.6456, 3.4249, 2.3745, 3.1647, 2.7060], device='cuda:1'), covar=tensor([0.0924, 0.1057, 0.1535, 0.0820, 0.1241, 0.1983, 0.1410, 0.3327], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0358, 0.0342, 0.0263, 0.0347, 0.0260, 0.0325, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:53:07,137 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7597, 3.2207, 5.1343, 2.4614, 2.7266, 3.7476, 3.1075, 3.7847], device='cuda:1'), covar=tensor([0.0437, 0.1257, 0.0434, 0.1358, 0.2079, 0.1640, 0.1406, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0224, 0.0226, 0.0177, 0.0227, 0.0272, 0.0216, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:53:14,216 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6994, 2.6407, 3.6954, 4.6165, 4.1338, 4.6573, 4.1146, 3.0947], device='cuda:1'), covar=tensor([0.0024, 0.0345, 0.0110, 0.0025, 0.0105, 0.0053, 0.0075, 0.0304], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0115, 0.0096, 0.0071, 0.0095, 0.0108, 0.0087, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:53:17,884 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0639, 3.0857, 4.4551, 2.2892, 2.6110, 3.3395, 2.9942, 3.4740], device='cuda:1'), covar=tensor([0.0509, 0.1186, 0.0361, 0.1424, 0.1979, 0.1500, 0.1357, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0224, 0.0227, 0.0177, 0.0227, 0.0272, 0.0216, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:53:19,067 INFO [finetune.py:992] (1/2) Epoch 8, batch 250, loss[loss=0.1613, simple_loss=0.2512, pruned_loss=0.03563, over 11622.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2673, pruned_loss=0.04523, over 1694428.24 frames. ], batch size: 48, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:53:38,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.23 vs. limit=5.0 2023-05-16 08:53:54,369 INFO [finetune.py:992] (1/2) Epoch 8, batch 300, loss[loss=0.179, simple_loss=0.2673, pruned_loss=0.04536, over 11229.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2664, pruned_loss=0.04483, over 1852776.94 frames. ], batch size: 55, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:54:00,201 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2980, 2.5418, 3.1616, 4.1906, 2.3567, 4.2100, 4.2313, 4.3950], device='cuda:1'), covar=tensor([0.0104, 0.1201, 0.0479, 0.0121, 0.1244, 0.0217, 0.0149, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0201, 0.0177, 0.0109, 0.0185, 0.0169, 0.0165, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:54:02,108 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.706e+02 3.258e+02 3.879e+02 6.810e+02, threshold=6.516e+02, percent-clipped=0.0 2023-05-16 08:54:03,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2833, 4.4114, 2.5046, 2.6330, 3.8930, 2.5199, 3.8288, 3.0701], device='cuda:1'), covar=tensor([0.0643, 0.0528, 0.1233, 0.1349, 0.0262, 0.1292, 0.0464, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0235, 0.0169, 0.0190, 0.0132, 0.0175, 0.0183, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:54:05,112 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5803, 2.4894, 3.6150, 4.4839, 4.0126, 4.4832, 3.9078, 2.9431], device='cuda:1'), covar=tensor([0.0025, 0.0351, 0.0132, 0.0032, 0.0103, 0.0064, 0.0103, 0.0335], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0116, 0.0097, 0.0072, 0.0096, 0.0109, 0.0088, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:54:11,423 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3503, 6.1212, 5.7368, 5.6869, 6.2208, 5.5184, 5.7994, 5.6778], device='cuda:1'), covar=tensor([0.1754, 0.1027, 0.1147, 0.2474, 0.1113, 0.2663, 0.1576, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0451, 0.0371, 0.0409, 0.0431, 0.0413, 0.0365, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 08:54:30,314 INFO [finetune.py:992] (1/2) Epoch 8, batch 350, loss[loss=0.1742, simple_loss=0.2561, pruned_loss=0.04612, over 12187.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2668, pruned_loss=0.04455, over 1966499.03 frames. ], batch size: 31, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:54:52,753 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0327, 5.8513, 5.4265, 5.3703, 5.9314, 5.3002, 5.5331, 5.4318], device='cuda:1'), covar=tensor([0.1527, 0.0962, 0.0986, 0.2146, 0.1013, 0.2235, 0.1916, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0449, 0.0371, 0.0408, 0.0431, 0.0413, 0.0365, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 08:55:09,935 INFO [finetune.py:992] (1/2) Epoch 8, batch 400, loss[loss=0.1595, simple_loss=0.2416, pruned_loss=0.0387, over 12037.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2663, pruned_loss=0.04444, over 2057933.44 frames. ], batch size: 28, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:55:17,164 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.883e+02 3.350e+02 4.048e+02 8.627e+02, threshold=6.701e+02, percent-clipped=3.0 2023-05-16 08:55:46,532 INFO [finetune.py:992] (1/2) Epoch 8, batch 450, loss[loss=0.1771, simple_loss=0.2594, pruned_loss=0.04734, over 10379.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2649, pruned_loss=0.04389, over 2130971.57 frames. ], batch size: 68, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:55:55,956 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186100.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:56:06,445 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186115.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 08:56:12,826 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186124.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:56:21,564 INFO [finetune.py:992] (1/2) Epoch 8, batch 500, loss[loss=0.1492, simple_loss=0.2298, pruned_loss=0.03428, over 12118.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2649, pruned_loss=0.04363, over 2192301.52 frames. ], batch size: 30, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:56:21,838 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8969, 3.3314, 5.2095, 2.9895, 3.0745, 3.8813, 3.3685, 3.9420], device='cuda:1'), covar=tensor([0.0436, 0.1320, 0.0285, 0.1128, 0.1867, 0.1410, 0.1385, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0229, 0.0232, 0.0180, 0.0232, 0.0278, 0.0220, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:56:28,709 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 2.866e+02 3.465e+02 4.025e+02 1.738e+03, threshold=6.929e+02, percent-clipped=2.0 2023-05-16 08:56:35,132 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186155.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 08:56:37,323 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186158.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:56:38,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3703, 3.5174, 3.1425, 3.2494, 2.8197, 2.6505, 3.5103, 2.3232], device='cuda:1'), covar=tensor([0.0377, 0.0115, 0.0175, 0.0179, 0.0415, 0.0390, 0.0152, 0.0452], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0151, 0.0150, 0.0175, 0.0193, 0.0189, 0.0157, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:56:39,440 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186161.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:56:43,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 08:56:46,994 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186172.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:56:57,532 INFO [finetune.py:992] (1/2) Epoch 8, batch 550, loss[loss=0.1672, simple_loss=0.266, pruned_loss=0.03415, over 12129.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2643, pruned_loss=0.04319, over 2237264.31 frames. ], batch size: 39, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:57:20,680 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186219.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:57:28,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-05-16 08:57:33,989 INFO [finetune.py:992] (1/2) Epoch 8, batch 600, loss[loss=0.1797, simple_loss=0.2648, pruned_loss=0.04724, over 12065.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2628, pruned_loss=0.04251, over 2271833.33 frames. ], batch size: 37, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:57:34,160 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4712, 4.9470, 5.4264, 4.7268, 5.0402, 4.8164, 5.4655, 5.0240], device='cuda:1'), covar=tensor([0.0211, 0.0353, 0.0244, 0.0267, 0.0334, 0.0282, 0.0198, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0240, 0.0260, 0.0238, 0.0237, 0.0235, 0.0213, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 08:57:41,032 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.802e+02 3.192e+02 3.860e+02 7.758e+02, threshold=6.385e+02, percent-clipped=2.0 2023-05-16 08:57:49,886 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186259.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:58:09,154 INFO [finetune.py:992] (1/2) Epoch 8, batch 650, loss[loss=0.1726, simple_loss=0.2646, pruned_loss=0.04033, over 11826.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2636, pruned_loss=0.04324, over 2283905.54 frames. ], batch size: 44, lr: 4.35e-03, grad_scale: 16.0 2023-05-16 08:58:33,268 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186320.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:58:34,832 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0110, 3.5388, 5.3082, 2.8858, 3.0470, 3.8315, 3.4986, 3.9769], device='cuda:1'), covar=tensor([0.0366, 0.1050, 0.0293, 0.1131, 0.1860, 0.1532, 0.1202, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0230, 0.0233, 0.0180, 0.0232, 0.0279, 0.0220, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:58:45,389 INFO [finetune.py:992] (1/2) Epoch 8, batch 700, loss[loss=0.1519, simple_loss=0.2403, pruned_loss=0.03169, over 12348.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2634, pruned_loss=0.04304, over 2298505.11 frames. ], batch size: 30, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 08:58:52,350 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.920e+02 2.942e+02 3.398e+02 4.313e+02 1.039e+03, threshold=6.795e+02, percent-clipped=2.0 2023-05-16 08:59:11,196 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9251, 3.4790, 5.2509, 2.8401, 2.9745, 3.8031, 3.4243, 3.8729], device='cuda:1'), covar=tensor([0.0331, 0.1064, 0.0348, 0.1156, 0.1844, 0.1419, 0.1225, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0231, 0.0233, 0.0180, 0.0232, 0.0279, 0.0220, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:59:21,703 INFO [finetune.py:992] (1/2) Epoch 8, batch 750, loss[loss=0.1711, simple_loss=0.2722, pruned_loss=0.03499, over 12192.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2631, pruned_loss=0.04271, over 2319436.74 frames. ], batch size: 35, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 08:59:41,999 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186415.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:59:43,400 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0770, 4.7478, 4.9455, 4.8721, 4.6484, 5.0329, 4.8111, 2.5078], device='cuda:1'), covar=tensor([0.0149, 0.0085, 0.0092, 0.0091, 0.0058, 0.0112, 0.0112, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0074, 0.0078, 0.0069, 0.0057, 0.0087, 0.0076, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 08:59:46,335 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186421.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 08:59:56,502 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7409, 2.8152, 4.3350, 4.6678, 3.0144, 2.5583, 2.9541, 2.0077], device='cuda:1'), covar=tensor([0.1441, 0.3153, 0.0557, 0.0401, 0.1207, 0.2339, 0.2698, 0.4232], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0375, 0.0264, 0.0289, 0.0256, 0.0290, 0.0365, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 08:59:57,631 INFO [finetune.py:992] (1/2) Epoch 8, batch 800, loss[loss=0.1846, simple_loss=0.2809, pruned_loss=0.04421, over 12351.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2622, pruned_loss=0.04239, over 2331206.19 frames. ], batch size: 36, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:00:05,529 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.737e+02 3.197e+02 3.965e+02 6.142e+02, threshold=6.393e+02, percent-clipped=0.0 2023-05-16 09:00:11,261 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186455.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 09:00:11,863 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186456.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:00:16,846 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186463.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:00:20,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 09:00:25,310 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186475.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:00:30,541 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186482.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:00:33,839 INFO [finetune.py:992] (1/2) Epoch 8, batch 850, loss[loss=0.1618, simple_loss=0.2438, pruned_loss=0.03989, over 12279.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.262, pruned_loss=0.04217, over 2345613.24 frames. ], batch size: 28, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:00:45,466 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186503.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 09:00:47,548 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3762, 5.1667, 5.3545, 5.3209, 4.9778, 5.0172, 4.7480, 5.3599], device='cuda:1'), covar=tensor([0.0827, 0.0694, 0.0759, 0.0747, 0.1996, 0.1407, 0.0642, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0656, 0.0558, 0.0590, 0.0795, 0.0690, 0.0519, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:00:51,594 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5840, 4.5073, 4.4281, 4.4212, 4.1149, 4.5357, 4.5655, 4.7772], device='cuda:1'), covar=tensor([0.0262, 0.0171, 0.0202, 0.0366, 0.0737, 0.0351, 0.0181, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0179, 0.0175, 0.0226, 0.0222, 0.0197, 0.0161, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 09:00:52,778 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186514.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:00:55,036 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186517.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:01:09,594 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186536.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:01:10,077 INFO [finetune.py:992] (1/2) Epoch 8, batch 900, loss[loss=0.1601, simple_loss=0.2494, pruned_loss=0.03536, over 11455.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2604, pruned_loss=0.04172, over 2355874.62 frames. ], batch size: 56, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:01:11,704 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186539.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:01:17,269 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 2.865e+02 3.396e+02 4.022e+02 7.338e+02, threshold=6.792e+02, percent-clipped=3.0 2023-05-16 09:01:31,867 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186567.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:01:39,774 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186578.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:01:45,270 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186586.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:01:45,741 INFO [finetune.py:992] (1/2) Epoch 8, batch 950, loss[loss=0.2133, simple_loss=0.2927, pruned_loss=0.06698, over 7734.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2608, pruned_loss=0.04179, over 2359690.06 frames. ], batch size: 98, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:01:54,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 09:01:56,184 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186600.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:02:03,546 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4713, 4.3236, 4.2583, 4.5661, 3.2457, 4.0344, 2.9613, 4.2120], device='cuda:1'), covar=tensor([0.1520, 0.0571, 0.0882, 0.0612, 0.1046, 0.0604, 0.1528, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0261, 0.0290, 0.0342, 0.0236, 0.0236, 0.0256, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:02:06,816 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:02:14,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 09:02:16,032 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186628.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:02:22,396 INFO [finetune.py:992] (1/2) Epoch 8, batch 1000, loss[loss=0.154, simple_loss=0.238, pruned_loss=0.03506, over 12124.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2601, pruned_loss=0.04187, over 2353948.66 frames. ], batch size: 30, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:02:29,249 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.785e+02 3.250e+02 3.932e+02 7.318e+02, threshold=6.500e+02, percent-clipped=1.0 2023-05-16 09:02:29,462 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186647.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:02:36,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 09:02:38,122 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 09:02:58,254 INFO [finetune.py:992] (1/2) Epoch 8, batch 1050, loss[loss=0.1679, simple_loss=0.2577, pruned_loss=0.0391, over 12210.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2604, pruned_loss=0.04225, over 2353599.62 frames. ], batch size: 35, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:03:13,917 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186709.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:03:21,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-05-16 09:03:33,889 INFO [finetune.py:992] (1/2) Epoch 8, batch 1100, loss[loss=0.1628, simple_loss=0.253, pruned_loss=0.03626, over 12284.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2608, pruned_loss=0.04242, over 2354490.16 frames. ], batch size: 33, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:03:41,714 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.786e+02 3.230e+02 3.834e+02 9.626e+02, threshold=6.461e+02, percent-clipped=1.0 2023-05-16 09:03:43,712 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-05-16 09:03:48,321 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186756.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:03:52,585 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0925, 6.0710, 5.8839, 5.3724, 5.2121, 5.9879, 5.6280, 5.4453], device='cuda:1'), covar=tensor([0.0708, 0.0891, 0.0615, 0.1580, 0.0648, 0.0694, 0.1635, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0523, 0.0491, 0.0603, 0.0392, 0.0674, 0.0731, 0.0547], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:03:58,201 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186770.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:04:03,121 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186777.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:04:10,229 INFO [finetune.py:992] (1/2) Epoch 8, batch 1150, loss[loss=0.1421, simple_loss=0.2268, pruned_loss=0.02867, over 12127.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.261, pruned_loss=0.04215, over 2366627.22 frames. ], batch size: 30, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:04:22,614 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186804.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:04:29,869 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186814.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:04:33,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-05-16 09:04:42,732 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186831.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:04:47,044 INFO [finetune.py:992] (1/2) Epoch 8, batch 1200, loss[loss=0.17, simple_loss=0.2626, pruned_loss=0.03873, over 12362.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2607, pruned_loss=0.042, over 2375810.05 frames. ], batch size: 38, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:04:49,392 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186840.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:04:50,143 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2971, 4.7397, 2.7925, 2.3302, 4.0227, 2.5898, 3.9534, 3.2621], device='cuda:1'), covar=tensor([0.0685, 0.0427, 0.1170, 0.1677, 0.0323, 0.1279, 0.0474, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0248, 0.0175, 0.0197, 0.0137, 0.0181, 0.0191, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:04:54,250 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 3.005e+02 3.396e+02 3.981e+02 6.798e+02, threshold=6.791e+02, percent-clipped=4.0 2023-05-16 09:05:04,881 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186862.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:05:05,629 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0548, 5.8469, 5.3488, 5.3734, 5.9619, 5.1881, 5.4873, 5.3990], device='cuda:1'), covar=tensor([0.1405, 0.0943, 0.1086, 0.1942, 0.0925, 0.2314, 0.1890, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0460, 0.0372, 0.0412, 0.0435, 0.0420, 0.0376, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 09:05:12,752 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186873.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:05:22,570 INFO [finetune.py:992] (1/2) Epoch 8, batch 1250, loss[loss=0.2122, simple_loss=0.2978, pruned_loss=0.06329, over 12039.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2613, pruned_loss=0.04236, over 2382232.67 frames. ], batch size: 42, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:05:23,501 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5346, 4.2686, 4.2892, 4.5761, 3.3289, 4.0941, 2.8875, 4.1585], device='cuda:1'), covar=tensor([0.1541, 0.0690, 0.0839, 0.0750, 0.1062, 0.0612, 0.1668, 0.1797], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0260, 0.0289, 0.0343, 0.0235, 0.0235, 0.0256, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:05:28,320 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186895.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:05:33,353 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186901.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 09:05:40,392 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186911.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:05:43,184 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186915.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:05:49,022 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186923.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:05:58,875 INFO [finetune.py:992] (1/2) Epoch 8, batch 1300, loss[loss=0.1861, simple_loss=0.2819, pruned_loss=0.04519, over 12134.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2615, pruned_loss=0.04247, over 2378390.72 frames. ], batch size: 36, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:06:00,434 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9037, 5.8997, 5.6523, 5.2199, 5.1645, 5.7999, 5.4214, 5.2559], device='cuda:1'), covar=tensor([0.0670, 0.0912, 0.0750, 0.1548, 0.0626, 0.0685, 0.1425, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0585, 0.0524, 0.0494, 0.0605, 0.0394, 0.0677, 0.0736, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:06:02,386 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186942.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:06:05,846 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 3.030e+02 3.437e+02 3.972e+02 7.661e+02, threshold=6.874e+02, percent-clipped=2.0 2023-05-16 09:06:17,452 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186963.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:06:22,545 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8394, 3.6536, 3.2143, 3.2227, 3.0118, 2.7365, 3.7445, 2.4353], device='cuda:1'), covar=tensor([0.0283, 0.0122, 0.0177, 0.0187, 0.0336, 0.0341, 0.0115, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0154, 0.0151, 0.0177, 0.0193, 0.0191, 0.0158, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:06:23,963 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186972.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:06:35,107 INFO [finetune.py:992] (1/2) Epoch 8, batch 1350, loss[loss=0.1573, simple_loss=0.2345, pruned_loss=0.04007, over 12280.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2605, pruned_loss=0.04205, over 2378854.81 frames. ], batch size: 28, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:06:51,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-05-16 09:06:52,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 09:07:02,182 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187025.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:07:08,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-16 09:07:10,596 INFO [finetune.py:992] (1/2) Epoch 8, batch 1400, loss[loss=0.1994, simple_loss=0.2971, pruned_loss=0.05085, over 10657.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2612, pruned_loss=0.04252, over 2367575.74 frames. ], batch size: 68, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:07:11,536 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187038.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:07:16,025 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7888, 3.6198, 3.2328, 3.1922, 3.0491, 2.7573, 3.6844, 2.3358], device='cuda:1'), covar=tensor([0.0304, 0.0137, 0.0200, 0.0217, 0.0327, 0.0341, 0.0132, 0.0445], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0154, 0.0152, 0.0176, 0.0193, 0.0190, 0.0159, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:07:17,856 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.675e+02 2.989e+02 3.804e+02 7.426e+02, threshold=5.978e+02, percent-clipped=2.0 2023-05-16 09:07:31,488 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187065.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:07:40,018 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187077.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:07:41,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 09:07:43,676 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8438, 4.7679, 4.6409, 4.7058, 4.3176, 4.8169, 4.8491, 5.0340], device='cuda:1'), covar=tensor([0.0223, 0.0159, 0.0195, 0.0323, 0.0777, 0.0333, 0.0155, 0.0185], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0184, 0.0179, 0.0230, 0.0227, 0.0201, 0.0165, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:07:46,660 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187086.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 09:07:47,186 INFO [finetune.py:992] (1/2) Epoch 8, batch 1450, loss[loss=0.1647, simple_loss=0.2438, pruned_loss=0.04278, over 12279.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2601, pruned_loss=0.0421, over 2375806.56 frames. ], batch size: 28, lr: 4.34e-03, grad_scale: 32.0 2023-05-16 09:07:52,938 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4212, 2.5531, 3.5885, 4.4604, 3.9134, 4.4453, 3.8136, 2.9457], device='cuda:1'), covar=tensor([0.0034, 0.0367, 0.0136, 0.0028, 0.0092, 0.0059, 0.0108, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0119, 0.0101, 0.0073, 0.0098, 0.0113, 0.0092, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:07:55,858 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187099.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:08:14,295 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187125.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:08:19,279 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187131.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:08:23,487 INFO [finetune.py:992] (1/2) Epoch 8, batch 1500, loss[loss=0.1708, simple_loss=0.2626, pruned_loss=0.03948, over 12276.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2602, pruned_loss=0.04197, over 2376547.69 frames. ], batch size: 33, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:08:30,862 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5010, 3.0680, 2.9246, 2.8978, 2.6703, 2.5364, 3.1354, 2.0442], device='cuda:1'), covar=tensor([0.0301, 0.0193, 0.0184, 0.0186, 0.0361, 0.0337, 0.0136, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0152, 0.0150, 0.0175, 0.0191, 0.0188, 0.0157, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 09:08:31,270 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 2.674e+02 3.261e+02 3.971e+02 7.100e+02, threshold=6.522e+02, percent-clipped=2.0 2023-05-16 09:08:49,127 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187173.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:08:53,392 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187179.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:08:56,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 09:08:58,975 INFO [finetune.py:992] (1/2) Epoch 8, batch 1550, loss[loss=0.1559, simple_loss=0.2431, pruned_loss=0.03433, over 11334.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2606, pruned_loss=0.04212, over 2383627.85 frames. ], batch size: 25, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:09:05,367 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187195.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:09:06,140 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187196.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 09:09:23,866 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187221.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:09:25,307 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187223.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:09:25,444 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0792, 4.0053, 4.1090, 4.4428, 2.9881, 3.7718, 2.5857, 3.9814], device='cuda:1'), covar=tensor([0.1973, 0.0793, 0.0925, 0.0643, 0.1342, 0.0751, 0.1982, 0.1494], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0263, 0.0292, 0.0347, 0.0238, 0.0238, 0.0259, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:09:34,982 INFO [finetune.py:992] (1/2) Epoch 8, batch 1600, loss[loss=0.1801, simple_loss=0.2748, pruned_loss=0.04272, over 12088.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2611, pruned_loss=0.04225, over 2384548.42 frames. ], batch size: 42, lr: 4.34e-03, grad_scale: 16.0 2023-05-16 09:09:38,663 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187242.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:09:39,285 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187243.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:09:42,830 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.917e+02 3.281e+02 3.882e+02 5.904e+02, threshold=6.562e+02, percent-clipped=0.0 2023-05-16 09:09:56,424 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187267.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:09:59,267 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187271.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:10:03,143 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.61 vs. limit=5.0 2023-05-16 09:10:11,259 INFO [finetune.py:992] (1/2) Epoch 8, batch 1650, loss[loss=0.1668, simple_loss=0.2546, pruned_loss=0.03954, over 12176.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.261, pruned_loss=0.04254, over 2382391.91 frames. ], batch size: 35, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:10:13,350 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187290.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:10:46,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 09:10:46,500 INFO [finetune.py:992] (1/2) Epoch 8, batch 1700, loss[loss=0.1825, simple_loss=0.2655, pruned_loss=0.04969, over 12191.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2616, pruned_loss=0.04282, over 2377510.22 frames. ], batch size: 35, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:10:55,036 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.737e+02 3.023e+02 3.663e+02 6.780e+02, threshold=6.046e+02, percent-clipped=1.0 2023-05-16 09:11:00,897 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3870, 4.2899, 4.2174, 4.2569, 3.9369, 4.4431, 4.3292, 4.5485], device='cuda:1'), covar=tensor([0.0270, 0.0197, 0.0205, 0.0379, 0.0705, 0.0409, 0.0217, 0.0210], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0184, 0.0180, 0.0231, 0.0228, 0.0201, 0.0165, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:11:05,210 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0768, 2.3589, 3.6631, 2.9700, 3.4377, 3.2281, 2.4904, 3.5644], device='cuda:1'), covar=tensor([0.0118, 0.0354, 0.0122, 0.0234, 0.0136, 0.0134, 0.0345, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0195, 0.0175, 0.0175, 0.0201, 0.0151, 0.0188, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:11:07,205 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187365.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:11:12,196 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187372.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:11:18,563 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187381.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 09:11:18,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 09:11:21,520 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6636, 5.4781, 5.5806, 5.6278, 5.2445, 5.2472, 5.0821, 5.5560], device='cuda:1'), covar=tensor([0.0586, 0.0570, 0.0725, 0.0663, 0.1953, 0.1286, 0.0537, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0660, 0.0569, 0.0598, 0.0808, 0.0701, 0.0522, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:11:22,786 INFO [finetune.py:992] (1/2) Epoch 8, batch 1750, loss[loss=0.1765, simple_loss=0.2657, pruned_loss=0.04365, over 12177.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.262, pruned_loss=0.04254, over 2372804.22 frames. ], batch size: 31, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:11:25,230 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8044, 3.0262, 4.7210, 5.0049, 2.9861, 2.7559, 3.1595, 2.2281], device='cuda:1'), covar=tensor([0.1460, 0.2843, 0.0441, 0.0346, 0.1254, 0.2148, 0.2549, 0.4074], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0373, 0.0263, 0.0292, 0.0256, 0.0289, 0.0362, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:11:27,928 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187394.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:11:42,396 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187413.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:11:57,059 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187433.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:11:59,653 INFO [finetune.py:992] (1/2) Epoch 8, batch 1800, loss[loss=0.193, simple_loss=0.2797, pruned_loss=0.05321, over 11613.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2609, pruned_loss=0.04192, over 2373732.18 frames. ], batch size: 48, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:12:07,497 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.838e+02 3.392e+02 3.885e+02 9.077e+02, threshold=6.783e+02, percent-clipped=5.0 2023-05-16 09:12:35,084 INFO [finetune.py:992] (1/2) Epoch 8, batch 1850, loss[loss=0.1573, simple_loss=0.2484, pruned_loss=0.03306, over 12144.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2619, pruned_loss=0.04258, over 2374177.77 frames. ], batch size: 34, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:12:41,471 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187496.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:12:53,458 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187512.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:12:58,436 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187519.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:01,236 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187523.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:11,115 INFO [finetune.py:992] (1/2) Epoch 8, batch 1900, loss[loss=0.2309, simple_loss=0.299, pruned_loss=0.08145, over 8296.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.262, pruned_loss=0.04245, over 2369301.75 frames. ], batch size: 99, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:13:16,090 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187544.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:18,776 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 2.883e+02 3.371e+02 4.186e+02 8.354e+02, threshold=6.742e+02, percent-clipped=2.0 2023-05-16 09:13:33,048 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187567.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:37,402 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187573.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:42,357 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187580.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:45,151 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187584.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:13:47,059 INFO [finetune.py:992] (1/2) Epoch 8, batch 1950, loss[loss=0.1871, simple_loss=0.2805, pruned_loss=0.04681, over 12176.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.262, pruned_loss=0.04248, over 2367237.04 frames. ], batch size: 35, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:14:07,078 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:14:22,425 INFO [finetune.py:992] (1/2) Epoch 8, batch 2000, loss[loss=0.197, simple_loss=0.2856, pruned_loss=0.05421, over 12362.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2621, pruned_loss=0.04251, over 2375674.24 frames. ], batch size: 38, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:14:30,923 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 2.688e+02 3.136e+02 3.958e+02 7.783e+02, threshold=6.272e+02, percent-clipped=1.0 2023-05-16 09:14:54,336 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187681.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 09:14:58,303 INFO [finetune.py:992] (1/2) Epoch 8, batch 2050, loss[loss=0.153, simple_loss=0.2326, pruned_loss=0.03671, over 11998.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.262, pruned_loss=0.04262, over 2372565.97 frames. ], batch size: 28, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:15:03,534 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187694.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:15:28,595 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187728.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:15:29,290 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187729.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:15:34,853 INFO [finetune.py:992] (1/2) Epoch 8, batch 2100, loss[loss=0.1881, simple_loss=0.2824, pruned_loss=0.04695, over 12149.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2606, pruned_loss=0.04181, over 2378208.07 frames. ], batch size: 36, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:15:38,288 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187742.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:15:42,441 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.818e+02 3.223e+02 3.937e+02 8.971e+02, threshold=6.446e+02, percent-clipped=3.0 2023-05-16 09:15:43,374 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0688, 2.3779, 2.3122, 2.2637, 2.1458, 2.0923, 2.3103, 1.8053], device='cuda:1'), covar=tensor([0.0282, 0.0157, 0.0176, 0.0155, 0.0308, 0.0233, 0.0159, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0156, 0.0154, 0.0180, 0.0198, 0.0194, 0.0161, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:16:05,379 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7071, 2.7816, 4.4554, 4.6869, 2.9164, 2.5621, 2.8310, 2.1509], device='cuda:1'), covar=tensor([0.1397, 0.2913, 0.0467, 0.0403, 0.1196, 0.2202, 0.2631, 0.3872], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0375, 0.0267, 0.0294, 0.0258, 0.0291, 0.0365, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:16:08,569 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7879, 3.0362, 5.3718, 2.4248, 2.6594, 4.0136, 3.2378, 4.0849], device='cuda:1'), covar=tensor([0.0426, 0.1399, 0.0210, 0.1523, 0.2213, 0.1452, 0.1485, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0233, 0.0239, 0.0182, 0.0237, 0.0285, 0.0223, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:16:09,727 INFO [finetune.py:992] (1/2) Epoch 8, batch 2150, loss[loss=0.1631, simple_loss=0.2443, pruned_loss=0.04096, over 12129.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.261, pruned_loss=0.04196, over 2372835.68 frames. ], batch size: 30, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:16:18,614 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6662, 2.6133, 3.9830, 4.1089, 2.7828, 2.5824, 2.6711, 2.1627], device='cuda:1'), covar=tensor([0.1375, 0.2744, 0.0536, 0.0447, 0.1250, 0.2205, 0.2660, 0.3839], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0375, 0.0266, 0.0294, 0.0258, 0.0291, 0.0365, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:16:39,360 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9350, 5.9988, 5.6835, 5.2067, 5.0926, 5.8300, 5.4745, 5.1989], device='cuda:1'), covar=tensor([0.0764, 0.0695, 0.0736, 0.1653, 0.0697, 0.0814, 0.1638, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0593, 0.0530, 0.0498, 0.0610, 0.0399, 0.0693, 0.0744, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:16:46,220 INFO [finetune.py:992] (1/2) Epoch 8, batch 2200, loss[loss=0.1932, simple_loss=0.2869, pruned_loss=0.04976, over 12151.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2614, pruned_loss=0.04182, over 2372827.83 frames. ], batch size: 38, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:16:53,917 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.731e+02 3.100e+02 3.737e+02 9.100e+02, threshold=6.201e+02, percent-clipped=2.0 2023-05-16 09:17:06,082 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3820, 4.7389, 2.9430, 2.6509, 4.0710, 2.6175, 3.9363, 3.2478], device='cuda:1'), covar=tensor([0.0625, 0.0479, 0.0945, 0.1402, 0.0256, 0.1235, 0.0497, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0250, 0.0174, 0.0196, 0.0139, 0.0180, 0.0192, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:17:07,452 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5589, 4.9411, 3.0227, 2.6360, 4.2667, 2.8356, 4.1237, 3.4561], device='cuda:1'), covar=tensor([0.0622, 0.0412, 0.1050, 0.1494, 0.0218, 0.1170, 0.0400, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0250, 0.0174, 0.0196, 0.0139, 0.0180, 0.0192, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:17:08,774 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187868.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:17:13,588 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187875.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:17:16,469 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187879.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:17:18,037 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9981, 4.9449, 4.7875, 4.8481, 4.4509, 5.0129, 4.9454, 5.2026], device='cuda:1'), covar=tensor([0.0261, 0.0137, 0.0201, 0.0295, 0.0758, 0.0301, 0.0146, 0.0170], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0187, 0.0183, 0.0235, 0.0231, 0.0205, 0.0167, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:17:18,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 09:17:22,262 INFO [finetune.py:992] (1/2) Epoch 8, batch 2250, loss[loss=0.1694, simple_loss=0.2643, pruned_loss=0.0372, over 12353.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2613, pruned_loss=0.04194, over 2379754.21 frames. ], batch size: 35, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:17:43,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6998, 4.6639, 4.5600, 4.1387, 4.3056, 4.6457, 4.3508, 4.1695], device='cuda:1'), covar=tensor([0.0738, 0.0894, 0.0668, 0.1454, 0.1721, 0.0820, 0.1491, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0530, 0.0498, 0.0607, 0.0399, 0.0692, 0.0742, 0.0556], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:17:50,371 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2449, 6.0267, 5.7219, 5.5158, 6.1197, 5.4628, 5.6078, 5.5759], device='cuda:1'), covar=tensor([0.1344, 0.0999, 0.0976, 0.1961, 0.0915, 0.2301, 0.1621, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0465, 0.0373, 0.0413, 0.0439, 0.0419, 0.0380, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:17:57,923 INFO [finetune.py:992] (1/2) Epoch 8, batch 2300, loss[loss=0.1806, simple_loss=0.2726, pruned_loss=0.0443, over 12352.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2614, pruned_loss=0.04208, over 2382634.49 frames. ], batch size: 35, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:18:06,256 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.922e+02 3.527e+02 4.326e+02 1.992e+03, threshold=7.054e+02, percent-clipped=7.0 2023-05-16 09:18:33,750 INFO [finetune.py:992] (1/2) Epoch 8, batch 2350, loss[loss=0.182, simple_loss=0.2739, pruned_loss=0.04506, over 12030.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2616, pruned_loss=0.04205, over 2387162.59 frames. ], batch size: 40, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:19:06,925 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188028.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:19:12,934 INFO [finetune.py:992] (1/2) Epoch 8, batch 2400, loss[loss=0.1971, simple_loss=0.2799, pruned_loss=0.05714, over 11860.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2616, pruned_loss=0.04218, over 2376695.44 frames. ], batch size: 44, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:19:20,877 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.762e+02 3.244e+02 3.839e+02 6.809e+02, threshold=6.487e+02, percent-clipped=0.0 2023-05-16 09:19:40,449 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188076.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:19:48,622 INFO [finetune.py:992] (1/2) Epoch 8, batch 2450, loss[loss=0.1835, simple_loss=0.2724, pruned_loss=0.04733, over 11331.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2609, pruned_loss=0.04214, over 2377436.22 frames. ], batch size: 55, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:20:02,485 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2206, 4.5725, 2.7218, 2.0636, 4.0423, 1.9770, 3.8328, 2.7897], device='cuda:1'), covar=tensor([0.0739, 0.0575, 0.1253, 0.2217, 0.0336, 0.1995, 0.0570, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0250, 0.0172, 0.0195, 0.0138, 0.0180, 0.0191, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:20:05,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 09:20:24,104 INFO [finetune.py:992] (1/2) Epoch 8, batch 2500, loss[loss=0.1661, simple_loss=0.2602, pruned_loss=0.03605, over 10457.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2608, pruned_loss=0.04215, over 2377189.16 frames. ], batch size: 68, lr: 4.33e-03, grad_scale: 16.0 2023-05-16 09:20:32,505 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 2.770e+02 3.184e+02 4.099e+02 5.447e+02, threshold=6.368e+02, percent-clipped=0.0 2023-05-16 09:20:46,845 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188168.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:20:51,850 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188175.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:20:54,758 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188179.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:21:00,272 INFO [finetune.py:992] (1/2) Epoch 8, batch 2550, loss[loss=0.1659, simple_loss=0.2552, pruned_loss=0.03828, over 12040.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2607, pruned_loss=0.04206, over 2368741.10 frames. ], batch size: 31, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:21:06,232 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0549, 5.8257, 5.5326, 5.3336, 5.9206, 5.2299, 5.4252, 5.4332], device='cuda:1'), covar=tensor([0.1319, 0.0927, 0.0859, 0.1871, 0.0898, 0.1937, 0.1884, 0.1057], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0466, 0.0372, 0.0414, 0.0440, 0.0420, 0.0381, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:21:07,707 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5113, 5.3859, 5.4717, 5.5096, 5.1544, 5.2016, 4.9366, 5.3843], device='cuda:1'), covar=tensor([0.0645, 0.0523, 0.0679, 0.0551, 0.1632, 0.1109, 0.0469, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0660, 0.0568, 0.0595, 0.0805, 0.0695, 0.0519, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:21:21,642 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188216.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:21:26,618 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188223.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:21:30,216 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188227.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:21:37,127 INFO [finetune.py:992] (1/2) Epoch 8, batch 2600, loss[loss=0.1785, simple_loss=0.2772, pruned_loss=0.03995, over 11713.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2595, pruned_loss=0.04147, over 2377713.64 frames. ], batch size: 48, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:21:44,176 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6645, 4.4152, 4.3973, 4.5332, 4.3588, 4.6229, 4.4288, 2.7806], device='cuda:1'), covar=tensor([0.0103, 0.0077, 0.0112, 0.0081, 0.0064, 0.0099, 0.0121, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0078, 0.0070, 0.0057, 0.0087, 0.0076, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:21:44,683 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 2.791e+02 3.137e+02 3.810e+02 9.850e+02, threshold=6.275e+02, percent-clipped=2.0 2023-05-16 09:21:54,992 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3773, 2.2252, 3.0609, 4.3424, 2.2530, 4.2751, 4.3857, 4.5243], device='cuda:1'), covar=tensor([0.0133, 0.1415, 0.0555, 0.0130, 0.1444, 0.0238, 0.0158, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0205, 0.0182, 0.0113, 0.0192, 0.0178, 0.0174, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:22:09,901 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188283.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:22:12,633 INFO [finetune.py:992] (1/2) Epoch 8, batch 2650, loss[loss=0.1698, simple_loss=0.2626, pruned_loss=0.03857, over 11275.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2594, pruned_loss=0.04145, over 2378325.65 frames. ], batch size: 55, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:22:33,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 09:22:48,918 INFO [finetune.py:992] (1/2) Epoch 8, batch 2700, loss[loss=0.1517, simple_loss=0.2401, pruned_loss=0.03158, over 12342.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2599, pruned_loss=0.04143, over 2376373.19 frames. ], batch size: 31, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:22:54,270 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188344.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:22:56,968 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.755e+02 3.326e+02 3.869e+02 7.495e+02, threshold=6.653e+02, percent-clipped=1.0 2023-05-16 09:23:23,971 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4483, 4.9775, 5.3960, 4.7391, 4.9920, 4.7818, 5.4491, 5.0508], device='cuda:1'), covar=tensor([0.0245, 0.0352, 0.0270, 0.0266, 0.0345, 0.0307, 0.0208, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0250, 0.0275, 0.0249, 0.0249, 0.0247, 0.0224, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:23:25,231 INFO [finetune.py:992] (1/2) Epoch 8, batch 2750, loss[loss=0.1777, simple_loss=0.2642, pruned_loss=0.04564, over 12308.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2604, pruned_loss=0.04153, over 2377828.23 frames. ], batch size: 34, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:24:01,864 INFO [finetune.py:992] (1/2) Epoch 8, batch 2800, loss[loss=0.194, simple_loss=0.296, pruned_loss=0.04597, over 11655.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2592, pruned_loss=0.04102, over 2383352.05 frames. ], batch size: 48, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:24:09,722 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.626e+02 3.087e+02 3.593e+02 6.799e+02, threshold=6.173e+02, percent-clipped=1.0 2023-05-16 09:24:13,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 09:24:33,552 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188481.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:24:37,599 INFO [finetune.py:992] (1/2) Epoch 8, batch 2850, loss[loss=0.1741, simple_loss=0.2642, pruned_loss=0.04202, over 11779.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2591, pruned_loss=0.04082, over 2382114.84 frames. ], batch size: 44, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:24:52,699 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5817, 5.1189, 5.5645, 4.8507, 5.1707, 4.9151, 5.5885, 5.2356], device='cuda:1'), covar=tensor([0.0217, 0.0320, 0.0190, 0.0247, 0.0336, 0.0296, 0.0195, 0.0213], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0249, 0.0275, 0.0249, 0.0248, 0.0247, 0.0224, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:25:13,832 INFO [finetune.py:992] (1/2) Epoch 8, batch 2900, loss[loss=0.152, simple_loss=0.2403, pruned_loss=0.0319, over 12110.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.258, pruned_loss=0.04054, over 2391732.85 frames. ], batch size: 33, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:25:17,536 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188542.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:25:21,490 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.114e+02 2.935e+02 3.309e+02 3.834e+02 6.179e+02, threshold=6.618e+02, percent-clipped=1.0 2023-05-16 09:25:35,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 09:25:50,036 INFO [finetune.py:992] (1/2) Epoch 8, batch 2950, loss[loss=0.1977, simple_loss=0.2859, pruned_loss=0.05474, over 12108.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2579, pruned_loss=0.04051, over 2386648.04 frames. ], batch size: 32, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:25:53,105 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5019, 3.5806, 3.2699, 3.2227, 2.9159, 2.7445, 3.5825, 2.4166], device='cuda:1'), covar=tensor([0.0384, 0.0111, 0.0183, 0.0188, 0.0433, 0.0341, 0.0125, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0162, 0.0158, 0.0184, 0.0204, 0.0199, 0.0167, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:26:07,697 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0610, 5.0343, 4.8570, 4.9398, 4.5192, 5.1191, 5.0192, 5.3587], device='cuda:1'), covar=tensor([0.0243, 0.0165, 0.0205, 0.0320, 0.0842, 0.0340, 0.0180, 0.0162], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0193, 0.0188, 0.0241, 0.0238, 0.0212, 0.0170, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:26:16,145 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1147, 2.4048, 3.6566, 3.0979, 3.5002, 3.1981, 2.5145, 3.5086], device='cuda:1'), covar=tensor([0.0122, 0.0361, 0.0152, 0.0244, 0.0148, 0.0166, 0.0379, 0.0145], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0198, 0.0178, 0.0178, 0.0204, 0.0153, 0.0191, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:26:24,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5007, 4.2456, 4.2052, 4.5831, 3.1666, 4.0473, 2.6941, 4.2571], device='cuda:1'), covar=tensor([0.1402, 0.0608, 0.0905, 0.0635, 0.1045, 0.0558, 0.1742, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0290, 0.0345, 0.0234, 0.0235, 0.0254, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:26:25,893 INFO [finetune.py:992] (1/2) Epoch 8, batch 3000, loss[loss=0.1643, simple_loss=0.2558, pruned_loss=0.03645, over 12139.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2577, pruned_loss=0.04051, over 2387146.59 frames. ], batch size: 34, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:26:25,893 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 09:26:44,282 INFO [finetune.py:1026] (1/2) Epoch 8, validation: loss=0.3187, simple_loss=0.3968, pruned_loss=0.1203, over 1020973.00 frames. 2023-05-16 09:26:44,283 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 09:26:45,896 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188639.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:26:48,239 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9378, 2.9893, 4.6497, 4.9622, 3.0378, 2.8023, 3.1712, 2.2196], device='cuda:1'), covar=tensor([0.1317, 0.2789, 0.0408, 0.0337, 0.1167, 0.2000, 0.2400, 0.3701], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0377, 0.0268, 0.0294, 0.0260, 0.0291, 0.0364, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:26:52,302 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.728e+02 3.186e+02 3.763e+02 6.061e+02, threshold=6.372e+02, percent-clipped=0.0 2023-05-16 09:26:54,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 09:27:20,034 INFO [finetune.py:992] (1/2) Epoch 8, batch 3050, loss[loss=0.1819, simple_loss=0.2729, pruned_loss=0.04548, over 11206.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2569, pruned_loss=0.04033, over 2394984.41 frames. ], batch size: 55, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:27:56,175 INFO [finetune.py:992] (1/2) Epoch 8, batch 3100, loss[loss=0.1314, simple_loss=0.2132, pruned_loss=0.02484, over 11753.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.256, pruned_loss=0.0403, over 2388816.45 frames. ], batch size: 26, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:28:03,968 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.878e+02 2.640e+02 3.122e+02 3.982e+02 7.416e+02, threshold=6.244e+02, percent-clipped=3.0 2023-05-16 09:28:31,447 INFO [finetune.py:992] (1/2) Epoch 8, batch 3150, loss[loss=0.1709, simple_loss=0.2658, pruned_loss=0.03798, over 12148.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2566, pruned_loss=0.04071, over 2374889.42 frames. ], batch size: 34, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:28:31,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7852, 2.5582, 4.7161, 5.1171, 3.3768, 2.6967, 3.0135, 1.9919], device='cuda:1'), covar=tensor([0.1484, 0.3708, 0.0400, 0.0316, 0.0955, 0.2236, 0.2856, 0.5064], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0376, 0.0267, 0.0293, 0.0259, 0.0290, 0.0362, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:29:08,901 INFO [finetune.py:992] (1/2) Epoch 8, batch 3200, loss[loss=0.1877, simple_loss=0.2676, pruned_loss=0.05392, over 7966.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2569, pruned_loss=0.04067, over 2374399.05 frames. ], batch size: 98, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:29:08,987 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188837.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:29:16,725 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 2.669e+02 3.069e+02 3.720e+02 5.873e+02, threshold=6.139e+02, percent-clipped=0.0 2023-05-16 09:29:33,873 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9779, 3.6534, 5.3249, 2.9477, 3.1279, 4.0551, 3.4738, 4.0287], device='cuda:1'), covar=tensor([0.0414, 0.0968, 0.0348, 0.1086, 0.1707, 0.1334, 0.1202, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0231, 0.0238, 0.0180, 0.0235, 0.0283, 0.0220, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:29:44,258 INFO [finetune.py:992] (1/2) Epoch 8, batch 3250, loss[loss=0.1798, simple_loss=0.2747, pruned_loss=0.04245, over 12286.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2579, pruned_loss=0.04126, over 2368009.87 frames. ], batch size: 37, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:30:10,308 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2947, 5.1293, 5.2217, 5.2638, 4.7385, 4.7772, 4.8044, 5.1269], device='cuda:1'), covar=tensor([0.0879, 0.0841, 0.0924, 0.0705, 0.2575, 0.1955, 0.0556, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0663, 0.0573, 0.0604, 0.0820, 0.0704, 0.0528, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:30:13,756 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188929.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:30:15,845 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188931.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:30:19,937 INFO [finetune.py:992] (1/2) Epoch 8, batch 3300, loss[loss=0.1594, simple_loss=0.2419, pruned_loss=0.03841, over 12331.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2591, pruned_loss=0.04189, over 2364734.13 frames. ], batch size: 30, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:30:21,572 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188939.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:30:27,771 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.793e+02 3.385e+02 3.964e+02 1.560e+03, threshold=6.770e+02, percent-clipped=7.0 2023-05-16 09:30:28,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-16 09:30:40,634 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5078, 5.3292, 5.4417, 5.4587, 5.0651, 5.1433, 4.9326, 5.3914], device='cuda:1'), covar=tensor([0.0653, 0.0550, 0.0716, 0.0548, 0.1657, 0.1154, 0.0509, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0664, 0.0573, 0.0603, 0.0819, 0.0703, 0.0527, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:30:44,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-16 09:30:54,644 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8391, 4.7166, 4.6724, 4.6107, 4.2611, 4.7748, 4.8334, 4.9998], device='cuda:1'), covar=tensor([0.0204, 0.0149, 0.0158, 0.0340, 0.0754, 0.0351, 0.0140, 0.0164], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0191, 0.0186, 0.0240, 0.0236, 0.0210, 0.0169, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:30:55,910 INFO [finetune.py:992] (1/2) Epoch 8, batch 3350, loss[loss=0.1754, simple_loss=0.2617, pruned_loss=0.04458, over 12118.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2597, pruned_loss=0.04199, over 2366291.70 frames. ], batch size: 38, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:30:55,977 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188987.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:30:58,341 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188990.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:30:59,821 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188992.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:31:19,816 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1650, 5.0281, 4.9773, 4.8840, 4.5538, 5.1491, 5.1566, 5.2965], device='cuda:1'), covar=tensor([0.0143, 0.0133, 0.0140, 0.0284, 0.0662, 0.0213, 0.0095, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0191, 0.0186, 0.0240, 0.0236, 0.0210, 0.0169, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:31:26,941 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189030.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:31:31,707 INFO [finetune.py:992] (1/2) Epoch 8, batch 3400, loss[loss=0.1686, simple_loss=0.255, pruned_loss=0.0411, over 12297.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2592, pruned_loss=0.04161, over 2375469.06 frames. ], batch size: 37, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:31:35,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-16 09:31:39,753 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.760e+02 3.345e+02 4.036e+02 1.056e+03, threshold=6.689e+02, percent-clipped=2.0 2023-05-16 09:31:50,793 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2024, 4.2100, 2.6959, 2.5363, 3.6571, 2.2677, 3.7115, 2.9492], device='cuda:1'), covar=tensor([0.0677, 0.0586, 0.1046, 0.1353, 0.0282, 0.1450, 0.0503, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0253, 0.0174, 0.0196, 0.0140, 0.0182, 0.0193, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:31:51,453 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189064.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:32:02,738 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189079.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:32:08,156 INFO [finetune.py:992] (1/2) Epoch 8, batch 3450, loss[loss=0.1833, simple_loss=0.2727, pruned_loss=0.04698, over 12095.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2593, pruned_loss=0.04182, over 2368027.73 frames. ], batch size: 32, lr: 4.32e-03, grad_scale: 16.0 2023-05-16 09:32:11,345 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189091.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:32:36,040 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189125.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:32:44,434 INFO [finetune.py:992] (1/2) Epoch 8, batch 3500, loss[loss=0.1856, simple_loss=0.2751, pruned_loss=0.04807, over 12105.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2598, pruned_loss=0.04178, over 2375268.22 frames. ], batch size: 32, lr: 4.31e-03, grad_scale: 32.0 2023-05-16 09:32:44,536 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189137.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:32:46,028 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3325, 3.3419, 3.1285, 3.0887, 2.7990, 2.7217, 3.4392, 1.9773], device='cuda:1'), covar=tensor([0.0395, 0.0142, 0.0179, 0.0188, 0.0348, 0.0342, 0.0130, 0.0517], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0159, 0.0156, 0.0180, 0.0200, 0.0194, 0.0163, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:32:46,716 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189140.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:32:52,136 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 2.761e+02 3.216e+02 3.789e+02 1.190e+03, threshold=6.432e+02, percent-clipped=2.0 2023-05-16 09:32:55,233 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7081, 3.1159, 5.0683, 2.5747, 2.8431, 3.8139, 3.1710, 3.8040], device='cuda:1'), covar=tensor([0.0464, 0.1214, 0.0279, 0.1203, 0.1860, 0.1431, 0.1330, 0.1136], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0230, 0.0236, 0.0179, 0.0233, 0.0282, 0.0219, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:33:18,284 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189185.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:33:19,663 INFO [finetune.py:992] (1/2) Epoch 8, batch 3550, loss[loss=0.1522, simple_loss=0.229, pruned_loss=0.03766, over 12189.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2599, pruned_loss=0.04187, over 2378422.67 frames. ], batch size: 29, lr: 4.31e-03, grad_scale: 32.0 2023-05-16 09:33:23,538 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2495, 2.2758, 3.1014, 4.1792, 2.1254, 4.2895, 4.2125, 4.4252], device='cuda:1'), covar=tensor([0.0126, 0.1249, 0.0464, 0.0121, 0.1392, 0.0218, 0.0145, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0199, 0.0178, 0.0111, 0.0187, 0.0173, 0.0169, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:33:56,131 INFO [finetune.py:992] (1/2) Epoch 8, batch 3600, loss[loss=0.1765, simple_loss=0.2649, pruned_loss=0.04404, over 12150.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2592, pruned_loss=0.04158, over 2375114.90 frames. ], batch size: 39, lr: 4.31e-03, grad_scale: 32.0 2023-05-16 09:34:03,902 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.606e+02 3.237e+02 3.912e+02 7.810e+02, threshold=6.473e+02, percent-clipped=1.0 2023-05-16 09:34:04,095 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189248.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:34:30,948 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189285.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:34:32,256 INFO [finetune.py:992] (1/2) Epoch 8, batch 3650, loss[loss=0.191, simple_loss=0.2809, pruned_loss=0.05055, over 11246.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2596, pruned_loss=0.0419, over 2380710.38 frames. ], batch size: 55, lr: 4.31e-03, grad_scale: 32.0 2023-05-16 09:34:32,333 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189287.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:34:48,148 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189309.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:35:07,878 INFO [finetune.py:992] (1/2) Epoch 8, batch 3700, loss[loss=0.1652, simple_loss=0.2637, pruned_loss=0.03336, over 12360.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.259, pruned_loss=0.04127, over 2386377.81 frames. ], batch size: 36, lr: 4.31e-03, grad_scale: 32.0 2023-05-16 09:35:14,712 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0301, 3.4004, 5.2801, 2.8436, 2.9817, 4.0308, 3.3440, 4.0433], device='cuda:1'), covar=tensor([0.0331, 0.1094, 0.0315, 0.1139, 0.1807, 0.1203, 0.1276, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0231, 0.0238, 0.0180, 0.0235, 0.0284, 0.0221, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:35:15,369 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189347.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:35:15,885 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.651e+02 3.048e+02 3.558e+02 6.280e+02, threshold=6.096e+02, percent-clipped=0.0 2023-05-16 09:35:22,895 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5369, 5.0229, 5.5019, 4.7994, 5.1013, 4.9059, 5.5638, 5.0763], device='cuda:1'), covar=tensor([0.0216, 0.0307, 0.0220, 0.0233, 0.0308, 0.0301, 0.0164, 0.0243], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0249, 0.0275, 0.0249, 0.0249, 0.0248, 0.0226, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:35:23,503 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3402, 6.1904, 5.7389, 5.6821, 6.2416, 5.4761, 5.8103, 5.7462], device='cuda:1'), covar=tensor([0.1533, 0.0883, 0.1089, 0.1811, 0.0943, 0.2157, 0.1478, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0466, 0.0375, 0.0414, 0.0440, 0.0420, 0.0378, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:35:43,114 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189386.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:35:43,754 INFO [finetune.py:992] (1/2) Epoch 8, batch 3750, loss[loss=0.174, simple_loss=0.2636, pruned_loss=0.04219, over 12192.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2592, pruned_loss=0.04127, over 2385476.81 frames. ], batch size: 31, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:35:59,334 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189408.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:36:01,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 09:36:08,264 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:36:16,931 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9434, 4.2767, 3.7756, 4.6218, 4.2480, 2.6955, 3.8950, 2.9492], device='cuda:1'), covar=tensor([0.0927, 0.0903, 0.1562, 0.0488, 0.1092, 0.1753, 0.1075, 0.3210], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0375, 0.0352, 0.0280, 0.0359, 0.0266, 0.0336, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:36:18,857 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189435.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:36:20,209 INFO [finetune.py:992] (1/2) Epoch 8, batch 3800, loss[loss=0.1537, simple_loss=0.2486, pruned_loss=0.02937, over 12283.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2585, pruned_loss=0.04085, over 2384647.17 frames. ], batch size: 37, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:36:24,629 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1782, 4.7971, 5.0077, 5.0267, 4.8135, 5.0266, 4.9686, 2.6077], device='cuda:1'), covar=tensor([0.0083, 0.0062, 0.0066, 0.0053, 0.0045, 0.0073, 0.0070, 0.0757], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0076, 0.0079, 0.0072, 0.0058, 0.0089, 0.0078, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:36:28,645 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.003e+02 2.751e+02 3.207e+02 3.788e+02 9.973e+02, threshold=6.415e+02, percent-clipped=4.0 2023-05-16 09:36:43,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-16 09:36:52,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8949, 3.4099, 5.2190, 2.6350, 2.8099, 3.8322, 3.3039, 3.8599], device='cuda:1'), covar=tensor([0.0394, 0.1185, 0.0315, 0.1206, 0.2132, 0.1650, 0.1441, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0231, 0.0238, 0.0180, 0.0235, 0.0284, 0.0222, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:36:55,376 INFO [finetune.py:992] (1/2) Epoch 8, batch 3850, loss[loss=0.1976, simple_loss=0.2795, pruned_loss=0.05785, over 12052.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2592, pruned_loss=0.04115, over 2385149.04 frames. ], batch size: 40, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:37:05,527 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189501.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:37:31,830 INFO [finetune.py:992] (1/2) Epoch 8, batch 3900, loss[loss=0.1688, simple_loss=0.2475, pruned_loss=0.04504, over 12186.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2589, pruned_loss=0.04123, over 2385746.28 frames. ], batch size: 31, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:37:35,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-05-16 09:37:40,307 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.778e+02 3.186e+02 3.785e+02 9.683e+02, threshold=6.372e+02, percent-clipped=2.0 2023-05-16 09:37:49,717 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189562.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:02,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.87 vs. limit=5.0 2023-05-16 09:38:06,498 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189585.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:07,777 INFO [finetune.py:992] (1/2) Epoch 8, batch 3950, loss[loss=0.176, simple_loss=0.2471, pruned_loss=0.05244, over 12018.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2592, pruned_loss=0.04151, over 2386464.77 frames. ], batch size: 28, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:38:07,955 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189587.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:15,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3636, 5.2288, 5.2682, 5.3425, 4.9747, 4.9786, 4.7902, 5.2474], device='cuda:1'), covar=tensor([0.0687, 0.0602, 0.0796, 0.0610, 0.1772, 0.1403, 0.0552, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0666, 0.0578, 0.0600, 0.0820, 0.0708, 0.0529, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:38:20,364 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189604.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:28,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 09:38:40,564 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189633.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:41,902 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189635.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:43,253 INFO [finetune.py:992] (1/2) Epoch 8, batch 4000, loss[loss=0.1593, simple_loss=0.2507, pruned_loss=0.03394, over 12364.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2588, pruned_loss=0.0416, over 2383838.93 frames. ], batch size: 35, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:38:44,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-05-16 09:38:45,542 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189640.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:38:51,550 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.907e+02 3.282e+02 4.224e+02 1.207e+03, threshold=6.563e+02, percent-clipped=1.0 2023-05-16 09:38:55,981 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3554, 4.2412, 4.2618, 4.4868, 2.8529, 3.9856, 2.8018, 4.2299], device='cuda:1'), covar=tensor([0.1534, 0.0672, 0.0931, 0.0726, 0.1252, 0.0612, 0.1711, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0263, 0.0296, 0.0352, 0.0239, 0.0238, 0.0257, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:39:18,620 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189686.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:19,187 INFO [finetune.py:992] (1/2) Epoch 8, batch 4050, loss[loss=0.1777, simple_loss=0.2623, pruned_loss=0.04653, over 12286.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2592, pruned_loss=0.04179, over 2388251.27 frames. ], batch size: 37, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:39:29,364 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189701.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:30,703 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189703.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:31,569 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189704.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:43,419 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189720.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:48,906 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189728.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:53,087 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189734.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:53,841 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189735.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:39:55,218 INFO [finetune.py:992] (1/2) Epoch 8, batch 4100, loss[loss=0.1656, simple_loss=0.2625, pruned_loss=0.03438, over 10507.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2605, pruned_loss=0.04226, over 2366510.41 frames. ], batch size: 68, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:40:03,529 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.851e+02 2.768e+02 3.461e+02 4.135e+02 7.228e+02, threshold=6.921e+02, percent-clipped=1.0 2023-05-16 09:40:15,255 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189765.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:40:17,264 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189768.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:40:27,848 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189783.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:40:30,599 INFO [finetune.py:992] (1/2) Epoch 8, batch 4150, loss[loss=0.1722, simple_loss=0.2608, pruned_loss=0.04178, over 12003.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2606, pruned_loss=0.0419, over 2370781.86 frames. ], batch size: 28, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:40:32,158 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189789.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:40:53,743 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1933, 5.1182, 4.9847, 5.0257, 4.7200, 5.1718, 5.1457, 5.4023], device='cuda:1'), covar=tensor([0.0150, 0.0125, 0.0164, 0.0279, 0.0654, 0.0219, 0.0122, 0.0138], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0192, 0.0187, 0.0242, 0.0237, 0.0210, 0.0171, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 09:41:07,506 INFO [finetune.py:992] (1/2) Epoch 8, batch 4200, loss[loss=0.1505, simple_loss=0.2406, pruned_loss=0.03025, over 12045.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2594, pruned_loss=0.0416, over 2371575.72 frames. ], batch size: 31, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:41:16,039 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.753e+02 3.083e+02 3.674e+02 8.354e+02, threshold=6.165e+02, percent-clipped=1.0 2023-05-16 09:41:22,632 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189857.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:41:31,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 09:41:43,591 INFO [finetune.py:992] (1/2) Epoch 8, batch 4250, loss[loss=0.211, simple_loss=0.2832, pruned_loss=0.06946, over 7780.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.26, pruned_loss=0.04177, over 2372111.82 frames. ], batch size: 97, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:41:55,930 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189904.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:42:19,166 INFO [finetune.py:992] (1/2) Epoch 8, batch 4300, loss[loss=0.1631, simple_loss=0.2485, pruned_loss=0.03889, over 12273.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.259, pruned_loss=0.04155, over 2365045.02 frames. ], batch size: 28, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:42:27,505 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.709e+02 3.186e+02 3.885e+02 8.568e+02, threshold=6.372e+02, percent-clipped=2.0 2023-05-16 09:42:29,793 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189952.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:42:55,450 INFO [finetune.py:992] (1/2) Epoch 8, batch 4350, loss[loss=0.1595, simple_loss=0.2473, pruned_loss=0.03592, over 12342.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2601, pruned_loss=0.04223, over 2358716.30 frames. ], batch size: 31, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:43:02,299 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189996.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:43:10,605 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190003.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:43:34,432 INFO [finetune.py:992] (1/2) Epoch 8, batch 4400, loss[loss=0.1752, simple_loss=0.253, pruned_loss=0.04872, over 12284.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.26, pruned_loss=0.04225, over 2360639.77 frames. ], batch size: 28, lr: 4.31e-03, grad_scale: 16.0 2023-05-16 09:43:42,848 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.004e+02 2.848e+02 3.319e+02 3.981e+02 8.022e+02, threshold=6.638e+02, percent-clipped=3.0 2023-05-16 09:43:44,379 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190051.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:43:50,657 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190060.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:44:03,588 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2521, 2.5772, 3.8544, 3.0810, 3.5752, 3.2288, 2.6799, 3.6436], device='cuda:1'), covar=tensor([0.0108, 0.0343, 0.0134, 0.0225, 0.0152, 0.0170, 0.0309, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0201, 0.0181, 0.0178, 0.0206, 0.0156, 0.0193, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:44:07,607 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190084.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:44:09,649 INFO [finetune.py:992] (1/2) Epoch 8, batch 4450, loss[loss=0.144, simple_loss=0.2327, pruned_loss=0.02763, over 12096.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.26, pruned_loss=0.04186, over 2363438.05 frames. ], batch size: 32, lr: 4.30e-03, grad_scale: 16.0 2023-05-16 09:44:30,319 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8624, 4.4901, 4.8415, 4.2850, 4.5293, 4.3662, 4.8606, 4.4582], device='cuda:1'), covar=tensor([0.0277, 0.0382, 0.0286, 0.0255, 0.0378, 0.0319, 0.0249, 0.0528], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0253, 0.0278, 0.0251, 0.0251, 0.0250, 0.0228, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:44:36,047 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1378, 4.7871, 5.0753, 5.0222, 4.9179, 4.9900, 4.9106, 2.8204], device='cuda:1'), covar=tensor([0.0075, 0.0066, 0.0061, 0.0054, 0.0043, 0.0087, 0.0082, 0.0665], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0076, 0.0080, 0.0072, 0.0058, 0.0089, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:44:45,602 INFO [finetune.py:992] (1/2) Epoch 8, batch 4500, loss[loss=0.1571, simple_loss=0.2354, pruned_loss=0.03936, over 12136.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2596, pruned_loss=0.0418, over 2363282.08 frames. ], batch size: 30, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:44:47,221 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190139.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 09:44:55,472 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.655e+02 3.257e+02 3.664e+02 9.318e+02, threshold=6.515e+02, percent-clipped=2.0 2023-05-16 09:44:56,435 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190151.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:45:00,661 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190157.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:45:21,861 INFO [finetune.py:992] (1/2) Epoch 8, batch 4550, loss[loss=0.1521, simple_loss=0.2445, pruned_loss=0.0298, over 12233.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2602, pruned_loss=0.04187, over 2362830.60 frames. ], batch size: 32, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:45:26,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-05-16 09:45:31,667 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190200.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 09:45:35,095 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190205.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:45:40,119 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190212.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:45:42,156 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8078, 2.6204, 3.7392, 4.7857, 4.1858, 4.6150, 4.1174, 3.1873], device='cuda:1'), covar=tensor([0.0029, 0.0386, 0.0144, 0.0031, 0.0082, 0.0069, 0.0103, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0119, 0.0101, 0.0074, 0.0099, 0.0114, 0.0091, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:45:42,216 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7148, 4.1576, 4.3004, 4.5530, 3.6956, 4.1571, 3.0507, 4.4550], device='cuda:1'), covar=tensor([0.1345, 0.0705, 0.0911, 0.0656, 0.0848, 0.0549, 0.1533, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0263, 0.0294, 0.0349, 0.0237, 0.0238, 0.0258, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:45:48,583 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2918, 4.8591, 5.2756, 4.5116, 4.9144, 4.6557, 5.3107, 5.0164], device='cuda:1'), covar=tensor([0.0251, 0.0372, 0.0226, 0.0298, 0.0403, 0.0332, 0.0209, 0.0300], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0253, 0.0277, 0.0251, 0.0250, 0.0250, 0.0227, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:45:57,595 INFO [finetune.py:992] (1/2) Epoch 8, batch 4600, loss[loss=0.164, simple_loss=0.2449, pruned_loss=0.04148, over 12179.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2603, pruned_loss=0.0418, over 2364150.95 frames. ], batch size: 31, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:46:06,859 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 2.690e+02 3.227e+02 3.913e+02 5.489e+02, threshold=6.454e+02, percent-clipped=0.0 2023-05-16 09:46:34,223 INFO [finetune.py:992] (1/2) Epoch 8, batch 4650, loss[loss=0.1444, simple_loss=0.2316, pruned_loss=0.02855, over 11755.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2601, pruned_loss=0.04184, over 2355179.76 frames. ], batch size: 26, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:46:40,846 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190296.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:46:55,169 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190316.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:47:10,131 INFO [finetune.py:992] (1/2) Epoch 8, batch 4700, loss[loss=0.1502, simple_loss=0.2424, pruned_loss=0.02895, over 12190.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2601, pruned_loss=0.04156, over 2350209.48 frames. ], batch size: 29, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:47:11,029 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1813, 6.0978, 5.9273, 5.3567, 5.3053, 6.0198, 5.6643, 5.4209], device='cuda:1'), covar=tensor([0.0602, 0.0779, 0.0646, 0.1793, 0.0644, 0.0876, 0.1570, 0.1136], device='cuda:1'), in_proj_covar=tensor([0.0595, 0.0531, 0.0496, 0.0610, 0.0396, 0.0691, 0.0749, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 09:47:15,207 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190344.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:47:19,553 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.904e+02 3.405e+02 3.927e+02 7.391e+02, threshold=6.810e+02, percent-clipped=1.0 2023-05-16 09:47:26,596 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190360.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:47:38,828 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190377.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:47:43,814 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190384.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:47:45,817 INFO [finetune.py:992] (1/2) Epoch 8, batch 4750, loss[loss=0.1864, simple_loss=0.2785, pruned_loss=0.04711, over 12304.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2603, pruned_loss=0.04201, over 2358058.51 frames. ], batch size: 34, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:47:49,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-16 09:47:55,131 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190399.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:48:01,850 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190408.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:48:04,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 09:48:19,630 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190432.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:48:23,112 INFO [finetune.py:992] (1/2) Epoch 8, batch 4800, loss[loss=0.1536, simple_loss=0.2401, pruned_loss=0.03358, over 12349.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2588, pruned_loss=0.04144, over 2364253.40 frames. ], batch size: 31, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:48:32,413 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.074e+02 2.843e+02 3.222e+02 3.876e+02 1.576e+03, threshold=6.444e+02, percent-clipped=3.0 2023-05-16 09:48:39,816 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190460.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:48:42,654 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3628, 4.8780, 5.3442, 4.6525, 4.9484, 4.7030, 5.3743, 5.0859], device='cuda:1'), covar=tensor([0.0228, 0.0364, 0.0236, 0.0256, 0.0378, 0.0320, 0.0191, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0256, 0.0279, 0.0252, 0.0252, 0.0251, 0.0230, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:48:58,825 INFO [finetune.py:992] (1/2) Epoch 8, batch 4850, loss[loss=0.17, simple_loss=0.2721, pruned_loss=0.03393, over 10489.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2587, pruned_loss=0.04138, over 2364682.18 frames. ], batch size: 68, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:49:04,584 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190495.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 09:49:07,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6922, 2.9617, 4.7238, 4.8948, 2.7429, 2.7174, 3.1180, 2.1077], device='cuda:1'), covar=tensor([0.1586, 0.2815, 0.0425, 0.0374, 0.1372, 0.2293, 0.2537, 0.4086], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0376, 0.0266, 0.0292, 0.0258, 0.0289, 0.0362, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:49:13,083 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190507.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:49:34,167 INFO [finetune.py:992] (1/2) Epoch 8, batch 4900, loss[loss=0.1648, simple_loss=0.2491, pruned_loss=0.0403, over 12173.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2594, pruned_loss=0.04157, over 2368504.84 frames. ], batch size: 29, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:49:44,297 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.174e+02 2.711e+02 3.144e+02 3.905e+02 9.316e+02, threshold=6.287e+02, percent-clipped=4.0 2023-05-16 09:50:11,195 INFO [finetune.py:992] (1/2) Epoch 8, batch 4950, loss[loss=0.1581, simple_loss=0.2569, pruned_loss=0.02967, over 12152.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2598, pruned_loss=0.04151, over 2368985.55 frames. ], batch size: 34, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:50:14,225 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190591.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 09:50:27,366 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190609.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:50:29,624 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3023, 4.6694, 4.0649, 5.0150, 4.5173, 3.0516, 4.3263, 3.0241], device='cuda:1'), covar=tensor([0.0785, 0.0779, 0.1469, 0.0443, 0.1090, 0.1532, 0.1041, 0.3238], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0384, 0.0360, 0.0288, 0.0369, 0.0271, 0.0343, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:50:47,353 INFO [finetune.py:992] (1/2) Epoch 8, batch 5000, loss[loss=0.1476, simple_loss=0.2311, pruned_loss=0.03211, over 12275.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2605, pruned_loss=0.04179, over 2367054.12 frames. ], batch size: 28, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:50:56,667 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.729e+02 3.218e+02 4.194e+02 6.963e+02, threshold=6.436e+02, percent-clipped=1.0 2023-05-16 09:50:58,262 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190652.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 09:51:11,011 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190670.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 09:51:12,321 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190672.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:51:14,553 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190675.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:51:23,624 INFO [finetune.py:992] (1/2) Epoch 8, batch 5050, loss[loss=0.1596, simple_loss=0.248, pruned_loss=0.03564, over 12251.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2598, pruned_loss=0.04116, over 2366968.63 frames. ], batch size: 32, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:51:28,903 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6786, 2.7265, 3.5110, 4.5014, 2.7498, 4.6798, 4.6040, 4.8537], device='cuda:1'), covar=tensor([0.0119, 0.1159, 0.0394, 0.0160, 0.1166, 0.0184, 0.0123, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0200, 0.0179, 0.0114, 0.0188, 0.0173, 0.0171, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:51:59,605 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190736.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:52:00,158 INFO [finetune.py:992] (1/2) Epoch 8, batch 5100, loss[loss=0.1467, simple_loss=0.233, pruned_loss=0.03021, over 12179.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2598, pruned_loss=0.04137, over 2373456.40 frames. ], batch size: 29, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:52:09,423 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 2.722e+02 3.204e+02 3.972e+02 7.518e+02, threshold=6.407e+02, percent-clipped=2.0 2023-05-16 09:52:13,113 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190755.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:52:35,758 INFO [finetune.py:992] (1/2) Epoch 8, batch 5150, loss[loss=0.1765, simple_loss=0.2728, pruned_loss=0.04011, over 12037.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2605, pruned_loss=0.04194, over 2366061.44 frames. ], batch size: 40, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:52:41,587 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190795.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 09:52:50,238 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190807.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:53:12,290 INFO [finetune.py:992] (1/2) Epoch 8, batch 5200, loss[loss=0.1829, simple_loss=0.2801, pruned_loss=0.04287, over 12042.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.26, pruned_loss=0.0417, over 2363691.70 frames. ], batch size: 40, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:53:16,726 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190843.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 09:53:21,683 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.026e+02 2.723e+02 3.198e+02 3.819e+02 9.971e+02, threshold=6.395e+02, percent-clipped=4.0 2023-05-16 09:53:25,437 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190855.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:53:49,106 INFO [finetune.py:992] (1/2) Epoch 8, batch 5250, loss[loss=0.1407, simple_loss=0.223, pruned_loss=0.02921, over 12270.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2589, pruned_loss=0.04131, over 2363801.70 frames. ], batch size: 28, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:54:24,895 INFO [finetune.py:992] (1/2) Epoch 8, batch 5300, loss[loss=0.1573, simple_loss=0.2385, pruned_loss=0.03802, over 11767.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2599, pruned_loss=0.04179, over 2354054.69 frames. ], batch size: 26, lr: 4.30e-03, grad_scale: 8.0 2023-05-16 09:54:31,977 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190947.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 09:54:33,934 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.953e+02 2.779e+02 3.150e+02 3.652e+02 5.658e+02, threshold=6.299e+02, percent-clipped=0.0 2023-05-16 09:54:44,378 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190965.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 09:54:50,166 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190972.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:55:00,571 INFO [finetune.py:992] (1/2) Epoch 8, batch 5350, loss[loss=0.1582, simple_loss=0.2498, pruned_loss=0.03325, over 12103.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2595, pruned_loss=0.0416, over 2357430.73 frames. ], batch size: 32, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:55:11,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-05-16 09:55:25,215 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191020.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:55:33,126 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191031.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:55:37,376 INFO [finetune.py:992] (1/2) Epoch 8, batch 5400, loss[loss=0.1712, simple_loss=0.2629, pruned_loss=0.03979, over 12131.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.26, pruned_loss=0.04146, over 2364968.77 frames. ], batch size: 38, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:55:46,718 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 3.027e+02 3.387e+02 3.939e+02 1.797e+03, threshold=6.773e+02, percent-clipped=3.0 2023-05-16 09:55:47,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 09:55:50,505 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191055.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:55:59,779 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191068.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:56:04,082 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2064, 4.6249, 4.1015, 4.8887, 4.4416, 2.7820, 4.3713, 2.8607], device='cuda:1'), covar=tensor([0.0808, 0.0751, 0.1236, 0.0447, 0.1038, 0.1717, 0.0810, 0.3582], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0382, 0.0358, 0.0285, 0.0365, 0.0270, 0.0340, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:56:13,103 INFO [finetune.py:992] (1/2) Epoch 8, batch 5450, loss[loss=0.1492, simple_loss=0.2308, pruned_loss=0.03386, over 12328.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2592, pruned_loss=0.04122, over 2370396.87 frames. ], batch size: 30, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:56:16,747 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7595, 2.3826, 3.6058, 3.7455, 3.0104, 2.6145, 2.5982, 2.3334], device='cuda:1'), covar=tensor([0.1265, 0.3111, 0.0595, 0.0459, 0.0877, 0.2031, 0.2713, 0.3696], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0379, 0.0268, 0.0295, 0.0262, 0.0292, 0.0364, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:56:24,221 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191103.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:56:36,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 09:56:36,942 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0992, 2.2424, 3.2816, 3.9410, 3.5706, 3.9393, 3.6492, 2.8127], device='cuda:1'), covar=tensor([0.0035, 0.0379, 0.0156, 0.0042, 0.0109, 0.0077, 0.0095, 0.0320], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0117, 0.0101, 0.0074, 0.0097, 0.0112, 0.0089, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:56:43,112 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191129.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:56:45,146 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2820, 4.8573, 5.2566, 4.6038, 4.8678, 4.7025, 5.2510, 4.9964], device='cuda:1'), covar=tensor([0.0283, 0.0364, 0.0322, 0.0274, 0.0378, 0.0325, 0.0280, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0255, 0.0277, 0.0250, 0.0249, 0.0247, 0.0229, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:56:45,186 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191132.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:56:45,886 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6282, 3.6372, 3.2178, 3.3116, 2.9476, 2.8097, 3.6281, 2.2143], device='cuda:1'), covar=tensor([0.0346, 0.0148, 0.0188, 0.0172, 0.0381, 0.0339, 0.0132, 0.0474], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0162, 0.0156, 0.0181, 0.0203, 0.0197, 0.0167, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:56:48,526 INFO [finetune.py:992] (1/2) Epoch 8, batch 5500, loss[loss=0.1885, simple_loss=0.2733, pruned_loss=0.05182, over 11999.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2598, pruned_loss=0.04187, over 2364463.51 frames. ], batch size: 40, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:56:50,825 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4816, 4.9922, 5.4493, 4.7847, 5.0200, 4.8359, 5.4890, 5.1383], device='cuda:1'), covar=tensor([0.0238, 0.0341, 0.0272, 0.0229, 0.0358, 0.0308, 0.0196, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0255, 0.0277, 0.0249, 0.0249, 0.0247, 0.0228, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:56:58,291 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.004e+02 2.739e+02 3.385e+02 4.146e+02 3.019e+03, threshold=6.771e+02, percent-clipped=2.0 2023-05-16 09:57:23,269 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191185.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:57:24,465 INFO [finetune.py:992] (1/2) Epoch 8, batch 5550, loss[loss=0.2394, simple_loss=0.3136, pruned_loss=0.08255, over 8265.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2599, pruned_loss=0.04175, over 2367065.21 frames. ], batch size: 98, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:57:28,963 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191193.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:57:32,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 09:57:47,116 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1800, 2.6798, 3.7473, 3.1318, 3.5963, 3.2671, 2.6689, 3.5982], device='cuda:1'), covar=tensor([0.0128, 0.0320, 0.0132, 0.0211, 0.0133, 0.0164, 0.0312, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0198, 0.0178, 0.0176, 0.0204, 0.0153, 0.0190, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:57:55,679 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6418, 2.7212, 3.5837, 4.5045, 3.8866, 4.5291, 3.9818, 3.0558], device='cuda:1'), covar=tensor([0.0023, 0.0321, 0.0144, 0.0034, 0.0112, 0.0056, 0.0090, 0.0323], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0117, 0.0100, 0.0073, 0.0097, 0.0112, 0.0089, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 09:58:00,390 INFO [finetune.py:992] (1/2) Epoch 8, batch 5600, loss[loss=0.1689, simple_loss=0.261, pruned_loss=0.0384, over 10591.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2604, pruned_loss=0.04182, over 2370534.93 frames. ], batch size: 68, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:58:06,885 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191246.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:58:07,583 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191247.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 09:58:09,413 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.817e+02 3.355e+02 3.946e+02 6.887e+02, threshold=6.710e+02, percent-clipped=1.0 2023-05-16 09:58:20,897 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191265.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:58:36,587 INFO [finetune.py:992] (1/2) Epoch 8, batch 5650, loss[loss=0.152, simple_loss=0.2317, pruned_loss=0.03614, over 12278.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2601, pruned_loss=0.04188, over 2375385.70 frames. ], batch size: 28, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:58:37,613 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-16 09:58:42,823 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191295.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 09:58:50,040 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6575, 2.5466, 3.9524, 4.1158, 2.8109, 2.5531, 2.7262, 2.1880], device='cuda:1'), covar=tensor([0.1479, 0.2958, 0.0570, 0.0442, 0.1189, 0.2301, 0.2568, 0.3694], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0378, 0.0268, 0.0294, 0.0261, 0.0292, 0.0364, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:58:55,534 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191313.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:59:03,492 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5069, 4.7532, 4.2014, 4.9811, 4.5933, 2.7905, 4.2267, 3.0849], device='cuda:1'), covar=tensor([0.0718, 0.0831, 0.1594, 0.0787, 0.1121, 0.1830, 0.1199, 0.3424], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0382, 0.0357, 0.0285, 0.0366, 0.0269, 0.0341, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:59:08,261 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191331.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:59:12,422 INFO [finetune.py:992] (1/2) Epoch 8, batch 5700, loss[loss=0.1586, simple_loss=0.2431, pruned_loss=0.03708, over 12281.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2603, pruned_loss=0.04227, over 2366648.05 frames. ], batch size: 33, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 09:59:15,497 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6504, 2.8754, 4.6233, 4.8862, 3.0680, 2.7111, 3.0656, 2.2250], device='cuda:1'), covar=tensor([0.1590, 0.3116, 0.0446, 0.0345, 0.1144, 0.2211, 0.2570, 0.3954], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0377, 0.0268, 0.0294, 0.0261, 0.0292, 0.0364, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 09:59:21,697 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 2.701e+02 3.181e+02 3.913e+02 6.720e+02, threshold=6.362e+02, percent-clipped=1.0 2023-05-16 09:59:32,630 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191365.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:59:42,443 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191379.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 09:59:47,449 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3247, 4.9887, 5.2408, 5.1271, 4.8359, 5.1568, 5.0102, 3.1455], device='cuda:1'), covar=tensor([0.0062, 0.0057, 0.0058, 0.0061, 0.0048, 0.0072, 0.0069, 0.0569], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0076, 0.0078, 0.0071, 0.0058, 0.0088, 0.0077, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 09:59:48,022 INFO [finetune.py:992] (1/2) Epoch 8, batch 5750, loss[loss=0.1413, simple_loss=0.2277, pruned_loss=0.02749, over 12256.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2599, pruned_loss=0.04208, over 2358475.52 frames. ], batch size: 28, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:00:15,210 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191424.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:00:16,757 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191426.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:00:25,234 INFO [finetune.py:992] (1/2) Epoch 8, batch 5800, loss[loss=0.1457, simple_loss=0.2232, pruned_loss=0.03407, over 11847.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2594, pruned_loss=0.04187, over 2368348.73 frames. ], batch size: 26, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:00:34,272 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.981e+02 2.692e+02 3.071e+02 3.732e+02 6.496e+02, threshold=6.141e+02, percent-clipped=1.0 2023-05-16 10:00:49,486 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6390, 2.8236, 4.7550, 4.9567, 2.8394, 2.6461, 2.9263, 2.1446], device='cuda:1'), covar=tensor([0.1559, 0.3135, 0.0395, 0.0350, 0.1267, 0.2237, 0.2748, 0.4084], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0380, 0.0269, 0.0296, 0.0263, 0.0293, 0.0366, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:01:00,496 INFO [finetune.py:992] (1/2) Epoch 8, batch 5850, loss[loss=0.1848, simple_loss=0.269, pruned_loss=0.05035, over 12356.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2602, pruned_loss=0.04231, over 2371682.52 frames. ], batch size: 38, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:01:01,250 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191488.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:01:05,011 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3859, 4.8203, 2.9142, 2.4584, 4.2454, 2.4901, 4.1966, 3.4275], device='cuda:1'), covar=tensor([0.0662, 0.0584, 0.1150, 0.1654, 0.0275, 0.1493, 0.0436, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0253, 0.0175, 0.0197, 0.0140, 0.0180, 0.0193, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:01:08,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.56 vs. limit=5.0 2023-05-16 10:01:36,104 INFO [finetune.py:992] (1/2) Epoch 8, batch 5900, loss[loss=0.184, simple_loss=0.2756, pruned_loss=0.04617, over 11823.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2603, pruned_loss=0.04209, over 2379335.40 frames. ], batch size: 44, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:01:39,039 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191541.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:01:46,012 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.085e+02 2.786e+02 3.233e+02 3.888e+02 1.854e+03, threshold=6.466e+02, percent-clipped=4.0 2023-05-16 10:01:49,865 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4292, 3.5796, 3.3185, 3.2175, 2.9549, 2.7032, 3.6327, 2.3171], device='cuda:1'), covar=tensor([0.0402, 0.0121, 0.0151, 0.0175, 0.0348, 0.0315, 0.0124, 0.0442], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0160, 0.0154, 0.0181, 0.0200, 0.0196, 0.0164, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:02:12,293 INFO [finetune.py:992] (1/2) Epoch 8, batch 5950, loss[loss=0.2365, simple_loss=0.3074, pruned_loss=0.08283, over 8167.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2607, pruned_loss=0.04255, over 2381144.47 frames. ], batch size: 102, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:02:32,922 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4969, 2.5292, 3.5544, 4.3955, 3.8257, 4.3334, 3.8975, 2.9351], device='cuda:1'), covar=tensor([0.0034, 0.0359, 0.0139, 0.0032, 0.0105, 0.0078, 0.0094, 0.0372], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0118, 0.0101, 0.0074, 0.0098, 0.0114, 0.0090, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:02:48,863 INFO [finetune.py:992] (1/2) Epoch 8, batch 6000, loss[loss=0.1978, simple_loss=0.2865, pruned_loss=0.05459, over 12120.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2609, pruned_loss=0.04238, over 2379763.08 frames. ], batch size: 38, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:02:48,863 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 10:03:07,072 INFO [finetune.py:1026] (1/2) Epoch 8, validation: loss=0.3192, simple_loss=0.397, pruned_loss=0.1207, over 1020973.00 frames. 2023-05-16 10:03:07,073 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 10:03:16,318 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.692e+02 2.721e+02 3.249e+02 3.799e+02 7.618e+02, threshold=6.498e+02, percent-clipped=2.0 2023-05-16 10:03:33,221 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0640, 4.7320, 4.9378, 4.8328, 4.6088, 4.9271, 4.7439, 2.7378], device='cuda:1'), covar=tensor([0.0074, 0.0073, 0.0069, 0.0073, 0.0056, 0.0082, 0.0114, 0.0757], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0079, 0.0072, 0.0058, 0.0089, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:03:43,555 INFO [finetune.py:992] (1/2) Epoch 8, batch 6050, loss[loss=0.1944, simple_loss=0.2888, pruned_loss=0.05002, over 11849.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.262, pruned_loss=0.04276, over 2382785.82 frames. ], batch size: 44, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:04:07,069 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9584, 4.6287, 4.8279, 4.7768, 4.5840, 4.8299, 4.7275, 2.9018], device='cuda:1'), covar=tensor([0.0096, 0.0064, 0.0073, 0.0072, 0.0056, 0.0086, 0.0072, 0.0676], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0077, 0.0078, 0.0071, 0.0058, 0.0088, 0.0078, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:04:07,718 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191721.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:04:09,972 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191724.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:04:19,025 INFO [finetune.py:992] (1/2) Epoch 8, batch 6100, loss[loss=0.1616, simple_loss=0.2592, pruned_loss=0.03201, over 12269.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04259, over 2382331.35 frames. ], batch size: 37, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:04:28,199 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.691e+02 3.267e+02 4.002e+02 6.127e+02, threshold=6.534e+02, percent-clipped=0.0 2023-05-16 10:04:40,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-16 10:04:43,752 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191772.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:04:54,416 INFO [finetune.py:992] (1/2) Epoch 8, batch 6150, loss[loss=0.1648, simple_loss=0.2597, pruned_loss=0.03493, over 12183.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2611, pruned_loss=0.04266, over 2373851.00 frames. ], batch size: 35, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:04:55,279 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191788.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:05:03,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-16 10:05:30,680 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191836.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:05:31,316 INFO [finetune.py:992] (1/2) Epoch 8, batch 6200, loss[loss=0.1491, simple_loss=0.2393, pruned_loss=0.02945, over 12251.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2607, pruned_loss=0.04226, over 2380582.46 frames. ], batch size: 32, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:05:34,427 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191841.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:05:40,552 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.958e+02 3.387e+02 4.141e+02 7.725e+02, threshold=6.774e+02, percent-clipped=2.0 2023-05-16 10:06:06,566 INFO [finetune.py:992] (1/2) Epoch 8, batch 6250, loss[loss=0.1735, simple_loss=0.2622, pruned_loss=0.04239, over 11335.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04256, over 2370811.34 frames. ], batch size: 55, lr: 4.29e-03, grad_scale: 8.0 2023-05-16 10:06:08,074 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191889.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:06:09,553 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2787, 5.0212, 5.2366, 5.1474, 4.9260, 5.1721, 5.0719, 3.1742], device='cuda:1'), covar=tensor([0.0065, 0.0051, 0.0052, 0.0053, 0.0045, 0.0070, 0.0055, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0076, 0.0078, 0.0071, 0.0057, 0.0087, 0.0077, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:06:41,762 INFO [finetune.py:992] (1/2) Epoch 8, batch 6300, loss[loss=0.156, simple_loss=0.2418, pruned_loss=0.0351, over 12183.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2605, pruned_loss=0.04213, over 2371995.33 frames. ], batch size: 29, lr: 4.28e-03, grad_scale: 8.0 2023-05-16 10:06:51,672 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.004e+02 2.808e+02 3.454e+02 4.208e+02 8.526e+02, threshold=6.909e+02, percent-clipped=2.0 2023-05-16 10:07:18,491 INFO [finetune.py:992] (1/2) Epoch 8, batch 6350, loss[loss=0.173, simple_loss=0.254, pruned_loss=0.04596, over 12074.00 frames. ], tot_loss[loss=0.173, simple_loss=0.261, pruned_loss=0.04245, over 2362913.21 frames. ], batch size: 32, lr: 4.28e-03, grad_scale: 8.0 2023-05-16 10:07:30,878 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2905, 6.0769, 5.6771, 5.6724, 6.1385, 5.4773, 5.6825, 5.7483], device='cuda:1'), covar=tensor([0.1350, 0.0872, 0.0880, 0.1674, 0.0865, 0.2087, 0.1660, 0.1294], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0468, 0.0376, 0.0417, 0.0446, 0.0427, 0.0382, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:07:46,023 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192021.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:07:57,031 INFO [finetune.py:992] (1/2) Epoch 8, batch 6400, loss[loss=0.1422, simple_loss=0.2299, pruned_loss=0.02728, over 11843.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2609, pruned_loss=0.04238, over 2359989.48 frames. ], batch size: 26, lr: 4.28e-03, grad_scale: 8.0 2023-05-16 10:08:06,314 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.917e+02 3.513e+02 4.206e+02 9.694e+02, threshold=7.025e+02, percent-clipped=2.0 2023-05-16 10:08:20,245 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192069.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:08:33,713 INFO [finetune.py:992] (1/2) Epoch 8, batch 6450, loss[loss=0.1891, simple_loss=0.2854, pruned_loss=0.04645, over 11810.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2598, pruned_loss=0.04183, over 2363599.30 frames. ], batch size: 44, lr: 4.28e-03, grad_scale: 8.0 2023-05-16 10:08:44,575 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192102.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:08:48,141 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1028, 4.5733, 4.0154, 4.8226, 4.5847, 2.9085, 4.3623, 2.9563], device='cuda:1'), covar=tensor([0.0864, 0.0780, 0.1399, 0.0504, 0.0993, 0.1543, 0.0852, 0.3403], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0377, 0.0354, 0.0281, 0.0361, 0.0265, 0.0337, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:09:09,813 INFO [finetune.py:992] (1/2) Epoch 8, batch 6500, loss[loss=0.1756, simple_loss=0.2744, pruned_loss=0.0384, over 12356.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2603, pruned_loss=0.04194, over 2365499.58 frames. ], batch size: 35, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:09:19,203 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.680e+02 3.163e+02 3.527e+02 5.515e+02, threshold=6.326e+02, percent-clipped=0.0 2023-05-16 10:09:22,924 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192155.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:09:28,506 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192163.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:09:39,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 10:09:40,150 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0399, 6.0332, 5.7995, 5.3680, 5.1211, 5.9361, 5.5560, 5.3467], device='cuda:1'), covar=tensor([0.0644, 0.0835, 0.0665, 0.1476, 0.0778, 0.0734, 0.1601, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0597, 0.0535, 0.0499, 0.0621, 0.0405, 0.0699, 0.0758, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:09:44,957 INFO [finetune.py:992] (1/2) Epoch 8, batch 6550, loss[loss=0.1798, simple_loss=0.2798, pruned_loss=0.03997, over 12352.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2611, pruned_loss=0.04225, over 2365782.96 frames. ], batch size: 35, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:10:05,910 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192216.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:10:07,956 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192219.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:10:09,424 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192221.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:10:21,134 INFO [finetune.py:992] (1/2) Epoch 8, batch 6600, loss[loss=0.1388, simple_loss=0.2196, pruned_loss=0.02895, over 12246.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.261, pruned_loss=0.0423, over 2365769.33 frames. ], batch size: 28, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:10:24,254 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0429, 3.8940, 4.0346, 4.3488, 2.9960, 3.7568, 2.6646, 4.0611], device='cuda:1'), covar=tensor([0.1693, 0.0753, 0.0911, 0.0630, 0.1061, 0.0682, 0.1755, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0264, 0.0293, 0.0352, 0.0237, 0.0239, 0.0259, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:10:30,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 2.904e+02 3.471e+02 4.736e+02 9.167e+02, threshold=6.942e+02, percent-clipped=6.0 2023-05-16 10:10:52,724 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192280.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:10:54,173 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:10:57,651 INFO [finetune.py:992] (1/2) Epoch 8, batch 6650, loss[loss=0.1576, simple_loss=0.2453, pruned_loss=0.03491, over 12153.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2613, pruned_loss=0.04233, over 2362972.88 frames. ], batch size: 34, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:11:33,303 INFO [finetune.py:992] (1/2) Epoch 8, batch 6700, loss[loss=0.1376, simple_loss=0.2194, pruned_loss=0.02794, over 11794.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2609, pruned_loss=0.04219, over 2368915.52 frames. ], batch size: 26, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:11:42,650 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.670e+02 3.043e+02 3.525e+02 1.080e+03, threshold=6.085e+02, percent-clipped=1.0 2023-05-16 10:12:09,497 INFO [finetune.py:992] (1/2) Epoch 8, batch 6750, loss[loss=0.1442, simple_loss=0.2272, pruned_loss=0.03053, over 12195.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2601, pruned_loss=0.04178, over 2367204.10 frames. ], batch size: 29, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:12:13,171 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192392.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:12:18,023 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192399.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:12:28,713 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8762, 2.3535, 3.2352, 2.8551, 3.1759, 2.9784, 2.3631, 3.1818], device='cuda:1'), covar=tensor([0.0106, 0.0314, 0.0145, 0.0186, 0.0139, 0.0141, 0.0294, 0.0127], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0205, 0.0183, 0.0181, 0.0211, 0.0157, 0.0196, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:12:42,460 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192433.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:12:45,158 INFO [finetune.py:992] (1/2) Epoch 8, batch 6800, loss[loss=0.205, simple_loss=0.2861, pruned_loss=0.06192, over 8018.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2612, pruned_loss=0.04248, over 2363147.95 frames. ], batch size: 97, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:12:54,371 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 2.674e+02 3.537e+02 4.460e+02 1.477e+03, threshold=7.075e+02, percent-clipped=4.0 2023-05-16 10:12:56,684 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192453.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:13:00,140 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192458.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:13:01,761 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192460.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:13:03,959 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6959, 2.9940, 5.0721, 2.7398, 2.7703, 3.8630, 3.2926, 3.9313], device='cuda:1'), covar=tensor([0.0449, 0.1356, 0.0297, 0.1126, 0.1829, 0.1336, 0.1343, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0234, 0.0242, 0.0180, 0.0235, 0.0287, 0.0223, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:13:20,837 INFO [finetune.py:992] (1/2) Epoch 8, batch 6850, loss[loss=0.1378, simple_loss=0.2174, pruned_loss=0.02908, over 12014.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2609, pruned_loss=0.04234, over 2363185.07 frames. ], batch size: 28, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:13:26,005 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192494.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:13:31,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 10:13:38,264 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192511.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:13:43,409 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2680, 4.6667, 4.1528, 4.9903, 4.5850, 2.9315, 4.4345, 3.0597], device='cuda:1'), covar=tensor([0.0783, 0.0770, 0.1278, 0.0400, 0.1064, 0.1536, 0.0897, 0.3091], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0374, 0.0352, 0.0279, 0.0360, 0.0263, 0.0336, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:13:57,441 INFO [finetune.py:992] (1/2) Epoch 8, batch 6900, loss[loss=0.2114, simple_loss=0.2979, pruned_loss=0.06246, over 12376.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2601, pruned_loss=0.04191, over 2370470.19 frames. ], batch size: 38, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:14:07,464 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.847e+02 3.151e+02 3.808e+02 6.868e+02, threshold=6.302e+02, percent-clipped=0.0 2023-05-16 10:14:14,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 10:14:15,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 10:14:25,153 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192575.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:14:26,608 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192577.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:14:33,735 INFO [finetune.py:992] (1/2) Epoch 8, batch 6950, loss[loss=0.2033, simple_loss=0.2909, pruned_loss=0.05789, over 11811.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2609, pruned_loss=0.04209, over 2369157.28 frames. ], batch size: 44, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:14:54,924 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192616.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:15:09,836 INFO [finetune.py:992] (1/2) Epoch 8, batch 7000, loss[loss=0.1569, simple_loss=0.2507, pruned_loss=0.03159, over 12273.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2606, pruned_loss=0.04204, over 2366963.97 frames. ], batch size: 37, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:15:19,104 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.679e+02 3.248e+02 3.813e+02 6.575e+02, threshold=6.495e+02, percent-clipped=2.0 2023-05-16 10:15:36,653 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192673.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:15:39,516 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192677.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:15:43,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2509, 4.5500, 2.7711, 2.3924, 3.9710, 2.4128, 3.8902, 3.1637], device='cuda:1'), covar=tensor([0.0657, 0.0516, 0.1101, 0.1594, 0.0303, 0.1376, 0.0442, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0253, 0.0175, 0.0199, 0.0139, 0.0180, 0.0195, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:15:46,231 INFO [finetune.py:992] (1/2) Epoch 8, batch 7050, loss[loss=0.2035, simple_loss=0.288, pruned_loss=0.05943, over 10501.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2618, pruned_loss=0.04228, over 2374549.49 frames. ], batch size: 68, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:15:46,454 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3516, 4.7319, 2.9520, 2.5493, 4.1289, 2.3490, 4.0106, 3.2923], device='cuda:1'), covar=tensor([0.0694, 0.0633, 0.1109, 0.1595, 0.0280, 0.1547, 0.0487, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0254, 0.0175, 0.0199, 0.0139, 0.0180, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:15:50,061 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4504, 4.7939, 3.0341, 2.6223, 4.1942, 2.5988, 4.0594, 3.4803], device='cuda:1'), covar=tensor([0.0626, 0.0479, 0.1059, 0.1530, 0.0297, 0.1379, 0.0495, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0254, 0.0176, 0.0200, 0.0140, 0.0180, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:15:59,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 10:16:20,685 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192734.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:16:21,374 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192735.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:16:21,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 10:16:22,651 INFO [finetune.py:992] (1/2) Epoch 8, batch 7100, loss[loss=0.1889, simple_loss=0.2819, pruned_loss=0.04798, over 11817.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2621, pruned_loss=0.04246, over 2363092.76 frames. ], batch size: 49, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:16:30,565 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192748.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:16:30,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 10:16:31,940 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.754e+02 3.372e+02 4.025e+02 1.084e+03, threshold=6.745e+02, percent-clipped=3.0 2023-05-16 10:16:35,395 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192755.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:16:37,644 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192758.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:16:58,088 INFO [finetune.py:992] (1/2) Epoch 8, batch 7150, loss[loss=0.1565, simple_loss=0.2353, pruned_loss=0.03884, over 12293.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2618, pruned_loss=0.04237, over 2368136.12 frames. ], batch size: 28, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:16:59,590 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192789.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:17:04,718 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192796.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:17:12,021 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192806.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:17:14,401 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4363, 3.5317, 3.2609, 3.0756, 2.7849, 2.7212, 3.4717, 2.1005], device='cuda:1'), covar=tensor([0.0358, 0.0114, 0.0154, 0.0190, 0.0364, 0.0344, 0.0122, 0.0522], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0161, 0.0154, 0.0182, 0.0200, 0.0198, 0.0165, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:17:15,790 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192811.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:17:30,351 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192831.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:17:34,990 INFO [finetune.py:992] (1/2) Epoch 8, batch 7200, loss[loss=0.1555, simple_loss=0.237, pruned_loss=0.03696, over 12171.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2615, pruned_loss=0.04239, over 2373563.62 frames. ], batch size: 29, lr: 4.28e-03, grad_scale: 16.0 2023-05-16 10:17:44,258 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.677e+02 3.312e+02 4.462e+02 1.222e+03, threshold=6.625e+02, percent-clipped=7.0 2023-05-16 10:17:50,855 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192859.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:18:02,114 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192875.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:18:03,627 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192877.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:18:10,458 INFO [finetune.py:992] (1/2) Epoch 8, batch 7250, loss[loss=0.1495, simple_loss=0.2495, pruned_loss=0.02473, over 12311.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2616, pruned_loss=0.04256, over 2371483.60 frames. ], batch size: 34, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:18:14,174 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192892.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 10:18:35,952 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192923.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:18:37,310 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192925.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:18:45,911 INFO [finetune.py:992] (1/2) Epoch 8, batch 7300, loss[loss=0.1802, simple_loss=0.2716, pruned_loss=0.04442, over 12123.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2625, pruned_loss=0.04296, over 2362083.48 frames. ], batch size: 39, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:18:55,087 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.770e+02 3.180e+02 3.835e+02 6.912e+02, threshold=6.361e+02, percent-clipped=1.0 2023-05-16 10:19:07,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-05-16 10:19:11,494 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192972.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:19:22,580 INFO [finetune.py:992] (1/2) Epoch 8, batch 7350, loss[loss=0.1634, simple_loss=0.2611, pruned_loss=0.03289, over 12303.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2621, pruned_loss=0.04281, over 2373663.87 frames. ], batch size: 34, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:19:36,711 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6603, 3.4614, 5.0302, 2.6630, 2.8143, 3.6879, 3.2914, 3.8068], device='cuda:1'), covar=tensor([0.0456, 0.1028, 0.0267, 0.1067, 0.1845, 0.1551, 0.1302, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0232, 0.0239, 0.0178, 0.0234, 0.0285, 0.0221, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:19:51,950 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-05-16 10:19:52,923 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193029.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:19:58,429 INFO [finetune.py:992] (1/2) Epoch 8, batch 7400, loss[loss=0.1589, simple_loss=0.2471, pruned_loss=0.03536, over 12281.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04257, over 2369457.75 frames. ], batch size: 33, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:20:01,548 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193041.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:06,485 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193048.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:07,826 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.810e+02 3.225e+02 3.988e+02 1.116e+03, threshold=6.451e+02, percent-clipped=6.0 2023-05-16 10:20:11,639 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193055.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:34,155 INFO [finetune.py:992] (1/2) Epoch 8, batch 7450, loss[loss=0.18, simple_loss=0.2746, pruned_loss=0.04268, over 12345.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2616, pruned_loss=0.04252, over 2362342.61 frames. ], batch size: 36, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:20:35,815 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193089.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:37,175 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193091.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:37,298 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7635, 2.3412, 2.9375, 3.6919, 2.2983, 3.8067, 3.7053, 3.8677], device='cuda:1'), covar=tensor([0.0186, 0.1183, 0.0565, 0.0162, 0.1234, 0.0277, 0.0279, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0198, 0.0177, 0.0112, 0.0186, 0.0172, 0.0169, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:20:40,802 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193096.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:45,905 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193102.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:20:46,449 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193103.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:21:11,261 INFO [finetune.py:992] (1/2) Epoch 8, batch 7500, loss[loss=0.1624, simple_loss=0.247, pruned_loss=0.03888, over 12174.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2604, pruned_loss=0.04201, over 2367304.02 frames. ], batch size: 31, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:21:11,336 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193137.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:21:20,444 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.655e+02 3.373e+02 4.142e+02 9.369e+02, threshold=6.746e+02, percent-clipped=3.0 2023-05-16 10:21:38,187 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3632, 4.9305, 5.3393, 4.6286, 5.0000, 4.7272, 5.3433, 4.9760], device='cuda:1'), covar=tensor([0.0212, 0.0341, 0.0215, 0.0239, 0.0311, 0.0274, 0.0184, 0.0309], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0256, 0.0275, 0.0252, 0.0249, 0.0249, 0.0231, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:21:46,384 INFO [finetune.py:992] (1/2) Epoch 8, batch 7550, loss[loss=0.1984, simple_loss=0.291, pruned_loss=0.05286, over 12047.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2609, pruned_loss=0.04221, over 2370521.26 frames. ], batch size: 42, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:21:46,458 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193187.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:21:49,623 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 10:22:03,673 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6549, 2.8492, 3.2915, 4.5230, 2.5520, 4.6830, 4.5892, 4.7202], device='cuda:1'), covar=tensor([0.0114, 0.1056, 0.0462, 0.0111, 0.1172, 0.0175, 0.0153, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0200, 0.0179, 0.0113, 0.0187, 0.0173, 0.0170, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:22:03,683 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:22:20,458 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9664, 3.1310, 4.3981, 2.4227, 2.4958, 3.3574, 2.9013, 3.4692], device='cuda:1'), covar=tensor([0.0527, 0.1070, 0.0352, 0.1169, 0.1918, 0.1367, 0.1379, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0233, 0.0241, 0.0180, 0.0236, 0.0286, 0.0223, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:22:21,699 INFO [finetune.py:992] (1/2) Epoch 8, batch 7600, loss[loss=0.1421, simple_loss=0.2251, pruned_loss=0.0295, over 12352.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2606, pruned_loss=0.04227, over 2377136.43 frames. ], batch size: 30, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:22:31,569 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.941e+02 3.431e+02 4.021e+02 6.757e+02, threshold=6.862e+02, percent-clipped=1.0 2023-05-16 10:22:47,992 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193272.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:22:48,074 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193272.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 10:22:56,620 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193284.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:22:58,693 INFO [finetune.py:992] (1/2) Epoch 8, batch 7650, loss[loss=0.1575, simple_loss=0.236, pruned_loss=0.03946, over 11772.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.261, pruned_loss=0.0426, over 2376563.66 frames. ], batch size: 26, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:23:21,983 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193320.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:23:28,418 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193329.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:23:33,883 INFO [finetune.py:992] (1/2) Epoch 8, batch 7700, loss[loss=0.1554, simple_loss=0.2529, pruned_loss=0.02892, over 12161.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.262, pruned_loss=0.04324, over 2362520.55 frames. ], batch size: 36, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:23:39,638 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193345.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:23:43,036 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.152e+02 2.912e+02 3.447e+02 4.156e+02 7.107e+02, threshold=6.894e+02, percent-clipped=1.0 2023-05-16 10:24:02,203 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193377.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:24:09,245 INFO [finetune.py:992] (1/2) Epoch 8, batch 7750, loss[loss=0.1588, simple_loss=0.2456, pruned_loss=0.03593, over 12299.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2621, pruned_loss=0.04336, over 2361894.34 frames. ], batch size: 33, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:24:12,356 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193391.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:24:17,240 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193397.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:24:35,350 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193421.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:24:46,558 INFO [finetune.py:992] (1/2) Epoch 8, batch 7800, loss[loss=0.1901, simple_loss=0.2738, pruned_loss=0.05316, over 12023.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2623, pruned_loss=0.04354, over 2365407.56 frames. ], batch size: 40, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:24:48,022 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193439.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:24:55,873 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 2.834e+02 3.445e+02 3.968e+02 8.886e+02, threshold=6.891e+02, percent-clipped=1.0 2023-05-16 10:24:58,977 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193454.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:25:07,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 10:25:18,884 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193482.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 10:25:22,205 INFO [finetune.py:992] (1/2) Epoch 8, batch 7850, loss[loss=0.1617, simple_loss=0.2462, pruned_loss=0.03859, over 12430.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2618, pruned_loss=0.0432, over 2365623.16 frames. ], batch size: 32, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:25:22,312 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193487.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:25:26,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 10:25:42,393 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193515.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:25:57,110 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193535.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:25:58,439 INFO [finetune.py:992] (1/2) Epoch 8, batch 7900, loss[loss=0.1482, simple_loss=0.2343, pruned_loss=0.03103, over 12120.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2614, pruned_loss=0.04297, over 2367441.54 frames. ], batch size: 30, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:26:08,196 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.779e+02 3.308e+02 3.802e+02 1.440e+03, threshold=6.616e+02, percent-clipped=1.0 2023-05-16 10:26:20,252 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193567.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:26:34,102 INFO [finetune.py:992] (1/2) Epoch 8, batch 7950, loss[loss=0.1575, simple_loss=0.2427, pruned_loss=0.03608, over 12285.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2612, pruned_loss=0.04324, over 2364816.61 frames. ], batch size: 33, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:26:54,260 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:27:03,382 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0583, 1.9256, 2.8848, 2.9703, 3.0062, 3.0453, 3.0219, 2.3304], device='cuda:1'), covar=tensor([0.0070, 0.0444, 0.0177, 0.0143, 0.0105, 0.0111, 0.0104, 0.0356], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0120, 0.0103, 0.0076, 0.0100, 0.0117, 0.0092, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:27:09,458 INFO [finetune.py:992] (1/2) Epoch 8, batch 8000, loss[loss=0.1895, simple_loss=0.2827, pruned_loss=0.04818, over 10673.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2609, pruned_loss=0.04315, over 2357501.50 frames. ], batch size: 68, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:27:11,726 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193640.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:27:18,679 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.127e+02 2.742e+02 3.203e+02 3.870e+02 1.111e+03, threshold=6.406e+02, percent-clipped=2.0 2023-05-16 10:27:37,298 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193676.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:27:46,193 INFO [finetune.py:992] (1/2) Epoch 8, batch 8050, loss[loss=0.267, simple_loss=0.3302, pruned_loss=0.1019, over 7814.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2612, pruned_loss=0.04299, over 2364175.06 frames. ], batch size: 98, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:27:53,481 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193697.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:28:15,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 10:28:21,677 INFO [finetune.py:992] (1/2) Epoch 8, batch 8100, loss[loss=0.1758, simple_loss=0.2667, pruned_loss=0.04245, over 12257.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2607, pruned_loss=0.04324, over 2356321.46 frames. ], batch size: 37, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:28:27,492 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193745.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:28:30,929 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.047e+02 2.724e+02 3.178e+02 4.113e+02 8.569e+02, threshold=6.357e+02, percent-clipped=4.0 2023-05-16 10:28:50,295 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193777.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:28:57,380 INFO [finetune.py:992] (1/2) Epoch 8, batch 8150, loss[loss=0.1595, simple_loss=0.2508, pruned_loss=0.0341, over 12117.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2612, pruned_loss=0.04325, over 2352873.96 frames. ], batch size: 33, lr: 4.27e-03, grad_scale: 16.0 2023-05-16 10:29:14,051 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193810.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:29:33,675 INFO [finetune.py:992] (1/2) Epoch 8, batch 8200, loss[loss=0.1681, simple_loss=0.2424, pruned_loss=0.04693, over 12286.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2614, pruned_loss=0.04327, over 2360068.94 frames. ], batch size: 28, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:29:42,871 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.208e+02 2.824e+02 3.337e+02 3.924e+02 1.543e+03, threshold=6.675e+02, percent-clipped=2.0 2023-05-16 10:29:52,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2023-05-16 10:29:54,937 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193867.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:30:06,875 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193884.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:30:08,872 INFO [finetune.py:992] (1/2) Epoch 8, batch 8250, loss[loss=0.1872, simple_loss=0.2729, pruned_loss=0.0507, over 12121.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2608, pruned_loss=0.04298, over 2362098.80 frames. ], batch size: 33, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:30:28,597 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1127, 2.0235, 2.7884, 3.0796, 2.9293, 3.1193, 2.8896, 2.3839], device='cuda:1'), covar=tensor([0.0073, 0.0400, 0.0175, 0.0086, 0.0124, 0.0105, 0.0124, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0120, 0.0103, 0.0076, 0.0100, 0.0117, 0.0092, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:30:29,183 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193915.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:30:44,548 INFO [finetune.py:992] (1/2) Epoch 8, batch 8300, loss[loss=0.1623, simple_loss=0.2466, pruned_loss=0.03899, over 12101.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2614, pruned_loss=0.04311, over 2358109.67 frames. ], batch size: 32, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:30:46,680 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193940.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:30:50,306 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193945.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:30:53,650 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 2.837e+02 3.419e+02 4.022e+02 1.405e+03, threshold=6.839e+02, percent-clipped=4.0 2023-05-16 10:31:08,591 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193971.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:31:21,241 INFO [finetune.py:992] (1/2) Epoch 8, batch 8350, loss[loss=0.1537, simple_loss=0.2399, pruned_loss=0.03381, over 12081.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2615, pruned_loss=0.04312, over 2365378.47 frames. ], batch size: 32, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:31:22,047 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193988.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:31:41,362 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5999, 2.2107, 2.9719, 2.6005, 2.8882, 2.8223, 2.1419, 2.9599], device='cuda:1'), covar=tensor([0.0112, 0.0295, 0.0168, 0.0209, 0.0130, 0.0144, 0.0292, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0200, 0.0181, 0.0179, 0.0206, 0.0155, 0.0192, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:31:59,569 INFO [finetune.py:992] (1/2) Epoch 8, batch 8400, loss[loss=0.1783, simple_loss=0.2726, pruned_loss=0.04196, over 12148.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2627, pruned_loss=0.04368, over 2356301.57 frames. ], batch size: 36, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:32:08,740 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 2.838e+02 3.415e+02 4.165e+02 1.425e+03, threshold=6.831e+02, percent-clipped=3.0 2023-05-16 10:32:12,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 10:32:15,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8259, 2.9693, 4.6627, 4.8220, 2.9472, 2.7214, 3.1039, 2.1972], device='cuda:1'), covar=tensor([0.1359, 0.2988, 0.0439, 0.0398, 0.1191, 0.2205, 0.2498, 0.3788], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0378, 0.0267, 0.0295, 0.0260, 0.0292, 0.0363, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:32:24,557 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194072.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:32:28,003 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194077.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 10:32:28,707 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2642, 6.0399, 5.4968, 5.5768, 6.1100, 5.4071, 5.6534, 5.6415], device='cuda:1'), covar=tensor([0.1356, 0.0865, 0.0869, 0.1784, 0.1029, 0.2417, 0.1558, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0466, 0.0377, 0.0416, 0.0443, 0.0426, 0.0378, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:32:34,901 INFO [finetune.py:992] (1/2) Epoch 8, batch 8450, loss[loss=0.2016, simple_loss=0.285, pruned_loss=0.05906, over 12109.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2632, pruned_loss=0.04393, over 2355033.15 frames. ], batch size: 38, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:32:51,281 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194110.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:32:59,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 10:33:03,276 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194125.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:33:09,188 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194133.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:33:11,808 INFO [finetune.py:992] (1/2) Epoch 8, batch 8500, loss[loss=0.1396, simple_loss=0.2243, pruned_loss=0.02744, over 12126.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2626, pruned_loss=0.04334, over 2365883.15 frames. ], batch size: 30, lr: 4.26e-03, grad_scale: 32.0 2023-05-16 10:33:20,968 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.571e+02 3.045e+02 3.848e+02 1.070e+03, threshold=6.091e+02, percent-clipped=1.0 2023-05-16 10:33:26,802 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194158.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:33:46,923 INFO [finetune.py:992] (1/2) Epoch 8, batch 8550, loss[loss=0.1711, simple_loss=0.2615, pruned_loss=0.04037, over 12263.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2626, pruned_loss=0.04349, over 2354591.79 frames. ], batch size: 32, lr: 4.26e-03, grad_scale: 32.0 2023-05-16 10:34:23,064 INFO [finetune.py:992] (1/2) Epoch 8, batch 8600, loss[loss=0.1581, simple_loss=0.2414, pruned_loss=0.03744, over 11837.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.262, pruned_loss=0.04304, over 2359387.24 frames. ], batch size: 26, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:34:25,260 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194240.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:34:32,993 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.656e+02 3.272e+02 4.250e+02 7.840e+02, threshold=6.543e+02, percent-clipped=3.0 2023-05-16 10:34:48,794 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194271.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:35:00,245 INFO [finetune.py:992] (1/2) Epoch 8, batch 8650, loss[loss=0.1825, simple_loss=0.2752, pruned_loss=0.04489, over 12156.00 frames. ], tot_loss[loss=0.174, simple_loss=0.262, pruned_loss=0.04306, over 2360138.98 frames. ], batch size: 36, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:35:06,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 10:35:21,993 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194318.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:35:22,591 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194319.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:35:35,220 INFO [finetune.py:992] (1/2) Epoch 8, batch 8700, loss[loss=0.1789, simple_loss=0.2695, pruned_loss=0.04412, over 11681.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2611, pruned_loss=0.04271, over 2364779.42 frames. ], batch size: 48, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:35:45,192 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.865e+02 3.268e+02 3.963e+02 7.088e+02, threshold=6.536e+02, percent-clipped=1.0 2023-05-16 10:35:46,821 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194353.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:35:53,386 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194362.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:36:05,527 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194379.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:36:11,050 INFO [finetune.py:992] (1/2) Epoch 8, batch 8750, loss[loss=0.1547, simple_loss=0.2419, pruned_loss=0.03376, over 12121.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.261, pruned_loss=0.04242, over 2365206.37 frames. ], batch size: 33, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:36:32,111 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194414.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:36:38,619 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194423.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 10:36:42,120 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194428.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:36:48,327 INFO [finetune.py:992] (1/2) Epoch 8, batch 8800, loss[loss=0.1934, simple_loss=0.2793, pruned_loss=0.05376, over 12113.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2612, pruned_loss=0.04237, over 2374860.93 frames. ], batch size: 38, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:36:58,330 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.631e+02 3.233e+02 4.193e+02 6.169e+02, threshold=6.466e+02, percent-clipped=0.0 2023-05-16 10:37:23,949 INFO [finetune.py:992] (1/2) Epoch 8, batch 8850, loss[loss=0.1715, simple_loss=0.265, pruned_loss=0.03903, over 12155.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2603, pruned_loss=0.04193, over 2376749.27 frames. ], batch size: 34, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:37:59,263 INFO [finetune.py:992] (1/2) Epoch 8, batch 8900, loss[loss=0.2815, simple_loss=0.3341, pruned_loss=0.1145, over 7846.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2606, pruned_loss=0.04229, over 2369475.52 frames. ], batch size: 98, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:38:02,187 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194540.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:38:10,023 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.719e+02 3.310e+02 4.071e+02 9.706e+02, threshold=6.620e+02, percent-clipped=3.0 2023-05-16 10:38:36,140 INFO [finetune.py:992] (1/2) Epoch 8, batch 8950, loss[loss=0.1765, simple_loss=0.2711, pruned_loss=0.04096, over 12200.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2605, pruned_loss=0.04213, over 2376415.49 frames. ], batch size: 35, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:38:36,935 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194588.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:38:48,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 10:38:59,202 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0258, 3.6196, 5.2905, 2.8749, 2.8995, 3.9558, 3.5254, 4.0944], device='cuda:1'), covar=tensor([0.0377, 0.0920, 0.0322, 0.1115, 0.1763, 0.1474, 0.1168, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0236, 0.0244, 0.0181, 0.0240, 0.0291, 0.0226, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:39:03,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2023-05-16 10:39:09,707 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194634.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:39:11,524 INFO [finetune.py:992] (1/2) Epoch 8, batch 9000, loss[loss=0.1914, simple_loss=0.2792, pruned_loss=0.05181, over 12344.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2596, pruned_loss=0.04164, over 2381337.17 frames. ], batch size: 31, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:39:11,524 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 10:39:29,678 INFO [finetune.py:1026] (1/2) Epoch 8, validation: loss=0.3234, simple_loss=0.3994, pruned_loss=0.1237, over 1020973.00 frames. 2023-05-16 10:39:29,679 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 10:39:40,953 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.865e+02 3.334e+02 3.958e+02 1.099e+03, threshold=6.669e+02, percent-clipped=1.0 2023-05-16 10:39:57,621 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194674.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:40:06,665 INFO [finetune.py:992] (1/2) Epoch 8, batch 9050, loss[loss=0.1789, simple_loss=0.2676, pruned_loss=0.04509, over 12311.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2597, pruned_loss=0.04176, over 2374555.17 frames. ], batch size: 34, lr: 4.26e-03, grad_scale: 16.0 2023-05-16 10:40:12,628 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194695.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:40:22,406 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194709.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:40:28,787 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194718.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 10:40:35,947 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194728.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:40:42,195 INFO [finetune.py:992] (1/2) Epoch 8, batch 9100, loss[loss=0.1739, simple_loss=0.2669, pruned_loss=0.04041, over 10557.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.26, pruned_loss=0.04165, over 2375795.77 frames. ], batch size: 68, lr: 4.25e-03, grad_scale: 16.0 2023-05-16 10:40:52,203 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.759e+02 3.384e+02 3.859e+02 6.643e+02, threshold=6.768e+02, percent-clipped=0.0 2023-05-16 10:41:10,295 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194776.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:41:18,666 INFO [finetune.py:992] (1/2) Epoch 8, batch 9150, loss[loss=0.1934, simple_loss=0.2787, pruned_loss=0.05407, over 11715.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.259, pruned_loss=0.0411, over 2381418.79 frames. ], batch size: 48, lr: 4.25e-03, grad_scale: 16.0 2023-05-16 10:41:24,968 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1453, 4.7311, 4.9825, 5.0079, 4.7427, 4.9471, 4.9086, 2.8331], device='cuda:1'), covar=tensor([0.0077, 0.0069, 0.0072, 0.0056, 0.0047, 0.0091, 0.0081, 0.0652], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0079, 0.0072, 0.0058, 0.0089, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:41:30,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 10:41:54,879 INFO [finetune.py:992] (1/2) Epoch 8, batch 9200, loss[loss=0.1992, simple_loss=0.2903, pruned_loss=0.05399, over 12137.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2589, pruned_loss=0.04101, over 2386457.04 frames. ], batch size: 36, lr: 4.25e-03, grad_scale: 16.0 2023-05-16 10:42:02,233 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194847.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:42:04,897 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.634e+02 3.159e+02 3.699e+02 5.868e+02, threshold=6.317e+02, percent-clipped=0.0 2023-05-16 10:42:30,419 INFO [finetune.py:992] (1/2) Epoch 8, batch 9250, loss[loss=0.2605, simple_loss=0.3206, pruned_loss=0.1002, over 7779.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2588, pruned_loss=0.0412, over 2380809.84 frames. ], batch size: 98, lr: 4.25e-03, grad_scale: 16.0 2023-05-16 10:42:45,467 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194908.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:42:47,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.18 vs. limit=5.0 2023-05-16 10:42:51,808 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8722, 4.7694, 4.8662, 4.8730, 4.5722, 4.5662, 4.4161, 4.7393], device='cuda:1'), covar=tensor([0.0634, 0.0508, 0.0715, 0.0570, 0.1561, 0.1165, 0.0517, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0518, 0.0674, 0.0585, 0.0608, 0.0828, 0.0717, 0.0533, 0.0474], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:43:02,996 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0714, 4.6363, 4.6823, 4.9661, 4.7659, 4.8911, 4.7513, 2.7333], device='cuda:1'), covar=tensor([0.0095, 0.0090, 0.0109, 0.0062, 0.0048, 0.0095, 0.0102, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0080, 0.0072, 0.0058, 0.0089, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:43:06,233 INFO [finetune.py:992] (1/2) Epoch 8, batch 9300, loss[loss=0.1797, simple_loss=0.2749, pruned_loss=0.04225, over 11249.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2592, pruned_loss=0.04157, over 2378089.10 frames. ], batch size: 55, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:43:17,503 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.938e+02 3.434e+02 4.053e+02 6.793e+02, threshold=6.868e+02, percent-clipped=2.0 2023-05-16 10:43:33,148 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194974.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:43:42,311 INFO [finetune.py:992] (1/2) Epoch 8, batch 9350, loss[loss=0.133, simple_loss=0.2137, pruned_loss=0.02617, over 12018.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2586, pruned_loss=0.0414, over 2376842.90 frames. ], batch size: 31, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:43:44,696 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194990.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:43:45,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 10:43:56,160 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0014, 3.8047, 3.9972, 3.7014, 3.8289, 3.6906, 3.9865, 3.6236], device='cuda:1'), covar=tensor([0.0324, 0.0408, 0.0362, 0.0263, 0.0335, 0.0317, 0.0288, 0.1402], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0259, 0.0279, 0.0255, 0.0254, 0.0251, 0.0233, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:43:58,313 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195009.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:43:59,854 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195011.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:44:03,338 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5188, 2.5333, 3.2044, 4.4972, 2.1152, 4.3888, 4.4828, 4.5843], device='cuda:1'), covar=tensor([0.0122, 0.1230, 0.0493, 0.0137, 0.1419, 0.0245, 0.0157, 0.0093], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0201, 0.0179, 0.0115, 0.0186, 0.0175, 0.0172, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:44:04,720 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195018.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:44:07,444 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195022.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:44:12,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-05-16 10:44:18,088 INFO [finetune.py:992] (1/2) Epoch 8, batch 9400, loss[loss=0.1587, simple_loss=0.2457, pruned_loss=0.03585, over 12017.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2593, pruned_loss=0.04203, over 2370199.76 frames. ], batch size: 31, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:44:23,298 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195044.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:44:28,844 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.895e+02 3.272e+02 4.035e+02 1.379e+03, threshold=6.543e+02, percent-clipped=4.0 2023-05-16 10:44:32,493 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195057.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:44:33,944 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0430, 4.6443, 4.1833, 4.2099, 4.7147, 4.1689, 4.2595, 4.1701], device='cuda:1'), covar=tensor([0.1676, 0.1005, 0.1161, 0.1862, 0.1095, 0.2187, 0.1650, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0477, 0.0381, 0.0425, 0.0452, 0.0431, 0.0385, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:44:38,704 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195066.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:44:42,341 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195071.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:44:43,117 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195072.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 10:44:53,746 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1334, 2.0312, 2.6856, 3.2093, 2.0520, 3.2538, 3.1287, 3.3086], device='cuda:1'), covar=tensor([0.0159, 0.1138, 0.0464, 0.0163, 0.1165, 0.0332, 0.0327, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0201, 0.0179, 0.0115, 0.0186, 0.0174, 0.0171, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:44:54,961 INFO [finetune.py:992] (1/2) Epoch 8, batch 9450, loss[loss=0.1747, simple_loss=0.2571, pruned_loss=0.04614, over 12336.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2598, pruned_loss=0.04207, over 2367062.48 frames. ], batch size: 30, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:45:08,026 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195105.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:45:27,072 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195132.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:45:30,271 INFO [finetune.py:992] (1/2) Epoch 8, batch 9500, loss[loss=0.1895, simple_loss=0.2797, pruned_loss=0.04969, over 11783.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2601, pruned_loss=0.04213, over 2371240.62 frames. ], batch size: 44, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:45:40,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.744e+02 3.314e+02 3.809e+02 6.264e+02, threshold=6.628e+02, percent-clipped=0.0 2023-05-16 10:46:05,311 INFO [finetune.py:992] (1/2) Epoch 8, batch 9550, loss[loss=0.1639, simple_loss=0.2579, pruned_loss=0.03497, over 12314.00 frames. ], tot_loss[loss=0.173, simple_loss=0.261, pruned_loss=0.04254, over 2359226.83 frames. ], batch size: 34, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:46:16,995 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195203.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:46:27,134 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3748, 3.5435, 3.2818, 3.7688, 3.4616, 2.6093, 3.3064, 2.8628], device='cuda:1'), covar=tensor([0.0858, 0.1113, 0.1577, 0.0587, 0.1315, 0.1551, 0.1102, 0.2568], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0372, 0.0352, 0.0282, 0.0361, 0.0264, 0.0335, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:46:31,932 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6677, 2.8379, 4.6383, 4.7565, 2.9168, 2.5977, 2.9057, 2.1323], device='cuda:1'), covar=tensor([0.1470, 0.2738, 0.0409, 0.0395, 0.1216, 0.2266, 0.2593, 0.3865], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0376, 0.0268, 0.0295, 0.0261, 0.0293, 0.0365, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:46:42,147 INFO [finetune.py:992] (1/2) Epoch 8, batch 9600, loss[loss=0.1418, simple_loss=0.2196, pruned_loss=0.03196, over 12196.00 frames. ], tot_loss[loss=0.172, simple_loss=0.26, pruned_loss=0.04202, over 2363721.21 frames. ], batch size: 29, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:46:52,604 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.821e+02 3.505e+02 4.259e+02 9.159e+02, threshold=7.010e+02, percent-clipped=6.0 2023-05-16 10:47:08,230 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9013, 3.3239, 2.4134, 2.1257, 2.9792, 2.2944, 3.1278, 2.6078], device='cuda:1'), covar=tensor([0.0551, 0.0701, 0.0864, 0.1263, 0.0363, 0.0997, 0.0560, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0253, 0.0174, 0.0198, 0.0140, 0.0179, 0.0196, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:47:09,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 10:47:17,319 INFO [finetune.py:992] (1/2) Epoch 8, batch 9650, loss[loss=0.1614, simple_loss=0.253, pruned_loss=0.03492, over 12152.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2605, pruned_loss=0.0421, over 2368720.88 frames. ], batch size: 34, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:47:19,543 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195290.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:47:26,786 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4712, 3.6350, 3.2544, 3.2053, 2.9138, 2.7106, 3.6616, 2.1576], device='cuda:1'), covar=tensor([0.0383, 0.0133, 0.0185, 0.0168, 0.0388, 0.0363, 0.0106, 0.0475], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0162, 0.0157, 0.0182, 0.0202, 0.0201, 0.0167, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:47:26,841 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3966, 4.5961, 4.1696, 4.9697, 4.5896, 2.8076, 4.2455, 3.0787], device='cuda:1'), covar=tensor([0.0685, 0.0814, 0.1267, 0.0447, 0.0931, 0.1743, 0.0977, 0.3196], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0374, 0.0353, 0.0284, 0.0362, 0.0265, 0.0337, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:47:40,477 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1396, 3.9681, 3.9021, 4.2890, 2.9635, 3.6922, 2.5822, 3.9427], device='cuda:1'), covar=tensor([0.1642, 0.0732, 0.1024, 0.0646, 0.1206, 0.0703, 0.1840, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0264, 0.0294, 0.0353, 0.0236, 0.0236, 0.0257, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:47:53,030 INFO [finetune.py:992] (1/2) Epoch 8, batch 9700, loss[loss=0.1691, simple_loss=0.2684, pruned_loss=0.03487, over 11233.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2597, pruned_loss=0.04202, over 2367053.98 frames. ], batch size: 55, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:47:53,841 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195338.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:47:54,655 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4689, 5.3118, 5.3973, 5.4650, 5.0597, 5.1127, 4.9353, 5.3284], device='cuda:1'), covar=tensor([0.0593, 0.0506, 0.0592, 0.0408, 0.1723, 0.1133, 0.0466, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0522, 0.0676, 0.0589, 0.0606, 0.0832, 0.0718, 0.0533, 0.0474], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:48:03,541 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 2.767e+02 3.197e+02 3.864e+02 7.277e+02, threshold=6.395e+02, percent-clipped=2.0 2023-05-16 10:48:03,657 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0356, 5.8279, 5.4168, 5.3358, 5.9515, 5.2497, 5.5303, 5.4070], device='cuda:1'), covar=tensor([0.1378, 0.1110, 0.1264, 0.1909, 0.1000, 0.2090, 0.1650, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0476, 0.0380, 0.0421, 0.0448, 0.0429, 0.0384, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 10:48:14,977 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195367.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 10:48:29,494 INFO [finetune.py:992] (1/2) Epoch 8, batch 9750, loss[loss=0.1639, simple_loss=0.2594, pruned_loss=0.03418, over 12100.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2603, pruned_loss=0.04193, over 2366974.79 frames. ], batch size: 33, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:48:33,153 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195392.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:48:39,062 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195400.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:48:43,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-05-16 10:48:43,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-16 10:48:58,023 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195427.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:49:04,958 INFO [finetune.py:992] (1/2) Epoch 8, batch 9800, loss[loss=0.1634, simple_loss=0.2468, pruned_loss=0.03996, over 12263.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2603, pruned_loss=0.04232, over 2364303.54 frames. ], batch size: 32, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:49:15,536 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.074e+02 2.933e+02 3.542e+02 4.192e+02 1.012e+03, threshold=7.085e+02, percent-clipped=3.0 2023-05-16 10:49:16,463 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195453.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:49:17,934 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1147, 4.4459, 3.8899, 4.7417, 4.3546, 2.8915, 4.1627, 2.9435], device='cuda:1'), covar=tensor([0.0813, 0.0840, 0.1439, 0.0518, 0.1093, 0.1619, 0.0956, 0.3302], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0376, 0.0355, 0.0284, 0.0364, 0.0266, 0.0339, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:49:18,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 10:49:27,270 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3143, 4.5748, 3.9983, 4.9276, 4.5423, 2.9859, 4.3489, 3.0786], device='cuda:1'), covar=tensor([0.0741, 0.0752, 0.1354, 0.0406, 0.0965, 0.1506, 0.0897, 0.2990], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0375, 0.0354, 0.0284, 0.0363, 0.0265, 0.0338, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:49:40,723 INFO [finetune.py:992] (1/2) Epoch 8, batch 9850, loss[loss=0.172, simple_loss=0.2716, pruned_loss=0.03626, over 11723.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2611, pruned_loss=0.04261, over 2364619.40 frames. ], batch size: 48, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:49:52,342 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195503.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:50:16,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 10:50:17,742 INFO [finetune.py:992] (1/2) Epoch 8, batch 9900, loss[loss=0.1658, simple_loss=0.2487, pruned_loss=0.04141, over 12001.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2605, pruned_loss=0.04258, over 2359702.01 frames. ], batch size: 28, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:50:27,670 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195551.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:50:28,283 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.980e+02 2.769e+02 3.203e+02 3.616e+02 6.223e+02, threshold=6.405e+02, percent-clipped=0.0 2023-05-16 10:50:52,660 INFO [finetune.py:992] (1/2) Epoch 8, batch 9950, loss[loss=0.1646, simple_loss=0.2561, pruned_loss=0.03651, over 12115.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2609, pruned_loss=0.04275, over 2358900.18 frames. ], batch size: 33, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:51:28,541 INFO [finetune.py:992] (1/2) Epoch 8, batch 10000, loss[loss=0.1448, simple_loss=0.2305, pruned_loss=0.02955, over 12004.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2607, pruned_loss=0.04254, over 2360573.60 frames. ], batch size: 28, lr: 4.25e-03, grad_scale: 8.0 2023-05-16 10:51:39,702 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.768e+02 3.381e+02 3.907e+02 5.631e+02, threshold=6.763e+02, percent-clipped=0.0 2023-05-16 10:51:50,222 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195667.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 10:52:04,748 INFO [finetune.py:992] (1/2) Epoch 8, batch 10050, loss[loss=0.1714, simple_loss=0.265, pruned_loss=0.03894, over 10443.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2601, pruned_loss=0.0421, over 2364848.22 frames. ], batch size: 68, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:52:14,101 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195700.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 10:52:20,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-05-16 10:52:24,711 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195715.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:52:33,302 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195727.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:52:40,246 INFO [finetune.py:992] (1/2) Epoch 8, batch 10100, loss[loss=0.1776, simple_loss=0.2753, pruned_loss=0.03998, over 12373.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2607, pruned_loss=0.04253, over 2359858.05 frames. ], batch size: 38, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:52:47,961 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195748.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 10:52:47,970 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195748.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:52:50,504 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.087e+02 2.741e+02 3.375e+02 4.030e+02 6.794e+02, threshold=6.750e+02, percent-clipped=1.0 2023-05-16 10:53:06,471 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195775.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:53:15,071 INFO [finetune.py:992] (1/2) Epoch 8, batch 10150, loss[loss=0.1349, simple_loss=0.2128, pruned_loss=0.02847, over 12031.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2611, pruned_loss=0.04246, over 2366483.14 frames. ], batch size: 28, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:53:36,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 10:53:52,386 INFO [finetune.py:992] (1/2) Epoch 8, batch 10200, loss[loss=0.175, simple_loss=0.2718, pruned_loss=0.03908, over 12280.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2608, pruned_loss=0.04213, over 2366116.66 frames. ], batch size: 37, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:53:57,999 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 10:54:03,062 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.806e+02 3.238e+02 4.023e+02 8.056e+02, threshold=6.475e+02, percent-clipped=2.0 2023-05-16 10:54:18,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 10:54:27,949 INFO [finetune.py:992] (1/2) Epoch 8, batch 10250, loss[loss=0.1446, simple_loss=0.2276, pruned_loss=0.03075, over 11995.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2604, pruned_loss=0.0423, over 2374289.34 frames. ], batch size: 28, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:55:03,414 INFO [finetune.py:992] (1/2) Epoch 8, batch 10300, loss[loss=0.1864, simple_loss=0.2788, pruned_loss=0.04696, over 11856.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2602, pruned_loss=0.04202, over 2379601.03 frames. ], batch size: 44, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:55:14,640 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.764e+02 3.180e+02 3.731e+02 8.237e+02, threshold=6.360e+02, percent-clipped=1.0 2023-05-16 10:55:21,570 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9212, 5.9322, 5.6675, 5.1432, 5.1018, 5.7937, 5.4680, 5.1821], device='cuda:1'), covar=tensor([0.0668, 0.0725, 0.0564, 0.1612, 0.0697, 0.0755, 0.1499, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0601, 0.0535, 0.0497, 0.0618, 0.0408, 0.0699, 0.0753, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:55:35,045 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2041, 6.1368, 6.0136, 5.5133, 5.2550, 6.1311, 5.7959, 5.5050], device='cuda:1'), covar=tensor([0.0755, 0.1105, 0.0572, 0.1404, 0.0699, 0.0677, 0.1358, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0601, 0.0536, 0.0496, 0.0618, 0.0408, 0.0698, 0.0753, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:55:40,008 INFO [finetune.py:992] (1/2) Epoch 8, batch 10350, loss[loss=0.1627, simple_loss=0.2567, pruned_loss=0.03439, over 12346.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.26, pruned_loss=0.04182, over 2379550.15 frames. ], batch size: 36, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:56:12,841 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3123, 5.0735, 5.2382, 5.2614, 4.8923, 4.9037, 4.6520, 5.1891], device='cuda:1'), covar=tensor([0.0684, 0.0666, 0.0823, 0.0570, 0.2027, 0.1418, 0.0642, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0680, 0.0591, 0.0609, 0.0831, 0.0718, 0.0536, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:56:18,473 INFO [finetune.py:992] (1/2) Epoch 8, batch 10400, loss[loss=0.1919, simple_loss=0.276, pruned_loss=0.0539, over 12058.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2599, pruned_loss=0.04196, over 2375791.16 frames. ], batch size: 42, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:56:26,286 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196048.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:56:27,034 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2236, 3.1865, 3.1539, 3.5124, 2.5892, 3.1385, 2.5407, 3.0038], device='cuda:1'), covar=tensor([0.1364, 0.0796, 0.0850, 0.0591, 0.0985, 0.0684, 0.1451, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0266, 0.0296, 0.0356, 0.0236, 0.0236, 0.0257, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:56:28,902 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.918e+02 3.403e+02 4.012e+02 6.635e+02, threshold=6.805e+02, percent-clipped=1.0 2023-05-16 10:56:54,490 INFO [finetune.py:992] (1/2) Epoch 8, batch 10450, loss[loss=0.1781, simple_loss=0.2705, pruned_loss=0.04282, over 12155.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2599, pruned_loss=0.04187, over 2376904.49 frames. ], batch size: 34, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:57:00,849 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196096.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:57:18,522 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2643, 5.0895, 5.2260, 5.2551, 4.9015, 4.9351, 4.7119, 5.1908], device='cuda:1'), covar=tensor([0.0750, 0.0562, 0.0766, 0.0559, 0.1760, 0.1307, 0.0571, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0684, 0.0592, 0.0611, 0.0833, 0.0722, 0.0538, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1') 2023-05-16 10:57:30,562 INFO [finetune.py:992] (1/2) Epoch 8, batch 10500, loss[loss=0.1674, simple_loss=0.2645, pruned_loss=0.03508, over 12373.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2602, pruned_loss=0.04207, over 2377313.26 frames. ], batch size: 38, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:57:41,283 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.777e+02 3.207e+02 3.747e+02 7.521e+02, threshold=6.415e+02, percent-clipped=1.0 2023-05-16 10:58:06,361 INFO [finetune.py:992] (1/2) Epoch 8, batch 10550, loss[loss=0.2307, simple_loss=0.3141, pruned_loss=0.07362, over 8203.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2602, pruned_loss=0.04178, over 2375308.09 frames. ], batch size: 98, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:58:28,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 10:58:42,561 INFO [finetune.py:992] (1/2) Epoch 8, batch 10600, loss[loss=0.1731, simple_loss=0.2676, pruned_loss=0.03936, over 12162.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2608, pruned_loss=0.04211, over 2368463.16 frames. ], batch size: 36, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:58:43,525 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6314, 3.2357, 4.9838, 2.5577, 2.7548, 3.6805, 3.1975, 3.7298], device='cuda:1'), covar=tensor([0.0420, 0.1202, 0.0292, 0.1172, 0.1956, 0.1440, 0.1306, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0235, 0.0244, 0.0180, 0.0237, 0.0291, 0.0224, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:58:53,127 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.800e+02 3.385e+02 4.224e+02 7.673e+02, threshold=6.769e+02, percent-clipped=3.0 2023-05-16 10:59:11,870 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0206, 4.8477, 4.9774, 5.0072, 4.6206, 4.6458, 4.4656, 4.9360], device='cuda:1'), covar=tensor([0.0790, 0.0655, 0.0768, 0.0650, 0.1942, 0.1534, 0.0644, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0680, 0.0589, 0.0609, 0.0832, 0.0721, 0.0535, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 10:59:16,343 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1010, 4.4803, 3.9718, 4.7193, 4.2685, 2.8910, 4.1175, 2.9737], device='cuda:1'), covar=tensor([0.0816, 0.0777, 0.1358, 0.0428, 0.1174, 0.1605, 0.0989, 0.3232], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0375, 0.0354, 0.0286, 0.0362, 0.0266, 0.0338, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 10:59:18,309 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9243, 3.5989, 5.2536, 2.6832, 2.8665, 3.9274, 3.3699, 3.9954], device='cuda:1'), covar=tensor([0.0402, 0.1004, 0.0281, 0.1231, 0.1867, 0.1393, 0.1314, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0234, 0.0244, 0.0179, 0.0236, 0.0290, 0.0224, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 10:59:18,770 INFO [finetune.py:992] (1/2) Epoch 8, batch 10650, loss[loss=0.1542, simple_loss=0.2352, pruned_loss=0.03659, over 12352.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2604, pruned_loss=0.04201, over 2367578.31 frames. ], batch size: 31, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 10:59:27,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 10:59:42,631 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196321.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 10:59:53,854 INFO [finetune.py:992] (1/2) Epoch 8, batch 10700, loss[loss=0.1753, simple_loss=0.2683, pruned_loss=0.04115, over 12165.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2606, pruned_loss=0.04194, over 2367945.90 frames. ], batch size: 36, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 11:00:04,520 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.733e+02 3.369e+02 4.182e+02 1.101e+03, threshold=6.737e+02, percent-clipped=2.0 2023-05-16 11:00:10,436 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196360.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:00:26,798 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196382.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:00:30,224 INFO [finetune.py:992] (1/2) Epoch 8, batch 10750, loss[loss=0.1447, simple_loss=0.2303, pruned_loss=0.02951, over 11746.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2601, pruned_loss=0.04119, over 2373688.57 frames. ], batch size: 26, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 11:00:54,692 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196421.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:01:01,454 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 11:01:06,079 INFO [finetune.py:992] (1/2) Epoch 8, batch 10800, loss[loss=0.1767, simple_loss=0.2664, pruned_loss=0.04348, over 12285.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.261, pruned_loss=0.04226, over 2361782.05 frames. ], batch size: 33, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 11:01:16,776 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 2.678e+02 3.444e+02 4.139e+02 7.546e+02, threshold=6.888e+02, percent-clipped=3.0 2023-05-16 11:01:21,665 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8750, 5.5618, 5.1246, 5.1441, 5.6797, 5.0023, 5.2066, 5.1288], device='cuda:1'), covar=tensor([0.1314, 0.1000, 0.1129, 0.1916, 0.0961, 0.2168, 0.1819, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0476, 0.0380, 0.0422, 0.0449, 0.0431, 0.0387, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:01:37,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 11:01:41,720 INFO [finetune.py:992] (1/2) Epoch 8, batch 10850, loss[loss=0.1592, simple_loss=0.2423, pruned_loss=0.03802, over 12287.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2609, pruned_loss=0.04228, over 2366140.70 frames. ], batch size: 28, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 11:01:46,080 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2148, 4.9476, 5.0638, 5.1249, 4.8329, 5.0672, 5.0431, 3.0623], device='cuda:1'), covar=tensor([0.0071, 0.0047, 0.0060, 0.0054, 0.0044, 0.0088, 0.0054, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0079, 0.0072, 0.0058, 0.0089, 0.0079, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:02:09,014 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 11:02:18,940 INFO [finetune.py:992] (1/2) Epoch 8, batch 10900, loss[loss=0.2141, simple_loss=0.2944, pruned_loss=0.06687, over 10620.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2599, pruned_loss=0.04209, over 2365049.15 frames. ], batch size: 68, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 11:02:29,323 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.316e+02 2.881e+02 3.421e+02 4.372e+02 1.882e+03, threshold=6.842e+02, percent-clipped=5.0 2023-05-16 11:02:48,885 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196578.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:02:55,034 INFO [finetune.py:992] (1/2) Epoch 8, batch 10950, loss[loss=0.2219, simple_loss=0.3104, pruned_loss=0.06666, over 10645.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2608, pruned_loss=0.04271, over 2352962.67 frames. ], batch size: 68, lr: 4.24e-03, grad_scale: 8.0 2023-05-16 11:03:14,484 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9689, 5.8686, 5.4203, 5.3638, 5.9504, 5.2498, 5.4155, 5.4561], device='cuda:1'), covar=tensor([0.1555, 0.0915, 0.0912, 0.1863, 0.0879, 0.2085, 0.1789, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0481, 0.0385, 0.0428, 0.0455, 0.0436, 0.0391, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:03:30,774 INFO [finetune.py:992] (1/2) Epoch 8, batch 11000, loss[loss=0.2567, simple_loss=0.3499, pruned_loss=0.08178, over 11024.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2622, pruned_loss=0.04359, over 2335281.77 frames. ], batch size: 55, lr: 4.23e-03, grad_scale: 8.0 2023-05-16 11:03:32,427 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196639.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:03:42,017 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.963e+02 3.550e+02 4.429e+02 1.214e+03, threshold=7.100e+02, percent-clipped=1.0 2023-05-16 11:04:00,590 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196677.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:04:07,484 INFO [finetune.py:992] (1/2) Epoch 8, batch 11050, loss[loss=0.2357, simple_loss=0.3199, pruned_loss=0.07574, over 10340.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2658, pruned_loss=0.04575, over 2307416.24 frames. ], batch size: 68, lr: 4.23e-03, grad_scale: 8.0 2023-05-16 11:04:28,034 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196716.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:04:30,163 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4816, 4.8241, 4.3115, 5.2767, 4.7080, 3.3233, 4.6272, 3.4601], device='cuda:1'), covar=tensor([0.0751, 0.0715, 0.1177, 0.0335, 0.1263, 0.1376, 0.0776, 0.2719], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0376, 0.0354, 0.0285, 0.0363, 0.0266, 0.0338, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:04:42,447 INFO [finetune.py:992] (1/2) Epoch 8, batch 11100, loss[loss=0.145, simple_loss=0.2218, pruned_loss=0.03409, over 12147.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2697, pruned_loss=0.04816, over 2253946.98 frames. ], batch size: 30, lr: 4.23e-03, grad_scale: 8.0 2023-05-16 11:04:53,517 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.108e+02 3.403e+02 4.029e+02 4.819e+02 7.876e+02, threshold=8.057e+02, percent-clipped=4.0 2023-05-16 11:05:13,379 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196779.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:05:15,566 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2500, 4.8375, 4.9706, 5.0574, 4.7368, 5.0936, 4.7930, 2.8883], device='cuda:1'), covar=tensor([0.0080, 0.0064, 0.0076, 0.0057, 0.0047, 0.0090, 0.0095, 0.0678], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0079, 0.0073, 0.0059, 0.0089, 0.0079, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:05:18,881 INFO [finetune.py:992] (1/2) Epoch 8, batch 11150, loss[loss=0.2643, simple_loss=0.3322, pruned_loss=0.09817, over 6825.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.2755, pruned_loss=0.05161, over 2201718.53 frames. ], batch size: 98, lr: 4.23e-03, grad_scale: 8.0 2023-05-16 11:05:45,738 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4406, 4.4290, 4.2948, 3.9604, 4.1137, 4.3900, 4.1731, 4.0056], device='cuda:1'), covar=tensor([0.0825, 0.0945, 0.0671, 0.1336, 0.2024, 0.0963, 0.1412, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0523, 0.0482, 0.0601, 0.0395, 0.0675, 0.0729, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 11:05:54,740 INFO [finetune.py:992] (1/2) Epoch 8, batch 11200, loss[loss=0.3401, simple_loss=0.3822, pruned_loss=0.149, over 6458.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2827, pruned_loss=0.05631, over 2137090.76 frames. ], batch size: 98, lr: 4.23e-03, grad_scale: 8.0 2023-05-16 11:05:56,979 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196840.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:06:05,863 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.285e+02 3.531e+02 4.319e+02 5.158e+02 1.234e+03, threshold=8.638e+02, percent-clipped=4.0 2023-05-16 11:06:06,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 2023-05-16 11:06:07,468 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:06:26,329 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196881.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:06:30,315 INFO [finetune.py:992] (1/2) Epoch 8, batch 11250, loss[loss=0.1806, simple_loss=0.2642, pruned_loss=0.04844, over 12343.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2899, pruned_loss=0.06133, over 2077537.94 frames. ], batch size: 31, lr: 4.23e-03, grad_scale: 8.0 2023-05-16 11:06:50,205 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196915.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:07:01,294 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2313, 1.9278, 2.3460, 2.1657, 2.2420, 2.3072, 1.7643, 2.2928], device='cuda:1'), covar=tensor([0.0094, 0.0263, 0.0106, 0.0147, 0.0130, 0.0108, 0.0235, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0197, 0.0176, 0.0172, 0.0201, 0.0150, 0.0185, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:07:03,856 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196934.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:07:05,774 INFO [finetune.py:992] (1/2) Epoch 8, batch 11300, loss[loss=0.2794, simple_loss=0.3459, pruned_loss=0.1064, over 6697.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2958, pruned_loss=0.06527, over 2023517.79 frames. ], batch size: 99, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:07:09,295 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196942.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:07:15,918 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.372e+02 3.452e+02 4.076e+02 5.044e+02 8.286e+02, threshold=8.152e+02, percent-clipped=0.0 2023-05-16 11:07:30,539 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3284, 5.1389, 5.2371, 5.2998, 4.9664, 4.9408, 4.7585, 5.1573], device='cuda:1'), covar=tensor([0.0727, 0.0603, 0.0726, 0.0571, 0.1827, 0.1384, 0.0561, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0660, 0.0571, 0.0589, 0.0800, 0.0695, 0.0519, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:07:34,113 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196977.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:07:37,485 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0973, 3.9030, 2.6555, 2.0726, 3.5909, 2.1168, 3.6989, 2.6133], device='cuda:1'), covar=tensor([0.0575, 0.0486, 0.0944, 0.1936, 0.0234, 0.1625, 0.0359, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0242, 0.0168, 0.0193, 0.0133, 0.0175, 0.0187, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:07:41,211 INFO [finetune.py:992] (1/2) Epoch 8, batch 11350, loss[loss=0.2675, simple_loss=0.3377, pruned_loss=0.09869, over 10462.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.3005, pruned_loss=0.06803, over 1987380.77 frames. ], batch size: 69, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:08:01,019 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197016.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:08:07,153 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197025.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:08:07,177 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7090, 4.4111, 3.9989, 4.1522, 4.4702, 3.9808, 4.1555, 3.9286], device='cuda:1'), covar=tensor([0.1522, 0.0991, 0.1483, 0.1635, 0.1056, 0.1901, 0.1540, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0466, 0.0376, 0.0414, 0.0442, 0.0422, 0.0377, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:08:15,531 INFO [finetune.py:992] (1/2) Epoch 8, batch 11400, loss[loss=0.2633, simple_loss=0.3225, pruned_loss=0.1021, over 7259.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3055, pruned_loss=0.07175, over 1927042.33 frames. ], batch size: 98, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:08:20,334 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9868, 2.2029, 2.6113, 3.0478, 2.2330, 3.1224, 2.8965, 3.1227], device='cuda:1'), covar=tensor([0.0139, 0.1086, 0.0454, 0.0159, 0.1111, 0.0228, 0.0290, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0197, 0.0177, 0.0111, 0.0184, 0.0169, 0.0166, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:08:25,272 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.064e+02 3.567e+02 4.120e+02 4.999e+02 1.164e+03, threshold=8.240e+02, percent-clipped=1.0 2023-05-16 11:08:34,357 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197064.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:08:49,873 INFO [finetune.py:992] (1/2) Epoch 8, batch 11450, loss[loss=0.2263, simple_loss=0.3085, pruned_loss=0.072, over 10460.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3084, pruned_loss=0.07396, over 1901988.55 frames. ], batch size: 69, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:09:22,250 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1862, 1.8983, 2.2824, 2.1524, 2.1973, 2.3003, 1.8365, 2.2424], device='cuda:1'), covar=tensor([0.0103, 0.0277, 0.0108, 0.0159, 0.0139, 0.0132, 0.0234, 0.0125], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0194, 0.0172, 0.0169, 0.0197, 0.0148, 0.0182, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:09:23,447 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197135.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:09:24,663 INFO [finetune.py:992] (1/2) Epoch 8, batch 11500, loss[loss=0.2297, simple_loss=0.3184, pruned_loss=0.07051, over 6998.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3121, pruned_loss=0.07705, over 1838224.32 frames. ], batch size: 102, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:09:34,834 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 3.515e+02 4.207e+02 4.859e+02 7.657e+02, threshold=8.415e+02, percent-clipped=0.0 2023-05-16 11:10:00,157 INFO [finetune.py:992] (1/2) Epoch 8, batch 11550, loss[loss=0.2416, simple_loss=0.3081, pruned_loss=0.08759, over 6877.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3128, pruned_loss=0.07782, over 1813160.86 frames. ], batch size: 99, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:10:16,013 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197210.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:10:29,842 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6561, 4.0188, 3.4924, 4.2681, 3.8664, 2.7262, 3.7230, 2.8800], device='cuda:1'), covar=tensor([0.0949, 0.0899, 0.1528, 0.0478, 0.1239, 0.1728, 0.1111, 0.3338], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0354, 0.0333, 0.0265, 0.0343, 0.0254, 0.0320, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:10:33,004 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197234.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:10:34,860 INFO [finetune.py:992] (1/2) Epoch 8, batch 11600, loss[loss=0.1998, simple_loss=0.2847, pruned_loss=0.05743, over 11482.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3134, pruned_loss=0.07866, over 1805595.89 frames. ], batch size: 48, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:10:34,988 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197237.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:10:45,625 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.588e+02 3.360e+02 3.974e+02 4.737e+02 1.581e+03, threshold=7.948e+02, percent-clipped=1.0 2023-05-16 11:11:07,459 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:11:11,476 INFO [finetune.py:992] (1/2) Epoch 8, batch 11650, loss[loss=0.3052, simple_loss=0.3644, pruned_loss=0.123, over 7151.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3141, pruned_loss=0.07985, over 1783198.56 frames. ], batch size: 98, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:11:45,976 INFO [finetune.py:992] (1/2) Epoch 8, batch 11700, loss[loss=0.2383, simple_loss=0.3211, pruned_loss=0.07771, over 12296.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3145, pruned_loss=0.08076, over 1757706.44 frames. ], batch size: 34, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:11:51,468 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4606, 2.6666, 3.9091, 2.3568, 2.5326, 3.3706, 2.7598, 3.3239], device='cuda:1'), covar=tensor([0.0545, 0.1268, 0.0248, 0.1279, 0.1657, 0.1025, 0.1350, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0224, 0.0226, 0.0173, 0.0225, 0.0273, 0.0211, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:11:56,332 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.471e+02 3.569e+02 3.981e+02 5.157e+02 1.224e+03, threshold=7.962e+02, percent-clipped=1.0 2023-05-16 11:12:20,623 INFO [finetune.py:992] (1/2) Epoch 8, batch 11750, loss[loss=0.264, simple_loss=0.3468, pruned_loss=0.09059, over 10423.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3157, pruned_loss=0.08212, over 1737695.13 frames. ], batch size: 68, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:12:40,324 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197415.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:12:52,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 11:12:54,424 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197435.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:12:55,538 INFO [finetune.py:992] (1/2) Epoch 8, batch 11800, loss[loss=0.2563, simple_loss=0.3136, pruned_loss=0.09953, over 6977.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3174, pruned_loss=0.08387, over 1709458.75 frames. ], batch size: 99, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:13:04,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 11:13:05,853 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.262e+02 3.453e+02 4.425e+02 5.306e+02 1.147e+03, threshold=8.850e+02, percent-clipped=3.0 2023-05-16 11:13:23,046 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197476.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:13:27,578 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197483.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:13:30,161 INFO [finetune.py:992] (1/2) Epoch 8, batch 11850, loss[loss=0.2264, simple_loss=0.3188, pruned_loss=0.06703, over 10332.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3181, pruned_loss=0.08421, over 1679283.53 frames. ], batch size: 68, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:13:46,329 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197510.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:13:50,356 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197516.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:14:05,197 INFO [finetune.py:992] (1/2) Epoch 8, batch 11900, loss[loss=0.1893, simple_loss=0.2888, pruned_loss=0.04487, over 11236.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3181, pruned_loss=0.0833, over 1666224.68 frames. ], batch size: 55, lr: 4.23e-03, grad_scale: 16.0 2023-05-16 11:14:05,401 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197537.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:14:10,811 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197545.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:14:16,131 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.200e+02 3.439e+02 3.960e+02 4.770e+02 1.222e+03, threshold=7.919e+02, percent-clipped=4.0 2023-05-16 11:14:20,175 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197558.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:14:33,112 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197577.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:14:38,964 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197585.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:14:40,174 INFO [finetune.py:992] (1/2) Epoch 8, batch 11950, loss[loss=0.1953, simple_loss=0.2754, pruned_loss=0.05757, over 7299.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3152, pruned_loss=0.08088, over 1656823.54 frames. ], batch size: 98, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:14:53,522 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197606.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:15:07,897 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4571, 3.1945, 3.1367, 3.2833, 2.7112, 3.1564, 2.5478, 2.7584], device='cuda:1'), covar=tensor([0.1644, 0.0918, 0.0984, 0.0721, 0.1057, 0.0816, 0.1945, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0259, 0.0285, 0.0341, 0.0231, 0.0230, 0.0256, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:15:15,136 INFO [finetune.py:992] (1/2) Epoch 8, batch 12000, loss[loss=0.2184, simple_loss=0.2905, pruned_loss=0.07313, over 7112.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3095, pruned_loss=0.07632, over 1668390.82 frames. ], batch size: 99, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:15:15,137 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 11:15:33,072 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8564, 5.6405, 5.6412, 5.4172, 4.8066, 5.7358, 5.2976, 5.3045], device='cuda:1'), covar=tensor([0.0436, 0.0944, 0.0525, 0.1161, 0.0522, 0.0612, 0.1510, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0548, 0.0494, 0.0452, 0.0559, 0.0372, 0.0629, 0.0674, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 11:15:34,456 INFO [finetune.py:1026] (1/2) Epoch 8, validation: loss=0.2946, simple_loss=0.3674, pruned_loss=0.1109, over 1020973.00 frames. 2023-05-16 11:15:34,457 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 11:15:37,269 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9182, 3.8808, 3.8772, 3.9629, 3.7743, 3.7687, 3.7393, 3.8555], device='cuda:1'), covar=tensor([0.0933, 0.0677, 0.0933, 0.0651, 0.1501, 0.1089, 0.0489, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0622, 0.0540, 0.0551, 0.0739, 0.0653, 0.0486, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:15:44,912 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.171e+02 3.035e+02 3.587e+02 4.219e+02 7.957e+02, threshold=7.174e+02, percent-clipped=2.0 2023-05-16 11:16:09,219 INFO [finetune.py:992] (1/2) Epoch 8, batch 12050, loss[loss=0.2201, simple_loss=0.3008, pruned_loss=0.06975, over 7267.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3052, pruned_loss=0.07296, over 1661815.92 frames. ], batch size: 99, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:16:42,024 INFO [finetune.py:992] (1/2) Epoch 8, batch 12100, loss[loss=0.2553, simple_loss=0.3217, pruned_loss=0.09449, over 7334.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3039, pruned_loss=0.07161, over 1680400.99 frames. ], batch size: 104, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:16:44,211 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7796, 2.9715, 4.1505, 4.4276, 3.0358, 2.7268, 2.8788, 1.9962], device='cuda:1'), covar=tensor([0.1391, 0.2408, 0.0465, 0.0326, 0.1116, 0.2206, 0.2514, 0.4500], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0360, 0.0258, 0.0278, 0.0248, 0.0282, 0.0351, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:16:51,590 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 3.331e+02 3.770e+02 4.406e+02 9.373e+02, threshold=7.541e+02, percent-clipped=1.0 2023-05-16 11:17:03,875 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197771.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:17:13,970 INFO [finetune.py:992] (1/2) Epoch 8, batch 12150, loss[loss=0.2194, simple_loss=0.3073, pruned_loss=0.0658, over 11751.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3049, pruned_loss=0.07206, over 1686624.53 frames. ], batch size: 44, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:17:45,952 INFO [finetune.py:992] (1/2) Epoch 8, batch 12200, loss[loss=0.266, simple_loss=0.3276, pruned_loss=0.1022, over 6837.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3059, pruned_loss=0.07321, over 1675052.59 frames. ], batch size: 100, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:17:55,228 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.087e+02 3.794e+02 4.591e+02 1.477e+03, threshold=7.588e+02, percent-clipped=2.0 2023-05-16 11:18:00,291 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7051, 3.4752, 3.5486, 3.6491, 3.5519, 3.7230, 3.5579, 2.6656], device='cuda:1'), covar=tensor([0.0086, 0.0076, 0.0118, 0.0074, 0.0066, 0.0104, 0.0076, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0069, 0.0072, 0.0065, 0.0052, 0.0080, 0.0070, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:18:03,473 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8212, 2.3817, 3.5477, 3.6277, 2.7979, 2.6830, 2.5685, 2.3578], device='cuda:1'), covar=tensor([0.1251, 0.2893, 0.0590, 0.0464, 0.0990, 0.2090, 0.2985, 0.3901], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0361, 0.0258, 0.0278, 0.0249, 0.0283, 0.0352, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:18:32,505 INFO [finetune.py:992] (1/2) Epoch 9, batch 0, loss[loss=0.1888, simple_loss=0.2813, pruned_loss=0.04818, over 12179.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2813, pruned_loss=0.04818, over 12179.00 frames. ], batch size: 36, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:18:32,505 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 11:18:44,353 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4602, 4.3685, 4.4850, 4.4640, 4.2309, 4.1966, 4.1680, 4.3058], device='cuda:1'), covar=tensor([0.0783, 0.0574, 0.0765, 0.0649, 0.1423, 0.1434, 0.0571, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0610, 0.0529, 0.0541, 0.0723, 0.0641, 0.0476, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:18:49,800 INFO [finetune.py:1026] (1/2) Epoch 9, validation: loss=0.2943, simple_loss=0.3667, pruned_loss=0.111, over 1020973.00 frames. 2023-05-16 11:18:49,801 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 11:18:50,591 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197872.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:19:05,315 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197892.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:19:11,814 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197901.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:19:25,828 INFO [finetune.py:992] (1/2) Epoch 9, batch 50, loss[loss=0.1654, simple_loss=0.2504, pruned_loss=0.04026, over 12112.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2676, pruned_loss=0.04434, over 540881.95 frames. ], batch size: 32, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:19:47,775 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.989e+02 3.499e+02 4.265e+02 8.208e+02, threshold=6.998e+02, percent-clipped=1.0 2023-05-16 11:19:48,632 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197953.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:20:01,266 INFO [finetune.py:992] (1/2) Epoch 9, batch 100, loss[loss=0.1803, simple_loss=0.269, pruned_loss=0.04576, over 12140.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2674, pruned_loss=0.04499, over 946715.74 frames. ], batch size: 36, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:20:25,398 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5945, 4.2929, 4.2110, 4.3394, 4.2391, 4.5836, 4.3627, 2.5036], device='cuda:1'), covar=tensor([0.0154, 0.0097, 0.0154, 0.0111, 0.0083, 0.0125, 0.0140, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0070, 0.0073, 0.0066, 0.0054, 0.0082, 0.0072, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:20:40,196 INFO [finetune.py:992] (1/2) Epoch 9, batch 150, loss[loss=0.166, simple_loss=0.2496, pruned_loss=0.04123, over 11369.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2672, pruned_loss=0.04481, over 1263590.21 frames. ], batch size: 25, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:21:02,897 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 2.862e+02 3.446e+02 4.094e+02 7.701e+02, threshold=6.892e+02, percent-clipped=0.0 2023-05-16 11:21:16,477 INFO [finetune.py:992] (1/2) Epoch 9, batch 200, loss[loss=0.1979, simple_loss=0.2924, pruned_loss=0.05169, over 12053.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2658, pruned_loss=0.04393, over 1520231.36 frames. ], batch size: 40, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:21:16,579 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198071.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:21:23,784 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198081.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:21:50,763 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198119.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:21:52,007 INFO [finetune.py:992] (1/2) Epoch 9, batch 250, loss[loss=0.1696, simple_loss=0.2576, pruned_loss=0.04078, over 12112.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2629, pruned_loss=0.04261, over 1714493.68 frames. ], batch size: 33, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:22:02,810 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0618, 5.0863, 4.8639, 4.8781, 4.6504, 5.0409, 5.0864, 5.2793], device='cuda:1'), covar=tensor([0.0303, 0.0134, 0.0216, 0.0333, 0.0745, 0.0244, 0.0138, 0.0175], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0173, 0.0169, 0.0218, 0.0214, 0.0192, 0.0155, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 11:22:07,800 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198142.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:22:14,503 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.981e+02 2.780e+02 3.190e+02 3.913e+02 1.023e+03, threshold=6.381e+02, percent-clipped=3.0 2023-05-16 11:22:27,858 INFO [finetune.py:992] (1/2) Epoch 9, batch 300, loss[loss=0.1838, simple_loss=0.2694, pruned_loss=0.04909, over 12054.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2626, pruned_loss=0.04249, over 1864858.69 frames. ], batch size: 42, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:22:28,279 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-05-16 11:22:28,778 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198172.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:22:39,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-05-16 11:22:49,861 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198201.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:23:04,141 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198220.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:23:04,726 INFO [finetune.py:992] (1/2) Epoch 9, batch 350, loss[loss=0.2006, simple_loss=0.2841, pruned_loss=0.05849, over 12063.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2616, pruned_loss=0.04216, over 1987796.19 frames. ], batch size: 40, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:23:24,061 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198248.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:23:24,794 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198249.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:23:26,766 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.757e+02 3.319e+02 3.988e+02 6.623e+02, threshold=6.637e+02, percent-clipped=1.0 2023-05-16 11:23:41,081 INFO [finetune.py:992] (1/2) Epoch 9, batch 400, loss[loss=0.1418, simple_loss=0.22, pruned_loss=0.03185, over 11837.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2614, pruned_loss=0.04218, over 2070076.43 frames. ], batch size: 26, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:23:54,782 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9872, 4.6748, 4.6522, 4.7568, 4.5251, 4.8888, 4.6542, 2.3346], device='cuda:1'), covar=tensor([0.0118, 0.0069, 0.0110, 0.0068, 0.0058, 0.0088, 0.0096, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0073, 0.0076, 0.0068, 0.0055, 0.0084, 0.0074, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:24:16,780 INFO [finetune.py:992] (1/2) Epoch 9, batch 450, loss[loss=0.1594, simple_loss=0.2406, pruned_loss=0.03913, over 12356.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2617, pruned_loss=0.04231, over 2140385.02 frames. ], batch size: 30, lr: 4.22e-03, grad_scale: 16.0 2023-05-16 11:24:39,782 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.276e+02 2.819e+02 3.326e+02 4.197e+02 7.125e+02, threshold=6.653e+02, percent-clipped=2.0 2023-05-16 11:24:52,703 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2636, 3.9577, 4.0665, 4.6307, 3.3057, 4.0427, 2.6743, 4.2689], device='cuda:1'), covar=tensor([0.1709, 0.0878, 0.1152, 0.0610, 0.1060, 0.0599, 0.1938, 0.1568], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0266, 0.0293, 0.0351, 0.0236, 0.0235, 0.0262, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:24:53,169 INFO [finetune.py:992] (1/2) Epoch 9, batch 500, loss[loss=0.1554, simple_loss=0.2271, pruned_loss=0.04185, over 12127.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2606, pruned_loss=0.0413, over 2198287.55 frames. ], batch size: 30, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:24:58,282 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9929, 5.9936, 5.7266, 5.2603, 5.1255, 5.8753, 5.5585, 5.2188], device='cuda:1'), covar=tensor([0.0724, 0.0835, 0.0607, 0.1661, 0.0646, 0.0764, 0.1448, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0511, 0.0465, 0.0581, 0.0381, 0.0652, 0.0701, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:25:08,415 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5290, 2.6688, 4.3763, 4.5263, 2.7641, 2.4598, 2.7254, 1.9513], device='cuda:1'), covar=tensor([0.1644, 0.3289, 0.0489, 0.0435, 0.1267, 0.2453, 0.2877, 0.4382], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0370, 0.0264, 0.0285, 0.0255, 0.0288, 0.0359, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:25:21,553 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8612, 4.4819, 4.8517, 4.2649, 4.4863, 4.3166, 4.8488, 4.5538], device='cuda:1'), covar=tensor([0.0286, 0.0400, 0.0281, 0.0288, 0.0421, 0.0344, 0.0239, 0.0479], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0246, 0.0265, 0.0243, 0.0242, 0.0241, 0.0222, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:25:29,266 INFO [finetune.py:992] (1/2) Epoch 9, batch 550, loss[loss=0.1586, simple_loss=0.2491, pruned_loss=0.03406, over 12169.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2607, pruned_loss=0.04125, over 2240669.28 frames. ], batch size: 31, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:25:41,410 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198437.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:25:52,023 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.772e+02 3.197e+02 3.691e+02 5.962e+02, threshold=6.394e+02, percent-clipped=0.0 2023-05-16 11:26:03,135 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7382, 2.6137, 3.8103, 4.6913, 4.1253, 4.6713, 3.9734, 3.5088], device='cuda:1'), covar=tensor([0.0023, 0.0399, 0.0103, 0.0031, 0.0088, 0.0061, 0.0094, 0.0293], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0120, 0.0101, 0.0072, 0.0097, 0.0114, 0.0092, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:26:05,769 INFO [finetune.py:992] (1/2) Epoch 9, batch 600, loss[loss=0.1929, simple_loss=0.2821, pruned_loss=0.05187, over 12121.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2614, pruned_loss=0.04152, over 2265765.88 frames. ], batch size: 38, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:26:09,610 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6982, 2.6842, 3.4391, 4.6835, 2.6354, 4.5025, 4.6398, 4.7131], device='cuda:1'), covar=tensor([0.0118, 0.1150, 0.0392, 0.0117, 0.1253, 0.0221, 0.0115, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0196, 0.0176, 0.0109, 0.0186, 0.0169, 0.0164, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:26:26,870 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198499.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:26:42,401 INFO [finetune.py:992] (1/2) Epoch 9, batch 650, loss[loss=0.1672, simple_loss=0.2549, pruned_loss=0.03975, over 10407.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2612, pruned_loss=0.04144, over 2285418.14 frames. ], batch size: 69, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:26:56,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-05-16 11:27:01,655 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198548.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:27:04,373 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.661e+02 3.289e+02 3.726e+02 5.491e+02, threshold=6.578e+02, percent-clipped=0.0 2023-05-16 11:27:10,249 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198560.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:27:17,795 INFO [finetune.py:992] (1/2) Epoch 9, batch 700, loss[loss=0.1827, simple_loss=0.2669, pruned_loss=0.04923, over 12154.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2609, pruned_loss=0.04148, over 2314412.98 frames. ], batch size: 36, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:27:35,944 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198596.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:27:49,910 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:27:53,995 INFO [finetune.py:992] (1/2) Epoch 9, batch 750, loss[loss=0.165, simple_loss=0.2471, pruned_loss=0.0414, over 12177.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2608, pruned_loss=0.04131, over 2326629.18 frames. ], batch size: 31, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:28:09,746 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7930, 3.6333, 3.3522, 3.3119, 3.1089, 2.9676, 3.7349, 2.3359], device='cuda:1'), covar=tensor([0.0317, 0.0132, 0.0172, 0.0173, 0.0326, 0.0306, 0.0096, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0156, 0.0151, 0.0179, 0.0196, 0.0192, 0.0161, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:28:16,592 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.769e+02 3.237e+02 3.971e+02 5.764e+02, threshold=6.473e+02, percent-clipped=0.0 2023-05-16 11:28:30,061 INFO [finetune.py:992] (1/2) Epoch 9, batch 800, loss[loss=0.1842, simple_loss=0.2683, pruned_loss=0.05009, over 12274.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2602, pruned_loss=0.04107, over 2342690.21 frames. ], batch size: 37, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:28:33,735 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198676.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:28:40,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 11:29:05,764 INFO [finetune.py:992] (1/2) Epoch 9, batch 850, loss[loss=0.1793, simple_loss=0.2743, pruned_loss=0.04212, over 10519.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2601, pruned_loss=0.04113, over 2351274.37 frames. ], batch size: 68, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:29:17,321 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198737.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:29:27,744 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.779e+02 3.276e+02 3.903e+02 7.256e+02, threshold=6.551e+02, percent-clipped=1.0 2023-05-16 11:29:41,439 INFO [finetune.py:992] (1/2) Epoch 9, batch 900, loss[loss=0.1601, simple_loss=0.2397, pruned_loss=0.04028, over 12341.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2596, pruned_loss=0.04103, over 2355412.67 frames. ], batch size: 30, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:29:51,554 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198785.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:30:18,286 INFO [finetune.py:992] (1/2) Epoch 9, batch 950, loss[loss=0.1774, simple_loss=0.2719, pruned_loss=0.04148, over 12349.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2599, pruned_loss=0.04107, over 2359185.05 frames. ], batch size: 36, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:30:40,146 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.863e+02 3.440e+02 4.069e+02 1.108e+03, threshold=6.879e+02, percent-clipped=2.0 2023-05-16 11:30:40,414 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198852.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:30:42,384 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198855.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:30:54,257 INFO [finetune.py:992] (1/2) Epoch 9, batch 1000, loss[loss=0.1838, simple_loss=0.2563, pruned_loss=0.05561, over 12126.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2593, pruned_loss=0.04072, over 2367510.78 frames. ], batch size: 30, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:31:02,908 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198883.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:31:24,252 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198913.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:31:27,121 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1606, 6.0923, 5.8508, 5.3814, 5.1349, 6.0141, 5.6700, 5.4296], device='cuda:1'), covar=tensor([0.0657, 0.1050, 0.0659, 0.1606, 0.0702, 0.0834, 0.1725, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0523, 0.0480, 0.0600, 0.0393, 0.0675, 0.0732, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 11:31:29,889 INFO [finetune.py:992] (1/2) Epoch 9, batch 1050, loss[loss=0.191, simple_loss=0.2798, pruned_loss=0.05109, over 11755.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2601, pruned_loss=0.04102, over 2370307.25 frames. ], batch size: 48, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:31:34,286 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198927.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:31:39,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-16 11:31:46,140 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198944.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:31:52,313 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.715e+02 3.357e+02 3.940e+02 9.082e+02, threshold=6.713e+02, percent-clipped=1.0 2023-05-16 11:32:03,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 11:32:05,600 INFO [finetune.py:992] (1/2) Epoch 9, batch 1100, loss[loss=0.1596, simple_loss=0.2484, pruned_loss=0.03543, over 12107.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2606, pruned_loss=0.04143, over 2367593.19 frames. ], batch size: 33, lr: 4.21e-03, grad_scale: 32.0 2023-05-16 11:32:05,684 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198971.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:32:17,944 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198988.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:32:30,956 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0385, 4.6921, 5.0785, 4.3277, 4.7898, 4.3762, 4.9867, 4.8698], device='cuda:1'), covar=tensor([0.0370, 0.0403, 0.0348, 0.0338, 0.0335, 0.0431, 0.0398, 0.0327], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0252, 0.0272, 0.0248, 0.0247, 0.0248, 0.0226, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:32:41,474 INFO [finetune.py:992] (1/2) Epoch 9, batch 1150, loss[loss=0.1624, simple_loss=0.2477, pruned_loss=0.03852, over 12289.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2603, pruned_loss=0.04118, over 2374077.66 frames. ], batch size: 33, lr: 4.21e-03, grad_scale: 32.0 2023-05-16 11:32:51,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4308, 3.4590, 3.1343, 3.1014, 2.7656, 2.6429, 3.4602, 2.0066], device='cuda:1'), covar=tensor([0.0350, 0.0131, 0.0194, 0.0199, 0.0357, 0.0352, 0.0120, 0.0513], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0156, 0.0151, 0.0179, 0.0197, 0.0192, 0.0160, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:33:05,172 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.724e+02 3.258e+02 3.920e+02 1.307e+03, threshold=6.516e+02, percent-clipped=3.0 2023-05-16 11:33:15,783 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7672, 5.5907, 5.1293, 5.1451, 5.6970, 4.9727, 5.2800, 5.1898], device='cuda:1'), covar=tensor([0.1398, 0.0929, 0.1111, 0.2036, 0.0856, 0.2164, 0.1532, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0464, 0.0378, 0.0422, 0.0443, 0.0424, 0.0379, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:33:17,847 INFO [finetune.py:992] (1/2) Epoch 9, batch 1200, loss[loss=0.1861, simple_loss=0.2752, pruned_loss=0.04851, over 12355.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2604, pruned_loss=0.04127, over 2373674.77 frames. ], batch size: 38, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:33:53,889 INFO [finetune.py:992] (1/2) Epoch 9, batch 1250, loss[loss=0.1499, simple_loss=0.2323, pruned_loss=0.03379, over 11824.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2604, pruned_loss=0.04143, over 2373092.14 frames. ], batch size: 26, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:34:12,596 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9525, 5.9484, 5.7047, 5.2034, 5.0266, 5.8256, 5.4780, 5.1913], device='cuda:1'), covar=tensor([0.0832, 0.0922, 0.0639, 0.1565, 0.0735, 0.0812, 0.1475, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0526, 0.0481, 0.0601, 0.0395, 0.0677, 0.0732, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 11:34:16,574 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.725e+02 3.192e+02 3.846e+02 8.183e+02, threshold=6.384e+02, percent-clipped=4.0 2023-05-16 11:34:18,133 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199155.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:34:30,043 INFO [finetune.py:992] (1/2) Epoch 9, batch 1300, loss[loss=0.2104, simple_loss=0.2851, pruned_loss=0.0678, over 7397.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2597, pruned_loss=0.04135, over 2378578.41 frames. ], batch size: 98, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:34:33,781 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199176.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:34:52,480 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.40 vs. limit=5.0 2023-05-16 11:34:52,899 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199203.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:34:56,515 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199208.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:35:05,596 INFO [finetune.py:992] (1/2) Epoch 9, batch 1350, loss[loss=0.1808, simple_loss=0.2739, pruned_loss=0.04382, over 12024.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2594, pruned_loss=0.04111, over 2376064.04 frames. ], batch size: 42, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:35:16,965 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199237.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:35:18,846 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199239.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:35:23,795 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1286, 2.5380, 3.6762, 3.1601, 3.5169, 3.2134, 2.6194, 3.6164], device='cuda:1'), covar=tensor([0.0150, 0.0330, 0.0126, 0.0218, 0.0138, 0.0170, 0.0370, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0199, 0.0179, 0.0178, 0.0206, 0.0155, 0.0192, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:35:28,441 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.776e+02 3.275e+02 3.775e+02 9.733e+02, threshold=6.551e+02, percent-clipped=4.0 2023-05-16 11:35:41,237 INFO [finetune.py:992] (1/2) Epoch 9, batch 1400, loss[loss=0.1882, simple_loss=0.2819, pruned_loss=0.04724, over 11305.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2596, pruned_loss=0.04128, over 2369136.27 frames. ], batch size: 55, lr: 4.21e-03, grad_scale: 16.0 2023-05-16 11:35:41,351 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199271.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:35:49,898 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199283.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:35:52,278 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6635, 2.6363, 4.0760, 4.3062, 2.8885, 2.5759, 2.6869, 2.1112], device='cuda:1'), covar=tensor([0.1492, 0.2966, 0.0560, 0.0466, 0.1180, 0.2217, 0.2851, 0.4050], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0375, 0.0267, 0.0288, 0.0259, 0.0292, 0.0364, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:36:02,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 11:36:16,312 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199319.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:36:17,646 INFO [finetune.py:992] (1/2) Epoch 9, batch 1450, loss[loss=0.1717, simple_loss=0.2634, pruned_loss=0.04001, over 10337.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2596, pruned_loss=0.04137, over 2367492.25 frames. ], batch size: 68, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:36:20,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 11:36:31,457 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4638, 2.7433, 3.2261, 4.4190, 2.3955, 4.3708, 4.3727, 4.5543], device='cuda:1'), covar=tensor([0.0121, 0.1025, 0.0488, 0.0163, 0.1292, 0.0199, 0.0156, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0181, 0.0115, 0.0190, 0.0175, 0.0171, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:36:40,396 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.705e+02 3.180e+02 3.708e+02 6.435e+02, threshold=6.359e+02, percent-clipped=0.0 2023-05-16 11:36:53,056 INFO [finetune.py:992] (1/2) Epoch 9, batch 1500, loss[loss=0.1643, simple_loss=0.2616, pruned_loss=0.03348, over 12362.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2599, pruned_loss=0.04132, over 2371612.91 frames. ], batch size: 35, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:37:14,582 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199400.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:37:29,782 INFO [finetune.py:992] (1/2) Epoch 9, batch 1550, loss[loss=0.1549, simple_loss=0.2388, pruned_loss=0.03549, over 12332.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2595, pruned_loss=0.04097, over 2374306.86 frames. ], batch size: 30, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:37:46,482 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5572, 2.7001, 3.2445, 4.4858, 2.4525, 4.5594, 4.5257, 4.6733], device='cuda:1'), covar=tensor([0.0149, 0.1187, 0.0498, 0.0160, 0.1382, 0.0185, 0.0136, 0.0087], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0200, 0.0180, 0.0114, 0.0189, 0.0174, 0.0170, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:37:46,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-16 11:37:52,558 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.003e+02 2.743e+02 3.183e+02 3.898e+02 7.323e+02, threshold=6.366e+02, percent-clipped=2.0 2023-05-16 11:37:56,389 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199458.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:37:58,492 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199461.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:37:59,126 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4626, 4.8312, 2.9324, 2.8272, 4.1575, 2.6158, 4.1780, 3.2675], device='cuda:1'), covar=tensor([0.0589, 0.0384, 0.1021, 0.1381, 0.0260, 0.1358, 0.0434, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0248, 0.0175, 0.0199, 0.0138, 0.0181, 0.0193, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:38:03,785 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 11:38:06,001 INFO [finetune.py:992] (1/2) Epoch 9, batch 1600, loss[loss=0.1834, simple_loss=0.2712, pruned_loss=0.04781, over 12314.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2603, pruned_loss=0.04123, over 2373972.27 frames. ], batch size: 34, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:38:14,749 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4965, 4.7921, 3.0174, 2.7117, 4.1424, 2.6004, 4.2142, 3.2622], device='cuda:1'), covar=tensor([0.0634, 0.0548, 0.1096, 0.1466, 0.0323, 0.1346, 0.0430, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0249, 0.0175, 0.0199, 0.0138, 0.0182, 0.0193, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:38:18,946 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6612, 4.6213, 4.4984, 4.1101, 4.2407, 4.5998, 4.3551, 4.1060], device='cuda:1'), covar=tensor([0.0845, 0.0901, 0.0682, 0.1603, 0.1809, 0.0882, 0.1367, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0529, 0.0485, 0.0609, 0.0398, 0.0683, 0.0738, 0.0547], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 11:38:22,584 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199494.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:38:32,482 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199508.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:38:40,514 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199519.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:38:41,663 INFO [finetune.py:992] (1/2) Epoch 9, batch 1650, loss[loss=0.1708, simple_loss=0.2552, pruned_loss=0.04316, over 12187.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2605, pruned_loss=0.04124, over 2375440.04 frames. ], batch size: 29, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:38:49,599 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199532.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:38:52,399 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199535.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:38:55,101 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199539.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:39:04,803 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.622e+02 3.079e+02 3.563e+02 8.648e+02, threshold=6.157e+02, percent-clipped=1.0 2023-05-16 11:39:06,442 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199555.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:39:06,978 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199556.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:39:17,503 INFO [finetune.py:992] (1/2) Epoch 9, batch 1700, loss[loss=0.1662, simple_loss=0.2595, pruned_loss=0.0365, over 12191.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2613, pruned_loss=0.04156, over 2377650.85 frames. ], batch size: 35, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:39:20,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 11:39:26,143 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199583.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:39:28,901 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199587.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:39:35,561 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199596.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:39:53,802 INFO [finetune.py:992] (1/2) Epoch 9, batch 1750, loss[loss=0.1598, simple_loss=0.2429, pruned_loss=0.03833, over 12351.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2606, pruned_loss=0.04134, over 2382927.09 frames. ], batch size: 31, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:39:56,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 11:40:01,100 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199631.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:40:16,863 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.823e+02 3.191e+02 3.852e+02 7.415e+02, threshold=6.382e+02, percent-clipped=2.0 2023-05-16 11:40:24,346 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1110, 2.5088, 3.7283, 3.0968, 3.5347, 3.2337, 2.5550, 3.5401], device='cuda:1'), covar=tensor([0.0132, 0.0324, 0.0140, 0.0206, 0.0129, 0.0172, 0.0335, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0200, 0.0179, 0.0179, 0.0206, 0.0154, 0.0192, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:40:29,887 INFO [finetune.py:992] (1/2) Epoch 9, batch 1800, loss[loss=0.1925, simple_loss=0.2835, pruned_loss=0.05075, over 12115.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2603, pruned_loss=0.04123, over 2388827.09 frames. ], batch size: 38, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:40:32,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9044, 4.8418, 4.7092, 4.7701, 4.3899, 4.9123, 4.9320, 5.0611], device='cuda:1'), covar=tensor([0.0324, 0.0170, 0.0231, 0.0341, 0.0825, 0.0323, 0.0182, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0190, 0.0184, 0.0238, 0.0235, 0.0211, 0.0169, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 11:41:06,302 INFO [finetune.py:992] (1/2) Epoch 9, batch 1850, loss[loss=0.1486, simple_loss=0.2323, pruned_loss=0.03242, over 12301.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2594, pruned_loss=0.04115, over 2392311.53 frames. ], batch size: 28, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:41:16,420 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2230, 5.1295, 5.0573, 5.0757, 4.7055, 5.1669, 5.2179, 5.4331], device='cuda:1'), covar=tensor([0.0289, 0.0139, 0.0181, 0.0302, 0.0784, 0.0265, 0.0132, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0190, 0.0184, 0.0239, 0.0235, 0.0211, 0.0170, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 11:41:28,993 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.717e+02 3.262e+02 3.822e+02 8.454e+02, threshold=6.525e+02, percent-clipped=2.0 2023-05-16 11:41:31,235 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199756.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:41:42,429 INFO [finetune.py:992] (1/2) Epoch 9, batch 1900, loss[loss=0.1797, simple_loss=0.2742, pruned_loss=0.04265, over 12352.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2599, pruned_loss=0.04143, over 2385131.14 frames. ], batch size: 36, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:42:13,438 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199814.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:42:15,696 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1106, 2.4190, 3.7383, 3.1021, 3.4540, 3.2461, 2.5680, 3.5188], device='cuda:1'), covar=tensor([0.0153, 0.0374, 0.0127, 0.0243, 0.0174, 0.0166, 0.0391, 0.0129], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0201, 0.0180, 0.0180, 0.0208, 0.0156, 0.0194, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:42:18,250 INFO [finetune.py:992] (1/2) Epoch 9, batch 1950, loss[loss=0.1818, simple_loss=0.2704, pruned_loss=0.04663, over 12301.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2599, pruned_loss=0.04126, over 2377490.78 frames. ], batch size: 34, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:42:26,111 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199832.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:42:36,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-16 11:42:39,400 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199850.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:42:41,351 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.711e+02 3.179e+02 3.779e+02 6.587e+02, threshold=6.359e+02, percent-clipped=1.0 2023-05-16 11:42:54,206 INFO [finetune.py:992] (1/2) Epoch 9, batch 2000, loss[loss=0.1857, simple_loss=0.2777, pruned_loss=0.04686, over 12353.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2594, pruned_loss=0.04114, over 2384111.74 frames. ], batch size: 36, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:43:01,028 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199880.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:43:08,720 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199891.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:43:22,304 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199909.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:43:30,678 INFO [finetune.py:992] (1/2) Epoch 9, batch 2050, loss[loss=0.1837, simple_loss=0.2692, pruned_loss=0.04906, over 12352.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2589, pruned_loss=0.04119, over 2383659.38 frames. ], batch size: 36, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:43:33,690 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4599, 5.2003, 5.2945, 5.3796, 4.9820, 5.0409, 4.7774, 5.2994], device='cuda:1'), covar=tensor([0.0633, 0.0602, 0.0898, 0.0580, 0.2007, 0.1224, 0.0624, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0672, 0.0579, 0.0589, 0.0817, 0.0707, 0.0525, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:43:53,836 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.235e+02 2.846e+02 3.318e+02 3.919e+02 6.845e+02, threshold=6.636e+02, percent-clipped=2.0 2023-05-16 11:44:06,104 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199970.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:44:06,564 INFO [finetune.py:992] (1/2) Epoch 9, batch 2100, loss[loss=0.2016, simple_loss=0.2944, pruned_loss=0.05444, over 11284.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2595, pruned_loss=0.04115, over 2379954.28 frames. ], batch size: 55, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:44:36,269 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7685, 3.3417, 5.1228, 2.6883, 2.8371, 3.7891, 3.3250, 3.7969], device='cuda:1'), covar=tensor([0.0398, 0.1180, 0.0319, 0.1249, 0.1982, 0.1619, 0.1338, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0231, 0.0238, 0.0180, 0.0234, 0.0283, 0.0220, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:44:44,124 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0645, 4.6896, 4.8507, 4.9051, 4.6541, 4.8673, 4.7530, 2.7551], device='cuda:1'), covar=tensor([0.0094, 0.0062, 0.0081, 0.0061, 0.0049, 0.0094, 0.0070, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0075, 0.0078, 0.0070, 0.0057, 0.0087, 0.0077, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:44:46,137 INFO [finetune.py:992] (1/2) Epoch 9, batch 2150, loss[loss=0.1571, simple_loss=0.2403, pruned_loss=0.037, over 11826.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2588, pruned_loss=0.04079, over 2384795.28 frames. ], batch size: 26, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:45:09,898 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.662e+02 3.100e+02 3.609e+02 6.112e+02, threshold=6.199e+02, percent-clipped=0.0 2023-05-16 11:45:12,282 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200056.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:45:22,847 INFO [finetune.py:992] (1/2) Epoch 9, batch 2200, loss[loss=0.1802, simple_loss=0.2713, pruned_loss=0.04452, over 10440.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2586, pruned_loss=0.04075, over 2384034.63 frames. ], batch size: 68, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:45:34,915 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0 2023-05-16 11:45:46,356 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200104.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:45:53,719 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200114.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:45:58,490 INFO [finetune.py:992] (1/2) Epoch 9, batch 2250, loss[loss=0.1886, simple_loss=0.2783, pruned_loss=0.04947, over 12108.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2594, pruned_loss=0.04086, over 2383301.40 frames. ], batch size: 39, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:46:19,628 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200150.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:46:21,555 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.841e+02 3.279e+02 4.097e+02 1.905e+03, threshold=6.557e+02, percent-clipped=6.0 2023-05-16 11:46:28,028 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200162.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:46:34,352 INFO [finetune.py:992] (1/2) Epoch 9, batch 2300, loss[loss=0.1524, simple_loss=0.2314, pruned_loss=0.03673, over 11803.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2599, pruned_loss=0.04127, over 2374537.23 frames. ], batch size: 26, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:46:38,652 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200177.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:46:49,380 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200191.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:46:53,983 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200198.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:47:10,431 INFO [finetune.py:992] (1/2) Epoch 9, batch 2350, loss[loss=0.1805, simple_loss=0.274, pruned_loss=0.04349, over 12247.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2601, pruned_loss=0.04156, over 2377283.75 frames. ], batch size: 37, lr: 4.20e-03, grad_scale: 16.0 2023-05-16 11:47:22,602 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200238.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:47:23,179 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200239.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:47:32,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.05 vs. limit=5.0 2023-05-16 11:47:33,028 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.910e+02 3.318e+02 3.930e+02 2.008e+03, threshold=6.636e+02, percent-clipped=7.0 2023-05-16 11:47:41,252 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200265.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:47:45,424 INFO [finetune.py:992] (1/2) Epoch 9, batch 2400, loss[loss=0.1654, simple_loss=0.2483, pruned_loss=0.04123, over 12194.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2604, pruned_loss=0.04183, over 2370306.33 frames. ], batch size: 35, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:47:51,902 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200279.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:48:21,837 INFO [finetune.py:992] (1/2) Epoch 9, batch 2450, loss[loss=0.1821, simple_loss=0.2695, pruned_loss=0.04733, over 12032.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2603, pruned_loss=0.04154, over 2372115.63 frames. ], batch size: 31, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:48:25,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 11:48:36,367 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200340.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:48:38,433 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0112, 4.9434, 4.7968, 4.8587, 4.5333, 4.9587, 4.9800, 5.1326], device='cuda:1'), covar=tensor([0.0214, 0.0142, 0.0205, 0.0288, 0.0738, 0.0273, 0.0173, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0195, 0.0189, 0.0244, 0.0241, 0.0215, 0.0173, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 11:48:45,374 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.761e+02 3.380e+02 3.986e+02 6.954e+02, threshold=6.761e+02, percent-clipped=1.0 2023-05-16 11:48:48,971 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1809, 5.0085, 5.1206, 5.1642, 4.7969, 4.8062, 4.6271, 5.0236], device='cuda:1'), covar=tensor([0.0708, 0.0610, 0.0875, 0.0578, 0.1901, 0.1299, 0.0580, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0664, 0.0569, 0.0581, 0.0802, 0.0704, 0.0518, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:48:58,194 INFO [finetune.py:992] (1/2) Epoch 9, batch 2500, loss[loss=0.1724, simple_loss=0.2605, pruned_loss=0.04214, over 12038.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.26, pruned_loss=0.04169, over 2371316.95 frames. ], batch size: 40, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:49:21,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-05-16 11:49:34,589 INFO [finetune.py:992] (1/2) Epoch 9, batch 2550, loss[loss=0.2056, simple_loss=0.2778, pruned_loss=0.06676, over 12150.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.26, pruned_loss=0.04161, over 2366373.96 frames. ], batch size: 36, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:49:57,448 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.735e+02 3.226e+02 3.817e+02 9.182e+02, threshold=6.453e+02, percent-clipped=2.0 2023-05-16 11:50:10,328 INFO [finetune.py:992] (1/2) Epoch 9, batch 2600, loss[loss=0.1886, simple_loss=0.2675, pruned_loss=0.0548, over 12151.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2587, pruned_loss=0.04119, over 2370382.79 frames. ], batch size: 34, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:50:46,264 INFO [finetune.py:992] (1/2) Epoch 9, batch 2650, loss[loss=0.1675, simple_loss=0.262, pruned_loss=0.03649, over 10304.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2589, pruned_loss=0.04086, over 2372905.38 frames. ], batch size: 68, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:50:54,991 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200533.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 11:51:08,742 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.749e+02 3.270e+02 3.848e+02 7.144e+02, threshold=6.540e+02, percent-clipped=2.0 2023-05-16 11:51:18,115 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200565.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:51:22,211 INFO [finetune.py:992] (1/2) Epoch 9, batch 2700, loss[loss=0.1649, simple_loss=0.2576, pruned_loss=0.03606, over 12099.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2589, pruned_loss=0.041, over 2373592.77 frames. ], batch size: 33, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:51:52,366 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200613.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:51:54,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1192, 5.9613, 5.4662, 5.5778, 6.0420, 5.4057, 5.5820, 5.5553], device='cuda:1'), covar=tensor([0.1329, 0.0872, 0.1085, 0.1677, 0.0922, 0.2057, 0.1671, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0469, 0.0376, 0.0419, 0.0446, 0.0424, 0.0383, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:51:58,020 INFO [finetune.py:992] (1/2) Epoch 9, batch 2750, loss[loss=0.1646, simple_loss=0.252, pruned_loss=0.03854, over 12337.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2597, pruned_loss=0.04142, over 2371664.45 frames. ], batch size: 31, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:52:08,754 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200635.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 11:52:16,013 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200645.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:52:21,445 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.726e+02 3.164e+02 4.050e+02 6.073e+02, threshold=6.328e+02, percent-clipped=0.0 2023-05-16 11:52:34,313 INFO [finetune.py:992] (1/2) Epoch 9, batch 2800, loss[loss=0.1647, simple_loss=0.2579, pruned_loss=0.03577, over 12160.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2588, pruned_loss=0.04094, over 2377469.13 frames. ], batch size: 34, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:52:59,249 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200706.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:53:10,425 INFO [finetune.py:992] (1/2) Epoch 9, batch 2850, loss[loss=0.1804, simple_loss=0.2762, pruned_loss=0.04229, over 12137.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2592, pruned_loss=0.04122, over 2379370.77 frames. ], batch size: 36, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:53:24,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5102, 5.3090, 5.4103, 5.4803, 5.0763, 5.1614, 4.9571, 5.4140], device='cuda:1'), covar=tensor([0.0619, 0.0615, 0.0799, 0.0563, 0.1860, 0.1166, 0.0546, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0670, 0.0577, 0.0593, 0.0817, 0.0716, 0.0527, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:53:28,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 11:53:33,240 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.724e+02 3.281e+02 3.924e+02 1.531e+03, threshold=6.562e+02, percent-clipped=4.0 2023-05-16 11:53:40,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 2023-05-16 11:53:46,048 INFO [finetune.py:992] (1/2) Epoch 9, batch 2900, loss[loss=0.2089, simple_loss=0.2941, pruned_loss=0.06187, over 11821.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2594, pruned_loss=0.04148, over 2378750.34 frames. ], batch size: 44, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:54:23,016 INFO [finetune.py:992] (1/2) Epoch 9, batch 2950, loss[loss=0.1627, simple_loss=0.2534, pruned_loss=0.036, over 12123.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2591, pruned_loss=0.04139, over 2379103.17 frames. ], batch size: 30, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:54:31,782 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200833.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 11:54:45,609 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.805e+02 3.316e+02 3.739e+02 7.927e+02, threshold=6.633e+02, percent-clipped=2.0 2023-05-16 11:54:58,962 INFO [finetune.py:992] (1/2) Epoch 9, batch 3000, loss[loss=0.153, simple_loss=0.24, pruned_loss=0.03297, over 12256.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2583, pruned_loss=0.04114, over 2377364.88 frames. ], batch size: 32, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:54:58,962 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 11:55:06,663 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8090, 1.8207, 2.7933, 3.8467, 1.8992, 3.9170, 3.7851, 3.9899], device='cuda:1'), covar=tensor([0.0142, 0.1451, 0.0496, 0.0135, 0.1415, 0.0176, 0.0251, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0206, 0.0182, 0.0117, 0.0192, 0.0177, 0.0174, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 11:55:12,084 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0917, 6.0251, 5.9601, 5.4258, 5.3754, 5.9863, 5.6164, 5.6547], device='cuda:1'), covar=tensor([0.0522, 0.0729, 0.0444, 0.1464, 0.0398, 0.0593, 0.1224, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0537, 0.0495, 0.0618, 0.0399, 0.0692, 0.0746, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 11:55:17,472 INFO [finetune.py:1026] (1/2) Epoch 9, validation: loss=0.3195, simple_loss=0.3965, pruned_loss=0.1213, over 1020973.00 frames. 2023-05-16 11:55:17,473 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 11:55:25,252 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200881.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 11:55:37,549 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6240, 4.9842, 3.3330, 2.9759, 4.2354, 2.5254, 4.1723, 3.5391], device='cuda:1'), covar=tensor([0.0612, 0.0462, 0.0918, 0.1155, 0.0289, 0.1351, 0.0408, 0.0636], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0251, 0.0175, 0.0199, 0.0139, 0.0181, 0.0194, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:55:46,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-16 11:55:51,015 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9519, 5.9200, 5.7274, 5.3027, 5.1818, 5.8840, 5.5120, 5.3199], device='cuda:1'), covar=tensor([0.0791, 0.0981, 0.0694, 0.1710, 0.0642, 0.0757, 0.1516, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0595, 0.0540, 0.0497, 0.0621, 0.0401, 0.0695, 0.0750, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 11:55:53,795 INFO [finetune.py:992] (1/2) Epoch 9, batch 3050, loss[loss=0.1656, simple_loss=0.2613, pruned_loss=0.03493, over 12356.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.259, pruned_loss=0.04164, over 2372078.65 frames. ], batch size: 36, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:55:58,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 11:56:03,872 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200935.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:56:16,432 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 2.822e+02 3.318e+02 4.013e+02 5.698e+02, threshold=6.637e+02, percent-clipped=0.0 2023-05-16 11:56:30,084 INFO [finetune.py:992] (1/2) Epoch 9, batch 3100, loss[loss=0.1865, simple_loss=0.2857, pruned_loss=0.04363, over 12365.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2591, pruned_loss=0.0414, over 2372188.47 frames. ], batch size: 36, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:56:37,502 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0133, 4.9692, 4.7890, 4.8579, 4.5103, 4.9955, 5.0019, 5.1764], device='cuda:1'), covar=tensor([0.0254, 0.0136, 0.0200, 0.0277, 0.0745, 0.0217, 0.0140, 0.0195], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0194, 0.0188, 0.0244, 0.0241, 0.0215, 0.0173, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 11:56:38,831 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200983.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:56:51,849 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201001.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:57:04,477 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 11:57:06,185 INFO [finetune.py:992] (1/2) Epoch 9, batch 3150, loss[loss=0.1757, simple_loss=0.275, pruned_loss=0.03815, over 12283.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.259, pruned_loss=0.04124, over 2373509.03 frames. ], batch size: 37, lr: 4.19e-03, grad_scale: 16.0 2023-05-16 11:57:15,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-05-16 11:57:29,728 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.942e+02 2.683e+02 3.143e+02 3.728e+02 6.490e+02, threshold=6.287e+02, percent-clipped=0.0 2023-05-16 11:57:42,699 INFO [finetune.py:992] (1/2) Epoch 9, batch 3200, loss[loss=0.1632, simple_loss=0.2561, pruned_loss=0.03514, over 10672.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2579, pruned_loss=0.04069, over 2373333.33 frames. ], batch size: 68, lr: 4.19e-03, grad_scale: 32.0 2023-05-16 11:57:42,889 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201071.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:58:02,377 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3059, 4.7088, 2.9102, 2.3642, 4.0573, 2.4109, 3.9564, 3.1424], device='cuda:1'), covar=tensor([0.0692, 0.0464, 0.1151, 0.1753, 0.0309, 0.1487, 0.0476, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0252, 0.0177, 0.0200, 0.0140, 0.0181, 0.0196, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:58:19,339 INFO [finetune.py:992] (1/2) Epoch 9, batch 3250, loss[loss=0.2164, simple_loss=0.2937, pruned_loss=0.06953, over 8166.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2583, pruned_loss=0.0412, over 2359322.01 frames. ], batch size: 98, lr: 4.19e-03, grad_scale: 32.0 2023-05-16 11:58:26,069 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1833, 3.9606, 3.9767, 4.3014, 2.7887, 3.7994, 2.4410, 3.9504], device='cuda:1'), covar=tensor([0.1594, 0.0756, 0.0982, 0.0787, 0.1285, 0.0646, 0.1951, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0265, 0.0292, 0.0354, 0.0236, 0.0236, 0.0259, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 11:58:27,410 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201132.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 11:58:38,547 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3467, 4.9300, 5.3863, 4.7065, 4.9932, 4.7896, 5.3722, 5.0186], device='cuda:1'), covar=tensor([0.0259, 0.0362, 0.0230, 0.0224, 0.0315, 0.0265, 0.0192, 0.0279], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0255, 0.0275, 0.0250, 0.0248, 0.0250, 0.0230, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 11:58:41,963 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.910e+02 3.291e+02 3.848e+02 5.664e+02, threshold=6.582e+02, percent-clipped=0.0 2023-05-16 11:58:54,693 INFO [finetune.py:992] (1/2) Epoch 9, batch 3300, loss[loss=0.2053, simple_loss=0.2805, pruned_loss=0.06511, over 8510.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2591, pruned_loss=0.04124, over 2366627.27 frames. ], batch size: 98, lr: 4.19e-03, grad_scale: 32.0 2023-05-16 11:59:31,516 INFO [finetune.py:992] (1/2) Epoch 9, batch 3350, loss[loss=0.1703, simple_loss=0.2618, pruned_loss=0.0394, over 11249.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2585, pruned_loss=0.04076, over 2376793.21 frames. ], batch size: 55, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 11:59:39,783 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 11:59:54,542 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.735e+02 3.170e+02 3.586e+02 6.394e+02, threshold=6.340e+02, percent-clipped=0.0 2023-05-16 11:59:54,735 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201253.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:00:00,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.88 vs. limit=5.0 2023-05-16 12:00:08,019 INFO [finetune.py:992] (1/2) Epoch 9, batch 3400, loss[loss=0.1676, simple_loss=0.2603, pruned_loss=0.03745, over 12113.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2584, pruned_loss=0.04081, over 2380641.11 frames. ], batch size: 33, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:00:10,477 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201274.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:00:29,597 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201301.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:00:38,733 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201314.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:00:43,494 INFO [finetune.py:992] (1/2) Epoch 9, batch 3450, loss[loss=0.1684, simple_loss=0.254, pruned_loss=0.04137, over 12183.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2585, pruned_loss=0.0409, over 2374715.11 frames. ], batch size: 31, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:00:54,409 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201335.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:00:56,624 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-05-16 12:01:04,240 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201349.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:01:07,016 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.752e+02 3.275e+02 4.140e+02 7.327e+02, threshold=6.550e+02, percent-clipped=2.0 2023-05-16 12:01:19,612 INFO [finetune.py:992] (1/2) Epoch 9, batch 3500, loss[loss=0.1531, simple_loss=0.2341, pruned_loss=0.03611, over 12029.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2585, pruned_loss=0.04079, over 2372326.74 frames. ], batch size: 31, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:01:56,025 INFO [finetune.py:992] (1/2) Epoch 9, batch 3550, loss[loss=0.1691, simple_loss=0.2559, pruned_loss=0.04121, over 12249.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2581, pruned_loss=0.04032, over 2379502.21 frames. ], batch size: 32, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:02:00,333 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201427.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:02:10,533 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5060, 4.1574, 4.2179, 4.5116, 3.0633, 4.1621, 2.7452, 4.2502], device='cuda:1'), covar=tensor([0.1464, 0.0718, 0.0878, 0.0584, 0.1102, 0.0472, 0.1665, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0262, 0.0288, 0.0350, 0.0234, 0.0234, 0.0256, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:02:19,019 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.890e+02 3.315e+02 3.847e+02 6.638e+02, threshold=6.631e+02, percent-clipped=2.0 2023-05-16 12:02:31,738 INFO [finetune.py:992] (1/2) Epoch 9, batch 3600, loss[loss=0.1631, simple_loss=0.2502, pruned_loss=0.03794, over 12142.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2576, pruned_loss=0.04003, over 2386627.43 frames. ], batch size: 34, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:03:08,098 INFO [finetune.py:992] (1/2) Epoch 9, batch 3650, loss[loss=0.1722, simple_loss=0.2499, pruned_loss=0.04725, over 11833.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.04005, over 2388820.72 frames. ], batch size: 26, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:03:13,094 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8832, 4.8806, 4.7296, 4.9171, 3.8047, 5.0054, 4.9416, 5.1004], device='cuda:1'), covar=tensor([0.0278, 0.0174, 0.0243, 0.0343, 0.1479, 0.0286, 0.0184, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0196, 0.0190, 0.0247, 0.0243, 0.0217, 0.0174, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 12:03:31,409 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.865e+02 3.266e+02 3.969e+02 6.770e+02, threshold=6.532e+02, percent-clipped=0.0 2023-05-16 12:03:42,573 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 12:03:44,094 INFO [finetune.py:992] (1/2) Epoch 9, batch 3700, loss[loss=0.196, simple_loss=0.2833, pruned_loss=0.05431, over 12337.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2592, pruned_loss=0.04091, over 2377697.16 frames. ], batch size: 36, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:03:49,897 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6751, 3.5735, 3.2610, 3.2036, 2.8654, 2.7614, 3.6441, 2.2139], device='cuda:1'), covar=tensor([0.0312, 0.0116, 0.0150, 0.0185, 0.0366, 0.0310, 0.0110, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0161, 0.0156, 0.0185, 0.0201, 0.0196, 0.0166, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:04:11,464 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201609.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:04:19,574 INFO [finetune.py:992] (1/2) Epoch 9, batch 3750, loss[loss=0.1561, simple_loss=0.2512, pruned_loss=0.03051, over 12279.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2598, pruned_loss=0.04119, over 2384709.34 frames. ], batch size: 33, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:04:26,827 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201630.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:04:36,916 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201644.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:04:43,134 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.871e+02 3.276e+02 3.780e+02 6.507e+02, threshold=6.551e+02, percent-clipped=1.0 2023-05-16 12:04:49,002 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201661.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:04:55,464 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6692, 2.9311, 3.8316, 4.6228, 4.0287, 4.6427, 3.9782, 3.4550], device='cuda:1'), covar=tensor([0.0030, 0.0333, 0.0117, 0.0033, 0.0096, 0.0062, 0.0102, 0.0319], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0121, 0.0103, 0.0074, 0.0100, 0.0115, 0.0091, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:04:55,991 INFO [finetune.py:992] (1/2) Epoch 9, batch 3800, loss[loss=0.146, simple_loss=0.2329, pruned_loss=0.0295, over 12339.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2594, pruned_loss=0.04111, over 2373498.70 frames. ], batch size: 31, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:05:18,895 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2760, 5.0857, 5.1923, 5.2584, 4.8203, 4.8901, 4.7270, 5.2206], device='cuda:1'), covar=tensor([0.0729, 0.0674, 0.0795, 0.0685, 0.2066, 0.1484, 0.0584, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0678, 0.0583, 0.0594, 0.0823, 0.0723, 0.0532, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:05:21,100 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201705.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:05:21,794 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201706.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:05:32,300 INFO [finetune.py:992] (1/2) Epoch 9, batch 3850, loss[loss=0.1935, simple_loss=0.2797, pruned_loss=0.0537, over 12134.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2591, pruned_loss=0.04113, over 2374529.93 frames. ], batch size: 38, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:05:33,228 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201722.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:05:36,751 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201727.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:05:50,943 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7441, 2.8841, 4.5342, 4.6566, 2.9014, 2.6684, 2.8561, 2.1870], device='cuda:1'), covar=tensor([0.1526, 0.2865, 0.0463, 0.0458, 0.1238, 0.2219, 0.2537, 0.3933], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0376, 0.0268, 0.0291, 0.0262, 0.0292, 0.0365, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:05:54,882 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.704e+02 3.228e+02 3.890e+02 7.833e+02, threshold=6.456e+02, percent-clipped=1.0 2023-05-16 12:06:05,679 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201767.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:06:08,321 INFO [finetune.py:992] (1/2) Epoch 9, batch 3900, loss[loss=0.15, simple_loss=0.2385, pruned_loss=0.03078, over 12149.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2584, pruned_loss=0.04104, over 2363662.33 frames. ], batch size: 30, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:06:10,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 12:06:11,165 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201775.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:06:23,406 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2286, 2.5388, 3.7528, 3.1813, 3.5575, 3.2256, 2.6088, 3.5835], device='cuda:1'), covar=tensor([0.0113, 0.0355, 0.0158, 0.0218, 0.0138, 0.0169, 0.0303, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0200, 0.0182, 0.0182, 0.0210, 0.0158, 0.0194, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:06:43,894 INFO [finetune.py:992] (1/2) Epoch 9, batch 3950, loss[loss=0.2036, simple_loss=0.293, pruned_loss=0.05711, over 12113.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2583, pruned_loss=0.04091, over 2369346.50 frames. ], batch size: 38, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:06:50,467 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8623, 3.3738, 3.6004, 3.7897, 3.6989, 3.7881, 3.6350, 2.5599], device='cuda:1'), covar=tensor([0.0084, 0.0118, 0.0131, 0.0068, 0.0058, 0.0106, 0.0091, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0076, 0.0078, 0.0071, 0.0057, 0.0087, 0.0077, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:06:59,737 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9570, 3.4943, 5.2010, 2.7648, 2.8709, 3.8991, 3.2344, 3.8934], device='cuda:1'), covar=tensor([0.0371, 0.1042, 0.0309, 0.1101, 0.1925, 0.1415, 0.1359, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0233, 0.0242, 0.0184, 0.0237, 0.0290, 0.0225, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:07:07,292 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.742e+02 3.284e+02 3.856e+02 7.422e+02, threshold=6.567e+02, percent-clipped=1.0 2023-05-16 12:07:20,255 INFO [finetune.py:992] (1/2) Epoch 9, batch 4000, loss[loss=0.1817, simple_loss=0.2727, pruned_loss=0.04531, over 12121.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2574, pruned_loss=0.04063, over 2368543.04 frames. ], batch size: 39, lr: 4.18e-03, grad_scale: 32.0 2023-05-16 12:07:31,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-05-16 12:07:47,287 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201909.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:07:56,290 INFO [finetune.py:992] (1/2) Epoch 9, batch 4050, loss[loss=0.2123, simple_loss=0.2996, pruned_loss=0.06253, over 11986.00 frames. ], tot_loss[loss=0.17, simple_loss=0.258, pruned_loss=0.04103, over 2364936.44 frames. ], batch size: 40, lr: 4.18e-03, grad_scale: 16.0 2023-05-16 12:08:02,883 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201930.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:08:20,088 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.994e+02 2.634e+02 3.114e+02 4.085e+02 9.505e+02, threshold=6.228e+02, percent-clipped=2.0 2023-05-16 12:08:22,273 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201957.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:08:30,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 12:08:31,939 INFO [finetune.py:992] (1/2) Epoch 9, batch 4100, loss[loss=0.1597, simple_loss=0.2422, pruned_loss=0.03862, over 12170.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2581, pruned_loss=0.04116, over 2366849.18 frames. ], batch size: 31, lr: 4.18e-03, grad_scale: 16.0 2023-05-16 12:08:33,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.06 vs. limit=5.0 2023-05-16 12:08:37,126 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2944, 6.1081, 5.6495, 5.6577, 6.1982, 5.5348, 5.6712, 5.7018], device='cuda:1'), covar=tensor([0.1362, 0.1035, 0.1059, 0.1749, 0.0920, 0.2189, 0.1954, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0478, 0.0379, 0.0428, 0.0454, 0.0430, 0.0384, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:08:37,877 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201978.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:08:46,456 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201990.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:08:56,450 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202000.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:09:00,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 2023-05-16 12:09:08,605 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202017.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:09:11,429 INFO [finetune.py:992] (1/2) Epoch 9, batch 4150, loss[loss=0.1538, simple_loss=0.2385, pruned_loss=0.03454, over 11992.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2583, pruned_loss=0.04112, over 2369871.81 frames. ], batch size: 28, lr: 4.18e-03, grad_scale: 16.0 2023-05-16 12:09:32,607 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202051.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:09:34,571 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.800e+02 3.305e+02 4.028e+02 6.958e+02, threshold=6.610e+02, percent-clipped=2.0 2023-05-16 12:09:40,322 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202062.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:09:47,155 INFO [finetune.py:992] (1/2) Epoch 9, batch 4200, loss[loss=0.1751, simple_loss=0.2643, pruned_loss=0.043, over 12361.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2603, pruned_loss=0.04204, over 2364593.18 frames. ], batch size: 35, lr: 4.18e-03, grad_scale: 16.0 2023-05-16 12:10:08,958 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6517, 4.8589, 4.3555, 5.1839, 4.8136, 3.0539, 4.4812, 3.2277], device='cuda:1'), covar=tensor([0.0612, 0.0751, 0.1251, 0.0548, 0.0914, 0.1472, 0.1010, 0.2894], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0379, 0.0358, 0.0289, 0.0368, 0.0266, 0.0341, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:10:17,369 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1084, 5.0506, 4.9224, 4.9331, 4.6367, 5.0775, 5.0803, 5.2778], device='cuda:1'), covar=tensor([0.0246, 0.0141, 0.0197, 0.0318, 0.0771, 0.0274, 0.0149, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0196, 0.0191, 0.0247, 0.0244, 0.0217, 0.0175, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 12:10:23,547 INFO [finetune.py:992] (1/2) Epoch 9, batch 4250, loss[loss=0.1725, simple_loss=0.2611, pruned_loss=0.04193, over 12141.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2591, pruned_loss=0.04163, over 2368469.23 frames. ], batch size: 34, lr: 4.18e-03, grad_scale: 16.0 2023-05-16 12:10:26,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 12:10:46,832 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 2.761e+02 3.356e+02 3.983e+02 6.332e+02, threshold=6.712e+02, percent-clipped=0.0 2023-05-16 12:10:53,341 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9973, 4.9375, 4.8570, 4.8744, 4.5546, 4.9527, 4.9715, 5.1784], device='cuda:1'), covar=tensor([0.0199, 0.0139, 0.0178, 0.0329, 0.0764, 0.0269, 0.0126, 0.0164], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0197, 0.0191, 0.0248, 0.0245, 0.0218, 0.0175, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 12:10:58,826 INFO [finetune.py:992] (1/2) Epoch 9, batch 4300, loss[loss=0.1647, simple_loss=0.2555, pruned_loss=0.03696, over 12318.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2589, pruned_loss=0.04155, over 2360169.52 frames. ], batch size: 34, lr: 4.18e-03, grad_scale: 16.0 2023-05-16 12:11:34,812 INFO [finetune.py:992] (1/2) Epoch 9, batch 4350, loss[loss=0.1716, simple_loss=0.2617, pruned_loss=0.04075, over 11313.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2592, pruned_loss=0.04173, over 2369091.79 frames. ], batch size: 55, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:11:53,570 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8782, 4.7640, 4.8277, 4.8709, 4.5309, 4.6010, 4.4460, 4.8106], device='cuda:1'), covar=tensor([0.0647, 0.0562, 0.0783, 0.0639, 0.1807, 0.1419, 0.0537, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0680, 0.0584, 0.0596, 0.0821, 0.0721, 0.0532, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:11:58,308 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.751e+02 3.212e+02 3.727e+02 1.205e+03, threshold=6.424e+02, percent-clipped=3.0 2023-05-16 12:12:10,927 INFO [finetune.py:992] (1/2) Epoch 9, batch 4400, loss[loss=0.1409, simple_loss=0.2188, pruned_loss=0.03148, over 12022.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.259, pruned_loss=0.04133, over 2372593.34 frames. ], batch size: 28, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:12:31,426 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202300.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:12:43,510 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202317.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:12:46,164 INFO [finetune.py:992] (1/2) Epoch 9, batch 4450, loss[loss=0.1595, simple_loss=0.2525, pruned_loss=0.03326, over 12153.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2605, pruned_loss=0.04181, over 2365027.76 frames. ], batch size: 34, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:13:03,782 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202346.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:13:05,200 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202348.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:13:10,075 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.951e+02 3.496e+02 3.984e+02 8.134e+02, threshold=6.992e+02, percent-clipped=1.0 2023-05-16 12:13:15,985 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202362.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:13:17,972 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202365.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:13:18,081 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3919, 4.9544, 5.4113, 4.6453, 5.0184, 4.8014, 5.3982, 5.0209], device='cuda:1'), covar=tensor([0.0241, 0.0366, 0.0247, 0.0259, 0.0330, 0.0285, 0.0202, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0254, 0.0276, 0.0250, 0.0248, 0.0250, 0.0229, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:13:22,061 INFO [finetune.py:992] (1/2) Epoch 9, batch 4500, loss[loss=0.1681, simple_loss=0.2543, pruned_loss=0.04092, over 12353.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2608, pruned_loss=0.04176, over 2373458.01 frames. ], batch size: 31, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:13:34,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.96 vs. limit=5.0 2023-05-16 12:13:50,921 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202410.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:13:58,710 INFO [finetune.py:992] (1/2) Epoch 9, batch 4550, loss[loss=0.1806, simple_loss=0.2685, pruned_loss=0.04637, over 11830.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2607, pruned_loss=0.04173, over 2373198.46 frames. ], batch size: 44, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:14:13,891 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4518, 2.9155, 3.8137, 2.4121, 2.6587, 3.0761, 2.9410, 3.1491], device='cuda:1'), covar=tensor([0.0466, 0.0968, 0.0449, 0.1234, 0.1520, 0.1313, 0.1141, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0234, 0.0243, 0.0184, 0.0237, 0.0291, 0.0224, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:14:22,242 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 2.840e+02 3.198e+02 3.742e+02 1.519e+03, threshold=6.396e+02, percent-clipped=2.0 2023-05-16 12:14:34,387 INFO [finetune.py:992] (1/2) Epoch 9, batch 4600, loss[loss=0.1466, simple_loss=0.2328, pruned_loss=0.0302, over 12342.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2605, pruned_loss=0.04127, over 2361101.05 frames. ], batch size: 31, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:15:05,669 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2551, 4.8089, 5.2596, 4.6222, 4.9144, 4.6802, 5.2639, 4.8721], device='cuda:1'), covar=tensor([0.0254, 0.0366, 0.0257, 0.0240, 0.0320, 0.0298, 0.0216, 0.0319], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0257, 0.0279, 0.0252, 0.0251, 0.0252, 0.0231, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:15:10,533 INFO [finetune.py:992] (1/2) Epoch 9, batch 4650, loss[loss=0.1745, simple_loss=0.2662, pruned_loss=0.04136, over 12150.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2598, pruned_loss=0.04092, over 2368407.72 frames. ], batch size: 36, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:15:14,995 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0785, 2.2806, 3.6637, 2.9985, 3.4209, 3.1672, 2.3621, 3.5062], device='cuda:1'), covar=tensor([0.0131, 0.0392, 0.0122, 0.0244, 0.0156, 0.0171, 0.0391, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0201, 0.0183, 0.0183, 0.0213, 0.0160, 0.0196, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:15:34,767 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.899e+02 3.227e+02 3.837e+02 6.733e+02, threshold=6.454e+02, percent-clipped=1.0 2023-05-16 12:15:39,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 12:15:46,584 INFO [finetune.py:992] (1/2) Epoch 9, batch 4700, loss[loss=0.1676, simple_loss=0.2596, pruned_loss=0.03781, over 12031.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2597, pruned_loss=0.04094, over 2369284.89 frames. ], batch size: 37, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:16:03,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 12:16:05,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7322, 2.9768, 3.8324, 4.6881, 4.0502, 4.6997, 4.0843, 3.4284], device='cuda:1'), covar=tensor([0.0028, 0.0312, 0.0129, 0.0033, 0.0104, 0.0059, 0.0098, 0.0304], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0121, 0.0104, 0.0075, 0.0100, 0.0116, 0.0092, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:16:22,524 INFO [finetune.py:992] (1/2) Epoch 9, batch 4750, loss[loss=0.1589, simple_loss=0.2434, pruned_loss=0.0372, over 12248.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2592, pruned_loss=0.04095, over 2374473.02 frames. ], batch size: 32, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:16:37,611 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9125, 4.4979, 4.7006, 4.7720, 4.5284, 4.7438, 4.6270, 2.7054], device='cuda:1'), covar=tensor([0.0096, 0.0080, 0.0083, 0.0071, 0.0061, 0.0097, 0.0082, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0076, 0.0079, 0.0072, 0.0058, 0.0089, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:16:41,040 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202646.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:16:46,562 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 2.882e+02 3.213e+02 4.113e+02 1.161e+03, threshold=6.425e+02, percent-clipped=2.0 2023-05-16 12:16:53,536 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2235, 6.0144, 5.5409, 5.5290, 6.0988, 5.4976, 5.6278, 5.6213], device='cuda:1'), covar=tensor([0.1394, 0.0861, 0.1115, 0.1923, 0.0910, 0.2050, 0.1574, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0473, 0.0374, 0.0426, 0.0449, 0.0428, 0.0384, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:16:58,512 INFO [finetune.py:992] (1/2) Epoch 9, batch 4800, loss[loss=0.1515, simple_loss=0.235, pruned_loss=0.03402, over 12186.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2606, pruned_loss=0.04217, over 2362560.25 frames. ], batch size: 29, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:17:15,850 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202694.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:17:21,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 12:17:34,875 INFO [finetune.py:992] (1/2) Epoch 9, batch 4850, loss[loss=0.1913, simple_loss=0.2817, pruned_loss=0.0505, over 11972.00 frames. ], tot_loss[loss=0.173, simple_loss=0.261, pruned_loss=0.04252, over 2363910.94 frames. ], batch size: 40, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:17:58,004 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 2.677e+02 3.036e+02 3.907e+02 8.573e+02, threshold=6.072e+02, percent-clipped=3.0 2023-05-16 12:18:10,315 INFO [finetune.py:992] (1/2) Epoch 9, batch 4900, loss[loss=0.1404, simple_loss=0.2279, pruned_loss=0.02642, over 12290.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2594, pruned_loss=0.04151, over 2366077.39 frames. ], batch size: 28, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:18:46,611 INFO [finetune.py:992] (1/2) Epoch 9, batch 4950, loss[loss=0.1532, simple_loss=0.2271, pruned_loss=0.03963, over 11443.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2605, pruned_loss=0.04166, over 2362463.88 frames. ], batch size: 25, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:18:53,736 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202830.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 12:19:10,494 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.757e+02 3.415e+02 4.105e+02 1.382e+03, threshold=6.831e+02, percent-clipped=3.0 2023-05-16 12:19:22,524 INFO [finetune.py:992] (1/2) Epoch 9, batch 5000, loss[loss=0.1651, simple_loss=0.2576, pruned_loss=0.03634, over 12354.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2605, pruned_loss=0.04193, over 2358317.85 frames. ], batch size: 35, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:19:36,947 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202891.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 12:19:44,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.58 vs. limit=5.0 2023-05-16 12:19:57,693 INFO [finetune.py:992] (1/2) Epoch 9, batch 5050, loss[loss=0.188, simple_loss=0.2747, pruned_loss=0.05068, over 12367.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2602, pruned_loss=0.04178, over 2366781.84 frames. ], batch size: 38, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:20:21,641 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.730e+02 3.199e+02 3.913e+02 1.181e+03, threshold=6.397e+02, percent-clipped=3.0 2023-05-16 12:20:31,940 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2555, 2.6497, 3.8533, 3.2531, 3.6860, 3.2741, 2.6434, 3.7228], device='cuda:1'), covar=tensor([0.0125, 0.0339, 0.0132, 0.0185, 0.0133, 0.0172, 0.0347, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0201, 0.0184, 0.0182, 0.0212, 0.0158, 0.0194, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:20:33,891 INFO [finetune.py:992] (1/2) Epoch 9, batch 5100, loss[loss=0.149, simple_loss=0.2396, pruned_loss=0.02926, over 12264.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2599, pruned_loss=0.0416, over 2374479.82 frames. ], batch size: 32, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:21:10,313 INFO [finetune.py:992] (1/2) Epoch 9, batch 5150, loss[loss=0.1517, simple_loss=0.2502, pruned_loss=0.02656, over 12147.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2597, pruned_loss=0.04174, over 2362911.90 frames. ], batch size: 36, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:21:22,513 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203038.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:21:33,503 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.075e+02 2.774e+02 3.219e+02 3.722e+02 9.351e+02, threshold=6.437e+02, percent-clipped=3.0 2023-05-16 12:21:45,384 INFO [finetune.py:992] (1/2) Epoch 9, batch 5200, loss[loss=0.1759, simple_loss=0.2732, pruned_loss=0.03932, over 12157.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2601, pruned_loss=0.04184, over 2367091.18 frames. ], batch size: 36, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:21:55,977 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203085.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:22:05,909 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203099.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:22:21,521 INFO [finetune.py:992] (1/2) Epoch 9, batch 5250, loss[loss=0.1672, simple_loss=0.2498, pruned_loss=0.04229, over 12191.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2601, pruned_loss=0.0415, over 2370566.82 frames. ], batch size: 29, lr: 4.17e-03, grad_scale: 16.0 2023-05-16 12:22:40,062 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203146.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:22:45,582 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.827e+02 2.746e+02 3.239e+02 3.710e+02 7.459e+02, threshold=6.478e+02, percent-clipped=1.0 2023-05-16 12:22:56,795 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-05-16 12:22:57,892 INFO [finetune.py:992] (1/2) Epoch 9, batch 5300, loss[loss=0.1599, simple_loss=0.234, pruned_loss=0.04288, over 11997.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2583, pruned_loss=0.04104, over 2376110.70 frames. ], batch size: 28, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:23:08,762 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203186.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 12:23:33,609 INFO [finetune.py:992] (1/2) Epoch 9, batch 5350, loss[loss=0.1491, simple_loss=0.2243, pruned_loss=0.03696, over 12007.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2589, pruned_loss=0.0412, over 2375715.77 frames. ], batch size: 28, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:23:41,429 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9326, 4.8738, 4.7600, 4.8320, 4.2676, 4.8802, 4.8909, 5.1166], device='cuda:1'), covar=tensor([0.0208, 0.0140, 0.0187, 0.0298, 0.0882, 0.0403, 0.0156, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0192, 0.0185, 0.0241, 0.0237, 0.0212, 0.0171, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 12:23:57,471 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.707e+02 3.219e+02 3.949e+02 9.779e+02, threshold=6.438e+02, percent-clipped=1.0 2023-05-16 12:24:09,265 INFO [finetune.py:992] (1/2) Epoch 9, batch 5400, loss[loss=0.153, simple_loss=0.2339, pruned_loss=0.03608, over 12244.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2592, pruned_loss=0.04145, over 2377378.27 frames. ], batch size: 32, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:24:21,208 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4375, 2.2554, 3.0305, 4.4342, 2.2134, 4.3908, 4.3820, 4.4602], device='cuda:1'), covar=tensor([0.0093, 0.1310, 0.0537, 0.0107, 0.1305, 0.0180, 0.0130, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0203, 0.0184, 0.0116, 0.0190, 0.0177, 0.0174, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:24:23,881 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7702, 4.7122, 4.6067, 4.6622, 4.2834, 4.8219, 4.7245, 4.9782], device='cuda:1'), covar=tensor([0.0215, 0.0152, 0.0217, 0.0330, 0.0826, 0.0303, 0.0174, 0.0202], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0192, 0.0185, 0.0240, 0.0237, 0.0211, 0.0171, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 12:24:44,760 INFO [finetune.py:992] (1/2) Epoch 9, batch 5450, loss[loss=0.1429, simple_loss=0.224, pruned_loss=0.03092, over 12332.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2596, pruned_loss=0.04159, over 2375845.33 frames. ], batch size: 30, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:24:57,045 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5757, 2.2488, 3.1682, 4.5919, 2.4480, 4.4832, 4.4454, 4.6075], device='cuda:1'), covar=tensor([0.0104, 0.1375, 0.0517, 0.0113, 0.1242, 0.0195, 0.0154, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0204, 0.0185, 0.0116, 0.0190, 0.0177, 0.0175, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:25:08,196 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 2.673e+02 3.050e+02 3.823e+02 9.002e+02, threshold=6.100e+02, percent-clipped=2.0 2023-05-16 12:25:19,996 INFO [finetune.py:992] (1/2) Epoch 9, batch 5500, loss[loss=0.1931, simple_loss=0.2861, pruned_loss=0.05009, over 12363.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2592, pruned_loss=0.04118, over 2380362.44 frames. ], batch size: 35, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:25:36,864 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203394.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:25:52,813 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.43 vs. limit=5.0 2023-05-16 12:25:56,688 INFO [finetune.py:992] (1/2) Epoch 9, batch 5550, loss[loss=0.1816, simple_loss=0.2674, pruned_loss=0.04793, over 12274.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2607, pruned_loss=0.04157, over 2377358.24 frames. ], batch size: 37, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:26:11,079 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203441.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:26:20,180 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.800e+02 3.328e+02 3.811e+02 1.182e+03, threshold=6.657e+02, percent-clipped=6.0 2023-05-16 12:26:29,457 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203467.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:26:32,206 INFO [finetune.py:992] (1/2) Epoch 9, batch 5600, loss[loss=0.1616, simple_loss=0.2506, pruned_loss=0.03633, over 12173.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2605, pruned_loss=0.04198, over 2366208.53 frames. ], batch size: 31, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:26:43,014 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203486.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 12:27:07,442 INFO [finetune.py:992] (1/2) Epoch 9, batch 5650, loss[loss=0.168, simple_loss=0.2605, pruned_loss=0.03776, over 12148.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2605, pruned_loss=0.04174, over 2370250.72 frames. ], batch size: 34, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:27:12,652 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203528.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:27:17,380 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203534.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 12:27:31,137 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.190e+02 2.781e+02 3.248e+02 4.061e+02 7.168e+02, threshold=6.495e+02, percent-clipped=2.0 2023-05-16 12:27:44,049 INFO [finetune.py:992] (1/2) Epoch 9, batch 5700, loss[loss=0.1864, simple_loss=0.2811, pruned_loss=0.04584, over 12077.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2604, pruned_loss=0.04148, over 2377782.76 frames. ], batch size: 42, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:28:19,513 INFO [finetune.py:992] (1/2) Epoch 9, batch 5750, loss[loss=0.1686, simple_loss=0.2608, pruned_loss=0.03817, over 12085.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2606, pruned_loss=0.04188, over 2376322.57 frames. ], batch size: 32, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:28:20,449 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2717, 2.4303, 2.9953, 4.2873, 2.2168, 4.3170, 4.2607, 4.3874], device='cuda:1'), covar=tensor([0.0183, 0.1353, 0.0588, 0.0149, 0.1429, 0.0205, 0.0162, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0205, 0.0186, 0.0116, 0.0191, 0.0178, 0.0175, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:28:24,740 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3433, 4.8442, 3.0936, 2.8147, 4.2228, 2.6777, 4.1289, 3.4191], device='cuda:1'), covar=tensor([0.0672, 0.0512, 0.1043, 0.1384, 0.0215, 0.1298, 0.0402, 0.0703], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0252, 0.0176, 0.0198, 0.0138, 0.0179, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:28:42,800 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.683e+02 3.193e+02 4.148e+02 8.960e+02, threshold=6.386e+02, percent-clipped=3.0 2023-05-16 12:28:54,515 INFO [finetune.py:992] (1/2) Epoch 9, batch 5800, loss[loss=0.1461, simple_loss=0.2396, pruned_loss=0.02632, over 12194.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2605, pruned_loss=0.04251, over 2365537.25 frames. ], batch size: 31, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:29:11,754 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203694.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:29:31,174 INFO [finetune.py:992] (1/2) Epoch 9, batch 5850, loss[loss=0.2054, simple_loss=0.2913, pruned_loss=0.05974, over 11635.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2609, pruned_loss=0.04288, over 2367051.07 frames. ], batch size: 48, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:29:41,332 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3679, 4.2029, 4.2393, 4.6558, 3.0156, 4.1157, 2.7360, 4.2671], device='cuda:1'), covar=tensor([0.1521, 0.0678, 0.0924, 0.0569, 0.1108, 0.0580, 0.1701, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0263, 0.0294, 0.0356, 0.0235, 0.0236, 0.0257, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:29:45,440 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203741.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:29:46,066 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203742.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:29:49,815 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203747.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:29:54,424 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 2.770e+02 3.365e+02 3.985e+02 7.334e+02, threshold=6.729e+02, percent-clipped=1.0 2023-05-16 12:30:03,063 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203766.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:30:06,238 INFO [finetune.py:992] (1/2) Epoch 9, batch 5900, loss[loss=0.1738, simple_loss=0.2666, pruned_loss=0.04051, over 11329.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2626, pruned_loss=0.04337, over 2359696.01 frames. ], batch size: 56, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:30:18,969 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203789.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:30:20,436 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3866, 5.0802, 5.3834, 5.3580, 4.5130, 4.5769, 4.7653, 5.1810], device='cuda:1'), covar=tensor([0.0836, 0.1088, 0.0804, 0.0886, 0.3642, 0.2373, 0.0690, 0.1526], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0690, 0.0601, 0.0607, 0.0847, 0.0733, 0.0545, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:30:21,085 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1801, 6.0443, 5.5231, 5.6152, 6.1184, 5.3931, 5.5542, 5.6149], device='cuda:1'), covar=tensor([0.1392, 0.0898, 0.1087, 0.1813, 0.0891, 0.1907, 0.1865, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0480, 0.0383, 0.0436, 0.0455, 0.0432, 0.0388, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:30:32,951 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203808.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:30:42,943 INFO [finetune.py:992] (1/2) Epoch 9, batch 5950, loss[loss=0.1704, simple_loss=0.2677, pruned_loss=0.03653, over 12150.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2626, pruned_loss=0.04317, over 2359578.85 frames. ], batch size: 36, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:30:43,200 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0764, 3.9519, 3.9778, 4.3146, 2.9600, 3.8185, 2.6466, 3.9539], device='cuda:1'), covar=tensor([0.1620, 0.0723, 0.0958, 0.0718, 0.1047, 0.0600, 0.1688, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0261, 0.0293, 0.0354, 0.0233, 0.0235, 0.0256, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:30:44,471 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203823.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:30:47,369 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203827.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:31:02,390 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203847.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:31:07,171 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.843e+02 2.790e+02 3.214e+02 3.808e+02 7.515e+02, threshold=6.427e+02, percent-clipped=3.0 2023-05-16 12:31:10,727 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4669, 5.0166, 5.4640, 4.8120, 5.0974, 4.8259, 5.4861, 5.1125], device='cuda:1'), covar=tensor([0.0242, 0.0331, 0.0207, 0.0210, 0.0307, 0.0299, 0.0170, 0.0258], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0259, 0.0283, 0.0253, 0.0253, 0.0256, 0.0236, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:31:19,273 INFO [finetune.py:992] (1/2) Epoch 9, batch 6000, loss[loss=0.1846, simple_loss=0.2744, pruned_loss=0.04737, over 12338.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2615, pruned_loss=0.04274, over 2364201.16 frames. ], batch size: 36, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:31:19,274 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 12:31:37,759 INFO [finetune.py:1026] (1/2) Epoch 9, validation: loss=0.3191, simple_loss=0.3958, pruned_loss=0.1212, over 1020973.00 frames. 2023-05-16 12:31:37,760 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 12:31:51,951 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2468, 3.4332, 3.1529, 3.0045, 2.8023, 2.4598, 3.4531, 2.2564], device='cuda:1'), covar=tensor([0.0381, 0.0141, 0.0155, 0.0223, 0.0361, 0.0392, 0.0142, 0.0477], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0162, 0.0156, 0.0186, 0.0201, 0.0198, 0.0168, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:31:58,190 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203900.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 12:32:03,702 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203908.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 12:32:13,534 INFO [finetune.py:992] (1/2) Epoch 9, batch 6050, loss[loss=0.1384, simple_loss=0.2241, pruned_loss=0.02633, over 12185.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2618, pruned_loss=0.04267, over 2365952.88 frames. ], batch size: 29, lr: 4.16e-03, grad_scale: 32.0 2023-05-16 12:32:26,236 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0548, 3.3781, 5.2624, 2.8520, 3.0598, 4.0650, 3.6055, 3.8666], device='cuda:1'), covar=tensor([0.0395, 0.1206, 0.0527, 0.1231, 0.1827, 0.1399, 0.1211, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0234, 0.0243, 0.0183, 0.0237, 0.0288, 0.0223, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:32:37,132 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.072e+02 2.790e+02 3.279e+02 3.879e+02 6.411e+02, threshold=6.558e+02, percent-clipped=0.0 2023-05-16 12:32:42,339 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203961.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 12:32:49,133 INFO [finetune.py:992] (1/2) Epoch 9, batch 6100, loss[loss=0.1716, simple_loss=0.2551, pruned_loss=0.04404, over 12085.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2617, pruned_loss=0.04286, over 2369024.91 frames. ], batch size: 32, lr: 4.16e-03, grad_scale: 32.0 2023-05-16 12:33:27,557 INFO [finetune.py:992] (1/2) Epoch 9, batch 6150, loss[loss=0.176, simple_loss=0.2621, pruned_loss=0.04495, over 12278.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2621, pruned_loss=0.04288, over 2374249.55 frames. ], batch size: 37, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:33:32,017 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0184, 5.9746, 5.7671, 5.3215, 5.2458, 5.9358, 5.5015, 5.3467], device='cuda:1'), covar=tensor([0.0747, 0.0970, 0.0637, 0.1562, 0.0571, 0.0750, 0.1723, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0597, 0.0531, 0.0493, 0.0614, 0.0400, 0.0697, 0.0748, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:33:33,593 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0275, 2.4087, 3.6024, 3.0103, 3.4579, 3.1309, 2.4572, 3.4792], device='cuda:1'), covar=tensor([0.0122, 0.0346, 0.0150, 0.0218, 0.0111, 0.0180, 0.0353, 0.0132], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0202, 0.0183, 0.0181, 0.0212, 0.0159, 0.0194, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:33:51,601 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.976e+02 3.458e+02 4.063e+02 7.171e+02, threshold=6.916e+02, percent-clipped=1.0 2023-05-16 12:34:03,555 INFO [finetune.py:992] (1/2) Epoch 9, batch 6200, loss[loss=0.1476, simple_loss=0.2288, pruned_loss=0.03318, over 12348.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2624, pruned_loss=0.04295, over 2362728.31 frames. ], batch size: 30, lr: 4.16e-03, grad_scale: 16.0 2023-05-16 12:34:08,008 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4659, 4.8208, 3.1071, 2.7376, 4.1605, 2.7483, 4.0109, 3.4559], device='cuda:1'), covar=tensor([0.0671, 0.0581, 0.1045, 0.1505, 0.0318, 0.1314, 0.0542, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0254, 0.0175, 0.0199, 0.0139, 0.0179, 0.0196, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:34:08,012 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204077.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:34:19,986 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9774, 6.0057, 5.6896, 5.2547, 5.1876, 5.8847, 5.5111, 5.2739], device='cuda:1'), covar=tensor([0.0817, 0.0780, 0.0674, 0.1448, 0.0629, 0.0763, 0.1570, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0596, 0.0529, 0.0492, 0.0610, 0.0399, 0.0695, 0.0744, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:34:21,545 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2495, 4.5507, 2.7622, 2.4604, 3.9457, 2.4448, 3.8426, 3.0336], device='cuda:1'), covar=tensor([0.0701, 0.0572, 0.1161, 0.1539, 0.0341, 0.1420, 0.0573, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0254, 0.0175, 0.0199, 0.0139, 0.0179, 0.0196, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:34:27,106 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204103.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:34:39,449 INFO [finetune.py:992] (1/2) Epoch 9, batch 6250, loss[loss=0.1383, simple_loss=0.2231, pruned_loss=0.02677, over 12290.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2617, pruned_loss=0.04232, over 2358490.36 frames. ], batch size: 28, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:34:40,228 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204122.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:34:41,001 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204123.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:34:51,828 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204138.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 12:34:58,843 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2722, 4.8527, 5.2805, 4.5891, 4.9066, 4.5683, 5.3229, 4.8888], device='cuda:1'), covar=tensor([0.0326, 0.0405, 0.0286, 0.0263, 0.0403, 0.0384, 0.0235, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0262, 0.0287, 0.0257, 0.0257, 0.0260, 0.0238, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:35:03,516 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 2.845e+02 3.277e+02 3.871e+02 1.275e+03, threshold=6.554e+02, percent-clipped=2.0 2023-05-16 12:35:09,554 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204163.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:35:15,193 INFO [finetune.py:992] (1/2) Epoch 9, batch 6300, loss[loss=0.1549, simple_loss=0.243, pruned_loss=0.03338, over 12091.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2614, pruned_loss=0.04236, over 2356632.04 frames. ], batch size: 32, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:35:15,268 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204171.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:35:20,518 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-05-16 12:35:27,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-16 12:35:32,204 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8516, 4.7850, 4.5672, 4.8749, 3.7844, 4.9485, 4.8482, 4.9613], device='cuda:1'), covar=tensor([0.0259, 0.0171, 0.0231, 0.0295, 0.1272, 0.0313, 0.0196, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0191, 0.0184, 0.0239, 0.0237, 0.0211, 0.0169, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 12:35:38,015 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204203.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 12:35:51,312 INFO [finetune.py:992] (1/2) Epoch 9, batch 6350, loss[loss=0.1588, simple_loss=0.251, pruned_loss=0.03329, over 12116.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2605, pruned_loss=0.04212, over 2359542.64 frames. ], batch size: 30, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:35:54,312 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204224.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:36:15,885 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.766e+02 3.267e+02 4.031e+02 1.123e+03, threshold=6.534e+02, percent-clipped=2.0 2023-05-16 12:36:16,727 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204256.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 12:36:27,212 INFO [finetune.py:992] (1/2) Epoch 9, batch 6400, loss[loss=0.2008, simple_loss=0.2815, pruned_loss=0.06009, over 12375.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2612, pruned_loss=0.04234, over 2358787.37 frames. ], batch size: 38, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:36:32,922 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0721, 4.9981, 4.8023, 4.8534, 4.5378, 4.9358, 4.9782, 5.1372], device='cuda:1'), covar=tensor([0.0194, 0.0132, 0.0204, 0.0306, 0.0770, 0.0244, 0.0136, 0.0174], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0191, 0.0184, 0.0239, 0.0237, 0.0210, 0.0169, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 12:36:59,002 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2973, 2.1707, 3.0887, 4.2997, 2.0083, 4.2094, 4.2508, 4.3858], device='cuda:1'), covar=tensor([0.0104, 0.1423, 0.0525, 0.0126, 0.1435, 0.0224, 0.0161, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0203, 0.0182, 0.0116, 0.0188, 0.0177, 0.0173, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:37:02,474 INFO [finetune.py:992] (1/2) Epoch 9, batch 6450, loss[loss=0.1553, simple_loss=0.2291, pruned_loss=0.04069, over 11763.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2615, pruned_loss=0.04241, over 2363583.71 frames. ], batch size: 26, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:37:11,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 12:37:26,652 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.828e+02 3.342e+02 4.052e+02 6.928e+02, threshold=6.683e+02, percent-clipped=1.0 2023-05-16 12:37:36,764 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2800, 2.8632, 3.8310, 3.3489, 3.6187, 3.3485, 2.8256, 3.6547], device='cuda:1'), covar=tensor([0.0114, 0.0297, 0.0141, 0.0211, 0.0153, 0.0185, 0.0307, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0200, 0.0180, 0.0180, 0.0211, 0.0157, 0.0192, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:37:39,324 INFO [finetune.py:992] (1/2) Epoch 9, batch 6500, loss[loss=0.1826, simple_loss=0.278, pruned_loss=0.04355, over 12351.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2604, pruned_loss=0.04178, over 2374584.76 frames. ], batch size: 35, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:37:59,284 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9214, 4.5733, 4.6387, 4.7085, 4.5316, 4.7890, 4.7491, 2.8075], device='cuda:1'), covar=tensor([0.0102, 0.0070, 0.0094, 0.0077, 0.0056, 0.0085, 0.0079, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0075, 0.0078, 0.0071, 0.0058, 0.0088, 0.0077, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:38:00,245 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0411, 5.9863, 5.7601, 5.1702, 5.1068, 5.9016, 5.5420, 5.2696], device='cuda:1'), covar=tensor([0.0577, 0.0783, 0.0593, 0.1503, 0.0681, 0.0676, 0.1485, 0.1099], device='cuda:1'), in_proj_covar=tensor([0.0602, 0.0535, 0.0497, 0.0619, 0.0403, 0.0698, 0.0754, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:38:02,541 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204403.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:38:04,530 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2923, 3.5006, 3.2770, 3.1232, 2.8386, 2.6640, 3.4992, 2.2075], device='cuda:1'), covar=tensor([0.0425, 0.0146, 0.0167, 0.0199, 0.0350, 0.0345, 0.0147, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0161, 0.0156, 0.0185, 0.0201, 0.0197, 0.0167, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:38:15,130 INFO [finetune.py:992] (1/2) Epoch 9, batch 6550, loss[loss=0.1949, simple_loss=0.2809, pruned_loss=0.05439, over 12127.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2611, pruned_loss=0.04217, over 2372239.84 frames. ], batch size: 39, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:38:16,067 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204422.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:38:19,635 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5043, 4.3213, 4.2101, 4.7105, 3.3091, 4.1772, 2.6225, 4.3848], device='cuda:1'), covar=tensor([0.1462, 0.0651, 0.0988, 0.0603, 0.1010, 0.0521, 0.1867, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0265, 0.0296, 0.0358, 0.0238, 0.0238, 0.0261, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:38:23,754 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204433.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 12:38:33,781 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1300, 4.9504, 5.0469, 5.1030, 4.7370, 4.7960, 4.5574, 5.0551], device='cuda:1'), covar=tensor([0.0716, 0.0574, 0.0892, 0.0631, 0.1768, 0.1333, 0.0554, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0696, 0.0606, 0.0617, 0.0849, 0.0737, 0.0546, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:38:36,458 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204451.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:38:39,188 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.755e+02 3.300e+02 4.124e+02 6.976e+02, threshold=6.601e+02, percent-clipped=1.0 2023-05-16 12:38:49,883 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204470.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:38:50,582 INFO [finetune.py:992] (1/2) Epoch 9, batch 6600, loss[loss=0.153, simple_loss=0.2513, pruned_loss=0.0273, over 12146.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2613, pruned_loss=0.0421, over 2372127.59 frames. ], batch size: 36, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:39:14,316 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204503.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:39:26,396 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204519.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:39:27,713 INFO [finetune.py:992] (1/2) Epoch 9, batch 6650, loss[loss=0.1784, simple_loss=0.2708, pruned_loss=0.04297, over 12259.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2605, pruned_loss=0.04185, over 2381740.82 frames. ], batch size: 32, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:39:48,927 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204551.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:39:51,779 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.747e+02 3.214e+02 3.903e+02 6.374e+02, threshold=6.428e+02, percent-clipped=0.0 2023-05-16 12:39:52,639 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204556.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 12:39:58,218 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1194, 5.9710, 5.5550, 5.4791, 6.0414, 5.3850, 5.5528, 5.5051], device='cuda:1'), covar=tensor([0.1308, 0.0773, 0.0829, 0.1740, 0.0855, 0.1811, 0.1598, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0483, 0.0381, 0.0432, 0.0457, 0.0431, 0.0385, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:40:00,143 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 12:40:03,099 INFO [finetune.py:992] (1/2) Epoch 9, batch 6700, loss[loss=0.2724, simple_loss=0.3441, pruned_loss=0.1003, over 8386.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2603, pruned_loss=0.04169, over 2377702.04 frames. ], batch size: 98, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:40:12,519 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 12:40:26,691 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204604.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 12:40:38,362 INFO [finetune.py:992] (1/2) Epoch 9, batch 6750, loss[loss=0.1604, simple_loss=0.2469, pruned_loss=0.03688, over 12022.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2594, pruned_loss=0.04136, over 2380729.14 frames. ], batch size: 31, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:41:03,409 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.632e+02 3.211e+02 3.961e+02 7.853e+02, threshold=6.423e+02, percent-clipped=1.0 2023-05-16 12:41:15,431 INFO [finetune.py:992] (1/2) Epoch 9, batch 6800, loss[loss=0.1818, simple_loss=0.2722, pruned_loss=0.0457, over 12148.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2598, pruned_loss=0.04138, over 2377738.20 frames. ], batch size: 34, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:41:19,850 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1345, 4.7467, 4.8658, 4.9581, 4.7621, 4.9523, 4.8874, 2.7152], device='cuda:1'), covar=tensor([0.0094, 0.0073, 0.0080, 0.0062, 0.0047, 0.0092, 0.0063, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0080, 0.0072, 0.0059, 0.0089, 0.0079, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:41:50,818 INFO [finetune.py:992] (1/2) Epoch 9, batch 6850, loss[loss=0.196, simple_loss=0.285, pruned_loss=0.05352, over 12008.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2603, pruned_loss=0.04128, over 2386205.68 frames. ], batch size: 40, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:41:59,732 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204733.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 12:42:02,603 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3380, 4.6187, 2.8036, 2.5163, 3.9048, 2.6287, 3.9184, 3.2566], device='cuda:1'), covar=tensor([0.0623, 0.0426, 0.1118, 0.1467, 0.0319, 0.1248, 0.0464, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0254, 0.0176, 0.0199, 0.0139, 0.0181, 0.0196, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:42:08,243 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5870, 2.3692, 3.2703, 4.4846, 2.2931, 4.5559, 4.5134, 4.6968], device='cuda:1'), covar=tensor([0.0097, 0.1238, 0.0398, 0.0152, 0.1247, 0.0195, 0.0139, 0.0078], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0182, 0.0115, 0.0187, 0.0176, 0.0172, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:42:15,224 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.697e+02 3.191e+02 3.818e+02 1.361e+03, threshold=6.382e+02, percent-clipped=3.0 2023-05-16 12:42:26,681 INFO [finetune.py:992] (1/2) Epoch 9, batch 6900, loss[loss=0.1666, simple_loss=0.253, pruned_loss=0.04011, over 12301.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2596, pruned_loss=0.04082, over 2384942.29 frames. ], batch size: 34, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:42:29,157 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2793, 4.0939, 4.1876, 4.5006, 3.0793, 3.9681, 2.5405, 4.2998], device='cuda:1'), covar=tensor([0.1543, 0.0747, 0.0879, 0.0565, 0.1084, 0.0568, 0.1825, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0263, 0.0292, 0.0355, 0.0235, 0.0237, 0.0257, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:42:34,017 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204781.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:43:02,357 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204819.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:43:03,636 INFO [finetune.py:992] (1/2) Epoch 9, batch 6950, loss[loss=0.1678, simple_loss=0.2524, pruned_loss=0.04154, over 12030.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2598, pruned_loss=0.0409, over 2380785.56 frames. ], batch size: 31, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:43:03,822 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204821.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:43:09,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.29 vs. limit=5.0 2023-05-16 12:43:27,950 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.567e+02 3.121e+02 3.938e+02 7.292e+02, threshold=6.242e+02, percent-clipped=1.0 2023-05-16 12:43:36,583 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204867.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:43:39,311 INFO [finetune.py:992] (1/2) Epoch 9, batch 7000, loss[loss=0.2366, simple_loss=0.3089, pruned_loss=0.08217, over 8119.00 frames. ], tot_loss[loss=0.171, simple_loss=0.26, pruned_loss=0.04106, over 2381350.37 frames. ], batch size: 100, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:43:47,445 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204882.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:44:15,059 INFO [finetune.py:992] (1/2) Epoch 9, batch 7050, loss[loss=0.1475, simple_loss=0.2317, pruned_loss=0.03166, over 12026.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2593, pruned_loss=0.04109, over 2382555.03 frames. ], batch size: 31, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:44:39,897 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.664e+02 3.244e+02 3.963e+02 8.511e+02, threshold=6.488e+02, percent-clipped=4.0 2023-05-16 12:44:51,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-05-16 12:44:52,104 INFO [finetune.py:992] (1/2) Epoch 9, batch 7100, loss[loss=0.1792, simple_loss=0.2697, pruned_loss=0.04433, over 12175.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2599, pruned_loss=0.04117, over 2375747.61 frames. ], batch size: 35, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:45:27,685 INFO [finetune.py:992] (1/2) Epoch 9, batch 7150, loss[loss=0.1577, simple_loss=0.2326, pruned_loss=0.0414, over 12289.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2593, pruned_loss=0.04126, over 2378260.76 frames. ], batch size: 28, lr: 4.15e-03, grad_scale: 16.0 2023-05-16 12:45:51,585 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.953e+02 2.688e+02 3.284e+02 3.844e+02 9.605e+02, threshold=6.568e+02, percent-clipped=3.0 2023-05-16 12:46:02,884 INFO [finetune.py:992] (1/2) Epoch 9, batch 7200, loss[loss=0.1874, simple_loss=0.2828, pruned_loss=0.04599, over 12017.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2599, pruned_loss=0.04148, over 2376588.78 frames. ], batch size: 40, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:46:31,256 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205109.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:46:37,751 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 12:46:39,122 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-05-16 12:46:39,388 INFO [finetune.py:992] (1/2) Epoch 9, batch 7250, loss[loss=0.1849, simple_loss=0.2719, pruned_loss=0.04897, over 12029.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2597, pruned_loss=0.04165, over 2366281.95 frames. ], batch size: 40, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:47:03,422 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.817e+02 3.407e+02 3.876e+02 9.349e+02, threshold=6.813e+02, percent-clipped=5.0 2023-05-16 12:47:07,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 12:47:14,319 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205170.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 12:47:14,819 INFO [finetune.py:992] (1/2) Epoch 9, batch 7300, loss[loss=0.1727, simple_loss=0.2722, pruned_loss=0.03661, over 12137.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2596, pruned_loss=0.04169, over 2372635.61 frames. ], batch size: 39, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:47:19,198 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205177.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:47:34,157 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205198.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:47:50,521 INFO [finetune.py:992] (1/2) Epoch 9, batch 7350, loss[loss=0.1469, simple_loss=0.228, pruned_loss=0.03288, over 11834.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2594, pruned_loss=0.04166, over 2361734.63 frames. ], batch size: 26, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:48:16,018 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.746e+02 3.298e+02 4.043e+02 7.027e+02, threshold=6.595e+02, percent-clipped=1.0 2023-05-16 12:48:19,160 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205259.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:48:27,522 INFO [finetune.py:992] (1/2) Epoch 9, batch 7400, loss[loss=0.1482, simple_loss=0.2318, pruned_loss=0.03233, over 12086.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2594, pruned_loss=0.04173, over 2358140.28 frames. ], batch size: 32, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:48:53,899 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2900, 4.1139, 4.0114, 4.4507, 3.1320, 3.9244, 2.7232, 4.2030], device='cuda:1'), covar=tensor([0.1496, 0.0671, 0.0885, 0.0710, 0.1058, 0.0555, 0.1650, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0263, 0.0293, 0.0354, 0.0235, 0.0237, 0.0258, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:48:54,659 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6211, 2.6953, 4.5431, 4.7308, 2.8986, 2.5766, 2.8833, 2.1683], device='cuda:1'), covar=tensor([0.1462, 0.3140, 0.0426, 0.0341, 0.1124, 0.2128, 0.2774, 0.3782], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0382, 0.0271, 0.0295, 0.0265, 0.0297, 0.0369, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:49:02,823 INFO [finetune.py:992] (1/2) Epoch 9, batch 7450, loss[loss=0.1778, simple_loss=0.2506, pruned_loss=0.05247, over 11782.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2586, pruned_loss=0.0414, over 2370048.53 frames. ], batch size: 26, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:49:26,873 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.873e+02 3.463e+02 4.047e+02 7.614e+02, threshold=6.926e+02, percent-clipped=2.0 2023-05-16 12:49:38,173 INFO [finetune.py:992] (1/2) Epoch 9, batch 7500, loss[loss=0.1814, simple_loss=0.2734, pruned_loss=0.04471, over 12357.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2586, pruned_loss=0.04147, over 2375849.86 frames. ], batch size: 36, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:50:15,122 INFO [finetune.py:992] (1/2) Epoch 9, batch 7550, loss[loss=0.1353, simple_loss=0.224, pruned_loss=0.0233, over 11986.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2589, pruned_loss=0.04142, over 2372059.51 frames. ], batch size: 28, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:50:39,122 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.733e+02 3.190e+02 3.709e+02 8.114e+02, threshold=6.380e+02, percent-clipped=1.0 2023-05-16 12:50:46,047 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205465.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 12:50:50,137 INFO [finetune.py:992] (1/2) Epoch 9, batch 7600, loss[loss=0.178, simple_loss=0.2642, pruned_loss=0.04585, over 12370.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2604, pruned_loss=0.04229, over 2367071.91 frames. ], batch size: 38, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:50:52,010 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-05-16 12:50:54,530 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205477.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:51:24,166 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3361, 4.6428, 2.8578, 2.5024, 3.9624, 2.3792, 4.0110, 3.0429], device='cuda:1'), covar=tensor([0.0697, 0.0480, 0.1169, 0.1691, 0.0281, 0.1496, 0.0430, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0259, 0.0178, 0.0202, 0.0141, 0.0183, 0.0200, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:51:26,109 INFO [finetune.py:992] (1/2) Epoch 9, batch 7650, loss[loss=0.1496, simple_loss=0.2318, pruned_loss=0.03377, over 12235.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2593, pruned_loss=0.04191, over 2370047.94 frames. ], batch size: 28, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:51:28,896 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205525.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:51:29,731 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2430, 5.1492, 5.1219, 5.0903, 4.7823, 5.1601, 5.1669, 5.4389], device='cuda:1'), covar=tensor([0.0190, 0.0123, 0.0164, 0.0313, 0.0672, 0.0304, 0.0167, 0.0143], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0191, 0.0185, 0.0239, 0.0237, 0.0211, 0.0168, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 12:51:50,147 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205554.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:51:50,748 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.107e+02 2.771e+02 3.229e+02 3.973e+02 7.489e+02, threshold=6.459e+02, percent-clipped=3.0 2023-05-16 12:51:55,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.30 vs. limit=5.0 2023-05-16 12:51:56,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6496, 3.7885, 3.4194, 3.2741, 2.9559, 2.9373, 3.7152, 2.3582], device='cuda:1'), covar=tensor([0.0318, 0.0123, 0.0182, 0.0168, 0.0353, 0.0282, 0.0108, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0160, 0.0154, 0.0183, 0.0200, 0.0197, 0.0166, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:52:02,276 INFO [finetune.py:992] (1/2) Epoch 9, batch 7700, loss[loss=0.1643, simple_loss=0.25, pruned_loss=0.03926, over 12125.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2599, pruned_loss=0.04219, over 2368752.14 frames. ], batch size: 30, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:52:38,368 INFO [finetune.py:992] (1/2) Epoch 9, batch 7750, loss[loss=0.1464, simple_loss=0.2312, pruned_loss=0.03074, over 12169.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2604, pruned_loss=0.04231, over 2364538.46 frames. ], batch size: 31, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:52:39,314 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1733, 1.9914, 2.3970, 2.1435, 2.2899, 2.3029, 1.9143, 2.3148], device='cuda:1'), covar=tensor([0.0108, 0.0281, 0.0163, 0.0180, 0.0144, 0.0164, 0.0237, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0202, 0.0184, 0.0183, 0.0214, 0.0159, 0.0195, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:53:02,431 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.764e+02 3.338e+02 3.986e+02 7.043e+02, threshold=6.676e+02, percent-clipped=1.0 2023-05-16 12:53:02,593 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9348, 3.9047, 3.8219, 4.0058, 3.7740, 3.7317, 3.6922, 3.8774], device='cuda:1'), covar=tensor([0.0978, 0.0651, 0.1192, 0.0666, 0.1548, 0.1277, 0.0541, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0692, 0.0602, 0.0617, 0.0840, 0.0732, 0.0540, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:53:02,669 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205655.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:53:14,380 INFO [finetune.py:992] (1/2) Epoch 9, batch 7800, loss[loss=0.1732, simple_loss=0.2663, pruned_loss=0.04006, over 12279.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2604, pruned_loss=0.04213, over 2367136.00 frames. ], batch size: 37, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:53:29,270 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6205, 4.8557, 3.2878, 3.0030, 4.2029, 2.9268, 4.1500, 3.5031], device='cuda:1'), covar=tensor([0.0586, 0.0459, 0.0972, 0.1272, 0.0258, 0.1207, 0.0482, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0259, 0.0177, 0.0201, 0.0141, 0.0182, 0.0199, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:53:46,730 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205716.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:53:50,024 INFO [finetune.py:992] (1/2) Epoch 9, batch 7850, loss[loss=0.1458, simple_loss=0.2326, pruned_loss=0.02946, over 11370.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04259, over 2361352.92 frames. ], batch size: 25, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:54:06,980 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205744.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:54:14,491 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 2.844e+02 3.312e+02 4.065e+02 1.026e+03, threshold=6.624e+02, percent-clipped=4.0 2023-05-16 12:54:21,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 12:54:21,788 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205765.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:54:23,394 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8919, 2.3601, 3.6162, 3.0310, 3.4681, 3.0524, 2.3292, 3.4615], device='cuda:1'), covar=tensor([0.0159, 0.0458, 0.0143, 0.0236, 0.0158, 0.0208, 0.0441, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0200, 0.0183, 0.0181, 0.0213, 0.0159, 0.0194, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:54:26,057 INFO [finetune.py:992] (1/2) Epoch 9, batch 7900, loss[loss=0.1779, simple_loss=0.2657, pruned_loss=0.045, over 12103.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2622, pruned_loss=0.04282, over 2354740.18 frames. ], batch size: 33, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:54:51,155 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205805.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:54:57,395 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205813.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:55:02,868 INFO [finetune.py:992] (1/2) Epoch 9, batch 7950, loss[loss=0.1409, simple_loss=0.2349, pruned_loss=0.02343, over 12187.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2622, pruned_loss=0.04299, over 2352827.80 frames. ], batch size: 31, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:55:26,359 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:55:26,946 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.901e+02 2.807e+02 3.233e+02 3.735e+02 6.641e+02, threshold=6.465e+02, percent-clipped=1.0 2023-05-16 12:55:38,085 INFO [finetune.py:992] (1/2) Epoch 9, batch 8000, loss[loss=0.1913, simple_loss=0.2818, pruned_loss=0.05046, over 12156.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2621, pruned_loss=0.04261, over 2363654.62 frames. ], batch size: 34, lr: 4.14e-03, grad_scale: 16.0 2023-05-16 12:55:51,909 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205890.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:56:00,348 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205902.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:56:13,912 INFO [finetune.py:992] (1/2) Epoch 9, batch 8050, loss[loss=0.1662, simple_loss=0.2558, pruned_loss=0.03827, over 12297.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2616, pruned_loss=0.04285, over 2362100.63 frames. ], batch size: 33, lr: 4.14e-03, grad_scale: 8.0 2023-05-16 12:56:35,760 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205951.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:56:40,444 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.735e+02 3.249e+02 3.846e+02 1.067e+03, threshold=6.497e+02, percent-clipped=3.0 2023-05-16 12:56:40,626 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0961, 6.0716, 5.8713, 5.3642, 5.2929, 6.0694, 5.5575, 5.4315], device='cuda:1'), covar=tensor([0.0645, 0.0913, 0.0710, 0.1644, 0.0627, 0.0751, 0.1739, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0539, 0.0502, 0.0622, 0.0404, 0.0701, 0.0755, 0.0556], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 12:56:51,076 INFO [finetune.py:992] (1/2) Epoch 9, batch 8100, loss[loss=0.1993, simple_loss=0.2936, pruned_loss=0.05249, over 12067.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2622, pruned_loss=0.04321, over 2354434.36 frames. ], batch size: 42, lr: 4.14e-03, grad_scale: 8.0 2023-05-16 12:57:07,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 12:57:22,338 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206011.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:57:29,339 INFO [finetune.py:992] (1/2) Epoch 9, batch 8150, loss[loss=0.1638, simple_loss=0.2602, pruned_loss=0.03368, over 12361.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2621, pruned_loss=0.0431, over 2359593.03 frames. ], batch size: 36, lr: 4.14e-03, grad_scale: 8.0 2023-05-16 12:57:49,281 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-16 12:57:54,552 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.743e+02 3.254e+02 3.827e+02 7.721e+02, threshold=6.508e+02, percent-clipped=1.0 2023-05-16 12:57:54,799 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206056.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:58:05,124 INFO [finetune.py:992] (1/2) Epoch 9, batch 8200, loss[loss=0.1713, simple_loss=0.2488, pruned_loss=0.04693, over 12011.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2617, pruned_loss=0.04297, over 2356656.12 frames. ], batch size: 28, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 12:58:06,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 12:58:22,410 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4746, 4.7210, 3.1651, 2.7290, 4.0685, 2.9154, 4.0740, 3.3207], device='cuda:1'), covar=tensor([0.0630, 0.0592, 0.1028, 0.1374, 0.0317, 0.1096, 0.0449, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0260, 0.0179, 0.0201, 0.0142, 0.0182, 0.0199, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 12:58:27,303 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206100.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:58:39,590 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206117.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:58:42,274 INFO [finetune.py:992] (1/2) Epoch 9, batch 8250, loss[loss=0.1455, simple_loss=0.2373, pruned_loss=0.02683, over 12092.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.261, pruned_loss=0.04255, over 2362943.42 frames. ], batch size: 32, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 12:58:48,066 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5559, 2.6494, 3.6482, 4.5081, 3.8413, 4.5161, 3.8934, 3.2981], device='cuda:1'), covar=tensor([0.0032, 0.0363, 0.0132, 0.0037, 0.0116, 0.0067, 0.0091, 0.0325], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0124, 0.0106, 0.0076, 0.0103, 0.0116, 0.0093, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 12:58:50,058 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1030, 5.0534, 4.9235, 5.0720, 4.0140, 5.1867, 5.0469, 5.2966], device='cuda:1'), covar=tensor([0.0302, 0.0194, 0.0229, 0.0347, 0.1263, 0.0502, 0.0213, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0193, 0.0186, 0.0240, 0.0238, 0.0213, 0.0169, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 12:59:03,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-16 12:59:07,279 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 2.848e+02 3.348e+02 4.019e+02 6.771e+02, threshold=6.696e+02, percent-clipped=1.0 2023-05-16 12:59:14,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 12:59:17,906 INFO [finetune.py:992] (1/2) Epoch 9, batch 8300, loss[loss=0.1691, simple_loss=0.2657, pruned_loss=0.03628, over 12290.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04264, over 2359824.99 frames. ], batch size: 37, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 12:59:35,704 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 12:59:36,833 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206197.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 12:59:47,875 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7801, 2.3392, 3.1951, 2.7088, 3.0823, 2.8875, 2.2729, 3.1284], device='cuda:1'), covar=tensor([0.0130, 0.0303, 0.0168, 0.0204, 0.0150, 0.0165, 0.0312, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0200, 0.0183, 0.0181, 0.0213, 0.0159, 0.0194, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 12:59:53,902 INFO [finetune.py:992] (1/2) Epoch 9, batch 8350, loss[loss=0.1511, simple_loss=0.2373, pruned_loss=0.03239, over 12123.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2608, pruned_loss=0.04257, over 2352735.93 frames. ], batch size: 30, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 12:59:57,277 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 13:00:13,260 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206246.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:00:20,438 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 2.699e+02 3.145e+02 4.092e+02 1.329e+03, threshold=6.291e+02, percent-clipped=4.0 2023-05-16 13:00:22,081 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206258.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:00:30,942 INFO [finetune.py:992] (1/2) Epoch 9, batch 8400, loss[loss=0.1884, simple_loss=0.2911, pruned_loss=0.04288, over 11456.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2614, pruned_loss=0.04255, over 2354797.18 frames. ], batch size: 55, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:00:46,781 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3313, 4.8502, 5.2674, 4.6354, 4.9589, 4.7348, 5.3053, 4.9244], device='cuda:1'), covar=tensor([0.0231, 0.0391, 0.0280, 0.0242, 0.0343, 0.0285, 0.0208, 0.0310], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0260, 0.0282, 0.0253, 0.0253, 0.0254, 0.0232, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:00:52,734 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7073, 2.7213, 4.0042, 4.1479, 2.8420, 2.6248, 2.7165, 2.1450], device='cuda:1'), covar=tensor([0.1368, 0.2669, 0.0526, 0.0472, 0.1174, 0.2113, 0.2549, 0.3766], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0381, 0.0271, 0.0292, 0.0263, 0.0296, 0.0368, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:00:59,777 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206311.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:01:06,768 INFO [finetune.py:992] (1/2) Epoch 9, batch 8450, loss[loss=0.1593, simple_loss=0.2552, pruned_loss=0.03167, over 12087.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2613, pruned_loss=0.04223, over 2365986.21 frames. ], batch size: 32, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:01:31,347 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 2.753e+02 3.191e+02 4.256e+02 8.777e+02, threshold=6.383e+02, percent-clipped=2.0 2023-05-16 13:01:33,620 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206359.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:01:42,111 INFO [finetune.py:992] (1/2) Epoch 9, batch 8500, loss[loss=0.236, simple_loss=0.3142, pruned_loss=0.07892, over 7996.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2613, pruned_loss=0.0423, over 2364828.70 frames. ], batch size: 98, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:02:04,523 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206400.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:02:13,035 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206412.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:02:19,388 INFO [finetune.py:992] (1/2) Epoch 9, batch 8550, loss[loss=0.1715, simple_loss=0.2555, pruned_loss=0.04378, over 12352.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2611, pruned_loss=0.04196, over 2366100.31 frames. ], batch size: 35, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:02:23,788 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9268, 3.3624, 5.2377, 2.7025, 2.7078, 3.9662, 3.4737, 4.0677], device='cuda:1'), covar=tensor([0.0396, 0.1135, 0.0306, 0.1165, 0.1947, 0.1342, 0.1242, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0228, 0.0240, 0.0179, 0.0232, 0.0285, 0.0219, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:02:38,228 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206448.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:02:43,930 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.816e+02 3.212e+02 3.831e+02 6.403e+02, threshold=6.425e+02, percent-clipped=1.0 2023-05-16 13:02:48,351 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5793, 2.2571, 2.9330, 2.6235, 2.8373, 2.7566, 2.2064, 2.9076], device='cuda:1'), covar=tensor([0.0116, 0.0309, 0.0182, 0.0194, 0.0146, 0.0178, 0.0322, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0200, 0.0184, 0.0182, 0.0213, 0.0160, 0.0195, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:02:54,505 INFO [finetune.py:992] (1/2) Epoch 9, batch 8600, loss[loss=0.1482, simple_loss=0.232, pruned_loss=0.03219, over 12339.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2608, pruned_loss=0.04186, over 2368481.78 frames. ], batch size: 30, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:03:10,253 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9510, 4.6003, 4.2253, 4.2609, 4.6557, 3.9873, 4.2880, 4.1410], device='cuda:1'), covar=tensor([0.1597, 0.1122, 0.1403, 0.1926, 0.1224, 0.2291, 0.1780, 0.1478], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0483, 0.0383, 0.0431, 0.0461, 0.0434, 0.0390, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:03:30,178 INFO [finetune.py:992] (1/2) Epoch 9, batch 8650, loss[loss=0.1626, simple_loss=0.2564, pruned_loss=0.03436, over 11266.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2608, pruned_loss=0.04178, over 2362456.83 frames. ], batch size: 55, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:03:34,860 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 13:03:39,480 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0823, 5.9116, 5.4973, 5.3843, 5.9736, 5.2395, 5.4767, 5.5014], device='cuda:1'), covar=tensor([0.1286, 0.0849, 0.0908, 0.1854, 0.0868, 0.2102, 0.1776, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0482, 0.0382, 0.0431, 0.0459, 0.0433, 0.0388, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:03:49,471 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206546.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:03:54,486 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206553.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:03:56,480 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.768e+02 3.154e+02 3.662e+02 7.102e+02, threshold=6.307e+02, percent-clipped=1.0 2023-05-16 13:04:07,008 INFO [finetune.py:992] (1/2) Epoch 9, batch 8700, loss[loss=0.1879, simple_loss=0.2784, pruned_loss=0.0487, over 12059.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2605, pruned_loss=0.0417, over 2362583.55 frames. ], batch size: 37, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:04:13,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 13:04:23,456 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206594.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:04:43,022 INFO [finetune.py:992] (1/2) Epoch 9, batch 8750, loss[loss=0.1591, simple_loss=0.2505, pruned_loss=0.03384, over 12256.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2604, pruned_loss=0.04171, over 2360201.88 frames. ], batch size: 32, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:04:46,747 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3277, 2.4204, 3.0908, 4.2956, 2.0271, 4.2524, 4.2439, 4.3537], device='cuda:1'), covar=tensor([0.0118, 0.1243, 0.0496, 0.0131, 0.1531, 0.0242, 0.0179, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0204, 0.0185, 0.0116, 0.0190, 0.0178, 0.0174, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:05:07,594 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.751e+02 3.304e+02 4.035e+02 8.012e+02, threshold=6.608e+02, percent-clipped=2.0 2023-05-16 13:05:18,985 INFO [finetune.py:992] (1/2) Epoch 9, batch 8800, loss[loss=0.1596, simple_loss=0.2463, pruned_loss=0.03645, over 12188.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2605, pruned_loss=0.04215, over 2361276.66 frames. ], batch size: 31, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:05:49,067 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206712.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:05:55,256 INFO [finetune.py:992] (1/2) Epoch 9, batch 8850, loss[loss=0.1637, simple_loss=0.2572, pruned_loss=0.03507, over 12267.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2599, pruned_loss=0.04165, over 2367770.69 frames. ], batch size: 37, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:06:20,284 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.742e+02 3.107e+02 3.864e+02 8.527e+02, threshold=6.215e+02, percent-clipped=1.0 2023-05-16 13:06:23,215 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206760.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:06:26,198 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6201, 2.5973, 3.2494, 4.6244, 2.1617, 4.5762, 4.5543, 4.6509], device='cuda:1'), covar=tensor([0.0136, 0.1265, 0.0471, 0.0108, 0.1580, 0.0177, 0.0150, 0.0078], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0203, 0.0185, 0.0116, 0.0190, 0.0177, 0.0174, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:06:26,867 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4304, 4.9772, 5.4400, 4.7179, 5.0782, 4.7813, 5.4921, 5.1023], device='cuda:1'), covar=tensor([0.0267, 0.0365, 0.0232, 0.0255, 0.0352, 0.0316, 0.0156, 0.0237], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0260, 0.0280, 0.0252, 0.0253, 0.0254, 0.0232, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:06:30,907 INFO [finetune.py:992] (1/2) Epoch 9, batch 8900, loss[loss=0.1573, simple_loss=0.2406, pruned_loss=0.03702, over 12111.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2599, pruned_loss=0.04161, over 2372702.93 frames. ], batch size: 30, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:06:56,214 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5174, 3.6287, 3.1418, 3.1022, 2.8152, 2.7035, 3.5862, 2.3931], device='cuda:1'), covar=tensor([0.0374, 0.0128, 0.0216, 0.0217, 0.0398, 0.0391, 0.0149, 0.0468], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0161, 0.0157, 0.0186, 0.0201, 0.0199, 0.0167, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:07:07,236 INFO [finetune.py:992] (1/2) Epoch 9, batch 8950, loss[loss=0.1983, simple_loss=0.2906, pruned_loss=0.05304, over 12361.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2599, pruned_loss=0.04171, over 2367264.81 frames. ], batch size: 38, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:07:17,227 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206834.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:07:27,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-05-16 13:07:30,523 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206853.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:07:32,401 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.757e+02 3.302e+02 3.769e+02 8.409e+02, threshold=6.604e+02, percent-clipped=4.0 2023-05-16 13:07:43,053 INFO [finetune.py:992] (1/2) Epoch 9, batch 9000, loss[loss=0.1792, simple_loss=0.2689, pruned_loss=0.04472, over 11876.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2603, pruned_loss=0.0419, over 2378288.08 frames. ], batch size: 44, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:07:43,053 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 13:07:51,155 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1007, 4.0518, 2.5684, 2.0345, 3.6042, 2.2553, 3.6984, 2.6237], device='cuda:1'), covar=tensor([0.0647, 0.0446, 0.1086, 0.1775, 0.0246, 0.1362, 0.0367, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0256, 0.0176, 0.0198, 0.0140, 0.0180, 0.0196, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:08:01,993 INFO [finetune.py:1026] (1/2) Epoch 9, validation: loss=0.3325, simple_loss=0.4032, pruned_loss=0.1308, over 1020973.00 frames. 2023-05-16 13:08:01,994 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 13:08:16,247 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6292, 2.6007, 3.2741, 4.5743, 2.2399, 4.5509, 4.5611, 4.6460], device='cuda:1'), covar=tensor([0.0104, 0.1119, 0.0405, 0.0116, 0.1275, 0.0189, 0.0137, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0201, 0.0183, 0.0115, 0.0187, 0.0175, 0.0172, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:08:18,917 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206895.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 13:08:23,064 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206901.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:08:37,885 INFO [finetune.py:992] (1/2) Epoch 9, batch 9050, loss[loss=0.1593, simple_loss=0.2449, pruned_loss=0.03684, over 11766.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2614, pruned_loss=0.04233, over 2374645.51 frames. ], batch size: 26, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:09:03,320 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.978e+02 3.533e+02 4.335e+02 8.865e+02, threshold=7.066e+02, percent-clipped=4.0 2023-05-16 13:09:13,739 INFO [finetune.py:992] (1/2) Epoch 9, batch 9100, loss[loss=0.1567, simple_loss=0.2337, pruned_loss=0.03987, over 12176.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2608, pruned_loss=0.04234, over 2371570.93 frames. ], batch size: 29, lr: 4.13e-03, grad_scale: 8.0 2023-05-16 13:09:48,941 INFO [finetune.py:992] (1/2) Epoch 9, batch 9150, loss[loss=0.1344, simple_loss=0.2236, pruned_loss=0.02262, over 12173.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2606, pruned_loss=0.04228, over 2371629.31 frames. ], batch size: 29, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:10:14,434 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.033e+02 2.676e+02 3.136e+02 3.998e+02 8.990e+02, threshold=6.272e+02, percent-clipped=3.0 2023-05-16 13:10:25,182 INFO [finetune.py:992] (1/2) Epoch 9, batch 9200, loss[loss=0.1746, simple_loss=0.2676, pruned_loss=0.04075, over 12023.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2608, pruned_loss=0.04197, over 2373563.87 frames. ], batch size: 40, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:10:36,358 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2844, 2.6256, 3.8362, 3.2287, 3.6637, 3.3547, 2.7798, 3.7052], device='cuda:1'), covar=tensor([0.0113, 0.0334, 0.0127, 0.0196, 0.0120, 0.0156, 0.0323, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0202, 0.0185, 0.0183, 0.0215, 0.0160, 0.0197, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:10:45,744 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0001, 3.0455, 4.4030, 2.3369, 2.4707, 3.3822, 2.9846, 3.5447], device='cuda:1'), covar=tensor([0.0577, 0.1153, 0.0415, 0.1241, 0.1905, 0.1550, 0.1414, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0231, 0.0243, 0.0180, 0.0235, 0.0289, 0.0222, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:10:58,309 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2745, 4.8703, 5.0013, 5.0840, 4.8910, 5.1663, 4.9897, 3.1598], device='cuda:1'), covar=tensor([0.0076, 0.0060, 0.0072, 0.0060, 0.0050, 0.0086, 0.0073, 0.0543], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0075, 0.0078, 0.0071, 0.0058, 0.0088, 0.0077, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:11:01,768 INFO [finetune.py:992] (1/2) Epoch 9, batch 9250, loss[loss=0.1698, simple_loss=0.2664, pruned_loss=0.03653, over 12181.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2601, pruned_loss=0.04219, over 2362145.94 frames. ], batch size: 35, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:11:26,265 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.738e+02 3.250e+02 4.161e+02 6.723e+02, threshold=6.501e+02, percent-clipped=1.0 2023-05-16 13:11:36,825 INFO [finetune.py:992] (1/2) Epoch 9, batch 9300, loss[loss=0.2427, simple_loss=0.3203, pruned_loss=0.08256, over 7902.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2609, pruned_loss=0.04241, over 2355764.48 frames. ], batch size: 97, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:11:40,533 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207176.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:11:50,475 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207190.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 13:12:13,435 INFO [finetune.py:992] (1/2) Epoch 9, batch 9350, loss[loss=0.1783, simple_loss=0.2751, pruned_loss=0.0407, over 12282.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2599, pruned_loss=0.04206, over 2351382.12 frames. ], batch size: 37, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:12:25,750 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207237.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:12:39,169 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.762e+02 3.159e+02 4.154e+02 6.329e+02, threshold=6.318e+02, percent-clipped=0.0 2023-05-16 13:12:49,862 INFO [finetune.py:992] (1/2) Epoch 9, batch 9400, loss[loss=0.1623, simple_loss=0.2535, pruned_loss=0.03558, over 12096.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2599, pruned_loss=0.04173, over 2356081.88 frames. ], batch size: 33, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:12:59,040 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5234, 4.3805, 4.3255, 4.4266, 3.7397, 4.6170, 4.5102, 4.6935], device='cuda:1'), covar=tensor([0.0279, 0.0202, 0.0242, 0.0344, 0.1136, 0.0325, 0.0195, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0195, 0.0187, 0.0241, 0.0239, 0.0212, 0.0171, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 13:13:25,065 INFO [finetune.py:992] (1/2) Epoch 9, batch 9450, loss[loss=0.1666, simple_loss=0.2615, pruned_loss=0.03578, over 12374.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2592, pruned_loss=0.04129, over 2363004.08 frames. ], batch size: 38, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:13:34,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 13:13:43,967 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3254, 5.1546, 5.2925, 5.3213, 4.9384, 4.9815, 4.7416, 5.2675], device='cuda:1'), covar=tensor([0.0759, 0.0557, 0.0652, 0.0585, 0.1750, 0.1290, 0.0532, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0693, 0.0596, 0.0608, 0.0835, 0.0727, 0.0541, 0.0474], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 13:13:50,125 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.085e+02 2.743e+02 3.330e+02 3.955e+02 7.509e+02, threshold=6.659e+02, percent-clipped=3.0 2023-05-16 13:13:52,529 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5061, 4.7998, 2.9784, 2.6032, 4.1001, 2.7676, 4.1015, 3.5060], device='cuda:1'), covar=tensor([0.0627, 0.0478, 0.1090, 0.1589, 0.0248, 0.1239, 0.0434, 0.0683], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0257, 0.0177, 0.0200, 0.0141, 0.0181, 0.0197, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:13:55,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0141, 4.9236, 4.8516, 4.8353, 4.5676, 5.0097, 5.0035, 5.2334], device='cuda:1'), covar=tensor([0.0208, 0.0157, 0.0190, 0.0294, 0.0707, 0.0317, 0.0134, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0195, 0.0187, 0.0241, 0.0239, 0.0213, 0.0171, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 13:13:59,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4604, 2.6887, 3.5269, 4.4467, 3.8305, 4.4522, 3.8371, 3.4232], device='cuda:1'), covar=tensor([0.0035, 0.0337, 0.0158, 0.0036, 0.0121, 0.0054, 0.0100, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0126, 0.0107, 0.0077, 0.0104, 0.0116, 0.0095, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:14:00,988 INFO [finetune.py:992] (1/2) Epoch 9, batch 9500, loss[loss=0.1418, simple_loss=0.2205, pruned_loss=0.03151, over 12241.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2589, pruned_loss=0.04087, over 2361464.95 frames. ], batch size: 28, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:14:10,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7067, 2.2194, 2.9729, 3.7386, 2.1076, 3.8499, 3.7209, 3.8320], device='cuda:1'), covar=tensor([0.0149, 0.1136, 0.0456, 0.0141, 0.1268, 0.0209, 0.0250, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0201, 0.0185, 0.0115, 0.0188, 0.0175, 0.0172, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:14:26,591 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7267, 2.8232, 4.6518, 4.7840, 2.8741, 2.6864, 2.9416, 2.0986], device='cuda:1'), covar=tensor([0.1580, 0.3034, 0.0434, 0.0399, 0.1347, 0.2300, 0.2746, 0.4092], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0381, 0.0272, 0.0295, 0.0263, 0.0295, 0.0370, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:14:29,042 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 13:14:37,860 INFO [finetune.py:992] (1/2) Epoch 9, batch 9550, loss[loss=0.1705, simple_loss=0.2628, pruned_loss=0.03906, over 12340.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2595, pruned_loss=0.04072, over 2361143.50 frames. ], batch size: 36, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:14:47,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 13:15:02,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.035e+02 2.805e+02 3.276e+02 4.126e+02 1.039e+03, threshold=6.552e+02, percent-clipped=2.0 2023-05-16 13:15:13,492 INFO [finetune.py:992] (1/2) Epoch 9, batch 9600, loss[loss=0.1728, simple_loss=0.2661, pruned_loss=0.03979, over 12265.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2599, pruned_loss=0.04107, over 2360535.98 frames. ], batch size: 37, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:15:27,441 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207490.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 13:15:29,710 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5708, 2.4624, 3.2999, 4.6028, 2.3705, 4.6144, 4.5971, 4.6454], device='cuda:1'), covar=tensor([0.0123, 0.1312, 0.0476, 0.0104, 0.1389, 0.0209, 0.0136, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0202, 0.0186, 0.0116, 0.0189, 0.0176, 0.0173, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:15:31,874 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9274, 3.2520, 2.4202, 2.2132, 2.8332, 2.3266, 3.1145, 2.6979], device='cuda:1'), covar=tensor([0.0563, 0.0803, 0.0850, 0.1400, 0.0317, 0.1107, 0.0488, 0.0704], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0257, 0.0176, 0.0199, 0.0141, 0.0180, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:15:49,497 INFO [finetune.py:992] (1/2) Epoch 9, batch 9650, loss[loss=0.1692, simple_loss=0.2559, pruned_loss=0.04121, over 12332.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.259, pruned_loss=0.0411, over 2361057.82 frames. ], batch size: 31, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:15:58,194 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207532.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:16:02,196 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=207538.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:16:14,942 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.667e+02 3.231e+02 3.855e+02 7.021e+02, threshold=6.462e+02, percent-clipped=2.0 2023-05-16 13:16:18,006 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3472, 5.2050, 5.2770, 5.3509, 4.9488, 5.0016, 4.8058, 5.2350], device='cuda:1'), covar=tensor([0.0684, 0.0532, 0.0773, 0.0537, 0.1827, 0.1176, 0.0513, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0688, 0.0595, 0.0605, 0.0831, 0.0725, 0.0540, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:16:18,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 13:16:25,739 INFO [finetune.py:992] (1/2) Epoch 9, batch 9700, loss[loss=0.1832, simple_loss=0.2709, pruned_loss=0.04777, over 12115.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2587, pruned_loss=0.0407, over 2369797.76 frames. ], batch size: 38, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:17:01,872 INFO [finetune.py:992] (1/2) Epoch 9, batch 9750, loss[loss=0.1673, simple_loss=0.2527, pruned_loss=0.04096, over 12146.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2593, pruned_loss=0.04107, over 2371700.40 frames. ], batch size: 39, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:17:26,415 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.711e+02 3.133e+02 4.067e+02 8.021e+02, threshold=6.267e+02, percent-clipped=3.0 2023-05-16 13:17:27,277 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9361, 5.9366, 5.6976, 5.2429, 5.1079, 5.8560, 5.4416, 5.2092], device='cuda:1'), covar=tensor([0.0745, 0.0835, 0.0679, 0.1795, 0.0744, 0.0787, 0.1556, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0547, 0.0506, 0.0623, 0.0410, 0.0707, 0.0758, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 13:17:37,213 INFO [finetune.py:992] (1/2) Epoch 9, batch 9800, loss[loss=0.2379, simple_loss=0.3104, pruned_loss=0.08273, over 7950.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2595, pruned_loss=0.04138, over 2371123.67 frames. ], batch size: 97, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:18:07,498 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-05-16 13:18:13,365 INFO [finetune.py:992] (1/2) Epoch 9, batch 9850, loss[loss=0.1672, simple_loss=0.263, pruned_loss=0.03568, over 12350.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2596, pruned_loss=0.04152, over 2363974.91 frames. ], batch size: 35, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:18:15,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 13:18:23,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-05-16 13:18:38,473 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 2.674e+02 3.274e+02 3.799e+02 1.142e+03, threshold=6.547e+02, percent-clipped=1.0 2023-05-16 13:18:49,853 INFO [finetune.py:992] (1/2) Epoch 9, batch 9900, loss[loss=0.17, simple_loss=0.256, pruned_loss=0.04197, over 11807.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2597, pruned_loss=0.04123, over 2371478.62 frames. ], batch size: 44, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:19:26,962 INFO [finetune.py:992] (1/2) Epoch 9, batch 9950, loss[loss=0.2013, simple_loss=0.2788, pruned_loss=0.06188, over 12051.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2584, pruned_loss=0.04067, over 2376289.71 frames. ], batch size: 37, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:19:34,917 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207832.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:19:51,787 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.069e+02 2.891e+02 3.446e+02 3.978e+02 7.477e+02, threshold=6.892e+02, percent-clipped=1.0 2023-05-16 13:19:53,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 13:20:02,356 INFO [finetune.py:992] (1/2) Epoch 9, batch 10000, loss[loss=0.1614, simple_loss=0.2503, pruned_loss=0.03631, over 12257.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2591, pruned_loss=0.04073, over 2378304.09 frames. ], batch size: 32, lr: 4.12e-03, grad_scale: 8.0 2023-05-16 13:20:08,836 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=207880.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:20:12,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 13:20:36,553 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4388, 3.6871, 3.1738, 3.1124, 2.8022, 2.7010, 3.7243, 2.3374], device='cuda:1'), covar=tensor([0.0415, 0.0145, 0.0220, 0.0227, 0.0465, 0.0423, 0.0125, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0162, 0.0157, 0.0186, 0.0201, 0.0198, 0.0168, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:20:38,511 INFO [finetune.py:992] (1/2) Epoch 9, batch 10050, loss[loss=0.1609, simple_loss=0.2461, pruned_loss=0.03782, over 12025.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.259, pruned_loss=0.04066, over 2384886.69 frames. ], batch size: 31, lr: 4.12e-03, grad_scale: 16.0 2023-05-16 13:20:51,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-05-16 13:21:03,752 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.712e+02 2.996e+02 4.038e+02 9.495e+02, threshold=5.992e+02, percent-clipped=2.0 2023-05-16 13:21:15,015 INFO [finetune.py:992] (1/2) Epoch 9, batch 10100, loss[loss=0.1574, simple_loss=0.2498, pruned_loss=0.03256, over 12284.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2593, pruned_loss=0.04103, over 2372762.08 frames. ], batch size: 33, lr: 4.12e-03, grad_scale: 16.0 2023-05-16 13:21:26,341 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0196, 4.9953, 4.8245, 5.0601, 3.9961, 5.0664, 5.1115, 5.1872], device='cuda:1'), covar=tensor([0.0344, 0.0171, 0.0239, 0.0297, 0.1163, 0.0381, 0.0173, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0192, 0.0186, 0.0238, 0.0236, 0.0211, 0.0170, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 13:21:32,014 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1144, 5.0475, 4.8846, 4.9848, 4.6179, 5.0466, 5.1287, 5.2719], device='cuda:1'), covar=tensor([0.0265, 0.0134, 0.0225, 0.0291, 0.0737, 0.0330, 0.0151, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0192, 0.0186, 0.0238, 0.0235, 0.0211, 0.0170, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 13:21:50,197 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7905, 5.7654, 5.5291, 4.8865, 4.9703, 5.6640, 5.2894, 5.0143], device='cuda:1'), covar=tensor([0.0746, 0.0925, 0.0696, 0.1607, 0.0769, 0.0753, 0.1560, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0599, 0.0536, 0.0497, 0.0611, 0.0402, 0.0691, 0.0745, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 13:21:53,583 INFO [finetune.py:992] (1/2) Epoch 9, batch 10150, loss[loss=0.1445, simple_loss=0.2303, pruned_loss=0.02934, over 12353.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2599, pruned_loss=0.04112, over 2375100.67 frames. ], batch size: 31, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:22:01,313 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208032.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:22:19,191 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.079e+02 2.808e+02 3.338e+02 3.944e+02 5.870e+02, threshold=6.676e+02, percent-clipped=0.0 2023-05-16 13:22:30,138 INFO [finetune.py:992] (1/2) Epoch 9, batch 10200, loss[loss=0.1505, simple_loss=0.2315, pruned_loss=0.03475, over 12165.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2598, pruned_loss=0.04117, over 2373592.51 frames. ], batch size: 29, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:22:43,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 13:22:46,089 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208093.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 13:23:05,535 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2881, 2.5980, 3.7032, 3.1170, 3.4565, 3.2348, 2.6200, 3.6208], device='cuda:1'), covar=tensor([0.0136, 0.0356, 0.0143, 0.0233, 0.0169, 0.0180, 0.0365, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0206, 0.0189, 0.0187, 0.0217, 0.0163, 0.0200, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:23:06,070 INFO [finetune.py:992] (1/2) Epoch 9, batch 10250, loss[loss=0.1572, simple_loss=0.253, pruned_loss=0.03072, over 12275.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2597, pruned_loss=0.04086, over 2379673.67 frames. ], batch size: 32, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:23:22,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-05-16 13:23:30,897 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.613e+02 3.110e+02 3.784e+02 7.618e+02, threshold=6.219e+02, percent-clipped=3.0 2023-05-16 13:23:41,553 INFO [finetune.py:992] (1/2) Epoch 9, batch 10300, loss[loss=0.1471, simple_loss=0.2324, pruned_loss=0.03094, over 12131.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2593, pruned_loss=0.04071, over 2386870.26 frames. ], batch size: 30, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:24:17,374 INFO [finetune.py:992] (1/2) Epoch 9, batch 10350, loss[loss=0.1759, simple_loss=0.2662, pruned_loss=0.04282, over 10261.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2589, pruned_loss=0.04083, over 2382707.35 frames. ], batch size: 68, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:24:21,272 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1114, 4.7545, 4.9220, 4.9161, 4.6919, 4.9795, 4.8111, 2.5847], device='cuda:1'), covar=tensor([0.0092, 0.0065, 0.0070, 0.0060, 0.0051, 0.0094, 0.0079, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0080, 0.0073, 0.0059, 0.0090, 0.0079, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:24:28,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-05-16 13:24:42,887 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.816e+02 3.307e+02 4.221e+02 9.934e+02, threshold=6.615e+02, percent-clipped=3.0 2023-05-16 13:24:49,469 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208265.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:24:53,614 INFO [finetune.py:992] (1/2) Epoch 9, batch 10400, loss[loss=0.1815, simple_loss=0.271, pruned_loss=0.04605, over 12041.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2587, pruned_loss=0.04071, over 2378472.03 frames. ], batch size: 40, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:25:29,365 INFO [finetune.py:992] (1/2) Epoch 9, batch 10450, loss[loss=0.167, simple_loss=0.2568, pruned_loss=0.03857, over 12037.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.258, pruned_loss=0.0402, over 2383715.40 frames. ], batch size: 31, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:25:33,079 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208326.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:25:38,682 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0675, 5.9404, 5.3151, 5.4585, 5.9859, 5.2549, 5.4514, 5.5605], device='cuda:1'), covar=tensor([0.1412, 0.0895, 0.1016, 0.1664, 0.1014, 0.2378, 0.1668, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0481, 0.0379, 0.0430, 0.0457, 0.0435, 0.0385, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:25:54,373 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3321, 4.6547, 2.8885, 2.5107, 4.0153, 2.3862, 4.0008, 3.2397], device='cuda:1'), covar=tensor([0.0648, 0.0492, 0.1165, 0.1576, 0.0264, 0.1407, 0.0443, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0254, 0.0175, 0.0195, 0.0138, 0.0178, 0.0193, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:25:54,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.807e+02 3.294e+02 3.880e+02 7.412e+02, threshold=6.587e+02, percent-clipped=1.0 2023-05-16 13:26:05,563 INFO [finetune.py:992] (1/2) Epoch 9, batch 10500, loss[loss=0.1445, simple_loss=0.226, pruned_loss=0.03149, over 12192.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2585, pruned_loss=0.04028, over 2381561.52 frames. ], batch size: 31, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:26:18,481 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208388.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 13:26:22,768 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1615, 4.7567, 4.9391, 4.9373, 4.8502, 5.0085, 4.9664, 3.0052], device='cuda:1'), covar=tensor([0.0076, 0.0073, 0.0076, 0.0064, 0.0047, 0.0086, 0.0067, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0077, 0.0079, 0.0072, 0.0059, 0.0089, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:26:42,091 INFO [finetune.py:992] (1/2) Epoch 9, batch 10550, loss[loss=0.1512, simple_loss=0.2366, pruned_loss=0.03288, over 12240.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2587, pruned_loss=0.04024, over 2379642.46 frames. ], batch size: 32, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:26:54,961 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8721, 3.8364, 3.8391, 3.9476, 3.5408, 3.5054, 3.6464, 3.8116], device='cuda:1'), covar=tensor([0.1318, 0.1092, 0.1675, 0.0957, 0.2519, 0.2354, 0.0738, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0704, 0.0603, 0.0614, 0.0842, 0.0743, 0.0550, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 13:27:06,868 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 2.736e+02 3.181e+02 3.822e+02 7.078e+02, threshold=6.362e+02, percent-clipped=1.0 2023-05-16 13:27:17,669 INFO [finetune.py:992] (1/2) Epoch 9, batch 10600, loss[loss=0.1646, simple_loss=0.2581, pruned_loss=0.03551, over 12112.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2583, pruned_loss=0.04017, over 2379177.41 frames. ], batch size: 42, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:27:37,901 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4815, 3.6283, 3.1772, 3.1467, 2.8693, 2.6682, 3.6441, 2.1766], device='cuda:1'), covar=tensor([0.0392, 0.0121, 0.0209, 0.0208, 0.0317, 0.0372, 0.0126, 0.0475], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0162, 0.0159, 0.0188, 0.0201, 0.0199, 0.0170, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:27:53,866 INFO [finetune.py:992] (1/2) Epoch 9, batch 10650, loss[loss=0.2495, simple_loss=0.3256, pruned_loss=0.08669, over 8357.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2584, pruned_loss=0.03997, over 2378470.54 frames. ], batch size: 98, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:28:05,167 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1320, 5.0082, 4.9735, 5.0557, 4.5838, 5.0267, 5.1533, 5.3745], device='cuda:1'), covar=tensor([0.0212, 0.0153, 0.0188, 0.0313, 0.0798, 0.0292, 0.0149, 0.0148], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0192, 0.0185, 0.0237, 0.0237, 0.0210, 0.0171, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 13:28:19,137 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.612e+02 3.191e+02 3.895e+02 7.455e+02, threshold=6.381e+02, percent-clipped=2.0 2023-05-16 13:28:25,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 13:28:29,722 INFO [finetune.py:992] (1/2) Epoch 9, batch 10700, loss[loss=0.2029, simple_loss=0.2871, pruned_loss=0.05936, over 12067.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2588, pruned_loss=0.04044, over 2385389.90 frames. ], batch size: 42, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:28:33,548 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208576.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:29:03,852 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2023-05-16 13:29:05,671 INFO [finetune.py:992] (1/2) Epoch 9, batch 10750, loss[loss=0.1565, simple_loss=0.2421, pruned_loss=0.03546, over 12178.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.258, pruned_loss=0.04032, over 2387856.43 frames. ], batch size: 31, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:29:05,798 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:29:17,154 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208637.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:29:30,957 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.664e+02 3.149e+02 3.821e+02 8.730e+02, threshold=6.297e+02, percent-clipped=2.0 2023-05-16 13:29:42,441 INFO [finetune.py:992] (1/2) Epoch 9, batch 10800, loss[loss=0.1922, simple_loss=0.2768, pruned_loss=0.05382, over 10318.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2581, pruned_loss=0.04086, over 2371299.45 frames. ], batch size: 68, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:29:54,415 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:30:11,558 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9916, 2.3186, 3.3878, 4.0515, 3.6112, 3.9672, 3.7406, 2.7293], device='cuda:1'), covar=tensor([0.0055, 0.0408, 0.0169, 0.0038, 0.0118, 0.0088, 0.0103, 0.0396], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0128, 0.0110, 0.0078, 0.0107, 0.0120, 0.0097, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:30:17,704 INFO [finetune.py:992] (1/2) Epoch 9, batch 10850, loss[loss=0.1961, simple_loss=0.282, pruned_loss=0.05512, over 12150.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2596, pruned_loss=0.04153, over 2361798.13 frames. ], batch size: 34, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:30:26,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 13:30:28,595 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208736.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:30:43,098 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.187e+02 2.779e+02 3.243e+02 4.269e+02 7.545e+02, threshold=6.485e+02, percent-clipped=7.0 2023-05-16 13:30:54,150 INFO [finetune.py:992] (1/2) Epoch 9, batch 10900, loss[loss=0.1679, simple_loss=0.2631, pruned_loss=0.03631, over 12194.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.26, pruned_loss=0.04204, over 2346557.04 frames. ], batch size: 35, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:31:12,761 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2741, 2.6480, 3.8791, 3.1421, 3.6164, 3.3243, 2.6306, 3.7121], device='cuda:1'), covar=tensor([0.0131, 0.0367, 0.0130, 0.0252, 0.0118, 0.0166, 0.0359, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0202, 0.0185, 0.0184, 0.0213, 0.0160, 0.0196, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:31:26,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-05-16 13:31:31,674 INFO [finetune.py:992] (1/2) Epoch 9, batch 10950, loss[loss=0.1585, simple_loss=0.2443, pruned_loss=0.0364, over 12364.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2606, pruned_loss=0.04217, over 2343279.81 frames. ], batch size: 30, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:31:55,983 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.795e+02 3.337e+02 4.140e+02 1.225e+03, threshold=6.674e+02, percent-clipped=5.0 2023-05-16 13:32:06,896 INFO [finetune.py:992] (1/2) Epoch 9, batch 11000, loss[loss=0.1725, simple_loss=0.2592, pruned_loss=0.04289, over 12276.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2631, pruned_loss=0.04358, over 2326287.26 frames. ], batch size: 37, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:32:23,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3216, 3.1777, 3.1949, 3.5028, 2.6332, 3.1930, 2.5756, 3.0846], device='cuda:1'), covar=tensor([0.1378, 0.0812, 0.0815, 0.0553, 0.0926, 0.0693, 0.1421, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0265, 0.0295, 0.0354, 0.0234, 0.0239, 0.0259, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:32:42,686 INFO [finetune.py:992] (1/2) Epoch 9, batch 11050, loss[loss=0.2309, simple_loss=0.3005, pruned_loss=0.0807, over 8250.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2662, pruned_loss=0.04532, over 2293123.33 frames. ], batch size: 97, lr: 4.11e-03, grad_scale: 16.0 2023-05-16 13:32:42,825 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208921.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:32:44,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 13:32:51,057 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208932.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:32:51,917 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3830, 3.2247, 3.1645, 3.5360, 2.6276, 3.1439, 2.4401, 3.0045], device='cuda:1'), covar=tensor([0.1374, 0.0804, 0.0876, 0.0665, 0.1043, 0.0733, 0.1588, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0265, 0.0295, 0.0355, 0.0234, 0.0240, 0.0259, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:33:08,141 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 3.079e+02 3.769e+02 4.599e+02 8.043e+02, threshold=7.539e+02, percent-clipped=5.0 2023-05-16 13:33:17,302 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208969.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:33:18,543 INFO [finetune.py:992] (1/2) Epoch 9, batch 11100, loss[loss=0.2482, simple_loss=0.315, pruned_loss=0.09068, over 7925.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2708, pruned_loss=0.04841, over 2238928.68 frames. ], batch size: 98, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:33:55,066 INFO [finetune.py:992] (1/2) Epoch 9, batch 11150, loss[loss=0.229, simple_loss=0.3077, pruned_loss=0.07511, over 11149.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2768, pruned_loss=0.05269, over 2171803.43 frames. ], batch size: 55, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:34:07,689 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 13:34:19,852 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.408e+02 3.523e+02 4.250e+02 5.223e+02 9.989e+02, threshold=8.500e+02, percent-clipped=4.0 2023-05-16 13:34:30,304 INFO [finetune.py:992] (1/2) Epoch 9, batch 11200, loss[loss=0.2445, simple_loss=0.3354, pruned_loss=0.07684, over 11040.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2831, pruned_loss=0.05675, over 2126949.08 frames. ], batch size: 55, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:34:42,335 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7021, 3.4733, 3.5053, 3.6678, 3.3584, 3.7585, 3.7590, 3.8065], device='cuda:1'), covar=tensor([0.0223, 0.0195, 0.0214, 0.0391, 0.0631, 0.0354, 0.0197, 0.0287], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0188, 0.0182, 0.0234, 0.0232, 0.0206, 0.0168, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 13:35:06,102 INFO [finetune.py:992] (1/2) Epoch 9, batch 11250, loss[loss=0.1836, simple_loss=0.2763, pruned_loss=0.04551, over 12265.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2892, pruned_loss=0.06079, over 2075503.31 frames. ], batch size: 33, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:35:22,554 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7275, 2.8446, 4.1824, 4.3532, 2.9738, 2.6931, 2.9292, 2.0131], device='cuda:1'), covar=tensor([0.1417, 0.2558, 0.0477, 0.0394, 0.1090, 0.2198, 0.2360, 0.4076], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0378, 0.0270, 0.0294, 0.0262, 0.0294, 0.0367, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:35:30,521 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 3.451e+02 3.991e+02 4.600e+02 9.298e+02, threshold=7.983e+02, percent-clipped=1.0 2023-05-16 13:35:41,553 INFO [finetune.py:992] (1/2) Epoch 9, batch 11300, loss[loss=0.2325, simple_loss=0.3173, pruned_loss=0.07391, over 11241.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2953, pruned_loss=0.06491, over 2021763.83 frames. ], batch size: 55, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:35:57,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-05-16 13:36:03,287 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209202.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:36:04,105 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1637, 1.9804, 2.3418, 2.0918, 2.2502, 2.2192, 1.7507, 2.2383], device='cuda:1'), covar=tensor([0.0086, 0.0227, 0.0113, 0.0155, 0.0127, 0.0121, 0.0226, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0198, 0.0180, 0.0180, 0.0208, 0.0155, 0.0192, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:36:16,462 INFO [finetune.py:992] (1/2) Epoch 9, batch 11350, loss[loss=0.2759, simple_loss=0.3498, pruned_loss=0.101, over 6887.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.3007, pruned_loss=0.06834, over 1957240.73 frames. ], batch size: 99, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:36:24,015 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209232.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:36:40,335 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209255.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:36:40,830 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.590e+02 3.481e+02 4.181e+02 4.952e+02 7.214e+02, threshold=8.362e+02, percent-clipped=0.0 2023-05-16 13:36:41,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9291, 2.3148, 2.2325, 2.1082, 2.0170, 2.0521, 2.0844, 1.6958], device='cuda:1'), covar=tensor([0.0271, 0.0145, 0.0146, 0.0183, 0.0233, 0.0184, 0.0155, 0.0342], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0157, 0.0155, 0.0183, 0.0196, 0.0196, 0.0167, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:36:43,676 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2733, 4.5799, 2.7267, 2.3285, 3.9565, 2.1975, 4.0087, 2.8969], device='cuda:1'), covar=tensor([0.0715, 0.0412, 0.1194, 0.1799, 0.0245, 0.1659, 0.0412, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0248, 0.0173, 0.0193, 0.0137, 0.0176, 0.0191, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:36:45,598 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209263.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:36:46,311 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3477, 3.0362, 2.9180, 3.3505, 2.6022, 3.0806, 2.5532, 2.7267], device='cuda:1'), covar=tensor([0.1556, 0.0923, 0.0741, 0.0541, 0.1014, 0.0823, 0.1428, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0264, 0.0293, 0.0351, 0.0234, 0.0238, 0.0258, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:36:50,715 INFO [finetune.py:992] (1/2) Epoch 9, batch 11400, loss[loss=0.256, simple_loss=0.329, pruned_loss=0.0915, over 7357.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.305, pruned_loss=0.0713, over 1910170.89 frames. ], batch size: 98, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:36:56,886 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209280.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:37:02,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-05-16 13:37:09,699 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209298.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:37:22,649 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209316.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:37:25,615 INFO [finetune.py:992] (1/2) Epoch 9, batch 11450, loss[loss=0.2624, simple_loss=0.3234, pruned_loss=0.1008, over 6405.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.309, pruned_loss=0.0746, over 1866688.67 frames. ], batch size: 98, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:37:50,140 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.446e+02 3.472e+02 3.871e+02 4.791e+02 8.062e+02, threshold=7.742e+02, percent-clipped=0.0 2023-05-16 13:37:52,330 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209359.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:38:00,954 INFO [finetune.py:992] (1/2) Epoch 9, batch 11500, loss[loss=0.3138, simple_loss=0.3695, pruned_loss=0.129, over 7014.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3121, pruned_loss=0.07701, over 1832339.03 frames. ], batch size: 98, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:38:35,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4999, 3.2915, 3.4143, 3.5558, 3.5493, 3.6862, 3.4736, 2.6642], device='cuda:1'), covar=tensor([0.0095, 0.0106, 0.0155, 0.0076, 0.0072, 0.0109, 0.0087, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0077, 0.0069, 0.0057, 0.0086, 0.0076, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:38:35,486 INFO [finetune.py:992] (1/2) Epoch 9, batch 11550, loss[loss=0.2912, simple_loss=0.3499, pruned_loss=0.1162, over 6857.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.313, pruned_loss=0.07789, over 1812589.19 frames. ], batch size: 99, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:38:59,548 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.537e+02 3.719e+02 4.229e+02 5.064e+02 1.141e+03, threshold=8.457e+02, percent-clipped=4.0 2023-05-16 13:39:06,521 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209466.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:39:09,700 INFO [finetune.py:992] (1/2) Epoch 9, batch 11600, loss[loss=0.2216, simple_loss=0.3025, pruned_loss=0.07038, over 11580.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3148, pruned_loss=0.08, over 1788427.05 frames. ], batch size: 48, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:39:38,516 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209510.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:39:46,479 INFO [finetune.py:992] (1/2) Epoch 9, batch 11650, loss[loss=0.2101, simple_loss=0.3026, pruned_loss=0.05879, over 12193.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.314, pruned_loss=0.08008, over 1769062.76 frames. ], batch size: 35, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:39:51,108 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209527.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:40:02,332 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3949, 4.0959, 4.0566, 4.2769, 4.0877, 4.3434, 4.3033, 2.4037], device='cuda:1'), covar=tensor([0.0095, 0.0092, 0.0159, 0.0075, 0.0075, 0.0115, 0.0084, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0076, 0.0069, 0.0057, 0.0086, 0.0075, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:40:07,183 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1375, 1.9493, 2.1721, 2.0317, 2.1330, 2.2207, 1.8090, 2.1717], device='cuda:1'), covar=tensor([0.0085, 0.0244, 0.0112, 0.0179, 0.0151, 0.0133, 0.0250, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0193, 0.0173, 0.0174, 0.0199, 0.0151, 0.0186, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:40:10,932 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.471e+02 3.255e+02 3.824e+02 4.485e+02 1.562e+03, threshold=7.649e+02, percent-clipped=1.0 2023-05-16 13:40:13,120 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209558.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:40:21,317 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2443, 3.4526, 3.1996, 3.0584, 2.8943, 2.7580, 3.2144, 1.9232], device='cuda:1'), covar=tensor([0.0397, 0.0123, 0.0140, 0.0179, 0.0297, 0.0301, 0.0185, 0.0557], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0155, 0.0152, 0.0181, 0.0194, 0.0194, 0.0164, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:40:21,779 INFO [finetune.py:992] (1/2) Epoch 9, batch 11700, loss[loss=0.1999, simple_loss=0.2924, pruned_loss=0.05371, over 11523.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3134, pruned_loss=0.07992, over 1758574.50 frames. ], batch size: 48, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:40:22,004 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209571.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:40:32,647 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2521, 3.4807, 3.2055, 3.4789, 3.3301, 2.4767, 3.1216, 2.8076], device='cuda:1'), covar=tensor([0.0967, 0.0916, 0.1412, 0.0823, 0.1504, 0.1893, 0.1281, 0.3063], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0357, 0.0332, 0.0275, 0.0347, 0.0254, 0.0322, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:40:46,173 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209606.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:40:47,451 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5918, 3.3527, 3.5427, 3.6195, 3.5472, 3.6882, 3.5420, 2.6274], device='cuda:1'), covar=tensor([0.0103, 0.0117, 0.0139, 0.0081, 0.0083, 0.0117, 0.0106, 0.0748], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0073, 0.0076, 0.0068, 0.0056, 0.0085, 0.0074, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:40:49,321 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209611.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:40:56,651 INFO [finetune.py:992] (1/2) Epoch 9, batch 11750, loss[loss=0.2536, simple_loss=0.3218, pruned_loss=0.09272, over 6983.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3133, pruned_loss=0.08076, over 1732397.78 frames. ], batch size: 98, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:41:19,751 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209654.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:41:20,963 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.757e+02 3.765e+02 4.358e+02 5.471e+02 1.515e+03, threshold=8.716e+02, percent-clipped=4.0 2023-05-16 13:41:28,579 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209667.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:41:31,084 INFO [finetune.py:992] (1/2) Epoch 9, batch 11800, loss[loss=0.2301, simple_loss=0.3128, pruned_loss=0.07367, over 10967.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3155, pruned_loss=0.08237, over 1721485.50 frames. ], batch size: 55, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:41:43,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.52 vs. limit=5.0 2023-05-16 13:41:43,513 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:42:06,036 INFO [finetune.py:992] (1/2) Epoch 9, batch 11850, loss[loss=0.2213, simple_loss=0.3099, pruned_loss=0.06639, over 10164.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3172, pruned_loss=0.08264, over 1711422.88 frames. ], batch size: 68, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:42:06,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-16 13:42:24,178 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5682, 3.4016, 3.5167, 3.5868, 3.5804, 3.7116, 3.5810, 2.7026], device='cuda:1'), covar=tensor([0.0094, 0.0098, 0.0136, 0.0082, 0.0071, 0.0105, 0.0093, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0072, 0.0075, 0.0068, 0.0056, 0.0084, 0.0074, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:42:25,393 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209749.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:42:29,823 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.633e+02 3.558e+02 4.060e+02 4.711e+02 6.540e+02, threshold=8.119e+02, percent-clipped=0.0 2023-05-16 13:42:39,979 INFO [finetune.py:992] (1/2) Epoch 9, batch 11900, loss[loss=0.2899, simple_loss=0.3534, pruned_loss=0.1132, over 6940.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3169, pruned_loss=0.082, over 1696012.30 frames. ], batch size: 99, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:43:15,114 INFO [finetune.py:992] (1/2) Epoch 9, batch 11950, loss[loss=0.1893, simple_loss=0.2922, pruned_loss=0.04325, over 10630.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3144, pruned_loss=0.08, over 1681378.43 frames. ], batch size: 69, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:43:15,969 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209822.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:43:26,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-05-16 13:43:39,977 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.202e+02 3.119e+02 3.617e+02 4.335e+02 9.745e+02, threshold=7.234e+02, percent-clipped=4.0 2023-05-16 13:43:41,507 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209858.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:43:47,166 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209866.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:43:50,512 INFO [finetune.py:992] (1/2) Epoch 9, batch 12000, loss[loss=0.2031, simple_loss=0.2885, pruned_loss=0.05884, over 6978.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3089, pruned_loss=0.0757, over 1675580.50 frames. ], batch size: 97, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:43:50,513 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 13:44:04,382 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7465, 2.4329, 3.3389, 3.3873, 2.6770, 2.6088, 2.4293, 2.3335], device='cuda:1'), covar=tensor([0.1173, 0.2870, 0.0592, 0.0524, 0.0927, 0.2119, 0.3206, 0.3788], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0366, 0.0262, 0.0284, 0.0256, 0.0289, 0.0360, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:44:08,689 INFO [finetune.py:1026] (1/2) Epoch 9, validation: loss=0.2842, simple_loss=0.3622, pruned_loss=0.103, over 1020973.00 frames. 2023-05-16 13:44:08,690 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 13:44:18,997 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8185, 4.4847, 4.1005, 4.1937, 4.5096, 3.9646, 4.2069, 4.0748], device='cuda:1'), covar=tensor([0.1640, 0.1010, 0.1063, 0.1745, 0.0955, 0.2011, 0.1454, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0455, 0.0364, 0.0405, 0.0433, 0.0410, 0.0363, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 13:44:33,569 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209906.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:44:37,038 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209911.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:44:43,495 INFO [finetune.py:992] (1/2) Epoch 9, batch 12050, loss[loss=0.1975, simple_loss=0.2854, pruned_loss=0.05481, over 11566.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3037, pruned_loss=0.07127, over 1707209.07 frames. ], batch size: 48, lr: 4.10e-03, grad_scale: 16.0 2023-05-16 13:44:51,089 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3736, 2.8865, 3.7050, 2.3308, 2.6099, 2.9683, 2.8727, 3.0857], device='cuda:1'), covar=tensor([0.0460, 0.1098, 0.0322, 0.1265, 0.1808, 0.1424, 0.1190, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0223, 0.0225, 0.0174, 0.0224, 0.0273, 0.0211, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:45:05,373 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209954.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:45:07,176 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 2.944e+02 3.345e+02 3.905e+02 1.431e+03, threshold=6.691e+02, percent-clipped=3.0 2023-05-16 13:45:08,591 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209959.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:45:08,752 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7086, 2.3309, 2.9693, 2.6149, 2.8503, 2.8440, 2.0917, 2.9453], device='cuda:1'), covar=tensor([0.0107, 0.0322, 0.0122, 0.0230, 0.0135, 0.0159, 0.0367, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0187, 0.0166, 0.0167, 0.0192, 0.0145, 0.0181, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:45:10,493 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209962.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:45:16,046 INFO [finetune.py:992] (1/2) Epoch 9, batch 12100, loss[loss=0.224, simple_loss=0.2929, pruned_loss=0.0776, over 6620.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.3023, pruned_loss=0.07037, over 1687829.02 frames. ], batch size: 101, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:45:39,081 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210002.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:45:39,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-05-16 13:45:51,166 INFO [finetune.py:992] (1/2) Epoch 9, batch 12150, loss[loss=0.2184, simple_loss=0.3115, pruned_loss=0.06262, over 11177.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3035, pruned_loss=0.07058, over 1708908.74 frames. ], batch size: 55, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:46:05,595 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210044.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:46:13,787 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.148e+02 3.271e+02 3.941e+02 4.882e+02 1.262e+03, threshold=7.882e+02, percent-clipped=10.0 2023-05-16 13:46:13,982 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210057.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:46:22,208 INFO [finetune.py:992] (1/2) Epoch 9, batch 12200, loss[loss=0.2561, simple_loss=0.3331, pruned_loss=0.08949, over 7090.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3053, pruned_loss=0.07229, over 1682412.70 frames. ], batch size: 98, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:47:06,196 INFO [finetune.py:992] (1/2) Epoch 10, batch 0, loss[loss=0.1647, simple_loss=0.256, pruned_loss=0.03668, over 12098.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.256, pruned_loss=0.03668, over 12098.00 frames. ], batch size: 32, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:47:06,197 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 13:47:23,506 INFO [finetune.py:1026] (1/2) Epoch 10, validation: loss=0.2866, simple_loss=0.3628, pruned_loss=0.1052, over 1020973.00 frames. 2023-05-16 13:47:23,507 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 13:47:32,890 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210118.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:47:35,573 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210122.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:47:40,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-16 13:47:58,748 INFO [finetune.py:992] (1/2) Epoch 10, batch 50, loss[loss=0.1688, simple_loss=0.2657, pruned_loss=0.03591, over 12301.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2722, pruned_loss=0.04675, over 535644.76 frames. ], batch size: 34, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:48:00,023 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 3.186e+02 3.722e+02 4.629e+02 6.830e+02, threshold=7.444e+02, percent-clipped=0.0 2023-05-16 13:48:06,478 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210166.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:48:09,316 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210170.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:48:21,974 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8605, 3.2962, 2.4925, 2.2309, 2.9939, 2.3410, 3.1636, 2.6707], device='cuda:1'), covar=tensor([0.0659, 0.0826, 0.1043, 0.1519, 0.0339, 0.1297, 0.0569, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0237, 0.0170, 0.0191, 0.0132, 0.0174, 0.0184, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:48:27,749 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2159, 5.1299, 4.9821, 5.0486, 4.6585, 5.0776, 5.0594, 5.2739], device='cuda:1'), covar=tensor([0.0156, 0.0120, 0.0170, 0.0310, 0.0682, 0.0299, 0.0155, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0166, 0.0162, 0.0209, 0.0206, 0.0183, 0.0151, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 13:48:34,916 INFO [finetune.py:992] (1/2) Epoch 10, batch 100, loss[loss=0.1553, simple_loss=0.2434, pruned_loss=0.03361, over 12354.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2683, pruned_loss=0.04548, over 942706.55 frames. ], batch size: 31, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:48:41,402 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210214.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:48:46,976 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0995, 2.5112, 3.5710, 3.1221, 3.4955, 3.1312, 2.4798, 3.4828], device='cuda:1'), covar=tensor([0.0144, 0.0351, 0.0135, 0.0223, 0.0150, 0.0196, 0.0372, 0.0151], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0189, 0.0168, 0.0169, 0.0194, 0.0147, 0.0183, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:49:11,062 INFO [finetune.py:992] (1/2) Epoch 10, batch 150, loss[loss=0.1861, simple_loss=0.2832, pruned_loss=0.04447, over 12339.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2672, pruned_loss=0.04456, over 1258354.51 frames. ], batch size: 36, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:49:12,470 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 2.678e+02 3.225e+02 3.794e+02 7.055e+02, threshold=6.451e+02, percent-clipped=0.0 2023-05-16 13:49:16,085 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210262.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:49:36,814 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1610, 3.4620, 3.5148, 3.9424, 2.6989, 3.4053, 2.4617, 3.4765], device='cuda:1'), covar=tensor([0.1702, 0.0929, 0.1096, 0.0773, 0.1227, 0.0791, 0.1996, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0262, 0.0289, 0.0343, 0.0231, 0.0236, 0.0257, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:49:46,200 INFO [finetune.py:992] (1/2) Epoch 10, batch 200, loss[loss=0.1806, simple_loss=0.2737, pruned_loss=0.0438, over 12187.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2666, pruned_loss=0.044, over 1502892.35 frames. ], batch size: 35, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:49:49,931 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210310.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:50:14,396 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210344.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:50:21,954 INFO [finetune.py:992] (1/2) Epoch 10, batch 250, loss[loss=0.1498, simple_loss=0.2341, pruned_loss=0.0328, over 11767.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2659, pruned_loss=0.0441, over 1680231.33 frames. ], batch size: 26, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:50:23,374 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 2.825e+02 3.486e+02 4.026e+02 8.795e+02, threshold=6.972e+02, percent-clipped=3.0 2023-05-16 13:50:44,191 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8648, 4.5563, 4.5809, 4.6531, 4.5488, 4.8144, 4.7479, 2.6394], device='cuda:1'), covar=tensor([0.0117, 0.0079, 0.0111, 0.0085, 0.0071, 0.0100, 0.0086, 0.0779], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0074, 0.0077, 0.0070, 0.0057, 0.0087, 0.0075, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 13:50:49,016 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210392.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:50:58,653 INFO [finetune.py:992] (1/2) Epoch 10, batch 300, loss[loss=0.1803, simple_loss=0.2727, pruned_loss=0.04395, over 12087.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2643, pruned_loss=0.04324, over 1837579.85 frames. ], batch size: 42, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:51:04,409 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210413.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:51:34,138 INFO [finetune.py:992] (1/2) Epoch 10, batch 350, loss[loss=0.2145, simple_loss=0.2981, pruned_loss=0.06545, over 8561.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2638, pruned_loss=0.04324, over 1954185.36 frames. ], batch size: 97, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:51:35,574 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 2.842e+02 3.213e+02 3.857e+02 1.790e+03, threshold=6.425e+02, percent-clipped=1.0 2023-05-16 13:52:10,256 INFO [finetune.py:992] (1/2) Epoch 10, batch 400, loss[loss=0.1949, simple_loss=0.2801, pruned_loss=0.05482, over 12361.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2631, pruned_loss=0.04289, over 2053578.91 frames. ], batch size: 38, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:52:45,816 INFO [finetune.py:992] (1/2) Epoch 10, batch 450, loss[loss=0.1873, simple_loss=0.2795, pruned_loss=0.04757, over 12284.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2637, pruned_loss=0.04299, over 2124505.33 frames. ], batch size: 37, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:52:47,259 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.061e+02 2.781e+02 3.270e+02 3.985e+02 2.182e+03, threshold=6.539e+02, percent-clipped=5.0 2023-05-16 13:53:21,565 INFO [finetune.py:992] (1/2) Epoch 10, batch 500, loss[loss=0.1933, simple_loss=0.2864, pruned_loss=0.05007, over 11122.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2627, pruned_loss=0.04242, over 2177607.72 frames. ], batch size: 55, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:53:41,646 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210632.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:53:52,884 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4028, 2.7390, 3.2685, 4.3862, 2.3501, 4.3263, 4.3892, 4.4928], device='cuda:1'), covar=tensor([0.0104, 0.1050, 0.0458, 0.0123, 0.1344, 0.0231, 0.0163, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0198, 0.0180, 0.0112, 0.0186, 0.0171, 0.0166, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 13:53:57,447 INFO [finetune.py:992] (1/2) Epoch 10, batch 550, loss[loss=0.1696, simple_loss=0.2692, pruned_loss=0.03497, over 12349.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2623, pruned_loss=0.0423, over 2222029.35 frames. ], batch size: 35, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:53:58,860 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.661e+02 2.995e+02 3.654e+02 5.423e+02, threshold=5.990e+02, percent-clipped=0.0 2023-05-16 13:54:05,431 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210666.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:54:17,233 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210682.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:54:25,104 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210693.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:54:33,345 INFO [finetune.py:992] (1/2) Epoch 10, batch 600, loss[loss=0.1794, simple_loss=0.2643, pruned_loss=0.04725, over 12003.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2628, pruned_loss=0.04237, over 2253891.60 frames. ], batch size: 40, lr: 4.09e-03, grad_scale: 16.0 2023-05-16 13:54:39,230 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210713.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:54:41,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.10 vs. limit=5.0 2023-05-16 13:54:49,252 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210727.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:54:58,785 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 13:55:00,590 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210743.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:55:09,063 INFO [finetune.py:992] (1/2) Epoch 10, batch 650, loss[loss=0.154, simple_loss=0.2406, pruned_loss=0.03372, over 12350.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2621, pruned_loss=0.04186, over 2275159.13 frames. ], batch size: 30, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:55:10,410 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.114e+02 2.764e+02 3.125e+02 3.796e+02 1.069e+03, threshold=6.251e+02, percent-clipped=3.0 2023-05-16 13:55:13,970 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210761.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:55:16,199 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210764.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:55:35,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.76 vs. limit=5.0 2023-05-16 13:55:45,238 INFO [finetune.py:992] (1/2) Epoch 10, batch 700, loss[loss=0.1652, simple_loss=0.2681, pruned_loss=0.03115, over 12199.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2618, pruned_loss=0.04163, over 2297232.89 frames. ], batch size: 35, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:55:57,502 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3588, 4.9491, 5.3453, 4.6453, 4.9951, 4.6948, 5.3144, 5.0253], device='cuda:1'), covar=tensor([0.0306, 0.0376, 0.0322, 0.0278, 0.0387, 0.0335, 0.0274, 0.0277], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0244, 0.0266, 0.0240, 0.0242, 0.0242, 0.0218, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:56:00,232 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210825.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:56:21,294 INFO [finetune.py:992] (1/2) Epoch 10, batch 750, loss[loss=0.1926, simple_loss=0.2806, pruned_loss=0.05237, over 12133.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2623, pruned_loss=0.04181, over 2321046.33 frames. ], batch size: 39, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:56:22,750 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 2.676e+02 3.272e+02 3.735e+02 7.216e+02, threshold=6.544e+02, percent-clipped=2.0 2023-05-16 13:56:57,492 INFO [finetune.py:992] (1/2) Epoch 10, batch 800, loss[loss=0.1734, simple_loss=0.2714, pruned_loss=0.03768, over 12358.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2614, pruned_loss=0.04171, over 2332806.35 frames. ], batch size: 35, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:57:27,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-05-16 13:57:32,611 INFO [finetune.py:992] (1/2) Epoch 10, batch 850, loss[loss=0.1499, simple_loss=0.2432, pruned_loss=0.02834, over 11982.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2615, pruned_loss=0.04192, over 2332864.40 frames. ], batch size: 28, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:57:34,559 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 2.920e+02 3.334e+02 3.958e+02 9.193e+02, threshold=6.668e+02, percent-clipped=2.0 2023-05-16 13:57:53,709 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210984.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:57:56,480 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210988.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:58:08,595 INFO [finetune.py:992] (1/2) Epoch 10, batch 900, loss[loss=0.1682, simple_loss=0.2601, pruned_loss=0.03818, over 12352.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2609, pruned_loss=0.04142, over 2339731.99 frames. ], batch size: 36, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:58:20,718 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211022.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:58:32,111 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211038.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:58:37,899 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211045.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:58:44,832 INFO [finetune.py:992] (1/2) Epoch 10, batch 950, loss[loss=0.1653, simple_loss=0.2596, pruned_loss=0.03549, over 12292.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2597, pruned_loss=0.04092, over 2347143.00 frames. ], batch size: 34, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:58:46,243 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.872e+02 3.267e+02 3.845e+02 7.278e+02, threshold=6.535e+02, percent-clipped=1.0 2023-05-16 13:58:48,544 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4712, 5.1534, 5.4716, 4.8279, 5.1067, 4.7115, 5.4300, 5.1820], device='cuda:1'), covar=tensor([0.0360, 0.0378, 0.0411, 0.0267, 0.0433, 0.0406, 0.0320, 0.0213], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0248, 0.0269, 0.0244, 0.0246, 0.0245, 0.0220, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 13:59:20,420 INFO [finetune.py:992] (1/2) Epoch 10, batch 1000, loss[loss=0.1639, simple_loss=0.2482, pruned_loss=0.03977, over 12111.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.259, pruned_loss=0.04062, over 2355041.57 frames. ], batch size: 33, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:59:27,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 13:59:29,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 13:59:31,776 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211120.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 13:59:56,499 INFO [finetune.py:992] (1/2) Epoch 10, batch 1050, loss[loss=0.1709, simple_loss=0.2675, pruned_loss=0.0371, over 12201.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.04057, over 2364703.03 frames. ], batch size: 35, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 13:59:57,916 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.801e+02 3.196e+02 3.803e+02 6.712e+02, threshold=6.392e+02, percent-clipped=1.0 2023-05-16 14:00:32,879 INFO [finetune.py:992] (1/2) Epoch 10, batch 1100, loss[loss=0.1784, simple_loss=0.273, pruned_loss=0.04195, over 12099.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2581, pruned_loss=0.04049, over 2365482.76 frames. ], batch size: 39, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 14:00:40,847 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3751, 2.8651, 3.9462, 3.3924, 3.7479, 3.4162, 2.7207, 3.7831], device='cuda:1'), covar=tensor([0.0122, 0.0327, 0.0131, 0.0201, 0.0130, 0.0203, 0.0389, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0200, 0.0180, 0.0178, 0.0207, 0.0155, 0.0192, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:01:08,070 INFO [finetune.py:992] (1/2) Epoch 10, batch 1150, loss[loss=0.1765, simple_loss=0.2701, pruned_loss=0.04144, over 12088.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2579, pruned_loss=0.04054, over 2375982.00 frames. ], batch size: 33, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 14:01:09,495 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.597e+02 3.087e+02 3.883e+02 5.758e+02, threshold=6.174e+02, percent-clipped=0.0 2023-05-16 14:01:32,443 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211288.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:01:43,869 INFO [finetune.py:992] (1/2) Epoch 10, batch 1200, loss[loss=0.2103, simple_loss=0.2994, pruned_loss=0.06056, over 11584.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2591, pruned_loss=0.04109, over 2374980.63 frames. ], batch size: 48, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 14:01:44,994 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 14:01:56,143 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211322.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:06,146 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211336.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:07,653 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211338.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:08,956 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211340.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:11,241 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2542, 5.2504, 5.0627, 4.5648, 4.6567, 5.1663, 4.8370, 4.6275], device='cuda:1'), covar=tensor([0.0713, 0.0759, 0.0625, 0.1423, 0.1195, 0.0763, 0.1511, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0604, 0.0541, 0.0502, 0.0615, 0.0410, 0.0692, 0.0763, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:02:20,142 INFO [finetune.py:992] (1/2) Epoch 10, batch 1250, loss[loss=0.1754, simple_loss=0.2678, pruned_loss=0.04151, over 11211.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.259, pruned_loss=0.04097, over 2376071.68 frames. ], batch size: 55, lr: 4.08e-03, grad_scale: 16.0 2023-05-16 14:02:20,350 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3560, 4.8536, 2.9782, 2.9572, 4.1467, 2.7672, 4.1304, 3.5358], device='cuda:1'), covar=tensor([0.0649, 0.0383, 0.1193, 0.1358, 0.0293, 0.1187, 0.0418, 0.0667], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0251, 0.0177, 0.0199, 0.0139, 0.0182, 0.0195, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:02:21,501 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.891e+02 3.376e+02 3.934e+02 8.670e+02, threshold=6.752e+02, percent-clipped=2.0 2023-05-16 14:02:30,900 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211370.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:42,017 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211386.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:42,124 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211386.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:02:55,708 INFO [finetune.py:992] (1/2) Epoch 10, batch 1300, loss[loss=0.1611, simple_loss=0.251, pruned_loss=0.03563, over 12119.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2577, pruned_loss=0.04073, over 2378598.04 frames. ], batch size: 39, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:03:00,798 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211412.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:03:07,031 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:03:08,795 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 14:03:26,571 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211447.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:03:32,000 INFO [finetune.py:992] (1/2) Epoch 10, batch 1350, loss[loss=0.1702, simple_loss=0.2676, pruned_loss=0.0364, over 11592.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2571, pruned_loss=0.04031, over 2380634.88 frames. ], batch size: 48, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:03:34,133 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 2.733e+02 3.111e+02 3.708e+02 5.238e+02, threshold=6.222e+02, percent-clipped=0.0 2023-05-16 14:03:41,375 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211468.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:03:45,033 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211473.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:04:02,770 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4403, 5.1927, 5.3227, 5.3415, 5.0072, 5.0197, 4.7887, 5.2811], device='cuda:1'), covar=tensor([0.0626, 0.0659, 0.0786, 0.0644, 0.1864, 0.1401, 0.0548, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0678, 0.0584, 0.0595, 0.0816, 0.0719, 0.0531, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:04:07,669 INFO [finetune.py:992] (1/2) Epoch 10, batch 1400, loss[loss=0.1613, simple_loss=0.2403, pruned_loss=0.04117, over 12181.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2574, pruned_loss=0.04031, over 2382410.06 frames. ], batch size: 29, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:04:39,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 14:04:43,929 INFO [finetune.py:992] (1/2) Epoch 10, batch 1450, loss[loss=0.1698, simple_loss=0.2541, pruned_loss=0.04276, over 12264.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2564, pruned_loss=0.03959, over 2385293.98 frames. ], batch size: 32, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:04:46,038 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.756e+02 3.139e+02 3.742e+02 1.134e+03, threshold=6.278e+02, percent-clipped=4.0 2023-05-16 14:04:52,446 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0649, 6.0424, 5.7499, 5.2156, 5.1031, 5.9131, 5.5784, 5.3362], device='cuda:1'), covar=tensor([0.0776, 0.0881, 0.0679, 0.1645, 0.0758, 0.0733, 0.1411, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0544, 0.0504, 0.0619, 0.0410, 0.0697, 0.0766, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:05:20,301 INFO [finetune.py:992] (1/2) Epoch 10, batch 1500, loss[loss=0.174, simple_loss=0.2622, pruned_loss=0.04289, over 12356.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.257, pruned_loss=0.03983, over 2391533.92 frames. ], batch size: 38, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:05:45,294 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211640.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:05:47,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3883, 4.4134, 4.1132, 4.5477, 2.9418, 4.2259, 2.8983, 4.3705], device='cuda:1'), covar=tensor([0.1603, 0.0610, 0.1011, 0.0702, 0.1188, 0.0550, 0.1631, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0263, 0.0292, 0.0349, 0.0234, 0.0237, 0.0257, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:05:53,154 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0982, 6.0663, 5.8683, 5.3806, 5.2319, 6.0186, 5.6046, 5.3837], device='cuda:1'), covar=tensor([0.0781, 0.1017, 0.0610, 0.1502, 0.0615, 0.0734, 0.1524, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0544, 0.0503, 0.0618, 0.0410, 0.0698, 0.0765, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:05:55,970 INFO [finetune.py:992] (1/2) Epoch 10, batch 1550, loss[loss=0.1853, simple_loss=0.2627, pruned_loss=0.05398, over 12114.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2575, pruned_loss=0.04007, over 2394845.57 frames. ], batch size: 38, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:05:58,005 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.636e+02 3.063e+02 3.605e+02 8.321e+02, threshold=6.125e+02, percent-clipped=1.0 2023-05-16 14:06:13,990 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8654, 3.2327, 5.0959, 2.7129, 2.8586, 3.9994, 3.3582, 3.7682], device='cuda:1'), covar=tensor([0.0373, 0.1132, 0.0348, 0.1045, 0.1803, 0.1324, 0.1232, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0233, 0.0241, 0.0180, 0.0237, 0.0289, 0.0221, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:06:20,245 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:06:31,969 INFO [finetune.py:992] (1/2) Epoch 10, batch 1600, loss[loss=0.1702, simple_loss=0.2568, pruned_loss=0.04179, over 12097.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2583, pruned_loss=0.04041, over 2383927.07 frames. ], batch size: 32, lr: 4.08e-03, grad_scale: 8.0 2023-05-16 14:06:44,905 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1485, 5.0756, 5.0046, 5.0140, 4.6478, 5.1361, 5.2359, 5.3297], device='cuda:1'), covar=tensor([0.0224, 0.0138, 0.0157, 0.0283, 0.0715, 0.0285, 0.0134, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0191, 0.0186, 0.0238, 0.0236, 0.0211, 0.0171, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:06:58,864 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211742.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:07:08,009 INFO [finetune.py:992] (1/2) Epoch 10, batch 1650, loss[loss=0.1668, simple_loss=0.2604, pruned_loss=0.03661, over 11808.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2582, pruned_loss=0.04033, over 2380658.16 frames. ], batch size: 44, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:07:10,071 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.665e+02 3.073e+02 3.607e+02 6.323e+02, threshold=6.146e+02, percent-clipped=2.0 2023-05-16 14:07:17,123 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211768.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:07:38,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 14:07:43,714 INFO [finetune.py:992] (1/2) Epoch 10, batch 1700, loss[loss=0.1484, simple_loss=0.2428, pruned_loss=0.02701, over 12310.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2596, pruned_loss=0.04051, over 2383833.14 frames. ], batch size: 34, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:07:46,119 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3871, 4.7709, 4.1524, 5.0732, 4.6147, 2.9723, 4.2836, 3.0507], device='cuda:1'), covar=tensor([0.0780, 0.0754, 0.1388, 0.0436, 0.1081, 0.1589, 0.1208, 0.3272], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0373, 0.0348, 0.0289, 0.0362, 0.0263, 0.0337, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:08:18,770 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2203, 5.1426, 5.0468, 5.1141, 4.6951, 5.2004, 5.2073, 5.3755], device='cuda:1'), covar=tensor([0.0158, 0.0130, 0.0151, 0.0244, 0.0688, 0.0247, 0.0131, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0190, 0.0186, 0.0237, 0.0235, 0.0210, 0.0170, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:08:20,026 INFO [finetune.py:992] (1/2) Epoch 10, batch 1750, loss[loss=0.1824, simple_loss=0.2759, pruned_loss=0.04443, over 12035.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2596, pruned_loss=0.04029, over 2381122.84 frames. ], batch size: 40, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:08:22,203 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.699e+02 3.136e+02 3.576e+02 9.797e+02, threshold=6.272e+02, percent-clipped=1.0 2023-05-16 14:08:25,893 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9085, 5.9266, 5.7279, 5.1637, 4.9930, 5.8220, 5.3873, 5.2982], device='cuda:1'), covar=tensor([0.0796, 0.0831, 0.0552, 0.1421, 0.0799, 0.0703, 0.1638, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0546, 0.0503, 0.0621, 0.0411, 0.0700, 0.0770, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:08:56,264 INFO [finetune.py:992] (1/2) Epoch 10, batch 1800, loss[loss=0.1561, simple_loss=0.2541, pruned_loss=0.02907, over 12117.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2593, pruned_loss=0.04047, over 2377703.59 frames. ], batch size: 33, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:09:32,014 INFO [finetune.py:992] (1/2) Epoch 10, batch 1850, loss[loss=0.1676, simple_loss=0.2566, pruned_loss=0.03933, over 12015.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2599, pruned_loss=0.04096, over 2376717.33 frames. ], batch size: 28, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:09:34,864 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.726e+02 3.285e+02 4.068e+02 5.793e+02, threshold=6.571e+02, percent-clipped=0.0 2023-05-16 14:09:36,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 14:10:11,302 INFO [finetune.py:992] (1/2) Epoch 10, batch 1900, loss[loss=0.1632, simple_loss=0.2433, pruned_loss=0.04157, over 12329.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2589, pruned_loss=0.04081, over 2378850.55 frames. ], batch size: 30, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:10:38,003 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212042.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:10:47,252 INFO [finetune.py:992] (1/2) Epoch 10, batch 1950, loss[loss=0.1946, simple_loss=0.283, pruned_loss=0.05313, over 12027.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2593, pruned_loss=0.04094, over 2376788.22 frames. ], batch size: 40, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:10:49,204 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 2.785e+02 3.345e+02 3.996e+02 7.616e+02, threshold=6.690e+02, percent-clipped=2.0 2023-05-16 14:10:56,440 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212068.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:11:12,134 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212090.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:11:22,826 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212104.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 14:11:23,377 INFO [finetune.py:992] (1/2) Epoch 10, batch 2000, loss[loss=0.1721, simple_loss=0.2633, pruned_loss=0.04046, over 12273.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2599, pruned_loss=0.04135, over 2370023.36 frames. ], batch size: 37, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:11:31,259 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212116.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:11:59,545 INFO [finetune.py:992] (1/2) Epoch 10, batch 2050, loss[loss=0.1663, simple_loss=0.2606, pruned_loss=0.03598, over 12105.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2592, pruned_loss=0.0409, over 2371907.07 frames. ], batch size: 42, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:12:02,301 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.686e+02 3.253e+02 3.860e+02 1.028e+03, threshold=6.506e+02, percent-clipped=1.0 2023-05-16 14:12:07,623 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212165.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 14:12:36,293 INFO [finetune.py:992] (1/2) Epoch 10, batch 2100, loss[loss=0.1516, simple_loss=0.2388, pruned_loss=0.0322, over 12325.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2591, pruned_loss=0.04061, over 2370796.38 frames. ], batch size: 31, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:13:05,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0136, 3.5116, 5.3070, 2.6847, 2.8794, 3.8511, 3.2941, 3.7503], device='cuda:1'), covar=tensor([0.0378, 0.1051, 0.0237, 0.1138, 0.1902, 0.1453, 0.1293, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0234, 0.0243, 0.0181, 0.0238, 0.0291, 0.0224, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:13:11,178 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212253.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:13:12,388 INFO [finetune.py:992] (1/2) Epoch 10, batch 2150, loss[loss=0.1659, simple_loss=0.2487, pruned_loss=0.04155, over 12285.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2585, pruned_loss=0.04069, over 2360869.55 frames. ], batch size: 33, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:13:14,583 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.827e+02 3.250e+02 3.793e+02 5.770e+02, threshold=6.500e+02, percent-clipped=0.0 2023-05-16 14:13:42,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 14:13:48,626 INFO [finetune.py:992] (1/2) Epoch 10, batch 2200, loss[loss=0.2, simple_loss=0.2865, pruned_loss=0.0568, over 8178.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2585, pruned_loss=0.04044, over 2365053.69 frames. ], batch size: 99, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:13:52,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2224, 2.7234, 3.7560, 3.2447, 3.6363, 3.3614, 2.6935, 3.6577], device='cuda:1'), covar=tensor([0.0133, 0.0311, 0.0123, 0.0243, 0.0142, 0.0178, 0.0319, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0203, 0.0185, 0.0182, 0.0212, 0.0158, 0.0195, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:13:55,255 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212314.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:14:21,479 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3511, 4.8075, 2.8586, 2.6540, 4.1762, 2.7452, 4.0617, 3.2633], device='cuda:1'), covar=tensor([0.0735, 0.0541, 0.1231, 0.1557, 0.0255, 0.1299, 0.0452, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0255, 0.0178, 0.0200, 0.0141, 0.0183, 0.0196, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:14:24,148 INFO [finetune.py:992] (1/2) Epoch 10, batch 2250, loss[loss=0.172, simple_loss=0.2657, pruned_loss=0.03914, over 12147.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2581, pruned_loss=0.04046, over 2367756.84 frames. ], batch size: 36, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:14:26,364 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.151e+02 2.714e+02 3.113e+02 3.776e+02 8.100e+02, threshold=6.226e+02, percent-clipped=2.0 2023-05-16 14:15:00,478 INFO [finetune.py:992] (1/2) Epoch 10, batch 2300, loss[loss=0.1675, simple_loss=0.2518, pruned_loss=0.04165, over 12177.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2588, pruned_loss=0.04062, over 2368856.92 frames. ], batch size: 31, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:15:15,499 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9426, 4.8881, 4.7640, 4.8212, 4.5183, 4.9639, 4.9104, 5.1886], device='cuda:1'), covar=tensor([0.0278, 0.0143, 0.0212, 0.0339, 0.0792, 0.0326, 0.0198, 0.0177], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0191, 0.0187, 0.0239, 0.0237, 0.0212, 0.0171, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:15:24,619 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212439.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:15:36,445 INFO [finetune.py:992] (1/2) Epoch 10, batch 2350, loss[loss=0.2023, simple_loss=0.2974, pruned_loss=0.05355, over 12301.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2587, pruned_loss=0.04043, over 2375165.75 frames. ], batch size: 34, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:15:38,545 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.030e+02 2.649e+02 3.231e+02 3.820e+02 8.982e+02, threshold=6.461e+02, percent-clipped=3.0 2023-05-16 14:15:40,086 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212460.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 14:15:53,425 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9964, 6.0116, 5.7489, 5.2152, 5.2023, 5.9170, 5.4842, 5.3188], device='cuda:1'), covar=tensor([0.0767, 0.0839, 0.0641, 0.1599, 0.0673, 0.0715, 0.1604, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0602, 0.0537, 0.0500, 0.0618, 0.0408, 0.0694, 0.0759, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:16:08,704 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212500.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:16:11,981 INFO [finetune.py:992] (1/2) Epoch 10, batch 2400, loss[loss=0.1977, simple_loss=0.2783, pruned_loss=0.05858, over 12366.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2587, pruned_loss=0.04059, over 2377907.69 frames. ], batch size: 38, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:16:48,160 INFO [finetune.py:992] (1/2) Epoch 10, batch 2450, loss[loss=0.1385, simple_loss=0.2256, pruned_loss=0.02571, over 12171.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2582, pruned_loss=0.04007, over 2385106.60 frames. ], batch size: 29, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:16:50,194 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.756e+02 3.071e+02 3.589e+02 6.075e+02, threshold=6.142e+02, percent-clipped=0.0 2023-05-16 14:17:22,442 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2300, 3.8165, 3.7688, 4.2062, 2.8590, 3.7759, 2.5135, 3.7959], device='cuda:1'), covar=tensor([0.1611, 0.0792, 0.0964, 0.0681, 0.1148, 0.0690, 0.1893, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0266, 0.0294, 0.0356, 0.0235, 0.0240, 0.0259, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:17:23,170 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8684, 3.7809, 3.2492, 3.3374, 3.0898, 2.9565, 3.6995, 2.3445], device='cuda:1'), covar=tensor([0.0309, 0.0133, 0.0200, 0.0185, 0.0333, 0.0330, 0.0117, 0.0454], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0159, 0.0157, 0.0183, 0.0199, 0.0195, 0.0165, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:17:24,323 INFO [finetune.py:992] (1/2) Epoch 10, batch 2500, loss[loss=0.1731, simple_loss=0.2752, pruned_loss=0.0355, over 12343.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2595, pruned_loss=0.04063, over 2381607.32 frames. ], batch size: 35, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:17:27,203 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212609.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 14:17:59,949 INFO [finetune.py:992] (1/2) Epoch 10, batch 2550, loss[loss=0.1428, simple_loss=0.2242, pruned_loss=0.03071, over 12349.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2588, pruned_loss=0.04043, over 2384160.39 frames. ], batch size: 30, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:18:02,082 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.656e+02 3.136e+02 3.615e+02 8.146e+02, threshold=6.272e+02, percent-clipped=2.0 2023-05-16 14:18:24,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-05-16 14:18:31,130 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212698.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:18:35,894 INFO [finetune.py:992] (1/2) Epoch 10, batch 2600, loss[loss=0.1574, simple_loss=0.2503, pruned_loss=0.03223, over 12301.00 frames. ], tot_loss[loss=0.17, simple_loss=0.259, pruned_loss=0.04053, over 2386971.34 frames. ], batch size: 33, lr: 4.07e-03, grad_scale: 8.0 2023-05-16 14:18:40,231 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212711.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:19:11,460 INFO [finetune.py:992] (1/2) Epoch 10, batch 2650, loss[loss=0.142, simple_loss=0.2258, pruned_loss=0.02907, over 12348.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2583, pruned_loss=0.04045, over 2374232.08 frames. ], batch size: 30, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:19:14,297 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.632e+02 3.258e+02 3.720e+02 9.493e+02, threshold=6.515e+02, percent-clipped=1.0 2023-05-16 14:19:15,156 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212759.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:19:15,756 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212760.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 14:19:24,499 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212772.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:19:25,890 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212774.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:19:40,459 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212795.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:19:47,875 INFO [finetune.py:992] (1/2) Epoch 10, batch 2700, loss[loss=0.1541, simple_loss=0.2337, pruned_loss=0.03721, over 12014.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2574, pruned_loss=0.04014, over 2381185.28 frames. ], batch size: 28, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:19:49,998 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212808.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:20:08,671 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2985, 4.6778, 3.9948, 4.9174, 4.5637, 2.8458, 4.2733, 3.0118], device='cuda:1'), covar=tensor([0.0844, 0.0693, 0.1479, 0.0613, 0.1055, 0.1616, 0.0941, 0.3127], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0376, 0.0352, 0.0292, 0.0363, 0.0265, 0.0339, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:20:09,294 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212835.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:20:14,406 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1339, 2.5878, 3.7239, 3.1258, 3.5546, 3.2859, 2.6196, 3.6396], device='cuda:1'), covar=tensor([0.0138, 0.0341, 0.0117, 0.0235, 0.0128, 0.0180, 0.0328, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0203, 0.0185, 0.0183, 0.0213, 0.0158, 0.0194, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:20:24,043 INFO [finetune.py:992] (1/2) Epoch 10, batch 2750, loss[loss=0.15, simple_loss=0.2353, pruned_loss=0.03235, over 12115.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.258, pruned_loss=0.04011, over 2387137.43 frames. ], batch size: 33, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:20:26,117 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.715e+02 3.169e+02 3.755e+02 8.631e+02, threshold=6.338e+02, percent-clipped=1.0 2023-05-16 14:20:41,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5217, 3.5162, 3.1384, 3.1583, 2.8372, 2.6895, 3.5159, 2.1233], device='cuda:1'), covar=tensor([0.0401, 0.0155, 0.0211, 0.0191, 0.0370, 0.0348, 0.0133, 0.0535], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0162, 0.0158, 0.0184, 0.0202, 0.0197, 0.0166, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:20:59,764 INFO [finetune.py:992] (1/2) Epoch 10, batch 2800, loss[loss=0.1425, simple_loss=0.2225, pruned_loss=0.03122, over 11793.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2589, pruned_loss=0.04082, over 2376817.15 frames. ], batch size: 26, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:21:02,706 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212909.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:21:21,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 14:21:35,069 INFO [finetune.py:992] (1/2) Epoch 10, batch 2850, loss[loss=0.1628, simple_loss=0.2509, pruned_loss=0.0373, over 12334.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2587, pruned_loss=0.0409, over 2374182.31 frames. ], batch size: 31, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:21:36,581 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212957.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:21:37,845 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.812e+02 3.269e+02 3.790e+02 1.926e+03, threshold=6.539e+02, percent-clipped=3.0 2023-05-16 14:21:53,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-05-16 14:22:11,599 INFO [finetune.py:992] (1/2) Epoch 10, batch 2900, loss[loss=0.1514, simple_loss=0.2471, pruned_loss=0.02788, over 12147.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2592, pruned_loss=0.04128, over 2373782.94 frames. ], batch size: 34, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:22:41,153 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-16 14:22:46,871 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213054.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:22:47,482 INFO [finetune.py:992] (1/2) Epoch 10, batch 2950, loss[loss=0.1848, simple_loss=0.2835, pruned_loss=0.04305, over 12149.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2591, pruned_loss=0.04108, over 2375702.15 frames. ], batch size: 36, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:22:50,249 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.807e+02 3.371e+02 4.240e+02 8.214e+02, threshold=6.743e+02, percent-clipped=3.0 2023-05-16 14:22:55,997 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213067.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:22:58,118 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0474, 4.9528, 4.8496, 5.0340, 3.8648, 5.1669, 5.1458, 5.1512], device='cuda:1'), covar=tensor([0.0220, 0.0180, 0.0213, 0.0294, 0.1230, 0.0247, 0.0140, 0.0201], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0189, 0.0184, 0.0236, 0.0235, 0.0209, 0.0168, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:23:15,483 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213095.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:23:22,359 INFO [finetune.py:992] (1/2) Epoch 10, batch 3000, loss[loss=0.1706, simple_loss=0.2635, pruned_loss=0.03884, over 12086.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2588, pruned_loss=0.04078, over 2377657.44 frames. ], batch size: 32, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:23:22,359 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 14:23:28,393 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8734, 2.3479, 3.6093, 3.9570, 3.7170, 3.8793, 3.8166, 2.6362], device='cuda:1'), covar=tensor([0.0066, 0.0422, 0.0140, 0.0052, 0.0117, 0.0105, 0.0113, 0.0500], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0123, 0.0104, 0.0075, 0.0102, 0.0114, 0.0094, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:23:29,194 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3536, 4.2141, 4.0261, 4.1976, 3.9789, 4.2530, 4.2794, 2.2644], device='cuda:1'), covar=tensor([0.0152, 0.0082, 0.0217, 0.0120, 0.0095, 0.0132, 0.0163, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0081, 0.0072, 0.0060, 0.0090, 0.0079, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:23:36,964 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5998, 2.7383, 3.9342, 4.2192, 3.0304, 2.6364, 2.6184, 2.0965], device='cuda:1'), covar=tensor([0.1460, 0.2636, 0.0607, 0.0396, 0.1056, 0.2043, 0.2996, 0.4548], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0382, 0.0271, 0.0294, 0.0265, 0.0299, 0.0372, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:23:38,031 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3165, 3.3092, 4.9222, 2.2916, 2.5888, 3.6591, 3.1855, 3.7282], device='cuda:1'), covar=tensor([0.0662, 0.1307, 0.0218, 0.1438, 0.2200, 0.1582, 0.1478, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0236, 0.0247, 0.0182, 0.0239, 0.0292, 0.0225, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:23:40,833 INFO [finetune.py:1026] (1/2) Epoch 10, validation: loss=0.3191, simple_loss=0.3958, pruned_loss=0.1211, over 1020973.00 frames. 2023-05-16 14:23:40,834 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 14:23:58,667 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213130.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:24:03,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 14:24:07,773 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213143.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:24:15,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 14:24:16,865 INFO [finetune.py:992] (1/2) Epoch 10, batch 3050, loss[loss=0.1988, simple_loss=0.2801, pruned_loss=0.05876, over 7696.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2588, pruned_loss=0.04073, over 2377772.63 frames. ], batch size: 98, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:24:19,695 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.181e+02 2.841e+02 3.499e+02 4.340e+02 7.854e+02, threshold=6.998e+02, percent-clipped=1.0 2023-05-16 14:24:52,534 INFO [finetune.py:992] (1/2) Epoch 10, batch 3100, loss[loss=0.1776, simple_loss=0.2692, pruned_loss=0.04296, over 12123.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2574, pruned_loss=0.04022, over 2378090.60 frames. ], batch size: 38, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:25:06,422 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3701, 4.7305, 2.8839, 2.6436, 4.0738, 2.6763, 3.9643, 3.1075], device='cuda:1'), covar=tensor([0.0662, 0.0449, 0.1122, 0.1528, 0.0253, 0.1277, 0.0515, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0252, 0.0176, 0.0198, 0.0141, 0.0180, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:25:28,654 INFO [finetune.py:992] (1/2) Epoch 10, batch 3150, loss[loss=0.1512, simple_loss=0.2377, pruned_loss=0.03234, over 12259.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2565, pruned_loss=0.0398, over 2383410.49 frames. ], batch size: 32, lr: 4.06e-03, grad_scale: 4.0 2023-05-16 14:25:31,434 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 2.677e+02 3.305e+02 3.988e+02 1.363e+03, threshold=6.611e+02, percent-clipped=2.0 2023-05-16 14:25:40,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2023-05-16 14:26:04,525 INFO [finetune.py:992] (1/2) Epoch 10, batch 3200, loss[loss=0.2318, simple_loss=0.3033, pruned_loss=0.08017, over 7981.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2573, pruned_loss=0.04016, over 2366180.36 frames. ], batch size: 98, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:26:09,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 14:26:39,369 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213354.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:26:39,877 INFO [finetune.py:992] (1/2) Epoch 10, batch 3250, loss[loss=0.1809, simple_loss=0.2751, pruned_loss=0.0433, over 12116.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2574, pruned_loss=0.04025, over 2374188.62 frames. ], batch size: 38, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:26:42,680 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.116e+02 2.529e+02 3.027e+02 3.616e+02 6.084e+02, threshold=6.053e+02, percent-clipped=0.0 2023-05-16 14:26:48,387 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213367.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:26:59,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-05-16 14:27:14,013 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213402.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:27:16,037 INFO [finetune.py:992] (1/2) Epoch 10, batch 3300, loss[loss=0.1646, simple_loss=0.2405, pruned_loss=0.0443, over 12345.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2583, pruned_loss=0.04041, over 2371348.17 frames. ], batch size: 31, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:27:23,273 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213415.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:27:33,895 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213430.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:27:37,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 14:27:52,223 INFO [finetune.py:992] (1/2) Epoch 10, batch 3350, loss[loss=0.1701, simple_loss=0.2659, pruned_loss=0.03714, over 11990.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2574, pruned_loss=0.03993, over 2376082.50 frames. ], batch size: 40, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:27:55,113 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.906e+02 2.748e+02 3.058e+02 3.642e+02 5.641e+02, threshold=6.115e+02, percent-clipped=0.0 2023-05-16 14:28:05,279 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1057, 2.5150, 3.5604, 3.0704, 3.4267, 3.1773, 2.4882, 3.5281], device='cuda:1'), covar=tensor([0.0119, 0.0341, 0.0175, 0.0222, 0.0141, 0.0192, 0.0348, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0201, 0.0184, 0.0181, 0.0211, 0.0157, 0.0193, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:28:08,667 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213478.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:28:27,415 INFO [finetune.py:992] (1/2) Epoch 10, batch 3400, loss[loss=0.1783, simple_loss=0.2714, pruned_loss=0.04261, over 12124.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2577, pruned_loss=0.04012, over 2378522.71 frames. ], batch size: 39, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:28:38,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.16 vs. limit=5.0 2023-05-16 14:29:03,105 INFO [finetune.py:992] (1/2) Epoch 10, batch 3450, loss[loss=0.1507, simple_loss=0.2376, pruned_loss=0.03185, over 12339.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2586, pruned_loss=0.04025, over 2376317.61 frames. ], batch size: 31, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:29:05,911 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.712e+02 3.310e+02 3.896e+02 1.850e+03, threshold=6.620e+02, percent-clipped=6.0 2023-05-16 14:29:11,039 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213566.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:29:11,070 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2319, 3.9319, 3.8881, 4.3938, 2.7044, 3.9740, 2.4365, 4.1208], device='cuda:1'), covar=tensor([0.1610, 0.0784, 0.1097, 0.0634, 0.1246, 0.0605, 0.1930, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0265, 0.0292, 0.0354, 0.0233, 0.0239, 0.0257, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:29:39,368 INFO [finetune.py:992] (1/2) Epoch 10, batch 3500, loss[loss=0.1747, simple_loss=0.2681, pruned_loss=0.04069, over 12345.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2572, pruned_loss=0.03977, over 2382617.66 frames. ], batch size: 36, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:29:40,651 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 14:29:55,225 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213627.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:29:55,982 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3453, 3.7955, 3.8774, 4.3737, 2.6412, 3.8276, 2.6816, 4.0407], device='cuda:1'), covar=tensor([0.1684, 0.0964, 0.1197, 0.0811, 0.1475, 0.0778, 0.1865, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0266, 0.0292, 0.0354, 0.0234, 0.0240, 0.0258, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:30:14,832 INFO [finetune.py:992] (1/2) Epoch 10, batch 3550, loss[loss=0.1605, simple_loss=0.2432, pruned_loss=0.03894, over 12110.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2571, pruned_loss=0.03971, over 2379946.03 frames. ], batch size: 30, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:30:17,581 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.610e+02 3.178e+02 3.849e+02 9.500e+02, threshold=6.356e+02, percent-clipped=2.0 2023-05-16 14:30:17,845 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3653, 3.3325, 3.0621, 3.0644, 2.7658, 2.5229, 3.2633, 2.1354], device='cuda:1'), covar=tensor([0.0424, 0.0120, 0.0182, 0.0190, 0.0337, 0.0347, 0.0136, 0.0481], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0160, 0.0158, 0.0184, 0.0201, 0.0197, 0.0168, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:30:24,905 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0491, 6.0594, 5.8463, 5.3648, 5.1753, 5.9693, 5.5895, 5.3533], device='cuda:1'), covar=tensor([0.0723, 0.0854, 0.0699, 0.1670, 0.0748, 0.0729, 0.1481, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0548, 0.0512, 0.0628, 0.0419, 0.0714, 0.0772, 0.0565], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:30:51,046 INFO [finetune.py:992] (1/2) Epoch 10, batch 3600, loss[loss=0.1606, simple_loss=0.2551, pruned_loss=0.03305, over 12309.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2565, pruned_loss=0.03946, over 2382762.30 frames. ], batch size: 34, lr: 4.06e-03, grad_scale: 8.0 2023-05-16 14:30:53,312 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5009, 2.4966, 3.0813, 4.4780, 2.3556, 4.5035, 4.5137, 4.6531], device='cuda:1'), covar=tensor([0.0140, 0.1165, 0.0545, 0.0148, 0.1275, 0.0212, 0.0142, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0203, 0.0185, 0.0116, 0.0189, 0.0177, 0.0174, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:31:13,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4632, 5.0619, 5.4451, 4.8076, 5.1335, 4.8364, 5.4789, 5.0415], device='cuda:1'), covar=tensor([0.0293, 0.0344, 0.0246, 0.0237, 0.0325, 0.0319, 0.0198, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0259, 0.0283, 0.0255, 0.0256, 0.0254, 0.0230, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:31:17,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2023-05-16 14:31:26,950 INFO [finetune.py:992] (1/2) Epoch 10, batch 3650, loss[loss=0.177, simple_loss=0.2667, pruned_loss=0.04361, over 12130.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2573, pruned_loss=0.04013, over 2382795.00 frames. ], batch size: 30, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:31:27,458 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-16 14:31:29,777 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.658e+02 3.103e+02 3.735e+02 8.103e+02, threshold=6.207e+02, percent-clipped=1.0 2023-05-16 14:31:36,217 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8523, 3.4964, 5.2101, 2.8963, 2.7851, 3.9033, 3.1945, 3.7837], device='cuda:1'), covar=tensor([0.0388, 0.1045, 0.0281, 0.1044, 0.1907, 0.1512, 0.1292, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0235, 0.0247, 0.0183, 0.0240, 0.0293, 0.0224, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:32:02,506 INFO [finetune.py:992] (1/2) Epoch 10, batch 3700, loss[loss=0.1475, simple_loss=0.2379, pruned_loss=0.0285, over 12128.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2579, pruned_loss=0.04044, over 2379423.00 frames. ], batch size: 30, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:32:09,481 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8170, 3.3465, 5.1674, 2.8870, 2.7372, 3.7704, 3.1216, 3.7635], device='cuda:1'), covar=tensor([0.0384, 0.1159, 0.0267, 0.1151, 0.1968, 0.1435, 0.1351, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0234, 0.0247, 0.0183, 0.0240, 0.0293, 0.0225, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:32:38,534 INFO [finetune.py:992] (1/2) Epoch 10, batch 3750, loss[loss=0.1419, simple_loss=0.2208, pruned_loss=0.03154, over 12169.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2575, pruned_loss=0.04031, over 2380021.00 frames. ], batch size: 29, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:32:41,259 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.981e+02 2.726e+02 3.266e+02 3.847e+02 1.047e+03, threshold=6.533e+02, percent-clipped=2.0 2023-05-16 14:32:48,575 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1478, 4.9883, 5.0455, 5.1368, 4.8135, 4.8186, 4.6075, 5.0347], device='cuda:1'), covar=tensor([0.0619, 0.0548, 0.0752, 0.0526, 0.1661, 0.1234, 0.0525, 0.1056], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0696, 0.0605, 0.0617, 0.0836, 0.0741, 0.0547, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:33:14,898 INFO [finetune.py:992] (1/2) Epoch 10, batch 3800, loss[loss=0.1301, simple_loss=0.215, pruned_loss=0.02256, over 12018.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.257, pruned_loss=0.0398, over 2385531.00 frames. ], batch size: 28, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:33:22,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-05-16 14:33:27,149 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213922.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:33:50,779 INFO [finetune.py:992] (1/2) Epoch 10, batch 3850, loss[loss=0.1857, simple_loss=0.2768, pruned_loss=0.04729, over 12289.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2566, pruned_loss=0.03969, over 2383992.33 frames. ], batch size: 37, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:33:53,668 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.760e+02 3.187e+02 3.831e+02 8.275e+02, threshold=6.375e+02, percent-clipped=2.0 2023-05-16 14:34:18,120 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2773, 6.2186, 6.0272, 5.5027, 5.3804, 6.1555, 5.7557, 5.5655], device='cuda:1'), covar=tensor([0.0708, 0.1017, 0.0721, 0.1622, 0.0678, 0.0751, 0.1615, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0622, 0.0556, 0.0518, 0.0637, 0.0422, 0.0721, 0.0789, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:34:30,252 INFO [finetune.py:992] (1/2) Epoch 10, batch 3900, loss[loss=0.19, simple_loss=0.2777, pruned_loss=0.05114, over 12142.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2569, pruned_loss=0.03969, over 2381005.73 frames. ], batch size: 36, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:34:47,430 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5350, 5.2930, 5.4220, 5.4545, 5.1250, 5.1577, 4.9071, 5.3822], device='cuda:1'), covar=tensor([0.0595, 0.0621, 0.0644, 0.0578, 0.1656, 0.1189, 0.0552, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0691, 0.0598, 0.0613, 0.0826, 0.0732, 0.0541, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:35:06,399 INFO [finetune.py:992] (1/2) Epoch 10, batch 3950, loss[loss=0.1542, simple_loss=0.2439, pruned_loss=0.03228, over 12085.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2579, pruned_loss=0.03982, over 2388728.94 frames. ], batch size: 32, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:35:09,282 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.613e+02 3.033e+02 3.734e+02 5.708e+02, threshold=6.065e+02, percent-clipped=0.0 2023-05-16 14:35:16,793 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6013, 2.8137, 3.8085, 4.3758, 3.8952, 4.4703, 3.9890, 3.1662], device='cuda:1'), covar=tensor([0.0026, 0.0332, 0.0128, 0.0048, 0.0115, 0.0058, 0.0122, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0121, 0.0103, 0.0075, 0.0100, 0.0113, 0.0094, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:35:42,211 INFO [finetune.py:992] (1/2) Epoch 10, batch 4000, loss[loss=0.1495, simple_loss=0.241, pruned_loss=0.02895, over 12035.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.03997, over 2376559.07 frames. ], batch size: 31, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:35:42,435 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4543, 3.6133, 3.2324, 3.1971, 2.9009, 2.8052, 3.5590, 2.2750], device='cuda:1'), covar=tensor([0.0384, 0.0128, 0.0188, 0.0188, 0.0377, 0.0336, 0.0161, 0.0504], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0161, 0.0158, 0.0184, 0.0201, 0.0197, 0.0168, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:35:46,514 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5686, 4.8641, 3.2285, 2.5676, 4.2984, 2.6770, 4.2274, 3.4943], device='cuda:1'), covar=tensor([0.0627, 0.0555, 0.0891, 0.1640, 0.0228, 0.1287, 0.0436, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0253, 0.0176, 0.0198, 0.0140, 0.0180, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:35:59,921 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214129.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:36:18,100 INFO [finetune.py:992] (1/2) Epoch 10, batch 4050, loss[loss=0.1687, simple_loss=0.2696, pruned_loss=0.03394, over 12183.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.03996, over 2379536.98 frames. ], batch size: 35, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:36:20,871 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.768e+02 3.378e+02 3.956e+02 1.343e+03, threshold=6.756e+02, percent-clipped=3.0 2023-05-16 14:36:43,659 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214190.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:36:54,748 INFO [finetune.py:992] (1/2) Epoch 10, batch 4100, loss[loss=0.1644, simple_loss=0.2497, pruned_loss=0.03954, over 12166.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2573, pruned_loss=0.03973, over 2379820.30 frames. ], batch size: 29, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:37:04,158 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214218.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:37:06,994 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214222.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:37:30,840 INFO [finetune.py:992] (1/2) Epoch 10, batch 4150, loss[loss=0.1952, simple_loss=0.2794, pruned_loss=0.05548, over 12006.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2579, pruned_loss=0.03995, over 2385734.73 frames. ], batch size: 40, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:37:33,638 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 2.613e+02 3.086e+02 3.878e+02 6.591e+02, threshold=6.171e+02, percent-clipped=0.0 2023-05-16 14:37:33,802 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5095, 5.3109, 5.3737, 5.4595, 5.0836, 5.0539, 4.8881, 5.3483], device='cuda:1'), covar=tensor([0.0549, 0.0526, 0.0768, 0.0444, 0.1587, 0.1214, 0.0463, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0692, 0.0600, 0.0617, 0.0831, 0.0735, 0.0543, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:37:41,692 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214270.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:37:48,340 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214279.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:37:57,509 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4035, 3.5972, 3.2553, 3.6728, 3.4644, 2.5747, 3.3042, 2.8226], device='cuda:1'), covar=tensor([0.0909, 0.0967, 0.1532, 0.0825, 0.1304, 0.1629, 0.1156, 0.2844], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0381, 0.0357, 0.0297, 0.0367, 0.0269, 0.0343, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:38:06,249 INFO [finetune.py:992] (1/2) Epoch 10, batch 4200, loss[loss=0.1793, simple_loss=0.2678, pruned_loss=0.04543, over 12118.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2583, pruned_loss=0.04021, over 2370951.53 frames. ], batch size: 38, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:38:13,436 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1151, 5.0247, 4.9482, 5.0467, 4.5998, 5.1345, 4.9841, 5.3500], device='cuda:1'), covar=tensor([0.0251, 0.0136, 0.0169, 0.0292, 0.0811, 0.0293, 0.0177, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0193, 0.0187, 0.0241, 0.0239, 0.0214, 0.0171, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:38:42,423 INFO [finetune.py:992] (1/2) Epoch 10, batch 4250, loss[loss=0.1849, simple_loss=0.2714, pruned_loss=0.04924, over 12026.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2582, pruned_loss=0.04039, over 2364165.44 frames. ], batch size: 40, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:38:45,170 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.824e+02 3.268e+02 4.058e+02 5.640e+02, threshold=6.536e+02, percent-clipped=0.0 2023-05-16 14:38:51,063 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7965, 4.4788, 4.6136, 4.6718, 4.5367, 4.7313, 4.6224, 2.8071], device='cuda:1'), covar=tensor([0.0134, 0.0073, 0.0098, 0.0081, 0.0059, 0.0116, 0.0103, 0.0711], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0078, 0.0081, 0.0073, 0.0060, 0.0091, 0.0081, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:39:09,400 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3576, 5.1236, 5.2438, 5.2801, 4.9039, 4.9519, 4.7033, 5.1907], device='cuda:1'), covar=tensor([0.0607, 0.0626, 0.0830, 0.0511, 0.1832, 0.1291, 0.0555, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0693, 0.0598, 0.0617, 0.0830, 0.0735, 0.0541, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:39:18,809 INFO [finetune.py:992] (1/2) Epoch 10, batch 4300, loss[loss=0.1584, simple_loss=0.2479, pruned_loss=0.03443, over 12086.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2584, pruned_loss=0.04028, over 2362360.79 frames. ], batch size: 32, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:39:33,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.89 vs. limit=5.0 2023-05-16 14:39:46,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 14:39:53,920 INFO [finetune.py:992] (1/2) Epoch 10, batch 4350, loss[loss=0.1701, simple_loss=0.2606, pruned_loss=0.03979, over 12342.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.258, pruned_loss=0.04051, over 2363490.16 frames. ], batch size: 36, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:39:56,689 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 2.611e+02 2.964e+02 3.774e+02 9.007e+02, threshold=5.928e+02, percent-clipped=5.0 2023-05-16 14:40:16,119 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214485.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:40:30,149 INFO [finetune.py:992] (1/2) Epoch 10, batch 4400, loss[loss=0.1334, simple_loss=0.2213, pruned_loss=0.0228, over 12332.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2585, pruned_loss=0.04039, over 2367065.58 frames. ], batch size: 31, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:40:41,832 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6475, 4.9620, 3.2292, 2.8134, 4.3816, 2.8187, 4.3485, 3.5154], device='cuda:1'), covar=tensor([0.0580, 0.0460, 0.1037, 0.1451, 0.0240, 0.1205, 0.0388, 0.0731], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0253, 0.0177, 0.0197, 0.0140, 0.0180, 0.0196, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:40:57,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-05-16 14:41:06,375 INFO [finetune.py:992] (1/2) Epoch 10, batch 4450, loss[loss=0.1956, simple_loss=0.2811, pruned_loss=0.05507, over 7990.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2585, pruned_loss=0.04071, over 2357978.44 frames. ], batch size: 98, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:41:09,101 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.774e+02 3.267e+02 3.826e+02 7.033e+02, threshold=6.534e+02, percent-clipped=2.0 2023-05-16 14:41:20,146 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214574.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:41:23,088 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5928, 3.8127, 3.3767, 3.2453, 2.9152, 2.8729, 3.7926, 2.2980], device='cuda:1'), covar=tensor([0.0389, 0.0129, 0.0189, 0.0197, 0.0415, 0.0369, 0.0118, 0.0478], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0160, 0.0158, 0.0183, 0.0200, 0.0196, 0.0168, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:41:34,305 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9348, 4.1846, 3.6649, 4.4055, 3.9269, 2.8072, 3.9044, 2.8065], device='cuda:1'), covar=tensor([0.0972, 0.0986, 0.1887, 0.0683, 0.1484, 0.1755, 0.1212, 0.3551], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0377, 0.0353, 0.0294, 0.0363, 0.0266, 0.0340, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:41:41,941 INFO [finetune.py:992] (1/2) Epoch 10, batch 4500, loss[loss=0.1625, simple_loss=0.2554, pruned_loss=0.0348, over 12101.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.259, pruned_loss=0.04073, over 2358615.07 frames. ], batch size: 32, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:41:44,222 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9714, 3.6426, 3.7372, 4.2260, 2.8929, 3.7991, 2.4448, 3.8589], device='cuda:1'), covar=tensor([0.1752, 0.0819, 0.0962, 0.0688, 0.1095, 0.0611, 0.1891, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0262, 0.0293, 0.0350, 0.0232, 0.0238, 0.0257, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:41:55,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5837, 2.9076, 4.4597, 4.6921, 2.8621, 2.6555, 2.8449, 2.0923], device='cuda:1'), covar=tensor([0.1586, 0.2911, 0.0490, 0.0434, 0.1283, 0.2286, 0.2765, 0.4132], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0382, 0.0270, 0.0293, 0.0266, 0.0300, 0.0373, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:42:18,443 INFO [finetune.py:992] (1/2) Epoch 10, batch 4550, loss[loss=0.1778, simple_loss=0.2806, pruned_loss=0.03746, over 12297.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2588, pruned_loss=0.04074, over 2360923.32 frames. ], batch size: 34, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:42:21,187 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.657e+02 3.127e+02 3.785e+02 6.147e+02, threshold=6.255e+02, percent-clipped=0.0 2023-05-16 14:42:23,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 14:42:34,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.27 vs. limit=5.0 2023-05-16 14:42:48,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-05-16 14:42:54,234 INFO [finetune.py:992] (1/2) Epoch 10, batch 4600, loss[loss=0.1471, simple_loss=0.2325, pruned_loss=0.03084, over 12210.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2578, pruned_loss=0.04037, over 2371439.60 frames. ], batch size: 29, lr: 4.05e-03, grad_scale: 8.0 2023-05-16 14:43:19,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-16 14:43:29,585 INFO [finetune.py:992] (1/2) Epoch 10, batch 4650, loss[loss=0.1839, simple_loss=0.2782, pruned_loss=0.04477, over 12366.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2577, pruned_loss=0.04038, over 2381317.90 frames. ], batch size: 35, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:43:29,720 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7356, 5.7056, 5.5265, 4.9721, 5.0553, 5.6662, 5.2142, 5.0682], device='cuda:1'), covar=tensor([0.0825, 0.1068, 0.0730, 0.1649, 0.0788, 0.0739, 0.1721, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0550, 0.0513, 0.0630, 0.0417, 0.0712, 0.0779, 0.0566], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:43:30,492 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8491, 4.7933, 4.6785, 4.7347, 4.3235, 4.8445, 4.8212, 5.0300], device='cuda:1'), covar=tensor([0.0249, 0.0164, 0.0219, 0.0333, 0.0824, 0.0337, 0.0172, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0194, 0.0187, 0.0243, 0.0239, 0.0214, 0.0173, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:43:32,374 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.776e+02 3.306e+02 3.949e+02 5.894e+02, threshold=6.612e+02, percent-clipped=0.0 2023-05-16 14:43:51,489 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214785.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:44:05,843 INFO [finetune.py:992] (1/2) Epoch 10, batch 4700, loss[loss=0.19, simple_loss=0.2721, pruned_loss=0.05393, over 12130.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2574, pruned_loss=0.04021, over 2379034.64 frames. ], batch size: 38, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:44:10,336 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2649, 4.0738, 4.1917, 4.5624, 3.2347, 4.1092, 2.6261, 4.2568], device='cuda:1'), covar=tensor([0.1713, 0.0758, 0.0853, 0.0661, 0.1064, 0.0610, 0.1842, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0263, 0.0294, 0.0352, 0.0234, 0.0238, 0.0259, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:44:26,358 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214833.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:44:30,112 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2907, 2.6650, 3.8229, 3.3175, 3.7326, 3.3364, 2.7230, 3.7326], device='cuda:1'), covar=tensor([0.0143, 0.0354, 0.0176, 0.0210, 0.0131, 0.0175, 0.0346, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0204, 0.0187, 0.0183, 0.0213, 0.0160, 0.0196, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:44:42,006 INFO [finetune.py:992] (1/2) Epoch 10, batch 4750, loss[loss=0.1993, simple_loss=0.2743, pruned_loss=0.06216, over 7730.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2573, pruned_loss=0.04042, over 2367329.85 frames. ], batch size: 99, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:44:44,879 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.892e+02 3.369e+02 3.981e+02 7.528e+02, threshold=6.739e+02, percent-clipped=2.0 2023-05-16 14:44:55,991 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214874.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:45:18,169 INFO [finetune.py:992] (1/2) Epoch 10, batch 4800, loss[loss=0.2109, simple_loss=0.2889, pruned_loss=0.06645, over 7823.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2575, pruned_loss=0.04063, over 2365359.81 frames. ], batch size: 98, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:45:30,410 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214922.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:45:53,966 INFO [finetune.py:992] (1/2) Epoch 10, batch 4850, loss[loss=0.1763, simple_loss=0.2718, pruned_loss=0.04039, over 12040.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.258, pruned_loss=0.04063, over 2357534.13 frames. ], batch size: 40, lr: 4.04e-03, grad_scale: 16.0 2023-05-16 14:45:56,629 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.716e+02 3.299e+02 3.755e+02 7.934e+02, threshold=6.598e+02, percent-clipped=1.0 2023-05-16 14:46:30,482 INFO [finetune.py:992] (1/2) Epoch 10, batch 4900, loss[loss=0.1774, simple_loss=0.2658, pruned_loss=0.04451, over 12323.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2574, pruned_loss=0.04042, over 2359569.89 frames. ], batch size: 30, lr: 4.04e-03, grad_scale: 16.0 2023-05-16 14:46:32,218 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8702, 3.3635, 5.1265, 2.6484, 2.9278, 3.8360, 3.2124, 3.9817], device='cuda:1'), covar=tensor([0.0380, 0.1165, 0.0348, 0.1134, 0.1836, 0.1446, 0.1324, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0234, 0.0249, 0.0182, 0.0240, 0.0295, 0.0225, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 14:47:02,302 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5628, 5.3278, 5.3703, 5.4764, 5.0612, 5.1333, 4.8932, 5.3862], device='cuda:1'), covar=tensor([0.0594, 0.0649, 0.0799, 0.0608, 0.2186, 0.1365, 0.0594, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0700, 0.0605, 0.0622, 0.0840, 0.0741, 0.0545, 0.0474], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 14:47:03,062 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9059, 4.8152, 4.7249, 4.7533, 4.4059, 4.9104, 4.8907, 5.0240], device='cuda:1'), covar=tensor([0.0238, 0.0150, 0.0202, 0.0316, 0.0775, 0.0248, 0.0149, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0194, 0.0188, 0.0245, 0.0240, 0.0215, 0.0174, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:47:06,477 INFO [finetune.py:992] (1/2) Epoch 10, batch 4950, loss[loss=0.2331, simple_loss=0.314, pruned_loss=0.07615, over 7794.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2582, pruned_loss=0.04035, over 2362369.58 frames. ], batch size: 98, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:47:09,871 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.873e+02 2.771e+02 3.187e+02 3.647e+02 6.173e+02, threshold=6.375e+02, percent-clipped=0.0 2023-05-16 14:47:12,859 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0761, 4.8580, 4.9366, 5.0209, 4.6427, 4.6834, 4.4643, 4.9357], device='cuda:1'), covar=tensor([0.0662, 0.0667, 0.0866, 0.0626, 0.1986, 0.1369, 0.0623, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0700, 0.0605, 0.0621, 0.0839, 0.0740, 0.0545, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:47:41,986 INFO [finetune.py:992] (1/2) Epoch 10, batch 5000, loss[loss=0.1701, simple_loss=0.2668, pruned_loss=0.03672, over 12158.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2593, pruned_loss=0.04103, over 2365720.86 frames. ], batch size: 36, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:47:43,652 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2609, 4.8228, 5.2188, 4.5656, 4.8237, 4.6210, 5.2625, 4.9200], device='cuda:1'), covar=tensor([0.0271, 0.0394, 0.0290, 0.0276, 0.0405, 0.0336, 0.0217, 0.0252], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0263, 0.0287, 0.0257, 0.0259, 0.0257, 0.0234, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:48:15,274 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2558, 4.5408, 2.7209, 2.4856, 3.9513, 2.3873, 3.9818, 3.0512], device='cuda:1'), covar=tensor([0.0741, 0.0572, 0.1270, 0.1570, 0.0278, 0.1437, 0.0464, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0258, 0.0180, 0.0200, 0.0143, 0.0183, 0.0199, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:48:17,131 INFO [finetune.py:992] (1/2) Epoch 10, batch 5050, loss[loss=0.1643, simple_loss=0.249, pruned_loss=0.03981, over 12176.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.259, pruned_loss=0.04092, over 2364264.43 frames. ], batch size: 31, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:48:20,565 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.832e+02 3.427e+02 4.150e+02 6.907e+02, threshold=6.854e+02, percent-clipped=3.0 2023-05-16 14:48:27,289 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0559, 2.5036, 3.7229, 3.1261, 3.5172, 3.2566, 2.5341, 3.5904], device='cuda:1'), covar=tensor([0.0108, 0.0301, 0.0108, 0.0209, 0.0126, 0.0161, 0.0326, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0204, 0.0188, 0.0184, 0.0213, 0.0160, 0.0197, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:48:44,167 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 14:48:53,163 INFO [finetune.py:992] (1/2) Epoch 10, batch 5100, loss[loss=0.1597, simple_loss=0.2497, pruned_loss=0.03482, over 12172.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.259, pruned_loss=0.04122, over 2361719.45 frames. ], batch size: 31, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:49:01,319 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4153, 4.3380, 4.2769, 4.3060, 3.9399, 4.4752, 4.4222, 4.5543], device='cuda:1'), covar=tensor([0.0229, 0.0159, 0.0219, 0.0395, 0.0839, 0.0327, 0.0188, 0.0213], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0191, 0.0184, 0.0241, 0.0236, 0.0211, 0.0171, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:49:11,609 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215230.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:49:29,926 INFO [finetune.py:992] (1/2) Epoch 10, batch 5150, loss[loss=0.1666, simple_loss=0.2607, pruned_loss=0.03631, over 12095.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2574, pruned_loss=0.04068, over 2362407.69 frames. ], batch size: 32, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:49:33,488 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 2.718e+02 3.092e+02 3.626e+02 6.998e+02, threshold=6.184e+02, percent-clipped=1.0 2023-05-16 14:49:54,279 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1925, 5.9305, 5.6414, 5.4446, 6.0625, 5.2483, 5.4615, 5.5181], device='cuda:1'), covar=tensor([0.1338, 0.0883, 0.0948, 0.1964, 0.0868, 0.2230, 0.1920, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0487, 0.0388, 0.0437, 0.0460, 0.0439, 0.0394, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:49:55,755 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215291.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:50:05,022 INFO [finetune.py:992] (1/2) Epoch 10, batch 5200, loss[loss=0.1869, simple_loss=0.2802, pruned_loss=0.04679, over 12110.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2574, pruned_loss=0.04076, over 2361730.43 frames. ], batch size: 38, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:50:19,359 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215325.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:50:24,965 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215333.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:50:41,084 INFO [finetune.py:992] (1/2) Epoch 10, batch 5250, loss[loss=0.1851, simple_loss=0.2763, pruned_loss=0.04698, over 12350.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2584, pruned_loss=0.04085, over 2359797.23 frames. ], batch size: 36, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:50:44,486 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.743e+02 3.102e+02 3.650e+02 7.283e+02, threshold=6.204e+02, percent-clipped=2.0 2023-05-16 14:50:51,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 14:51:03,095 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215386.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 14:51:09,359 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215394.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 14:51:17,320 INFO [finetune.py:992] (1/2) Epoch 10, batch 5300, loss[loss=0.176, simple_loss=0.2563, pruned_loss=0.04784, over 12349.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2582, pruned_loss=0.04044, over 2369687.73 frames. ], batch size: 30, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:51:32,647 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-05-16 14:51:52,973 INFO [finetune.py:992] (1/2) Epoch 10, batch 5350, loss[loss=0.1534, simple_loss=0.2328, pruned_loss=0.03698, over 12289.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.258, pruned_loss=0.04042, over 2363866.71 frames. ], batch size: 28, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:51:56,655 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 2.929e+02 3.222e+02 3.897e+02 6.738e+02, threshold=6.444e+02, percent-clipped=2.0 2023-05-16 14:52:28,093 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215503.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:52:29,285 INFO [finetune.py:992] (1/2) Epoch 10, batch 5400, loss[loss=0.1392, simple_loss=0.2214, pruned_loss=0.02845, over 12346.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2582, pruned_loss=0.04051, over 2370235.32 frames. ], batch size: 30, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:53:00,872 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215548.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:53:05,527 INFO [finetune.py:992] (1/2) Epoch 10, batch 5450, loss[loss=0.214, simple_loss=0.3005, pruned_loss=0.06374, over 10550.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2577, pruned_loss=0.04035, over 2376824.26 frames. ], batch size: 68, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:53:09,058 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 2.735e+02 3.191e+02 3.702e+02 7.831e+02, threshold=6.383e+02, percent-clipped=1.0 2023-05-16 14:53:12,099 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215564.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:53:12,779 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215565.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:53:27,760 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215586.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 14:53:29,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 14:53:41,301 INFO [finetune.py:992] (1/2) Epoch 10, batch 5500, loss[loss=0.1778, simple_loss=0.2706, pruned_loss=0.0425, over 12152.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2583, pruned_loss=0.04074, over 2367851.23 frames. ], batch size: 34, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:53:42,156 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4777, 5.0128, 5.4089, 4.7304, 5.0395, 4.8319, 5.4646, 5.1096], device='cuda:1'), covar=tensor([0.0216, 0.0351, 0.0237, 0.0254, 0.0310, 0.0287, 0.0166, 0.0231], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0263, 0.0287, 0.0257, 0.0259, 0.0258, 0.0234, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:53:44,321 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215609.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:53:56,968 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215626.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:54:03,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 14:54:17,700 INFO [finetune.py:992] (1/2) Epoch 10, batch 5550, loss[loss=0.2261, simple_loss=0.2994, pruned_loss=0.07639, over 8095.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2582, pruned_loss=0.04044, over 2371814.72 frames. ], batch size: 98, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:54:21,189 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.670e+02 3.182e+02 3.606e+02 3.832e+03, threshold=6.365e+02, percent-clipped=3.0 2023-05-16 14:54:22,978 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4482, 4.5305, 4.1431, 4.8538, 4.5876, 2.8206, 4.2894, 3.0421], device='cuda:1'), covar=tensor([0.0662, 0.0833, 0.1309, 0.0527, 0.0947, 0.1626, 0.0961, 0.3154], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0379, 0.0354, 0.0294, 0.0365, 0.0267, 0.0343, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:54:25,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 14:54:36,860 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215681.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:54:42,511 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215689.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 14:54:53,684 INFO [finetune.py:992] (1/2) Epoch 10, batch 5600, loss[loss=0.1983, simple_loss=0.2845, pruned_loss=0.0561, over 12118.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2591, pruned_loss=0.04091, over 2369621.17 frames. ], batch size: 38, lr: 4.04e-03, grad_scale: 8.0 2023-05-16 14:55:22,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 14:55:27,556 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1212, 5.0842, 5.0096, 4.9777, 4.6145, 5.1305, 5.1361, 5.2825], device='cuda:1'), covar=tensor([0.0247, 0.0130, 0.0173, 0.0308, 0.0769, 0.0295, 0.0133, 0.0174], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0191, 0.0184, 0.0241, 0.0235, 0.0212, 0.0170, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 14:55:29,451 INFO [finetune.py:992] (1/2) Epoch 10, batch 5650, loss[loss=0.1321, simple_loss=0.2171, pruned_loss=0.02357, over 11995.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2581, pruned_loss=0.04064, over 2374879.49 frames. ], batch size: 28, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:55:33,060 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.023e+02 2.706e+02 3.144e+02 3.811e+02 8.174e+02, threshold=6.288e+02, percent-clipped=6.0 2023-05-16 14:56:05,976 INFO [finetune.py:992] (1/2) Epoch 10, batch 5700, loss[loss=0.1744, simple_loss=0.2674, pruned_loss=0.0407, over 10540.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2592, pruned_loss=0.04112, over 2371449.30 frames. ], batch size: 68, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:56:18,785 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3631, 4.8997, 5.3167, 4.6728, 4.9147, 4.6973, 5.3895, 5.0175], device='cuda:1'), covar=tensor([0.0289, 0.0375, 0.0270, 0.0298, 0.0357, 0.0350, 0.0200, 0.0276], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0264, 0.0288, 0.0257, 0.0259, 0.0258, 0.0234, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:56:23,825 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215829.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:56:39,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 14:56:41,854 INFO [finetune.py:992] (1/2) Epoch 10, batch 5750, loss[loss=0.2038, simple_loss=0.2842, pruned_loss=0.06172, over 12160.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2595, pruned_loss=0.0415, over 2365570.84 frames. ], batch size: 39, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:56:44,825 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215859.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:56:45,449 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.768e+02 3.361e+02 4.210e+02 8.925e+02, threshold=6.721e+02, percent-clipped=2.0 2023-05-16 14:57:03,993 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215886.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 14:57:06,774 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215890.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:57:16,546 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215904.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 14:57:17,102 INFO [finetune.py:992] (1/2) Epoch 10, batch 5800, loss[loss=0.1756, simple_loss=0.2614, pruned_loss=0.0449, over 12363.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2588, pruned_loss=0.04121, over 2375041.45 frames. ], batch size: 30, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:57:29,066 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215921.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:57:38,270 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215934.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:57:52,887 INFO [finetune.py:992] (1/2) Epoch 10, batch 5850, loss[loss=0.2165, simple_loss=0.299, pruned_loss=0.067, over 11145.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2588, pruned_loss=0.04136, over 2373521.43 frames. ], batch size: 55, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:57:56,867 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 2.721e+02 3.210e+02 4.010e+02 7.242e+02, threshold=6.420e+02, percent-clipped=2.0 2023-05-16 14:58:11,824 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215981.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:58:17,414 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215989.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 14:58:31,768 INFO [finetune.py:992] (1/2) Epoch 10, batch 5900, loss[loss=0.1543, simple_loss=0.2524, pruned_loss=0.02812, over 12154.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.261, pruned_loss=0.04205, over 2366072.23 frames. ], batch size: 36, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:58:39,269 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0754, 2.5108, 3.6757, 3.0953, 3.5231, 3.2515, 2.4775, 3.5190], device='cuda:1'), covar=tensor([0.0125, 0.0355, 0.0141, 0.0215, 0.0148, 0.0164, 0.0350, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0203, 0.0185, 0.0181, 0.0210, 0.0157, 0.0193, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:58:48,684 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216028.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:58:49,239 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216029.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 14:58:54,924 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216037.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:59:04,318 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3989, 4.6149, 4.2907, 5.0545, 4.6335, 2.5574, 4.1239, 3.0198], device='cuda:1'), covar=tensor([0.0690, 0.0824, 0.1081, 0.0450, 0.0885, 0.1828, 0.1111, 0.3124], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0384, 0.0358, 0.0299, 0.0368, 0.0270, 0.0346, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 14:59:05,354 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 14:59:08,233 INFO [finetune.py:992] (1/2) Epoch 10, batch 5950, loss[loss=0.1638, simple_loss=0.257, pruned_loss=0.03528, over 11718.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2603, pruned_loss=0.04166, over 2364723.87 frames. ], batch size: 48, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:59:11,723 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.189e+02 2.666e+02 3.056e+02 3.587e+02 8.191e+02, threshold=6.112e+02, percent-clipped=1.0 2023-05-16 14:59:13,854 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1142, 5.9821, 5.6047, 5.5037, 6.1567, 5.4544, 5.5945, 5.5724], device='cuda:1'), covar=tensor([0.1323, 0.1068, 0.0916, 0.1948, 0.0852, 0.2021, 0.1844, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0477, 0.0377, 0.0425, 0.0446, 0.0427, 0.0383, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 14:59:32,585 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216089.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 14:59:42,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 14:59:44,515 INFO [finetune.py:992] (1/2) Epoch 10, batch 6000, loss[loss=0.2356, simple_loss=0.3115, pruned_loss=0.07984, over 8414.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2604, pruned_loss=0.04173, over 2369445.02 frames. ], batch size: 98, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 14:59:44,516 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 15:00:03,003 INFO [finetune.py:1026] (1/2) Epoch 10, validation: loss=0.3144, simple_loss=0.3924, pruned_loss=0.1182, over 1020973.00 frames. 2023-05-16 15:00:03,004 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 15:00:38,506 INFO [finetune.py:992] (1/2) Epoch 10, batch 6050, loss[loss=0.2088, simple_loss=0.2993, pruned_loss=0.05909, over 11818.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2598, pruned_loss=0.04155, over 2372477.02 frames. ], batch size: 44, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:00:42,132 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216159.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:00:42,701 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.596e+02 3.153e+02 3.700e+02 5.951e+02, threshold=6.306e+02, percent-clipped=0.0 2023-05-16 15:00:52,985 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1919, 5.9451, 5.6289, 5.5031, 6.1338, 5.3043, 5.5696, 5.5553], device='cuda:1'), covar=tensor([0.1323, 0.1033, 0.1033, 0.2037, 0.0927, 0.2198, 0.1916, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0475, 0.0375, 0.0424, 0.0446, 0.0426, 0.0382, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:01:00,424 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7104, 2.7541, 4.4184, 4.5449, 2.9496, 2.6024, 2.9982, 2.1204], device='cuda:1'), covar=tensor([0.1532, 0.3022, 0.0494, 0.0434, 0.1196, 0.2362, 0.2533, 0.4055], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0380, 0.0269, 0.0294, 0.0265, 0.0299, 0.0372, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:01:00,918 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216185.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:01:15,309 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216204.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:01:15,864 INFO [finetune.py:992] (1/2) Epoch 10, batch 6100, loss[loss=0.1825, simple_loss=0.2746, pruned_loss=0.04518, over 10461.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2599, pruned_loss=0.04157, over 2374597.45 frames. ], batch size: 68, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:01:17,333 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216207.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:01:27,354 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216221.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:01:31,207 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.61 vs. limit=5.0 2023-05-16 15:01:49,238 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216252.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:01:51,228 INFO [finetune.py:992] (1/2) Epoch 10, batch 6150, loss[loss=0.1667, simple_loss=0.2513, pruned_loss=0.041, over 12123.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2593, pruned_loss=0.04129, over 2371656.27 frames. ], batch size: 30, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:01:54,547 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.108e+02 2.856e+02 3.329e+02 3.835e+02 6.242e+02, threshold=6.658e+02, percent-clipped=0.0 2023-05-16 15:02:01,054 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216269.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:02:09,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-05-16 15:02:18,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-05-16 15:02:26,042 INFO [finetune.py:992] (1/2) Epoch 10, batch 6200, loss[loss=0.1846, simple_loss=0.2741, pruned_loss=0.0475, over 12055.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2604, pruned_loss=0.04183, over 2368738.87 frames. ], batch size: 42, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:03:02,658 INFO [finetune.py:992] (1/2) Epoch 10, batch 6250, loss[loss=0.145, simple_loss=0.2292, pruned_loss=0.03035, over 11988.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2605, pruned_loss=0.04171, over 2365643.20 frames. ], batch size: 28, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:03:06,156 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.634e+02 3.207e+02 4.091e+02 8.473e+02, threshold=6.414e+02, percent-clipped=5.0 2023-05-16 15:03:23,145 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216384.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:03:26,028 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1120, 4.9886, 4.9302, 4.9626, 4.5872, 5.0835, 5.1568, 5.2564], device='cuda:1'), covar=tensor([0.0167, 0.0159, 0.0207, 0.0369, 0.0769, 0.0337, 0.0126, 0.0185], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0195, 0.0188, 0.0246, 0.0242, 0.0216, 0.0172, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:03:38,235 INFO [finetune.py:992] (1/2) Epoch 10, batch 6300, loss[loss=0.1718, simple_loss=0.2661, pruned_loss=0.03869, over 11676.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2598, pruned_loss=0.04125, over 2367977.50 frames. ], batch size: 48, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:03:59,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5541, 4.3823, 4.3095, 4.6563, 3.4665, 4.1697, 2.8821, 4.3220], device='cuda:1'), covar=tensor([0.1390, 0.0583, 0.0861, 0.0590, 0.0964, 0.0529, 0.1552, 0.1331], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0265, 0.0295, 0.0353, 0.0235, 0.0240, 0.0258, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:04:13,776 INFO [finetune.py:992] (1/2) Epoch 10, batch 6350, loss[loss=0.1526, simple_loss=0.2352, pruned_loss=0.03502, over 12295.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2599, pruned_loss=0.04095, over 2374811.92 frames. ], batch size: 28, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:04:17,376 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.032e+02 2.790e+02 3.027e+02 3.610e+02 6.704e+02, threshold=6.054e+02, percent-clipped=1.0 2023-05-16 15:04:18,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 15:04:24,545 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0588, 6.0004, 5.7967, 5.3944, 5.2929, 5.9836, 5.5774, 5.3374], device='cuda:1'), covar=tensor([0.0709, 0.0973, 0.0663, 0.1631, 0.0635, 0.0663, 0.1377, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0550, 0.0517, 0.0630, 0.0414, 0.0714, 0.0774, 0.0569], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 15:04:35,980 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216485.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:04:50,462 INFO [finetune.py:992] (1/2) Epoch 10, batch 6400, loss[loss=0.1667, simple_loss=0.259, pruned_loss=0.03723, over 12126.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2599, pruned_loss=0.04087, over 2377181.38 frames. ], batch size: 39, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:04:58,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 15:05:10,428 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216533.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:05:26,274 INFO [finetune.py:992] (1/2) Epoch 10, batch 6450, loss[loss=0.1632, simple_loss=0.2495, pruned_loss=0.03838, over 11817.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2589, pruned_loss=0.04076, over 2383758.27 frames. ], batch size: 44, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:05:29,777 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.061e+02 2.703e+02 3.234e+02 4.071e+02 9.218e+02, threshold=6.469e+02, percent-clipped=5.0 2023-05-16 15:05:51,690 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216591.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:05:59,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 15:06:01,440 INFO [finetune.py:992] (1/2) Epoch 10, batch 6500, loss[loss=0.1672, simple_loss=0.2539, pruned_loss=0.04028, over 12117.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2598, pruned_loss=0.04102, over 2383416.67 frames. ], batch size: 30, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:06:03,857 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1292, 3.6348, 5.3802, 2.7016, 3.0712, 3.8620, 3.7578, 3.8989], device='cuda:1'), covar=tensor([0.0421, 0.0984, 0.0249, 0.1207, 0.1759, 0.1628, 0.1031, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0236, 0.0249, 0.0181, 0.0239, 0.0295, 0.0224, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:06:15,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.28 vs. limit=5.0 2023-05-16 15:06:35,877 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216652.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:06:37,804 INFO [finetune.py:992] (1/2) Epoch 10, batch 6550, loss[loss=0.1697, simple_loss=0.2651, pruned_loss=0.03714, over 12069.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2599, pruned_loss=0.04087, over 2386918.91 frames. ], batch size: 40, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:06:37,950 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216655.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:06:41,342 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 2.848e+02 3.255e+02 3.871e+02 6.183e+02, threshold=6.511e+02, percent-clipped=0.0 2023-05-16 15:06:58,230 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216684.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:07:13,055 INFO [finetune.py:992] (1/2) Epoch 10, batch 6600, loss[loss=0.1481, simple_loss=0.2368, pruned_loss=0.02969, over 12118.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.26, pruned_loss=0.0411, over 2368001.85 frames. ], batch size: 33, lr: 4.03e-03, grad_scale: 8.0 2023-05-16 15:07:13,226 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216705.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:07:21,114 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216716.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:07:32,392 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216732.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:07:49,372 INFO [finetune.py:992] (1/2) Epoch 10, batch 6650, loss[loss=0.1635, simple_loss=0.2537, pruned_loss=0.03663, over 12185.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2608, pruned_loss=0.04142, over 2370308.67 frames. ], batch size: 31, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:07:53,064 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.856e+02 3.449e+02 3.977e+02 7.392e+02, threshold=6.898e+02, percent-clipped=2.0 2023-05-16 15:07:57,554 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216766.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:08:26,170 INFO [finetune.py:992] (1/2) Epoch 10, batch 6700, loss[loss=0.1681, simple_loss=0.2633, pruned_loss=0.03644, over 12012.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2596, pruned_loss=0.04074, over 2374331.26 frames. ], batch size: 40, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:08:37,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1767, 3.9438, 2.5571, 2.3550, 3.4983, 2.4105, 3.6318, 2.9131], device='cuda:1'), covar=tensor([0.0629, 0.0630, 0.1072, 0.1404, 0.0357, 0.1273, 0.0509, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0254, 0.0177, 0.0198, 0.0142, 0.0182, 0.0197, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:09:01,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 15:09:01,825 INFO [finetune.py:992] (1/2) Epoch 10, batch 6750, loss[loss=0.1915, simple_loss=0.2903, pruned_loss=0.04635, over 12374.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2592, pruned_loss=0.04053, over 2383455.64 frames. ], batch size: 38, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:09:05,470 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.677e+02 3.140e+02 3.668e+02 5.488e+02, threshold=6.280e+02, percent-clipped=0.0 2023-05-16 15:09:38,010 INFO [finetune.py:992] (1/2) Epoch 10, batch 6800, loss[loss=0.167, simple_loss=0.2556, pruned_loss=0.03919, over 12098.00 frames. ], tot_loss[loss=0.171, simple_loss=0.26, pruned_loss=0.04098, over 2378202.59 frames. ], batch size: 32, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:10:08,079 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216947.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:10:13,669 INFO [finetune.py:992] (1/2) Epoch 10, batch 6850, loss[loss=0.1763, simple_loss=0.2697, pruned_loss=0.04145, over 12365.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2593, pruned_loss=0.04087, over 2379267.65 frames. ], batch size: 35, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:10:17,031 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.810e+02 3.435e+02 3.923e+02 9.714e+02, threshold=6.870e+02, percent-clipped=3.0 2023-05-16 15:10:20,676 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:10:48,868 INFO [finetune.py:992] (1/2) Epoch 10, batch 6900, loss[loss=0.2045, simple_loss=0.295, pruned_loss=0.05699, over 10292.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2601, pruned_loss=0.0413, over 2375844.45 frames. ], batch size: 68, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:10:53,247 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217011.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 15:10:54,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5162, 5.1266, 5.5051, 4.8592, 5.0914, 4.9347, 5.5525, 5.1592], device='cuda:1'), covar=tensor([0.0229, 0.0342, 0.0241, 0.0235, 0.0326, 0.0284, 0.0179, 0.0201], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0268, 0.0293, 0.0262, 0.0266, 0.0261, 0.0236, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:11:03,864 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217026.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:11:11,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2589, 5.1626, 5.1142, 5.1291, 4.7601, 5.2733, 5.2154, 5.3880], device='cuda:1'), covar=tensor([0.0257, 0.0131, 0.0176, 0.0285, 0.0726, 0.0359, 0.0148, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0197, 0.0189, 0.0247, 0.0242, 0.0215, 0.0173, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:11:24,437 INFO [finetune.py:992] (1/2) Epoch 10, batch 6950, loss[loss=0.188, simple_loss=0.2734, pruned_loss=0.05131, over 11130.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2587, pruned_loss=0.04066, over 2385438.94 frames. ], batch size: 55, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:11:28,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.811e+02 3.227e+02 3.716e+02 6.951e+02, threshold=6.455e+02, percent-clipped=1.0 2023-05-16 15:11:29,374 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217061.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:12:01,373 INFO [finetune.py:992] (1/2) Epoch 10, batch 7000, loss[loss=0.1797, simple_loss=0.2658, pruned_loss=0.04677, over 12128.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2583, pruned_loss=0.04056, over 2387677.94 frames. ], batch size: 39, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:12:17,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.88 vs. limit=5.0 2023-05-16 15:12:36,437 INFO [finetune.py:992] (1/2) Epoch 10, batch 7050, loss[loss=0.1726, simple_loss=0.2633, pruned_loss=0.0409, over 12304.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2585, pruned_loss=0.04065, over 2389142.87 frames. ], batch size: 34, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:12:39,933 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.825e+02 3.318e+02 4.081e+02 8.200e+02, threshold=6.636e+02, percent-clipped=0.0 2023-05-16 15:12:53,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 15:12:56,192 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1642, 4.9717, 5.1050, 5.1344, 4.7835, 4.8342, 4.6052, 5.0621], device='cuda:1'), covar=tensor([0.0600, 0.0592, 0.0818, 0.0526, 0.1797, 0.1170, 0.0542, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0690, 0.0595, 0.0617, 0.0832, 0.0728, 0.0539, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:13:12,713 INFO [finetune.py:992] (1/2) Epoch 10, batch 7100, loss[loss=0.1815, simple_loss=0.2689, pruned_loss=0.04703, over 11998.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2581, pruned_loss=0.0403, over 2391480.45 frames. ], batch size: 40, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:13:43,128 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217247.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:13:48,455 INFO [finetune.py:992] (1/2) Epoch 10, batch 7150, loss[loss=0.1574, simple_loss=0.2509, pruned_loss=0.03192, over 12029.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2584, pruned_loss=0.04061, over 2382506.54 frames. ], batch size: 31, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:13:51,964 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 2.573e+02 3.098e+02 4.027e+02 8.429e+02, threshold=6.196e+02, percent-clipped=2.0 2023-05-16 15:14:02,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 15:14:14,731 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7490, 3.0889, 3.9477, 4.6364, 4.1553, 4.7663, 4.1294, 3.3808], device='cuda:1'), covar=tensor([0.0024, 0.0301, 0.0105, 0.0028, 0.0082, 0.0047, 0.0106, 0.0291], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0124, 0.0105, 0.0075, 0.0102, 0.0115, 0.0096, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:14:16,733 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217295.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:14:23,738 INFO [finetune.py:992] (1/2) Epoch 10, batch 7200, loss[loss=0.1967, simple_loss=0.2848, pruned_loss=0.05424, over 12113.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.04056, over 2381385.41 frames. ], batch size: 38, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:14:27,990 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217311.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 15:14:34,901 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217321.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:14:59,444 INFO [finetune.py:992] (1/2) Epoch 10, batch 7250, loss[loss=0.187, simple_loss=0.2854, pruned_loss=0.04431, over 12012.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.04002, over 2384703.04 frames. ], batch size: 40, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:15:02,431 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217359.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 15:15:02,997 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.680e+02 3.204e+02 3.936e+02 1.205e+03, threshold=6.408e+02, percent-clipped=3.0 2023-05-16 15:15:03,864 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217361.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:15:23,853 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217388.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:15:35,816 INFO [finetune.py:992] (1/2) Epoch 10, batch 7300, loss[loss=0.1711, simple_loss=0.2624, pruned_loss=0.03988, over 12290.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2585, pruned_loss=0.0409, over 2373472.59 frames. ], batch size: 37, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:15:38,714 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217409.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:16:07,292 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217449.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:16:11,448 INFO [finetune.py:992] (1/2) Epoch 10, batch 7350, loss[loss=0.1468, simple_loss=0.2316, pruned_loss=0.03097, over 12336.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2591, pruned_loss=0.04153, over 2367473.78 frames. ], batch size: 31, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:16:14,994 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.128e+02 2.746e+02 3.216e+02 3.952e+02 7.752e+02, threshold=6.432e+02, percent-clipped=2.0 2023-05-16 15:16:24,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-05-16 15:16:37,666 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 15:16:37,908 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217492.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:16:47,579 INFO [finetune.py:992] (1/2) Epoch 10, batch 7400, loss[loss=0.166, simple_loss=0.2555, pruned_loss=0.03826, over 12191.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2593, pruned_loss=0.04177, over 2370650.14 frames. ], batch size: 35, lr: 4.02e-03, grad_scale: 16.0 2023-05-16 15:17:06,293 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4259, 4.7280, 4.1546, 5.0078, 4.6298, 2.9897, 4.3403, 3.2600], device='cuda:1'), covar=tensor([0.0727, 0.0691, 0.1366, 0.0403, 0.1052, 0.1485, 0.1007, 0.2796], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0381, 0.0356, 0.0296, 0.0365, 0.0267, 0.0346, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:17:22,072 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217553.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:17:23,270 INFO [finetune.py:992] (1/2) Epoch 10, batch 7450, loss[loss=0.1488, simple_loss=0.2298, pruned_loss=0.03389, over 12278.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2598, pruned_loss=0.04171, over 2375485.74 frames. ], batch size: 28, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:17:26,951 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2163, 4.8409, 5.2379, 4.4592, 4.8526, 4.5116, 5.2109, 4.9450], device='cuda:1'), covar=tensor([0.0323, 0.0417, 0.0316, 0.0353, 0.0389, 0.0377, 0.0319, 0.0297], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0267, 0.0291, 0.0261, 0.0264, 0.0259, 0.0235, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:17:27,442 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.751e+02 3.168e+02 3.931e+02 8.735e+02, threshold=6.336e+02, percent-clipped=4.0 2023-05-16 15:17:45,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 15:17:47,786 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2797, 2.7033, 3.7877, 3.2719, 3.6471, 3.4823, 2.7960, 3.7238], device='cuda:1'), covar=tensor([0.0114, 0.0326, 0.0154, 0.0223, 0.0149, 0.0148, 0.0303, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0205, 0.0187, 0.0185, 0.0213, 0.0160, 0.0195, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:17:50,157 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-05-16 15:17:58,945 INFO [finetune.py:992] (1/2) Epoch 10, batch 7500, loss[loss=0.1881, simple_loss=0.2825, pruned_loss=0.04687, over 11254.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2603, pruned_loss=0.04222, over 2366120.91 frames. ], batch size: 55, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:18:10,416 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:18:34,875 INFO [finetune.py:992] (1/2) Epoch 10, batch 7550, loss[loss=0.1618, simple_loss=0.2541, pruned_loss=0.0348, over 12352.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.26, pruned_loss=0.04194, over 2372504.82 frames. ], batch size: 35, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:18:39,784 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.719e+02 3.185e+02 3.698e+02 8.240e+02, threshold=6.370e+02, percent-clipped=1.0 2023-05-16 15:18:42,023 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217664.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:18:45,380 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217669.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:19:10,692 INFO [finetune.py:992] (1/2) Epoch 10, batch 7600, loss[loss=0.181, simple_loss=0.2753, pruned_loss=0.04336, over 12034.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2602, pruned_loss=0.04223, over 2365551.86 frames. ], batch size: 40, lr: 4.02e-03, grad_scale: 8.0 2023-05-16 15:19:19,571 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5346, 4.8643, 4.2289, 5.0831, 4.6449, 3.0969, 4.4926, 3.1953], device='cuda:1'), covar=tensor([0.0777, 0.0683, 0.1458, 0.0446, 0.1281, 0.1518, 0.0937, 0.3019], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0381, 0.0357, 0.0298, 0.0366, 0.0267, 0.0346, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:19:25,208 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217725.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:19:38,438 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217744.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:19:46,206 INFO [finetune.py:992] (1/2) Epoch 10, batch 7650, loss[loss=0.1821, simple_loss=0.2785, pruned_loss=0.04283, over 12051.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2599, pruned_loss=0.04188, over 2366632.02 frames. ], batch size: 45, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:19:46,444 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7296, 3.8597, 3.3025, 3.3376, 3.1063, 3.0518, 3.8317, 2.5793], device='cuda:1'), covar=tensor([0.0369, 0.0116, 0.0205, 0.0211, 0.0372, 0.0313, 0.0111, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0159, 0.0156, 0.0182, 0.0199, 0.0194, 0.0167, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:19:48,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 15:19:50,505 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.731e+02 3.164e+02 3.954e+02 1.132e+03, threshold=6.329e+02, percent-clipped=2.0 2023-05-16 15:20:01,523 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2215, 5.1429, 5.0565, 5.1495, 4.7110, 5.2565, 5.1948, 5.3741], device='cuda:1'), covar=tensor([0.0348, 0.0152, 0.0253, 0.0275, 0.0820, 0.0354, 0.0155, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0196, 0.0188, 0.0245, 0.0240, 0.0215, 0.0171, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:20:04,359 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5624, 3.8243, 3.3516, 3.2640, 3.0705, 2.9447, 3.7834, 2.4388], device='cuda:1'), covar=tensor([0.0372, 0.0101, 0.0181, 0.0202, 0.0356, 0.0309, 0.0134, 0.0456], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0159, 0.0156, 0.0182, 0.0199, 0.0195, 0.0167, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:20:23,912 INFO [finetune.py:992] (1/2) Epoch 10, batch 7700, loss[loss=0.1883, simple_loss=0.2749, pruned_loss=0.05085, over 12191.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2582, pruned_loss=0.04125, over 2366933.12 frames. ], batch size: 35, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:20:54,629 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217848.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:20:56,124 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4807, 3.7073, 3.2656, 3.1713, 2.9829, 2.8446, 3.6074, 2.3283], device='cuda:1'), covar=tensor([0.0385, 0.0093, 0.0164, 0.0180, 0.0318, 0.0284, 0.0121, 0.0411], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0159, 0.0155, 0.0182, 0.0199, 0.0194, 0.0167, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:20:59,401 INFO [finetune.py:992] (1/2) Epoch 10, batch 7750, loss[loss=0.1529, simple_loss=0.2363, pruned_loss=0.03474, over 12193.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2589, pruned_loss=0.04141, over 2368990.65 frames. ], batch size: 29, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:21:03,597 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.707e+02 2.759e+02 3.257e+02 3.915e+02 7.146e+02, threshold=6.515e+02, percent-clipped=1.0 2023-05-16 15:21:27,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2023-05-16 15:21:28,774 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9204, 4.8421, 4.7901, 4.8332, 4.4543, 4.9643, 4.9027, 5.0616], device='cuda:1'), covar=tensor([0.0236, 0.0150, 0.0179, 0.0329, 0.0742, 0.0315, 0.0158, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0195, 0.0188, 0.0245, 0.0239, 0.0216, 0.0171, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:21:30,229 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217898.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:21:35,072 INFO [finetune.py:992] (1/2) Epoch 10, batch 7800, loss[loss=0.167, simple_loss=0.2597, pruned_loss=0.03709, over 11565.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2587, pruned_loss=0.04138, over 2363939.42 frames. ], batch size: 48, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:22:11,811 INFO [finetune.py:992] (1/2) Epoch 10, batch 7850, loss[loss=0.1816, simple_loss=0.2694, pruned_loss=0.04689, over 12062.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2585, pruned_loss=0.04132, over 2370072.31 frames. ], batch size: 42, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:22:14,734 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217959.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 15:22:15,901 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.816e+02 3.296e+02 4.410e+02 8.166e+02, threshold=6.592e+02, percent-clipped=3.0 2023-05-16 15:22:32,413 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1406, 3.2848, 2.9973, 2.9638, 2.6770, 2.5533, 3.3314, 2.0804], device='cuda:1'), covar=tensor([0.0501, 0.0155, 0.0226, 0.0224, 0.0450, 0.0355, 0.0129, 0.0509], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0159, 0.0155, 0.0182, 0.0200, 0.0195, 0.0167, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:22:38,137 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0665, 5.9312, 5.5872, 5.5003, 6.0915, 5.4159, 5.6623, 5.5297], device='cuda:1'), covar=tensor([0.1349, 0.0970, 0.1007, 0.1839, 0.0798, 0.2130, 0.1830, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0495, 0.0395, 0.0443, 0.0467, 0.0446, 0.0398, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:22:50,564 INFO [finetune.py:992] (1/2) Epoch 10, batch 7900, loss[loss=0.1512, simple_loss=0.2425, pruned_loss=0.02995, over 12174.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2582, pruned_loss=0.04072, over 2372517.83 frames. ], batch size: 31, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:23:01,341 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218020.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:23:10,658 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2151, 4.7583, 5.0524, 5.0408, 4.8031, 5.1202, 4.9992, 2.7373], device='cuda:1'), covar=tensor([0.0080, 0.0070, 0.0065, 0.0055, 0.0046, 0.0078, 0.0063, 0.0717], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0080, 0.0072, 0.0060, 0.0089, 0.0079, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:23:18,367 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218044.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:23:22,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6662, 2.8077, 3.7469, 4.5188, 3.9229, 4.5050, 4.0062, 3.2675], device='cuda:1'), covar=tensor([0.0027, 0.0337, 0.0130, 0.0031, 0.0122, 0.0074, 0.0099, 0.0323], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0124, 0.0106, 0.0076, 0.0102, 0.0116, 0.0095, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:23:23,358 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7074, 2.9228, 3.8781, 4.5549, 4.0024, 4.5972, 4.1267, 3.5957], device='cuda:1'), covar=tensor([0.0030, 0.0335, 0.0113, 0.0034, 0.0108, 0.0057, 0.0078, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0124, 0.0106, 0.0076, 0.0102, 0.0116, 0.0095, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:23:26,575 INFO [finetune.py:992] (1/2) Epoch 10, batch 7950, loss[loss=0.2591, simple_loss=0.3204, pruned_loss=0.09889, over 7383.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2584, pruned_loss=0.04063, over 2371023.22 frames. ], batch size: 97, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:23:30,786 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.590e+02 3.175e+02 4.044e+02 9.490e+02, threshold=6.351e+02, percent-clipped=3.0 2023-05-16 15:23:33,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0 2023-05-16 15:23:46,947 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5427, 3.8303, 3.3304, 3.3114, 3.1116, 2.9534, 3.7845, 2.6152], device='cuda:1'), covar=tensor([0.0388, 0.0112, 0.0200, 0.0188, 0.0355, 0.0316, 0.0151, 0.0380], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0160, 0.0155, 0.0182, 0.0200, 0.0195, 0.0167, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:23:48,392 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4324, 4.7975, 3.0679, 2.5389, 4.1629, 2.5833, 4.0518, 3.3239], device='cuda:1'), covar=tensor([0.0671, 0.0444, 0.0929, 0.1461, 0.0342, 0.1152, 0.0453, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0253, 0.0176, 0.0198, 0.0140, 0.0179, 0.0196, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:23:53,230 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218092.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:23:59,175 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218100.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:24:02,448 INFO [finetune.py:992] (1/2) Epoch 10, batch 8000, loss[loss=0.1388, simple_loss=0.2257, pruned_loss=0.02597, over 12179.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2594, pruned_loss=0.04135, over 2374768.19 frames. ], batch size: 31, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:24:32,923 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218148.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:24:37,729 INFO [finetune.py:992] (1/2) Epoch 10, batch 8050, loss[loss=0.171, simple_loss=0.2642, pruned_loss=0.03891, over 12102.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2601, pruned_loss=0.04192, over 2363124.98 frames. ], batch size: 33, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:24:41,972 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.803e+02 3.296e+02 3.767e+02 8.425e+02, threshold=6.591e+02, percent-clipped=3.0 2023-05-16 15:24:42,250 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218161.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:24:57,239 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218182.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:25:07,059 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218196.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:25:14,225 INFO [finetune.py:992] (1/2) Epoch 10, batch 8100, loss[loss=0.1867, simple_loss=0.2819, pruned_loss=0.0457, over 12351.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2595, pruned_loss=0.04116, over 2377400.75 frames. ], batch size: 36, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:25:18,022 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218210.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:25:42,034 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218243.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:25:49,814 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218254.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:25:50,443 INFO [finetune.py:992] (1/2) Epoch 10, batch 8150, loss[loss=0.14, simple_loss=0.2233, pruned_loss=0.02839, over 11758.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2596, pruned_loss=0.04105, over 2372196.58 frames. ], batch size: 26, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:25:54,531 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.787e+02 3.392e+02 4.159e+02 9.116e+02, threshold=6.785e+02, percent-clipped=3.0 2023-05-16 15:26:01,825 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218271.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 15:26:25,829 INFO [finetune.py:992] (1/2) Epoch 10, batch 8200, loss[loss=0.1799, simple_loss=0.2727, pruned_loss=0.04355, over 12154.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2596, pruned_loss=0.04142, over 2372868.18 frames. ], batch size: 36, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:26:36,376 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218320.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:27:01,517 INFO [finetune.py:992] (1/2) Epoch 10, batch 8250, loss[loss=0.1686, simple_loss=0.2657, pruned_loss=0.03579, over 11860.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2606, pruned_loss=0.04175, over 2369981.94 frames. ], batch size: 44, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:27:05,743 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.826e+02 3.147e+02 3.997e+02 2.145e+03, threshold=6.293e+02, percent-clipped=2.0 2023-05-16 15:27:11,575 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218368.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:27:38,230 INFO [finetune.py:992] (1/2) Epoch 10, batch 8300, loss[loss=0.1899, simple_loss=0.2791, pruned_loss=0.05036, over 12275.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2595, pruned_loss=0.04117, over 2373979.54 frames. ], batch size: 37, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:28:06,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 15:28:07,296 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2534, 4.9743, 5.1518, 5.1352, 4.9173, 5.2012, 5.0436, 2.7286], device='cuda:1'), covar=tensor([0.0082, 0.0048, 0.0060, 0.0051, 0.0043, 0.0066, 0.0056, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0079, 0.0072, 0.0059, 0.0089, 0.0079, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:28:13,561 INFO [finetune.py:992] (1/2) Epoch 10, batch 8350, loss[loss=0.1843, simple_loss=0.2867, pruned_loss=0.04097, over 12159.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2601, pruned_loss=0.04136, over 2361776.58 frames. ], batch size: 36, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:28:14,339 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218456.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:28:17,978 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.649e+02 3.148e+02 3.755e+02 8.260e+02, threshold=6.296e+02, percent-clipped=2.0 2023-05-16 15:28:49,594 INFO [finetune.py:992] (1/2) Epoch 10, batch 8400, loss[loss=0.1636, simple_loss=0.2643, pruned_loss=0.03147, over 12359.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2599, pruned_loss=0.04142, over 2364597.49 frames. ], batch size: 35, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:28:53,369 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218509.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:29:04,273 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8503, 3.0071, 4.6942, 4.8856, 2.8523, 2.7102, 3.1105, 2.1867], device='cuda:1'), covar=tensor([0.1469, 0.2851, 0.0405, 0.0373, 0.1280, 0.2378, 0.2567, 0.3892], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0377, 0.0268, 0.0295, 0.0265, 0.0297, 0.0369, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:29:13,993 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218538.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:29:25,390 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218554.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:29:25,960 INFO [finetune.py:992] (1/2) Epoch 10, batch 8450, loss[loss=0.1825, simple_loss=0.278, pruned_loss=0.04354, over 10506.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2595, pruned_loss=0.04118, over 2361421.83 frames. ], batch size: 68, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:29:30,156 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.285e+02 2.848e+02 3.305e+02 3.959e+02 6.215e+02, threshold=6.610e+02, percent-clipped=0.0 2023-05-16 15:29:33,785 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218566.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 15:29:36,160 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7798, 4.2307, 3.8945, 4.4963, 3.6854, 4.1244, 2.7424, 4.4986], device='cuda:1'), covar=tensor([0.1135, 0.0595, 0.1246, 0.0879, 0.0814, 0.0480, 0.1578, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0267, 0.0300, 0.0358, 0.0238, 0.0241, 0.0261, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:29:36,859 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218570.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:29:39,757 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218574.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:29:57,663 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6316, 3.2612, 4.9336, 2.4332, 2.4878, 3.6080, 3.1192, 3.7404], device='cuda:1'), covar=tensor([0.0408, 0.1175, 0.0427, 0.1239, 0.2104, 0.1562, 0.1338, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0235, 0.0246, 0.0180, 0.0236, 0.0294, 0.0223, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:29:59,966 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218602.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:30:01,933 INFO [finetune.py:992] (1/2) Epoch 10, batch 8500, loss[loss=0.1517, simple_loss=0.2322, pruned_loss=0.03558, over 12355.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2596, pruned_loss=0.04165, over 2358661.66 frames. ], batch size: 30, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:30:15,732 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9555, 4.4359, 3.8047, 4.7062, 4.2449, 2.7311, 4.0283, 2.9567], device='cuda:1'), covar=tensor([0.0980, 0.0796, 0.1583, 0.0541, 0.1271, 0.1704, 0.1150, 0.3149], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0378, 0.0355, 0.0299, 0.0365, 0.0267, 0.0345, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:30:19,235 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7760, 4.6358, 4.5918, 4.5899, 4.2919, 4.6622, 4.7404, 4.9364], device='cuda:1'), covar=tensor([0.0263, 0.0187, 0.0209, 0.0447, 0.0847, 0.0438, 0.0191, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0197, 0.0189, 0.0247, 0.0241, 0.0218, 0.0172, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:30:24,097 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218635.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:30:38,837 INFO [finetune.py:992] (1/2) Epoch 10, batch 8550, loss[loss=0.15, simple_loss=0.2277, pruned_loss=0.03611, over 12008.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.259, pruned_loss=0.04123, over 2369863.29 frames. ], batch size: 28, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:30:42,992 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.716e+02 3.290e+02 3.941e+02 1.023e+03, threshold=6.580e+02, percent-clipped=1.0 2023-05-16 15:30:48,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 15:30:51,927 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1645, 4.8287, 5.0129, 4.9931, 4.9109, 5.1185, 4.9456, 2.6850], device='cuda:1'), covar=tensor([0.0074, 0.0060, 0.0072, 0.0062, 0.0041, 0.0077, 0.0066, 0.0745], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0080, 0.0072, 0.0059, 0.0090, 0.0079, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:31:14,702 INFO [finetune.py:992] (1/2) Epoch 10, batch 8600, loss[loss=0.1957, simple_loss=0.2856, pruned_loss=0.05289, over 12111.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2603, pruned_loss=0.04197, over 2356930.35 frames. ], batch size: 39, lr: 4.01e-03, grad_scale: 8.0 2023-05-16 15:31:41,065 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9779, 2.4732, 3.5483, 2.9437, 3.3656, 3.1209, 2.4574, 3.4928], device='cuda:1'), covar=tensor([0.0140, 0.0352, 0.0188, 0.0251, 0.0146, 0.0195, 0.0363, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0204, 0.0185, 0.0183, 0.0214, 0.0159, 0.0194, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:31:41,795 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7508, 2.9072, 3.8755, 4.6263, 4.0189, 4.6124, 4.0023, 3.3655], device='cuda:1'), covar=tensor([0.0028, 0.0350, 0.0136, 0.0037, 0.0119, 0.0060, 0.0137, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0125, 0.0107, 0.0076, 0.0103, 0.0117, 0.0097, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:31:50,212 INFO [finetune.py:992] (1/2) Epoch 10, batch 8650, loss[loss=0.1596, simple_loss=0.2454, pruned_loss=0.03694, over 12418.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2589, pruned_loss=0.04123, over 2367429.77 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:31:51,138 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218756.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:31:54,418 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.573e+02 2.817e+02 3.099e+02 3.578e+02 8.400e+02, threshold=6.197e+02, percent-clipped=5.0 2023-05-16 15:32:27,113 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218804.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:32:27,791 INFO [finetune.py:992] (1/2) Epoch 10, batch 8700, loss[loss=0.1655, simple_loss=0.2489, pruned_loss=0.04109, over 12180.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2592, pruned_loss=0.04169, over 2363549.26 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:32:51,432 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218838.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:33:03,201 INFO [finetune.py:992] (1/2) Epoch 10, batch 8750, loss[loss=0.2465, simple_loss=0.3235, pruned_loss=0.0848, over 8016.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2601, pruned_loss=0.04191, over 2363767.91 frames. ], batch size: 97, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:33:07,469 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.025e+02 2.606e+02 3.159e+02 3.755e+02 6.243e+02, threshold=6.317e+02, percent-clipped=2.0 2023-05-16 15:33:10,409 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218865.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:33:11,186 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218866.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:33:25,173 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218886.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:33:38,622 INFO [finetune.py:992] (1/2) Epoch 10, batch 8800, loss[loss=0.1794, simple_loss=0.2721, pruned_loss=0.04328, over 12090.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2606, pruned_loss=0.04194, over 2362420.90 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:33:44,507 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9164, 3.8489, 3.3405, 3.4284, 3.1142, 2.9085, 3.8725, 2.5816], device='cuda:1'), covar=tensor([0.0303, 0.0153, 0.0204, 0.0189, 0.0337, 0.0329, 0.0109, 0.0412], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0160, 0.0155, 0.0180, 0.0198, 0.0196, 0.0166, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:33:45,091 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218914.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:33:56,853 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218930.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:34:15,522 INFO [finetune.py:992] (1/2) Epoch 10, batch 8850, loss[loss=0.1588, simple_loss=0.2444, pruned_loss=0.03658, over 12035.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2588, pruned_loss=0.04107, over 2370945.90 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:34:19,768 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.641e+02 3.077e+02 3.873e+02 7.892e+02, threshold=6.154e+02, percent-clipped=4.0 2023-05-16 15:34:22,885 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:34:36,397 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218984.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:34:51,340 INFO [finetune.py:992] (1/2) Epoch 10, batch 8900, loss[loss=0.1635, simple_loss=0.2611, pruned_loss=0.03292, over 12148.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.259, pruned_loss=0.04087, over 2370637.61 frames. ], batch size: 34, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:35:03,231 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219021.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:35:06,857 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219026.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:35:20,457 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219045.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:35:27,311 INFO [finetune.py:992] (1/2) Epoch 10, batch 8950, loss[loss=0.1701, simple_loss=0.2529, pruned_loss=0.04364, over 11554.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.259, pruned_loss=0.0411, over 2364582.52 frames. ], batch size: 48, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:35:32,436 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.720e+02 3.240e+02 3.849e+02 1.255e+03, threshold=6.481e+02, percent-clipped=2.0 2023-05-16 15:35:40,499 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219072.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:35:47,729 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:36:04,403 INFO [finetune.py:992] (1/2) Epoch 10, batch 9000, loss[loss=0.1353, simple_loss=0.2171, pruned_loss=0.02671, over 12250.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2587, pruned_loss=0.04072, over 2373317.10 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:36:04,403 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 15:36:23,645 INFO [finetune.py:1026] (1/2) Epoch 10, validation: loss=0.33, simple_loss=0.4019, pruned_loss=0.1291, over 1020973.00 frames. 2023-05-16 15:36:23,646 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 15:36:44,032 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219133.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:36:51,075 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6259, 3.2574, 5.0961, 2.6097, 2.7429, 3.7955, 3.1299, 3.8966], device='cuda:1'), covar=tensor([0.0431, 0.1187, 0.0251, 0.1165, 0.1906, 0.1319, 0.1362, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0236, 0.0248, 0.0181, 0.0237, 0.0295, 0.0224, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:36:51,826 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7331, 2.9326, 4.6230, 4.7852, 2.8965, 2.7307, 3.0610, 2.1939], device='cuda:1'), covar=tensor([0.1483, 0.2815, 0.0451, 0.0398, 0.1264, 0.2162, 0.2633, 0.4012], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0382, 0.0272, 0.0299, 0.0268, 0.0301, 0.0375, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:36:59,336 INFO [finetune.py:992] (1/2) Epoch 10, batch 9050, loss[loss=0.261, simple_loss=0.3236, pruned_loss=0.09919, over 8156.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2601, pruned_loss=0.04141, over 2371954.76 frames. ], batch size: 98, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:37:03,549 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.910e+02 3.386e+02 3.882e+02 8.426e+02, threshold=6.771e+02, percent-clipped=1.0 2023-05-16 15:37:06,536 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219165.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:37:13,478 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2522, 5.0194, 5.1440, 5.2105, 4.8537, 4.9259, 4.6360, 5.1170], device='cuda:1'), covar=tensor([0.0629, 0.0676, 0.0850, 0.0589, 0.1769, 0.1316, 0.0643, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0694, 0.0601, 0.0624, 0.0839, 0.0736, 0.0546, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:37:35,169 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219203.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:37:36,356 INFO [finetune.py:992] (1/2) Epoch 10, batch 9100, loss[loss=0.2513, simple_loss=0.3176, pruned_loss=0.09249, over 8037.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2606, pruned_loss=0.0418, over 2362753.51 frames. ], batch size: 98, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:37:42,236 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219213.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:37:54,315 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219230.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:38:11,905 INFO [finetune.py:992] (1/2) Epoch 10, batch 9150, loss[loss=0.1785, simple_loss=0.2722, pruned_loss=0.04236, over 12273.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2603, pruned_loss=0.04136, over 2367785.24 frames. ], batch size: 37, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:38:16,101 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.605e+02 3.021e+02 3.487e+02 6.301e+02, threshold=6.043e+02, percent-clipped=0.0 2023-05-16 15:38:18,443 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219264.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:38:28,246 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219278.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:38:47,341 INFO [finetune.py:992] (1/2) Epoch 10, batch 9200, loss[loss=0.1792, simple_loss=0.27, pruned_loss=0.04417, over 11845.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2594, pruned_loss=0.04086, over 2375014.13 frames. ], batch size: 44, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:38:59,231 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219321.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:39:13,310 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219340.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:39:14,923 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219342.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:39:23,755 INFO [finetune.py:992] (1/2) Epoch 10, batch 9250, loss[loss=0.1853, simple_loss=0.2744, pruned_loss=0.04809, over 12294.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2595, pruned_loss=0.04094, over 2371224.40 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:39:27,994 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.725e+02 3.129e+02 3.820e+02 8.515e+02, threshold=6.257e+02, percent-clipped=3.0 2023-05-16 15:39:36,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8716, 3.6467, 3.6998, 3.8893, 3.5549, 3.9101, 3.8661, 4.0088], device='cuda:1'), covar=tensor([0.0227, 0.0220, 0.0178, 0.0347, 0.0571, 0.0371, 0.0193, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0193, 0.0186, 0.0244, 0.0239, 0.0216, 0.0170, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:39:39,554 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219377.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:39:58,330 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219403.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:39:59,513 INFO [finetune.py:992] (1/2) Epoch 10, batch 9300, loss[loss=0.1543, simple_loss=0.2437, pruned_loss=0.03249, over 12257.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2598, pruned_loss=0.04105, over 2366837.40 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:40:16,076 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219428.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:40:34,930 INFO [finetune.py:992] (1/2) Epoch 10, batch 9350, loss[loss=0.1556, simple_loss=0.2451, pruned_loss=0.03302, over 12263.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2597, pruned_loss=0.04143, over 2355446.72 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:40:39,260 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.892e+02 3.242e+02 3.758e+02 1.096e+03, threshold=6.484e+02, percent-clipped=4.0 2023-05-16 15:40:45,391 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2023-05-16 15:41:01,310 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219490.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:41:11,718 INFO [finetune.py:992] (1/2) Epoch 10, batch 9400, loss[loss=0.1893, simple_loss=0.2825, pruned_loss=0.04812, over 12147.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2592, pruned_loss=0.04074, over 2367885.02 frames. ], batch size: 36, lr: 4.00e-03, grad_scale: 8.0 2023-05-16 15:41:35,369 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0384, 4.6036, 4.1240, 4.2528, 4.7123, 4.1408, 4.2745, 4.1215], device='cuda:1'), covar=tensor([0.1611, 0.1147, 0.1673, 0.1964, 0.1134, 0.2070, 0.1611, 0.1501], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0493, 0.0398, 0.0445, 0.0468, 0.0444, 0.0400, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:41:44,661 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219551.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:41:47,290 INFO [finetune.py:992] (1/2) Epoch 10, batch 9450, loss[loss=0.1833, simple_loss=0.2755, pruned_loss=0.04554, over 12048.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2588, pruned_loss=0.04058, over 2368908.37 frames. ], batch size: 42, lr: 4.00e-03, grad_scale: 16.0 2023-05-16 15:41:50,206 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219559.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:41:51,496 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.827e+02 2.762e+02 3.070e+02 3.810e+02 6.920e+02, threshold=6.141e+02, percent-clipped=1.0 2023-05-16 15:41:58,759 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0661, 5.7215, 5.3568, 5.3442, 5.9002, 5.1896, 5.3343, 5.2588], device='cuda:1'), covar=tensor([0.1495, 0.1164, 0.0897, 0.1822, 0.0932, 0.2061, 0.2025, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0491, 0.0396, 0.0442, 0.0466, 0.0441, 0.0398, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:42:11,742 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4321, 2.5565, 3.5348, 4.3155, 3.8128, 4.2617, 3.6874, 2.9667], device='cuda:1'), covar=tensor([0.0040, 0.0396, 0.0154, 0.0040, 0.0121, 0.0102, 0.0133, 0.0386], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0126, 0.0109, 0.0078, 0.0105, 0.0118, 0.0098, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:42:23,219 INFO [finetune.py:992] (1/2) Epoch 10, batch 9500, loss[loss=0.1784, simple_loss=0.2707, pruned_loss=0.04304, over 12095.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2599, pruned_loss=0.04098, over 2361219.68 frames. ], batch size: 42, lr: 4.00e-03, grad_scale: 16.0 2023-05-16 15:42:34,793 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:42:49,566 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219640.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:43:00,008 INFO [finetune.py:992] (1/2) Epoch 10, batch 9550, loss[loss=0.1767, simple_loss=0.2714, pruned_loss=0.04094, over 12348.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2594, pruned_loss=0.04054, over 2362827.76 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 16.0 2023-05-16 15:43:04,261 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.626e+02 3.089e+02 3.885e+02 9.046e+02, threshold=6.179e+02, percent-clipped=3.0 2023-05-16 15:43:10,066 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219669.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:43:15,697 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219677.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:43:18,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 15:43:23,602 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:43:30,566 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219698.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 15:43:35,476 INFO [finetune.py:992] (1/2) Epoch 10, batch 9600, loss[loss=0.1681, simple_loss=0.2601, pruned_loss=0.03803, over 12179.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2597, pruned_loss=0.04068, over 2366620.46 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 16.0 2023-05-16 15:43:49,973 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219725.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:43:52,128 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219728.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:43:56,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 15:44:01,069 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3854, 6.2211, 5.7021, 5.6843, 6.2718, 5.5659, 5.8509, 5.7013], device='cuda:1'), covar=tensor([0.1500, 0.0901, 0.1117, 0.1938, 0.0900, 0.2034, 0.1578, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0490, 0.0395, 0.0444, 0.0465, 0.0441, 0.0398, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:44:08,287 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3301, 2.6878, 3.9265, 3.3365, 3.6504, 3.4610, 2.9287, 3.8326], device='cuda:1'), covar=tensor([0.0137, 0.0322, 0.0130, 0.0223, 0.0173, 0.0154, 0.0284, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0206, 0.0189, 0.0186, 0.0217, 0.0162, 0.0198, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:44:10,822 INFO [finetune.py:992] (1/2) Epoch 10, batch 9650, loss[loss=0.1981, simple_loss=0.2868, pruned_loss=0.05473, over 12352.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2595, pruned_loss=0.04097, over 2360251.16 frames. ], batch size: 35, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:44:15,603 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.169e+02 2.820e+02 3.328e+02 3.898e+02 8.382e+02, threshold=6.656e+02, percent-clipped=4.0 2023-05-16 15:44:17,280 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219763.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:44:27,241 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219776.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:44:47,910 INFO [finetune.py:992] (1/2) Epoch 10, batch 9700, loss[loss=0.196, simple_loss=0.2912, pruned_loss=0.05038, over 12132.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2589, pruned_loss=0.04079, over 2368582.68 frames. ], batch size: 38, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:44:58,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1248, 2.4776, 3.6806, 3.0822, 3.4394, 3.2238, 2.6449, 3.5632], device='cuda:1'), covar=tensor([0.0138, 0.0375, 0.0163, 0.0255, 0.0122, 0.0184, 0.0356, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0206, 0.0189, 0.0187, 0.0217, 0.0163, 0.0198, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:45:01,474 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219824.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:45:15,795 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4644, 5.2600, 5.4400, 5.4098, 5.0958, 5.0698, 4.8709, 5.3685], device='cuda:1'), covar=tensor([0.0604, 0.0571, 0.0627, 0.0557, 0.1510, 0.1312, 0.0583, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0694, 0.0597, 0.0625, 0.0836, 0.0735, 0.0544, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:45:17,126 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219846.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:45:20,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 15:45:23,393 INFO [finetune.py:992] (1/2) Epoch 10, batch 9750, loss[loss=0.2055, simple_loss=0.294, pruned_loss=0.05847, over 12102.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2574, pruned_loss=0.04013, over 2374075.01 frames. ], batch size: 33, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:45:24,953 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219857.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:45:26,193 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219859.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:45:27,480 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 2.682e+02 3.236e+02 3.689e+02 1.131e+03, threshold=6.473e+02, percent-clipped=3.0 2023-05-16 15:45:28,984 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0505, 6.0630, 5.7528, 5.2758, 5.1592, 5.9065, 5.5601, 5.2478], device='cuda:1'), covar=tensor([0.0690, 0.0758, 0.0638, 0.1527, 0.0675, 0.0757, 0.1517, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0613, 0.0557, 0.0516, 0.0630, 0.0412, 0.0716, 0.0773, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 15:45:40,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 15:45:58,396 INFO [finetune.py:992] (1/2) Epoch 10, batch 9800, loss[loss=0.1581, simple_loss=0.2517, pruned_loss=0.03221, over 12125.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2588, pruned_loss=0.04062, over 2373447.08 frames. ], batch size: 39, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:46:00,523 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219907.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:46:09,182 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219918.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:46:09,906 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4278, 2.3933, 3.0760, 4.3432, 2.1439, 4.3722, 4.4952, 4.5907], device='cuda:1'), covar=tensor([0.0137, 0.1346, 0.0528, 0.0167, 0.1501, 0.0208, 0.0183, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0201, 0.0184, 0.0116, 0.0187, 0.0179, 0.0174, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:46:35,399 INFO [finetune.py:992] (1/2) Epoch 10, batch 9850, loss[loss=0.172, simple_loss=0.262, pruned_loss=0.04099, over 12291.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04014, over 2372493.15 frames. ], batch size: 33, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:46:39,493 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.736e+02 3.238e+02 3.957e+02 6.493e+02, threshold=6.477e+02, percent-clipped=2.0 2023-05-16 15:46:40,468 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0119, 4.3224, 3.8284, 4.6022, 4.1511, 2.7133, 4.0244, 2.8167], device='cuda:1'), covar=tensor([0.0870, 0.0912, 0.1417, 0.0508, 0.1149, 0.1701, 0.1036, 0.3463], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0374, 0.0351, 0.0297, 0.0361, 0.0263, 0.0341, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:47:05,889 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219998.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:47:13,718 INFO [finetune.py:992] (1/2) Epoch 10, batch 9900, loss[loss=0.1467, simple_loss=0.2255, pruned_loss=0.0339, over 12188.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2586, pruned_loss=0.04049, over 2372987.56 frames. ], batch size: 29, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:47:17,008 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 15:47:42,738 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220046.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 15:47:49,601 INFO [finetune.py:992] (1/2) Epoch 10, batch 9950, loss[loss=0.1797, simple_loss=0.2661, pruned_loss=0.04668, over 12037.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2591, pruned_loss=0.04048, over 2375059.67 frames. ], batch size: 42, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:47:54,578 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.719e+02 3.158e+02 3.980e+02 6.647e+02, threshold=6.316e+02, percent-clipped=1.0 2023-05-16 15:48:13,059 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220087.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:48:21,044 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220098.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:48:25,767 INFO [finetune.py:992] (1/2) Epoch 10, batch 10000, loss[loss=0.1805, simple_loss=0.2645, pruned_loss=0.04827, over 11229.00 frames. ], tot_loss[loss=0.17, simple_loss=0.259, pruned_loss=0.04048, over 2373682.00 frames. ], batch size: 55, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:48:35,852 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220119.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:48:39,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 2023-05-16 15:48:46,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 15:48:55,062 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220146.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:48:56,428 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220148.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:48:59,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2023-05-16 15:49:01,135 INFO [finetune.py:992] (1/2) Epoch 10, batch 10050, loss[loss=0.1543, simple_loss=0.24, pruned_loss=0.0343, over 12338.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.259, pruned_loss=0.04043, over 2375596.89 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:49:04,121 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220159.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:49:04,845 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7943, 3.4170, 5.1957, 2.6629, 2.8006, 3.8963, 3.2830, 3.9086], device='cuda:1'), covar=tensor([0.0445, 0.1098, 0.0268, 0.1242, 0.1976, 0.1447, 0.1345, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0238, 0.0250, 0.0183, 0.0238, 0.0297, 0.0226, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:49:05,274 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.261e+02 2.862e+02 3.244e+02 4.147e+02 1.035e+03, threshold=6.489e+02, percent-clipped=3.0 2023-05-16 15:49:14,527 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0787, 6.0538, 5.8819, 5.1410, 5.2963, 5.9600, 5.5324, 5.3357], device='cuda:1'), covar=tensor([0.0706, 0.0888, 0.0586, 0.1691, 0.0583, 0.0779, 0.1567, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0605, 0.0555, 0.0509, 0.0623, 0.0408, 0.0713, 0.0767, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 15:49:19,671 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2218, 4.5114, 4.0438, 4.8362, 4.5082, 2.8078, 4.2406, 3.2041], device='cuda:1'), covar=tensor([0.0859, 0.0870, 0.1442, 0.0522, 0.1056, 0.1646, 0.0982, 0.3186], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0373, 0.0351, 0.0299, 0.0362, 0.0263, 0.0342, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:49:29,542 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220194.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:49:37,553 INFO [finetune.py:992] (1/2) Epoch 10, batch 10100, loss[loss=0.1832, simple_loss=0.2757, pruned_loss=0.04535, over 10487.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2596, pruned_loss=0.0407, over 2370326.61 frames. ], batch size: 68, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:49:43,901 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220213.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:49:55,609 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220229.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:50:04,076 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1544, 6.0820, 5.9841, 5.4569, 5.3067, 6.0754, 5.6489, 5.4911], device='cuda:1'), covar=tensor([0.0663, 0.0960, 0.0556, 0.1326, 0.0590, 0.0674, 0.1381, 0.0939], device='cuda:1'), in_proj_covar=tensor([0.0605, 0.0555, 0.0509, 0.0623, 0.0408, 0.0711, 0.0767, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 15:50:13,920 INFO [finetune.py:992] (1/2) Epoch 10, batch 10150, loss[loss=0.1761, simple_loss=0.271, pruned_loss=0.04058, over 11297.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2597, pruned_loss=0.041, over 2363687.14 frames. ], batch size: 55, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:50:16,236 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220258.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:50:18,222 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.689e+02 3.151e+02 3.706e+02 1.036e+03, threshold=6.302e+02, percent-clipped=2.0 2023-05-16 15:50:39,187 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220290.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:50:49,863 INFO [finetune.py:992] (1/2) Epoch 10, batch 10200, loss[loss=0.1722, simple_loss=0.2592, pruned_loss=0.04258, over 12038.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2588, pruned_loss=0.04066, over 2366037.46 frames. ], batch size: 45, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:50:59,986 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220319.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:51:02,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 15:51:24,943 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6849, 3.8202, 3.3545, 3.3500, 3.1622, 2.9431, 3.8330, 2.4516], device='cuda:1'), covar=tensor([0.0361, 0.0089, 0.0194, 0.0181, 0.0321, 0.0332, 0.0102, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0163, 0.0160, 0.0184, 0.0203, 0.0201, 0.0170, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:51:26,759 INFO [finetune.py:992] (1/2) Epoch 10, batch 10250, loss[loss=0.1638, simple_loss=0.2645, pruned_loss=0.0316, over 12096.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04017, over 2376557.12 frames. ], batch size: 33, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:51:31,014 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.634e+02 2.920e+02 3.635e+02 1.121e+03, threshold=5.839e+02, percent-clipped=1.0 2023-05-16 15:51:36,158 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4731, 2.5207, 3.6336, 4.3243, 3.8416, 4.3559, 3.7028, 3.0836], device='cuda:1'), covar=tensor([0.0037, 0.0383, 0.0157, 0.0041, 0.0116, 0.0068, 0.0132, 0.0328], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0123, 0.0106, 0.0077, 0.0102, 0.0115, 0.0095, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:51:40,261 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3399, 5.1597, 5.3002, 5.3166, 4.9257, 4.9900, 4.7352, 5.2530], device='cuda:1'), covar=tensor([0.0720, 0.0591, 0.0813, 0.0567, 0.1821, 0.1211, 0.0657, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0689, 0.0594, 0.0619, 0.0833, 0.0729, 0.0541, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:52:02,288 INFO [finetune.py:992] (1/2) Epoch 10, batch 10300, loss[loss=0.1797, simple_loss=0.2691, pruned_loss=0.04512, over 12158.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2585, pruned_loss=0.04053, over 2369894.16 frames. ], batch size: 36, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:52:03,188 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9733, 3.4913, 3.6511, 4.1765, 2.9031, 3.4440, 2.2450, 3.5542], device='cuda:1'), covar=tensor([0.1756, 0.0970, 0.1115, 0.0632, 0.1271, 0.0814, 0.2212, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0265, 0.0298, 0.0355, 0.0237, 0.0240, 0.0262, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:52:12,230 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220419.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:52:29,312 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220443.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:52:29,387 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5119, 5.3633, 5.4901, 5.4711, 5.1185, 5.1761, 4.9595, 5.4507], device='cuda:1'), covar=tensor([0.0707, 0.0546, 0.0794, 0.0587, 0.1871, 0.1224, 0.0595, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0690, 0.0597, 0.0620, 0.0835, 0.0729, 0.0541, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 15:52:37,229 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220454.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:52:37,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-05-16 15:52:37,776 INFO [finetune.py:992] (1/2) Epoch 10, batch 10350, loss[loss=0.1772, simple_loss=0.2604, pruned_loss=0.04706, over 11568.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2583, pruned_loss=0.04063, over 2374579.34 frames. ], batch size: 48, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:52:38,811 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1163, 4.4946, 3.9033, 4.7520, 4.4077, 2.8942, 4.0535, 2.9184], device='cuda:1'), covar=tensor([0.0836, 0.0779, 0.1621, 0.0497, 0.1059, 0.1518, 0.1048, 0.3193], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0372, 0.0350, 0.0297, 0.0360, 0.0262, 0.0341, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:52:41,954 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.058e+02 2.853e+02 3.399e+02 3.948e+02 7.256e+02, threshold=6.798e+02, percent-clipped=6.0 2023-05-16 15:52:46,218 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220467.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:52:48,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-16 15:52:50,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 15:53:14,399 INFO [finetune.py:992] (1/2) Epoch 10, batch 10400, loss[loss=0.1766, simple_loss=0.2702, pruned_loss=0.04151, over 12118.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.258, pruned_loss=0.04027, over 2374971.94 frames. ], batch size: 38, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:53:16,764 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6054, 2.6608, 3.2463, 4.5029, 2.3806, 4.4793, 4.5970, 4.7226], device='cuda:1'), covar=tensor([0.0121, 0.1235, 0.0479, 0.0182, 0.1393, 0.0232, 0.0142, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0202, 0.0184, 0.0116, 0.0188, 0.0179, 0.0175, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:53:20,249 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220513.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:53:31,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2023-05-16 15:53:50,009 INFO [finetune.py:992] (1/2) Epoch 10, batch 10450, loss[loss=0.1976, simple_loss=0.2804, pruned_loss=0.05745, over 12068.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.258, pruned_loss=0.04022, over 2374332.95 frames. ], batch size: 42, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:53:50,177 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220555.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:53:54,259 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 2.943e+02 3.296e+02 3.873e+02 5.866e+02, threshold=6.592e+02, percent-clipped=0.0 2023-05-16 15:53:54,351 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220561.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:54:11,536 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220585.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:54:23,252 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8577, 5.8153, 5.6411, 5.0803, 5.1073, 5.7479, 5.3414, 5.0862], device='cuda:1'), covar=tensor([0.0678, 0.0904, 0.0600, 0.1473, 0.0674, 0.0758, 0.1516, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0613, 0.0558, 0.0513, 0.0628, 0.0410, 0.0717, 0.0774, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 15:54:25,955 INFO [finetune.py:992] (1/2) Epoch 10, batch 10500, loss[loss=0.1678, simple_loss=0.2474, pruned_loss=0.04408, over 12355.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2581, pruned_loss=0.04046, over 2378588.68 frames. ], batch size: 30, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:54:31,884 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3162, 3.4416, 3.2109, 3.1663, 2.8889, 2.6733, 3.5141, 2.2134], device='cuda:1'), covar=tensor([0.0427, 0.0176, 0.0205, 0.0205, 0.0405, 0.0434, 0.0134, 0.0510], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0166, 0.0162, 0.0187, 0.0206, 0.0206, 0.0173, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:54:32,445 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220614.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:54:34,071 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220616.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 15:54:36,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 2023-05-16 15:55:00,057 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220651.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:55:02,704 INFO [finetune.py:992] (1/2) Epoch 10, batch 10550, loss[loss=0.1757, simple_loss=0.2649, pruned_loss=0.04325, over 12343.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2585, pruned_loss=0.04, over 2378893.30 frames. ], batch size: 36, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:55:06,783 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.732e+02 3.114e+02 3.768e+02 7.868e+02, threshold=6.228e+02, percent-clipped=1.0 2023-05-16 15:55:06,985 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0679, 5.0107, 4.9014, 4.9712, 4.5277, 5.1261, 5.0476, 5.2542], device='cuda:1'), covar=tensor([0.0256, 0.0138, 0.0169, 0.0288, 0.0782, 0.0279, 0.0149, 0.0179], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0192, 0.0186, 0.0244, 0.0239, 0.0216, 0.0170, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 15:55:34,767 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7753, 2.4527, 3.5830, 3.7530, 2.8943, 2.6900, 2.6730, 2.3652], device='cuda:1'), covar=tensor([0.1219, 0.2589, 0.0602, 0.0439, 0.0905, 0.1884, 0.2337, 0.3362], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0380, 0.0271, 0.0297, 0.0266, 0.0299, 0.0372, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:55:37,819 INFO [finetune.py:992] (1/2) Epoch 10, batch 10600, loss[loss=0.1787, simple_loss=0.2644, pruned_loss=0.04653, over 12362.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2589, pruned_loss=0.04028, over 2384395.25 frames. ], batch size: 38, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:55:43,123 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220712.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:56:04,975 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220743.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:56:12,726 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220754.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:56:13,261 INFO [finetune.py:992] (1/2) Epoch 10, batch 10650, loss[loss=0.24, simple_loss=0.3154, pruned_loss=0.0823, over 7727.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2584, pruned_loss=0.0402, over 2381566.84 frames. ], batch size: 98, lr: 3.99e-03, grad_scale: 16.0 2023-05-16 15:56:17,465 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.734e+02 3.095e+02 3.683e+02 6.434e+02, threshold=6.190e+02, percent-clipped=1.0 2023-05-16 15:56:37,091 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.02 vs. limit=5.0 2023-05-16 15:56:39,634 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220791.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:56:48,420 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220802.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:56:50,486 INFO [finetune.py:992] (1/2) Epoch 10, batch 10700, loss[loss=0.1632, simple_loss=0.2519, pruned_loss=0.03726, over 12294.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2586, pruned_loss=0.04015, over 2377756.94 frames. ], batch size: 34, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 15:57:02,816 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220822.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:57:26,106 INFO [finetune.py:992] (1/2) Epoch 10, batch 10750, loss[loss=0.1818, simple_loss=0.273, pruned_loss=0.04532, over 12307.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2579, pruned_loss=0.03998, over 2375661.80 frames. ], batch size: 34, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 15:57:30,325 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.819e+02 3.336e+02 4.154e+02 5.728e+02, threshold=6.671e+02, percent-clipped=0.0 2023-05-16 15:57:46,241 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220883.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:57:47,606 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220885.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:58:00,819 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220904.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:58:01,330 INFO [finetune.py:992] (1/2) Epoch 10, batch 10800, loss[loss=0.19, simple_loss=0.2734, pruned_loss=0.05336, over 12278.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2587, pruned_loss=0.04051, over 2375695.52 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 15:58:05,623 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220911.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 15:58:07,623 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220914.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:58:14,141 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5564, 2.7870, 4.4235, 4.5588, 2.9233, 2.5500, 2.8284, 2.1566], device='cuda:1'), covar=tensor([0.1583, 0.2912, 0.0496, 0.0430, 0.1204, 0.2358, 0.2691, 0.3790], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0378, 0.0269, 0.0296, 0.0264, 0.0296, 0.0370, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 15:58:21,682 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220933.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:58:38,199 INFO [finetune.py:992] (1/2) Epoch 10, batch 10850, loss[loss=0.1461, simple_loss=0.2315, pruned_loss=0.03034, over 12344.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2578, pruned_loss=0.04032, over 2369635.08 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 15:58:42,548 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 2.730e+02 3.365e+02 4.016e+02 7.164e+02, threshold=6.730e+02, percent-clipped=2.0 2023-05-16 15:58:43,403 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220962.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:58:45,626 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:58:59,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 15:59:09,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 15:59:14,783 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0215, 3.5533, 5.3086, 2.7494, 2.9113, 4.1091, 3.3877, 4.1145], device='cuda:1'), covar=tensor([0.0471, 0.1064, 0.0337, 0.1247, 0.1907, 0.1387, 0.1292, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0237, 0.0250, 0.0183, 0.0237, 0.0296, 0.0226, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 15:59:15,177 INFO [finetune.py:992] (1/2) Epoch 10, batch 10900, loss[loss=0.1632, simple_loss=0.2444, pruned_loss=0.04105, over 11986.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2588, pruned_loss=0.04054, over 2375485.41 frames. ], batch size: 28, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 15:59:16,762 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221007.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 15:59:35,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.84 vs. limit=5.0 2023-05-16 15:59:44,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-05-16 15:59:50,674 INFO [finetune.py:992] (1/2) Epoch 10, batch 10950, loss[loss=0.1502, simple_loss=0.2319, pruned_loss=0.03427, over 12251.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2597, pruned_loss=0.04129, over 2369294.46 frames. ], batch size: 28, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 15:59:54,857 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.682e+02 3.296e+02 4.099e+02 7.687e+02, threshold=6.592e+02, percent-clipped=1.0 2023-05-16 16:00:27,504 INFO [finetune.py:992] (1/2) Epoch 10, batch 11000, loss[loss=0.1726, simple_loss=0.269, pruned_loss=0.03808, over 12311.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2631, pruned_loss=0.04333, over 2334840.30 frames. ], batch size: 34, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:00:41,716 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5803, 4.5385, 4.4899, 4.0287, 4.2369, 4.5551, 4.2415, 4.0872], device='cuda:1'), covar=tensor([0.0831, 0.0967, 0.0669, 0.1427, 0.1839, 0.0835, 0.1398, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0560, 0.0518, 0.0635, 0.0411, 0.0720, 0.0778, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 16:01:02,728 INFO [finetune.py:992] (1/2) Epoch 10, batch 11050, loss[loss=0.1856, simple_loss=0.2916, pruned_loss=0.03974, over 11141.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2665, pruned_loss=0.04539, over 2288446.39 frames. ], batch size: 55, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:01:07,037 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 3.083e+02 3.628e+02 4.356e+02 1.327e+03, threshold=7.256e+02, percent-clipped=10.0 2023-05-16 16:01:19,865 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221178.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:01:39,603 INFO [finetune.py:992] (1/2) Epoch 10, batch 11100, loss[loss=0.2263, simple_loss=0.3071, pruned_loss=0.07272, over 7867.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2691, pruned_loss=0.04704, over 2246065.19 frames. ], batch size: 97, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:01:43,978 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:02:14,506 INFO [finetune.py:992] (1/2) Epoch 10, batch 11150, loss[loss=0.2407, simple_loss=0.335, pruned_loss=0.07322, over 11761.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2747, pruned_loss=0.05045, over 2198034.00 frames. ], batch size: 44, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:02:18,179 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221259.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:02:18,950 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221260.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:02:19,479 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.155e+02 3.250e+02 3.791e+02 4.612e+02 9.960e+02, threshold=7.583e+02, percent-clipped=2.0 2023-05-16 16:02:21,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.40 vs. limit=5.0 2023-05-16 16:02:32,596 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3675, 2.8659, 4.0854, 3.4054, 3.7779, 3.6675, 2.8889, 3.9536], device='cuda:1'), covar=tensor([0.0128, 0.0295, 0.0106, 0.0198, 0.0143, 0.0127, 0.0297, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0205, 0.0187, 0.0186, 0.0216, 0.0162, 0.0196, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:02:50,811 INFO [finetune.py:992] (1/2) Epoch 10, batch 11200, loss[loss=0.21, simple_loss=0.3051, pruned_loss=0.05745, over 11608.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2805, pruned_loss=0.05436, over 2140677.05 frames. ], batch size: 48, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:02:52,299 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221307.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:02:53,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 16:03:26,529 INFO [finetune.py:992] (1/2) Epoch 10, batch 11250, loss[loss=0.1587, simple_loss=0.2573, pruned_loss=0.03003, over 12268.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2875, pruned_loss=0.05897, over 2086434.30 frames. ], batch size: 32, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:03:26,604 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221355.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:03:30,559 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.422e+02 3.396e+02 4.025e+02 5.294e+02 9.338e+02, threshold=8.049e+02, percent-clipped=5.0 2023-05-16 16:03:45,913 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6970, 2.1795, 2.9039, 3.7056, 2.1730, 3.8340, 3.8081, 3.8285], device='cuda:1'), covar=tensor([0.0154, 0.1325, 0.0473, 0.0145, 0.1293, 0.0199, 0.0203, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0197, 0.0180, 0.0114, 0.0183, 0.0175, 0.0171, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:03:59,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-05-16 16:04:00,871 INFO [finetune.py:992] (1/2) Epoch 10, batch 11300, loss[loss=0.2076, simple_loss=0.3025, pruned_loss=0.05637, over 11077.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2944, pruned_loss=0.06371, over 2006147.59 frames. ], batch size: 55, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:04:18,279 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6702, 4.5314, 4.5192, 4.5772, 4.1202, 4.7386, 4.6827, 4.7808], device='cuda:1'), covar=tensor([0.0198, 0.0165, 0.0190, 0.0331, 0.0794, 0.0259, 0.0135, 0.0202], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0183, 0.0177, 0.0232, 0.0226, 0.0203, 0.0161, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 16:04:36,478 INFO [finetune.py:992] (1/2) Epoch 10, batch 11350, loss[loss=0.2108, simple_loss=0.298, pruned_loss=0.06181, over 11008.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2985, pruned_loss=0.06618, over 1976403.08 frames. ], batch size: 55, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:04:40,614 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.540e+02 3.534e+02 4.230e+02 4.906e+02 1.074e+03, threshold=8.461e+02, percent-clipped=2.0 2023-05-16 16:04:52,811 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221478.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:05:11,506 INFO [finetune.py:992] (1/2) Epoch 10, batch 11400, loss[loss=0.1966, simple_loss=0.2836, pruned_loss=0.05485, over 12353.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.3034, pruned_loss=0.06956, over 1919506.35 frames. ], batch size: 36, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:05:12,987 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221507.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:05:15,982 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 16:05:24,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 2023-05-16 16:05:25,928 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221526.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:05:46,554 INFO [finetune.py:992] (1/2) Epoch 10, batch 11450, loss[loss=0.2221, simple_loss=0.3092, pruned_loss=0.06748, over 10398.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3069, pruned_loss=0.07294, over 1881451.28 frames. ], batch size: 68, lr: 3.98e-03, grad_scale: 32.0 2023-05-16 16:05:48,180 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8899, 2.2274, 2.7102, 2.8581, 2.9916, 2.9776, 2.8959, 2.3739], device='cuda:1'), covar=tensor([0.0071, 0.0325, 0.0198, 0.0088, 0.0101, 0.0101, 0.0116, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0122, 0.0105, 0.0076, 0.0100, 0.0114, 0.0094, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:05:50,169 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221560.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:05:50,665 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.263e+02 3.595e+02 4.162e+02 4.999e+02 1.118e+03, threshold=8.323e+02, percent-clipped=4.0 2023-05-16 16:05:53,002 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6006, 5.5309, 5.3396, 4.9185, 4.8353, 5.4971, 5.1705, 4.9962], device='cuda:1'), covar=tensor([0.0577, 0.0870, 0.0640, 0.1368, 0.0917, 0.0630, 0.1307, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0589, 0.0535, 0.0495, 0.0602, 0.0396, 0.0686, 0.0734, 0.0547], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:05:55,844 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221568.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:06:04,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 16:06:11,867 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4637, 3.1968, 3.4977, 3.5031, 3.5414, 3.6726, 3.3813, 2.6504], device='cuda:1'), covar=tensor([0.0089, 0.0127, 0.0137, 0.0079, 0.0062, 0.0112, 0.0078, 0.0745], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0077, 0.0081, 0.0072, 0.0059, 0.0090, 0.0079, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:06:21,253 INFO [finetune.py:992] (1/2) Epoch 10, batch 11500, loss[loss=0.2141, simple_loss=0.298, pruned_loss=0.06511, over 10516.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3091, pruned_loss=0.07424, over 1860730.04 frames. ], batch size: 69, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:06:24,008 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221608.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:06:40,861 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3657, 3.9880, 4.2316, 4.2922, 4.1971, 4.3525, 4.2794, 2.4895], device='cuda:1'), covar=tensor([0.0095, 0.0099, 0.0101, 0.0070, 0.0054, 0.0112, 0.0071, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0081, 0.0072, 0.0059, 0.0089, 0.0079, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:06:41,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-05-16 16:06:55,868 INFO [finetune.py:992] (1/2) Epoch 10, batch 11550, loss[loss=0.2185, simple_loss=0.293, pruned_loss=0.07199, over 10273.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3105, pruned_loss=0.07582, over 1818045.39 frames. ], batch size: 68, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:07:00,389 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.705e+02 3.400e+02 4.212e+02 4.832e+02 8.048e+02, threshold=8.424e+02, percent-clipped=0.0 2023-05-16 16:07:05,334 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7242, 2.6995, 3.9748, 4.1478, 2.9234, 2.7134, 2.7596, 2.1043], device='cuda:1'), covar=tensor([0.1381, 0.2496, 0.0495, 0.0441, 0.1109, 0.2120, 0.2609, 0.4185], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0375, 0.0266, 0.0292, 0.0260, 0.0294, 0.0369, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:07:30,307 INFO [finetune.py:992] (1/2) Epoch 10, batch 11600, loss[loss=0.2216, simple_loss=0.3073, pruned_loss=0.06796, over 10373.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3121, pruned_loss=0.07727, over 1802856.56 frames. ], batch size: 68, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:07:38,259 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0509, 3.2271, 4.8042, 5.1525, 3.1262, 3.0053, 3.1844, 2.2645], device='cuda:1'), covar=tensor([0.1423, 0.2706, 0.0390, 0.0292, 0.1245, 0.2084, 0.2556, 0.4397], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0374, 0.0265, 0.0291, 0.0259, 0.0293, 0.0368, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:07:51,202 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-05-16 16:08:01,351 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1161, 4.5122, 3.8803, 4.8262, 4.3426, 2.8714, 4.1945, 3.0068], device='cuda:1'), covar=tensor([0.0910, 0.0770, 0.1499, 0.0417, 0.1443, 0.1833, 0.1121, 0.3406], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0357, 0.0337, 0.0281, 0.0348, 0.0256, 0.0327, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:08:07,791 INFO [finetune.py:992] (1/2) Epoch 10, batch 11650, loss[loss=0.1844, simple_loss=0.2753, pruned_loss=0.04677, over 11070.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3116, pruned_loss=0.07742, over 1792037.29 frames. ], batch size: 55, lr: 3.98e-03, grad_scale: 16.0 2023-05-16 16:08:10,164 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221758.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:08:12,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.437e+02 3.386e+02 3.820e+02 4.524e+02 6.802e+02, threshold=7.640e+02, percent-clipped=0.0 2023-05-16 16:08:43,806 INFO [finetune.py:992] (1/2) Epoch 10, batch 11700, loss[loss=0.222, simple_loss=0.3032, pruned_loss=0.07042, over 11105.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3115, pruned_loss=0.07769, over 1783709.54 frames. ], batch size: 55, lr: 3.97e-03, grad_scale: 16.0 2023-05-16 16:08:53,165 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221819.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:09:12,175 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4987, 4.4835, 4.3513, 3.9544, 4.1346, 4.4660, 4.1995, 4.0335], device='cuda:1'), covar=tensor([0.0879, 0.0971, 0.0685, 0.1338, 0.1787, 0.0814, 0.1463, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0587, 0.0532, 0.0491, 0.0599, 0.0395, 0.0677, 0.0727, 0.0542], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:09:17,510 INFO [finetune.py:992] (1/2) Epoch 10, batch 11750, loss[loss=0.2039, simple_loss=0.2835, pruned_loss=0.0621, over 11669.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.311, pruned_loss=0.07783, over 1781619.01 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 16.0 2023-05-16 16:09:22,864 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.552e+02 3.467e+02 3.945e+02 4.631e+02 8.258e+02, threshold=7.889e+02, percent-clipped=2.0 2023-05-16 16:09:23,688 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221863.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:09:29,284 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4358, 3.0903, 3.0703, 3.3879, 2.6505, 3.1185, 2.5836, 2.7778], device='cuda:1'), covar=tensor([0.1520, 0.0949, 0.0871, 0.0609, 0.0974, 0.0808, 0.1692, 0.0622], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0261, 0.0291, 0.0347, 0.0232, 0.0236, 0.0256, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:09:41,852 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221890.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:09:51,662 INFO [finetune.py:992] (1/2) Epoch 10, batch 11800, loss[loss=0.2096, simple_loss=0.3057, pruned_loss=0.0567, over 10304.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3132, pruned_loss=0.08009, over 1745945.13 frames. ], batch size: 68, lr: 3.97e-03, grad_scale: 16.0 2023-05-16 16:10:23,967 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221951.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:10:26,408 INFO [finetune.py:992] (1/2) Epoch 10, batch 11850, loss[loss=0.2565, simple_loss=0.3308, pruned_loss=0.09108, over 6466.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3159, pruned_loss=0.08172, over 1718206.28 frames. ], batch size: 104, lr: 3.97e-03, grad_scale: 16.0 2023-05-16 16:10:31,134 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.758e+02 3.642e+02 4.323e+02 4.856e+02 1.009e+03, threshold=8.645e+02, percent-clipped=1.0 2023-05-16 16:10:47,892 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4665, 4.4387, 4.3586, 4.0034, 4.1239, 4.4442, 4.2128, 4.0399], device='cuda:1'), covar=tensor([0.0834, 0.0989, 0.0650, 0.1256, 0.1898, 0.0794, 0.1322, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0527, 0.0488, 0.0594, 0.0391, 0.0673, 0.0719, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 16:11:05,064 INFO [finetune.py:992] (1/2) Epoch 10, batch 11900, loss[loss=0.1911, simple_loss=0.2837, pruned_loss=0.04928, over 12133.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3162, pruned_loss=0.08122, over 1692605.12 frames. ], batch size: 34, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:11:07,918 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222009.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:11:20,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 16:11:39,923 INFO [finetune.py:992] (1/2) Epoch 10, batch 11950, loss[loss=0.1991, simple_loss=0.2805, pruned_loss=0.05884, over 6499.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3132, pruned_loss=0.07888, over 1676076.07 frames. ], batch size: 98, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:11:45,241 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.111e+02 3.118e+02 3.605e+02 4.252e+02 6.600e+02, threshold=7.210e+02, percent-clipped=0.0 2023-05-16 16:11:50,233 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222070.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:12:14,982 INFO [finetune.py:992] (1/2) Epoch 10, batch 12000, loss[loss=0.1979, simple_loss=0.2798, pruned_loss=0.05798, over 7014.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3076, pruned_loss=0.07452, over 1686699.02 frames. ], batch size: 98, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:12:14,982 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 16:12:33,169 INFO [finetune.py:1026] (1/2) Epoch 10, validation: loss=0.2883, simple_loss=0.3634, pruned_loss=0.1066, over 1020973.00 frames. 2023-05-16 16:12:33,170 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 16:12:39,353 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222114.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:13:08,135 INFO [finetune.py:992] (1/2) Epoch 10, batch 12050, loss[loss=0.2363, simple_loss=0.3062, pruned_loss=0.08316, over 6798.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3036, pruned_loss=0.07175, over 1690740.13 frames. ], batch size: 98, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:13:13,303 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.833e+02 3.308e+02 3.914e+02 7.827e+02, threshold=6.617e+02, percent-clipped=2.0 2023-05-16 16:13:13,455 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222163.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:13:26,596 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-05-16 16:13:27,493 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222185.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:13:40,360 INFO [finetune.py:992] (1/2) Epoch 10, batch 12100, loss[loss=0.2098, simple_loss=0.2859, pruned_loss=0.06683, over 6963.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3021, pruned_loss=0.07072, over 1680437.22 frames. ], batch size: 98, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:13:40,553 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222205.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:13:44,259 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:14:07,631 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222246.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:14:07,724 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222246.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 16:14:13,166 INFO [finetune.py:992] (1/2) Epoch 10, batch 12150, loss[loss=0.2372, simple_loss=0.3078, pruned_loss=0.08332, over 6714.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3028, pruned_loss=0.07094, over 1676720.84 frames. ], batch size: 98, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:14:18,161 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.270e+02 3.038e+02 3.502e+02 4.308e+02 1.208e+03, threshold=7.004e+02, percent-clipped=4.0 2023-05-16 16:14:20,249 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222266.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:14:45,034 INFO [finetune.py:992] (1/2) Epoch 10, batch 12200, loss[loss=0.248, simple_loss=0.3167, pruned_loss=0.08967, over 6788.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.304, pruned_loss=0.07211, over 1655259.71 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:15:28,749 INFO [finetune.py:992] (1/2) Epoch 11, batch 0, loss[loss=0.2036, simple_loss=0.2948, pruned_loss=0.05618, over 10728.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2948, pruned_loss=0.05618, over 10728.00 frames. ], batch size: 68, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:15:28,750 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 16:15:36,503 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0760, 5.6102, 5.2790, 5.3071, 5.6879, 5.1741, 5.0581, 5.3602], device='cuda:1'), covar=tensor([0.1262, 0.1036, 0.1289, 0.1888, 0.0790, 0.1878, 0.2467, 0.0984], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0457, 0.0371, 0.0415, 0.0431, 0.0406, 0.0364, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 16:15:44,371 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2927, 4.2079, 4.2252, 4.2237, 3.8064, 4.2532, 4.3153, 4.3750], device='cuda:1'), covar=tensor([0.0346, 0.0166, 0.0221, 0.0352, 0.0725, 0.0414, 0.0225, 0.0238], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0166, 0.0161, 0.0210, 0.0205, 0.0185, 0.0147, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 16:15:47,008 INFO [finetune.py:1026] (1/2) Epoch 11, validation: loss=0.2863, simple_loss=0.3621, pruned_loss=0.1053, over 1020973.00 frames. 2023-05-16 16:15:47,009 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 16:15:52,929 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0509, 2.2905, 3.0623, 3.9785, 2.2055, 4.0491, 4.0160, 4.1353], device='cuda:1'), covar=tensor([0.0112, 0.1262, 0.0443, 0.0112, 0.1323, 0.0241, 0.0168, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0190, 0.0171, 0.0108, 0.0175, 0.0165, 0.0161, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:16:04,070 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.436e+02 3.276e+02 3.874e+02 4.757e+02 1.252e+03, threshold=7.749e+02, percent-clipped=3.0 2023-05-16 16:16:05,465 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222365.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:16:22,434 INFO [finetune.py:992] (1/2) Epoch 11, batch 50, loss[loss=0.182, simple_loss=0.2804, pruned_loss=0.04176, over 11521.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2688, pruned_loss=0.04397, over 531917.97 frames. ], batch size: 48, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:16:40,358 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222414.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:16:57,983 INFO [finetune.py:992] (1/2) Epoch 11, batch 100, loss[loss=0.1828, simple_loss=0.2718, pruned_loss=0.04687, over 12141.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2672, pruned_loss=0.04359, over 947264.76 frames. ], batch size: 34, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:17:14,279 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222462.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:17:14,909 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 2.885e+02 3.237e+02 3.919e+02 6.319e+02, threshold=6.473e+02, percent-clipped=0.0 2023-05-16 16:17:19,403 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1660, 2.6997, 3.8222, 3.2001, 3.6419, 3.3526, 2.8135, 3.7403], device='cuda:1'), covar=tensor([0.0151, 0.0365, 0.0167, 0.0268, 0.0181, 0.0202, 0.0334, 0.0125], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0195, 0.0173, 0.0175, 0.0201, 0.0153, 0.0187, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:17:24,240 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5918, 5.3475, 5.4817, 5.5297, 5.2204, 5.2195, 4.9642, 5.4933], device='cuda:1'), covar=tensor([0.0671, 0.0740, 0.0865, 0.0624, 0.1871, 0.1291, 0.0609, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0646, 0.0560, 0.0578, 0.0767, 0.0682, 0.0508, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:17:33,847 INFO [finetune.py:992] (1/2) Epoch 11, batch 150, loss[loss=0.1749, simple_loss=0.265, pruned_loss=0.04237, over 12208.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2647, pruned_loss=0.04289, over 1271505.31 frames. ], batch size: 35, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:17:55,929 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4098, 5.0111, 5.3973, 4.7048, 5.0376, 4.7745, 5.4844, 5.0673], device='cuda:1'), covar=tensor([0.0304, 0.0335, 0.0273, 0.0269, 0.0343, 0.0315, 0.0208, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0245, 0.0267, 0.0242, 0.0242, 0.0239, 0.0218, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:17:58,814 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222523.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:18:09,725 INFO [finetune.py:992] (1/2) Epoch 11, batch 200, loss[loss=0.1899, simple_loss=0.2845, pruned_loss=0.0476, over 12025.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2634, pruned_loss=0.0423, over 1523690.38 frames. ], batch size: 40, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:18:11,195 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222541.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 16:18:14,871 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222546.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:18:25,420 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222561.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:18:26,777 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.057e+02 2.661e+02 3.019e+02 3.700e+02 1.022e+03, threshold=6.039e+02, percent-clipped=2.0 2023-05-16 16:18:32,069 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222570.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:18:36,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-05-16 16:18:42,027 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222584.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 16:18:45,163 INFO [finetune.py:992] (1/2) Epoch 11, batch 250, loss[loss=0.1894, simple_loss=0.2854, pruned_loss=0.04667, over 12148.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2629, pruned_loss=0.04215, over 1714429.50 frames. ], batch size: 36, lr: 3.97e-03, grad_scale: 8.0 2023-05-16 16:18:48,805 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222594.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:19:08,997 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3580, 4.0221, 3.9299, 4.3033, 2.8140, 3.7511, 2.4152, 4.0392], device='cuda:1'), covar=tensor([0.1690, 0.0744, 0.1109, 0.0748, 0.1316, 0.0730, 0.2081, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0270, 0.0298, 0.0355, 0.0238, 0.0242, 0.0262, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:19:15,223 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222631.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:19:21,180 INFO [finetune.py:992] (1/2) Epoch 11, batch 300, loss[loss=0.1475, simple_loss=0.2306, pruned_loss=0.03221, over 12141.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2611, pruned_loss=0.04161, over 1862060.11 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:19:38,885 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.989e+02 2.847e+02 3.403e+02 4.061e+02 8.881e+02, threshold=6.805e+02, percent-clipped=3.0 2023-05-16 16:19:40,274 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222665.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:19:44,550 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1380, 6.1162, 5.8621, 5.3499, 5.1957, 6.0049, 5.6411, 5.3686], device='cuda:1'), covar=tensor([0.0633, 0.0945, 0.0629, 0.1640, 0.0689, 0.0720, 0.1445, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0541, 0.0498, 0.0607, 0.0398, 0.0685, 0.0740, 0.0547], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 16:19:50,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-05-16 16:19:57,313 INFO [finetune.py:992] (1/2) Epoch 11, batch 350, loss[loss=0.1635, simple_loss=0.2582, pruned_loss=0.03442, over 12103.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2613, pruned_loss=0.04162, over 1962236.81 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:20:14,648 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222713.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:20:33,064 INFO [finetune.py:992] (1/2) Epoch 11, batch 400, loss[loss=0.1655, simple_loss=0.2622, pruned_loss=0.03437, over 12163.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2606, pruned_loss=0.04111, over 2055659.59 frames. ], batch size: 36, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:20:47,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 16:20:50,138 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.851e+02 2.690e+02 3.204e+02 3.852e+02 5.513e+02, threshold=6.408e+02, percent-clipped=1.0 2023-05-16 16:21:08,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-05-16 16:21:09,025 INFO [finetune.py:992] (1/2) Epoch 11, batch 450, loss[loss=0.1475, simple_loss=0.2198, pruned_loss=0.03762, over 11989.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2596, pruned_loss=0.04068, over 2134793.40 frames. ], batch size: 28, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:21:45,579 INFO [finetune.py:992] (1/2) Epoch 11, batch 500, loss[loss=0.1815, simple_loss=0.2755, pruned_loss=0.04375, over 12164.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2601, pruned_loss=0.04096, over 2182114.71 frames. ], batch size: 36, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:21:47,185 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222841.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 16:22:01,185 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222861.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:22:02,451 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.741e+02 3.201e+02 3.737e+02 1.157e+03, threshold=6.403e+02, percent-clipped=2.0 2023-05-16 16:22:13,796 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222879.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 16:22:20,823 INFO [finetune.py:992] (1/2) Epoch 11, batch 550, loss[loss=0.1549, simple_loss=0.2479, pruned_loss=0.031, over 12045.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2592, pruned_loss=0.0407, over 2221551.93 frames. ], batch size: 31, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:22:20,892 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222889.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:22:30,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-05-16 16:22:35,379 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222909.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:22:38,512 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7195, 2.8416, 4.6921, 4.8325, 2.8041, 2.5288, 3.0657, 2.1465], device='cuda:1'), covar=tensor([0.1566, 0.3061, 0.0421, 0.0426, 0.1407, 0.2589, 0.2658, 0.4148], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0382, 0.0269, 0.0296, 0.0265, 0.0299, 0.0375, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:22:45,987 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 16:22:47,606 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222926.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:22:57,489 INFO [finetune.py:992] (1/2) Epoch 11, batch 600, loss[loss=0.1566, simple_loss=0.2437, pruned_loss=0.03474, over 12186.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2577, pruned_loss=0.03997, over 2260291.54 frames. ], batch size: 31, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:22:57,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-05-16 16:22:58,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222940.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:23:15,000 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.030e+02 2.809e+02 3.240e+02 4.121e+02 5.870e+02, threshold=6.479e+02, percent-clipped=0.0 2023-05-16 16:23:19,528 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222969.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:23:19,651 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1112, 4.5396, 3.8806, 4.8220, 4.3266, 2.9093, 4.2673, 2.9098], device='cuda:1'), covar=tensor([0.0942, 0.0863, 0.1601, 0.0494, 0.1369, 0.1664, 0.0933, 0.3428], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0370, 0.0346, 0.0286, 0.0359, 0.0265, 0.0336, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:23:30,046 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222984.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:23:32,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 16:23:33,370 INFO [finetune.py:992] (1/2) Epoch 11, batch 650, loss[loss=0.1479, simple_loss=0.2454, pruned_loss=0.02517, over 12356.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2578, pruned_loss=0.04004, over 2287056.56 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:23:42,318 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223001.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:24:02,438 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9814, 4.9319, 4.8766, 4.8501, 4.5620, 5.0243, 4.9636, 5.2539], device='cuda:1'), covar=tensor([0.0268, 0.0162, 0.0188, 0.0337, 0.0752, 0.0281, 0.0166, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0185, 0.0178, 0.0233, 0.0230, 0.0205, 0.0163, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 16:24:02,475 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223030.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:24:08,665 INFO [finetune.py:992] (1/2) Epoch 11, batch 700, loss[loss=0.1886, simple_loss=0.2722, pruned_loss=0.05251, over 12131.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2578, pruned_loss=0.04011, over 2312842.08 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:24:13,144 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223045.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:24:25,710 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.739e+02 3.214e+02 3.661e+02 1.810e+03, threshold=6.428e+02, percent-clipped=2.0 2023-05-16 16:24:26,642 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223064.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:24:44,725 INFO [finetune.py:992] (1/2) Epoch 11, batch 750, loss[loss=0.1907, simple_loss=0.2857, pruned_loss=0.04787, over 12155.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2583, pruned_loss=0.04009, over 2321287.43 frames. ], batch size: 34, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:24:49,985 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9675, 5.9042, 5.6732, 5.1790, 4.9830, 5.7988, 5.4789, 5.1542], device='cuda:1'), covar=tensor([0.0736, 0.0974, 0.0824, 0.1581, 0.0814, 0.0751, 0.1582, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0604, 0.0547, 0.0508, 0.0618, 0.0405, 0.0698, 0.0751, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 16:25:11,211 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223125.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:25:20,905 INFO [finetune.py:992] (1/2) Epoch 11, batch 800, loss[loss=0.1565, simple_loss=0.2386, pruned_loss=0.0372, over 12129.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2586, pruned_loss=0.04012, over 2330125.67 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:25:25,344 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7569, 3.4051, 5.1044, 2.6009, 2.7352, 3.7839, 3.2294, 3.8342], device='cuda:1'), covar=tensor([0.0502, 0.1103, 0.0325, 0.1174, 0.1965, 0.1481, 0.1324, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0234, 0.0242, 0.0181, 0.0235, 0.0290, 0.0222, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:25:37,927 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.699e+02 3.208e+02 3.913e+02 7.854e+02, threshold=6.416e+02, percent-clipped=1.0 2023-05-16 16:25:45,217 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223173.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 16:25:45,788 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9193, 5.7218, 5.8109, 4.9840, 5.0771, 5.9338, 5.1353, 5.2845], device='cuda:1'), covar=tensor([0.1281, 0.1953, 0.1160, 0.2849, 0.1207, 0.1159, 0.3276, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0611, 0.0552, 0.0512, 0.0625, 0.0411, 0.0703, 0.0760, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 16:25:49,378 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223179.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 16:25:54,367 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9417, 2.9324, 4.7759, 5.0344, 3.0599, 2.7583, 3.0482, 2.3264], device='cuda:1'), covar=tensor([0.1495, 0.3183, 0.0478, 0.0404, 0.1255, 0.2426, 0.2796, 0.4204], device='cuda:1'), in_proj_covar=tensor([0.0288, 0.0382, 0.0269, 0.0296, 0.0264, 0.0299, 0.0374, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:25:56,194 INFO [finetune.py:992] (1/2) Epoch 11, batch 850, loss[loss=0.1875, simple_loss=0.2803, pruned_loss=0.04738, over 11265.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2583, pruned_loss=0.04007, over 2340279.53 frames. ], batch size: 55, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:26:18,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5057, 2.5426, 3.5798, 4.3239, 3.7319, 4.3830, 3.8079, 2.8885], device='cuda:1'), covar=tensor([0.0030, 0.0400, 0.0145, 0.0047, 0.0133, 0.0074, 0.0116, 0.0398], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0122, 0.0104, 0.0075, 0.0099, 0.0113, 0.0093, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:26:23,679 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223226.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:26:24,302 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223227.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:26:30,334 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223234.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 16:26:33,551 INFO [finetune.py:992] (1/2) Epoch 11, batch 900, loss[loss=0.1725, simple_loss=0.2655, pruned_loss=0.03974, over 12124.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2582, pruned_loss=0.04011, over 2348940.09 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:26:50,590 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 2.660e+02 3.181e+02 3.660e+02 7.233e+02, threshold=6.361e+02, percent-clipped=3.0 2023-05-16 16:26:58,408 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223274.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:27:09,017 INFO [finetune.py:992] (1/2) Epoch 11, batch 950, loss[loss=0.1421, simple_loss=0.227, pruned_loss=0.02865, over 12024.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2586, pruned_loss=0.04044, over 2362683.45 frames. ], batch size: 31, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:27:13,974 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223296.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:27:33,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2023-05-16 16:27:34,393 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223325.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:27:44,284 INFO [finetune.py:992] (1/2) Epoch 11, batch 1000, loss[loss=0.1597, simple_loss=0.2413, pruned_loss=0.03907, over 12138.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2587, pruned_loss=0.0404, over 2360606.93 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:27:45,144 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223340.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:28:00,447 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6950, 4.4283, 4.2534, 4.6113, 3.2930, 4.2475, 2.7678, 4.3987], device='cuda:1'), covar=tensor([0.1428, 0.0580, 0.0986, 0.0767, 0.1092, 0.0520, 0.1831, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0268, 0.0299, 0.0358, 0.0239, 0.0244, 0.0262, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:28:01,561 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.720e+02 3.145e+02 3.668e+02 6.745e+02, threshold=6.290e+02, percent-clipped=1.0 2023-05-16 16:28:07,984 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9462, 5.8980, 5.6539, 5.1034, 5.0248, 5.7890, 5.4682, 5.2112], device='cuda:1'), covar=tensor([0.0779, 0.0986, 0.0696, 0.1638, 0.0802, 0.0851, 0.1673, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0608, 0.0553, 0.0513, 0.0625, 0.0410, 0.0703, 0.0762, 0.0562], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 16:28:09,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 16:28:21,552 INFO [finetune.py:992] (1/2) Epoch 11, batch 1050, loss[loss=0.1562, simple_loss=0.2485, pruned_loss=0.03197, over 12304.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.259, pruned_loss=0.04046, over 2368335.13 frames. ], batch size: 34, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:28:43,654 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:28:44,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-05-16 16:28:50,824 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2077, 4.5875, 3.8953, 4.8332, 4.4984, 2.7101, 4.3285, 2.9248], device='cuda:1'), covar=tensor([0.0942, 0.0965, 0.1409, 0.0635, 0.1136, 0.1806, 0.1014, 0.3323], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0371, 0.0349, 0.0289, 0.0361, 0.0266, 0.0338, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:28:57,078 INFO [finetune.py:992] (1/2) Epoch 11, batch 1100, loss[loss=0.1615, simple_loss=0.2546, pruned_loss=0.03416, over 12109.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2594, pruned_loss=0.0407, over 2371499.60 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:29:14,115 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.738e+02 3.282e+02 3.607e+02 7.657e+02, threshold=6.564e+02, percent-clipped=2.0 2023-05-16 16:29:25,672 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3588, 4.7435, 2.7882, 2.7338, 4.1271, 2.5681, 3.9642, 3.1823], device='cuda:1'), covar=tensor([0.0716, 0.0543, 0.1320, 0.1519, 0.0216, 0.1470, 0.0503, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0254, 0.0179, 0.0202, 0.0140, 0.0185, 0.0197, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:29:32,507 INFO [finetune.py:992] (1/2) Epoch 11, batch 1150, loss[loss=0.1719, simple_loss=0.2677, pruned_loss=0.03808, over 11675.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2582, pruned_loss=0.04051, over 2369040.25 frames. ], batch size: 48, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:30:00,830 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223529.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 16:30:08,306 INFO [finetune.py:992] (1/2) Epoch 11, batch 1200, loss[loss=0.1498, simple_loss=0.2357, pruned_loss=0.032, over 12188.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2589, pruned_loss=0.04075, over 2375475.57 frames. ], batch size: 31, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:30:25,837 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.885e+02 3.352e+02 3.938e+02 8.392e+02, threshold=6.704e+02, percent-clipped=4.0 2023-05-16 16:30:36,610 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223578.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:30:44,086 INFO [finetune.py:992] (1/2) Epoch 11, batch 1250, loss[loss=0.1882, simple_loss=0.2784, pruned_loss=0.04901, over 12021.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2591, pruned_loss=0.04093, over 2383224.14 frames. ], batch size: 40, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:30:49,125 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223596.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:09,715 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223625.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:17,045 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7464, 2.5522, 4.0039, 4.1576, 2.7350, 2.6079, 2.7518, 2.2140], device='cuda:1'), covar=tensor([0.1454, 0.3043, 0.0576, 0.0487, 0.1240, 0.2362, 0.2768, 0.4042], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0387, 0.0272, 0.0299, 0.0267, 0.0304, 0.0379, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:31:19,612 INFO [finetune.py:992] (1/2) Epoch 11, batch 1300, loss[loss=0.1379, simple_loss=0.2209, pruned_loss=0.02746, over 12034.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2587, pruned_loss=0.04055, over 2384237.24 frames. ], batch size: 28, lr: 3.96e-03, grad_scale: 8.0 2023-05-16 16:31:19,838 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223639.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:20,461 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223640.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:23,249 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223644.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:36,577 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.575e+02 3.052e+02 3.546e+02 8.927e+02, threshold=6.105e+02, percent-clipped=2.0 2023-05-16 16:31:43,763 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223673.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:55,876 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:31:56,523 INFO [finetune.py:992] (1/2) Epoch 11, batch 1350, loss[loss=0.1531, simple_loss=0.2465, pruned_loss=0.02989, over 12350.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2591, pruned_loss=0.04049, over 2379536.50 frames. ], batch size: 36, lr: 3.95e-03, grad_scale: 8.0 2023-05-16 16:32:18,437 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223720.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:32:28,556 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2887, 2.8279, 2.7803, 2.7538, 2.4767, 2.4264, 2.8258, 1.9465], device='cuda:1'), covar=tensor([0.0358, 0.0206, 0.0229, 0.0224, 0.0436, 0.0302, 0.0165, 0.0506], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0163, 0.0160, 0.0182, 0.0204, 0.0198, 0.0168, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:32:31,885 INFO [finetune.py:992] (1/2) Epoch 11, batch 1400, loss[loss=0.1523, simple_loss=0.2305, pruned_loss=0.03708, over 12352.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2573, pruned_loss=0.03986, over 2379592.38 frames. ], batch size: 30, lr: 3.95e-03, grad_scale: 8.0 2023-05-16 16:32:48,663 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.710e+02 3.305e+02 4.108e+02 9.896e+02, threshold=6.610e+02, percent-clipped=5.0 2023-05-16 16:32:52,400 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223768.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:33:07,213 INFO [finetune.py:992] (1/2) Epoch 11, batch 1450, loss[loss=0.1643, simple_loss=0.2553, pruned_loss=0.03668, over 12098.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2568, pruned_loss=0.03956, over 2381486.03 frames. ], batch size: 33, lr: 3.95e-03, grad_scale: 8.0 2023-05-16 16:33:10,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-16 16:33:36,740 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223829.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 16:33:44,139 INFO [finetune.py:992] (1/2) Epoch 11, batch 1500, loss[loss=0.1736, simple_loss=0.2623, pruned_loss=0.04243, over 11315.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2578, pruned_loss=0.03994, over 2380119.89 frames. ], batch size: 55, lr: 3.95e-03, grad_scale: 8.0 2023-05-16 16:34:01,159 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 2.778e+02 3.304e+02 3.965e+02 3.242e+03, threshold=6.607e+02, percent-clipped=6.0 2023-05-16 16:34:11,150 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223877.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 16:34:13,336 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2014, 5.0389, 5.1270, 5.1935, 4.8404, 4.9298, 4.6524, 5.1454], device='cuda:1'), covar=tensor([0.0757, 0.0608, 0.0900, 0.0588, 0.1853, 0.1372, 0.0569, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0692, 0.0601, 0.0621, 0.0831, 0.0729, 0.0543, 0.0474], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:34:13,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.20 vs. limit=5.0 2023-05-16 16:34:19,633 INFO [finetune.py:992] (1/2) Epoch 11, batch 1550, loss[loss=0.1508, simple_loss=0.2344, pruned_loss=0.03365, over 12349.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.259, pruned_loss=0.04024, over 2386608.38 frames. ], batch size: 31, lr: 3.95e-03, grad_scale: 8.0 2023-05-16 16:34:20,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 16:34:51,527 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223934.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:34:54,976 INFO [finetune.py:992] (1/2) Epoch 11, batch 1600, loss[loss=0.1477, simple_loss=0.2337, pruned_loss=0.03082, over 12137.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.259, pruned_loss=0.04035, over 2381530.30 frames. ], batch size: 30, lr: 3.95e-03, grad_scale: 8.0 2023-05-16 16:35:11,589 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.692e+02 3.217e+02 3.636e+02 1.307e+03, threshold=6.433e+02, percent-clipped=3.0 2023-05-16 16:35:28,108 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2229, 4.9786, 5.1379, 5.1756, 4.8305, 4.8824, 4.5468, 5.1157], device='cuda:1'), covar=tensor([0.0653, 0.0631, 0.0805, 0.0563, 0.1854, 0.1271, 0.0627, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0695, 0.0602, 0.0623, 0.0834, 0.0732, 0.0546, 0.0474], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:35:31,548 INFO [finetune.py:992] (1/2) Epoch 11, batch 1650, loss[loss=0.1668, simple_loss=0.2515, pruned_loss=0.04104, over 12249.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2594, pruned_loss=0.04039, over 2380906.33 frames. ], batch size: 32, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:35:48,454 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6767, 3.7754, 3.3511, 3.3263, 3.0963, 2.9817, 3.8050, 2.5232], device='cuda:1'), covar=tensor([0.0342, 0.0117, 0.0199, 0.0184, 0.0365, 0.0328, 0.0136, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0165, 0.0161, 0.0184, 0.0205, 0.0200, 0.0170, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:36:10,085 INFO [finetune.py:992] (1/2) Epoch 11, batch 1700, loss[loss=0.1868, simple_loss=0.2856, pruned_loss=0.04405, over 12350.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2584, pruned_loss=0.04005, over 2379962.14 frames. ], batch size: 35, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:36:27,400 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.666e+02 3.190e+02 3.702e+02 6.745e+02, threshold=6.379e+02, percent-clipped=1.0 2023-05-16 16:36:34,086 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7221, 2.6287, 3.2872, 4.5486, 2.5787, 4.6159, 4.6634, 4.7830], device='cuda:1'), covar=tensor([0.0109, 0.1232, 0.0471, 0.0170, 0.1208, 0.0219, 0.0137, 0.0103], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0182, 0.0117, 0.0188, 0.0176, 0.0173, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:36:45,452 INFO [finetune.py:992] (1/2) Epoch 11, batch 1750, loss[loss=0.1852, simple_loss=0.2574, pruned_loss=0.05652, over 11878.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2582, pruned_loss=0.04042, over 2373436.38 frames. ], batch size: 26, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:36:59,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2023-05-16 16:37:02,549 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5039, 4.3551, 4.2440, 4.6143, 3.1289, 4.2399, 2.8059, 4.2445], device='cuda:1'), covar=tensor([0.1523, 0.0572, 0.0795, 0.0507, 0.1098, 0.0494, 0.1749, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0270, 0.0300, 0.0359, 0.0239, 0.0243, 0.0263, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:37:21,883 INFO [finetune.py:992] (1/2) Epoch 11, batch 1800, loss[loss=0.1746, simple_loss=0.2575, pruned_loss=0.04586, over 12077.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2578, pruned_loss=0.04029, over 2371725.32 frames. ], batch size: 32, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:37:38,946 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.217e+02 2.842e+02 3.339e+02 3.794e+02 8.326e+02, threshold=6.677e+02, percent-clipped=2.0 2023-05-16 16:37:57,018 INFO [finetune.py:992] (1/2) Epoch 11, batch 1850, loss[loss=0.1646, simple_loss=0.2552, pruned_loss=0.03698, over 12353.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2577, pruned_loss=0.04027, over 2370520.67 frames. ], batch size: 36, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:38:13,249 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7415, 4.3235, 4.3606, 4.6185, 3.1423, 4.2176, 2.9804, 4.2448], device='cuda:1'), covar=tensor([0.1357, 0.0619, 0.0767, 0.0683, 0.1073, 0.0518, 0.1592, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0270, 0.0300, 0.0359, 0.0238, 0.0243, 0.0263, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:38:13,916 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224212.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:38:17,473 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4759, 4.8368, 2.9619, 2.7504, 4.2307, 2.7522, 4.0904, 3.4171], device='cuda:1'), covar=tensor([0.0701, 0.0506, 0.1195, 0.1529, 0.0240, 0.1337, 0.0513, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0253, 0.0177, 0.0200, 0.0141, 0.0183, 0.0196, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:38:26,877 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 16:38:29,443 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224234.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:38:32,775 INFO [finetune.py:992] (1/2) Epoch 11, batch 1900, loss[loss=0.1831, simple_loss=0.2792, pruned_loss=0.04356, over 10523.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.257, pruned_loss=0.04006, over 2371468.04 frames. ], batch size: 68, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:38:49,363 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.710e+02 3.083e+02 3.655e+02 7.153e+02, threshold=6.166e+02, percent-clipped=1.0 2023-05-16 16:38:58,180 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224273.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:39:04,126 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=224282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:39:08,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-16 16:39:09,045 INFO [finetune.py:992] (1/2) Epoch 11, batch 1950, loss[loss=0.156, simple_loss=0.2481, pruned_loss=0.03196, over 12153.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2576, pruned_loss=0.04027, over 2375557.36 frames. ], batch size: 34, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:39:44,488 INFO [finetune.py:992] (1/2) Epoch 11, batch 2000, loss[loss=0.1458, simple_loss=0.2438, pruned_loss=0.02394, over 12325.00 frames. ], tot_loss[loss=0.169, simple_loss=0.258, pruned_loss=0.03997, over 2378236.10 frames. ], batch size: 34, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:40:01,802 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.828e+02 2.747e+02 3.182e+02 3.685e+02 6.124e+02, threshold=6.365e+02, percent-clipped=0.0 2023-05-16 16:40:20,056 INFO [finetune.py:992] (1/2) Epoch 11, batch 2050, loss[loss=0.1754, simple_loss=0.2669, pruned_loss=0.04194, over 11971.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.0401, over 2381584.74 frames. ], batch size: 42, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:40:56,638 INFO [finetune.py:992] (1/2) Epoch 11, batch 2100, loss[loss=0.171, simple_loss=0.2539, pruned_loss=0.04407, over 12125.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.258, pruned_loss=0.04014, over 2381670.24 frames. ], batch size: 30, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:41:13,657 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.918e+02 3.370e+02 3.984e+02 9.339e+02, threshold=6.740e+02, percent-clipped=3.0 2023-05-16 16:41:19,270 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0112, 2.3573, 2.9262, 3.8943, 2.2778, 3.9787, 3.9892, 4.1401], device='cuda:1'), covar=tensor([0.0144, 0.1340, 0.0567, 0.0173, 0.1375, 0.0315, 0.0201, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0204, 0.0184, 0.0118, 0.0188, 0.0178, 0.0176, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:41:31,932 INFO [finetune.py:992] (1/2) Epoch 11, batch 2150, loss[loss=0.1825, simple_loss=0.2639, pruned_loss=0.05055, over 12246.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2587, pruned_loss=0.04055, over 2375169.67 frames. ], batch size: 32, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:42:07,231 INFO [finetune.py:992] (1/2) Epoch 11, batch 2200, loss[loss=0.1693, simple_loss=0.262, pruned_loss=0.03826, over 12365.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2578, pruned_loss=0.04012, over 2379695.53 frames. ], batch size: 35, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:42:11,645 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5742, 4.8090, 4.1759, 5.0691, 4.7613, 3.0453, 4.3111, 3.1504], device='cuda:1'), covar=tensor([0.0705, 0.0746, 0.1353, 0.0473, 0.0948, 0.1488, 0.0995, 0.3146], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0377, 0.0354, 0.0296, 0.0367, 0.0270, 0.0344, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:42:23,541 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224562.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:42:24,048 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 2.732e+02 3.415e+02 4.168e+02 1.123e+03, threshold=6.829e+02, percent-clipped=5.0 2023-05-16 16:42:25,699 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8875, 3.3599, 2.3464, 2.1229, 3.0088, 2.3091, 3.1464, 2.6569], device='cuda:1'), covar=tensor([0.0537, 0.0713, 0.0968, 0.1407, 0.0259, 0.1084, 0.0492, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0253, 0.0176, 0.0200, 0.0141, 0.0182, 0.0196, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:42:29,027 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224568.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:42:43,658 INFO [finetune.py:992] (1/2) Epoch 11, batch 2250, loss[loss=0.1563, simple_loss=0.2475, pruned_loss=0.03253, over 12284.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2588, pruned_loss=0.04048, over 2371795.73 frames. ], batch size: 33, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:43:08,371 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224623.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:43:15,229 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2962, 4.4349, 4.0003, 4.9143, 4.6556, 2.8230, 4.2320, 2.9280], device='cuda:1'), covar=tensor([0.0851, 0.1048, 0.1371, 0.0530, 0.0906, 0.1760, 0.1007, 0.3569], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0379, 0.0357, 0.0298, 0.0369, 0.0272, 0.0345, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:43:19,159 INFO [finetune.py:992] (1/2) Epoch 11, batch 2300, loss[loss=0.1602, simple_loss=0.2549, pruned_loss=0.03273, over 12167.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2574, pruned_loss=0.0399, over 2381078.06 frames. ], batch size: 36, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:43:36,347 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.733e+02 3.158e+02 3.825e+02 8.863e+02, threshold=6.315e+02, percent-clipped=2.0 2023-05-16 16:43:44,015 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3466, 4.5418, 2.6111, 2.0559, 4.0771, 2.1925, 3.8557, 3.0350], device='cuda:1'), covar=tensor([0.0622, 0.0600, 0.1213, 0.1938, 0.0304, 0.1602, 0.0518, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0254, 0.0176, 0.0200, 0.0141, 0.0182, 0.0197, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:43:54,396 INFO [finetune.py:992] (1/2) Epoch 11, batch 2350, loss[loss=0.2291, simple_loss=0.3093, pruned_loss=0.07448, over 7967.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2588, pruned_loss=0.04043, over 2379356.58 frames. ], batch size: 97, lr: 3.95e-03, grad_scale: 16.0 2023-05-16 16:44:14,269 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8360, 2.4850, 3.4396, 2.9299, 3.2747, 3.0353, 2.4738, 3.3265], device='cuda:1'), covar=tensor([0.0162, 0.0348, 0.0175, 0.0259, 0.0162, 0.0194, 0.0368, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0202, 0.0183, 0.0183, 0.0210, 0.0162, 0.0195, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:44:21,921 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224726.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:44:30,993 INFO [finetune.py:992] (1/2) Epoch 11, batch 2400, loss[loss=0.1509, simple_loss=0.2237, pruned_loss=0.03909, over 11767.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2586, pruned_loss=0.04032, over 2379975.54 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:44:31,886 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224740.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:44:38,147 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3806, 4.7542, 2.8323, 2.5783, 4.0984, 2.5744, 3.9857, 3.2725], device='cuda:1'), covar=tensor([0.0724, 0.0574, 0.1276, 0.1742, 0.0269, 0.1419, 0.0520, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0254, 0.0177, 0.0200, 0.0141, 0.0183, 0.0198, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:44:47,588 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.760e+02 3.237e+02 3.987e+02 7.740e+02, threshold=6.473e+02, percent-clipped=4.0 2023-05-16 16:45:04,773 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224787.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:45:05,989 INFO [finetune.py:992] (1/2) Epoch 11, batch 2450, loss[loss=0.1723, simple_loss=0.2713, pruned_loss=0.03665, over 12200.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2589, pruned_loss=0.0402, over 2378830.92 frames. ], batch size: 35, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:45:14,838 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224801.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:45:41,785 INFO [finetune.py:992] (1/2) Epoch 11, batch 2500, loss[loss=0.2043, simple_loss=0.2887, pruned_loss=0.05996, over 12366.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04013, over 2377696.62 frames. ], batch size: 38, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:45:59,441 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.514e+02 2.989e+02 3.631e+02 1.476e+03, threshold=5.979e+02, percent-clipped=2.0 2023-05-16 16:46:03,089 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224868.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:46:18,568 INFO [finetune.py:992] (1/2) Epoch 11, batch 2550, loss[loss=0.1675, simple_loss=0.2477, pruned_loss=0.04369, over 12295.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2578, pruned_loss=0.04009, over 2384471.18 frames. ], batch size: 28, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:46:27,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3186, 4.7146, 2.9172, 2.6796, 4.0730, 2.5165, 4.0713, 3.2995], device='cuda:1'), covar=tensor([0.0727, 0.1005, 0.1110, 0.1483, 0.0310, 0.1384, 0.0447, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0255, 0.0177, 0.0200, 0.0142, 0.0183, 0.0198, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:46:29,931 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9297, 3.5639, 5.2670, 2.9227, 3.0444, 4.0209, 3.4030, 4.0699], device='cuda:1'), covar=tensor([0.0367, 0.1033, 0.0231, 0.1043, 0.1693, 0.1325, 0.1236, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0235, 0.0247, 0.0182, 0.0237, 0.0293, 0.0225, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:46:37,456 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=224916.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:46:38,910 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224918.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:46:53,576 INFO [finetune.py:992] (1/2) Epoch 11, batch 2600, loss[loss=0.1472, simple_loss=0.228, pruned_loss=0.03324, over 12279.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2571, pruned_loss=0.03957, over 2388466.90 frames. ], batch size: 28, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:47:10,460 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.594e+02 2.935e+02 3.512e+02 9.403e+02, threshold=5.870e+02, percent-clipped=2.0 2023-05-16 16:47:28,766 INFO [finetune.py:992] (1/2) Epoch 11, batch 2650, loss[loss=0.1641, simple_loss=0.2563, pruned_loss=0.03598, over 12092.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2573, pruned_loss=0.03968, over 2377694.94 frames. ], batch size: 32, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:48:05,444 INFO [finetune.py:992] (1/2) Epoch 11, batch 2700, loss[loss=0.1563, simple_loss=0.2475, pruned_loss=0.03255, over 12096.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2572, pruned_loss=0.0396, over 2373656.12 frames. ], batch size: 33, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:48:22,302 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.025e+02 2.698e+02 3.150e+02 3.686e+02 6.435e+02, threshold=6.301e+02, percent-clipped=2.0 2023-05-16 16:48:24,509 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2867, 6.1562, 6.0601, 5.4909, 5.2620, 6.1498, 5.7620, 5.5248], device='cuda:1'), covar=tensor([0.0589, 0.0881, 0.0614, 0.1691, 0.0680, 0.0721, 0.1445, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0563, 0.0519, 0.0634, 0.0416, 0.0717, 0.0773, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 16:48:36,049 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:48:37,686 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4871, 3.6134, 3.1774, 3.1518, 2.8905, 2.7626, 3.5903, 2.3650], device='cuda:1'), covar=tensor([0.0415, 0.0135, 0.0200, 0.0221, 0.0401, 0.0370, 0.0126, 0.0493], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0165, 0.0161, 0.0186, 0.0207, 0.0200, 0.0171, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:48:40,996 INFO [finetune.py:992] (1/2) Epoch 11, batch 2750, loss[loss=0.168, simple_loss=0.2555, pruned_loss=0.04023, over 12138.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2569, pruned_loss=0.0394, over 2365698.85 frames. ], batch size: 30, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:48:45,798 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225096.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:48:50,883 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225103.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:49:15,880 INFO [finetune.py:992] (1/2) Epoch 11, batch 2800, loss[loss=0.1703, simple_loss=0.2648, pruned_loss=0.03786, over 12274.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2582, pruned_loss=0.03986, over 2369198.85 frames. ], batch size: 33, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:49:18,867 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225142.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:49:34,286 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.666e+02 3.138e+02 3.783e+02 6.283e+02, threshold=6.276e+02, percent-clipped=0.0 2023-05-16 16:49:35,249 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225164.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:49:52,808 INFO [finetune.py:992] (1/2) Epoch 11, batch 2850, loss[loss=0.1922, simple_loss=0.2845, pruned_loss=0.04995, over 12140.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2586, pruned_loss=0.04006, over 2373726.47 frames. ], batch size: 39, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:50:03,245 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225203.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:50:13,695 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225218.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:50:28,562 INFO [finetune.py:992] (1/2) Epoch 11, batch 2900, loss[loss=0.1574, simple_loss=0.2402, pruned_loss=0.03735, over 12328.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2578, pruned_loss=0.03961, over 2377472.39 frames. ], batch size: 30, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:50:45,561 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.705e+02 3.062e+02 3.716e+02 6.501e+02, threshold=6.124e+02, percent-clipped=1.0 2023-05-16 16:50:47,668 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225266.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:51:04,585 INFO [finetune.py:992] (1/2) Epoch 11, batch 2950, loss[loss=0.1534, simple_loss=0.2408, pruned_loss=0.03299, over 12241.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2579, pruned_loss=0.03973, over 2374970.16 frames. ], batch size: 32, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:51:15,515 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9064, 3.3407, 5.1957, 2.7825, 2.9620, 4.0042, 3.2964, 4.0562], device='cuda:1'), covar=tensor([0.0409, 0.1176, 0.0332, 0.1097, 0.1790, 0.1393, 0.1344, 0.0994], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0235, 0.0247, 0.0182, 0.0238, 0.0294, 0.0225, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:51:36,781 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2250, 4.6060, 4.0537, 4.8281, 4.4869, 2.9302, 4.1937, 3.0847], device='cuda:1'), covar=tensor([0.0792, 0.0718, 0.1389, 0.0575, 0.1102, 0.1595, 0.1015, 0.3166], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0380, 0.0359, 0.0300, 0.0369, 0.0272, 0.0346, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:51:40,655 INFO [finetune.py:992] (1/2) Epoch 11, batch 3000, loss[loss=0.1536, simple_loss=0.2481, pruned_loss=0.02953, over 12353.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2576, pruned_loss=0.03953, over 2376344.59 frames. ], batch size: 35, lr: 3.94e-03, grad_scale: 16.0 2023-05-16 16:51:40,655 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 16:51:59,596 INFO [finetune.py:1026] (1/2) Epoch 11, validation: loss=0.3119, simple_loss=0.3915, pruned_loss=0.1162, over 1020973.00 frames. 2023-05-16 16:51:59,597 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 16:52:16,696 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.992e+02 2.700e+02 3.015e+02 3.753e+02 6.118e+02, threshold=6.030e+02, percent-clipped=0.0 2023-05-16 16:52:30,215 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225382.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:52:35,497 INFO [finetune.py:992] (1/2) Epoch 11, batch 3050, loss[loss=0.1967, simple_loss=0.2829, pruned_loss=0.05522, over 12113.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2573, pruned_loss=0.03993, over 2378386.20 frames. ], batch size: 33, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:52:40,518 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225396.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:52:48,679 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5256, 2.6171, 3.6266, 4.5766, 3.9205, 4.5218, 3.8807, 3.1727], device='cuda:1'), covar=tensor([0.0044, 0.0396, 0.0159, 0.0035, 0.0127, 0.0092, 0.0139, 0.0359], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0123, 0.0105, 0.0077, 0.0102, 0.0115, 0.0095, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 16:53:05,365 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225430.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:11,620 INFO [finetune.py:992] (1/2) Epoch 11, batch 3100, loss[loss=0.1382, simple_loss=0.2239, pruned_loss=0.02625, over 12203.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2573, pruned_loss=0.0401, over 2381750.78 frames. ], batch size: 29, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:53:11,827 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225439.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:15,183 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225444.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:22,307 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5755, 5.1436, 5.4849, 4.8251, 5.1299, 4.9442, 5.5304, 5.1209], device='cuda:1'), covar=tensor([0.0257, 0.0347, 0.0304, 0.0264, 0.0378, 0.0374, 0.0299, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0262, 0.0287, 0.0257, 0.0257, 0.0259, 0.0232, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:53:25,741 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225459.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:28,004 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225462.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:29,233 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 2.825e+02 3.231e+02 3.810e+02 6.061e+02, threshold=6.463e+02, percent-clipped=1.0 2023-05-16 16:53:46,891 INFO [finetune.py:992] (1/2) Epoch 11, batch 3150, loss[loss=0.1754, simple_loss=0.2595, pruned_loss=0.04561, over 12311.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.257, pruned_loss=0.03982, over 2380853.17 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:53:53,144 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225498.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:54,731 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225500.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:53:58,140 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225505.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:54:10,220 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225522.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:54:10,935 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225523.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:54:11,031 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6948, 2.8415, 4.5698, 4.8401, 2.9968, 2.7233, 2.9690, 2.2473], device='cuda:1'), covar=tensor([0.1575, 0.3139, 0.0490, 0.0419, 0.1235, 0.2271, 0.2675, 0.3920], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0387, 0.0274, 0.0301, 0.0268, 0.0303, 0.0378, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:54:22,573 INFO [finetune.py:992] (1/2) Epoch 11, batch 3200, loss[loss=0.1887, simple_loss=0.2724, pruned_loss=0.05249, over 10662.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2576, pruned_loss=0.04004, over 2386169.64 frames. ], batch size: 68, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:54:37,555 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225559.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 16:54:40,876 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.979e+02 3.359e+02 3.894e+02 1.470e+03, threshold=6.719e+02, percent-clipped=3.0 2023-05-16 16:54:42,622 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225566.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:54:54,495 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225583.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:54:58,450 INFO [finetune.py:992] (1/2) Epoch 11, batch 3250, loss[loss=0.1922, simple_loss=0.2876, pruned_loss=0.04838, over 12312.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2571, pruned_loss=0.03956, over 2387169.97 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:55:21,169 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225620.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 16:55:34,489 INFO [finetune.py:992] (1/2) Epoch 11, batch 3300, loss[loss=0.1566, simple_loss=0.2469, pruned_loss=0.03317, over 12205.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2568, pruned_loss=0.03954, over 2388223.83 frames. ], batch size: 29, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:55:50,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.62 vs. limit=5.0 2023-05-16 16:55:52,483 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.640e+02 3.098e+02 3.639e+02 5.733e+02, threshold=6.196e+02, percent-clipped=0.0 2023-05-16 16:56:10,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 2023-05-16 16:56:10,933 INFO [finetune.py:992] (1/2) Epoch 11, batch 3350, loss[loss=0.1628, simple_loss=0.2473, pruned_loss=0.03919, over 12326.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2567, pruned_loss=0.03934, over 2386057.28 frames. ], batch size: 31, lr: 3.94e-03, grad_scale: 8.0 2023-05-16 16:56:47,199 INFO [finetune.py:992] (1/2) Epoch 11, batch 3400, loss[loss=0.1638, simple_loss=0.2558, pruned_loss=0.03588, over 12082.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.257, pruned_loss=0.03988, over 2375440.01 frames. ], batch size: 42, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 16:57:01,289 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225759.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:57:04,702 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.798e+02 3.269e+02 3.862e+02 6.913e+02, threshold=6.537e+02, percent-clipped=1.0 2023-05-16 16:57:10,636 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6615, 3.6517, 3.1632, 3.2208, 2.9529, 2.8125, 3.7486, 2.4701], device='cuda:1'), covar=tensor([0.0355, 0.0138, 0.0233, 0.0200, 0.0384, 0.0364, 0.0127, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0164, 0.0163, 0.0185, 0.0205, 0.0198, 0.0171, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 16:57:22,142 INFO [finetune.py:992] (1/2) Epoch 11, batch 3450, loss[loss=0.1549, simple_loss=0.2448, pruned_loss=0.03249, over 12132.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2568, pruned_loss=0.03982, over 2385227.52 frames. ], batch size: 33, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 16:57:26,589 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225795.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:57:28,799 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225798.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:57:34,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.81 vs. limit=5.0 2023-05-16 16:57:35,247 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225807.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:57:43,242 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225818.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:57:58,474 INFO [finetune.py:992] (1/2) Epoch 11, batch 3500, loss[loss=0.1707, simple_loss=0.2717, pruned_loss=0.03487, over 12355.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2565, pruned_loss=0.03979, over 2389926.22 frames. ], batch size: 36, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 16:58:03,492 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225846.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:58:14,553 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225861.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:58:15,445 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225862.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:58:16,558 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.805e+02 3.086e+02 3.880e+02 7.528e+02, threshold=6.172e+02, percent-clipped=2.0 2023-05-16 16:58:26,490 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225878.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:58:33,828 INFO [finetune.py:992] (1/2) Epoch 11, batch 3550, loss[loss=0.1911, simple_loss=0.2827, pruned_loss=0.04979, over 12127.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2577, pruned_loss=0.04064, over 2376557.33 frames. ], batch size: 39, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 16:58:38,364 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225895.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:58:42,466 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0575, 5.8947, 5.5803, 5.4191, 5.9833, 5.3893, 5.4789, 5.4665], device='cuda:1'), covar=tensor([0.1475, 0.1038, 0.0908, 0.1985, 0.0957, 0.2030, 0.2035, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0492, 0.0391, 0.0447, 0.0463, 0.0436, 0.0396, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 16:58:52,143 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225915.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 16:58:57,919 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225923.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:59:08,896 INFO [finetune.py:992] (1/2) Epoch 11, batch 3600, loss[loss=0.141, simple_loss=0.2222, pruned_loss=0.02994, over 11712.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2575, pruned_loss=0.04044, over 2374236.22 frames. ], batch size: 26, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 16:59:21,240 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225956.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 16:59:26,638 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 2.804e+02 3.266e+02 3.834e+02 8.704e+02, threshold=6.532e+02, percent-clipped=2.0 2023-05-16 16:59:44,822 INFO [finetune.py:992] (1/2) Epoch 11, batch 3650, loss[loss=0.1841, simple_loss=0.2739, pruned_loss=0.04712, over 12045.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2573, pruned_loss=0.04038, over 2372620.68 frames. ], batch size: 40, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:00:16,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 17:00:24,363 INFO [finetune.py:992] (1/2) Epoch 11, batch 3700, loss[loss=0.1915, simple_loss=0.2745, pruned_loss=0.05424, over 12119.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2563, pruned_loss=0.03992, over 2366990.08 frames. ], batch size: 39, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:00:26,019 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226041.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:00:30,872 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6381, 5.3794, 5.0270, 4.9397, 5.4974, 4.7456, 4.9700, 4.9148], device='cuda:1'), covar=tensor([0.1576, 0.1091, 0.1065, 0.2204, 0.1038, 0.2393, 0.1971, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0490, 0.0390, 0.0446, 0.0461, 0.0434, 0.0392, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:00:42,175 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 2.762e+02 3.228e+02 3.974e+02 8.639e+02, threshold=6.455e+02, percent-clipped=4.0 2023-05-16 17:00:59,584 INFO [finetune.py:992] (1/2) Epoch 11, batch 3750, loss[loss=0.1669, simple_loss=0.2561, pruned_loss=0.03886, over 12192.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2562, pruned_loss=0.03989, over 2362841.25 frames. ], batch size: 35, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:01:03,825 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226095.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:01:08,930 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226102.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:01:17,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-16 17:01:20,973 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226118.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:01:25,441 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6686, 3.1205, 3.6008, 4.6348, 3.9209, 4.6541, 3.8823, 3.3344], device='cuda:1'), covar=tensor([0.0036, 0.0324, 0.0160, 0.0045, 0.0115, 0.0069, 0.0132, 0.0348], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0124, 0.0107, 0.0078, 0.0104, 0.0118, 0.0097, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:01:35,799 INFO [finetune.py:992] (1/2) Epoch 11, batch 3800, loss[loss=0.1611, simple_loss=0.2431, pruned_loss=0.03961, over 12356.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2561, pruned_loss=0.03987, over 2366810.25 frames. ], batch size: 30, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:01:39,357 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226143.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:01:51,855 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226161.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:01:53,856 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.852e+02 3.373e+02 4.117e+02 5.606e+02, threshold=6.746e+02, percent-clipped=0.0 2023-05-16 17:01:55,320 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226166.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:02:03,901 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226178.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:02:08,759 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9727, 4.8211, 4.7245, 4.7973, 4.4481, 4.8397, 4.9666, 5.2185], device='cuda:1'), covar=tensor([0.0289, 0.0181, 0.0222, 0.0383, 0.0857, 0.0350, 0.0165, 0.0180], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0195, 0.0188, 0.0243, 0.0240, 0.0214, 0.0173, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 17:02:11,337 INFO [finetune.py:992] (1/2) Epoch 11, batch 3850, loss[loss=0.2028, simple_loss=0.3009, pruned_loss=0.0524, over 12040.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.257, pruned_loss=0.04019, over 2372479.81 frames. ], batch size: 42, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:02:25,895 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226209.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:02:30,285 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226215.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 17:02:32,323 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226218.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:02:38,003 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226226.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:02:42,819 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226233.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:02:46,688 INFO [finetune.py:992] (1/2) Epoch 11, batch 3900, loss[loss=0.1695, simple_loss=0.2635, pruned_loss=0.03776, over 11706.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2584, pruned_loss=0.04102, over 2358326.58 frames. ], batch size: 48, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:02:51,070 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1656, 6.1303, 5.8864, 5.5026, 5.2846, 6.0485, 5.7169, 5.3439], device='cuda:1'), covar=tensor([0.0598, 0.0864, 0.0648, 0.1506, 0.0615, 0.0671, 0.1472, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0551, 0.0509, 0.0622, 0.0408, 0.0699, 0.0757, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:02:55,269 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226251.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:03:04,696 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226263.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 17:03:05,283 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 2.742e+02 3.251e+02 4.151e+02 6.960e+02, threshold=6.503e+02, percent-clipped=2.0 2023-05-16 17:03:22,897 INFO [finetune.py:992] (1/2) Epoch 11, batch 3950, loss[loss=0.1371, simple_loss=0.2225, pruned_loss=0.02582, over 12184.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2566, pruned_loss=0.04007, over 2370087.28 frames. ], batch size: 29, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:03:27,189 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226294.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:03:58,772 INFO [finetune.py:992] (1/2) Epoch 11, batch 4000, loss[loss=0.1564, simple_loss=0.2439, pruned_loss=0.03444, over 12140.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2575, pruned_loss=0.04013, over 2376750.53 frames. ], batch size: 30, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:04:16,258 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 2.667e+02 2.947e+02 3.775e+02 7.891e+02, threshold=5.895e+02, percent-clipped=1.0 2023-05-16 17:04:31,381 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1513, 4.4103, 4.0312, 4.8319, 4.3383, 2.7785, 4.2490, 3.0075], device='cuda:1'), covar=tensor([0.0812, 0.0893, 0.1484, 0.0485, 0.1155, 0.1700, 0.0965, 0.3112], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0373, 0.0352, 0.0293, 0.0363, 0.0268, 0.0338, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:04:33,981 INFO [finetune.py:992] (1/2) Epoch 11, batch 4050, loss[loss=0.1941, simple_loss=0.2867, pruned_loss=0.05076, over 12357.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2584, pruned_loss=0.04078, over 2365638.04 frames. ], batch size: 36, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:04:39,756 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226397.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:05:10,052 INFO [finetune.py:992] (1/2) Epoch 11, batch 4100, loss[loss=0.1321, simple_loss=0.2121, pruned_loss=0.02609, over 12275.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2581, pruned_loss=0.04052, over 2370223.56 frames. ], batch size: 28, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:05:14,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 17:05:27,625 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0582, 5.9921, 5.6048, 5.4755, 6.0307, 5.2950, 5.4805, 5.4773], device='cuda:1'), covar=tensor([0.1464, 0.0921, 0.1054, 0.2070, 0.1001, 0.2101, 0.2079, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0480, 0.0384, 0.0439, 0.0454, 0.0424, 0.0387, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:05:28,199 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.182e+02 2.822e+02 3.304e+02 3.943e+02 5.377e+02, threshold=6.609e+02, percent-clipped=0.0 2023-05-16 17:05:45,790 INFO [finetune.py:992] (1/2) Epoch 11, batch 4150, loss[loss=0.1698, simple_loss=0.2564, pruned_loss=0.04156, over 12187.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2583, pruned_loss=0.04069, over 2363624.83 frames. ], batch size: 35, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:06:06,059 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226518.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:06:20,783 INFO [finetune.py:992] (1/2) Epoch 11, batch 4200, loss[loss=0.197, simple_loss=0.2994, pruned_loss=0.04728, over 12154.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.259, pruned_loss=0.04093, over 2366235.69 frames. ], batch size: 36, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:06:30,027 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226551.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:06:38,979 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.677e+02 3.287e+02 3.935e+02 7.058e+02, threshold=6.573e+02, percent-clipped=2.0 2023-05-16 17:06:39,901 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2074, 4.3568, 2.6721, 2.4840, 3.7536, 2.3599, 3.9356, 3.0316], device='cuda:1'), covar=tensor([0.0754, 0.0554, 0.1148, 0.1535, 0.0295, 0.1431, 0.0443, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0256, 0.0178, 0.0201, 0.0143, 0.0183, 0.0199, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:06:40,402 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226566.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:06:40,556 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6302, 2.2748, 3.2235, 4.5234, 2.5085, 4.5929, 4.6341, 4.6749], device='cuda:1'), covar=tensor([0.0102, 0.1336, 0.0488, 0.0126, 0.1264, 0.0190, 0.0133, 0.0074], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0197, 0.0177, 0.0114, 0.0182, 0.0171, 0.0171, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:06:57,269 INFO [finetune.py:992] (1/2) Epoch 11, batch 4250, loss[loss=0.1387, simple_loss=0.2284, pruned_loss=0.02449, over 12000.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2592, pruned_loss=0.04089, over 2367767.06 frames. ], batch size: 28, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:06:57,353 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226589.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:07:04,460 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226599.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:07:33,241 INFO [finetune.py:992] (1/2) Epoch 11, batch 4300, loss[loss=0.1391, simple_loss=0.2191, pruned_loss=0.02953, over 12323.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2583, pruned_loss=0.04048, over 2375000.44 frames. ], batch size: 28, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:07:50,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.747e+02 3.282e+02 3.834e+02 8.024e+02, threshold=6.565e+02, percent-clipped=1.0 2023-05-16 17:07:54,698 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8670, 4.7378, 4.8074, 4.8522, 4.4736, 4.5418, 4.3322, 4.7872], device='cuda:1'), covar=tensor([0.0758, 0.0588, 0.0824, 0.0565, 0.1825, 0.1215, 0.0600, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0701, 0.0606, 0.0623, 0.0839, 0.0736, 0.0548, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:08:09,133 INFO [finetune.py:992] (1/2) Epoch 11, batch 4350, loss[loss=0.1693, simple_loss=0.2691, pruned_loss=0.03476, over 12280.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2585, pruned_loss=0.0406, over 2365479.37 frames. ], batch size: 37, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:08:14,920 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226697.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:08:39,047 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8188, 4.7199, 4.6619, 4.6958, 4.3241, 4.8026, 4.7905, 4.9758], device='cuda:1'), covar=tensor([0.0253, 0.0162, 0.0180, 0.0368, 0.0814, 0.0317, 0.0157, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0196, 0.0186, 0.0243, 0.0240, 0.0213, 0.0173, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 17:08:45,364 INFO [finetune.py:992] (1/2) Epoch 11, batch 4400, loss[loss=0.1335, simple_loss=0.2171, pruned_loss=0.02496, over 12340.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.04, over 2367880.63 frames. ], batch size: 30, lr: 3.93e-03, grad_scale: 8.0 2023-05-16 17:08:49,711 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226745.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:08:51,229 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5005, 5.0005, 5.4738, 4.7956, 5.0779, 4.8686, 5.4738, 5.0832], device='cuda:1'), covar=tensor([0.0249, 0.0363, 0.0252, 0.0226, 0.0334, 0.0329, 0.0231, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0260, 0.0284, 0.0257, 0.0255, 0.0258, 0.0231, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:09:03,158 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.812e+02 2.735e+02 3.143e+02 3.705e+02 6.632e+02, threshold=6.287e+02, percent-clipped=1.0 2023-05-16 17:09:20,952 INFO [finetune.py:992] (1/2) Epoch 11, batch 4450, loss[loss=0.1787, simple_loss=0.2719, pruned_loss=0.04275, over 11620.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.257, pruned_loss=0.0396, over 2369310.34 frames. ], batch size: 48, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:09:28,217 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7513, 3.4535, 5.2295, 2.6563, 2.8212, 3.7799, 3.3581, 3.8277], device='cuda:1'), covar=tensor([0.0423, 0.1121, 0.0270, 0.1230, 0.1954, 0.1552, 0.1301, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0236, 0.0252, 0.0183, 0.0239, 0.0297, 0.0224, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:09:39,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 17:09:46,104 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1454, 2.5240, 3.6969, 3.0801, 3.5218, 3.2139, 2.5859, 3.5653], device='cuda:1'), covar=tensor([0.0132, 0.0361, 0.0151, 0.0245, 0.0145, 0.0194, 0.0346, 0.0136], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0202, 0.0184, 0.0186, 0.0213, 0.0162, 0.0195, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:09:57,086 INFO [finetune.py:992] (1/2) Epoch 11, batch 4500, loss[loss=0.174, simple_loss=0.2711, pruned_loss=0.03838, over 12262.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2577, pruned_loss=0.0399, over 2367031.51 frames. ], batch size: 37, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:10:14,846 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 2.671e+02 3.036e+02 3.765e+02 8.868e+02, threshold=6.071e+02, percent-clipped=2.0 2023-05-16 17:10:33,105 INFO [finetune.py:992] (1/2) Epoch 11, batch 4550, loss[loss=0.1487, simple_loss=0.2419, pruned_loss=0.02772, over 12200.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2588, pruned_loss=0.04025, over 2363710.71 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:10:33,212 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226889.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:10:49,672 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2774, 2.8006, 4.8524, 2.3065, 2.5218, 3.9087, 3.0013, 3.7212], device='cuda:1'), covar=tensor([0.0499, 0.1245, 0.0234, 0.1238, 0.1756, 0.0956, 0.1304, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0235, 0.0251, 0.0183, 0.0239, 0.0296, 0.0223, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:11:05,297 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1695, 4.8109, 5.1857, 4.4663, 4.8313, 4.5248, 5.1551, 4.9019], device='cuda:1'), covar=tensor([0.0340, 0.0374, 0.0350, 0.0292, 0.0351, 0.0357, 0.0308, 0.0336], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0259, 0.0285, 0.0257, 0.0256, 0.0258, 0.0231, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:11:07,364 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226937.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:11:08,640 INFO [finetune.py:992] (1/2) Epoch 11, batch 4600, loss[loss=0.1758, simple_loss=0.2737, pruned_loss=0.03896, over 12191.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2575, pruned_loss=0.03993, over 2366938.86 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:11:26,236 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.738e+02 3.287e+02 4.050e+02 6.950e+02, threshold=6.573e+02, percent-clipped=1.0 2023-05-16 17:11:44,309 INFO [finetune.py:992] (1/2) Epoch 11, batch 4650, loss[loss=0.1555, simple_loss=0.2418, pruned_loss=0.03453, over 12352.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2575, pruned_loss=0.04007, over 2362698.84 frames. ], batch size: 30, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:11:44,463 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226989.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:11:52,660 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.33 vs. limit=5.0 2023-05-16 17:12:08,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 17:12:20,737 INFO [finetune.py:992] (1/2) Epoch 11, batch 4700, loss[loss=0.1382, simple_loss=0.227, pruned_loss=0.02473, over 12370.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2568, pruned_loss=0.03993, over 2361645.50 frames. ], batch size: 30, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:12:28,789 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227050.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:12:38,482 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.827e+02 3.209e+02 4.002e+02 7.614e+02, threshold=6.418e+02, percent-clipped=2.0 2023-05-16 17:12:53,489 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1368, 3.7370, 5.4562, 3.0300, 3.0264, 4.1216, 3.5190, 4.2230], device='cuda:1'), covar=tensor([0.0327, 0.0994, 0.0322, 0.1026, 0.1738, 0.1402, 0.1157, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0233, 0.0249, 0.0181, 0.0237, 0.0294, 0.0222, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:12:56,131 INFO [finetune.py:992] (1/2) Epoch 11, batch 4750, loss[loss=0.1722, simple_loss=0.2737, pruned_loss=0.03535, over 12277.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.257, pruned_loss=0.03974, over 2366835.42 frames. ], batch size: 37, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:13:32,444 INFO [finetune.py:992] (1/2) Epoch 11, batch 4800, loss[loss=0.1469, simple_loss=0.2315, pruned_loss=0.03116, over 12034.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2573, pruned_loss=0.03994, over 2370268.65 frames. ], batch size: 31, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:13:50,613 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.763e+02 3.237e+02 3.912e+02 6.148e+02, threshold=6.474e+02, percent-clipped=0.0 2023-05-16 17:14:08,358 INFO [finetune.py:992] (1/2) Epoch 11, batch 4850, loss[loss=0.1541, simple_loss=0.2461, pruned_loss=0.03109, over 12180.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.257, pruned_loss=0.03988, over 2375474.84 frames. ], batch size: 29, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:14:41,663 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 17:14:43,917 INFO [finetune.py:992] (1/2) Epoch 11, batch 4900, loss[loss=0.2231, simple_loss=0.2904, pruned_loss=0.07785, over 7976.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2568, pruned_loss=0.0401, over 2369467.47 frames. ], batch size: 98, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:14:46,449 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-16 17:14:52,414 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1768, 6.0750, 5.5856, 5.5801, 6.1166, 5.5222, 5.6299, 5.5868], device='cuda:1'), covar=tensor([0.1287, 0.0873, 0.0935, 0.1814, 0.0825, 0.2061, 0.1715, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0489, 0.0392, 0.0446, 0.0463, 0.0433, 0.0393, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:15:01,558 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.765e+02 3.274e+02 4.144e+02 1.019e+03, threshold=6.547e+02, percent-clipped=2.0 2023-05-16 17:15:19,758 INFO [finetune.py:992] (1/2) Epoch 11, batch 4950, loss[loss=0.1696, simple_loss=0.2667, pruned_loss=0.03623, over 12354.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2569, pruned_loss=0.04022, over 2369622.09 frames. ], batch size: 30, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:15:34,574 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2813, 4.6069, 2.7588, 2.2808, 4.0110, 2.1282, 3.9961, 2.7916], device='cuda:1'), covar=tensor([0.0706, 0.0524, 0.1273, 0.2198, 0.0348, 0.1970, 0.0518, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0257, 0.0179, 0.0202, 0.0145, 0.0184, 0.0200, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:15:55,682 INFO [finetune.py:992] (1/2) Epoch 11, batch 5000, loss[loss=0.1646, simple_loss=0.2643, pruned_loss=0.0324, over 12267.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.257, pruned_loss=0.04002, over 2364957.47 frames. ], batch size: 37, lr: 3.92e-03, grad_scale: 8.0 2023-05-16 17:15:59,932 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227345.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:16:09,806 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2587, 5.2263, 5.0791, 4.6474, 4.7642, 5.2176, 4.8924, 4.7244], device='cuda:1'), covar=tensor([0.0725, 0.0935, 0.0655, 0.1672, 0.1010, 0.0802, 0.1498, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0556, 0.0514, 0.0630, 0.0408, 0.0708, 0.0762, 0.0568], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:16:13,207 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.824e+02 3.448e+02 4.234e+02 1.052e+03, threshold=6.897e+02, percent-clipped=5.0 2023-05-16 17:16:31,083 INFO [finetune.py:992] (1/2) Epoch 11, batch 5050, loss[loss=0.1642, simple_loss=0.249, pruned_loss=0.0397, over 12038.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2562, pruned_loss=0.03956, over 2365947.22 frames. ], batch size: 31, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:16:35,489 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6066, 4.6014, 4.4753, 4.0694, 4.2475, 4.5976, 4.3496, 4.1688], device='cuda:1'), covar=tensor([0.0864, 0.0956, 0.0711, 0.1562, 0.1786, 0.0887, 0.1490, 0.1090], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0557, 0.0514, 0.0629, 0.0408, 0.0707, 0.0761, 0.0569], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:16:47,792 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227411.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:17:04,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-16 17:17:07,541 INFO [finetune.py:992] (1/2) Epoch 11, batch 5100, loss[loss=0.1384, simple_loss=0.2247, pruned_loss=0.02604, over 12204.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2556, pruned_loss=0.03921, over 2370981.84 frames. ], batch size: 29, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:17:23,169 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2362, 4.5880, 3.9528, 4.8516, 4.3412, 2.7537, 4.1991, 3.0280], device='cuda:1'), covar=tensor([0.0732, 0.0696, 0.1520, 0.0424, 0.1234, 0.1697, 0.0972, 0.3110], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0376, 0.0355, 0.0296, 0.0364, 0.0268, 0.0340, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:17:25,710 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 2.628e+02 2.984e+02 3.553e+02 6.120e+02, threshold=5.968e+02, percent-clipped=0.0 2023-05-16 17:17:31,604 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227472.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:17:43,460 INFO [finetune.py:992] (1/2) Epoch 11, batch 5150, loss[loss=0.1481, simple_loss=0.2263, pruned_loss=0.03496, over 12261.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2561, pruned_loss=0.03964, over 2368168.57 frames. ], batch size: 28, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:17:45,978 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 17:18:09,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 17:18:18,559 INFO [finetune.py:992] (1/2) Epoch 11, batch 5200, loss[loss=0.187, simple_loss=0.2854, pruned_loss=0.04424, over 12119.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2572, pruned_loss=0.03984, over 2360869.05 frames. ], batch size: 38, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:18:36,912 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.643e+02 3.110e+02 3.966e+02 7.037e+02, threshold=6.219e+02, percent-clipped=5.0 2023-05-16 17:18:54,568 INFO [finetune.py:992] (1/2) Epoch 11, batch 5250, loss[loss=0.1442, simple_loss=0.233, pruned_loss=0.0277, over 12257.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2567, pruned_loss=0.03952, over 2366699.80 frames. ], batch size: 32, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:18:59,649 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6633, 3.1667, 5.0194, 2.7404, 2.9450, 3.7533, 3.2683, 3.6337], device='cuda:1'), covar=tensor([0.0417, 0.1116, 0.0330, 0.1029, 0.1732, 0.1474, 0.1182, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0231, 0.0246, 0.0179, 0.0235, 0.0291, 0.0219, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:19:15,345 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9351, 5.9639, 5.7620, 5.1950, 5.1728, 5.8188, 5.5084, 5.2382], device='cuda:1'), covar=tensor([0.0620, 0.0719, 0.0635, 0.1604, 0.0687, 0.0759, 0.1281, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0559, 0.0516, 0.0633, 0.0411, 0.0713, 0.0766, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:19:30,881 INFO [finetune.py:992] (1/2) Epoch 11, batch 5300, loss[loss=0.1506, simple_loss=0.2334, pruned_loss=0.0339, over 12269.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2575, pruned_loss=0.0399, over 2374180.87 frames. ], batch size: 28, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:19:35,309 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227645.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:19:48,920 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.109e+02 2.684e+02 3.181e+02 3.822e+02 7.017e+02, threshold=6.362e+02, percent-clipped=2.0 2023-05-16 17:19:54,692 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2264, 6.1582, 6.0416, 5.4715, 5.3398, 6.1525, 5.7836, 5.5621], device='cuda:1'), covar=tensor([0.0618, 0.0943, 0.0618, 0.1723, 0.0577, 0.0709, 0.1420, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0561, 0.0519, 0.0634, 0.0413, 0.0715, 0.0770, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:20:06,767 INFO [finetune.py:992] (1/2) Epoch 11, batch 5350, loss[loss=0.1383, simple_loss=0.2231, pruned_loss=0.02672, over 12338.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.256, pruned_loss=0.03951, over 2378248.30 frames. ], batch size: 30, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:20:09,629 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=227693.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:20:09,772 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227693.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:20:42,905 INFO [finetune.py:992] (1/2) Epoch 11, batch 5400, loss[loss=0.1685, simple_loss=0.2513, pruned_loss=0.04286, over 12369.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2562, pruned_loss=0.03972, over 2380543.70 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:20:53,987 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227754.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:21:01,444 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.711e+02 3.096e+02 3.743e+02 6.834e+02, threshold=6.192e+02, percent-clipped=3.0 2023-05-16 17:21:03,581 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227767.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:21:19,113 INFO [finetune.py:992] (1/2) Epoch 11, batch 5450, loss[loss=0.1883, simple_loss=0.2781, pruned_loss=0.04921, over 12120.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2555, pruned_loss=0.03932, over 2380215.12 frames. ], batch size: 38, lr: 3.92e-03, grad_scale: 16.0 2023-05-16 17:21:46,786 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227827.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:21:54,871 INFO [finetune.py:992] (1/2) Epoch 11, batch 5500, loss[loss=0.1665, simple_loss=0.2626, pruned_loss=0.03524, over 12299.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2553, pruned_loss=0.03947, over 2378290.22 frames. ], batch size: 37, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:21:55,034 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4288, 4.9896, 5.4193, 4.7092, 5.0344, 4.8238, 5.4395, 5.0657], device='cuda:1'), covar=tensor([0.0249, 0.0330, 0.0236, 0.0241, 0.0342, 0.0256, 0.0175, 0.0225], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0257, 0.0282, 0.0253, 0.0255, 0.0256, 0.0231, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:22:12,512 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.638e+02 3.047e+02 3.700e+02 7.213e+02, threshold=6.093e+02, percent-clipped=2.0 2023-05-16 17:22:26,333 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9428, 4.3292, 3.7588, 4.5524, 4.0461, 2.7184, 3.8754, 2.8690], device='cuda:1'), covar=tensor([0.0872, 0.0741, 0.1411, 0.0515, 0.1283, 0.1611, 0.1092, 0.3163], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0378, 0.0356, 0.0299, 0.0367, 0.0270, 0.0343, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:22:30,490 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227888.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:22:30,992 INFO [finetune.py:992] (1/2) Epoch 11, batch 5550, loss[loss=0.1708, simple_loss=0.2646, pruned_loss=0.03848, over 11533.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2563, pruned_loss=0.03983, over 2365384.81 frames. ], batch size: 48, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:22:59,725 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1528, 2.5392, 3.6981, 3.1599, 3.4707, 3.3350, 2.6580, 3.5591], device='cuda:1'), covar=tensor([0.0128, 0.0351, 0.0152, 0.0246, 0.0142, 0.0172, 0.0369, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0203, 0.0187, 0.0188, 0.0217, 0.0165, 0.0196, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:23:07,317 INFO [finetune.py:992] (1/2) Epoch 11, batch 5600, loss[loss=0.1478, simple_loss=0.2404, pruned_loss=0.02761, over 12340.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2561, pruned_loss=0.03983, over 2356830.19 frames. ], batch size: 36, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:23:24,839 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.752e+02 3.255e+02 3.896e+02 8.050e+02, threshold=6.511e+02, percent-clipped=3.0 2023-05-16 17:23:42,512 INFO [finetune.py:992] (1/2) Epoch 11, batch 5650, loss[loss=0.172, simple_loss=0.2614, pruned_loss=0.04131, over 12266.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2559, pruned_loss=0.03958, over 2357942.48 frames. ], batch size: 37, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:23:54,412 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4652, 5.2781, 5.4573, 5.4607, 5.0771, 5.1432, 4.8602, 5.4032], device='cuda:1'), covar=tensor([0.0755, 0.0613, 0.0719, 0.0564, 0.1987, 0.1338, 0.0579, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0703, 0.0602, 0.0619, 0.0842, 0.0734, 0.0548, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:23:59,320 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228007.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:24:21,536 INFO [finetune.py:992] (1/2) Epoch 11, batch 5700, loss[loss=0.1761, simple_loss=0.2742, pruned_loss=0.03904, over 10453.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.257, pruned_loss=0.04026, over 2345736.97 frames. ], batch size: 68, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:24:28,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4821, 5.0651, 5.4557, 4.7822, 5.0957, 4.8868, 5.5062, 5.0782], device='cuda:1'), covar=tensor([0.0244, 0.0329, 0.0263, 0.0238, 0.0317, 0.0297, 0.0178, 0.0317], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0260, 0.0285, 0.0256, 0.0257, 0.0259, 0.0233, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:24:29,346 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228049.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:24:39,953 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.655e+02 3.162e+02 3.777e+02 6.559e+02, threshold=6.324e+02, percent-clipped=1.0 2023-05-16 17:24:42,189 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228067.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:24:42,984 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228068.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:24:51,536 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228080.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:24:57,688 INFO [finetune.py:992] (1/2) Epoch 11, batch 5750, loss[loss=0.1815, simple_loss=0.2631, pruned_loss=0.05, over 12145.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2581, pruned_loss=0.04041, over 2349855.62 frames. ], batch size: 39, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:25:16,247 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228115.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:25:18,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-16 17:25:27,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.12 vs. limit=5.0 2023-05-16 17:25:33,600 INFO [finetune.py:992] (1/2) Epoch 11, batch 5800, loss[loss=0.1683, simple_loss=0.2628, pruned_loss=0.03696, over 11549.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2588, pruned_loss=0.04082, over 2346380.80 frames. ], batch size: 48, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:25:35,265 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228141.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:25:38,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 17:25:42,401 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1717, 4.7014, 4.9535, 5.0054, 4.8300, 4.9782, 4.9101, 2.9058], device='cuda:1'), covar=tensor([0.0099, 0.0062, 0.0083, 0.0055, 0.0045, 0.0095, 0.0084, 0.0708], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0074, 0.0061, 0.0093, 0.0082, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:25:51,482 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.760e+02 3.160e+02 4.031e+02 1.052e+03, threshold=6.321e+02, percent-clipped=2.0 2023-05-16 17:26:04,937 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228183.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:26:09,246 INFO [finetune.py:992] (1/2) Epoch 11, batch 5850, loss[loss=0.1685, simple_loss=0.2542, pruned_loss=0.04136, over 12029.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2578, pruned_loss=0.04039, over 2356018.82 frames. ], batch size: 31, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:26:45,406 INFO [finetune.py:992] (1/2) Epoch 11, batch 5900, loss[loss=0.1827, simple_loss=0.2764, pruned_loss=0.04447, over 12195.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.257, pruned_loss=0.03973, over 2362890.79 frames. ], batch size: 35, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:27:02,780 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.734e+02 3.192e+02 3.742e+02 7.430e+02, threshold=6.385e+02, percent-clipped=3.0 2023-05-16 17:27:05,157 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4037, 5.2537, 5.3433, 5.3878, 5.0182, 5.0645, 4.8204, 5.2989], device='cuda:1'), covar=tensor([0.0813, 0.0611, 0.0767, 0.0600, 0.1870, 0.1259, 0.0558, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0699, 0.0601, 0.0614, 0.0839, 0.0733, 0.0546, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:27:10,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 17:27:14,461 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2889, 5.1319, 5.1777, 5.2367, 4.8760, 4.8956, 4.7027, 5.1673], device='cuda:1'), covar=tensor([0.0666, 0.0566, 0.0839, 0.0588, 0.1742, 0.1350, 0.0532, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0698, 0.0600, 0.0614, 0.0837, 0.0732, 0.0545, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:27:21,370 INFO [finetune.py:992] (1/2) Epoch 11, batch 5950, loss[loss=0.1602, simple_loss=0.2544, pruned_loss=0.03303, over 12286.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2572, pruned_loss=0.03959, over 2370292.79 frames. ], batch size: 33, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:27:48,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-16 17:27:49,801 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228329.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:27:53,347 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228333.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:27:57,355 INFO [finetune.py:992] (1/2) Epoch 11, batch 6000, loss[loss=0.129, simple_loss=0.2135, pruned_loss=0.02218, over 12021.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2576, pruned_loss=0.03981, over 2371722.78 frames. ], batch size: 28, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:27:57,355 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 17:28:15,633 INFO [finetune.py:1026] (1/2) Epoch 11, validation: loss=0.3167, simple_loss=0.3935, pruned_loss=0.12, over 1020973.00 frames. 2023-05-16 17:28:15,634 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 17:28:21,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 17:28:22,755 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228349.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:28:32,791 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228363.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:28:33,398 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.594e+02 3.119e+02 3.794e+02 1.855e+03, threshold=6.238e+02, percent-clipped=1.0 2023-05-16 17:28:51,428 INFO [finetune.py:992] (1/2) Epoch 11, batch 6050, loss[loss=0.1614, simple_loss=0.2445, pruned_loss=0.03914, over 12191.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2578, pruned_loss=0.04001, over 2374608.51 frames. ], batch size: 29, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:28:52,282 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228390.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:28:55,145 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228394.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:28:56,989 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228397.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:29:07,026 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 17:29:20,411 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228429.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 17:29:25,358 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228436.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:29:27,444 INFO [finetune.py:992] (1/2) Epoch 11, batch 6100, loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05887, over 11876.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2595, pruned_loss=0.04117, over 2364364.70 frames. ], batch size: 44, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:29:44,823 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.773e+02 3.303e+02 3.873e+02 8.250e+02, threshold=6.606e+02, percent-clipped=3.0 2023-05-16 17:29:56,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 17:29:58,016 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228483.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:30:02,164 INFO [finetune.py:992] (1/2) Epoch 11, batch 6150, loss[loss=0.1625, simple_loss=0.2521, pruned_loss=0.03645, over 12350.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2593, pruned_loss=0.04106, over 2365009.12 frames. ], batch size: 35, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:30:03,027 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228490.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 17:30:20,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-16 17:30:32,208 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228531.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:30:37,757 INFO [finetune.py:992] (1/2) Epoch 11, batch 6200, loss[loss=0.1861, simple_loss=0.2779, pruned_loss=0.04714, over 12347.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2585, pruned_loss=0.04063, over 2373239.10 frames. ], batch size: 35, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:30:55,315 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 2.786e+02 3.277e+02 4.123e+02 6.920e+02, threshold=6.554e+02, percent-clipped=1.0 2023-05-16 17:30:59,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9917, 3.9345, 3.9674, 4.0447, 3.7774, 3.8283, 3.6957, 3.9927], device='cuda:1'), covar=tensor([0.1117, 0.0680, 0.1345, 0.0743, 0.1740, 0.1345, 0.0659, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0697, 0.0602, 0.0619, 0.0842, 0.0734, 0.0546, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:31:04,131 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3384, 4.5654, 2.7943, 2.1683, 4.0821, 2.3663, 3.9042, 3.1087], device='cuda:1'), covar=tensor([0.0634, 0.0477, 0.1086, 0.1703, 0.0272, 0.1330, 0.0427, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0252, 0.0175, 0.0197, 0.0140, 0.0179, 0.0196, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:31:09,912 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7913, 3.2732, 5.1881, 2.4914, 2.7787, 3.7822, 3.2290, 3.9696], device='cuda:1'), covar=tensor([0.0425, 0.1254, 0.0337, 0.1275, 0.2043, 0.1598, 0.1362, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0234, 0.0250, 0.0182, 0.0238, 0.0292, 0.0222, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:31:13,927 INFO [finetune.py:992] (1/2) Epoch 11, batch 6250, loss[loss=0.1763, simple_loss=0.2756, pruned_loss=0.03851, over 12060.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2591, pruned_loss=0.04085, over 2365086.36 frames. ], batch size: 42, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:31:23,484 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9442, 5.9145, 5.7515, 5.1363, 5.1068, 5.8464, 5.4055, 5.2883], device='cuda:1'), covar=tensor([0.0735, 0.0929, 0.0626, 0.1512, 0.0681, 0.0743, 0.1524, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0611, 0.0553, 0.0507, 0.0622, 0.0409, 0.0703, 0.0759, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:31:34,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.18 vs. limit=5.0 2023-05-16 17:31:49,561 INFO [finetune.py:992] (1/2) Epoch 11, batch 6300, loss[loss=0.1444, simple_loss=0.2359, pruned_loss=0.02648, over 12037.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2593, pruned_loss=0.04059, over 2364272.48 frames. ], batch size: 31, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:32:06,846 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228663.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:32:07,360 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.706e+02 3.069e+02 3.698e+02 7.170e+02, threshold=6.138e+02, percent-clipped=2.0 2023-05-16 17:32:14,490 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1965, 4.0545, 3.9960, 4.2983, 2.9496, 3.8478, 2.7722, 3.9727], device='cuda:1'), covar=tensor([0.1649, 0.0698, 0.0855, 0.0661, 0.1150, 0.0654, 0.1698, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0267, 0.0296, 0.0357, 0.0239, 0.0242, 0.0259, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:32:22,996 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228685.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:32:25,783 INFO [finetune.py:992] (1/2) Epoch 11, batch 6350, loss[loss=0.1701, simple_loss=0.2671, pruned_loss=0.03657, over 10784.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04014, over 2372760.16 frames. ], batch size: 68, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:32:25,902 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228689.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:32:36,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 17:32:41,149 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228711.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:32:51,486 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3340, 4.9135, 5.3052, 4.5982, 4.9556, 4.7121, 5.3393, 5.0042], device='cuda:1'), covar=tensor([0.0240, 0.0328, 0.0253, 0.0252, 0.0335, 0.0284, 0.0199, 0.0283], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0256, 0.0284, 0.0254, 0.0255, 0.0257, 0.0232, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:32:59,337 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228736.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:33:01,384 INFO [finetune.py:992] (1/2) Epoch 11, batch 6400, loss[loss=0.1444, simple_loss=0.2225, pruned_loss=0.03314, over 12168.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2591, pruned_loss=0.04053, over 2373113.17 frames. ], batch size: 29, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:33:18,899 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.844e+02 3.278e+02 3.824e+02 1.219e+03, threshold=6.557e+02, percent-clipped=5.0 2023-05-16 17:33:24,028 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0275, 4.3956, 3.9320, 4.8049, 4.2703, 2.9135, 4.1089, 2.9644], device='cuda:1'), covar=tensor([0.0972, 0.0864, 0.1454, 0.0493, 0.1284, 0.1676, 0.1144, 0.3259], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0384, 0.0362, 0.0306, 0.0374, 0.0274, 0.0347, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:33:33,023 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228784.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:33:33,724 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228785.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 17:33:36,318 INFO [finetune.py:992] (1/2) Epoch 11, batch 6450, loss[loss=0.2058, simple_loss=0.289, pruned_loss=0.06133, over 12018.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2601, pruned_loss=0.04113, over 2377739.54 frames. ], batch size: 40, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:34:12,198 INFO [finetune.py:992] (1/2) Epoch 11, batch 6500, loss[loss=0.2012, simple_loss=0.2867, pruned_loss=0.05786, over 10492.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2612, pruned_loss=0.04157, over 2369462.64 frames. ], batch size: 68, lr: 3.91e-03, grad_scale: 16.0 2023-05-16 17:34:30,686 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.116e+02 2.836e+02 3.404e+02 4.180e+02 7.755e+02, threshold=6.809e+02, percent-clipped=2.0 2023-05-16 17:34:48,415 INFO [finetune.py:992] (1/2) Epoch 11, batch 6550, loss[loss=0.1943, simple_loss=0.287, pruned_loss=0.05081, over 12046.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2613, pruned_loss=0.04139, over 2375506.94 frames. ], batch size: 40, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:35:21,409 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228935.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:35:24,024 INFO [finetune.py:992] (1/2) Epoch 11, batch 6600, loss[loss=0.1556, simple_loss=0.2495, pruned_loss=0.03087, over 12290.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2595, pruned_loss=0.04075, over 2374125.76 frames. ], batch size: 33, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:35:32,735 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9518, 4.8645, 4.7177, 4.8195, 4.3873, 4.9509, 4.9313, 5.0683], device='cuda:1'), covar=tensor([0.0227, 0.0150, 0.0204, 0.0338, 0.0784, 0.0321, 0.0164, 0.0201], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0197, 0.0188, 0.0246, 0.0243, 0.0216, 0.0175, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 17:35:40,571 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8686, 5.8492, 5.6483, 5.2161, 5.0777, 5.7838, 5.3537, 5.1878], device='cuda:1'), covar=tensor([0.0885, 0.0933, 0.0671, 0.1703, 0.0770, 0.0774, 0.1899, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0558, 0.0510, 0.0625, 0.0413, 0.0710, 0.0772, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:35:41,803 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.806e+02 3.298e+02 4.086e+02 1.232e+03, threshold=6.595e+02, percent-clipped=2.0 2023-05-16 17:35:57,697 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228985.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:36:00,414 INFO [finetune.py:992] (1/2) Epoch 11, batch 6650, loss[loss=0.1723, simple_loss=0.2603, pruned_loss=0.04221, over 12286.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2586, pruned_loss=0.04053, over 2376935.58 frames. ], batch size: 34, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:36:00,544 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228989.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:36:05,583 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228996.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:36:27,017 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9526, 4.8793, 4.7132, 4.7680, 4.4561, 4.9097, 4.8551, 5.1378], device='cuda:1'), covar=tensor([0.0335, 0.0184, 0.0278, 0.0384, 0.0829, 0.0338, 0.0197, 0.0197], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0198, 0.0190, 0.0249, 0.0246, 0.0219, 0.0177, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 17:36:28,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 17:36:32,510 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229033.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:36:35,385 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229037.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:36:36,755 INFO [finetune.py:992] (1/2) Epoch 11, batch 6700, loss[loss=0.1281, simple_loss=0.2069, pruned_loss=0.02464, over 12338.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2569, pruned_loss=0.03952, over 2383618.01 frames. ], batch size: 30, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:36:54,253 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.704e+02 3.020e+02 3.596e+02 6.128e+02, threshold=6.041e+02, percent-clipped=0.0 2023-05-16 17:37:09,278 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229085.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 17:37:11,855 INFO [finetune.py:992] (1/2) Epoch 11, batch 6750, loss[loss=0.2128, simple_loss=0.2976, pruned_loss=0.06395, over 11856.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.259, pruned_loss=0.04042, over 2375186.93 frames. ], batch size: 44, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:37:24,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-05-16 17:37:41,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-16 17:37:44,146 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229133.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 17:37:48,229 INFO [finetune.py:992] (1/2) Epoch 11, batch 6800, loss[loss=0.2041, simple_loss=0.2869, pruned_loss=0.06065, over 12125.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2583, pruned_loss=0.04022, over 2372285.89 frames. ], batch size: 38, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:38:06,452 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.994e+02 2.625e+02 3.230e+02 3.758e+02 8.293e+02, threshold=6.460e+02, percent-clipped=1.0 2023-05-16 17:38:24,409 INFO [finetune.py:992] (1/2) Epoch 11, batch 6850, loss[loss=0.158, simple_loss=0.2517, pruned_loss=0.03215, over 12287.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2578, pruned_loss=0.03967, over 2381000.81 frames. ], batch size: 37, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:38:28,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-05-16 17:39:00,057 INFO [finetune.py:992] (1/2) Epoch 11, batch 6900, loss[loss=0.1879, simple_loss=0.2756, pruned_loss=0.05009, over 11650.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2594, pruned_loss=0.04044, over 2359220.98 frames. ], batch size: 48, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:39:10,381 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7885, 2.9621, 4.7158, 4.8834, 2.9735, 2.6907, 3.0605, 2.1873], device='cuda:1'), covar=tensor([0.1552, 0.3133, 0.0439, 0.0429, 0.1275, 0.2393, 0.2584, 0.4061], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0383, 0.0271, 0.0299, 0.0267, 0.0300, 0.0373, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:39:18,633 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.742e+02 3.213e+02 3.908e+02 1.203e+03, threshold=6.427e+02, percent-clipped=1.0 2023-05-16 17:39:36,537 INFO [finetune.py:992] (1/2) Epoch 11, batch 6950, loss[loss=0.145, simple_loss=0.2258, pruned_loss=0.03208, over 12348.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2585, pruned_loss=0.0401, over 2362270.48 frames. ], batch size: 30, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:39:38,029 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229291.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:39:45,239 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229300.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:40:06,581 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3341, 5.1177, 5.2708, 5.2667, 4.9149, 4.9303, 4.6828, 5.2283], device='cuda:1'), covar=tensor([0.0723, 0.0666, 0.0791, 0.0685, 0.2141, 0.1439, 0.0632, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0709, 0.0607, 0.0627, 0.0851, 0.0743, 0.0553, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:40:12,866 INFO [finetune.py:992] (1/2) Epoch 11, batch 7000, loss[loss=0.2048, simple_loss=0.2927, pruned_loss=0.05847, over 11119.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2586, pruned_loss=0.04036, over 2364599.76 frames. ], batch size: 55, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:40:28,248 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229361.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:40:30,051 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 2.844e+02 3.336e+02 3.992e+02 8.859e+02, threshold=6.672e+02, percent-clipped=3.0 2023-05-16 17:40:47,960 INFO [finetune.py:992] (1/2) Epoch 11, batch 7050, loss[loss=0.1936, simple_loss=0.2949, pruned_loss=0.04616, over 11611.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04014, over 2375399.35 frames. ], batch size: 48, lr: 3.90e-03, grad_scale: 32.0 2023-05-16 17:40:56,924 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6592, 3.3703, 5.0860, 2.5722, 2.7290, 3.8643, 2.9547, 3.9297], device='cuda:1'), covar=tensor([0.0481, 0.1158, 0.0401, 0.1198, 0.2015, 0.1338, 0.1544, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0233, 0.0250, 0.0181, 0.0236, 0.0293, 0.0221, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:41:10,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 17:41:14,335 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7391, 2.5833, 4.0020, 4.1420, 2.9162, 2.5969, 2.8468, 2.2532], device='cuda:1'), covar=tensor([0.1335, 0.3006, 0.0554, 0.0434, 0.1060, 0.2167, 0.2475, 0.3721], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0386, 0.0272, 0.0300, 0.0268, 0.0302, 0.0375, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:41:23,996 INFO [finetune.py:992] (1/2) Epoch 11, batch 7100, loss[loss=0.1629, simple_loss=0.2588, pruned_loss=0.03347, over 12283.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.258, pruned_loss=0.04019, over 2369000.58 frames. ], batch size: 37, lr: 3.90e-03, grad_scale: 32.0 2023-05-16 17:41:27,752 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229443.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:41:33,248 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9116, 3.7238, 3.7348, 3.8346, 3.5463, 3.9404, 3.8831, 4.0215], device='cuda:1'), covar=tensor([0.0238, 0.0198, 0.0198, 0.0389, 0.0612, 0.0351, 0.0189, 0.0243], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0198, 0.0189, 0.0248, 0.0244, 0.0217, 0.0177, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 17:41:42,118 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.714e+02 3.188e+02 3.777e+02 1.024e+03, threshold=6.376e+02, percent-clipped=2.0 2023-05-16 17:41:51,602 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229477.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:41:59,992 INFO [finetune.py:992] (1/2) Epoch 11, batch 7150, loss[loss=0.1526, simple_loss=0.2396, pruned_loss=0.03286, over 12335.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2585, pruned_loss=0.04063, over 2367618.29 frames. ], batch size: 30, lr: 3.90e-03, grad_scale: 32.0 2023-05-16 17:42:10,650 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229504.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:42:19,992 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 17:42:34,845 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229538.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:42:35,366 INFO [finetune.py:992] (1/2) Epoch 11, batch 7200, loss[loss=0.1645, simple_loss=0.2555, pruned_loss=0.03678, over 12092.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2586, pruned_loss=0.04057, over 2368263.09 frames. ], batch size: 32, lr: 3.90e-03, grad_scale: 32.0 2023-05-16 17:42:48,999 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9779, 3.7312, 5.3613, 2.6760, 2.8384, 3.8455, 3.2835, 3.8816], device='cuda:1'), covar=tensor([0.0416, 0.1004, 0.0248, 0.1250, 0.2002, 0.1530, 0.1383, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0234, 0.0251, 0.0182, 0.0238, 0.0294, 0.0223, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:42:53,799 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.105e+02 2.606e+02 3.037e+02 3.666e+02 6.952e+02, threshold=6.074e+02, percent-clipped=1.0 2023-05-16 17:43:03,370 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229577.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:43:11,576 INFO [finetune.py:992] (1/2) Epoch 11, batch 7250, loss[loss=0.1642, simple_loss=0.2558, pruned_loss=0.03629, over 12356.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.04061, over 2360656.80 frames. ], batch size: 35, lr: 3.90e-03, grad_scale: 16.0 2023-05-16 17:43:13,102 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229591.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:43:31,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-05-16 17:43:37,530 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6025, 3.2786, 5.0751, 2.4671, 2.7002, 3.7664, 3.0354, 3.7743], device='cuda:1'), covar=tensor([0.0441, 0.1235, 0.0330, 0.1254, 0.2036, 0.1397, 0.1502, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0233, 0.0251, 0.0182, 0.0237, 0.0293, 0.0222, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:43:47,412 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229638.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:43:47,895 INFO [finetune.py:992] (1/2) Epoch 11, batch 7300, loss[loss=0.1697, simple_loss=0.2665, pruned_loss=0.03644, over 12167.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.04048, over 2354656.66 frames. ], batch size: 36, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:43:47,968 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229639.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:43:59,909 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229656.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:44:02,140 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229659.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 17:44:06,890 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.730e+02 3.253e+02 4.003e+02 5.938e+02, threshold=6.506e+02, percent-clipped=0.0 2023-05-16 17:44:15,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-05-16 17:44:23,038 INFO [finetune.py:992] (1/2) Epoch 11, batch 7350, loss[loss=0.194, simple_loss=0.2979, pruned_loss=0.04511, over 12048.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2595, pruned_loss=0.04063, over 2368926.21 frames. ], batch size: 37, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:44:28,177 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8850, 3.5900, 3.7192, 3.7984, 3.7672, 3.8329, 3.6606, 2.6032], device='cuda:1'), covar=tensor([0.0089, 0.0102, 0.0113, 0.0083, 0.0060, 0.0114, 0.0088, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0079, 0.0083, 0.0075, 0.0061, 0.0093, 0.0083, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:44:46,115 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229720.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 17:44:59,245 INFO [finetune.py:992] (1/2) Epoch 11, batch 7400, loss[loss=0.147, simple_loss=0.2479, pruned_loss=0.02302, over 12351.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.258, pruned_loss=0.0399, over 2374475.64 frames. ], batch size: 36, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:45:18,587 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.667e+02 3.178e+02 3.720e+02 7.407e+02, threshold=6.356e+02, percent-clipped=1.0 2023-05-16 17:45:34,699 INFO [finetune.py:992] (1/2) Epoch 11, batch 7450, loss[loss=0.147, simple_loss=0.2338, pruned_loss=0.03006, over 12024.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2586, pruned_loss=0.04024, over 2374863.88 frames. ], batch size: 31, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:45:41,857 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229799.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:46:06,021 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229833.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:46:10,198 INFO [finetune.py:992] (1/2) Epoch 11, batch 7500, loss[loss=0.1674, simple_loss=0.252, pruned_loss=0.04141, over 12020.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2589, pruned_loss=0.04038, over 2370534.42 frames. ], batch size: 31, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:46:22,378 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6249, 2.8613, 4.2290, 4.5363, 2.8859, 2.5778, 2.8175, 2.1853], device='cuda:1'), covar=tensor([0.1426, 0.2556, 0.0537, 0.0412, 0.1210, 0.2227, 0.2678, 0.3684], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0385, 0.0272, 0.0300, 0.0268, 0.0301, 0.0375, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:46:29,851 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 2.803e+02 3.268e+02 4.138e+02 6.772e+02, threshold=6.536e+02, percent-clipped=2.0 2023-05-16 17:46:46,287 INFO [finetune.py:992] (1/2) Epoch 11, batch 7550, loss[loss=0.1708, simple_loss=0.2626, pruned_loss=0.03949, over 12283.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2595, pruned_loss=0.04071, over 2372582.19 frames. ], batch size: 33, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:46:49,987 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229894.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:46:56,599 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2981, 2.7201, 3.9707, 3.3560, 3.7417, 3.5001, 2.8480, 3.8161], device='cuda:1'), covar=tensor([0.0142, 0.0308, 0.0108, 0.0232, 0.0108, 0.0159, 0.0290, 0.0108], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0206, 0.0191, 0.0189, 0.0218, 0.0167, 0.0199, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:47:18,601 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229933.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:47:22,841 INFO [finetune.py:992] (1/2) Epoch 11, batch 7600, loss[loss=0.1358, simple_loss=0.2175, pruned_loss=0.02711, over 12355.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2578, pruned_loss=0.03968, over 2380607.54 frames. ], batch size: 30, lr: 3.90e-03, grad_scale: 8.0 2023-05-16 17:47:34,680 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229955.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:47:35,348 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229956.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:47:42,262 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 2.765e+02 3.248e+02 4.055e+02 1.373e+03, threshold=6.497e+02, percent-clipped=4.0 2023-05-16 17:47:56,879 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229987.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:47:58,162 INFO [finetune.py:992] (1/2) Epoch 11, batch 7650, loss[loss=0.1538, simple_loss=0.248, pruned_loss=0.02977, over 12086.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2575, pruned_loss=0.03974, over 2377250.25 frames. ], batch size: 32, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:48:12,826 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230004.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:48:20,757 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230015.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 17:48:37,920 INFO [finetune.py:992] (1/2) Epoch 11, batch 7700, loss[loss=0.1631, simple_loss=0.2553, pruned_loss=0.03541, over 12116.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2583, pruned_loss=0.04024, over 2369523.15 frames. ], batch size: 33, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:48:44,512 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230048.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:48:51,532 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3435, 3.5178, 3.3082, 3.6475, 3.4932, 2.6826, 3.2820, 2.8338], device='cuda:1'), covar=tensor([0.0812, 0.0929, 0.1338, 0.0863, 0.1128, 0.1487, 0.1033, 0.2716], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0378, 0.0356, 0.0303, 0.0368, 0.0268, 0.0343, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:48:52,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 17:48:56,783 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 2.911e+02 3.424e+02 4.220e+02 6.889e+02, threshold=6.848e+02, percent-clipped=3.0 2023-05-16 17:49:13,216 INFO [finetune.py:992] (1/2) Epoch 11, batch 7750, loss[loss=0.1277, simple_loss=0.2125, pruned_loss=0.02148, over 11982.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2576, pruned_loss=0.04009, over 2376990.21 frames. ], batch size: 28, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:49:20,195 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230099.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:49:31,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5555, 4.5299, 4.4423, 4.1138, 4.1730, 4.5408, 4.2650, 4.1035], device='cuda:1'), covar=tensor([0.0844, 0.0955, 0.0719, 0.1400, 0.1943, 0.0795, 0.1429, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0553, 0.0510, 0.0624, 0.0412, 0.0709, 0.0768, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:49:35,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.27 vs. limit=5.0 2023-05-16 17:49:44,301 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230133.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:49:48,990 INFO [finetune.py:992] (1/2) Epoch 11, batch 7800, loss[loss=0.1787, simple_loss=0.2722, pruned_loss=0.04266, over 11844.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2585, pruned_loss=0.04022, over 2374380.41 frames. ], batch size: 44, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:49:54,842 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230147.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:49:56,989 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230150.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:50:07,996 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 2.602e+02 3.039e+02 3.458e+02 5.413e+02, threshold=6.079e+02, percent-clipped=0.0 2023-05-16 17:50:19,625 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230181.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:50:25,208 INFO [finetune.py:992] (1/2) Epoch 11, batch 7850, loss[loss=0.1514, simple_loss=0.238, pruned_loss=0.03243, over 12169.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2581, pruned_loss=0.04033, over 2365766.67 frames. ], batch size: 31, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:50:41,195 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:50:56,840 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230233.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:51:00,884 INFO [finetune.py:992] (1/2) Epoch 11, batch 7900, loss[loss=0.1684, simple_loss=0.2492, pruned_loss=0.04378, over 12182.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.258, pruned_loss=0.04053, over 2363117.72 frames. ], batch size: 31, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:51:02,617 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230241.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:51:09,047 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230250.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:51:20,371 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 2.861e+02 3.357e+02 3.950e+02 8.975e+02, threshold=6.714e+02, percent-clipped=3.0 2023-05-16 17:51:25,592 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6828, 3.2972, 5.0308, 2.6315, 2.7417, 3.7729, 3.0381, 3.7927], device='cuda:1'), covar=tensor([0.0415, 0.1153, 0.0307, 0.1185, 0.1972, 0.1477, 0.1485, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0232, 0.0251, 0.0180, 0.0238, 0.0293, 0.0220, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:51:31,779 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230281.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:51:37,364 INFO [finetune.py:992] (1/2) Epoch 11, batch 7950, loss[loss=0.1843, simple_loss=0.2866, pruned_loss=0.04102, over 12054.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2584, pruned_loss=0.04058, over 2368076.57 frames. ], batch size: 37, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:51:39,266 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0 2023-05-16 17:51:41,878 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5262, 3.7830, 3.8167, 4.3319, 2.9006, 3.8178, 2.3572, 3.8995], device='cuda:1'), covar=tensor([0.1578, 0.1009, 0.1327, 0.0891, 0.1373, 0.0700, 0.2261, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0269, 0.0295, 0.0360, 0.0240, 0.0243, 0.0260, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:51:46,938 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230302.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:51:52,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8910, 2.4166, 3.3657, 2.8838, 3.2191, 3.0760, 2.4494, 3.2634], device='cuda:1'), covar=tensor([0.0147, 0.0360, 0.0161, 0.0255, 0.0152, 0.0178, 0.0343, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0206, 0.0192, 0.0189, 0.0219, 0.0167, 0.0200, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:51:55,967 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230315.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 17:52:13,565 INFO [finetune.py:992] (1/2) Epoch 11, batch 8000, loss[loss=0.1534, simple_loss=0.2322, pruned_loss=0.0373, over 11793.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2584, pruned_loss=0.04062, over 2364661.50 frames. ], batch size: 26, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:52:16,427 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230343.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:52:30,660 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230363.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 17:52:32,616 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.609e+02 2.925e+02 3.511e+02 7.142e+02, threshold=5.850e+02, percent-clipped=2.0 2023-05-16 17:52:49,082 INFO [finetune.py:992] (1/2) Epoch 11, batch 8050, loss[loss=0.1666, simple_loss=0.2622, pruned_loss=0.03553, over 12281.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2579, pruned_loss=0.04036, over 2362960.64 frames. ], batch size: 37, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:53:25,167 INFO [finetune.py:992] (1/2) Epoch 11, batch 8100, loss[loss=0.1539, simple_loss=0.232, pruned_loss=0.03785, over 11768.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2589, pruned_loss=0.04149, over 2347143.35 frames. ], batch size: 26, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:53:44,598 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.665e+02 3.226e+02 3.957e+02 7.795e+02, threshold=6.451e+02, percent-clipped=2.0 2023-05-16 17:53:46,197 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230468.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:54:00,771 INFO [finetune.py:992] (1/2) Epoch 11, batch 8150, loss[loss=0.199, simple_loss=0.2895, pruned_loss=0.05426, over 10649.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2591, pruned_loss=0.04132, over 2351792.22 frames. ], batch size: 68, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:54:05,208 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4745, 4.8880, 3.2881, 2.8661, 4.2333, 2.9438, 4.2487, 3.6074], device='cuda:1'), covar=tensor([0.0694, 0.0452, 0.0969, 0.1437, 0.0259, 0.1129, 0.0402, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0257, 0.0178, 0.0202, 0.0141, 0.0182, 0.0201, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 17:54:12,802 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230506.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:54:26,347 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230525.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:54:29,138 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230529.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 17:54:31,793 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230533.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:54:35,918 INFO [finetune.py:992] (1/2) Epoch 11, batch 8200, loss[loss=0.1624, simple_loss=0.2538, pruned_loss=0.03547, over 12184.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2598, pruned_loss=0.04206, over 2349183.83 frames. ], batch size: 35, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:54:43,989 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230550.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:54:55,578 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.736e+02 3.152e+02 4.001e+02 7.952e+02, threshold=6.304e+02, percent-clipped=2.0 2023-05-16 17:55:09,932 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230586.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:55:11,887 INFO [finetune.py:992] (1/2) Epoch 11, batch 8250, loss[loss=0.1835, simple_loss=0.2799, pruned_loss=0.04352, over 12157.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2594, pruned_loss=0.04199, over 2353337.58 frames. ], batch size: 36, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:55:15,657 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230594.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:55:17,691 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230597.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:55:18,381 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230598.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:55:48,510 INFO [finetune.py:992] (1/2) Epoch 11, batch 8300, loss[loss=0.1533, simple_loss=0.2392, pruned_loss=0.0337, over 12283.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2598, pruned_loss=0.04182, over 2358036.79 frames. ], batch size: 28, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:55:51,488 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230643.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:56:02,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 17:56:07,449 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.789e+02 3.208e+02 4.062e+02 2.917e+03, threshold=6.416e+02, percent-clipped=6.0 2023-05-16 17:56:11,329 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7596, 2.8816, 4.6399, 4.7840, 2.9389, 2.5866, 3.0252, 2.2273], device='cuda:1'), covar=tensor([0.1565, 0.3148, 0.0479, 0.0476, 0.1321, 0.2549, 0.2768, 0.4133], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0389, 0.0278, 0.0304, 0.0271, 0.0305, 0.0380, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:56:16,834 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1364, 6.1296, 5.8911, 5.3913, 5.1932, 6.0168, 5.6655, 5.4516], device='cuda:1'), covar=tensor([0.0607, 0.0713, 0.0650, 0.1602, 0.0663, 0.0745, 0.1440, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0613, 0.0558, 0.0513, 0.0630, 0.0418, 0.0710, 0.0775, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 17:56:23,409 INFO [finetune.py:992] (1/2) Epoch 11, batch 8350, loss[loss=0.1808, simple_loss=0.2838, pruned_loss=0.03892, over 12191.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2597, pruned_loss=0.04152, over 2365534.35 frames. ], batch size: 35, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:56:24,905 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230691.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:56:32,407 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5320, 2.3112, 3.6954, 4.5845, 3.8895, 4.5443, 3.8707, 2.9244], device='cuda:1'), covar=tensor([0.0037, 0.0468, 0.0166, 0.0037, 0.0122, 0.0069, 0.0110, 0.0428], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0122, 0.0105, 0.0078, 0.0102, 0.0116, 0.0096, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:57:00,091 INFO [finetune.py:992] (1/2) Epoch 11, batch 8400, loss[loss=0.1736, simple_loss=0.2714, pruned_loss=0.03794, over 12155.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2588, pruned_loss=0.04078, over 2372119.17 frames. ], batch size: 34, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:57:01,719 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4351, 3.5954, 3.3355, 3.2288, 3.0256, 2.8586, 3.7271, 2.0270], device='cuda:1'), covar=tensor([0.0467, 0.0190, 0.0180, 0.0198, 0.0342, 0.0288, 0.0125, 0.0577], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0165, 0.0166, 0.0188, 0.0209, 0.0202, 0.0172, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:57:19,674 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 2.634e+02 3.184e+02 3.774e+02 7.292e+02, threshold=6.369e+02, percent-clipped=3.0 2023-05-16 17:57:27,634 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8230, 2.9800, 4.7860, 4.9195, 2.7127, 2.6603, 3.0141, 2.3579], device='cuda:1'), covar=tensor([0.1513, 0.3070, 0.0427, 0.0386, 0.1432, 0.2340, 0.2760, 0.3808], device='cuda:1'), in_proj_covar=tensor([0.0288, 0.0384, 0.0274, 0.0301, 0.0268, 0.0301, 0.0375, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 17:57:35,733 INFO [finetune.py:992] (1/2) Epoch 11, batch 8450, loss[loss=0.1937, simple_loss=0.28, pruned_loss=0.05373, over 11645.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.0406, over 2377331.43 frames. ], batch size: 48, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:57:37,949 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230792.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:57:42,814 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6399, 2.7263, 3.6864, 4.6690, 3.8989, 4.6559, 3.8631, 3.3477], device='cuda:1'), covar=tensor([0.0034, 0.0379, 0.0149, 0.0036, 0.0127, 0.0056, 0.0126, 0.0327], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0122, 0.0105, 0.0078, 0.0102, 0.0116, 0.0096, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:57:47,993 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230806.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:57:48,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-16 17:58:00,516 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230824.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 17:58:11,079 INFO [finetune.py:992] (1/2) Epoch 11, batch 8500, loss[loss=0.1846, simple_loss=0.2763, pruned_loss=0.04645, over 11992.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2584, pruned_loss=0.04036, over 2385692.96 frames. ], batch size: 42, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:58:18,566 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1507, 3.9309, 3.9572, 4.3687, 2.9410, 3.9987, 2.7383, 4.1420], device='cuda:1'), covar=tensor([0.1592, 0.0783, 0.0994, 0.0592, 0.1192, 0.0556, 0.1745, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0269, 0.0295, 0.0359, 0.0239, 0.0242, 0.0258, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 17:58:21,237 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230853.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:58:21,807 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:58:30,852 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.607e+02 3.151e+02 3.693e+02 1.366e+03, threshold=6.301e+02, percent-clipped=2.0 2023-05-16 17:58:41,381 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230881.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:58:46,988 INFO [finetune.py:992] (1/2) Epoch 11, batch 8550, loss[loss=0.1757, simple_loss=0.2714, pruned_loss=0.04, over 11619.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.258, pruned_loss=0.04032, over 2386747.42 frames. ], batch size: 48, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:58:47,084 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230889.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:58:47,863 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230890.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:58:53,013 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230897.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:59:22,828 INFO [finetune.py:992] (1/2) Epoch 11, batch 8600, loss[loss=0.1794, simple_loss=0.2693, pruned_loss=0.0448, over 11147.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2583, pruned_loss=0.04043, over 2386271.49 frames. ], batch size: 55, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 17:59:27,066 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230945.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:59:31,489 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230951.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 17:59:41,760 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.076e+02 2.776e+02 3.433e+02 4.051e+02 7.621e+02, threshold=6.865e+02, percent-clipped=5.0 2023-05-16 17:59:57,894 INFO [finetune.py:992] (1/2) Epoch 11, batch 8650, loss[loss=0.151, simple_loss=0.2401, pruned_loss=0.03093, over 12161.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.258, pruned_loss=0.04057, over 2382810.88 frames. ], batch size: 34, lr: 3.89e-03, grad_scale: 8.0 2023-05-16 18:00:10,464 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5144, 4.1629, 4.1147, 4.5729, 3.1900, 4.1158, 2.5998, 4.3556], device='cuda:1'), covar=tensor([0.1436, 0.0704, 0.0975, 0.0746, 0.1143, 0.0578, 0.1881, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0265, 0.0290, 0.0354, 0.0236, 0.0239, 0.0254, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:00:17,891 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.97 vs. limit=5.0 2023-05-16 18:00:34,265 INFO [finetune.py:992] (1/2) Epoch 11, batch 8700, loss[loss=0.1554, simple_loss=0.2405, pruned_loss=0.03509, over 12040.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2567, pruned_loss=0.03985, over 2387677.03 frames. ], batch size: 31, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:00:54,146 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.745e+02 3.296e+02 3.890e+02 9.227e+02, threshold=6.593e+02, percent-clipped=2.0 2023-05-16 18:01:10,442 INFO [finetune.py:992] (1/2) Epoch 11, batch 8750, loss[loss=0.1644, simple_loss=0.2581, pruned_loss=0.03537, over 12085.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2568, pruned_loss=0.0395, over 2387290.15 frames. ], batch size: 42, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:01:21,354 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 18:01:35,368 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231124.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 18:01:42,879 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-05-16 18:01:45,815 INFO [finetune.py:992] (1/2) Epoch 11, batch 8800, loss[loss=0.1737, simple_loss=0.2585, pruned_loss=0.04443, over 12345.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2583, pruned_loss=0.04023, over 2386660.21 frames. ], batch size: 36, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:01:52,138 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231148.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:01:59,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 2023-05-16 18:02:05,562 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.994e+02 2.928e+02 3.202e+02 4.207e+02 8.695e+02, threshold=6.405e+02, percent-clipped=1.0 2023-05-16 18:02:10,021 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231172.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:02:16,256 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231181.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:02:22,462 INFO [finetune.py:992] (1/2) Epoch 11, batch 8850, loss[loss=0.1718, simple_loss=0.2603, pruned_loss=0.0417, over 12346.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2576, pruned_loss=0.04001, over 2384441.84 frames. ], batch size: 36, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:02:22,625 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231189.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:02:51,089 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231229.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:02:55,344 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3111, 4.9219, 5.2881, 4.6055, 4.9113, 4.6708, 5.3077, 4.9215], device='cuda:1'), covar=tensor([0.0282, 0.0359, 0.0287, 0.0257, 0.0368, 0.0361, 0.0225, 0.0355], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0262, 0.0288, 0.0260, 0.0259, 0.0263, 0.0237, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:02:56,693 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231237.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:02:58,109 INFO [finetune.py:992] (1/2) Epoch 11, batch 8900, loss[loss=0.1412, simple_loss=0.2313, pruned_loss=0.02557, over 12348.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2579, pruned_loss=0.04019, over 2387007.79 frames. ], batch size: 30, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:03:03,066 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231246.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:03:06,641 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231251.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:03:16,931 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.751e+02 3.110e+02 3.886e+02 7.148e+02, threshold=6.221e+02, percent-clipped=1.0 2023-05-16 18:03:33,151 INFO [finetune.py:992] (1/2) Epoch 11, batch 8950, loss[loss=0.1679, simple_loss=0.2428, pruned_loss=0.04653, over 11841.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2587, pruned_loss=0.04049, over 2390150.60 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:03:40,502 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6836, 3.0271, 4.5616, 4.7383, 2.8801, 2.7320, 3.0906, 2.2365], device='cuda:1'), covar=tensor([0.1604, 0.2802, 0.0518, 0.0492, 0.1312, 0.2356, 0.2550, 0.3895], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0389, 0.0277, 0.0304, 0.0272, 0.0306, 0.0380, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:03:50,301 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231312.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:03:58,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-05-16 18:04:08,731 INFO [finetune.py:992] (1/2) Epoch 11, batch 9000, loss[loss=0.1368, simple_loss=0.2213, pruned_loss=0.02615, over 12168.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2592, pruned_loss=0.04115, over 2383963.14 frames. ], batch size: 29, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:04:08,732 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 18:04:22,777 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9580, 4.5188, 4.8790, 4.4266, 4.5697, 4.5049, 4.9433, 4.9306], device='cuda:1'), covar=tensor([0.0247, 0.0407, 0.0308, 0.0252, 0.0422, 0.0380, 0.0237, 0.0176], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0263, 0.0289, 0.0261, 0.0261, 0.0264, 0.0238, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:04:26,869 INFO [finetune.py:1026] (1/2) Epoch 11, validation: loss=0.3233, simple_loss=0.3971, pruned_loss=0.1248, over 1020973.00 frames. 2023-05-16 18:04:26,870 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 18:04:45,843 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 2.737e+02 3.170e+02 3.842e+02 1.182e+03, threshold=6.339e+02, percent-clipped=5.0 2023-05-16 18:05:02,082 INFO [finetune.py:992] (1/2) Epoch 11, batch 9050, loss[loss=0.1564, simple_loss=0.2481, pruned_loss=0.0324, over 12093.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2586, pruned_loss=0.04122, over 2380836.04 frames. ], batch size: 32, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:05:39,128 INFO [finetune.py:992] (1/2) Epoch 11, batch 9100, loss[loss=0.1341, simple_loss=0.2147, pruned_loss=0.02675, over 11815.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2577, pruned_loss=0.04042, over 2377651.05 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:05:45,540 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231448.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:05:58,198 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.760e+02 3.235e+02 3.788e+02 1.428e+03, threshold=6.471e+02, percent-clipped=2.0 2023-05-16 18:06:14,298 INFO [finetune.py:992] (1/2) Epoch 11, batch 9150, loss[loss=0.1757, simple_loss=0.2656, pruned_loss=0.04284, over 12096.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2583, pruned_loss=0.04053, over 2371719.88 frames. ], batch size: 39, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:06:16,609 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5282, 2.5648, 3.2529, 4.4355, 2.4080, 4.3942, 4.5417, 4.6425], device='cuda:1'), covar=tensor([0.0158, 0.1202, 0.0493, 0.0171, 0.1229, 0.0232, 0.0162, 0.0104], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0183, 0.0117, 0.0189, 0.0178, 0.0176, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:06:19,353 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231496.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:06:35,153 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7181, 2.8435, 4.3698, 4.6273, 2.9611, 2.5467, 2.9949, 2.0034], device='cuda:1'), covar=tensor([0.1417, 0.2689, 0.0486, 0.0397, 0.1195, 0.2281, 0.2395, 0.4051], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0385, 0.0276, 0.0301, 0.0270, 0.0303, 0.0378, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:06:48,335 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2010, 6.0798, 5.9362, 5.4269, 5.2627, 6.0416, 5.6690, 5.4430], device='cuda:1'), covar=tensor([0.0601, 0.1058, 0.0619, 0.1552, 0.0598, 0.0681, 0.1412, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0562, 0.0519, 0.0632, 0.0417, 0.0714, 0.0779, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 18:06:49,687 INFO [finetune.py:992] (1/2) Epoch 11, batch 9200, loss[loss=0.1347, simple_loss=0.2204, pruned_loss=0.0245, over 12173.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2583, pruned_loss=0.04043, over 2368179.32 frames. ], batch size: 29, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:06:54,554 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231546.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:07:07,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4478, 4.8153, 2.9281, 2.7225, 4.1837, 2.8419, 4.0973, 3.4595], device='cuda:1'), covar=tensor([0.0737, 0.0476, 0.1168, 0.1544, 0.0269, 0.1280, 0.0424, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0256, 0.0178, 0.0200, 0.0140, 0.0182, 0.0198, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:07:07,851 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2284, 5.1645, 5.0488, 4.6020, 4.7225, 5.1683, 4.8031, 4.6018], device='cuda:1'), covar=tensor([0.0834, 0.1081, 0.0682, 0.1582, 0.1123, 0.0778, 0.1610, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0562, 0.0520, 0.0631, 0.0418, 0.0715, 0.0779, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 18:07:09,112 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.671e+02 3.119e+02 3.638e+02 8.693e+02, threshold=6.238e+02, percent-clipped=1.0 2023-05-16 18:07:26,044 INFO [finetune.py:992] (1/2) Epoch 11, batch 9250, loss[loss=0.1468, simple_loss=0.2258, pruned_loss=0.03389, over 12362.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2583, pruned_loss=0.04021, over 2374398.76 frames. ], batch size: 30, lr: 3.88e-03, grad_scale: 8.0 2023-05-16 18:07:29,734 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231594.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:07:30,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-05-16 18:07:39,311 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231607.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:07:45,108 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5057, 3.4305, 3.1951, 3.0630, 2.8412, 2.6364, 3.5238, 2.2559], device='cuda:1'), covar=tensor([0.0367, 0.0134, 0.0183, 0.0192, 0.0386, 0.0356, 0.0131, 0.0430], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0165, 0.0166, 0.0188, 0.0207, 0.0204, 0.0171, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:08:01,677 INFO [finetune.py:992] (1/2) Epoch 11, batch 9300, loss[loss=0.1521, simple_loss=0.2368, pruned_loss=0.0337, over 12339.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.258, pruned_loss=0.03985, over 2383962.67 frames. ], batch size: 30, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:08:07,497 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231647.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:08:09,632 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4637, 3.4581, 3.1523, 2.9993, 2.7819, 2.5342, 3.5131, 2.2547], device='cuda:1'), covar=tensor([0.0368, 0.0164, 0.0203, 0.0218, 0.0426, 0.0392, 0.0124, 0.0426], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0165, 0.0165, 0.0188, 0.0207, 0.0204, 0.0171, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:08:20,739 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.565e+02 3.061e+02 3.853e+02 6.352e+02, threshold=6.123e+02, percent-clipped=2.0 2023-05-16 18:08:28,144 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1295, 6.0517, 5.8270, 5.3531, 5.2486, 5.9740, 5.6371, 5.3476], device='cuda:1'), covar=tensor([0.0580, 0.0901, 0.0702, 0.1566, 0.0607, 0.0731, 0.1595, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0612, 0.0555, 0.0514, 0.0626, 0.0412, 0.0708, 0.0772, 0.0569], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 18:08:37,118 INFO [finetune.py:992] (1/2) Epoch 11, batch 9350, loss[loss=0.1496, simple_loss=0.2347, pruned_loss=0.03227, over 12287.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2578, pruned_loss=0.03961, over 2382934.51 frames. ], batch size: 28, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:08:38,665 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1697, 6.0620, 5.6680, 5.5678, 6.0933, 5.4907, 5.6883, 5.6234], device='cuda:1'), covar=tensor([0.1521, 0.0950, 0.1051, 0.2130, 0.1020, 0.2172, 0.1551, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0501, 0.0394, 0.0450, 0.0467, 0.0440, 0.0399, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:08:51,728 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231708.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:08:52,510 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5329, 2.9319, 3.8658, 2.2769, 2.5992, 3.0823, 2.9186, 3.3027], device='cuda:1'), covar=tensor([0.0536, 0.1102, 0.0371, 0.1248, 0.1665, 0.1524, 0.1152, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0230, 0.0247, 0.0181, 0.0236, 0.0291, 0.0220, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:08:56,688 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4290, 2.6131, 3.5998, 4.3733, 3.8676, 4.3979, 3.6452, 3.0652], device='cuda:1'), covar=tensor([0.0038, 0.0388, 0.0143, 0.0039, 0.0119, 0.0071, 0.0165, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0122, 0.0105, 0.0078, 0.0102, 0.0116, 0.0096, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:09:14,127 INFO [finetune.py:992] (1/2) Epoch 11, batch 9400, loss[loss=0.1672, simple_loss=0.2589, pruned_loss=0.03774, over 12090.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2584, pruned_loss=0.04007, over 2371559.57 frames. ], batch size: 32, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:09:16,351 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4598, 3.5436, 3.1491, 3.0849, 2.8257, 2.6458, 3.5734, 2.1901], device='cuda:1'), covar=tensor([0.0395, 0.0148, 0.0215, 0.0217, 0.0421, 0.0421, 0.0152, 0.0463], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0165, 0.0166, 0.0188, 0.0207, 0.0204, 0.0171, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:09:33,274 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.554e+02 2.912e+02 3.615e+02 7.355e+02, threshold=5.823e+02, percent-clipped=1.0 2023-05-16 18:09:49,448 INFO [finetune.py:992] (1/2) Epoch 11, batch 9450, loss[loss=0.167, simple_loss=0.2648, pruned_loss=0.03459, over 12290.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2586, pruned_loss=0.04041, over 2373767.48 frames. ], batch size: 37, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:10:24,883 INFO [finetune.py:992] (1/2) Epoch 11, batch 9500, loss[loss=0.1708, simple_loss=0.2636, pruned_loss=0.03897, over 12097.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2586, pruned_loss=0.04025, over 2370880.28 frames. ], batch size: 33, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:10:38,747 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-05-16 18:10:45,345 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 2.719e+02 3.179e+02 3.902e+02 9.199e+02, threshold=6.358e+02, percent-clipped=2.0 2023-05-16 18:10:56,896 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3719, 4.1562, 4.1454, 4.4581, 3.1622, 4.0110, 2.6626, 4.1510], device='cuda:1'), covar=tensor([0.1444, 0.0617, 0.0796, 0.0445, 0.0989, 0.0583, 0.1684, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0266, 0.0292, 0.0356, 0.0238, 0.0242, 0.0257, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:11:01,616 INFO [finetune.py:992] (1/2) Epoch 11, batch 9550, loss[loss=0.1369, simple_loss=0.2304, pruned_loss=0.0217, over 12113.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.0402, over 2368015.62 frames. ], batch size: 30, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:11:04,635 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4941, 2.2587, 3.1636, 4.4285, 2.3088, 4.4274, 4.5511, 4.5934], device='cuda:1'), covar=tensor([0.0127, 0.1421, 0.0496, 0.0162, 0.1361, 0.0207, 0.0140, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0184, 0.0117, 0.0190, 0.0177, 0.0176, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:11:14,480 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231907.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:11:36,723 INFO [finetune.py:992] (1/2) Epoch 11, batch 9600, loss[loss=0.1696, simple_loss=0.256, pruned_loss=0.04157, over 12075.00 frames. ], tot_loss[loss=0.169, simple_loss=0.258, pruned_loss=0.03994, over 2372072.76 frames. ], batch size: 32, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:11:47,843 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231955.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:11:55,413 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 2.672e+02 3.125e+02 3.880e+02 1.029e+03, threshold=6.250e+02, percent-clipped=2.0 2023-05-16 18:12:08,478 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1397, 5.0627, 4.8941, 4.9876, 4.6121, 5.1376, 5.0982, 5.2738], device='cuda:1'), covar=tensor([0.0233, 0.0140, 0.0198, 0.0323, 0.0781, 0.0298, 0.0163, 0.0148], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0198, 0.0190, 0.0249, 0.0246, 0.0216, 0.0177, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 18:12:11,954 INFO [finetune.py:992] (1/2) Epoch 11, batch 9650, loss[loss=0.2026, simple_loss=0.2929, pruned_loss=0.05617, over 11548.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.258, pruned_loss=0.04016, over 2372383.88 frames. ], batch size: 48, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:12:12,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 18:12:18,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-05-16 18:12:25,403 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232003.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 18:12:43,732 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1708, 3.9782, 4.0332, 4.3328, 3.2734, 3.8612, 2.5715, 4.1119], device='cuda:1'), covar=tensor([0.1565, 0.0777, 0.0903, 0.0633, 0.1001, 0.0692, 0.1838, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0267, 0.0294, 0.0358, 0.0239, 0.0243, 0.0259, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:12:51,277 INFO [finetune.py:992] (1/2) Epoch 11, batch 9700, loss[loss=0.1526, simple_loss=0.2468, pruned_loss=0.02917, over 12116.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2585, pruned_loss=0.04045, over 2367988.47 frames. ], batch size: 33, lr: 3.88e-03, grad_scale: 16.0 2023-05-16 18:12:57,640 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2491, 5.0817, 5.1966, 5.2507, 4.8880, 4.9363, 4.6731, 5.1722], device='cuda:1'), covar=tensor([0.0740, 0.0604, 0.0783, 0.0562, 0.1944, 0.1325, 0.0527, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0545, 0.0708, 0.0612, 0.0627, 0.0851, 0.0752, 0.0556, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:13:10,112 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 2.676e+02 3.220e+02 3.837e+02 6.586e+02, threshold=6.439e+02, percent-clipped=2.0 2023-05-16 18:13:20,243 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1683, 2.4393, 3.6595, 3.1241, 3.4537, 3.1694, 2.5501, 3.4706], device='cuda:1'), covar=tensor([0.0135, 0.0404, 0.0162, 0.0245, 0.0166, 0.0214, 0.0360, 0.0156], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0206, 0.0191, 0.0189, 0.0219, 0.0168, 0.0197, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:13:26,479 INFO [finetune.py:992] (1/2) Epoch 11, batch 9750, loss[loss=0.1779, simple_loss=0.2775, pruned_loss=0.03915, over 12177.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2585, pruned_loss=0.04022, over 2373025.60 frames. ], batch size: 35, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:14:02,360 INFO [finetune.py:992] (1/2) Epoch 11, batch 9800, loss[loss=0.1641, simple_loss=0.2525, pruned_loss=0.03784, over 12352.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.03999, over 2370849.29 frames. ], batch size: 31, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:14:21,616 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2655, 4.8368, 5.2375, 4.5663, 4.8823, 4.6438, 5.2829, 4.9059], device='cuda:1'), covar=tensor([0.0253, 0.0397, 0.0275, 0.0273, 0.0376, 0.0333, 0.0187, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0263, 0.0286, 0.0259, 0.0260, 0.0262, 0.0235, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:14:22,137 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.630e+02 3.078e+02 3.591e+02 8.413e+02, threshold=6.157e+02, percent-clipped=1.0 2023-05-16 18:14:38,081 INFO [finetune.py:992] (1/2) Epoch 11, batch 9850, loss[loss=0.1526, simple_loss=0.2327, pruned_loss=0.03628, over 11808.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.04006, over 2369490.30 frames. ], batch size: 26, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:15:13,676 INFO [finetune.py:992] (1/2) Epoch 11, batch 9900, loss[loss=0.1615, simple_loss=0.2462, pruned_loss=0.03844, over 12186.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2563, pruned_loss=0.03973, over 2367024.99 frames. ], batch size: 31, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:15:18,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-05-16 18:15:19,478 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232247.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:15:32,852 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.591e+02 3.141e+02 3.790e+02 6.209e+02, threshold=6.282e+02, percent-clipped=1.0 2023-05-16 18:15:49,744 INFO [finetune.py:992] (1/2) Epoch 11, batch 9950, loss[loss=0.1584, simple_loss=0.2462, pruned_loss=0.03527, over 12080.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2566, pruned_loss=0.0397, over 2375285.14 frames. ], batch size: 32, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:16:00,430 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232303.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:16:04,210 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232308.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:16:25,510 INFO [finetune.py:992] (1/2) Epoch 11, batch 10000, loss[loss=0.1384, simple_loss=0.2256, pruned_loss=0.02558, over 12352.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2567, pruned_loss=0.03963, over 2371336.79 frames. ], batch size: 31, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:16:33,909 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232351.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:16:44,833 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.705e+02 3.211e+02 4.031e+02 1.098e+03, threshold=6.422e+02, percent-clipped=4.0 2023-05-16 18:16:52,893 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1406, 5.9598, 5.5746, 5.4273, 6.0219, 5.3685, 5.4617, 5.5015], device='cuda:1'), covar=tensor([0.1372, 0.0994, 0.1078, 0.1978, 0.1041, 0.2291, 0.1965, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0495, 0.0391, 0.0442, 0.0461, 0.0440, 0.0396, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:16:55,810 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232381.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:17:01,431 INFO [finetune.py:992] (1/2) Epoch 11, batch 10050, loss[loss=0.1582, simple_loss=0.237, pruned_loss=0.03972, over 12140.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2569, pruned_loss=0.03935, over 2377780.62 frames. ], batch size: 30, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:17:06,783 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 18:17:38,395 INFO [finetune.py:992] (1/2) Epoch 11, batch 10100, loss[loss=0.1408, simple_loss=0.2323, pruned_loss=0.02466, over 12261.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2575, pruned_loss=0.03974, over 2369660.51 frames. ], batch size: 32, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:17:40,765 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232442.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 18:17:57,700 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.646e+02 2.968e+02 3.640e+02 9.346e+02, threshold=5.937e+02, percent-clipped=2.0 2023-05-16 18:18:04,733 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232476.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:18:13,580 INFO [finetune.py:992] (1/2) Epoch 11, batch 10150, loss[loss=0.1842, simple_loss=0.2752, pruned_loss=0.04665, over 12115.00 frames. ], tot_loss[loss=0.169, simple_loss=0.258, pruned_loss=0.04002, over 2374297.81 frames. ], batch size: 39, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:18:32,880 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8012, 4.4447, 4.5803, 4.6495, 4.5067, 4.7088, 4.6020, 2.5283], device='cuda:1'), covar=tensor([0.0093, 0.0074, 0.0091, 0.0067, 0.0047, 0.0092, 0.0088, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0075, 0.0061, 0.0092, 0.0083, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:18:34,274 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232518.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:18:47,847 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232537.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:18:49,010 INFO [finetune.py:992] (1/2) Epoch 11, batch 10200, loss[loss=0.2595, simple_loss=0.3292, pruned_loss=0.09486, over 8033.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2587, pruned_loss=0.04044, over 2370678.01 frames. ], batch size: 98, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:19:08,219 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 2.823e+02 3.310e+02 4.105e+02 7.764e+02, threshold=6.621e+02, percent-clipped=2.0 2023-05-16 18:19:12,618 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1492, 5.1276, 4.9942, 5.0317, 4.7294, 5.0826, 5.1033, 5.3018], device='cuda:1'), covar=tensor([0.0277, 0.0130, 0.0188, 0.0357, 0.0709, 0.0323, 0.0176, 0.0166], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0195, 0.0188, 0.0247, 0.0243, 0.0215, 0.0176, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 18:19:18,496 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232579.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:19:25,338 INFO [finetune.py:992] (1/2) Epoch 11, batch 10250, loss[loss=0.1838, simple_loss=0.2653, pruned_loss=0.05115, over 12368.00 frames. ], tot_loss[loss=0.17, simple_loss=0.259, pruned_loss=0.04049, over 2372135.49 frames. ], batch size: 38, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:19:35,633 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232603.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:20:01,065 INFO [finetune.py:992] (1/2) Epoch 11, batch 10300, loss[loss=0.1753, simple_loss=0.2669, pruned_loss=0.0418, over 11830.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2582, pruned_loss=0.04015, over 2370801.14 frames. ], batch size: 44, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:20:04,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 18:20:19,848 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 2.566e+02 2.999e+02 3.659e+02 5.204e+02, threshold=5.997e+02, percent-clipped=0.0 2023-05-16 18:20:21,347 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8751, 4.5188, 4.6328, 4.7372, 4.5963, 4.7673, 4.7269, 2.5358], device='cuda:1'), covar=tensor([0.0082, 0.0069, 0.0094, 0.0067, 0.0050, 0.0097, 0.0069, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0079, 0.0082, 0.0074, 0.0061, 0.0092, 0.0083, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:20:35,897 INFO [finetune.py:992] (1/2) Epoch 11, batch 10350, loss[loss=0.1629, simple_loss=0.2557, pruned_loss=0.03511, over 12151.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.04056, over 2373568.48 frames. ], batch size: 36, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:20:44,601 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5503, 3.6154, 3.2844, 3.1716, 2.9287, 2.7832, 3.6601, 2.2543], device='cuda:1'), covar=tensor([0.0415, 0.0145, 0.0217, 0.0238, 0.0430, 0.0380, 0.0145, 0.0536], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0165, 0.0165, 0.0188, 0.0206, 0.0202, 0.0171, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:21:11,095 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232737.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:21:12,395 INFO [finetune.py:992] (1/2) Epoch 11, batch 10400, loss[loss=0.1734, simple_loss=0.2549, pruned_loss=0.04598, over 11683.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2589, pruned_loss=0.04062, over 2365529.30 frames. ], batch size: 48, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:21:13,403 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6285, 4.3386, 4.6175, 4.1814, 4.3599, 4.1255, 4.6353, 4.1969], device='cuda:1'), covar=tensor([0.0273, 0.0367, 0.0263, 0.0235, 0.0369, 0.0340, 0.0212, 0.0711], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0262, 0.0287, 0.0257, 0.0258, 0.0261, 0.0233, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:21:20,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-16 18:21:31,592 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.598e+02 3.079e+02 3.556e+02 7.043e+02, threshold=6.158e+02, percent-clipped=2.0 2023-05-16 18:21:42,363 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1256, 3.9697, 3.9652, 4.3798, 2.9955, 3.8250, 2.4664, 4.0507], device='cuda:1'), covar=tensor([0.1655, 0.0760, 0.0990, 0.0667, 0.1164, 0.0647, 0.1947, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0268, 0.0295, 0.0359, 0.0240, 0.0243, 0.0258, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:21:42,957 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1058, 5.9227, 5.5480, 5.5547, 6.0586, 5.4353, 5.5052, 5.5118], device='cuda:1'), covar=tensor([0.1452, 0.0941, 0.0977, 0.1735, 0.0938, 0.2088, 0.2012, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0499, 0.0394, 0.0445, 0.0463, 0.0443, 0.0399, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:21:47,782 INFO [finetune.py:992] (1/2) Epoch 11, batch 10450, loss[loss=0.2704, simple_loss=0.3433, pruned_loss=0.09876, over 8123.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.259, pruned_loss=0.04066, over 2366385.45 frames. ], batch size: 98, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:21:54,377 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0354, 3.6148, 5.3722, 2.7591, 3.0106, 3.8853, 3.4704, 3.9850], device='cuda:1'), covar=tensor([0.0330, 0.0923, 0.0243, 0.1123, 0.1867, 0.1483, 0.1201, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0233, 0.0251, 0.0182, 0.0237, 0.0294, 0.0222, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:22:18,716 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232832.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:22:23,618 INFO [finetune.py:992] (1/2) Epoch 11, batch 10500, loss[loss=0.1939, simple_loss=0.2698, pruned_loss=0.05897, over 12046.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2584, pruned_loss=0.04026, over 2374986.25 frames. ], batch size: 40, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:22:43,735 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 2.625e+02 3.024e+02 3.562e+02 9.074e+02, threshold=6.047e+02, percent-clipped=2.0 2023-05-16 18:22:49,476 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232874.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:22:59,804 INFO [finetune.py:992] (1/2) Epoch 11, batch 10550, loss[loss=0.1485, simple_loss=0.242, pruned_loss=0.02749, over 12239.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2579, pruned_loss=0.04003, over 2372594.40 frames. ], batch size: 32, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:23:04,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-16 18:23:09,833 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1778, 6.1104, 5.9200, 5.4091, 5.1583, 6.0458, 5.6852, 5.4111], device='cuda:1'), covar=tensor([0.0582, 0.0893, 0.0591, 0.1462, 0.0662, 0.0663, 0.1427, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0557, 0.0519, 0.0635, 0.0422, 0.0717, 0.0781, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 18:23:09,857 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232903.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:23:32,788 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 18:23:35,255 INFO [finetune.py:992] (1/2) Epoch 11, batch 10600, loss[loss=0.1647, simple_loss=0.2602, pruned_loss=0.03462, over 12001.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2576, pruned_loss=0.0397, over 2380172.62 frames. ], batch size: 40, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:23:43,741 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232951.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:23:51,749 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5722, 3.6439, 3.2769, 3.1655, 3.0160, 2.6687, 3.7712, 2.3262], device='cuda:1'), covar=tensor([0.0380, 0.0135, 0.0234, 0.0199, 0.0386, 0.0389, 0.0110, 0.0458], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0166, 0.0166, 0.0189, 0.0206, 0.0203, 0.0172, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:23:54,290 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.546e+02 3.052e+02 3.532e+02 7.559e+02, threshold=6.103e+02, percent-clipped=1.0 2023-05-16 18:24:10,503 INFO [finetune.py:992] (1/2) Epoch 11, batch 10650, loss[loss=0.1839, simple_loss=0.2774, pruned_loss=0.04527, over 12030.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2569, pruned_loss=0.03967, over 2373581.00 frames. ], batch size: 40, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:24:11,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232990.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:24:32,821 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233018.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:24:46,283 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233037.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:24:47,551 INFO [finetune.py:992] (1/2) Epoch 11, batch 10700, loss[loss=0.1454, simple_loss=0.23, pruned_loss=0.03035, over 11853.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2574, pruned_loss=0.04009, over 2363044.43 frames. ], batch size: 26, lr: 3.87e-03, grad_scale: 16.0 2023-05-16 18:24:56,433 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233051.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:25:07,316 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.769e+02 2.764e+02 3.107e+02 3.687e+02 8.150e+02, threshold=6.215e+02, percent-clipped=3.0 2023-05-16 18:25:16,025 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233079.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:25:20,163 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233085.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:25:23,004 INFO [finetune.py:992] (1/2) Epoch 11, batch 10750, loss[loss=0.1642, simple_loss=0.2555, pruned_loss=0.03646, over 12192.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2573, pruned_loss=0.03979, over 2369422.65 frames. ], batch size: 35, lr: 3.87e-03, grad_scale: 8.0 2023-05-16 18:25:49,146 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9917, 4.5987, 4.6899, 4.8514, 4.6362, 4.8579, 4.7194, 2.8176], device='cuda:1'), covar=tensor([0.0096, 0.0067, 0.0098, 0.0061, 0.0047, 0.0105, 0.0091, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0080, 0.0083, 0.0075, 0.0061, 0.0093, 0.0084, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:25:53,540 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233132.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:25:58,825 INFO [finetune.py:992] (1/2) Epoch 11, batch 10800, loss[loss=0.2002, simple_loss=0.2958, pruned_loss=0.05234, over 12147.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2581, pruned_loss=0.03987, over 2365868.46 frames. ], batch size: 39, lr: 3.87e-03, grad_scale: 8.0 2023-05-16 18:26:19,294 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.655e+02 3.240e+02 3.813e+02 6.450e+02, threshold=6.480e+02, percent-clipped=1.0 2023-05-16 18:26:24,379 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233174.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:26:28,471 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233180.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:26:34,609 INFO [finetune.py:992] (1/2) Epoch 11, batch 10850, loss[loss=0.16, simple_loss=0.2411, pruned_loss=0.03942, over 12346.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2588, pruned_loss=0.04061, over 2367029.13 frames. ], batch size: 31, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:27:00,366 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233222.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:27:12,429 INFO [finetune.py:992] (1/2) Epoch 11, batch 10900, loss[loss=0.1754, simple_loss=0.2698, pruned_loss=0.04049, over 11749.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2577, pruned_loss=0.04022, over 2368404.61 frames. ], batch size: 44, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:27:26,889 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233259.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:27:32,449 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.720e+02 3.240e+02 3.874e+02 5.388e+02, threshold=6.479e+02, percent-clipped=0.0 2023-05-16 18:27:48,694 INFO [finetune.py:992] (1/2) Epoch 11, batch 10950, loss[loss=0.1533, simple_loss=0.2357, pruned_loss=0.03547, over 12166.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2578, pruned_loss=0.04045, over 2363370.24 frames. ], batch size: 29, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:28:11,416 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233320.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:28:16,524 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5892, 2.8030, 4.4387, 4.6970, 2.8890, 2.5276, 2.9123, 2.0214], device='cuda:1'), covar=tensor([0.1586, 0.2929, 0.0499, 0.0368, 0.1300, 0.2340, 0.2807, 0.4341], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0382, 0.0276, 0.0300, 0.0270, 0.0302, 0.0379, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:28:24,716 INFO [finetune.py:992] (1/2) Epoch 11, batch 11000, loss[loss=0.1581, simple_loss=0.2482, pruned_loss=0.03403, over 12120.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2582, pruned_loss=0.04077, over 2359058.68 frames. ], batch size: 33, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:28:29,798 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233346.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:28:37,943 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6670, 4.0506, 3.6806, 4.3591, 3.8365, 2.5952, 3.6890, 2.9308], device='cuda:1'), covar=tensor([0.0858, 0.0878, 0.1272, 0.0447, 0.1348, 0.1758, 0.1107, 0.2760], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0380, 0.0358, 0.0308, 0.0369, 0.0270, 0.0346, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:28:44,639 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.841e+02 3.437e+02 4.261e+02 8.985e+02, threshold=6.874e+02, percent-clipped=4.0 2023-05-16 18:28:49,556 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233374.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:28:59,923 INFO [finetune.py:992] (1/2) Epoch 11, batch 11050, loss[loss=0.232, simple_loss=0.3027, pruned_loss=0.0806, over 8005.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2626, pruned_loss=0.04313, over 2315161.28 frames. ], batch size: 98, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:29:10,154 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4465, 5.2096, 5.3428, 5.4062, 5.0379, 5.1428, 4.8731, 5.3012], device='cuda:1'), covar=tensor([0.0634, 0.0583, 0.0723, 0.0486, 0.1695, 0.1088, 0.0555, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0702, 0.0608, 0.0618, 0.0847, 0.0746, 0.0554, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:29:36,238 INFO [finetune.py:992] (1/2) Epoch 11, batch 11100, loss[loss=0.2432, simple_loss=0.3238, pruned_loss=0.08131, over 10531.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2666, pruned_loss=0.04523, over 2284784.42 frames. ], batch size: 68, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:29:55,776 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 3.180e+02 3.861e+02 4.760e+02 8.899e+02, threshold=7.722e+02, percent-clipped=6.0 2023-05-16 18:30:11,651 INFO [finetune.py:992] (1/2) Epoch 11, batch 11150, loss[loss=0.2944, simple_loss=0.3699, pruned_loss=0.1095, over 6573.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.272, pruned_loss=0.04839, over 2234058.65 frames. ], batch size: 99, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:30:24,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2023-05-16 18:30:47,986 INFO [finetune.py:992] (1/2) Epoch 11, batch 11200, loss[loss=0.2994, simple_loss=0.3643, pruned_loss=0.1172, over 6899.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2798, pruned_loss=0.05383, over 2142614.85 frames. ], batch size: 97, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:31:07,935 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.326e+02 3.488e+02 4.042e+02 5.103e+02 1.073e+03, threshold=8.084e+02, percent-clipped=2.0 2023-05-16 18:31:10,346 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1029, 5.8405, 5.4152, 5.4503, 5.8931, 5.2470, 5.3953, 5.4207], device='cuda:1'), covar=tensor([0.1233, 0.0838, 0.1187, 0.1845, 0.0933, 0.2157, 0.1941, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0488, 0.0387, 0.0437, 0.0457, 0.0431, 0.0391, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:31:20,670 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6543, 3.0787, 3.4469, 3.5855, 3.5069, 3.5886, 3.3374, 2.6935], device='cuda:1'), covar=tensor([0.0086, 0.0170, 0.0159, 0.0094, 0.0075, 0.0145, 0.0107, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0079, 0.0083, 0.0074, 0.0061, 0.0093, 0.0083, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:31:23,958 INFO [finetune.py:992] (1/2) Epoch 11, batch 11250, loss[loss=0.1592, simple_loss=0.2458, pruned_loss=0.03626, over 12363.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2851, pruned_loss=0.05721, over 2097258.87 frames. ], batch size: 30, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:31:42,076 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:31:54,579 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2868, 2.6779, 3.7681, 4.0826, 4.0047, 4.1053, 3.8123, 3.0556], device='cuda:1'), covar=tensor([0.0041, 0.0400, 0.0123, 0.0066, 0.0090, 0.0103, 0.0110, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0121, 0.0102, 0.0077, 0.0100, 0.0115, 0.0094, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:31:55,361 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3150, 4.6763, 2.8792, 2.3612, 4.1279, 2.0889, 4.0694, 3.0004], device='cuda:1'), covar=tensor([0.0616, 0.0375, 0.1062, 0.1710, 0.0227, 0.1667, 0.0364, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0248, 0.0172, 0.0195, 0.0137, 0.0176, 0.0192, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:31:58,987 INFO [finetune.py:992] (1/2) Epoch 11, batch 11300, loss[loss=0.2107, simple_loss=0.3031, pruned_loss=0.05915, over 10306.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2921, pruned_loss=0.06216, over 2027583.89 frames. ], batch size: 68, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:31:59,157 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233639.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:32:04,021 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233646.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 18:32:05,525 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8944, 2.5791, 3.4667, 3.5490, 2.8518, 2.6962, 2.5928, 2.4400], device='cuda:1'), covar=tensor([0.1114, 0.2332, 0.0660, 0.0482, 0.0916, 0.1891, 0.2427, 0.3325], device='cuda:1'), in_proj_covar=tensor([0.0288, 0.0378, 0.0271, 0.0295, 0.0266, 0.0298, 0.0375, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:32:11,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.06 vs. limit=5.0 2023-05-16 18:32:19,032 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.452e+02 3.573e+02 4.203e+02 4.687e+02 1.009e+03, threshold=8.407e+02, percent-clipped=1.0 2023-05-16 18:32:24,013 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233674.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:32:33,997 INFO [finetune.py:992] (1/2) Epoch 11, batch 11350, loss[loss=0.1705, simple_loss=0.2468, pruned_loss=0.04713, over 12291.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2975, pruned_loss=0.06542, over 1974092.04 frames. ], batch size: 28, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:32:37,463 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233694.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:32:41,735 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233700.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 18:32:48,942 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233710.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:32:57,144 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233722.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:33:02,823 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 18:33:09,362 INFO [finetune.py:992] (1/2) Epoch 11, batch 11400, loss[loss=0.1952, simple_loss=0.2822, pruned_loss=0.05414, over 11609.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.3008, pruned_loss=0.0674, over 1943727.56 frames. ], batch size: 48, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:33:28,735 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.455e+02 3.541e+02 4.081e+02 5.202e+02 1.297e+03, threshold=8.162e+02, percent-clipped=1.0 2023-05-16 18:33:31,659 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233771.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:33:43,587 INFO [finetune.py:992] (1/2) Epoch 11, batch 11450, loss[loss=0.2714, simple_loss=0.3323, pruned_loss=0.1052, over 6897.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3047, pruned_loss=0.07065, over 1901329.33 frames. ], batch size: 98, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:34:01,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-05-16 18:34:19,043 INFO [finetune.py:992] (1/2) Epoch 11, batch 11500, loss[loss=0.2594, simple_loss=0.3221, pruned_loss=0.09833, over 6951.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3073, pruned_loss=0.07307, over 1851831.46 frames. ], batch size: 98, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:34:38,556 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.233e+02 3.519e+02 4.029e+02 4.678e+02 7.673e+02, threshold=8.058e+02, percent-clipped=0.0 2023-05-16 18:34:54,103 INFO [finetune.py:992] (1/2) Epoch 11, batch 11550, loss[loss=0.2097, simple_loss=0.3027, pruned_loss=0.05839, over 10310.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3089, pruned_loss=0.07483, over 1829331.95 frames. ], batch size: 68, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:35:13,042 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233915.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:35:28,833 INFO [finetune.py:992] (1/2) Epoch 11, batch 11600, loss[loss=0.247, simple_loss=0.3224, pruned_loss=0.08576, over 7048.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3107, pruned_loss=0.07637, over 1794649.96 frames. ], batch size: 98, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:35:45,758 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233963.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:35:48,494 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.535e+02 3.403e+02 3.814e+02 4.898e+02 8.762e+02, threshold=7.629e+02, percent-clipped=1.0 2023-05-16 18:35:59,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 18:35:59,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 18:36:03,531 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2950, 2.9189, 3.6635, 2.2463, 2.6031, 3.0401, 2.9310, 3.1486], device='cuda:1'), covar=tensor([0.0552, 0.1170, 0.0331, 0.1327, 0.1760, 0.1377, 0.1151, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0227, 0.0239, 0.0176, 0.0230, 0.0284, 0.0215, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:36:05,418 INFO [finetune.py:992] (1/2) Epoch 11, batch 11650, loss[loss=0.2067, simple_loss=0.2983, pruned_loss=0.05751, over 10642.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3111, pruned_loss=0.07753, over 1755072.82 frames. ], batch size: 69, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:36:09,961 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233995.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 18:36:27,897 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7027, 2.2535, 2.8087, 3.7083, 2.2354, 3.7540, 3.7100, 3.7950], device='cuda:1'), covar=tensor([0.0123, 0.1390, 0.0513, 0.0135, 0.1375, 0.0210, 0.0227, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0197, 0.0178, 0.0112, 0.0184, 0.0171, 0.0172, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:36:33,964 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234025.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:36:43,297 INFO [finetune.py:992] (1/2) Epoch 11, batch 11700, loss[loss=0.2261, simple_loss=0.2952, pruned_loss=0.07852, over 6777.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.31, pruned_loss=0.07717, over 1740783.77 frames. ], batch size: 98, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:36:53,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.93 vs. limit=5.0 2023-05-16 18:37:02,246 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234066.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:37:02,774 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.425e+02 3.573e+02 4.120e+02 4.948e+02 7.827e+02, threshold=8.240e+02, percent-clipped=1.0 2023-05-16 18:37:16,574 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234086.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:37:18,378 INFO [finetune.py:992] (1/2) Epoch 11, batch 11750, loss[loss=0.2891, simple_loss=0.3323, pruned_loss=0.1229, over 5963.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3109, pruned_loss=0.0784, over 1715626.97 frames. ], batch size: 99, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:37:35,485 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4860, 3.1749, 3.1227, 3.4420, 2.7047, 3.1587, 2.6713, 2.8949], device='cuda:1'), covar=tensor([0.1332, 0.0757, 0.0763, 0.0501, 0.0965, 0.0711, 0.1452, 0.0599], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0260, 0.0286, 0.0342, 0.0231, 0.0237, 0.0252, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:37:53,759 INFO [finetune.py:992] (1/2) Epoch 11, batch 11800, loss[loss=0.2787, simple_loss=0.3378, pruned_loss=0.1098, over 6423.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3124, pruned_loss=0.08002, over 1698313.32 frames. ], batch size: 98, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:38:11,449 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8963, 3.8879, 3.9199, 3.9635, 3.7716, 3.8200, 3.7138, 3.8679], device='cuda:1'), covar=tensor([0.1107, 0.0636, 0.1168, 0.0728, 0.1578, 0.1056, 0.0592, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0648, 0.0564, 0.0570, 0.0774, 0.0686, 0.0513, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:38:13,318 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.460e+02 3.496e+02 4.148e+02 4.897e+02 9.284e+02, threshold=8.296e+02, percent-clipped=3.0 2023-05-16 18:38:21,621 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6877, 3.2843, 3.3985, 3.6070, 3.5111, 3.6511, 3.5408, 2.5833], device='cuda:1'), covar=tensor([0.0109, 0.0150, 0.0191, 0.0095, 0.0081, 0.0156, 0.0126, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0080, 0.0072, 0.0059, 0.0090, 0.0080, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:38:28,019 INFO [finetune.py:992] (1/2) Epoch 11, batch 11850, loss[loss=0.1962, simple_loss=0.2953, pruned_loss=0.04849, over 10177.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.315, pruned_loss=0.08094, over 1691205.25 frames. ], batch size: 68, lr: 3.86e-03, grad_scale: 8.0 2023-05-16 18:38:38,060 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234202.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:38:58,131 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234231.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:39:03,482 INFO [finetune.py:992] (1/2) Epoch 11, batch 11900, loss[loss=0.2362, simple_loss=0.3158, pruned_loss=0.07834, over 7292.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3137, pruned_loss=0.07935, over 1681720.13 frames. ], batch size: 98, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:39:20,866 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234263.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:39:23,309 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.247e+02 3.250e+02 3.654e+02 4.294e+02 1.197e+03, threshold=7.308e+02, percent-clipped=1.0 2023-05-16 18:39:38,864 INFO [finetune.py:992] (1/2) Epoch 11, batch 11950, loss[loss=0.2131, simple_loss=0.2878, pruned_loss=0.06923, over 6771.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3104, pruned_loss=0.07633, over 1677445.36 frames. ], batch size: 99, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:39:41,128 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234292.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:39:43,165 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234295.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 18:40:13,538 INFO [finetune.py:992] (1/2) Epoch 11, batch 12000, loss[loss=0.176, simple_loss=0.2644, pruned_loss=0.04378, over 7049.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3048, pruned_loss=0.07226, over 1678382.75 frames. ], batch size: 97, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:40:13,538 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 18:40:31,663 INFO [finetune.py:1026] (1/2) Epoch 11, validation: loss=0.2893, simple_loss=0.3643, pruned_loss=0.1072, over 1020973.00 frames. 2023-05-16 18:40:31,664 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 18:40:34,466 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234343.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:40:36,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 18:40:50,680 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234366.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:40:51,167 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.340e+02 2.882e+02 3.344e+02 3.924e+02 1.313e+03, threshold=6.687e+02, percent-clipped=3.0 2023-05-16 18:41:00,812 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234381.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:41:07,054 INFO [finetune.py:992] (1/2) Epoch 11, batch 12050, loss[loss=0.204, simple_loss=0.2844, pruned_loss=0.06177, over 7077.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.3008, pruned_loss=0.0693, over 1681133.60 frames. ], batch size: 99, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:41:23,683 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234414.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:41:39,567 INFO [finetune.py:992] (1/2) Epoch 11, batch 12100, loss[loss=0.2351, simple_loss=0.304, pruned_loss=0.08314, over 6757.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2996, pruned_loss=0.06853, over 1680294.45 frames. ], batch size: 99, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:41:58,106 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.997e+02 3.482e+02 4.109e+02 7.850e+02, threshold=6.964e+02, percent-clipped=2.0 2023-05-16 18:42:12,187 INFO [finetune.py:992] (1/2) Epoch 11, batch 12150, loss[loss=0.2307, simple_loss=0.3083, pruned_loss=0.07651, over 7105.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.3005, pruned_loss=0.06864, over 1688833.14 frames. ], batch size: 98, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:42:23,814 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9608, 2.1446, 2.8068, 2.8552, 2.9450, 2.9667, 2.8769, 2.3346], device='cuda:1'), covar=tensor([0.0088, 0.0403, 0.0189, 0.0082, 0.0119, 0.0110, 0.0126, 0.0409], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0119, 0.0100, 0.0074, 0.0097, 0.0112, 0.0091, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:42:30,313 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0021, 2.2462, 2.3130, 2.2509, 2.1306, 1.8874, 2.2645, 1.6617], device='cuda:1'), covar=tensor([0.0292, 0.0173, 0.0194, 0.0182, 0.0305, 0.0255, 0.0161, 0.0381], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0149, 0.0150, 0.0173, 0.0189, 0.0187, 0.0156, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-05-16 18:42:44,102 INFO [finetune.py:992] (1/2) Epoch 11, batch 12200, loss[loss=0.2291, simple_loss=0.3004, pruned_loss=0.0789, over 7531.00 frames. ], tot_loss[loss=0.221, simple_loss=0.3022, pruned_loss=0.06993, over 1670349.88 frames. ], batch size: 99, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:42:55,925 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234558.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 18:43:01,280 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.292e+02 3.263e+02 3.711e+02 4.394e+02 2.112e+03, threshold=7.422e+02, percent-clipped=3.0 2023-05-16 18:43:28,050 INFO [finetune.py:992] (1/2) Epoch 12, batch 0, loss[loss=0.1627, simple_loss=0.2476, pruned_loss=0.03893, over 12003.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2476, pruned_loss=0.03893, over 12003.00 frames. ], batch size: 28, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:43:28,050 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 18:43:33,214 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5825, 4.3968, 4.1509, 4.4235, 4.3623, 4.3227, 4.5979, 2.0781], device='cuda:1'), covar=tensor([0.0090, 0.0070, 0.0167, 0.0076, 0.0055, 0.0170, 0.0066, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0075, 0.0079, 0.0070, 0.0058, 0.0088, 0.0078, 0.0095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:43:35,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6768, 5.6019, 5.5389, 5.1666, 4.9972, 5.6705, 5.3149, 5.3504], device='cuda:1'), covar=tensor([0.0672, 0.1068, 0.0535, 0.1586, 0.0556, 0.0585, 0.1160, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0510, 0.0474, 0.0579, 0.0384, 0.0647, 0.0695, 0.0520], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 18:43:45,303 INFO [finetune.py:1026] (1/2) Epoch 12, validation: loss=0.2833, simple_loss=0.36, pruned_loss=0.1033, over 1020973.00 frames. 2023-05-16 18:43:45,304 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 18:43:54,314 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234587.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:44:00,836 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0117, 3.0785, 4.4339, 2.2682, 2.5037, 3.2845, 2.8505, 3.3666], device='cuda:1'), covar=tensor([0.0512, 0.1322, 0.0423, 0.1503, 0.2311, 0.1755, 0.1629, 0.1547], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0222, 0.0228, 0.0173, 0.0224, 0.0274, 0.0209, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:44:21,546 INFO [finetune.py:992] (1/2) Epoch 12, batch 50, loss[loss=0.1752, simple_loss=0.2705, pruned_loss=0.03993, over 12156.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2682, pruned_loss=0.04543, over 530750.56 frames. ], batch size: 34, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:44:42,452 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5124, 4.7903, 4.0553, 4.9590, 4.5667, 2.8441, 4.2522, 2.8917], device='cuda:1'), covar=tensor([0.0782, 0.0694, 0.1535, 0.0514, 0.1052, 0.1890, 0.1158, 0.4115], device='cuda:1'), in_proj_covar=tensor([0.0288, 0.0357, 0.0339, 0.0283, 0.0346, 0.0257, 0.0327, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:44:52,092 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.975e+02 3.374e+02 4.225e+02 3.070e+03, threshold=6.748e+02, percent-clipped=4.0 2023-05-16 18:44:53,765 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234669.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:44:57,224 INFO [finetune.py:992] (1/2) Epoch 12, batch 100, loss[loss=0.1504, simple_loss=0.2291, pruned_loss=0.03586, over 11770.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2681, pruned_loss=0.04411, over 931378.35 frames. ], batch size: 26, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:45:03,122 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234681.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:45:17,795 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-05-16 18:45:21,083 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0448, 2.3210, 3.5934, 2.9952, 3.4662, 3.1131, 2.3501, 3.4971], device='cuda:1'), covar=tensor([0.0157, 0.0459, 0.0196, 0.0282, 0.0140, 0.0206, 0.0405, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0194, 0.0174, 0.0176, 0.0200, 0.0155, 0.0184, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:45:33,376 INFO [finetune.py:992] (1/2) Epoch 12, batch 150, loss[loss=0.1764, simple_loss=0.2692, pruned_loss=0.04181, over 12280.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2671, pruned_loss=0.04382, over 1250216.64 frames. ], batch size: 37, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:45:34,889 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234726.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:45:37,668 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234729.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:45:38,508 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234730.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 18:46:04,845 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.565e+02 3.090e+02 3.582e+02 7.594e+02, threshold=6.180e+02, percent-clipped=1.0 2023-05-16 18:46:09,943 INFO [finetune.py:992] (1/2) Epoch 12, batch 200, loss[loss=0.1959, simple_loss=0.2914, pruned_loss=0.05018, over 12142.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2646, pruned_loss=0.04307, over 1496154.84 frames. ], batch size: 38, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:46:15,997 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234782.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:46:19,547 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234787.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:46:20,234 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1849, 2.5024, 3.5800, 4.1932, 3.7005, 4.1282, 3.6078, 2.7938], device='cuda:1'), covar=tensor([0.0045, 0.0374, 0.0125, 0.0039, 0.0089, 0.0078, 0.0112, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0121, 0.0101, 0.0075, 0.0099, 0.0113, 0.0093, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:46:30,395 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6564, 3.6730, 3.2206, 3.2239, 2.9883, 2.7712, 3.7433, 2.4610], device='cuda:1'), covar=tensor([0.0345, 0.0148, 0.0229, 0.0203, 0.0368, 0.0380, 0.0121, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0151, 0.0153, 0.0175, 0.0193, 0.0188, 0.0159, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:46:37,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-16 18:46:46,153 INFO [finetune.py:992] (1/2) Epoch 12, batch 250, loss[loss=0.2136, simple_loss=0.2888, pruned_loss=0.0692, over 8152.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2631, pruned_loss=0.04249, over 1681193.88 frames. ], batch size: 100, lr: 3.85e-03, grad_scale: 4.0 2023-05-16 18:47:00,916 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234843.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:47:11,427 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234858.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:47:17,821 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3673, 4.2902, 4.2521, 4.2450, 4.0263, 4.4599, 4.3438, 4.5184], device='cuda:1'), covar=tensor([0.0200, 0.0164, 0.0214, 0.0403, 0.0675, 0.0350, 0.0193, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0180, 0.0176, 0.0231, 0.0226, 0.0201, 0.0166, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 18:47:18,375 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.787e+02 3.242e+02 3.922e+02 7.095e+02, threshold=6.485e+02, percent-clipped=2.0 2023-05-16 18:47:21,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-05-16 18:47:22,655 INFO [finetune.py:992] (1/2) Epoch 12, batch 300, loss[loss=0.1652, simple_loss=0.2642, pruned_loss=0.03312, over 12317.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2618, pruned_loss=0.0419, over 1836698.40 frames. ], batch size: 34, lr: 3.85e-03, grad_scale: 4.0 2023-05-16 18:47:28,502 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234882.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:47:32,630 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234887.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:47:46,273 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234906.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:47:58,889 INFO [finetune.py:992] (1/2) Epoch 12, batch 350, loss[loss=0.1885, simple_loss=0.2877, pruned_loss=0.04458, over 12162.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2612, pruned_loss=0.04139, over 1958980.88 frames. ], batch size: 36, lr: 3.85e-03, grad_scale: 4.0 2023-05-16 18:48:06,660 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234935.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:48:12,518 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234943.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:48:21,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 18:48:21,569 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4018, 2.5203, 3.7768, 4.4255, 3.8666, 4.3214, 3.7637, 2.9169], device='cuda:1'), covar=tensor([0.0053, 0.0510, 0.0161, 0.0038, 0.0132, 0.0093, 0.0156, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0123, 0.0104, 0.0076, 0.0101, 0.0115, 0.0095, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:48:22,907 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234958.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:48:24,348 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234960.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:48:29,998 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.564e+02 3.005e+02 3.635e+02 9.262e+02, threshold=6.010e+02, percent-clipped=1.0 2023-05-16 18:48:34,368 INFO [finetune.py:992] (1/2) Epoch 12, batch 400, loss[loss=0.1437, simple_loss=0.2306, pruned_loss=0.02846, over 12358.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2617, pruned_loss=0.04193, over 2054874.16 frames. ], batch size: 30, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:49:07,321 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235019.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:49:08,674 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235021.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:49:10,657 INFO [finetune.py:992] (1/2) Epoch 12, batch 450, loss[loss=0.1595, simple_loss=0.2568, pruned_loss=0.03115, over 12272.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2611, pruned_loss=0.0416, over 2131817.18 frames. ], batch size: 37, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:49:11,444 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235025.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 18:49:40,276 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3905, 3.6590, 3.4562, 3.2625, 3.0783, 2.8644, 3.5389, 2.2310], device='cuda:1'), covar=tensor([0.0427, 0.0110, 0.0139, 0.0172, 0.0324, 0.0319, 0.0130, 0.0493], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0155, 0.0157, 0.0179, 0.0198, 0.0192, 0.0164, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:49:42,218 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.777e+02 3.203e+02 3.888e+02 1.136e+03, threshold=6.405e+02, percent-clipped=5.0 2023-05-16 18:49:46,480 INFO [finetune.py:992] (1/2) Epoch 12, batch 500, loss[loss=0.1383, simple_loss=0.2294, pruned_loss=0.02363, over 12349.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2606, pruned_loss=0.04121, over 2189998.89 frames. ], batch size: 30, lr: 3.85e-03, grad_scale: 8.0 2023-05-16 18:49:50,701 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3665, 4.2647, 4.2081, 4.2708, 3.9951, 4.4506, 4.3734, 4.5651], device='cuda:1'), covar=tensor([0.0292, 0.0170, 0.0243, 0.0375, 0.0765, 0.0331, 0.0212, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0186, 0.0182, 0.0238, 0.0234, 0.0208, 0.0171, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 18:49:52,045 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:50:21,688 INFO [finetune.py:992] (1/2) Epoch 12, batch 550, loss[loss=0.2048, simple_loss=0.2903, pruned_loss=0.05966, over 12045.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.26, pruned_loss=0.04088, over 2235250.44 frames. ], batch size: 40, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:50:32,279 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235138.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:50:32,532 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.71 vs. limit=5.0 2023-05-16 18:50:53,592 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.532e+02 3.050e+02 3.463e+02 5.793e+02, threshold=6.100e+02, percent-clipped=0.0 2023-05-16 18:50:57,770 INFO [finetune.py:992] (1/2) Epoch 12, batch 600, loss[loss=0.2027, simple_loss=0.291, pruned_loss=0.05718, over 12062.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2598, pruned_loss=0.04075, over 2262784.56 frames. ], batch size: 37, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:51:34,097 INFO [finetune.py:992] (1/2) Epoch 12, batch 650, loss[loss=0.1883, simple_loss=0.2742, pruned_loss=0.05119, over 12111.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2593, pruned_loss=0.04064, over 2291400.55 frames. ], batch size: 39, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:51:44,322 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235238.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:51:56,532 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235255.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:52:05,688 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.751e+02 3.353e+02 3.897e+02 6.243e+02, threshold=6.706e+02, percent-clipped=1.0 2023-05-16 18:52:10,054 INFO [finetune.py:992] (1/2) Epoch 12, batch 700, loss[loss=0.1956, simple_loss=0.288, pruned_loss=0.05158, over 12021.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2595, pruned_loss=0.04044, over 2305814.81 frames. ], batch size: 42, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:52:38,852 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235314.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:52:40,244 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235316.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:52:40,361 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235316.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:52:45,963 INFO [finetune.py:992] (1/2) Epoch 12, batch 750, loss[loss=0.1837, simple_loss=0.2814, pruned_loss=0.04302, over 12350.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.259, pruned_loss=0.04027, over 2327594.49 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:52:46,759 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235325.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:53:06,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 18:53:07,360 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2506, 2.6242, 3.8112, 3.2960, 3.6041, 3.4158, 2.7310, 3.6399], device='cuda:1'), covar=tensor([0.0136, 0.0357, 0.0161, 0.0234, 0.0205, 0.0157, 0.0322, 0.0161], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0200, 0.0182, 0.0183, 0.0210, 0.0161, 0.0191, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:53:12,220 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235360.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 18:53:15,731 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1125, 4.4580, 2.5927, 2.2576, 3.8756, 2.2908, 3.8129, 2.9201], device='cuda:1'), covar=tensor([0.0766, 0.0424, 0.1217, 0.1764, 0.0349, 0.1603, 0.0480, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0249, 0.0175, 0.0200, 0.0139, 0.0184, 0.0195, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:53:17,625 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.741e+02 3.316e+02 4.028e+02 1.402e+03, threshold=6.632e+02, percent-clipped=1.0 2023-05-16 18:53:19,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0 2023-05-16 18:53:21,207 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235373.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:53:21,856 INFO [finetune.py:992] (1/2) Epoch 12, batch 800, loss[loss=0.1858, simple_loss=0.2761, pruned_loss=0.04775, over 11209.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2594, pruned_loss=0.0404, over 2344589.62 frames. ], batch size: 55, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:53:27,646 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235382.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:53:55,458 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235421.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:53:57,303 INFO [finetune.py:992] (1/2) Epoch 12, batch 850, loss[loss=0.2017, simple_loss=0.2874, pruned_loss=0.058, over 12038.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2593, pruned_loss=0.04056, over 2359343.94 frames. ], batch size: 42, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:54:01,644 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235430.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:54:08,069 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235438.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:54:16,956 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-16 18:54:29,373 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.777e+02 3.296e+02 3.961e+02 7.490e+02, threshold=6.592e+02, percent-clipped=2.0 2023-05-16 18:54:33,578 INFO [finetune.py:992] (1/2) Epoch 12, batch 900, loss[loss=0.1917, simple_loss=0.28, pruned_loss=0.05168, over 11584.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2592, pruned_loss=0.04078, over 2362929.02 frames. ], batch size: 48, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:54:42,372 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235486.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:55:10,327 INFO [finetune.py:992] (1/2) Epoch 12, batch 950, loss[loss=0.1564, simple_loss=0.2571, pruned_loss=0.02789, over 12344.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2593, pruned_loss=0.0405, over 2364865.72 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:55:20,481 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235538.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:55:25,622 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3942, 4.8184, 3.1039, 3.0162, 4.1324, 2.7224, 4.1678, 3.5015], device='cuda:1'), covar=tensor([0.0729, 0.0591, 0.1028, 0.1284, 0.0305, 0.1293, 0.0487, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0247, 0.0173, 0.0198, 0.0138, 0.0181, 0.0193, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 18:55:26,962 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0974, 6.0901, 5.8062, 5.4682, 5.2105, 6.0089, 5.5665, 5.3111], device='cuda:1'), covar=tensor([0.0695, 0.0852, 0.0622, 0.1536, 0.0640, 0.0695, 0.1580, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0612, 0.0550, 0.0513, 0.0631, 0.0412, 0.0699, 0.0762, 0.0565], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 18:55:41,852 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.824e+02 3.256e+02 3.898e+02 8.712e+02, threshold=6.511e+02, percent-clipped=2.0 2023-05-16 18:55:46,189 INFO [finetune.py:992] (1/2) Epoch 12, batch 1000, loss[loss=0.1738, simple_loss=0.2772, pruned_loss=0.03516, over 12366.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.259, pruned_loss=0.04041, over 2369836.58 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:55:54,783 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235586.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:56:13,640 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235611.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:56:15,555 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235614.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:56:17,013 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235616.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:56:22,613 INFO [finetune.py:992] (1/2) Epoch 12, batch 1050, loss[loss=0.1784, simple_loss=0.2701, pruned_loss=0.04335, over 12133.00 frames. ], tot_loss[loss=0.169, simple_loss=0.258, pruned_loss=0.03996, over 2372535.12 frames. ], batch size: 38, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:56:33,574 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2604, 6.1876, 5.9540, 5.4797, 5.2942, 6.1189, 5.7331, 5.4692], device='cuda:1'), covar=tensor([0.0633, 0.0979, 0.0654, 0.1653, 0.0683, 0.0787, 0.1619, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0552, 0.0515, 0.0633, 0.0414, 0.0703, 0.0767, 0.0568], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 18:56:38,802 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1759, 2.1133, 3.5517, 4.1811, 3.8568, 4.0862, 3.6386, 2.8287], device='cuda:1'), covar=tensor([0.0056, 0.0523, 0.0158, 0.0047, 0.0105, 0.0077, 0.0142, 0.0408], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0122, 0.0104, 0.0076, 0.0100, 0.0115, 0.0094, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 18:56:50,320 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235662.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:56:51,704 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235664.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:56:54,505 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 2.852e+02 3.251e+02 3.804e+02 7.432e+02, threshold=6.502e+02, percent-clipped=1.0 2023-05-16 18:56:58,794 INFO [finetune.py:992] (1/2) Epoch 12, batch 1100, loss[loss=0.1839, simple_loss=0.2752, pruned_loss=0.04628, over 12127.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2588, pruned_loss=0.04024, over 2375940.16 frames. ], batch size: 39, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:57:02,663 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235679.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:57:27,426 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2655, 4.5947, 4.0679, 4.9828, 4.5085, 2.9851, 4.2737, 2.9582], device='cuda:1'), covar=tensor([0.0837, 0.0808, 0.1388, 0.0486, 0.1258, 0.1636, 0.0990, 0.3375], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0372, 0.0352, 0.0298, 0.0361, 0.0266, 0.0339, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:57:28,572 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235716.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 18:57:33,981 INFO [finetune.py:992] (1/2) Epoch 12, batch 1150, loss[loss=0.168, simple_loss=0.2547, pruned_loss=0.0407, over 12181.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2585, pruned_loss=0.04024, over 2379951.76 frames. ], batch size: 31, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:57:46,268 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235740.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:57:53,566 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2979, 4.5874, 4.0357, 4.9326, 4.5836, 2.8002, 4.1347, 3.1285], device='cuda:1'), covar=tensor([0.0810, 0.0811, 0.1444, 0.0451, 0.1039, 0.1765, 0.1125, 0.3185], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0375, 0.0354, 0.0301, 0.0364, 0.0268, 0.0342, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 18:58:05,942 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.666e+02 3.094e+02 3.777e+02 6.961e+02, threshold=6.188e+02, percent-clipped=4.0 2023-05-16 18:58:10,223 INFO [finetune.py:992] (1/2) Epoch 12, batch 1200, loss[loss=0.1594, simple_loss=0.2509, pruned_loss=0.03399, over 12153.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.258, pruned_loss=0.04026, over 2380614.13 frames. ], batch size: 36, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:58:22,477 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235791.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:58:46,424 INFO [finetune.py:992] (1/2) Epoch 12, batch 1250, loss[loss=0.1722, simple_loss=0.2667, pruned_loss=0.03879, over 12194.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2582, pruned_loss=0.03999, over 2380413.62 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:59:06,638 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235852.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:59:17,930 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.652e+02 3.056e+02 3.662e+02 6.029e+02, threshold=6.111e+02, percent-clipped=0.0 2023-05-16 18:59:22,126 INFO [finetune.py:992] (1/2) Epoch 12, batch 1300, loss[loss=0.1709, simple_loss=0.2603, pruned_loss=0.04072, over 12284.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2572, pruned_loss=0.03975, over 2381267.51 frames. ], batch size: 37, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 18:59:49,261 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235911.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 18:59:55,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-16 18:59:58,263 INFO [finetune.py:992] (1/2) Epoch 12, batch 1350, loss[loss=0.1475, simple_loss=0.2317, pruned_loss=0.03167, over 12171.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2571, pruned_loss=0.03945, over 2383815.78 frames. ], batch size: 29, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 19:00:01,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-05-16 19:00:24,239 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235959.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:00:30,493 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.594e+02 3.047e+02 3.701e+02 8.074e+02, threshold=6.093e+02, percent-clipped=3.0 2023-05-16 19:00:34,776 INFO [finetune.py:992] (1/2) Epoch 12, batch 1400, loss[loss=0.2057, simple_loss=0.306, pruned_loss=0.05277, over 11857.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2584, pruned_loss=0.0401, over 2371977.86 frames. ], batch size: 49, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 19:00:45,792 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0620, 2.4993, 3.6371, 3.1158, 3.4298, 3.2578, 2.6801, 3.5229], device='cuda:1'), covar=tensor([0.0146, 0.0336, 0.0173, 0.0245, 0.0184, 0.0173, 0.0318, 0.0156], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0201, 0.0184, 0.0185, 0.0212, 0.0162, 0.0192, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:01:07,937 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236016.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:01:09,411 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236018.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:01:13,402 INFO [finetune.py:992] (1/2) Epoch 12, batch 1450, loss[loss=0.1879, simple_loss=0.2769, pruned_loss=0.04948, over 11278.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2581, pruned_loss=0.04, over 2376108.13 frames. ], batch size: 55, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 19:01:22,083 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236035.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:01:43,084 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236064.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:01:45,410 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1997, 2.6066, 3.8305, 3.2990, 3.6401, 3.4224, 2.6370, 3.6753], device='cuda:1'), covar=tensor([0.0132, 0.0324, 0.0123, 0.0219, 0.0143, 0.0159, 0.0358, 0.0125], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0201, 0.0184, 0.0185, 0.0212, 0.0163, 0.0192, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:01:45,846 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.819e+02 3.335e+02 3.913e+02 5.807e+02, threshold=6.670e+02, percent-clipped=0.0 2023-05-16 19:01:50,046 INFO [finetune.py:992] (1/2) Epoch 12, batch 1500, loss[loss=0.1697, simple_loss=0.2571, pruned_loss=0.0411, over 12023.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2593, pruned_loss=0.04056, over 2372094.32 frames. ], batch size: 31, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 19:01:53,806 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236079.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:02:05,595 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7362, 2.8722, 4.7703, 4.8166, 2.9930, 2.6278, 2.9379, 2.2356], device='cuda:1'), covar=tensor([0.1547, 0.2939, 0.0407, 0.0435, 0.1311, 0.2466, 0.2676, 0.3971], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0383, 0.0273, 0.0296, 0.0270, 0.0302, 0.0378, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:02:13,521 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7155, 2.8524, 4.5107, 4.5976, 2.9001, 2.5869, 2.8979, 2.1437], device='cuda:1'), covar=tensor([0.1527, 0.3158, 0.0494, 0.0487, 0.1330, 0.2381, 0.2752, 0.4074], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0383, 0.0273, 0.0296, 0.0270, 0.0302, 0.0378, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:02:14,168 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7164, 2.8669, 3.9037, 4.6575, 4.0709, 4.6338, 3.9657, 3.2990], device='cuda:1'), covar=tensor([0.0028, 0.0357, 0.0122, 0.0032, 0.0083, 0.0068, 0.0125, 0.0315], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0122, 0.0103, 0.0076, 0.0099, 0.0115, 0.0095, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:02:26,072 INFO [finetune.py:992] (1/2) Epoch 12, batch 1550, loss[loss=0.1639, simple_loss=0.2589, pruned_loss=0.03448, over 12359.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.259, pruned_loss=0.04021, over 2385264.60 frames. ], batch size: 36, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 19:02:42,366 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236147.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:02:57,377 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.682e+02 3.126e+02 4.093e+02 1.267e+03, threshold=6.252e+02, percent-clipped=5.0 2023-05-16 19:03:01,720 INFO [finetune.py:992] (1/2) Epoch 12, batch 1600, loss[loss=0.1707, simple_loss=0.2592, pruned_loss=0.04107, over 12038.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2589, pruned_loss=0.03994, over 2378841.29 frames. ], batch size: 40, lr: 3.84e-03, grad_scale: 8.0 2023-05-16 19:03:38,230 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3002, 4.6337, 3.9301, 5.0061, 4.5278, 3.0088, 4.1976, 3.1337], device='cuda:1'), covar=tensor([0.0876, 0.0872, 0.1651, 0.0468, 0.1197, 0.1644, 0.1061, 0.3184], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0375, 0.0354, 0.0303, 0.0364, 0.0269, 0.0343, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:03:38,606 INFO [finetune.py:992] (1/2) Epoch 12, batch 1650, loss[loss=0.1645, simple_loss=0.2529, pruned_loss=0.03811, over 12267.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2585, pruned_loss=0.0395, over 2386501.57 frames. ], batch size: 32, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:03:43,758 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2577, 2.6825, 3.8329, 3.2264, 3.6099, 3.3616, 2.6504, 3.6598], device='cuda:1'), covar=tensor([0.0131, 0.0354, 0.0134, 0.0263, 0.0137, 0.0191, 0.0354, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0202, 0.0186, 0.0187, 0.0214, 0.0164, 0.0194, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:03:52,403 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0542, 2.4595, 3.6151, 2.9662, 3.3604, 3.2157, 2.3658, 3.4578], device='cuda:1'), covar=tensor([0.0149, 0.0391, 0.0178, 0.0299, 0.0168, 0.0192, 0.0414, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0202, 0.0186, 0.0187, 0.0215, 0.0164, 0.0193, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:03:54,632 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3226, 3.8183, 4.0354, 4.2669, 2.6864, 3.5978, 2.6095, 3.9047], device='cuda:1'), covar=tensor([0.1623, 0.0912, 0.0956, 0.0701, 0.1488, 0.0863, 0.1983, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0269, 0.0298, 0.0355, 0.0240, 0.0245, 0.0262, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:04:03,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 19:04:10,982 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.699e+02 3.140e+02 3.673e+02 8.053e+02, threshold=6.280e+02, percent-clipped=3.0 2023-05-16 19:04:14,573 INFO [finetune.py:992] (1/2) Epoch 12, batch 1700, loss[loss=0.1471, simple_loss=0.2243, pruned_loss=0.03492, over 11762.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2578, pruned_loss=0.03948, over 2383227.63 frames. ], batch size: 26, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:04:36,765 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5017, 4.8880, 3.0774, 2.6837, 4.1645, 2.7832, 4.0906, 3.3645], device='cuda:1'), covar=tensor([0.0737, 0.0526, 0.1170, 0.1720, 0.0307, 0.1253, 0.0536, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0254, 0.0177, 0.0200, 0.0141, 0.0184, 0.0198, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:04:50,314 INFO [finetune.py:992] (1/2) Epoch 12, batch 1750, loss[loss=0.1537, simple_loss=0.2414, pruned_loss=0.03294, over 12290.00 frames. ], tot_loss[loss=0.168, simple_loss=0.257, pruned_loss=0.03952, over 2387521.47 frames. ], batch size: 33, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:04:58,231 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236335.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:05:22,390 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.127e+02 2.700e+02 3.222e+02 3.974e+02 5.690e+02, threshold=6.445e+02, percent-clipped=0.0 2023-05-16 19:05:26,036 INFO [finetune.py:992] (1/2) Epoch 12, batch 1800, loss[loss=0.1472, simple_loss=0.2253, pruned_loss=0.03456, over 12170.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2563, pruned_loss=0.03943, over 2395917.68 frames. ], batch size: 29, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:05:26,124 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236374.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:05:33,121 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236383.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:05:46,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-05-16 19:06:02,432 INFO [finetune.py:992] (1/2) Epoch 12, batch 1850, loss[loss=0.1576, simple_loss=0.2415, pruned_loss=0.03687, over 11333.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2562, pruned_loss=0.03949, over 2385192.64 frames. ], batch size: 25, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:06:18,948 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236447.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:06:20,433 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6092, 4.4852, 4.4374, 4.4679, 4.1333, 4.6423, 4.5429, 4.7997], device='cuda:1'), covar=tensor([0.0255, 0.0156, 0.0211, 0.0326, 0.0796, 0.0358, 0.0188, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0193, 0.0186, 0.0246, 0.0240, 0.0216, 0.0175, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 19:06:34,313 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.812e+02 3.208e+02 3.818e+02 6.942e+02, threshold=6.416e+02, percent-clipped=0.0 2023-05-16 19:06:37,923 INFO [finetune.py:992] (1/2) Epoch 12, batch 1900, loss[loss=0.173, simple_loss=0.2715, pruned_loss=0.03723, over 12038.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2574, pruned_loss=0.03971, over 2377359.11 frames. ], batch size: 40, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:06:53,655 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236495.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:07:15,025 INFO [finetune.py:992] (1/2) Epoch 12, batch 1950, loss[loss=0.1808, simple_loss=0.2746, pruned_loss=0.04346, over 12110.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2577, pruned_loss=0.0398, over 2374394.65 frames. ], batch size: 39, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:07:46,999 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.574e+02 3.108e+02 3.850e+02 7.109e+02, threshold=6.217e+02, percent-clipped=4.0 2023-05-16 19:07:50,489 INFO [finetune.py:992] (1/2) Epoch 12, batch 2000, loss[loss=0.1566, simple_loss=0.2484, pruned_loss=0.03236, over 12149.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2578, pruned_loss=0.03964, over 2376273.81 frames. ], batch size: 34, lr: 3.83e-03, grad_scale: 8.0 2023-05-16 19:08:26,760 INFO [finetune.py:992] (1/2) Epoch 12, batch 2050, loss[loss=0.1467, simple_loss=0.2241, pruned_loss=0.03463, over 12331.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.257, pruned_loss=0.03959, over 2381127.36 frames. ], batch size: 30, lr: 3.83e-03, grad_scale: 8.0 2023-05-16 19:08:59,859 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.605e+02 3.083e+02 3.688e+02 8.642e+02, threshold=6.167e+02, percent-clipped=2.0 2023-05-16 19:09:03,388 INFO [finetune.py:992] (1/2) Epoch 12, batch 2100, loss[loss=0.1936, simple_loss=0.2765, pruned_loss=0.05533, over 11592.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2569, pruned_loss=0.03956, over 2377789.16 frames. ], batch size: 48, lr: 3.83e-03, grad_scale: 8.0 2023-05-16 19:09:03,530 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236674.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:09:25,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0419, 4.7338, 4.7834, 4.8767, 4.8030, 4.9227, 4.8809, 2.7522], device='cuda:1'), covar=tensor([0.0123, 0.0066, 0.0094, 0.0072, 0.0042, 0.0104, 0.0074, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0077, 0.0080, 0.0073, 0.0059, 0.0091, 0.0081, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:09:37,531 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236722.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:09:38,985 INFO [finetune.py:992] (1/2) Epoch 12, batch 2150, loss[loss=0.1497, simple_loss=0.2322, pruned_loss=0.03361, over 12292.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2572, pruned_loss=0.03954, over 2380336.87 frames. ], batch size: 28, lr: 3.83e-03, grad_scale: 8.0 2023-05-16 19:09:53,525 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4360, 4.8138, 3.0082, 2.9372, 4.1060, 2.5897, 4.1671, 3.4665], device='cuda:1'), covar=tensor([0.0705, 0.0427, 0.1196, 0.1396, 0.0294, 0.1352, 0.0413, 0.0713], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0254, 0.0176, 0.0200, 0.0142, 0.0183, 0.0197, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:09:55,771 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6564, 2.6993, 4.1324, 4.2762, 2.9038, 2.5271, 2.7961, 2.2244], device='cuda:1'), covar=tensor([0.1557, 0.3055, 0.0547, 0.0459, 0.1242, 0.2535, 0.2687, 0.3989], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0386, 0.0275, 0.0299, 0.0271, 0.0303, 0.0381, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:10:11,906 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.100e+02 2.781e+02 3.230e+02 3.738e+02 9.151e+02, threshold=6.461e+02, percent-clipped=3.0 2023-05-16 19:10:15,467 INFO [finetune.py:992] (1/2) Epoch 12, batch 2200, loss[loss=0.1887, simple_loss=0.2752, pruned_loss=0.05108, over 12357.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2564, pruned_loss=0.03911, over 2385855.33 frames. ], batch size: 38, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:10:26,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-05-16 19:10:46,916 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3512, 5.1282, 5.2811, 5.3422, 4.9648, 4.9728, 4.7247, 5.2073], device='cuda:1'), covar=tensor([0.0627, 0.0692, 0.0833, 0.0563, 0.2021, 0.1281, 0.0551, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0706, 0.0609, 0.0613, 0.0840, 0.0744, 0.0543, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:10:52,386 INFO [finetune.py:992] (1/2) Epoch 12, batch 2250, loss[loss=0.14, simple_loss=0.2225, pruned_loss=0.02871, over 12295.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2561, pruned_loss=0.03914, over 2382170.61 frames. ], batch size: 28, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:11:06,046 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0937, 5.8709, 5.4873, 5.3617, 5.9614, 5.2389, 5.3852, 5.4005], device='cuda:1'), covar=tensor([0.1361, 0.0987, 0.1052, 0.1939, 0.0928, 0.1976, 0.1985, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0496, 0.0391, 0.0443, 0.0460, 0.0434, 0.0394, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:11:25,238 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.624e+02 2.991e+02 3.746e+02 7.102e+02, threshold=5.982e+02, percent-clipped=2.0 2023-05-16 19:11:28,195 INFO [finetune.py:992] (1/2) Epoch 12, batch 2300, loss[loss=0.1578, simple_loss=0.2502, pruned_loss=0.03274, over 12157.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2557, pruned_loss=0.0389, over 2383576.95 frames. ], batch size: 34, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:11:41,359 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236892.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:12:04,928 INFO [finetune.py:992] (1/2) Epoch 12, batch 2350, loss[loss=0.1476, simple_loss=0.2311, pruned_loss=0.03201, over 12026.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2556, pruned_loss=0.03895, over 2378134.62 frames. ], batch size: 28, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:12:26,117 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236953.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:12:37,631 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2134, 2.0766, 2.9964, 3.1117, 3.0168, 3.1284, 2.9839, 2.3642], device='cuda:1'), covar=tensor([0.0072, 0.0414, 0.0165, 0.0079, 0.0144, 0.0107, 0.0142, 0.0350], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0124, 0.0106, 0.0078, 0.0103, 0.0117, 0.0098, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:12:38,061 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.618e+02 2.916e+02 3.600e+02 7.295e+02, threshold=5.831e+02, percent-clipped=3.0 2023-05-16 19:12:40,930 INFO [finetune.py:992] (1/2) Epoch 12, batch 2400, loss[loss=0.1882, simple_loss=0.2792, pruned_loss=0.04856, over 11835.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2561, pruned_loss=0.03914, over 2369267.28 frames. ], batch size: 44, lr: 3.83e-03, grad_scale: 8.0 2023-05-16 19:13:16,073 INFO [finetune.py:992] (1/2) Epoch 12, batch 2450, loss[loss=0.1547, simple_loss=0.2444, pruned_loss=0.03253, over 12043.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2556, pruned_loss=0.03896, over 2377778.42 frames. ], batch size: 31, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:13:21,269 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3599, 2.4886, 3.0388, 4.2765, 2.3611, 4.3270, 4.3494, 4.3850], device='cuda:1'), covar=tensor([0.0113, 0.1214, 0.0525, 0.0134, 0.1248, 0.0202, 0.0128, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0183, 0.0117, 0.0190, 0.0178, 0.0176, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:13:49,918 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.894e+02 3.316e+02 3.902e+02 6.635e+02, threshold=6.633e+02, percent-clipped=2.0 2023-05-16 19:13:52,168 INFO [finetune.py:992] (1/2) Epoch 12, batch 2500, loss[loss=0.1733, simple_loss=0.2715, pruned_loss=0.03757, over 11574.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2572, pruned_loss=0.03977, over 2370406.29 frames. ], batch size: 48, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:14:28,242 INFO [finetune.py:992] (1/2) Epoch 12, batch 2550, loss[loss=0.1893, simple_loss=0.2746, pruned_loss=0.052, over 8156.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.257, pruned_loss=0.03969, over 2366874.01 frames. ], batch size: 98, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:14:41,823 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9705, 5.8057, 5.3879, 5.3112, 5.9191, 5.1096, 5.3512, 5.2773], device='cuda:1'), covar=tensor([0.1664, 0.1101, 0.1383, 0.2022, 0.0891, 0.2617, 0.1958, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0500, 0.0397, 0.0446, 0.0466, 0.0441, 0.0400, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:15:01,498 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.724e+02 3.089e+02 3.882e+02 6.578e+02, threshold=6.177e+02, percent-clipped=0.0 2023-05-16 19:15:03,629 INFO [finetune.py:992] (1/2) Epoch 12, batch 2600, loss[loss=0.1473, simple_loss=0.2361, pruned_loss=0.02929, over 12138.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2571, pruned_loss=0.03956, over 2372481.70 frames. ], batch size: 30, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:15:05,854 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2280, 6.0125, 5.5776, 5.5138, 6.1045, 5.2279, 5.5761, 5.5147], device='cuda:1'), covar=tensor([0.1519, 0.0853, 0.1039, 0.1916, 0.0906, 0.2473, 0.1754, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0498, 0.0394, 0.0446, 0.0464, 0.0439, 0.0398, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:15:21,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 19:15:30,301 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2846, 2.3650, 3.0601, 4.2232, 2.3348, 4.3118, 4.2742, 4.3559], device='cuda:1'), covar=tensor([0.0144, 0.1286, 0.0520, 0.0139, 0.1314, 0.0198, 0.0157, 0.0096], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0201, 0.0182, 0.0117, 0.0190, 0.0178, 0.0176, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:15:39,931 INFO [finetune.py:992] (1/2) Epoch 12, batch 2650, loss[loss=0.1538, simple_loss=0.2355, pruned_loss=0.036, over 12342.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2576, pruned_loss=0.03973, over 2373209.91 frames. ], batch size: 31, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:15:57,488 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237248.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:16:13,839 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.563e+02 3.095e+02 3.616e+02 7.529e+02, threshold=6.190e+02, percent-clipped=2.0 2023-05-16 19:16:15,980 INFO [finetune.py:992] (1/2) Epoch 12, batch 2700, loss[loss=0.144, simple_loss=0.2302, pruned_loss=0.02888, over 11987.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2557, pruned_loss=0.03917, over 2382609.66 frames. ], batch size: 28, lr: 3.83e-03, grad_scale: 4.0 2023-05-16 19:16:25,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-16 19:16:48,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-16 19:16:51,222 INFO [finetune.py:992] (1/2) Epoch 12, batch 2750, loss[loss=0.1843, simple_loss=0.2716, pruned_loss=0.04854, over 12358.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2562, pruned_loss=0.03918, over 2380788.04 frames. ], batch size: 38, lr: 3.82e-03, grad_scale: 4.0 2023-05-16 19:17:24,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7002, 3.0291, 3.8982, 4.5710, 4.0009, 4.6095, 4.0640, 3.3639], device='cuda:1'), covar=tensor([0.0033, 0.0316, 0.0141, 0.0054, 0.0126, 0.0079, 0.0121, 0.0323], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0122, 0.0105, 0.0077, 0.0102, 0.0115, 0.0095, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:17:25,097 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.699e+02 3.189e+02 3.843e+02 1.550e+03, threshold=6.378e+02, percent-clipped=4.0 2023-05-16 19:17:27,224 INFO [finetune.py:992] (1/2) Epoch 12, batch 2800, loss[loss=0.1831, simple_loss=0.2744, pruned_loss=0.04586, over 12077.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2561, pruned_loss=0.0394, over 2373304.56 frames. ], batch size: 42, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:18:03,962 INFO [finetune.py:992] (1/2) Epoch 12, batch 2850, loss[loss=0.1491, simple_loss=0.2352, pruned_loss=0.03147, over 12121.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2555, pruned_loss=0.0392, over 2373599.49 frames. ], batch size: 30, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:18:37,460 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.825e+02 2.552e+02 3.036e+02 3.726e+02 9.226e+02, threshold=6.072e+02, percent-clipped=2.0 2023-05-16 19:18:39,528 INFO [finetune.py:992] (1/2) Epoch 12, batch 2900, loss[loss=0.1817, simple_loss=0.2772, pruned_loss=0.04311, over 11224.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2559, pruned_loss=0.03947, over 2368506.15 frames. ], batch size: 55, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:18:52,458 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237491.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:19:15,815 INFO [finetune.py:992] (1/2) Epoch 12, batch 2950, loss[loss=0.1422, simple_loss=0.2299, pruned_loss=0.02727, over 12175.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2556, pruned_loss=0.03898, over 2375113.26 frames. ], batch size: 29, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:19:16,075 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6677, 3.3412, 5.0054, 2.5467, 2.6230, 3.7419, 3.1701, 3.7012], device='cuda:1'), covar=tensor([0.0395, 0.1099, 0.0316, 0.1157, 0.2049, 0.1474, 0.1352, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0236, 0.0251, 0.0185, 0.0239, 0.0296, 0.0223, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:19:23,905 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2297, 4.9165, 5.1524, 5.0623, 4.8776, 5.1908, 5.0168, 2.8379], device='cuda:1'), covar=tensor([0.0082, 0.0063, 0.0064, 0.0062, 0.0047, 0.0084, 0.0072, 0.0669], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0078, 0.0081, 0.0073, 0.0060, 0.0091, 0.0082, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:19:23,988 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9416, 3.6086, 5.2977, 2.7430, 2.7570, 3.9190, 3.3863, 3.9282], device='cuda:1'), covar=tensor([0.0377, 0.1043, 0.0319, 0.1172, 0.2042, 0.1507, 0.1275, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0236, 0.0251, 0.0185, 0.0240, 0.0297, 0.0224, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:19:30,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-05-16 19:19:33,685 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237548.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 19:19:36,679 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237552.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:19:44,420 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237563.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:19:49,810 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.638e+02 3.009e+02 3.582e+02 2.121e+03, threshold=6.018e+02, percent-clipped=2.0 2023-05-16 19:19:51,957 INFO [finetune.py:992] (1/2) Epoch 12, batch 3000, loss[loss=0.1462, simple_loss=0.2342, pruned_loss=0.02917, over 12141.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2565, pruned_loss=0.03943, over 2368023.51 frames. ], batch size: 30, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:19:51,957 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 19:20:10,386 INFO [finetune.py:1026] (1/2) Epoch 12, validation: loss=0.3162, simple_loss=0.3931, pruned_loss=0.1196, over 1020973.00 frames. 2023-05-16 19:20:10,387 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 19:20:26,416 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237596.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:20:47,085 INFO [finetune.py:992] (1/2) Epoch 12, batch 3050, loss[loss=0.1258, simple_loss=0.2093, pruned_loss=0.02111, over 12010.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2559, pruned_loss=0.03932, over 2363070.67 frames. ], batch size: 28, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:20:47,290 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237624.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:21:20,224 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.634e+02 3.161e+02 3.725e+02 7.420e+02, threshold=6.322e+02, percent-clipped=3.0 2023-05-16 19:21:22,470 INFO [finetune.py:992] (1/2) Epoch 12, batch 3100, loss[loss=0.1461, simple_loss=0.2323, pruned_loss=0.02997, over 12295.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2563, pruned_loss=0.03942, over 2371038.22 frames. ], batch size: 28, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:21:24,255 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3776, 4.7426, 4.0511, 5.0769, 4.5860, 2.9668, 4.3091, 3.2333], device='cuda:1'), covar=tensor([0.0816, 0.0731, 0.1471, 0.0437, 0.1189, 0.1650, 0.1097, 0.2962], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0376, 0.0356, 0.0306, 0.0366, 0.0269, 0.0343, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:21:49,036 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3710, 2.8185, 3.9731, 3.3403, 3.6812, 3.4622, 2.7955, 3.7405], device='cuda:1'), covar=tensor([0.0144, 0.0367, 0.0130, 0.0221, 0.0166, 0.0181, 0.0351, 0.0138], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0207, 0.0190, 0.0190, 0.0219, 0.0166, 0.0197, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:21:58,078 INFO [finetune.py:992] (1/2) Epoch 12, batch 3150, loss[loss=0.1761, simple_loss=0.2616, pruned_loss=0.04537, over 12124.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.256, pruned_loss=0.03941, over 2369195.98 frames. ], batch size: 39, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:22:32,922 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.636e+02 3.028e+02 3.612e+02 1.044e+03, threshold=6.055e+02, percent-clipped=1.0 2023-05-16 19:22:35,072 INFO [finetune.py:992] (1/2) Epoch 12, batch 3200, loss[loss=0.144, simple_loss=0.2337, pruned_loss=0.02716, over 12200.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2564, pruned_loss=0.03964, over 2371418.34 frames. ], batch size: 29, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:22:46,729 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237790.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:23:10,854 INFO [finetune.py:992] (1/2) Epoch 12, batch 3250, loss[loss=0.1819, simple_loss=0.2672, pruned_loss=0.04832, over 11776.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2563, pruned_loss=0.0393, over 2380769.67 frames. ], batch size: 44, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:23:26,795 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237847.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:23:29,685 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237851.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:23:31,068 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4885, 5.3023, 5.4246, 5.4912, 5.1327, 5.1473, 4.8762, 5.4590], device='cuda:1'), covar=tensor([0.0743, 0.0644, 0.0815, 0.0574, 0.1840, 0.1247, 0.0569, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0706, 0.0613, 0.0623, 0.0844, 0.0743, 0.0547, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:23:43,412 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.882e+02 3.204e+02 3.746e+02 1.309e+03, threshold=6.408e+02, percent-clipped=2.0 2023-05-16 19:23:45,419 INFO [finetune.py:992] (1/2) Epoch 12, batch 3300, loss[loss=0.1637, simple_loss=0.2522, pruned_loss=0.0376, over 12047.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2569, pruned_loss=0.03957, over 2372857.63 frames. ], batch size: 42, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:24:09,775 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5371, 5.1325, 5.5224, 4.8449, 5.1153, 4.9009, 5.5700, 5.2142], device='cuda:1'), covar=tensor([0.0224, 0.0357, 0.0228, 0.0246, 0.0354, 0.0325, 0.0184, 0.0207], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0263, 0.0288, 0.0261, 0.0262, 0.0262, 0.0235, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:24:13,334 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237912.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:24:17,867 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 19:24:19,008 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237919.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:24:22,406 INFO [finetune.py:992] (1/2) Epoch 12, batch 3350, loss[loss=0.1622, simple_loss=0.2563, pruned_loss=0.03404, over 12303.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2557, pruned_loss=0.03909, over 2378791.84 frames. ], batch size: 34, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:24:34,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 19:24:55,995 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 2.823e+02 3.216e+02 3.693e+02 6.256e+02, threshold=6.432e+02, percent-clipped=0.0 2023-05-16 19:24:57,692 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237973.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:24:58,222 INFO [finetune.py:992] (1/2) Epoch 12, batch 3400, loss[loss=0.1622, simple_loss=0.2456, pruned_loss=0.03941, over 12343.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2552, pruned_loss=0.03898, over 2380856.12 frames. ], batch size: 30, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:25:27,945 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6343, 2.7901, 3.2820, 4.5539, 2.3258, 4.4990, 4.5423, 4.5873], device='cuda:1'), covar=tensor([0.0141, 0.1107, 0.0441, 0.0132, 0.1341, 0.0243, 0.0168, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0205, 0.0185, 0.0120, 0.0191, 0.0181, 0.0179, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:25:37,061 INFO [finetune.py:992] (1/2) Epoch 12, batch 3450, loss[loss=0.165, simple_loss=0.2572, pruned_loss=0.03638, over 12348.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2566, pruned_loss=0.03941, over 2383665.40 frames. ], batch size: 35, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:26:06,955 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238064.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:26:11,656 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.658e+02 2.911e+02 3.575e+02 8.675e+02, threshold=5.823e+02, percent-clipped=1.0 2023-05-16 19:26:13,815 INFO [finetune.py:992] (1/2) Epoch 12, batch 3500, loss[loss=0.2032, simple_loss=0.2765, pruned_loss=0.06491, over 8206.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2562, pruned_loss=0.03931, over 2374380.45 frames. ], batch size: 99, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:26:14,009 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238074.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:26:24,000 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1903, 2.7243, 3.5109, 4.0711, 3.6515, 4.0216, 3.7601, 2.7707], device='cuda:1'), covar=tensor([0.0044, 0.0311, 0.0151, 0.0060, 0.0127, 0.0078, 0.0116, 0.0391], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0122, 0.0105, 0.0077, 0.0101, 0.0115, 0.0096, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:26:43,790 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2435, 3.2740, 4.6571, 2.4340, 2.6420, 3.4641, 3.1398, 3.4893], device='cuda:1'), covar=tensor([0.0539, 0.1143, 0.0376, 0.1332, 0.2042, 0.1605, 0.1382, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0238, 0.0252, 0.0187, 0.0241, 0.0299, 0.0226, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:26:49,059 INFO [finetune.py:992] (1/2) Epoch 12, batch 3550, loss[loss=0.1784, simple_loss=0.2723, pruned_loss=0.04223, over 12012.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.257, pruned_loss=0.03976, over 2370593.36 frames. ], batch size: 40, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:26:50,039 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238125.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:26:50,720 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238126.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:26:55,659 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2170, 4.8298, 5.1957, 4.5506, 4.8498, 4.5602, 5.2567, 4.8776], device='cuda:1'), covar=tensor([0.0263, 0.0385, 0.0287, 0.0291, 0.0357, 0.0407, 0.0221, 0.0351], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0262, 0.0286, 0.0260, 0.0260, 0.0261, 0.0234, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:26:57,216 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238135.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:27:04,905 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238146.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:27:05,693 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238147.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:27:07,200 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1591, 6.0199, 5.8672, 5.4393, 5.3507, 6.0329, 5.5974, 5.4124], device='cuda:1'), covar=tensor([0.0744, 0.1337, 0.0832, 0.1693, 0.0647, 0.0770, 0.1680, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0564, 0.0524, 0.0642, 0.0418, 0.0720, 0.0793, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 19:27:22,612 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.776e+02 3.155e+02 3.758e+02 5.495e+02, threshold=6.310e+02, percent-clipped=0.0 2023-05-16 19:27:25,398 INFO [finetune.py:992] (1/2) Epoch 12, batch 3600, loss[loss=0.1482, simple_loss=0.2346, pruned_loss=0.03088, over 12341.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2561, pruned_loss=0.0392, over 2379609.11 frames. ], batch size: 31, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:27:34,744 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238187.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:27:40,288 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238195.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:27:58,671 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238219.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:28:02,058 INFO [finetune.py:992] (1/2) Epoch 12, batch 3650, loss[loss=0.1879, simple_loss=0.2802, pruned_loss=0.04776, over 12152.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2562, pruned_loss=0.03905, over 2387514.63 frames. ], batch size: 36, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:28:27,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-05-16 19:28:32,680 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238267.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:28:33,370 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238268.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:28:35,470 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.557e+02 3.166e+02 3.787e+02 8.400e+02, threshold=6.332e+02, percent-clipped=1.0 2023-05-16 19:28:37,674 INFO [finetune.py:992] (1/2) Epoch 12, batch 3700, loss[loss=0.183, simple_loss=0.2681, pruned_loss=0.04894, over 12123.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2556, pruned_loss=0.03908, over 2379756.77 frames. ], batch size: 39, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:29:13,319 INFO [finetune.py:992] (1/2) Epoch 12, batch 3750, loss[loss=0.187, simple_loss=0.272, pruned_loss=0.05101, over 12360.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2557, pruned_loss=0.03891, over 2382565.87 frames. ], batch size: 38, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:29:29,317 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0786, 2.5758, 3.6374, 3.1882, 3.4842, 3.2908, 2.6500, 3.5034], device='cuda:1'), covar=tensor([0.0153, 0.0367, 0.0176, 0.0234, 0.0161, 0.0184, 0.0372, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0207, 0.0191, 0.0190, 0.0219, 0.0168, 0.0198, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:29:31,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-05-16 19:29:47,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.858e+02 3.226e+02 3.875e+02 2.077e+03, threshold=6.452e+02, percent-clipped=4.0 2023-05-16 19:29:49,472 INFO [finetune.py:992] (1/2) Epoch 12, batch 3800, loss[loss=0.1715, simple_loss=0.2625, pruned_loss=0.0403, over 12188.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2563, pruned_loss=0.0392, over 2378373.37 frames. ], batch size: 35, lr: 3.82e-03, grad_scale: 8.0 2023-05-16 19:30:20,171 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5511, 3.3623, 4.9872, 2.7540, 2.6928, 3.6815, 3.0991, 3.8104], device='cuda:1'), covar=tensor([0.0458, 0.1105, 0.0264, 0.1132, 0.2016, 0.1529, 0.1447, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0236, 0.0249, 0.0185, 0.0239, 0.0297, 0.0223, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:30:22,753 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238420.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:30:22,839 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238420.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:30:25,409 INFO [finetune.py:992] (1/2) Epoch 12, batch 3850, loss[loss=0.1781, simple_loss=0.2664, pruned_loss=0.04489, over 12076.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2558, pruned_loss=0.03926, over 2375056.63 frames. ], batch size: 42, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:30:29,887 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238430.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:30:41,143 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238446.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:30:44,618 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9364, 4.5868, 4.1669, 4.2012, 4.6770, 4.0278, 4.1966, 4.0412], device='cuda:1'), covar=tensor([0.1589, 0.1162, 0.1409, 0.2116, 0.1102, 0.2325, 0.1973, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0492, 0.0388, 0.0435, 0.0458, 0.0431, 0.0393, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:30:53,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-05-16 19:30:58,815 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.697e+02 3.146e+02 3.990e+02 1.679e+03, threshold=6.292e+02, percent-clipped=3.0 2023-05-16 19:31:00,954 INFO [finetune.py:992] (1/2) Epoch 12, batch 3900, loss[loss=0.153, simple_loss=0.2398, pruned_loss=0.03307, over 12345.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2561, pruned_loss=0.03933, over 2365662.74 frames. ], batch size: 31, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:31:06,240 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238481.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 19:31:06,803 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238482.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:31:11,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 19:31:16,252 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238494.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:31:37,283 INFO [finetune.py:992] (1/2) Epoch 12, batch 3950, loss[loss=0.1868, simple_loss=0.2773, pruned_loss=0.04817, over 11263.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2556, pruned_loss=0.03921, over 2364165.14 frames. ], batch size: 55, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:31:58,661 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9467, 4.8496, 4.7854, 4.8876, 4.4228, 5.0276, 4.8858, 5.1562], device='cuda:1'), covar=tensor([0.0248, 0.0158, 0.0195, 0.0353, 0.0852, 0.0311, 0.0176, 0.0184], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0200, 0.0192, 0.0252, 0.0249, 0.0220, 0.0180, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 19:32:07,227 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0775, 6.0340, 5.7811, 5.2818, 5.2282, 5.9402, 5.5891, 5.3352], device='cuda:1'), covar=tensor([0.0684, 0.0939, 0.0747, 0.1693, 0.0707, 0.0724, 0.1557, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0561, 0.0523, 0.0639, 0.0418, 0.0716, 0.0789, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 19:32:08,733 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238568.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:32:10,602 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.647e+02 3.036e+02 3.542e+02 7.006e+02, threshold=6.071e+02, percent-clipped=2.0 2023-05-16 19:32:12,716 INFO [finetune.py:992] (1/2) Epoch 12, batch 4000, loss[loss=0.1696, simple_loss=0.2593, pruned_loss=0.03999, over 12153.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2559, pruned_loss=0.03899, over 2376186.44 frames. ], batch size: 36, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:32:20,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 19:32:43,476 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238616.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:32:47,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-16 19:32:49,127 INFO [finetune.py:992] (1/2) Epoch 12, batch 4050, loss[loss=0.1542, simple_loss=0.2271, pruned_loss=0.04062, over 11916.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2563, pruned_loss=0.03943, over 2364021.96 frames. ], batch size: 26, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:33:07,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 19:33:07,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 19:33:23,039 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.680e+02 3.070e+02 3.796e+02 8.784e+02, threshold=6.139e+02, percent-clipped=3.0 2023-05-16 19:33:25,218 INFO [finetune.py:992] (1/2) Epoch 12, batch 4100, loss[loss=0.1516, simple_loss=0.2533, pruned_loss=0.02497, over 12177.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2556, pruned_loss=0.03919, over 2365058.03 frames. ], batch size: 35, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:33:30,348 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238681.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:33:35,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 19:33:58,399 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238720.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:34:01,071 INFO [finetune.py:992] (1/2) Epoch 12, batch 4150, loss[loss=0.1798, simple_loss=0.2699, pruned_loss=0.04492, over 12262.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.256, pruned_loss=0.03938, over 2359843.87 frames. ], batch size: 37, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:34:05,591 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238730.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:34:12,884 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5514, 4.3133, 4.2936, 4.5490, 4.3897, 4.5261, 4.3643, 2.1264], device='cuda:1'), covar=tensor([0.0203, 0.0116, 0.0173, 0.0112, 0.0097, 0.0181, 0.0144, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0079, 0.0082, 0.0074, 0.0061, 0.0093, 0.0083, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:34:14,364 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238742.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:34:19,564 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 19:34:33,453 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238768.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:34:35,366 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.514e+02 2.973e+02 3.546e+02 5.463e+02, threshold=5.946e+02, percent-clipped=0.0 2023-05-16 19:34:37,468 INFO [finetune.py:992] (1/2) Epoch 12, batch 4200, loss[loss=0.1676, simple_loss=0.2678, pruned_loss=0.03372, over 12157.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2566, pruned_loss=0.03968, over 2362270.57 frames. ], batch size: 36, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:34:39,003 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238776.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 19:34:40,411 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238778.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:34:43,345 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238782.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:35:12,601 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4961, 3.5548, 3.2285, 3.1589, 2.8551, 2.7763, 3.5925, 2.3775], device='cuda:1'), covar=tensor([0.0392, 0.0147, 0.0197, 0.0176, 0.0387, 0.0354, 0.0135, 0.0463], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0163, 0.0167, 0.0192, 0.0208, 0.0203, 0.0176, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:35:13,784 INFO [finetune.py:992] (1/2) Epoch 12, batch 4250, loss[loss=0.1752, simple_loss=0.268, pruned_loss=0.04125, over 12147.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2571, pruned_loss=0.03965, over 2365139.98 frames. ], batch size: 34, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:35:18,047 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238830.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:35:24,575 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7611, 3.7561, 3.3629, 3.3175, 3.0536, 3.0466, 3.7496, 2.4894], device='cuda:1'), covar=tensor([0.0370, 0.0158, 0.0203, 0.0187, 0.0389, 0.0340, 0.0135, 0.0451], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0163, 0.0166, 0.0191, 0.0207, 0.0202, 0.0175, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:35:28,094 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5941, 2.4359, 3.9120, 4.5852, 4.1238, 4.4700, 4.0888, 3.1721], device='cuda:1'), covar=tensor([0.0061, 0.0492, 0.0142, 0.0044, 0.0114, 0.0093, 0.0122, 0.0360], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0122, 0.0106, 0.0077, 0.0101, 0.0115, 0.0096, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:35:47,022 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 2.596e+02 3.163e+02 3.758e+02 5.764e+02, threshold=6.325e+02, percent-clipped=0.0 2023-05-16 19:35:49,186 INFO [finetune.py:992] (1/2) Epoch 12, batch 4300, loss[loss=0.2136, simple_loss=0.2916, pruned_loss=0.06779, over 8292.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2576, pruned_loss=0.03977, over 2369109.87 frames. ], batch size: 98, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:36:12,994 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1687, 5.0557, 4.9732, 4.9816, 4.7143, 5.0894, 5.1540, 5.3209], device='cuda:1'), covar=tensor([0.0213, 0.0161, 0.0200, 0.0336, 0.0711, 0.0253, 0.0156, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0200, 0.0193, 0.0252, 0.0248, 0.0220, 0.0180, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 19:36:24,943 INFO [finetune.py:992] (1/2) Epoch 12, batch 4350, loss[loss=0.1401, simple_loss=0.2297, pruned_loss=0.02526, over 12087.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2569, pruned_loss=0.03964, over 2368131.83 frames. ], batch size: 32, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:36:32,110 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3519, 2.8999, 2.8072, 2.7230, 2.5522, 2.4672, 2.9765, 1.9769], device='cuda:1'), covar=tensor([0.0364, 0.0176, 0.0211, 0.0202, 0.0371, 0.0314, 0.0152, 0.0480], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0162, 0.0167, 0.0191, 0.0207, 0.0202, 0.0175, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:36:32,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 19:36:33,475 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238936.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:36:58,688 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.500e+02 2.722e+02 3.105e+02 3.870e+02 7.844e+02, threshold=6.211e+02, percent-clipped=1.0 2023-05-16 19:37:00,885 INFO [finetune.py:992] (1/2) Epoch 12, batch 4400, loss[loss=0.1934, simple_loss=0.278, pruned_loss=0.05438, over 12104.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2565, pruned_loss=0.03966, over 2360081.29 frames. ], batch size: 42, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:37:17,366 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238997.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:37:36,407 INFO [finetune.py:992] (1/2) Epoch 12, batch 4450, loss[loss=0.1738, simple_loss=0.2639, pruned_loss=0.04181, over 12369.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2558, pruned_loss=0.03946, over 2371037.90 frames. ], batch size: 38, lr: 3.81e-03, grad_scale: 16.0 2023-05-16 19:37:45,629 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239037.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:38:10,407 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.792e+02 3.466e+02 4.003e+02 6.008e+02, threshold=6.932e+02, percent-clipped=0.0 2023-05-16 19:38:12,510 INFO [finetune.py:992] (1/2) Epoch 12, batch 4500, loss[loss=0.1607, simple_loss=0.2581, pruned_loss=0.03162, over 12347.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2561, pruned_loss=0.03961, over 2366647.42 frames. ], batch size: 36, lr: 3.81e-03, grad_scale: 16.0 2023-05-16 19:38:14,098 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239076.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:38:48,047 INFO [finetune.py:992] (1/2) Epoch 12, batch 4550, loss[loss=0.173, simple_loss=0.2652, pruned_loss=0.0404, over 11675.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2569, pruned_loss=0.03996, over 2367268.80 frames. ], batch size: 48, lr: 3.81e-03, grad_scale: 16.0 2023-05-16 19:38:48,113 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239124.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:39:07,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-16 19:39:21,057 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.710e+02 3.133e+02 3.780e+02 8.254e+02, threshold=6.265e+02, percent-clipped=2.0 2023-05-16 19:39:23,231 INFO [finetune.py:992] (1/2) Epoch 12, batch 4600, loss[loss=0.1631, simple_loss=0.2553, pruned_loss=0.03546, over 12093.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2567, pruned_loss=0.03983, over 2378874.74 frames. ], batch size: 32, lr: 3.81e-03, grad_scale: 16.0 2023-05-16 19:39:23,334 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1658, 5.9756, 5.6113, 5.4583, 6.0753, 5.4819, 5.5392, 5.5753], device='cuda:1'), covar=tensor([0.1617, 0.0985, 0.1089, 0.2278, 0.0943, 0.2047, 0.1817, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0499, 0.0395, 0.0440, 0.0470, 0.0436, 0.0399, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:39:59,407 INFO [finetune.py:992] (1/2) Epoch 12, batch 4650, loss[loss=0.1381, simple_loss=0.2278, pruned_loss=0.02419, over 12115.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2567, pruned_loss=0.04004, over 2364328.25 frames. ], batch size: 30, lr: 3.81e-03, grad_scale: 16.0 2023-05-16 19:40:21,617 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 19:40:33,917 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.878e+02 2.619e+02 3.215e+02 4.293e+02 1.000e+03, threshold=6.429e+02, percent-clipped=5.0 2023-05-16 19:40:35,240 INFO [finetune.py:992] (1/2) Epoch 12, batch 4700, loss[loss=0.1535, simple_loss=0.2465, pruned_loss=0.03021, over 12147.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2576, pruned_loss=0.0402, over 2365862.79 frames. ], batch size: 36, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:40:48,416 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239292.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 19:41:11,123 INFO [finetune.py:992] (1/2) Epoch 12, batch 4750, loss[loss=0.1961, simple_loss=0.2889, pruned_loss=0.05172, over 12078.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2585, pruned_loss=0.04053, over 2347024.83 frames. ], batch size: 42, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:41:21,106 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239337.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:41:38,301 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0464, 4.5050, 3.8441, 4.7109, 4.1753, 2.8253, 4.0284, 3.0060], device='cuda:1'), covar=tensor([0.0916, 0.0667, 0.1481, 0.0461, 0.1374, 0.1710, 0.1077, 0.3077], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0375, 0.0356, 0.0306, 0.0364, 0.0268, 0.0342, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:41:46,345 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.662e+02 3.103e+02 3.697e+02 5.990e+02, threshold=6.207e+02, percent-clipped=0.0 2023-05-16 19:41:47,710 INFO [finetune.py:992] (1/2) Epoch 12, batch 4800, loss[loss=0.1687, simple_loss=0.2545, pruned_loss=0.04149, over 12019.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2576, pruned_loss=0.04025, over 2354434.81 frames. ], batch size: 31, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:41:55,608 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239385.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:42:23,428 INFO [finetune.py:992] (1/2) Epoch 12, batch 4850, loss[loss=0.1707, simple_loss=0.2639, pruned_loss=0.03874, over 11610.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2574, pruned_loss=0.03992, over 2353645.07 frames. ], batch size: 48, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:42:29,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.21 vs. limit=5.0 2023-05-16 19:42:46,811 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6772, 2.7030, 3.3010, 4.4936, 2.4966, 4.5835, 4.5767, 4.7174], device='cuda:1'), covar=tensor([0.0122, 0.1252, 0.0493, 0.0168, 0.1378, 0.0200, 0.0141, 0.0090], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0207, 0.0186, 0.0121, 0.0194, 0.0182, 0.0180, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:42:57,297 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.566e+02 2.984e+02 3.406e+02 6.973e+02, threshold=5.968e+02, percent-clipped=1.0 2023-05-16 19:42:58,750 INFO [finetune.py:992] (1/2) Epoch 12, batch 4900, loss[loss=0.1783, simple_loss=0.2591, pruned_loss=0.04873, over 12375.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2566, pruned_loss=0.03921, over 2359298.54 frames. ], batch size: 38, lr: 3.81e-03, grad_scale: 8.0 2023-05-16 19:43:02,396 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1051, 3.7012, 3.9603, 4.3520, 3.0089, 3.8106, 2.3426, 4.0560], device='cuda:1'), covar=tensor([0.1779, 0.0972, 0.0998, 0.0790, 0.1190, 0.0714, 0.2158, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0267, 0.0298, 0.0356, 0.0239, 0.0244, 0.0262, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:43:35,465 INFO [finetune.py:992] (1/2) Epoch 12, batch 4950, loss[loss=0.1597, simple_loss=0.2497, pruned_loss=0.03483, over 12308.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2562, pruned_loss=0.03924, over 2361425.12 frames. ], batch size: 34, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:44:09,763 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 2.810e+02 3.291e+02 4.221e+02 1.405e+03, threshold=6.582e+02, percent-clipped=5.0 2023-05-16 19:44:11,187 INFO [finetune.py:992] (1/2) Epoch 12, batch 5000, loss[loss=0.1798, simple_loss=0.2695, pruned_loss=0.04503, over 12300.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2567, pruned_loss=0.03928, over 2367562.47 frames. ], batch size: 34, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:44:24,378 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239592.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 19:44:46,645 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239622.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:44:47,844 INFO [finetune.py:992] (1/2) Epoch 12, batch 5050, loss[loss=0.1845, simple_loss=0.2898, pruned_loss=0.03962, over 12278.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2575, pruned_loss=0.0395, over 2362359.79 frames. ], batch size: 37, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:44:59,517 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239640.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:44:59,657 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6651, 3.7818, 3.3414, 3.2504, 3.0933, 2.9777, 3.8022, 2.5509], device='cuda:1'), covar=tensor([0.0370, 0.0139, 0.0237, 0.0204, 0.0386, 0.0344, 0.0139, 0.0454], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0161, 0.0167, 0.0191, 0.0206, 0.0202, 0.0174, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:45:22,687 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.849e+02 3.125e+02 3.864e+02 7.121e+02, threshold=6.249e+02, percent-clipped=1.0 2023-05-16 19:45:24,108 INFO [finetune.py:992] (1/2) Epoch 12, batch 5100, loss[loss=0.1867, simple_loss=0.2795, pruned_loss=0.04692, over 12048.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2576, pruned_loss=0.0396, over 2366263.93 frames. ], batch size: 37, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:45:30,535 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239683.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:45:31,283 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3548, 4.5906, 3.0630, 2.6749, 3.9986, 2.4343, 3.9313, 3.1114], device='cuda:1'), covar=tensor([0.0729, 0.0513, 0.0996, 0.1388, 0.0260, 0.1369, 0.0505, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0256, 0.0175, 0.0199, 0.0140, 0.0180, 0.0198, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:45:59,777 INFO [finetune.py:992] (1/2) Epoch 12, batch 5150, loss[loss=0.1686, simple_loss=0.2695, pruned_loss=0.03381, over 12192.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2573, pruned_loss=0.03933, over 2366776.81 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:46:14,133 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4838, 5.1512, 5.4810, 4.7958, 5.1221, 4.8739, 5.4684, 5.1112], device='cuda:1'), covar=tensor([0.0289, 0.0373, 0.0320, 0.0282, 0.0386, 0.0351, 0.0313, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0261, 0.0288, 0.0260, 0.0259, 0.0261, 0.0235, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:46:34,240 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.573e+02 3.136e+02 3.793e+02 5.974e+02, threshold=6.273e+02, percent-clipped=0.0 2023-05-16 19:46:35,697 INFO [finetune.py:992] (1/2) Epoch 12, batch 5200, loss[loss=0.1656, simple_loss=0.2589, pruned_loss=0.03615, over 12154.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2569, pruned_loss=0.03939, over 2371967.17 frames. ], batch size: 34, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:46:47,789 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2627, 2.7417, 3.8023, 3.2790, 3.5804, 3.3292, 2.8278, 3.6432], device='cuda:1'), covar=tensor([0.0147, 0.0309, 0.0124, 0.0212, 0.0163, 0.0179, 0.0307, 0.0130], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0191, 0.0192, 0.0221, 0.0168, 0.0198, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:47:12,306 INFO [finetune.py:992] (1/2) Epoch 12, batch 5250, loss[loss=0.1669, simple_loss=0.2485, pruned_loss=0.04263, over 12253.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2563, pruned_loss=0.03923, over 2380743.60 frames. ], batch size: 32, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:47:13,187 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2399, 2.1950, 3.0557, 4.0611, 1.9234, 4.2714, 4.2313, 4.3500], device='cuda:1'), covar=tensor([0.0129, 0.1477, 0.0524, 0.0181, 0.1651, 0.0205, 0.0160, 0.0097], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0206, 0.0186, 0.0121, 0.0194, 0.0183, 0.0180, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:47:46,313 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.888e+02 3.226e+02 3.799e+02 6.835e+02, threshold=6.451e+02, percent-clipped=1.0 2023-05-16 19:47:47,718 INFO [finetune.py:992] (1/2) Epoch 12, batch 5300, loss[loss=0.1507, simple_loss=0.2256, pruned_loss=0.0379, over 12290.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2565, pruned_loss=0.03932, over 2375150.20 frames. ], batch size: 28, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:47:50,647 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7203, 5.6110, 5.5667, 4.9415, 4.9044, 5.7638, 4.8376, 5.0495], device='cuda:1'), covar=tensor([0.1220, 0.1605, 0.1124, 0.2827, 0.1309, 0.1366, 0.3512, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0566, 0.0525, 0.0640, 0.0418, 0.0715, 0.0787, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 19:48:24,063 INFO [finetune.py:992] (1/2) Epoch 12, batch 5350, loss[loss=0.177, simple_loss=0.262, pruned_loss=0.04604, over 12044.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.256, pruned_loss=0.03906, over 2378940.70 frames. ], batch size: 42, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:48:59,109 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.651e+02 3.240e+02 3.822e+02 6.055e+02, threshold=6.479e+02, percent-clipped=0.0 2023-05-16 19:49:00,576 INFO [finetune.py:992] (1/2) Epoch 12, batch 5400, loss[loss=0.1978, simple_loss=0.284, pruned_loss=0.05581, over 12123.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2565, pruned_loss=0.03938, over 2374614.25 frames. ], batch size: 39, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:49:03,473 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239978.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:49:07,205 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9844, 2.5184, 3.3766, 3.9812, 3.6081, 3.9342, 3.4766, 2.7367], device='cuda:1'), covar=tensor([0.0056, 0.0366, 0.0181, 0.0046, 0.0120, 0.0083, 0.0144, 0.0397], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0124, 0.0106, 0.0078, 0.0102, 0.0116, 0.0097, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:49:35,703 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4801, 4.8699, 3.2362, 2.8674, 4.1591, 2.6771, 4.1939, 3.4350], device='cuda:1'), covar=tensor([0.0819, 0.0498, 0.1168, 0.1610, 0.0298, 0.1391, 0.0503, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0259, 0.0179, 0.0202, 0.0142, 0.0182, 0.0202, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:49:36,338 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1533, 5.9181, 5.4388, 5.5664, 5.9937, 5.2475, 5.5786, 5.5107], device='cuda:1'), covar=tensor([0.1456, 0.1077, 0.1070, 0.1976, 0.1017, 0.2345, 0.1642, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0506, 0.0397, 0.0439, 0.0476, 0.0441, 0.0402, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:49:39,030 INFO [finetune.py:992] (1/2) Epoch 12, batch 5450, loss[loss=0.2007, simple_loss=0.2917, pruned_loss=0.05483, over 11983.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2569, pruned_loss=0.03958, over 2366026.27 frames. ], batch size: 42, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:50:13,741 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.697e+02 3.215e+02 4.095e+02 7.076e+02, threshold=6.429e+02, percent-clipped=3.0 2023-05-16 19:50:15,182 INFO [finetune.py:992] (1/2) Epoch 12, batch 5500, loss[loss=0.1753, simple_loss=0.2677, pruned_loss=0.04144, over 11679.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2565, pruned_loss=0.03929, over 2372357.15 frames. ], batch size: 48, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:50:17,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7057, 3.4186, 5.1222, 2.6827, 2.7727, 3.6619, 3.2371, 3.8762], device='cuda:1'), covar=tensor([0.0466, 0.1132, 0.0315, 0.1250, 0.1953, 0.1640, 0.1398, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0238, 0.0253, 0.0186, 0.0240, 0.0298, 0.0225, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 19:50:30,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-16 19:50:51,049 INFO [finetune.py:992] (1/2) Epoch 12, batch 5550, loss[loss=0.1605, simple_loss=0.2482, pruned_loss=0.03639, over 12113.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2569, pruned_loss=0.0399, over 2365471.23 frames. ], batch size: 33, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:50:57,540 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7832, 5.7361, 5.5411, 5.1372, 4.9517, 5.6783, 5.2888, 5.1363], device='cuda:1'), covar=tensor([0.0760, 0.0955, 0.0684, 0.1527, 0.0853, 0.0712, 0.1471, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0564, 0.0523, 0.0639, 0.0415, 0.0714, 0.0785, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 19:51:14,730 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7740, 2.7629, 4.1262, 4.3038, 3.0032, 2.7051, 2.8410, 2.1713], device='cuda:1'), covar=tensor([0.1466, 0.2886, 0.0543, 0.0461, 0.1233, 0.2313, 0.2667, 0.4029], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0386, 0.0272, 0.0299, 0.0270, 0.0303, 0.0379, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:51:25,103 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.669e+02 3.151e+02 3.800e+02 8.227e+02, threshold=6.301e+02, percent-clipped=3.0 2023-05-16 19:51:26,581 INFO [finetune.py:992] (1/2) Epoch 12, batch 5600, loss[loss=0.1595, simple_loss=0.2499, pruned_loss=0.03453, over 12358.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2571, pruned_loss=0.03975, over 2374931.90 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:51:28,925 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6260, 4.9185, 4.2180, 5.2293, 4.7240, 3.3333, 4.4696, 3.2883], device='cuda:1'), covar=tensor([0.0706, 0.0662, 0.1426, 0.0438, 0.1139, 0.1366, 0.0961, 0.3124], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0372, 0.0353, 0.0305, 0.0362, 0.0266, 0.0340, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:51:59,456 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0910, 5.9024, 5.4790, 5.5972, 6.0017, 5.2806, 5.4709, 5.5212], device='cuda:1'), covar=tensor([0.1467, 0.1019, 0.0959, 0.1724, 0.0964, 0.2222, 0.1708, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0510, 0.0400, 0.0443, 0.0478, 0.0444, 0.0405, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:52:02,934 INFO [finetune.py:992] (1/2) Epoch 12, batch 5650, loss[loss=0.1875, simple_loss=0.2713, pruned_loss=0.05191, over 12363.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2559, pruned_loss=0.03931, over 2376675.75 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:52:37,551 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.686e+02 3.253e+02 3.697e+02 6.012e+02, threshold=6.505e+02, percent-clipped=0.0 2023-05-16 19:52:38,898 INFO [finetune.py:992] (1/2) Epoch 12, batch 5700, loss[loss=0.1638, simple_loss=0.2488, pruned_loss=0.03944, over 12105.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2559, pruned_loss=0.03928, over 2372533.98 frames. ], batch size: 33, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:52:41,906 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240278.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:52:52,598 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9112, 5.5818, 5.1770, 5.2126, 5.6952, 5.0168, 5.1426, 5.1225], device='cuda:1'), covar=tensor([0.1470, 0.1081, 0.1248, 0.1669, 0.1016, 0.2087, 0.1948, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0505, 0.0396, 0.0436, 0.0474, 0.0439, 0.0401, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:53:14,241 INFO [finetune.py:992] (1/2) Epoch 12, batch 5750, loss[loss=0.1563, simple_loss=0.2458, pruned_loss=0.03337, over 12351.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2564, pruned_loss=0.0394, over 2371749.16 frames. ], batch size: 31, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:53:14,422 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9482, 4.8402, 4.7872, 4.8010, 4.4550, 4.9921, 4.9593, 5.1754], device='cuda:1'), covar=tensor([0.0266, 0.0183, 0.0194, 0.0372, 0.0814, 0.0294, 0.0153, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0201, 0.0193, 0.0252, 0.0249, 0.0220, 0.0180, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1') 2023-05-16 19:53:15,766 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=240326.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:53:25,914 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240340.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 19:53:49,052 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.047e+02 2.763e+02 3.303e+02 4.001e+02 5.915e+02, threshold=6.607e+02, percent-clipped=0.0 2023-05-16 19:53:50,505 INFO [finetune.py:992] (1/2) Epoch 12, batch 5800, loss[loss=0.1639, simple_loss=0.2573, pruned_loss=0.03526, over 12163.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2564, pruned_loss=0.03934, over 2375999.79 frames. ], batch size: 31, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:54:10,875 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=240401.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 19:54:26,965 INFO [finetune.py:992] (1/2) Epoch 12, batch 5850, loss[loss=0.1712, simple_loss=0.2653, pruned_loss=0.03853, over 11335.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2565, pruned_loss=0.0392, over 2385371.32 frames. ], batch size: 55, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:54:44,599 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2237, 2.9750, 2.9260, 2.8373, 2.6679, 2.5422, 2.9332, 1.9379], device='cuda:1'), covar=tensor([0.0402, 0.0173, 0.0175, 0.0202, 0.0323, 0.0250, 0.0161, 0.0472], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0161, 0.0165, 0.0192, 0.0206, 0.0202, 0.0175, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:55:00,852 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.627e+02 3.128e+02 3.726e+02 9.792e+02, threshold=6.257e+02, percent-clipped=3.0 2023-05-16 19:55:02,339 INFO [finetune.py:992] (1/2) Epoch 12, batch 5900, loss[loss=0.1541, simple_loss=0.233, pruned_loss=0.03759, over 12017.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2567, pruned_loss=0.0392, over 2389352.05 frames. ], batch size: 28, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:55:28,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-05-16 19:55:38,246 INFO [finetune.py:992] (1/2) Epoch 12, batch 5950, loss[loss=0.1679, simple_loss=0.2628, pruned_loss=0.03645, over 12155.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.256, pruned_loss=0.03886, over 2389433.54 frames. ], batch size: 34, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:56:04,594 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1736, 5.0559, 4.9276, 5.0639, 4.3284, 5.1670, 5.1737, 5.3228], device='cuda:1'), covar=tensor([0.0217, 0.0163, 0.0202, 0.0295, 0.1109, 0.0302, 0.0168, 0.0188], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0202, 0.0193, 0.0252, 0.0249, 0.0221, 0.0180, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 19:56:13,079 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.622e+02 3.199e+02 3.651e+02 6.632e+02, threshold=6.397e+02, percent-clipped=1.0 2023-05-16 19:56:14,467 INFO [finetune.py:992] (1/2) Epoch 12, batch 6000, loss[loss=0.1627, simple_loss=0.2578, pruned_loss=0.0338, over 12299.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2569, pruned_loss=0.03892, over 2381079.51 frames. ], batch size: 34, lr: 3.80e-03, grad_scale: 8.0 2023-05-16 19:56:14,467 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 19:56:32,704 INFO [finetune.py:1026] (1/2) Epoch 12, validation: loss=0.3152, simple_loss=0.3923, pruned_loss=0.1191, over 1020973.00 frames. 2023-05-16 19:56:32,705 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 19:56:56,727 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0409, 4.9444, 4.8431, 4.9409, 4.5090, 5.0484, 5.0426, 5.2622], device='cuda:1'), covar=tensor([0.0210, 0.0152, 0.0186, 0.0313, 0.0781, 0.0272, 0.0140, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0202, 0.0194, 0.0253, 0.0250, 0.0222, 0.0181, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 19:57:09,270 INFO [finetune.py:992] (1/2) Epoch 12, batch 6050, loss[loss=0.1554, simple_loss=0.2505, pruned_loss=0.0302, over 12262.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.257, pruned_loss=0.03914, over 2380021.35 frames. ], batch size: 37, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 19:57:10,136 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0088, 5.9554, 5.7201, 5.1531, 5.1373, 5.8772, 5.5416, 5.2756], device='cuda:1'), covar=tensor([0.0727, 0.0883, 0.0692, 0.1677, 0.0716, 0.0698, 0.1428, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0630, 0.0572, 0.0527, 0.0646, 0.0421, 0.0724, 0.0795, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 19:57:14,445 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6596, 2.8501, 4.5036, 4.6618, 2.9205, 2.6495, 2.9682, 2.1861], device='cuda:1'), covar=tensor([0.1630, 0.3059, 0.0472, 0.0397, 0.1326, 0.2454, 0.2639, 0.3816], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0389, 0.0275, 0.0302, 0.0272, 0.0306, 0.0382, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:57:43,808 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.690e+02 3.291e+02 4.006e+02 1.196e+03, threshold=6.583e+02, percent-clipped=4.0 2023-05-16 19:57:45,256 INFO [finetune.py:992] (1/2) Epoch 12, batch 6100, loss[loss=0.1694, simple_loss=0.262, pruned_loss=0.03845, over 12368.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.258, pruned_loss=0.03982, over 2375127.54 frames. ], batch size: 36, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 19:57:51,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8766, 2.2838, 3.5730, 2.8742, 3.4297, 3.0067, 2.3211, 3.3927], device='cuda:1'), covar=tensor([0.0201, 0.0464, 0.0171, 0.0315, 0.0162, 0.0208, 0.0443, 0.0170], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0191, 0.0192, 0.0221, 0.0167, 0.0197, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 19:58:01,015 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240696.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 19:58:20,712 INFO [finetune.py:992] (1/2) Epoch 12, batch 6150, loss[loss=0.191, simple_loss=0.2727, pruned_loss=0.0547, over 11619.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2576, pruned_loss=0.0399, over 2369578.37 frames. ], batch size: 48, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 19:58:23,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 19:58:36,147 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9781, 4.5938, 4.7656, 4.8498, 4.6970, 4.8837, 4.7245, 2.4830], device='cuda:1'), covar=tensor([0.0124, 0.0104, 0.0112, 0.0084, 0.0068, 0.0119, 0.0115, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0082, 0.0074, 0.0061, 0.0094, 0.0083, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 19:58:54,920 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.179e+02 2.787e+02 3.331e+02 4.158e+02 7.198e+02, threshold=6.662e+02, percent-clipped=3.0 2023-05-16 19:58:56,395 INFO [finetune.py:992] (1/2) Epoch 12, batch 6200, loss[loss=0.154, simple_loss=0.2501, pruned_loss=0.0289, over 12360.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2588, pruned_loss=0.04037, over 2358302.49 frames. ], batch size: 35, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 19:59:32,520 INFO [finetune.py:992] (1/2) Epoch 12, batch 6250, loss[loss=0.1435, simple_loss=0.2293, pruned_loss=0.0288, over 12365.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2579, pruned_loss=0.04008, over 2365470.84 frames. ], batch size: 30, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 19:59:45,657 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2447, 4.8035, 5.0474, 5.0989, 4.8880, 5.0506, 5.0146, 3.2624], device='cuda:1'), covar=tensor([0.0093, 0.0066, 0.0073, 0.0062, 0.0048, 0.0107, 0.0080, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0082, 0.0074, 0.0061, 0.0093, 0.0082, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:00:06,438 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.812e+02 3.369e+02 3.939e+02 7.796e+02, threshold=6.738e+02, percent-clipped=1.0 2023-05-16 20:00:07,822 INFO [finetune.py:992] (1/2) Epoch 12, batch 6300, loss[loss=0.1619, simple_loss=0.2428, pruned_loss=0.04051, over 12134.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2574, pruned_loss=0.03986, over 2367934.72 frames. ], batch size: 30, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:00:43,811 INFO [finetune.py:992] (1/2) Epoch 12, batch 6350, loss[loss=0.1572, simple_loss=0.2444, pruned_loss=0.03501, over 12011.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2572, pruned_loss=0.04006, over 2367332.85 frames. ], batch size: 31, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:00:50,339 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240932.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:01:18,683 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.842e+02 2.526e+02 2.951e+02 3.583e+02 3.182e+03, threshold=5.902e+02, percent-clipped=2.0 2023-05-16 20:01:20,174 INFO [finetune.py:992] (1/2) Epoch 12, batch 6400, loss[loss=0.1646, simple_loss=0.2551, pruned_loss=0.03708, over 12278.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2565, pruned_loss=0.03969, over 2372037.01 frames. ], batch size: 33, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:01:28,841 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240986.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:01:33,869 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=240993.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 20:01:35,795 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240996.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:01:56,576 INFO [finetune.py:992] (1/2) Epoch 12, batch 6450, loss[loss=0.1573, simple_loss=0.2413, pruned_loss=0.0366, over 12270.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2569, pruned_loss=0.03952, over 2376481.57 frames. ], batch size: 32, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:02:10,819 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241044.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:02:13,040 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241047.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:02:31,314 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.843e+02 2.681e+02 3.309e+02 3.959e+02 7.143e+02, threshold=6.619e+02, percent-clipped=4.0 2023-05-16 20:02:32,734 INFO [finetune.py:992] (1/2) Epoch 12, batch 6500, loss[loss=0.1895, simple_loss=0.2893, pruned_loss=0.04489, over 12272.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2567, pruned_loss=0.03923, over 2384302.71 frames. ], batch size: 37, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:03:08,527 INFO [finetune.py:992] (1/2) Epoch 12, batch 6550, loss[loss=0.1596, simple_loss=0.2465, pruned_loss=0.03631, over 12368.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2575, pruned_loss=0.03954, over 2377384.15 frames. ], batch size: 38, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:03:43,401 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.717e+02 3.207e+02 3.960e+02 7.122e+02, threshold=6.413e+02, percent-clipped=1.0 2023-05-16 20:03:44,745 INFO [finetune.py:992] (1/2) Epoch 12, batch 6600, loss[loss=0.1739, simple_loss=0.2692, pruned_loss=0.03931, over 12104.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2576, pruned_loss=0.03961, over 2377919.14 frames. ], batch size: 38, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:03:52,209 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7008, 2.5371, 4.6884, 5.0750, 3.2888, 2.5878, 2.8151, 2.0613], device='cuda:1'), covar=tensor([0.1712, 0.3756, 0.0464, 0.0299, 0.1059, 0.2491, 0.3358, 0.5497], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0389, 0.0275, 0.0303, 0.0273, 0.0306, 0.0384, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:04:21,372 INFO [finetune.py:992] (1/2) Epoch 12, batch 6650, loss[loss=0.1789, simple_loss=0.27, pruned_loss=0.04394, over 11695.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2571, pruned_loss=0.0396, over 2371190.81 frames. ], batch size: 48, lr: 3.79e-03, grad_scale: 8.0 2023-05-16 20:04:55,416 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 2.914e+02 3.361e+02 3.942e+02 8.822e+02, threshold=6.721e+02, percent-clipped=4.0 2023-05-16 20:04:56,900 INFO [finetune.py:992] (1/2) Epoch 12, batch 6700, loss[loss=0.1824, simple_loss=0.27, pruned_loss=0.04744, over 11306.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2562, pruned_loss=0.03927, over 2373504.80 frames. ], batch size: 55, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:04:59,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-05-16 20:05:07,048 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241288.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:05:09,187 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9782, 5.9678, 5.7592, 5.2307, 5.1216, 5.9048, 5.5181, 5.2974], device='cuda:1'), covar=tensor([0.0807, 0.0966, 0.0738, 0.1547, 0.0682, 0.0713, 0.1461, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0561, 0.0518, 0.0632, 0.0411, 0.0708, 0.0772, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 20:05:33,161 INFO [finetune.py:992] (1/2) Epoch 12, batch 6750, loss[loss=0.1818, simple_loss=0.2682, pruned_loss=0.04767, over 11893.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2555, pruned_loss=0.03909, over 2379953.58 frames. ], batch size: 44, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:05:45,965 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241342.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:06:08,332 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.614e+02 3.057e+02 3.687e+02 7.195e+02, threshold=6.114e+02, percent-clipped=1.0 2023-05-16 20:06:09,766 INFO [finetune.py:992] (1/2) Epoch 12, batch 6800, loss[loss=0.1816, simple_loss=0.2785, pruned_loss=0.0424, over 12115.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2561, pruned_loss=0.03906, over 2373877.44 frames. ], batch size: 38, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:06:26,975 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2099, 5.1160, 4.9929, 5.1303, 4.6746, 5.1776, 5.1566, 5.4219], device='cuda:1'), covar=tensor([0.0251, 0.0149, 0.0199, 0.0302, 0.0818, 0.0336, 0.0158, 0.0175], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0199, 0.0192, 0.0251, 0.0246, 0.0220, 0.0178, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 20:06:33,112 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241406.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:06:45,789 INFO [finetune.py:992] (1/2) Epoch 12, batch 6850, loss[loss=0.1473, simple_loss=0.2308, pruned_loss=0.03191, over 11805.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2557, pruned_loss=0.0389, over 2381476.34 frames. ], batch size: 26, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:06:58,167 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0197, 2.5203, 3.6347, 2.9842, 3.4209, 3.1353, 2.4776, 3.5037], device='cuda:1'), covar=tensor([0.0171, 0.0393, 0.0154, 0.0281, 0.0177, 0.0200, 0.0427, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0192, 0.0191, 0.0221, 0.0167, 0.0198, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:07:17,305 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241467.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:07:20,528 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.775e+02 3.412e+02 3.852e+02 9.667e+02, threshold=6.824e+02, percent-clipped=3.0 2023-05-16 20:07:21,937 INFO [finetune.py:992] (1/2) Epoch 12, batch 6900, loss[loss=0.1594, simple_loss=0.2451, pruned_loss=0.0368, over 12095.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2572, pruned_loss=0.03955, over 2370400.47 frames. ], batch size: 32, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:07:42,766 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5555, 2.4566, 3.8902, 4.4209, 4.0953, 4.3239, 3.9740, 3.1576], device='cuda:1'), covar=tensor([0.0050, 0.0490, 0.0127, 0.0047, 0.0106, 0.0102, 0.0141, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0124, 0.0107, 0.0079, 0.0102, 0.0116, 0.0098, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:07:57,834 INFO [finetune.py:992] (1/2) Epoch 12, batch 6950, loss[loss=0.1918, simple_loss=0.2846, pruned_loss=0.04957, over 12045.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2583, pruned_loss=0.03992, over 2367360.80 frames. ], batch size: 40, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:08:02,225 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5600, 2.6606, 3.6917, 4.4353, 3.8858, 4.3333, 3.8231, 2.9874], device='cuda:1'), covar=tensor([0.0032, 0.0356, 0.0145, 0.0036, 0.0104, 0.0078, 0.0121, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0124, 0.0107, 0.0080, 0.0103, 0.0117, 0.0098, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:08:31,995 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.054e+02 2.695e+02 3.162e+02 3.755e+02 7.672e+02, threshold=6.324e+02, percent-clipped=2.0 2023-05-16 20:08:33,381 INFO [finetune.py:992] (1/2) Epoch 12, batch 7000, loss[loss=0.199, simple_loss=0.2785, pruned_loss=0.05978, over 12095.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2579, pruned_loss=0.03986, over 2374562.24 frames. ], batch size: 32, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:08:43,535 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241588.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:08:45,774 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1973, 4.6459, 4.0315, 4.8913, 4.5278, 2.3634, 4.0291, 2.9058], device='cuda:1'), covar=tensor([0.0847, 0.0690, 0.1409, 0.0492, 0.1034, 0.1977, 0.1195, 0.3290], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0383, 0.0361, 0.0313, 0.0372, 0.0273, 0.0347, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:08:52,222 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4249, 4.7070, 2.8192, 2.7749, 4.0044, 2.7131, 4.0078, 3.2980], device='cuda:1'), covar=tensor([0.0652, 0.0491, 0.1165, 0.1378, 0.0316, 0.1215, 0.0477, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0256, 0.0175, 0.0198, 0.0141, 0.0180, 0.0198, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:08:53,243 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0343, 2.1791, 3.3850, 3.8808, 3.5932, 3.8658, 3.4937, 2.6142], device='cuda:1'), covar=tensor([0.0055, 0.0452, 0.0170, 0.0063, 0.0132, 0.0102, 0.0152, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0125, 0.0107, 0.0080, 0.0103, 0.0117, 0.0098, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:09:08,282 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:09:10,201 INFO [finetune.py:992] (1/2) Epoch 12, batch 7050, loss[loss=0.1794, simple_loss=0.2817, pruned_loss=0.03854, over 11600.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2585, pruned_loss=0.0403, over 2366627.46 frames. ], batch size: 48, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:09:15,497 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6363, 4.3006, 4.6141, 4.1201, 4.3465, 4.1707, 4.6267, 4.2812], device='cuda:1'), covar=tensor([0.0302, 0.0401, 0.0310, 0.0265, 0.0383, 0.0325, 0.0250, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0262, 0.0287, 0.0262, 0.0259, 0.0262, 0.0237, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:09:18,837 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241636.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:09:23,015 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241642.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:09:28,159 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3742, 4.7240, 2.8147, 2.6771, 4.0605, 2.6042, 4.0008, 3.4046], device='cuda:1'), covar=tensor([0.0659, 0.0490, 0.1133, 0.1394, 0.0300, 0.1265, 0.0506, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0254, 0.0174, 0.0197, 0.0140, 0.0180, 0.0197, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:09:36,669 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8294, 3.8005, 3.3280, 3.2890, 3.0663, 2.9144, 3.7632, 2.4173], device='cuda:1'), covar=tensor([0.0331, 0.0108, 0.0176, 0.0193, 0.0370, 0.0368, 0.0124, 0.0443], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0159, 0.0162, 0.0189, 0.0201, 0.0201, 0.0171, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:09:45,058 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.716e+02 3.212e+02 3.977e+02 7.981e+02, threshold=6.423e+02, percent-clipped=1.0 2023-05-16 20:09:46,258 INFO [finetune.py:992] (1/2) Epoch 12, batch 7100, loss[loss=0.1948, simple_loss=0.279, pruned_loss=0.05532, over 12295.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2591, pruned_loss=0.04032, over 2371706.83 frames. ], batch size: 34, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:09:52,273 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241682.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:09:57,649 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241690.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:10:12,716 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2819, 3.0313, 4.7376, 2.3392, 2.5178, 3.5358, 3.0391, 3.7145], device='cuda:1'), covar=tensor([0.0602, 0.1328, 0.0468, 0.1382, 0.2212, 0.1804, 0.1489, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0233, 0.0249, 0.0181, 0.0237, 0.0294, 0.0220, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:10:21,795 INFO [finetune.py:992] (1/2) Epoch 12, batch 7150, loss[loss=0.1997, simple_loss=0.2967, pruned_loss=0.05133, over 12376.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2589, pruned_loss=0.04009, over 2368192.42 frames. ], batch size: 38, lr: 3.79e-03, grad_scale: 16.0 2023-05-16 20:10:49,603 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241762.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 20:10:56,606 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.664e+02 3.124e+02 3.838e+02 7.600e+02, threshold=6.248e+02, percent-clipped=2.0 2023-05-16 20:10:57,993 INFO [finetune.py:992] (1/2) Epoch 12, batch 7200, loss[loss=0.1624, simple_loss=0.239, pruned_loss=0.04295, over 12356.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2592, pruned_loss=0.04015, over 2365187.61 frames. ], batch size: 30, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:11:19,956 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241803.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:11:34,450 INFO [finetune.py:992] (1/2) Epoch 12, batch 7250, loss[loss=0.156, simple_loss=0.2517, pruned_loss=0.03014, over 12310.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2585, pruned_loss=0.03983, over 2365410.24 frames. ], batch size: 33, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:11:39,579 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241831.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:12:02,887 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241864.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:12:08,393 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.596e+02 3.129e+02 3.845e+02 1.160e+03, threshold=6.258e+02, percent-clipped=4.0 2023-05-16 20:12:09,829 INFO [finetune.py:992] (1/2) Epoch 12, batch 7300, loss[loss=0.1609, simple_loss=0.2583, pruned_loss=0.03173, over 12169.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2582, pruned_loss=0.03993, over 2359020.61 frames. ], batch size: 36, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:12:23,626 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241892.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 20:12:27,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 20:12:43,615 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8947, 3.5742, 5.3359, 2.7144, 2.7908, 3.9276, 3.3729, 3.9533], device='cuda:1'), covar=tensor([0.0395, 0.0993, 0.0217, 0.1117, 0.2005, 0.1462, 0.1240, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0234, 0.0249, 0.0182, 0.0237, 0.0294, 0.0221, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:12:46,201 INFO [finetune.py:992] (1/2) Epoch 12, batch 7350, loss[loss=0.1505, simple_loss=0.2409, pruned_loss=0.03002, over 12346.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2576, pruned_loss=0.03984, over 2369831.61 frames. ], batch size: 30, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:13:04,813 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6140, 2.4310, 3.3183, 4.5364, 2.4869, 4.4686, 4.4672, 4.6547], device='cuda:1'), covar=tensor([0.0155, 0.1315, 0.0479, 0.0138, 0.1256, 0.0245, 0.0205, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0205, 0.0185, 0.0120, 0.0191, 0.0182, 0.0180, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:13:20,763 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.598e+02 2.904e+02 3.475e+02 6.796e+02, threshold=5.808e+02, percent-clipped=1.0 2023-05-16 20:13:22,265 INFO [finetune.py:992] (1/2) Epoch 12, batch 7400, loss[loss=0.1502, simple_loss=0.2353, pruned_loss=0.03253, over 12036.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04011, over 2358581.73 frames. ], batch size: 31, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:13:24,546 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241977.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:14:01,351 INFO [finetune.py:992] (1/2) Epoch 12, batch 7450, loss[loss=0.1488, simple_loss=0.2298, pruned_loss=0.03393, over 12163.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2581, pruned_loss=0.04033, over 2357282.90 frames. ], batch size: 29, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:14:28,478 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242062.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:14:35,407 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 2.768e+02 3.267e+02 3.959e+02 8.297e+02, threshold=6.535e+02, percent-clipped=4.0 2023-05-16 20:14:36,874 INFO [finetune.py:992] (1/2) Epoch 12, batch 7500, loss[loss=0.1602, simple_loss=0.2597, pruned_loss=0.03035, over 12296.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2578, pruned_loss=0.04032, over 2356543.25 frames. ], batch size: 34, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:14:41,240 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9309, 4.5811, 4.7212, 4.8903, 4.7572, 4.8747, 4.7998, 2.4177], device='cuda:1'), covar=tensor([0.0109, 0.0079, 0.0095, 0.0055, 0.0049, 0.0094, 0.0074, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0082, 0.0074, 0.0060, 0.0093, 0.0082, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:15:03,160 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242110.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:15:12,886 INFO [finetune.py:992] (1/2) Epoch 12, batch 7550, loss[loss=0.1607, simple_loss=0.2564, pruned_loss=0.03254, over 12277.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2581, pruned_loss=0.04052, over 2366148.05 frames. ], batch size: 37, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:15:33,468 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2040, 4.8621, 5.2367, 4.4901, 4.9128, 4.5610, 5.2031, 4.8396], device='cuda:1'), covar=tensor([0.0343, 0.0456, 0.0406, 0.0341, 0.0398, 0.0393, 0.0332, 0.0404], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0263, 0.0289, 0.0263, 0.0261, 0.0264, 0.0239, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:15:33,529 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242153.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:15:37,695 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242159.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:15:47,472 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 2.622e+02 3.220e+02 3.769e+02 1.536e+03, threshold=6.441e+02, percent-clipped=4.0 2023-05-16 20:15:48,921 INFO [finetune.py:992] (1/2) Epoch 12, batch 7600, loss[loss=0.1545, simple_loss=0.2492, pruned_loss=0.02984, over 12336.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2584, pruned_loss=0.04086, over 2363699.30 frames. ], batch size: 31, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:15:58,287 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242187.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 20:16:00,662 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3644, 3.1880, 3.1812, 3.5324, 2.5467, 3.1594, 2.6500, 2.9980], device='cuda:1'), covar=tensor([0.1327, 0.0706, 0.0753, 0.0507, 0.0975, 0.0725, 0.1465, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0265, 0.0296, 0.0358, 0.0239, 0.0243, 0.0259, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:16:18,243 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242214.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:16:25,317 INFO [finetune.py:992] (1/2) Epoch 12, batch 7650, loss[loss=0.185, simple_loss=0.2748, pruned_loss=0.04757, over 12351.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2584, pruned_loss=0.04091, over 2365041.43 frames. ], batch size: 35, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:16:26,867 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242226.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:17:00,127 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.730e+02 3.254e+02 3.873e+02 7.340e+02, threshold=6.508e+02, percent-clipped=1.0 2023-05-16 20:17:01,532 INFO [finetune.py:992] (1/2) Epoch 12, batch 7700, loss[loss=0.151, simple_loss=0.2451, pruned_loss=0.02844, over 12284.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2581, pruned_loss=0.04069, over 2367556.76 frames. ], batch size: 33, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:17:03,748 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242277.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:17:11,010 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242287.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:17:26,779 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242309.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:17:38,001 INFO [finetune.py:992] (1/2) Epoch 12, batch 7750, loss[loss=0.14, simple_loss=0.2199, pruned_loss=0.03002, over 11412.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2576, pruned_loss=0.04043, over 2360807.57 frames. ], batch size: 25, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:17:38,753 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242325.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:18:10,947 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242370.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:18:12,115 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.829e+02 3.213e+02 3.821e+02 6.496e+02, threshold=6.427e+02, percent-clipped=0.0 2023-05-16 20:18:13,537 INFO [finetune.py:992] (1/2) Epoch 12, batch 7800, loss[loss=0.1634, simple_loss=0.266, pruned_loss=0.03039, over 12364.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.258, pruned_loss=0.04055, over 2362177.15 frames. ], batch size: 36, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:18:20,554 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0611, 4.9330, 4.8083, 4.9319, 4.5950, 4.9953, 5.0250, 5.2141], device='cuda:1'), covar=tensor([0.0230, 0.0177, 0.0203, 0.0369, 0.0738, 0.0404, 0.0188, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0202, 0.0194, 0.0254, 0.0247, 0.0222, 0.0180, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 20:18:49,461 INFO [finetune.py:992] (1/2) Epoch 12, batch 7850, loss[loss=0.1495, simple_loss=0.2414, pruned_loss=0.02878, over 12288.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.258, pruned_loss=0.04045, over 2373836.13 frames. ], batch size: 33, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:19:14,261 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242459.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:19:24,098 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.795e+02 3.203e+02 3.816e+02 9.207e+02, threshold=6.406e+02, percent-clipped=4.0 2023-05-16 20:19:24,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 20:19:25,610 INFO [finetune.py:992] (1/2) Epoch 12, batch 7900, loss[loss=0.1592, simple_loss=0.2415, pruned_loss=0.03846, over 11828.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2582, pruned_loss=0.0409, over 2367073.54 frames. ], batch size: 26, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:19:32,114 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242483.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:19:34,881 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242487.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:19:48,941 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242507.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:19:50,399 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242509.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:19:59,132 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5281, 2.8180, 3.2494, 4.4524, 2.2540, 4.4710, 4.4967, 4.6185], device='cuda:1'), covar=tensor([0.0139, 0.1118, 0.0497, 0.0147, 0.1377, 0.0216, 0.0155, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0203, 0.0184, 0.0119, 0.0190, 0.0181, 0.0178, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:20:01,099 INFO [finetune.py:992] (1/2) Epoch 12, batch 7950, loss[loss=0.1449, simple_loss=0.2266, pruned_loss=0.03159, over 12001.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2569, pruned_loss=0.04016, over 2370981.13 frames. ], batch size: 28, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:20:01,952 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0839, 5.9982, 5.5944, 5.6006, 6.0456, 5.3516, 5.5285, 5.5201], device='cuda:1'), covar=tensor([0.1369, 0.0908, 0.1095, 0.1674, 0.0871, 0.2077, 0.1758, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0496, 0.0396, 0.0440, 0.0469, 0.0436, 0.0396, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:20:08,978 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242535.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:20:16,145 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242544.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:20:23,972 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242555.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:20:35,873 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.774e+02 3.170e+02 3.692e+02 7.403e+02, threshold=6.340e+02, percent-clipped=1.0 2023-05-16 20:20:37,250 INFO [finetune.py:992] (1/2) Epoch 12, batch 8000, loss[loss=0.1505, simple_loss=0.2379, pruned_loss=0.03151, over 11823.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2568, pruned_loss=0.03983, over 2374152.88 frames. ], batch size: 26, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:20:42,876 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242582.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:20:57,510 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242602.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:21:08,081 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242616.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:21:13,417 INFO [finetune.py:992] (1/2) Epoch 12, batch 8050, loss[loss=0.1594, simple_loss=0.2522, pruned_loss=0.03332, over 12355.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2567, pruned_loss=0.04007, over 2369316.50 frames. ], batch size: 36, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:21:17,224 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8198, 3.5289, 3.6554, 3.7524, 3.4857, 3.8402, 3.8573, 3.9283], device='cuda:1'), covar=tensor([0.0242, 0.0247, 0.0194, 0.0463, 0.0579, 0.0510, 0.0194, 0.0259], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0200, 0.0192, 0.0251, 0.0246, 0.0221, 0.0179, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 20:21:41,381 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242663.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:21:42,722 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242665.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:21:43,560 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5042, 2.2063, 3.8874, 4.3739, 4.0164, 4.3777, 3.8710, 2.9815], device='cuda:1'), covar=tensor([0.0053, 0.0546, 0.0144, 0.0062, 0.0108, 0.0083, 0.0141, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0124, 0.0105, 0.0079, 0.0103, 0.0116, 0.0097, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:21:47,628 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 2.732e+02 3.297e+02 3.901e+02 7.745e+02, threshold=6.593e+02, percent-clipped=1.0 2023-05-16 20:21:48,986 INFO [finetune.py:992] (1/2) Epoch 12, batch 8100, loss[loss=0.1663, simple_loss=0.2518, pruned_loss=0.04036, over 12250.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2575, pruned_loss=0.04015, over 2377625.44 frames. ], batch size: 32, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:21:54,803 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4775, 4.8897, 4.3410, 5.1394, 4.7394, 3.1333, 4.3916, 3.2732], device='cuda:1'), covar=tensor([0.0738, 0.0712, 0.1395, 0.0439, 0.0965, 0.1524, 0.0964, 0.3058], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0377, 0.0352, 0.0311, 0.0365, 0.0268, 0.0341, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:22:04,116 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.59 vs. limit=5.0 2023-05-16 20:22:23,705 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5009, 2.4400, 3.6600, 4.3466, 3.8374, 4.4118, 3.7095, 2.9917], device='cuda:1'), covar=tensor([0.0045, 0.0441, 0.0139, 0.0054, 0.0115, 0.0076, 0.0158, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0124, 0.0106, 0.0079, 0.0104, 0.0117, 0.0098, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:22:24,906 INFO [finetune.py:992] (1/2) Epoch 12, batch 8150, loss[loss=0.1613, simple_loss=0.2566, pruned_loss=0.033, over 12142.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2579, pruned_loss=0.04038, over 2380817.01 frames. ], batch size: 38, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:22:43,512 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3677, 6.1223, 5.5897, 5.7350, 6.1850, 5.4631, 5.7448, 5.6773], device='cuda:1'), covar=tensor([0.1498, 0.0934, 0.1165, 0.1689, 0.0953, 0.2033, 0.1750, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0496, 0.0396, 0.0442, 0.0469, 0.0437, 0.0396, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:22:59,370 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 2.796e+02 3.192e+02 4.090e+02 9.009e+02, threshold=6.384e+02, percent-clipped=4.0 2023-05-16 20:22:59,558 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242772.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:23:00,818 INFO [finetune.py:992] (1/2) Epoch 12, batch 8200, loss[loss=0.1609, simple_loss=0.2545, pruned_loss=0.03367, over 12186.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2579, pruned_loss=0.04037, over 2379788.52 frames. ], batch size: 35, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:23:08,347 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3550, 2.5605, 3.0658, 4.1997, 2.3489, 4.2674, 4.2978, 4.4352], device='cuda:1'), covar=tensor([0.0117, 0.1128, 0.0513, 0.0144, 0.1304, 0.0243, 0.0140, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0201, 0.0182, 0.0118, 0.0188, 0.0181, 0.0177, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:23:24,271 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4620, 4.7726, 3.0489, 2.8128, 4.0152, 2.5725, 4.0176, 3.3074], device='cuda:1'), covar=tensor([0.0726, 0.0580, 0.1171, 0.1489, 0.0376, 0.1462, 0.0561, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0258, 0.0177, 0.0200, 0.0143, 0.0183, 0.0201, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:23:24,329 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3208, 4.6570, 4.0231, 4.9988, 4.5606, 2.9255, 4.2583, 3.0627], device='cuda:1'), covar=tensor([0.0815, 0.0788, 0.1506, 0.0467, 0.1103, 0.1649, 0.1040, 0.3434], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0380, 0.0357, 0.0314, 0.0369, 0.0272, 0.0345, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:23:26,309 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242809.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:23:37,152 INFO [finetune.py:992] (1/2) Epoch 12, batch 8250, loss[loss=0.1876, simple_loss=0.2874, pruned_loss=0.04384, over 11322.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2577, pruned_loss=0.03971, over 2386048.58 frames. ], batch size: 55, lr: 3.78e-03, grad_scale: 16.0 2023-05-16 20:23:41,031 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8470, 3.4920, 5.1875, 2.6260, 2.6123, 3.8996, 3.1465, 3.9517], device='cuda:1'), covar=tensor([0.0389, 0.1089, 0.0262, 0.1223, 0.2124, 0.1387, 0.1426, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0232, 0.0248, 0.0181, 0.0236, 0.0293, 0.0220, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:23:44,333 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2821, 4.3519, 2.8008, 2.5867, 3.8065, 2.4484, 3.8377, 3.0261], device='cuda:1'), covar=tensor([0.0689, 0.0614, 0.1227, 0.1555, 0.0332, 0.1483, 0.0496, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0258, 0.0177, 0.0199, 0.0142, 0.0182, 0.0200, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:23:44,350 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242833.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:23:48,384 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242839.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:23:52,108 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 20:24:01,109 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242857.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:24:01,222 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1069, 4.9752, 4.9150, 4.9531, 4.6545, 5.1021, 5.0294, 5.2950], device='cuda:1'), covar=tensor([0.0224, 0.0161, 0.0173, 0.0394, 0.0747, 0.0274, 0.0159, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0200, 0.0192, 0.0252, 0.0246, 0.0221, 0.0179, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 20:24:05,573 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5958, 4.2179, 4.3294, 4.5058, 4.3868, 4.5885, 4.4029, 2.3265], device='cuda:1'), covar=tensor([0.0166, 0.0111, 0.0145, 0.0095, 0.0074, 0.0135, 0.0120, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0079, 0.0081, 0.0074, 0.0060, 0.0093, 0.0082, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:24:11,666 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.670e+02 3.128e+02 3.945e+02 6.328e+02, threshold=6.256e+02, percent-clipped=0.0 2023-05-16 20:24:13,103 INFO [finetune.py:992] (1/2) Epoch 12, batch 8300, loss[loss=0.1587, simple_loss=0.2449, pruned_loss=0.03626, over 12288.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2587, pruned_loss=0.04038, over 2379029.21 frames. ], batch size: 33, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:24:19,045 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242882.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:24:39,955 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242911.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:24:47,280 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242921.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:24:48,016 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4900, 5.0733, 5.4968, 4.8091, 5.1093, 4.8938, 5.5009, 5.0757], device='cuda:1'), covar=tensor([0.0251, 0.0383, 0.0244, 0.0241, 0.0369, 0.0316, 0.0200, 0.0290], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0265, 0.0289, 0.0264, 0.0264, 0.0264, 0.0240, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:24:49,293 INFO [finetune.py:992] (1/2) Epoch 12, batch 8350, loss[loss=0.1698, simple_loss=0.2693, pruned_loss=0.03513, over 12025.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2585, pruned_loss=0.04017, over 2378562.24 frames. ], batch size: 40, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:24:53,475 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242930.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:24:59,542 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 20:25:10,539 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2328, 4.0438, 4.2267, 4.6291, 3.1188, 3.9154, 2.6117, 4.1307], device='cuda:1'), covar=tensor([0.1862, 0.0738, 0.0936, 0.0529, 0.1206, 0.0668, 0.1935, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0264, 0.0296, 0.0356, 0.0239, 0.0240, 0.0258, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:25:13,309 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242958.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:25:14,072 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4849, 5.2868, 5.4158, 5.4585, 5.0791, 5.1250, 4.8509, 5.4033], device='cuda:1'), covar=tensor([0.0557, 0.0526, 0.0765, 0.0519, 0.1698, 0.1151, 0.0522, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0708, 0.0618, 0.0628, 0.0852, 0.0750, 0.0554, 0.0486], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:25:18,290 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:25:18,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-16 20:25:23,170 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.692e+02 2.758e+02 3.242e+02 3.889e+02 6.335e+02, threshold=6.485e+02, percent-clipped=1.0 2023-05-16 20:25:24,688 INFO [finetune.py:992] (1/2) Epoch 12, batch 8400, loss[loss=0.1727, simple_loss=0.2718, pruned_loss=0.03677, over 12297.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2582, pruned_loss=0.04003, over 2384727.01 frames. ], batch size: 34, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:25:31,300 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242982.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 20:25:53,261 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243013.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:26:00,961 INFO [finetune.py:992] (1/2) Epoch 12, batch 8450, loss[loss=0.1755, simple_loss=0.2738, pruned_loss=0.03863, over 10523.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2585, pruned_loss=0.04006, over 2378188.75 frames. ], batch size: 68, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:26:08,302 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6791, 2.8116, 4.4339, 4.6019, 2.8141, 2.5588, 2.8107, 2.1509], device='cuda:1'), covar=tensor([0.1619, 0.3053, 0.0504, 0.0428, 0.1362, 0.2468, 0.2842, 0.3995], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0388, 0.0275, 0.0303, 0.0274, 0.0307, 0.0382, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:26:15,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.25 vs. limit=5.0 2023-05-16 20:26:35,249 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.724e+02 3.204e+02 3.577e+02 5.925e+02, threshold=6.408e+02, percent-clipped=0.0 2023-05-16 20:26:36,753 INFO [finetune.py:992] (1/2) Epoch 12, batch 8500, loss[loss=0.1694, simple_loss=0.2624, pruned_loss=0.03818, over 12182.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2587, pruned_loss=0.03999, over 2377753.36 frames. ], batch size: 35, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:27:12,122 INFO [finetune.py:992] (1/2) Epoch 12, batch 8550, loss[loss=0.1753, simple_loss=0.2696, pruned_loss=0.04051, over 12156.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.258, pruned_loss=0.03985, over 2376511.60 frames. ], batch size: 36, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:27:15,700 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243128.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:27:20,681 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5538, 2.7430, 3.1653, 4.4348, 2.4273, 4.4501, 4.5087, 4.6408], device='cuda:1'), covar=tensor([0.0106, 0.1147, 0.0527, 0.0144, 0.1286, 0.0218, 0.0151, 0.0083], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0204, 0.0186, 0.0120, 0.0192, 0.0183, 0.0180, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:27:23,499 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243139.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:27:24,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 20:27:46,820 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.711e+02 3.242e+02 3.742e+02 1.088e+03, threshold=6.484e+02, percent-clipped=3.0 2023-05-16 20:27:48,281 INFO [finetune.py:992] (1/2) Epoch 12, batch 8600, loss[loss=0.1318, simple_loss=0.2217, pruned_loss=0.02094, over 12346.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2581, pruned_loss=0.03991, over 2376817.53 frames. ], batch size: 30, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:27:57,586 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243187.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:28:15,641 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243211.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:28:24,845 INFO [finetune.py:992] (1/2) Epoch 12, batch 8650, loss[loss=0.1718, simple_loss=0.261, pruned_loss=0.04132, over 11889.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2581, pruned_loss=0.04019, over 2369342.74 frames. ], batch size: 44, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:28:48,874 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243258.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:28:49,526 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243259.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:28:59,525 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 2.814e+02 3.184e+02 3.863e+02 6.884e+02, threshold=6.368e+02, percent-clipped=2.0 2023-05-16 20:29:00,960 INFO [finetune.py:992] (1/2) Epoch 12, batch 8700, loss[loss=0.1514, simple_loss=0.2421, pruned_loss=0.03037, over 12113.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2583, pruned_loss=0.03996, over 2368539.62 frames. ], batch size: 33, lr: 3.77e-03, grad_scale: 32.0 2023-05-16 20:29:03,219 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243277.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:29:14,452 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243293.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:29:23,855 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243306.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:29:36,494 INFO [finetune.py:992] (1/2) Epoch 12, batch 8750, loss[loss=0.1847, simple_loss=0.2676, pruned_loss=0.05087, over 11789.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2586, pruned_loss=0.04006, over 2369619.27 frames. ], batch size: 44, lr: 3.77e-03, grad_scale: 32.0 2023-05-16 20:29:48,060 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4137, 5.2175, 5.3316, 5.3818, 5.0091, 5.0205, 4.8131, 5.3399], device='cuda:1'), covar=tensor([0.0688, 0.0628, 0.0818, 0.0651, 0.1873, 0.1421, 0.0541, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0712, 0.0620, 0.0634, 0.0860, 0.0755, 0.0556, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:29:58,669 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243354.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:30:10,893 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.851e+02 3.290e+02 4.022e+02 1.120e+03, threshold=6.581e+02, percent-clipped=4.0 2023-05-16 20:30:12,367 INFO [finetune.py:992] (1/2) Epoch 12, batch 8800, loss[loss=0.1323, simple_loss=0.2143, pruned_loss=0.02517, over 12372.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2588, pruned_loss=0.04021, over 2370633.90 frames. ], batch size: 30, lr: 3.77e-03, grad_scale: 32.0 2023-05-16 20:30:20,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 20:30:48,549 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6636, 4.2437, 4.4253, 4.5376, 4.4053, 4.5393, 4.5111, 2.4655], device='cuda:1'), covar=tensor([0.0095, 0.0090, 0.0090, 0.0074, 0.0058, 0.0111, 0.0093, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0079, 0.0082, 0.0075, 0.0061, 0.0094, 0.0083, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:30:49,065 INFO [finetune.py:992] (1/2) Epoch 12, batch 8850, loss[loss=0.1805, simple_loss=0.2711, pruned_loss=0.04493, over 11319.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2571, pruned_loss=0.0396, over 2373037.09 frames. ], batch size: 55, lr: 3.77e-03, grad_scale: 32.0 2023-05-16 20:30:51,928 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243428.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:31:11,331 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1644, 6.1059, 5.8675, 5.4084, 5.2419, 5.9915, 5.6414, 5.4088], device='cuda:1'), covar=tensor([0.0584, 0.0740, 0.0659, 0.1616, 0.0621, 0.0728, 0.1486, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0561, 0.0519, 0.0634, 0.0415, 0.0713, 0.0782, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 20:31:12,136 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3322, 2.5934, 3.5728, 4.1204, 3.8170, 4.2112, 3.7504, 2.9745], device='cuda:1'), covar=tensor([0.0046, 0.0376, 0.0171, 0.0081, 0.0112, 0.0088, 0.0149, 0.0408], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0125, 0.0107, 0.0080, 0.0104, 0.0119, 0.0097, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:31:14,202 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5549, 5.5125, 5.2824, 4.8559, 4.8946, 5.4280, 5.0766, 4.8936], device='cuda:1'), covar=tensor([0.0632, 0.0768, 0.0706, 0.1593, 0.0887, 0.0783, 0.1476, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0561, 0.0519, 0.0633, 0.0415, 0.0713, 0.0781, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 20:31:23,250 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.679e+02 3.186e+02 3.892e+02 5.819e+02, threshold=6.373e+02, percent-clipped=0.0 2023-05-16 20:31:24,688 INFO [finetune.py:992] (1/2) Epoch 12, batch 8900, loss[loss=0.241, simple_loss=0.3261, pruned_loss=0.07795, over 8043.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2565, pruned_loss=0.03921, over 2366839.16 frames. ], batch size: 98, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:31:26,077 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243476.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:31:49,671 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3262, 4.8533, 5.1909, 5.1647, 4.9713, 5.1606, 5.1075, 2.8745], device='cuda:1'), covar=tensor([0.0074, 0.0066, 0.0063, 0.0047, 0.0040, 0.0085, 0.0101, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0079, 0.0081, 0.0074, 0.0060, 0.0093, 0.0082, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:32:00,882 INFO [finetune.py:992] (1/2) Epoch 12, batch 8950, loss[loss=0.1691, simple_loss=0.2652, pruned_loss=0.03644, over 12315.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2567, pruned_loss=0.0396, over 2361344.36 frames. ], batch size: 34, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:32:36,324 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.807e+02 3.224e+02 3.679e+02 7.560e+02, threshold=6.449e+02, percent-clipped=1.0 2023-05-16 20:32:36,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 20:32:37,012 INFO [finetune.py:992] (1/2) Epoch 12, batch 9000, loss[loss=0.1603, simple_loss=0.2391, pruned_loss=0.04075, over 12326.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2572, pruned_loss=0.03961, over 2370870.99 frames. ], batch size: 30, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:32:37,012 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 20:32:46,838 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1984, 3.3809, 3.0567, 3.5543, 3.3295, 2.4305, 3.2409, 2.7978], device='cuda:1'), covar=tensor([0.0915, 0.1021, 0.1508, 0.0670, 0.1128, 0.1718, 0.1093, 0.2686], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0380, 0.0356, 0.0313, 0.0369, 0.0271, 0.0344, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:32:55,019 INFO [finetune.py:1026] (1/2) Epoch 12, validation: loss=0.3264, simple_loss=0.3973, pruned_loss=0.1277, over 1020973.00 frames. 2023-05-16 20:32:55,020 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 20:32:57,988 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243577.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:32:58,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0804, 2.4435, 3.6446, 3.0078, 3.4524, 3.1721, 2.3845, 3.5545], device='cuda:1'), covar=tensor([0.0140, 0.0346, 0.0140, 0.0249, 0.0146, 0.0193, 0.0396, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0206, 0.0190, 0.0189, 0.0218, 0.0166, 0.0198, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:33:27,703 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1124, 4.9699, 4.8623, 4.9195, 4.5887, 5.0043, 5.0050, 5.2735], device='cuda:1'), covar=tensor([0.0218, 0.0149, 0.0186, 0.0318, 0.0750, 0.0284, 0.0134, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0201, 0.0194, 0.0253, 0.0247, 0.0223, 0.0179, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:1') 2023-05-16 20:33:31,806 INFO [finetune.py:992] (1/2) Epoch 12, batch 9050, loss[loss=0.1599, simple_loss=0.2556, pruned_loss=0.03209, over 12154.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2575, pruned_loss=0.03968, over 2368023.44 frames. ], batch size: 36, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:33:32,602 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243625.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:33:38,486 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2013, 3.6240, 3.3505, 3.2700, 2.9055, 2.7464, 3.6507, 2.0626], device='cuda:1'), covar=tensor([0.0524, 0.0156, 0.0193, 0.0214, 0.0384, 0.0353, 0.0129, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0163, 0.0164, 0.0192, 0.0205, 0.0203, 0.0175, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:33:50,376 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243649.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:34:07,274 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.482e+02 3.089e+02 3.852e+02 1.074e+03, threshold=6.178e+02, percent-clipped=3.0 2023-05-16 20:34:07,979 INFO [finetune.py:992] (1/2) Epoch 12, batch 9100, loss[loss=0.1532, simple_loss=0.2447, pruned_loss=0.03079, over 12107.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2566, pruned_loss=0.03954, over 2374261.97 frames. ], batch size: 33, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:34:38,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 20:34:43,344 INFO [finetune.py:992] (1/2) Epoch 12, batch 9150, loss[loss=0.1759, simple_loss=0.2672, pruned_loss=0.04232, over 11756.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2572, pruned_loss=0.03978, over 2376398.96 frames. ], batch size: 44, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:35:11,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-05-16 20:35:18,962 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.646e+02 3.221e+02 3.838e+02 1.025e+03, threshold=6.442e+02, percent-clipped=3.0 2023-05-16 20:35:19,703 INFO [finetune.py:992] (1/2) Epoch 12, batch 9200, loss[loss=0.1491, simple_loss=0.2392, pruned_loss=0.02947, over 12253.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2565, pruned_loss=0.03951, over 2380660.17 frames. ], batch size: 32, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:35:56,673 INFO [finetune.py:992] (1/2) Epoch 12, batch 9250, loss[loss=0.1763, simple_loss=0.2759, pruned_loss=0.03832, over 12072.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2566, pruned_loss=0.03931, over 2387154.63 frames. ], batch size: 40, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:36:26,113 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9347, 2.3400, 3.4281, 2.8789, 3.2591, 3.0416, 2.3278, 3.3164], device='cuda:1'), covar=tensor([0.0134, 0.0387, 0.0147, 0.0280, 0.0166, 0.0195, 0.0382, 0.0138], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0207, 0.0191, 0.0190, 0.0218, 0.0166, 0.0199, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:36:31,437 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.684e+02 3.034e+02 3.618e+02 6.950e+02, threshold=6.067e+02, percent-clipped=3.0 2023-05-16 20:36:32,180 INFO [finetune.py:992] (1/2) Epoch 12, batch 9300, loss[loss=0.1765, simple_loss=0.2678, pruned_loss=0.04258, over 12159.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2557, pruned_loss=0.03894, over 2384921.57 frames. ], batch size: 34, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:37:09,172 INFO [finetune.py:992] (1/2) Epoch 12, batch 9350, loss[loss=0.2231, simple_loss=0.3055, pruned_loss=0.07038, over 8264.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2569, pruned_loss=0.03949, over 2374846.63 frames. ], batch size: 100, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:37:26,885 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243949.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:37:29,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 20:37:31,141 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0455, 6.0314, 5.8582, 5.2683, 5.2500, 5.9419, 5.5541, 5.3943], device='cuda:1'), covar=tensor([0.0845, 0.0913, 0.0650, 0.1560, 0.0688, 0.0804, 0.1676, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0626, 0.0567, 0.0523, 0.0639, 0.0414, 0.0720, 0.0787, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 20:37:43,792 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.953e+02 3.600e+02 4.345e+02 1.678e+03, threshold=7.200e+02, percent-clipped=9.0 2023-05-16 20:37:44,510 INFO [finetune.py:992] (1/2) Epoch 12, batch 9400, loss[loss=0.1689, simple_loss=0.2667, pruned_loss=0.03556, over 12301.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2578, pruned_loss=0.03982, over 2365527.69 frames. ], batch size: 34, lr: 3.77e-03, grad_scale: 16.0 2023-05-16 20:38:00,840 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243997.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:38:07,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9231, 4.3740, 4.3619, 4.8216, 3.6087, 4.3728, 3.0191, 4.4763], device='cuda:1'), covar=tensor([0.1213, 0.0513, 0.0733, 0.0473, 0.0867, 0.0452, 0.1507, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0264, 0.0297, 0.0357, 0.0240, 0.0241, 0.0261, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:38:09,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 20:38:24,494 INFO [finetune.py:992] (1/2) Epoch 12, batch 9450, loss[loss=0.1468, simple_loss=0.2329, pruned_loss=0.03031, over 11411.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2576, pruned_loss=0.03947, over 2371991.04 frames. ], batch size: 25, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:38:59,123 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0430, 5.7198, 5.3966, 5.2401, 5.8666, 5.1305, 5.3765, 5.3155], device='cuda:1'), covar=tensor([0.1680, 0.1106, 0.1037, 0.2450, 0.1095, 0.2584, 0.2041, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0505, 0.0398, 0.0450, 0.0476, 0.0437, 0.0399, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:39:00,429 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 2.575e+02 3.107e+02 4.027e+02 6.079e+02, threshold=6.214e+02, percent-clipped=0.0 2023-05-16 20:39:01,151 INFO [finetune.py:992] (1/2) Epoch 12, batch 9500, loss[loss=0.1521, simple_loss=0.252, pruned_loss=0.02611, over 12141.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2576, pruned_loss=0.03938, over 2374517.07 frames. ], batch size: 34, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:39:11,533 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0252, 3.3912, 5.3141, 2.8764, 2.9331, 3.8011, 3.3750, 3.9272], device='cuda:1'), covar=tensor([0.0344, 0.1190, 0.0297, 0.1093, 0.1880, 0.1548, 0.1276, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0236, 0.0252, 0.0182, 0.0239, 0.0296, 0.0223, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:39:27,834 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1648, 6.1527, 5.9376, 5.3552, 5.2925, 6.0400, 5.6656, 5.4655], device='cuda:1'), covar=tensor([0.0749, 0.0883, 0.0658, 0.1595, 0.0640, 0.0699, 0.1613, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0567, 0.0525, 0.0642, 0.0416, 0.0724, 0.0790, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 20:39:37,079 INFO [finetune.py:992] (1/2) Epoch 12, batch 9550, loss[loss=0.1615, simple_loss=0.2558, pruned_loss=0.0336, over 12367.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2575, pruned_loss=0.03931, over 2378914.37 frames. ], batch size: 38, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:40:12,225 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4297, 2.4991, 3.2561, 4.2849, 2.3561, 4.4386, 4.3956, 4.4386], device='cuda:1'), covar=tensor([0.0135, 0.1268, 0.0486, 0.0169, 0.1311, 0.0198, 0.0164, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0205, 0.0185, 0.0121, 0.0192, 0.0184, 0.0179, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:40:12,707 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.660e+02 3.055e+02 3.615e+02 7.485e+02, threshold=6.109e+02, percent-clipped=2.0 2023-05-16 20:40:13,436 INFO [finetune.py:992] (1/2) Epoch 12, batch 9600, loss[loss=0.1598, simple_loss=0.2483, pruned_loss=0.03571, over 12094.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2571, pruned_loss=0.03928, over 2382023.38 frames. ], batch size: 32, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:40:41,237 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1848, 4.4646, 4.0895, 4.9108, 4.4683, 2.7802, 4.1129, 3.1402], device='cuda:1'), covar=tensor([0.0882, 0.0911, 0.1407, 0.0427, 0.1024, 0.1753, 0.1102, 0.3154], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0381, 0.0357, 0.0313, 0.0368, 0.0270, 0.0344, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:40:50,226 INFO [finetune.py:992] (1/2) Epoch 12, batch 9650, loss[loss=0.1589, simple_loss=0.2518, pruned_loss=0.03297, over 12411.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2567, pruned_loss=0.03889, over 2387550.15 frames. ], batch size: 32, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:41:22,564 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7855, 2.8077, 3.9634, 4.6698, 4.1559, 4.6448, 4.1120, 3.5010], device='cuda:1'), covar=tensor([0.0036, 0.0382, 0.0121, 0.0036, 0.0097, 0.0071, 0.0109, 0.0309], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0125, 0.0106, 0.0079, 0.0105, 0.0117, 0.0097, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:41:22,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 20:41:25,338 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.732e+02 3.120e+02 3.808e+02 5.883e+02, threshold=6.241e+02, percent-clipped=0.0 2023-05-16 20:41:26,076 INFO [finetune.py:992] (1/2) Epoch 12, batch 9700, loss[loss=0.2464, simple_loss=0.3166, pruned_loss=0.08814, over 8082.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2568, pruned_loss=0.03912, over 2385233.84 frames. ], batch size: 99, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:41:41,121 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1067, 4.8043, 5.1267, 5.0430, 4.2735, 4.4436, 4.4905, 4.8138], device='cuda:1'), covar=tensor([0.1010, 0.1159, 0.0960, 0.0947, 0.3547, 0.2151, 0.0858, 0.1913], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0714, 0.0618, 0.0631, 0.0855, 0.0750, 0.0559, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:42:02,518 INFO [finetune.py:992] (1/2) Epoch 12, batch 9750, loss[loss=0.1751, simple_loss=0.2694, pruned_loss=0.04043, over 12341.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2569, pruned_loss=0.03917, over 2385546.43 frames. ], batch size: 36, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:42:23,995 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=244353.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:42:38,002 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.796e+02 3.294e+02 3.819e+02 9.587e+02, threshold=6.587e+02, percent-clipped=2.0 2023-05-16 20:42:38,711 INFO [finetune.py:992] (1/2) Epoch 12, batch 9800, loss[loss=0.18, simple_loss=0.2776, pruned_loss=0.04124, over 12158.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2569, pruned_loss=0.03931, over 2381590.15 frames. ], batch size: 36, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:42:40,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-05-16 20:42:51,712 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3875, 4.7206, 3.0368, 2.7773, 4.0448, 2.5613, 4.0438, 3.1448], device='cuda:1'), covar=tensor([0.0680, 0.0616, 0.1037, 0.1405, 0.0304, 0.1348, 0.0472, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0258, 0.0176, 0.0200, 0.0143, 0.0181, 0.0200, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:42:56,178 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1487, 3.8364, 3.9683, 4.3572, 3.1170, 3.9059, 2.5007, 4.0934], device='cuda:1'), covar=tensor([0.1680, 0.0761, 0.0891, 0.0685, 0.1117, 0.0597, 0.1928, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0263, 0.0296, 0.0357, 0.0241, 0.0241, 0.0261, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:43:07,863 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=244414.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:43:14,899 INFO [finetune.py:992] (1/2) Epoch 12, batch 9850, loss[loss=0.1636, simple_loss=0.2607, pruned_loss=0.03323, over 12150.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2561, pruned_loss=0.03894, over 2388676.31 frames. ], batch size: 36, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:43:26,835 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-16 20:43:50,739 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.647e+02 3.033e+02 3.603e+02 6.141e+02, threshold=6.066e+02, percent-clipped=0.0 2023-05-16 20:43:51,449 INFO [finetune.py:992] (1/2) Epoch 12, batch 9900, loss[loss=0.1824, simple_loss=0.2868, pruned_loss=0.03905, over 12356.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2564, pruned_loss=0.03909, over 2381423.01 frames. ], batch size: 36, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:43:57,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-16 20:44:28,026 INFO [finetune.py:992] (1/2) Epoch 12, batch 9950, loss[loss=0.1668, simple_loss=0.2648, pruned_loss=0.03439, over 12103.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.257, pruned_loss=0.03975, over 2365645.77 frames. ], batch size: 39, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:45:02,736 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.695e+02 3.297e+02 3.922e+02 9.943e+02, threshold=6.593e+02, percent-clipped=6.0 2023-05-16 20:45:03,465 INFO [finetune.py:992] (1/2) Epoch 12, batch 10000, loss[loss=0.1657, simple_loss=0.2539, pruned_loss=0.03875, over 11880.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2565, pruned_loss=0.03955, over 2364646.94 frames. ], batch size: 44, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:45:39,700 INFO [finetune.py:992] (1/2) Epoch 12, batch 10050, loss[loss=0.1675, simple_loss=0.2588, pruned_loss=0.03813, over 12312.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2563, pruned_loss=0.03938, over 2376763.52 frames. ], batch size: 34, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:46:15,311 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.632e+02 3.142e+02 3.893e+02 5.782e+02, threshold=6.284e+02, percent-clipped=0.0 2023-05-16 20:46:16,039 INFO [finetune.py:992] (1/2) Epoch 12, batch 10100, loss[loss=0.1757, simple_loss=0.2639, pruned_loss=0.04371, over 12150.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2554, pruned_loss=0.03879, over 2382019.44 frames. ], batch size: 34, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:46:33,316 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2525, 4.7523, 4.0797, 5.0490, 4.4549, 2.6473, 4.1062, 3.0023], device='cuda:1'), covar=tensor([0.0784, 0.0713, 0.1464, 0.0480, 0.1186, 0.1779, 0.1188, 0.3283], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0378, 0.0355, 0.0312, 0.0366, 0.0269, 0.0344, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:46:40,980 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=244709.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:46:46,682 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=244717.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:46:51,498 INFO [finetune.py:992] (1/2) Epoch 12, batch 10150, loss[loss=0.1399, simple_loss=0.224, pruned_loss=0.02788, over 12168.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2545, pruned_loss=0.03848, over 2378782.06 frames. ], batch size: 29, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:47:12,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-16 20:47:27,287 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.627e+02 3.161e+02 3.759e+02 7.063e+02, threshold=6.321e+02, percent-clipped=1.0 2023-05-16 20:47:27,951 INFO [finetune.py:992] (1/2) Epoch 12, batch 10200, loss[loss=0.201, simple_loss=0.2878, pruned_loss=0.05713, over 11612.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2548, pruned_loss=0.03843, over 2382200.86 frames. ], batch size: 48, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:47:31,214 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=244778.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:47:32,677 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1089, 2.3974, 3.6318, 3.0437, 3.4873, 3.1625, 2.4147, 3.4607], device='cuda:1'), covar=tensor([0.0149, 0.0395, 0.0129, 0.0220, 0.0144, 0.0177, 0.0375, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0206, 0.0190, 0.0187, 0.0218, 0.0166, 0.0197, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:47:52,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5520, 2.6389, 3.4287, 4.4357, 2.3753, 4.4223, 4.5088, 4.6725], device='cuda:1'), covar=tensor([0.0140, 0.1268, 0.0452, 0.0155, 0.1373, 0.0261, 0.0177, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0201, 0.0183, 0.0120, 0.0190, 0.0182, 0.0177, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:48:04,515 INFO [finetune.py:992] (1/2) Epoch 12, batch 10250, loss[loss=0.1705, simple_loss=0.2628, pruned_loss=0.03915, over 12040.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.255, pruned_loss=0.03826, over 2385377.59 frames. ], batch size: 40, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:48:39,154 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.708e+02 3.036e+02 3.649e+02 6.724e+02, threshold=6.073e+02, percent-clipped=1.0 2023-05-16 20:48:39,914 INFO [finetune.py:992] (1/2) Epoch 12, batch 10300, loss[loss=0.1356, simple_loss=0.2162, pruned_loss=0.02754, over 12278.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2546, pruned_loss=0.03794, over 2391883.99 frames. ], batch size: 28, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:48:46,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 20:48:49,230 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=244887.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:48:53,970 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6417, 5.2292, 5.6158, 4.9682, 5.2872, 5.0648, 5.6458, 5.2242], device='cuda:1'), covar=tensor([0.0233, 0.0291, 0.0207, 0.0216, 0.0306, 0.0281, 0.0173, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0267, 0.0292, 0.0263, 0.0265, 0.0266, 0.0239, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:49:16,452 INFO [finetune.py:992] (1/2) Epoch 12, batch 10350, loss[loss=0.171, simple_loss=0.2658, pruned_loss=0.03811, over 12158.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.256, pruned_loss=0.03892, over 2376023.71 frames. ], batch size: 34, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:49:33,684 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=244948.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:49:37,926 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=244954.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:49:39,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4222, 3.6632, 3.1920, 3.1519, 2.9361, 2.6286, 3.7725, 2.2893], device='cuda:1'), covar=tensor([0.0406, 0.0155, 0.0224, 0.0210, 0.0406, 0.0419, 0.0132, 0.0509], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0163, 0.0165, 0.0191, 0.0204, 0.0201, 0.0174, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:49:51,294 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.813e+02 2.759e+02 3.200e+02 3.746e+02 7.063e+02, threshold=6.400e+02, percent-clipped=2.0 2023-05-16 20:49:52,031 INFO [finetune.py:992] (1/2) Epoch 12, batch 10400, loss[loss=0.1575, simple_loss=0.2504, pruned_loss=0.03233, over 12203.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2562, pruned_loss=0.0391, over 2377460.83 frames. ], batch size: 35, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:50:12,997 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3725, 4.6859, 2.9226, 2.7891, 3.9457, 2.5575, 4.0526, 3.3488], device='cuda:1'), covar=tensor([0.0650, 0.0506, 0.1038, 0.1397, 0.0329, 0.1358, 0.0466, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0261, 0.0179, 0.0203, 0.0145, 0.0183, 0.0202, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:50:17,153 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245009.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:50:21,430 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245015.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:50:27,509 INFO [finetune.py:992] (1/2) Epoch 12, batch 10450, loss[loss=0.1976, simple_loss=0.2904, pruned_loss=0.05242, over 10346.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2572, pruned_loss=0.03955, over 2365140.59 frames. ], batch size: 68, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:50:51,729 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245057.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:51:03,630 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.694e+02 3.120e+02 3.983e+02 1.025e+03, threshold=6.241e+02, percent-clipped=4.0 2023-05-16 20:51:03,790 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245073.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:51:04,385 INFO [finetune.py:992] (1/2) Epoch 12, batch 10500, loss[loss=0.1627, simple_loss=0.2424, pruned_loss=0.04153, over 12278.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2556, pruned_loss=0.03862, over 2373243.83 frames. ], batch size: 28, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:51:22,106 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.92 vs. limit=5.0 2023-05-16 20:51:40,157 INFO [finetune.py:992] (1/2) Epoch 12, batch 10550, loss[loss=0.2419, simple_loss=0.3125, pruned_loss=0.08572, over 8100.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2563, pruned_loss=0.03921, over 2361695.70 frames. ], batch size: 99, lr: 3.76e-03, grad_scale: 16.0 2023-05-16 20:51:53,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 20:52:14,923 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.721e+02 3.204e+02 3.875e+02 6.361e+02, threshold=6.407e+02, percent-clipped=1.0 2023-05-16 20:52:15,672 INFO [finetune.py:992] (1/2) Epoch 12, batch 10600, loss[loss=0.144, simple_loss=0.2221, pruned_loss=0.03293, over 12209.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2568, pruned_loss=0.03932, over 2365985.73 frames. ], batch size: 29, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:52:53,151 INFO [finetune.py:992] (1/2) Epoch 12, batch 10650, loss[loss=0.161, simple_loss=0.2469, pruned_loss=0.03755, over 12011.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2568, pruned_loss=0.03906, over 2366963.82 frames. ], batch size: 31, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:53:06,611 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245243.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:53:11,926 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9182, 2.3366, 3.5907, 2.9449, 3.5124, 3.0580, 2.3093, 3.4144], device='cuda:1'), covar=tensor([0.0208, 0.0470, 0.0210, 0.0302, 0.0162, 0.0234, 0.0463, 0.0164], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0209, 0.0193, 0.0190, 0.0221, 0.0168, 0.0200, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:53:28,251 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.009e+02 2.654e+02 3.047e+02 3.752e+02 8.474e+02, threshold=6.094e+02, percent-clipped=2.0 2023-05-16 20:53:28,993 INFO [finetune.py:992] (1/2) Epoch 12, batch 10700, loss[loss=0.1633, simple_loss=0.264, pruned_loss=0.03124, over 12363.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2571, pruned_loss=0.03897, over 2366681.46 frames. ], batch size: 35, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:53:53,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4738, 2.5734, 3.6823, 4.3588, 3.8206, 4.3568, 3.7550, 3.1160], device='cuda:1'), covar=tensor([0.0034, 0.0401, 0.0137, 0.0046, 0.0116, 0.0084, 0.0134, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0127, 0.0108, 0.0080, 0.0106, 0.0120, 0.0099, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:53:54,465 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245310.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:54:04,501 INFO [finetune.py:992] (1/2) Epoch 12, batch 10750, loss[loss=0.1686, simple_loss=0.2633, pruned_loss=0.03697, over 12047.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2573, pruned_loss=0.03891, over 2369235.10 frames. ], batch size: 42, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:54:14,641 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0546, 3.6237, 3.8370, 4.3060, 2.8847, 3.7376, 2.3825, 3.9691], device='cuda:1'), covar=tensor([0.1644, 0.0904, 0.1029, 0.0648, 0.1158, 0.0654, 0.2001, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0267, 0.0300, 0.0362, 0.0243, 0.0243, 0.0265, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:54:16,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4689, 2.5447, 3.2507, 4.2751, 2.3460, 4.3878, 4.4515, 4.5609], device='cuda:1'), covar=tensor([0.0145, 0.1298, 0.0506, 0.0201, 0.1393, 0.0206, 0.0151, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0203, 0.0184, 0.0121, 0.0189, 0.0182, 0.0177, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:54:33,166 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6130, 3.8022, 3.3875, 3.3361, 3.1513, 2.9087, 3.8347, 2.6241], device='cuda:1'), covar=tensor([0.0391, 0.0117, 0.0213, 0.0194, 0.0336, 0.0351, 0.0128, 0.0426], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0163, 0.0166, 0.0191, 0.0204, 0.0201, 0.0174, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:54:40,587 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 2.769e+02 3.205e+02 3.871e+02 6.976e+02, threshold=6.409e+02, percent-clipped=1.0 2023-05-16 20:54:40,730 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245373.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:54:41,350 INFO [finetune.py:992] (1/2) Epoch 12, batch 10800, loss[loss=0.1774, simple_loss=0.268, pruned_loss=0.04337, over 12297.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2575, pruned_loss=0.03915, over 2371824.81 frames. ], batch size: 34, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:54:48,607 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2670, 4.6925, 4.1806, 5.0368, 4.4959, 2.9240, 4.3009, 3.1043], device='cuda:1'), covar=tensor([0.0850, 0.0761, 0.1408, 0.0453, 0.1264, 0.1654, 0.1117, 0.3129], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0378, 0.0356, 0.0311, 0.0366, 0.0268, 0.0344, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:54:54,150 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2302, 3.9708, 4.1207, 4.5617, 3.0365, 3.9498, 2.6760, 4.1707], device='cuda:1'), covar=tensor([0.1548, 0.0674, 0.0778, 0.0599, 0.1079, 0.0567, 0.1651, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0267, 0.0300, 0.0362, 0.0243, 0.0243, 0.0264, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 20:55:14,322 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0371, 5.7793, 5.2839, 5.3494, 5.9528, 5.2697, 5.4401, 5.3564], device='cuda:1'), covar=tensor([0.1401, 0.1092, 0.1094, 0.1889, 0.0947, 0.2287, 0.1773, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0503, 0.0399, 0.0452, 0.0476, 0.0443, 0.0398, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:55:15,021 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245421.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:55:16,999 INFO [finetune.py:992] (1/2) Epoch 12, batch 10850, loss[loss=0.2038, simple_loss=0.2889, pruned_loss=0.05929, over 12086.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2572, pruned_loss=0.03925, over 2372619.08 frames. ], batch size: 42, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:55:22,189 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0640, 2.2018, 2.6055, 3.0292, 2.1797, 3.1595, 3.0687, 3.2580], device='cuda:1'), covar=tensor([0.0180, 0.1045, 0.0497, 0.0243, 0.1057, 0.0325, 0.0322, 0.0150], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0203, 0.0184, 0.0122, 0.0190, 0.0182, 0.0178, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:55:43,544 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245460.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:55:53,051 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.872e+02 2.731e+02 3.179e+02 3.624e+02 5.889e+02, threshold=6.357e+02, percent-clipped=0.0 2023-05-16 20:55:53,708 INFO [finetune.py:992] (1/2) Epoch 12, batch 10900, loss[loss=0.1937, simple_loss=0.2808, pruned_loss=0.05329, over 12033.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2583, pruned_loss=0.03983, over 2364469.28 frames. ], batch size: 37, lr: 3.75e-03, grad_scale: 32.0 2023-05-16 20:56:28,718 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245521.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:56:30,642 INFO [finetune.py:992] (1/2) Epoch 12, batch 10950, loss[loss=0.133, simple_loss=0.2171, pruned_loss=0.02446, over 12007.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2585, pruned_loss=0.04045, over 2357858.15 frames. ], batch size: 28, lr: 3.75e-03, grad_scale: 32.0 2023-05-16 20:56:44,453 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245543.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:56:46,442 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245546.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:57:02,192 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4779, 2.5373, 3.2959, 4.3505, 2.4098, 4.3778, 4.4344, 4.5450], device='cuda:1'), covar=tensor([0.0116, 0.1219, 0.0476, 0.0160, 0.1256, 0.0211, 0.0161, 0.0096], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0201, 0.0182, 0.0121, 0.0188, 0.0181, 0.0176, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:57:05,480 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.160e+02 2.725e+02 3.222e+02 4.178e+02 8.072e+02, threshold=6.445e+02, percent-clipped=5.0 2023-05-16 20:57:06,127 INFO [finetune.py:992] (1/2) Epoch 12, batch 11000, loss[loss=0.249, simple_loss=0.3447, pruned_loss=0.07666, over 10748.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2612, pruned_loss=0.04196, over 2338591.62 frames. ], batch size: 70, lr: 3.75e-03, grad_scale: 32.0 2023-05-16 20:57:14,780 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4538, 2.6080, 3.2423, 4.3436, 2.4459, 4.4239, 4.5086, 4.5748], device='cuda:1'), covar=tensor([0.0139, 0.1149, 0.0471, 0.0192, 0.1241, 0.0246, 0.0136, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0200, 0.0182, 0.0121, 0.0188, 0.0180, 0.0176, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 20:57:18,334 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245591.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:57:24,094 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8579, 3.2722, 2.4652, 2.1526, 2.8745, 2.3122, 3.1425, 2.6144], device='cuda:1'), covar=tensor([0.0632, 0.0828, 0.0991, 0.1481, 0.0360, 0.1169, 0.0553, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0266, 0.0182, 0.0206, 0.0146, 0.0185, 0.0205, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:57:30,003 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245607.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 20:57:32,710 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245610.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:57:42,943 INFO [finetune.py:992] (1/2) Epoch 12, batch 11050, loss[loss=0.2221, simple_loss=0.301, pruned_loss=0.07158, over 7939.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2646, pruned_loss=0.04383, over 2307543.42 frames. ], batch size: 97, lr: 3.75e-03, grad_scale: 32.0 2023-05-16 20:58:06,920 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245658.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:58:13,883 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3634, 6.1472, 5.8118, 5.7306, 6.2434, 5.6088, 5.8020, 5.7045], device='cuda:1'), covar=tensor([0.1519, 0.1072, 0.1168, 0.2025, 0.0967, 0.2299, 0.1702, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0498, 0.0393, 0.0446, 0.0472, 0.0440, 0.0393, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 20:58:17,295 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.303e+02 3.015e+02 3.784e+02 4.411e+02 6.473e+02, threshold=7.569e+02, percent-clipped=1.0 2023-05-16 20:58:17,989 INFO [finetune.py:992] (1/2) Epoch 12, batch 11100, loss[loss=0.2309, simple_loss=0.3111, pruned_loss=0.0754, over 8233.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2675, pruned_loss=0.04578, over 2267793.34 frames. ], batch size: 100, lr: 3.75e-03, grad_scale: 32.0 2023-05-16 20:58:21,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-05-16 20:58:26,558 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245686.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 20:58:52,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-05-16 20:58:54,210 INFO [finetune.py:992] (1/2) Epoch 12, batch 11150, loss[loss=0.2918, simple_loss=0.3551, pruned_loss=0.1142, over 7139.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2715, pruned_loss=0.04884, over 2212374.73 frames. ], batch size: 98, lr: 3.75e-03, grad_scale: 32.0 2023-05-16 20:59:10,505 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245747.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 20:59:14,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.21 vs. limit=5.0 2023-05-16 20:59:29,231 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.300e+02 3.344e+02 3.973e+02 4.906e+02 8.289e+02, threshold=7.946e+02, percent-clipped=2.0 2023-05-16 20:59:29,250 INFO [finetune.py:992] (1/2) Epoch 12, batch 11200, loss[loss=0.3224, simple_loss=0.3787, pruned_loss=0.133, over 6988.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2773, pruned_loss=0.0523, over 2150613.02 frames. ], batch size: 100, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 20:59:43,288 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245792.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 20:59:45,403 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3859, 4.2685, 4.3034, 4.3450, 3.9786, 4.5123, 4.4768, 4.5638], device='cuda:1'), covar=tensor([0.0220, 0.0169, 0.0177, 0.0410, 0.0664, 0.0257, 0.0165, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0193, 0.0186, 0.0242, 0.0237, 0.0213, 0.0173, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 20:59:48,226 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0329, 4.9886, 4.9423, 5.0167, 4.6206, 5.0853, 5.0707, 5.2811], device='cuda:1'), covar=tensor([0.0218, 0.0153, 0.0164, 0.0306, 0.0714, 0.0413, 0.0162, 0.0170], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0193, 0.0186, 0.0242, 0.0237, 0.0213, 0.0173, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 21:00:00,167 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245816.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:00:05,660 INFO [finetune.py:992] (1/2) Epoch 12, batch 11250, loss[loss=0.1775, simple_loss=0.265, pruned_loss=0.04507, over 12194.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2841, pruned_loss=0.05672, over 2094179.33 frames. ], batch size: 31, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:00:06,304 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 21:00:15,738 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9686, 2.1664, 2.2564, 2.2020, 2.1158, 2.0346, 2.2049, 1.6432], device='cuda:1'), covar=tensor([0.0286, 0.0160, 0.0152, 0.0189, 0.0235, 0.0192, 0.0169, 0.0357], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0161, 0.0162, 0.0189, 0.0202, 0.0198, 0.0172, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:00:26,492 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245853.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:00:26,568 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6632, 4.0146, 3.6239, 4.2478, 3.7095, 2.6554, 3.7228, 2.8173], device='cuda:1'), covar=tensor([0.0973, 0.0867, 0.1385, 0.0574, 0.1541, 0.1733, 0.1129, 0.3223], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0368, 0.0346, 0.0302, 0.0355, 0.0262, 0.0334, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:00:41,266 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.328e+02 3.283e+02 4.161e+02 5.005e+02 1.339e+03, threshold=8.323e+02, percent-clipped=2.0 2023-05-16 21:00:41,284 INFO [finetune.py:992] (1/2) Epoch 12, batch 11300, loss[loss=0.2968, simple_loss=0.3532, pruned_loss=0.1203, over 6603.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2907, pruned_loss=0.06088, over 2031476.10 frames. ], batch size: 98, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:00:52,919 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245891.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:00:53,632 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5161, 4.1811, 4.1143, 4.5185, 3.2588, 4.0717, 2.9349, 4.1480], device='cuda:1'), covar=tensor([0.1513, 0.0685, 0.0926, 0.0733, 0.1126, 0.0594, 0.1623, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0264, 0.0295, 0.0355, 0.0240, 0.0240, 0.0260, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:01:00,481 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245902.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 21:01:15,888 INFO [finetune.py:992] (1/2) Epoch 12, batch 11350, loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09987, over 7431.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2963, pruned_loss=0.06446, over 1972715.72 frames. ], batch size: 98, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:01:18,304 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-16 21:01:35,697 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245952.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:01:50,524 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.992e+02 3.504e+02 4.241e+02 5.083e+02 1.276e+03, threshold=8.481e+02, percent-clipped=3.0 2023-05-16 21:01:50,542 INFO [finetune.py:992] (1/2) Epoch 12, batch 11400, loss[loss=0.242, simple_loss=0.3185, pruned_loss=0.08275, over 6445.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.3012, pruned_loss=0.06787, over 1927505.66 frames. ], batch size: 98, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:02:16,184 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7827, 2.2344, 2.5446, 2.8618, 2.2371, 2.8942, 2.8260, 2.9760], device='cuda:1'), covar=tensor([0.0172, 0.0973, 0.0469, 0.0238, 0.1061, 0.0270, 0.0327, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0198, 0.0179, 0.0119, 0.0187, 0.0177, 0.0172, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:02:29,779 INFO [finetune.py:992] (1/2) Epoch 12, batch 11450, loss[loss=0.2813, simple_loss=0.3472, pruned_loss=0.1077, over 6818.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.3029, pruned_loss=0.06929, over 1900131.69 frames. ], batch size: 98, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:02:41,846 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246042.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 21:02:59,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 21:03:03,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-05-16 21:03:04,525 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.706e+02 3.486e+02 3.950e+02 4.755e+02 6.665e+02, threshold=7.899e+02, percent-clipped=0.0 2023-05-16 21:03:04,544 INFO [finetune.py:992] (1/2) Epoch 12, batch 11500, loss[loss=0.2528, simple_loss=0.3271, pruned_loss=0.08925, over 6970.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3063, pruned_loss=0.07245, over 1854286.22 frames. ], batch size: 99, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:03:07,729 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-05-16 21:03:30,598 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246111.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:03:34,674 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246116.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:03:40,072 INFO [finetune.py:992] (1/2) Epoch 12, batch 11550, loss[loss=0.2047, simple_loss=0.2853, pruned_loss=0.0621, over 12251.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3083, pruned_loss=0.07444, over 1820327.13 frames. ], batch size: 28, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:03:46,451 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8937, 2.1638, 2.8513, 2.7976, 2.9705, 2.9065, 2.9273, 2.3622], device='cuda:1'), covar=tensor([0.0071, 0.0383, 0.0164, 0.0092, 0.0113, 0.0111, 0.0129, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0123, 0.0103, 0.0077, 0.0102, 0.0115, 0.0095, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:03:56,257 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246148.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:04:07,117 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246164.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:04:12,597 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246172.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:04:14,366 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.699e+02 3.686e+02 4.061e+02 4.741e+02 8.417e+02, threshold=8.121e+02, percent-clipped=2.0 2023-05-16 21:04:14,385 INFO [finetune.py:992] (1/2) Epoch 12, batch 11600, loss[loss=0.2212, simple_loss=0.3045, pruned_loss=0.06892, over 10430.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3094, pruned_loss=0.07558, over 1794437.79 frames. ], batch size: 68, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:04:30,350 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246196.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 21:04:35,056 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246202.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 21:04:49,003 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246221.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:04:51,084 INFO [finetune.py:992] (1/2) Epoch 12, batch 11650, loss[loss=0.2087, simple_loss=0.2939, pruned_loss=0.0617, over 10391.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3092, pruned_loss=0.07623, over 1765874.59 frames. ], batch size: 68, lr: 3.75e-03, grad_scale: 16.0 2023-05-16 21:05:08,442 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246247.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:05:10,588 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246250.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:05:15,226 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246257.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 21:05:26,398 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.526e+02 3.461e+02 3.908e+02 4.540e+02 6.500e+02, threshold=7.816e+02, percent-clipped=0.0 2023-05-16 21:05:26,427 INFO [finetune.py:992] (1/2) Epoch 12, batch 11700, loss[loss=0.2343, simple_loss=0.3117, pruned_loss=0.0784, over 7089.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3087, pruned_loss=0.07638, over 1746945.66 frames. ], batch size: 98, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:05:32,035 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246282.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:05:36,115 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7732, 3.7769, 3.7771, 3.8646, 3.6618, 3.7119, 3.5843, 3.7587], device='cuda:1'), covar=tensor([0.1271, 0.0725, 0.1288, 0.0738, 0.1669, 0.1243, 0.0625, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0669, 0.0582, 0.0597, 0.0791, 0.0701, 0.0526, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:06:01,226 INFO [finetune.py:992] (1/2) Epoch 12, batch 11750, loss[loss=0.2442, simple_loss=0.3095, pruned_loss=0.08943, over 6857.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3093, pruned_loss=0.07741, over 1724060.27 frames. ], batch size: 99, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:06:13,557 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246342.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 21:06:31,192 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8625, 3.7289, 3.8505, 3.6070, 3.7533, 3.6072, 3.8078, 3.5078], device='cuda:1'), covar=tensor([0.0414, 0.0380, 0.0378, 0.0277, 0.0374, 0.0335, 0.0393, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0245, 0.0267, 0.0243, 0.0242, 0.0245, 0.0219, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:06:36,454 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.683e+02 3.527e+02 4.253e+02 4.983e+02 1.025e+03, threshold=8.505e+02, percent-clipped=3.0 2023-05-16 21:06:36,473 INFO [finetune.py:992] (1/2) Epoch 12, batch 11800, loss[loss=0.2481, simple_loss=0.3177, pruned_loss=0.08919, over 6764.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3114, pruned_loss=0.07922, over 1703089.70 frames. ], batch size: 98, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:06:47,420 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246390.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 21:07:11,785 INFO [finetune.py:992] (1/2) Epoch 12, batch 11850, loss[loss=0.2728, simple_loss=0.3418, pruned_loss=0.1019, over 6336.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3133, pruned_loss=0.0796, over 1701114.34 frames. ], batch size: 98, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:07:13,337 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7930, 3.8007, 3.7625, 3.8726, 3.6974, 3.7256, 3.6151, 3.7787], device='cuda:1'), covar=tensor([0.1304, 0.0653, 0.1603, 0.0689, 0.1369, 0.1061, 0.0513, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0657, 0.0572, 0.0586, 0.0774, 0.0688, 0.0517, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:07:23,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-16 21:07:28,162 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246448.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:07:37,862 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9860, 2.1518, 2.2661, 2.2400, 2.0891, 1.9085, 2.2124, 1.6408], device='cuda:1'), covar=tensor([0.0339, 0.0199, 0.0183, 0.0210, 0.0302, 0.0238, 0.0184, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0153, 0.0154, 0.0181, 0.0193, 0.0189, 0.0164, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:07:41,033 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246467.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:07:41,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-16 21:07:43,763 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3681, 2.9087, 3.7272, 2.2290, 2.4356, 3.0488, 2.8570, 3.1989], device='cuda:1'), covar=tensor([0.0584, 0.1138, 0.0315, 0.1520, 0.1945, 0.1352, 0.1311, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0225, 0.0233, 0.0175, 0.0228, 0.0277, 0.0212, 0.0250], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:07:44,350 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2802, 4.9083, 4.9829, 5.1793, 5.0198, 5.1241, 5.2283, 2.5722], device='cuda:1'), covar=tensor([0.0061, 0.0066, 0.0093, 0.0053, 0.0044, 0.0098, 0.0057, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0076, 0.0079, 0.0071, 0.0058, 0.0089, 0.0079, 0.0094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:07:45,471 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.582e+02 3.580e+02 4.280e+02 5.047e+02 8.856e+02, threshold=8.560e+02, percent-clipped=2.0 2023-05-16 21:07:45,490 INFO [finetune.py:992] (1/2) Epoch 12, batch 11900, loss[loss=0.264, simple_loss=0.3299, pruned_loss=0.09908, over 6948.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3134, pruned_loss=0.07915, over 1683837.98 frames. ], batch size: 100, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:08:01,864 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246496.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:08:09,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.89 vs. limit=5.0 2023-05-16 21:08:20,797 INFO [finetune.py:992] (1/2) Epoch 12, batch 11950, loss[loss=0.1952, simple_loss=0.2824, pruned_loss=0.05406, over 11087.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3098, pruned_loss=0.07602, over 1680845.16 frames. ], batch size: 55, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:08:37,886 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246547.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:08:41,235 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246552.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 21:08:55,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7858, 3.8063, 3.7851, 3.8833, 3.7056, 3.7559, 3.6136, 3.7989], device='cuda:1'), covar=tensor([0.1497, 0.0736, 0.1386, 0.0739, 0.1668, 0.1119, 0.0557, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0649, 0.0566, 0.0581, 0.0763, 0.0680, 0.0510, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:08:55,984 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.981e+02 3.497e+02 4.415e+02 8.803e+02, threshold=6.994e+02, percent-clipped=1.0 2023-05-16 21:08:56,013 INFO [finetune.py:992] (1/2) Epoch 12, batch 12000, loss[loss=0.2047, simple_loss=0.282, pruned_loss=0.06374, over 6751.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3055, pruned_loss=0.0729, over 1659501.16 frames. ], batch size: 98, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:08:56,013 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 21:09:12,631 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3031, 5.2481, 5.3255, 5.3730, 4.9876, 5.1223, 4.9780, 5.1671], device='cuda:1'), covar=tensor([0.0939, 0.0463, 0.0511, 0.0518, 0.1690, 0.1246, 0.0401, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0649, 0.0566, 0.0581, 0.0763, 0.0680, 0.0510, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:09:14,139 INFO [finetune.py:1026] (1/2) Epoch 12, validation: loss=0.2869, simple_loss=0.3623, pruned_loss=0.1058, over 1020973.00 frames. 2023-05-16 21:09:14,140 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 21:09:16,293 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246577.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:09:17,063 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4751, 3.0468, 3.7023, 2.3754, 2.5959, 3.0352, 2.9742, 3.1516], device='cuda:1'), covar=tensor([0.0454, 0.0998, 0.0282, 0.1241, 0.1825, 0.1323, 0.1122, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0224, 0.0231, 0.0175, 0.0227, 0.0275, 0.0211, 0.0248], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:09:29,250 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246595.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:09:49,050 INFO [finetune.py:992] (1/2) Epoch 12, batch 12050, loss[loss=0.2157, simple_loss=0.2912, pruned_loss=0.07017, over 7374.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.301, pruned_loss=0.06937, over 1670643.60 frames. ], batch size: 97, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:10:22,289 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.133e+02 2.926e+02 3.521e+02 4.054e+02 6.335e+02, threshold=7.041e+02, percent-clipped=0.0 2023-05-16 21:10:22,312 INFO [finetune.py:992] (1/2) Epoch 12, batch 12100, loss[loss=0.207, simple_loss=0.2957, pruned_loss=0.05914, over 10332.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2994, pruned_loss=0.06818, over 1672818.89 frames. ], batch size: 69, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:10:24,338 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246677.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:10:35,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-16 21:10:54,393 INFO [finetune.py:992] (1/2) Epoch 12, batch 12150, loss[loss=0.1947, simple_loss=0.2874, pruned_loss=0.05102, over 10345.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.3005, pruned_loss=0.0686, over 1684261.61 frames. ], batch size: 69, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:10:59,660 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8169, 2.2143, 2.7600, 2.7244, 2.8472, 2.9037, 2.7899, 2.4275], device='cuda:1'), covar=tensor([0.0087, 0.0360, 0.0175, 0.0101, 0.0145, 0.0106, 0.0150, 0.0339], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0121, 0.0099, 0.0075, 0.0100, 0.0111, 0.0093, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:11:03,457 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246738.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:11:20,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7669, 3.0726, 2.3714, 2.1589, 2.7199, 2.2647, 2.9578, 2.5681], device='cuda:1'), covar=tensor([0.0632, 0.0570, 0.0974, 0.1515, 0.0263, 0.1320, 0.0519, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0242, 0.0171, 0.0195, 0.0136, 0.0178, 0.0189, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:11:21,807 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246767.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:11:25,894 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 3.313e+02 3.843e+02 4.649e+02 1.350e+03, threshold=7.686e+02, percent-clipped=5.0 2023-05-16 21:11:25,913 INFO [finetune.py:992] (1/2) Epoch 12, batch 12200, loss[loss=0.213, simple_loss=0.2861, pruned_loss=0.06996, over 6511.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.3021, pruned_loss=0.06976, over 1676574.88 frames. ], batch size: 98, lr: 3.74e-03, grad_scale: 16.0 2023-05-16 21:11:37,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 21:12:10,846 INFO [finetune.py:992] (1/2) Epoch 13, batch 0, loss[loss=0.2297, simple_loss=0.2969, pruned_loss=0.08121, over 12129.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.2969, pruned_loss=0.08121, over 12129.00 frames. ], batch size: 38, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:12:10,846 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 21:12:17,308 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4267, 5.8714, 5.6071, 5.5140, 5.9317, 5.2571, 5.2307, 5.5163], device='cuda:1'), covar=tensor([0.0968, 0.0725, 0.0894, 0.1454, 0.0632, 0.2034, 0.2344, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0461, 0.0371, 0.0418, 0.0437, 0.0406, 0.0364, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 21:12:28,456 INFO [finetune.py:1026] (1/2) Epoch 13, validation: loss=0.2846, simple_loss=0.3601, pruned_loss=0.1046, over 1020973.00 frames. 2023-05-16 21:12:28,457 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 21:12:33,563 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246815.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:12:45,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6156, 2.6945, 4.4598, 4.5731, 2.7264, 2.4815, 2.8471, 1.9397], device='cuda:1'), covar=tensor([0.1792, 0.3503, 0.0485, 0.0447, 0.1454, 0.2787, 0.3070, 0.4858], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0369, 0.0260, 0.0286, 0.0259, 0.0295, 0.0370, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:12:51,669 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 21:12:59,878 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246852.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 21:13:01,312 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1663, 5.9111, 5.6149, 5.4771, 6.0100, 5.2982, 5.4932, 5.5545], device='cuda:1'), covar=tensor([0.1542, 0.0933, 0.1115, 0.2004, 0.0884, 0.2297, 0.1941, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0465, 0.0375, 0.0423, 0.0441, 0.0411, 0.0369, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 21:13:02,832 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9263, 4.8653, 4.7767, 4.8021, 4.4779, 4.8369, 4.9032, 5.0953], device='cuda:1'), covar=tensor([0.0249, 0.0174, 0.0198, 0.0380, 0.0802, 0.0331, 0.0162, 0.0226], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0173, 0.0168, 0.0217, 0.0213, 0.0192, 0.0156, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 21:13:04,073 INFO [finetune.py:992] (1/2) Epoch 13, batch 50, loss[loss=0.1765, simple_loss=0.273, pruned_loss=0.04003, over 12263.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2692, pruned_loss=0.045, over 542772.07 frames. ], batch size: 37, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:13:16,862 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.322e+02 2.937e+02 3.620e+02 4.068e+02 8.964e+02, threshold=7.240e+02, percent-clipped=1.0 2023-05-16 21:13:18,358 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246877.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:13:32,063 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4517, 3.5413, 3.1834, 3.1982, 2.8582, 2.7470, 3.6072, 2.3623], device='cuda:1'), covar=tensor([0.0440, 0.0163, 0.0221, 0.0213, 0.0463, 0.0377, 0.0169, 0.0593], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0151, 0.0153, 0.0180, 0.0192, 0.0189, 0.0163, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:13:35,411 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246900.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 21:13:41,127 INFO [finetune.py:992] (1/2) Epoch 13, batch 100, loss[loss=0.1481, simple_loss=0.2439, pruned_loss=0.02617, over 12140.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2653, pruned_loss=0.04291, over 947993.58 frames. ], batch size: 34, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:13:53,427 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246925.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:14:00,032 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246934.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:14:17,271 INFO [finetune.py:992] (1/2) Epoch 13, batch 150, loss[loss=0.1554, simple_loss=0.247, pruned_loss=0.03187, over 12084.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2628, pruned_loss=0.04224, over 1259999.42 frames. ], batch size: 32, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:14:29,053 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 2.751e+02 3.175e+02 3.588e+02 8.004e+02, threshold=6.350e+02, percent-clipped=1.0 2023-05-16 21:14:43,259 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246995.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:14:52,563 INFO [finetune.py:992] (1/2) Epoch 13, batch 200, loss[loss=0.1449, simple_loss=0.2291, pruned_loss=0.03034, over 12182.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2618, pruned_loss=0.04143, over 1509807.08 frames. ], batch size: 29, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:15:11,152 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247033.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:15:14,979 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6565, 3.3260, 5.0661, 2.6216, 2.7683, 3.6109, 3.0523, 3.6731], device='cuda:1'), covar=tensor([0.0482, 0.1243, 0.0425, 0.1365, 0.2310, 0.1938, 0.1601, 0.1489], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0229, 0.0237, 0.0180, 0.0232, 0.0282, 0.0217, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:15:29,838 INFO [finetune.py:992] (1/2) Epoch 13, batch 250, loss[loss=0.2021, simple_loss=0.292, pruned_loss=0.05615, over 12041.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2613, pruned_loss=0.04147, over 1703283.22 frames. ], batch size: 42, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:15:40,571 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3307, 4.6789, 2.9010, 2.7528, 3.8836, 2.5499, 3.8931, 3.0507], device='cuda:1'), covar=tensor([0.0848, 0.0578, 0.1270, 0.1537, 0.0336, 0.1570, 0.0569, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0249, 0.0175, 0.0198, 0.0139, 0.0181, 0.0193, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:15:41,769 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.777e+02 3.206e+02 3.750e+02 2.533e+03, threshold=6.411e+02, percent-clipped=1.0 2023-05-16 21:16:05,085 INFO [finetune.py:992] (1/2) Epoch 13, batch 300, loss[loss=0.1569, simple_loss=0.2399, pruned_loss=0.03691, over 12242.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2613, pruned_loss=0.04141, over 1847891.65 frames. ], batch size: 32, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:16:15,994 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2956, 6.1475, 5.7167, 5.7577, 6.2060, 5.5169, 5.7248, 5.7877], device='cuda:1'), covar=tensor([0.1258, 0.0847, 0.1044, 0.1699, 0.0820, 0.2142, 0.1579, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0472, 0.0381, 0.0429, 0.0448, 0.0419, 0.0374, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:16:40,547 INFO [finetune.py:992] (1/2) Epoch 13, batch 350, loss[loss=0.1723, simple_loss=0.2578, pruned_loss=0.04342, over 12180.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2608, pruned_loss=0.04107, over 1971461.22 frames. ], batch size: 35, lr: 3.74e-03, grad_scale: 8.0 2023-05-16 21:16:52,659 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.818e+02 3.215e+02 3.898e+02 9.522e+02, threshold=6.431e+02, percent-clipped=2.0 2023-05-16 21:16:56,281 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6965, 3.3183, 5.1441, 2.6080, 2.7897, 3.7550, 3.1876, 3.8135], device='cuda:1'), covar=tensor([0.0437, 0.1207, 0.0356, 0.1285, 0.2095, 0.1679, 0.1442, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0233, 0.0242, 0.0182, 0.0235, 0.0287, 0.0221, 0.0259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:17:05,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-16 21:17:17,900 INFO [finetune.py:992] (1/2) Epoch 13, batch 400, loss[loss=0.1496, simple_loss=0.2305, pruned_loss=0.03429, over 12005.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2595, pruned_loss=0.04066, over 2069713.03 frames. ], batch size: 28, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:17:23,187 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-16 21:17:28,620 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0534, 4.9308, 4.8540, 4.8780, 4.5851, 4.9653, 4.9787, 5.2731], device='cuda:1'), covar=tensor([0.0212, 0.0164, 0.0203, 0.0351, 0.0755, 0.0285, 0.0173, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0185, 0.0179, 0.0232, 0.0228, 0.0205, 0.0167, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 21:17:39,350 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3726, 4.7139, 2.8906, 2.5807, 3.9390, 2.6023, 3.9629, 3.0939], device='cuda:1'), covar=tensor([0.0745, 0.0671, 0.1325, 0.1714, 0.0285, 0.1463, 0.0557, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0251, 0.0177, 0.0199, 0.0140, 0.0181, 0.0194, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:17:53,513 INFO [finetune.py:992] (1/2) Epoch 13, batch 450, loss[loss=0.1792, simple_loss=0.2688, pruned_loss=0.0448, over 11629.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2589, pruned_loss=0.04052, over 2139033.15 frames. ], batch size: 48, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:18:05,456 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.692e+02 3.169e+02 3.840e+02 7.314e+02, threshold=6.338e+02, percent-clipped=2.0 2023-05-16 21:18:16,362 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247290.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:18:28,763 INFO [finetune.py:992] (1/2) Epoch 13, batch 500, loss[loss=0.2051, simple_loss=0.2902, pruned_loss=0.06001, over 12120.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2589, pruned_loss=0.04035, over 2197658.99 frames. ], batch size: 39, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:18:32,252 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3719, 4.9111, 5.3083, 4.5980, 5.0296, 4.7425, 5.3676, 5.0280], device='cuda:1'), covar=tensor([0.0287, 0.0452, 0.0367, 0.0314, 0.0368, 0.0388, 0.0232, 0.0328], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0250, 0.0274, 0.0249, 0.0250, 0.0252, 0.0225, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:18:47,176 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=247333.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:19:05,217 INFO [finetune.py:992] (1/2) Epoch 13, batch 550, loss[loss=0.1407, simple_loss=0.2263, pruned_loss=0.02753, over 11732.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2586, pruned_loss=0.04013, over 2245015.22 frames. ], batch size: 26, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:19:16,746 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 2.711e+02 3.240e+02 3.910e+02 6.136e+02, threshold=6.479e+02, percent-clipped=0.0 2023-05-16 21:19:19,828 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247379.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:19:21,118 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247381.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:19:40,653 INFO [finetune.py:992] (1/2) Epoch 13, batch 600, loss[loss=0.1538, simple_loss=0.2359, pruned_loss=0.03584, over 12360.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2585, pruned_loss=0.04022, over 2273416.30 frames. ], batch size: 30, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:19:53,587 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247426.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:19:54,535 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-16 21:20:03,343 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247440.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:20:15,687 INFO [finetune.py:992] (1/2) Epoch 13, batch 650, loss[loss=0.1825, simple_loss=0.2755, pruned_loss=0.04473, over 11766.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2593, pruned_loss=0.04089, over 2284459.10 frames. ], batch size: 44, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:20:21,775 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4307, 2.5621, 3.1704, 4.3355, 2.1641, 4.2508, 4.4569, 4.5153], device='cuda:1'), covar=tensor([0.0144, 0.1314, 0.0549, 0.0178, 0.1545, 0.0320, 0.0164, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0180, 0.0116, 0.0187, 0.0176, 0.0171, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:20:27,740 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 2.699e+02 3.159e+02 3.694e+02 7.659e+02, threshold=6.318e+02, percent-clipped=2.0 2023-05-16 21:20:37,247 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247487.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:20:39,413 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5786, 2.3342, 3.7279, 4.4596, 3.9919, 4.3009, 3.7578, 3.1805], device='cuda:1'), covar=tensor([0.0035, 0.0436, 0.0120, 0.0037, 0.0101, 0.0091, 0.0124, 0.0345], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0123, 0.0102, 0.0076, 0.0101, 0.0114, 0.0095, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:20:47,250 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247500.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:20:52,799 INFO [finetune.py:992] (1/2) Epoch 13, batch 700, loss[loss=0.1579, simple_loss=0.2481, pruned_loss=0.03384, over 12093.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2591, pruned_loss=0.04033, over 2304782.83 frames. ], batch size: 32, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:21:05,275 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247525.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:21:25,703 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247554.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:21:28,447 INFO [finetune.py:992] (1/2) Epoch 13, batch 750, loss[loss=0.1466, simple_loss=0.2354, pruned_loss=0.02888, over 12114.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2581, pruned_loss=0.03989, over 2320007.53 frames. ], batch size: 33, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:21:30,767 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247561.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:21:40,615 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.608e+02 3.056e+02 3.668e+02 6.856e+02, threshold=6.112e+02, percent-clipped=0.0 2023-05-16 21:21:48,667 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247586.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:21:51,550 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=247590.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:21:54,631 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8750, 3.4227, 5.2660, 2.8766, 2.9753, 3.9512, 3.2902, 3.9433], device='cuda:1'), covar=tensor([0.0366, 0.1160, 0.0290, 0.1124, 0.1952, 0.1410, 0.1355, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0237, 0.0247, 0.0185, 0.0239, 0.0293, 0.0224, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:22:04,534 INFO [finetune.py:992] (1/2) Epoch 13, batch 800, loss[loss=0.1298, simple_loss=0.2158, pruned_loss=0.02194, over 12277.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2575, pruned_loss=0.03975, over 2324061.43 frames. ], batch size: 28, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:22:07,515 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1679, 4.6251, 4.1937, 4.9928, 4.6035, 2.4574, 3.8240, 2.9790], device='cuda:1'), covar=tensor([0.1038, 0.0932, 0.1538, 0.0653, 0.1233, 0.2275, 0.1551, 0.3671], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0376, 0.0352, 0.0305, 0.0364, 0.0271, 0.0341, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:22:09,631 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247615.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:22:26,576 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247638.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:22:31,717 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1164, 5.0087, 4.8687, 4.9475, 4.6360, 5.1354, 5.0960, 5.3861], device='cuda:1'), covar=tensor([0.0326, 0.0173, 0.0223, 0.0347, 0.0872, 0.0311, 0.0198, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0193, 0.0186, 0.0242, 0.0236, 0.0212, 0.0174, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-16 21:22:41,667 INFO [finetune.py:992] (1/2) Epoch 13, batch 850, loss[loss=0.1663, simple_loss=0.2492, pruned_loss=0.04166, over 12176.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2571, pruned_loss=0.03933, over 2344015.85 frames. ], batch size: 31, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:22:53,631 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 2.684e+02 3.226e+02 3.703e+02 6.597e+02, threshold=6.452e+02, percent-clipped=2.0 2023-05-16 21:23:16,903 INFO [finetune.py:992] (1/2) Epoch 13, batch 900, loss[loss=0.175, simple_loss=0.279, pruned_loss=0.03555, over 12296.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2569, pruned_loss=0.03926, over 2355785.16 frames. ], batch size: 34, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:23:36,042 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247735.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:23:52,218 INFO [finetune.py:992] (1/2) Epoch 13, batch 950, loss[loss=0.1869, simple_loss=0.2812, pruned_loss=0.04633, over 12300.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2578, pruned_loss=0.0399, over 2349547.85 frames. ], batch size: 34, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:24:05,180 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.716e+02 3.279e+02 3.877e+02 9.094e+02, threshold=6.558e+02, percent-clipped=4.0 2023-05-16 21:24:10,097 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247782.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:24:11,538 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247784.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:24:13,552 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6613, 2.9986, 3.8864, 4.5404, 4.1314, 4.5711, 3.8186, 3.4530], device='cuda:1'), covar=tensor([0.0042, 0.0340, 0.0124, 0.0040, 0.0111, 0.0074, 0.0142, 0.0321], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0123, 0.0101, 0.0077, 0.0102, 0.0114, 0.0095, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:24:25,572 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4615, 5.0753, 5.4005, 4.6560, 4.9554, 4.7843, 5.4122, 5.0719], device='cuda:1'), covar=tensor([0.0331, 0.0376, 0.0360, 0.0297, 0.0440, 0.0433, 0.0313, 0.0306], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0256, 0.0281, 0.0254, 0.0256, 0.0258, 0.0232, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:24:26,411 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8988, 2.9647, 4.6889, 4.9510, 3.1306, 2.7099, 3.1259, 2.2114], device='cuda:1'), covar=tensor([0.1466, 0.3136, 0.0458, 0.0357, 0.1254, 0.2417, 0.2630, 0.4106], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0381, 0.0269, 0.0294, 0.0267, 0.0303, 0.0379, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:24:29,072 INFO [finetune.py:992] (1/2) Epoch 13, batch 1000, loss[loss=0.1615, simple_loss=0.2566, pruned_loss=0.03318, over 12287.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2572, pruned_loss=0.03934, over 2355726.69 frames. ], batch size: 33, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:24:43,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8800, 4.7413, 4.8350, 4.8792, 4.5606, 4.6126, 4.3298, 4.7803], device='cuda:1'), covar=tensor([0.0775, 0.0639, 0.0879, 0.0585, 0.1952, 0.1460, 0.0582, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0695, 0.0609, 0.0623, 0.0832, 0.0736, 0.0547, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:24:55,879 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247845.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:25:00,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-16 21:25:03,479 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247856.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:25:04,876 INFO [finetune.py:992] (1/2) Epoch 13, batch 1050, loss[loss=0.1822, simple_loss=0.2767, pruned_loss=0.04385, over 12376.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2568, pruned_loss=0.03897, over 2364794.33 frames. ], batch size: 38, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:25:16,972 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.693e+02 3.221e+02 3.799e+02 6.659e+02, threshold=6.442e+02, percent-clipped=1.0 2023-05-16 21:25:21,262 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247881.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:25:40,643 INFO [finetune.py:992] (1/2) Epoch 13, batch 1100, loss[loss=0.1508, simple_loss=0.2404, pruned_loss=0.03058, over 12374.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2559, pruned_loss=0.03863, over 2375581.91 frames. ], batch size: 30, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:25:42,205 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247910.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:26:17,640 INFO [finetune.py:992] (1/2) Epoch 13, batch 1150, loss[loss=0.1802, simple_loss=0.2696, pruned_loss=0.04545, over 12113.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2568, pruned_loss=0.03909, over 2370680.83 frames. ], batch size: 38, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:26:19,135 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4960, 5.2800, 5.4090, 5.4492, 5.0527, 5.1379, 4.8387, 5.4221], device='cuda:1'), covar=tensor([0.0744, 0.0635, 0.0808, 0.0590, 0.2068, 0.1326, 0.0594, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0699, 0.0611, 0.0627, 0.0838, 0.0736, 0.0549, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:26:29,598 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.864e+02 2.775e+02 3.254e+02 3.807e+02 5.397e+02, threshold=6.508e+02, percent-clipped=0.0 2023-05-16 21:26:36,380 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247984.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:26:45,081 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247996.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:26:56,669 INFO [finetune.py:992] (1/2) Epoch 13, batch 1200, loss[loss=0.1772, simple_loss=0.2693, pruned_loss=0.04253, over 11363.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.256, pruned_loss=0.03861, over 2381012.11 frames. ], batch size: 55, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:27:16,164 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248035.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:27:22,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2264, 4.5890, 3.9390, 4.7889, 4.5016, 2.4922, 4.0937, 2.9722], device='cuda:1'), covar=tensor([0.0794, 0.0798, 0.1460, 0.0850, 0.1011, 0.1973, 0.1223, 0.3409], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0377, 0.0355, 0.0307, 0.0365, 0.0271, 0.0342, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:27:23,362 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248045.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:27:31,127 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3613, 2.2782, 3.0299, 4.1855, 2.0406, 4.2047, 4.2955, 4.3806], device='cuda:1'), covar=tensor([0.0126, 0.1512, 0.0591, 0.0162, 0.1582, 0.0328, 0.0172, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0201, 0.0180, 0.0116, 0.0187, 0.0176, 0.0172, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:27:31,899 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248057.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:27:32,428 INFO [finetune.py:992] (1/2) Epoch 13, batch 1250, loss[loss=0.1407, simple_loss=0.2256, pruned_loss=0.0279, over 12194.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2562, pruned_loss=0.0387, over 2381431.42 frames. ], batch size: 29, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:27:43,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 21:27:45,372 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.704e+02 3.068e+02 3.784e+02 5.230e+02, threshold=6.136e+02, percent-clipped=0.0 2023-05-16 21:27:50,554 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:27:50,614 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248082.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:27:51,244 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248083.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:28:09,343 INFO [finetune.py:992] (1/2) Epoch 13, batch 1300, loss[loss=0.167, simple_loss=0.2635, pruned_loss=0.03524, over 12181.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2567, pruned_loss=0.03867, over 2379130.50 frames. ], batch size: 35, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:28:25,282 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248130.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:28:32,414 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248140.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:28:34,753 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248143.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:28:43,930 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248156.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:28:45,185 INFO [finetune.py:992] (1/2) Epoch 13, batch 1350, loss[loss=0.1739, simple_loss=0.263, pruned_loss=0.04236, over 12367.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2567, pruned_loss=0.03871, over 2381765.10 frames. ], batch size: 35, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:28:57,392 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.440e+02 2.860e+02 3.425e+02 6.326e+02, threshold=5.721e+02, percent-clipped=1.0 2023-05-16 21:29:01,736 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248181.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:29:09,714 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6584, 2.5390, 3.6958, 4.4722, 4.0040, 4.5326, 3.8669, 3.3905], device='cuda:1'), covar=tensor([0.0033, 0.0398, 0.0130, 0.0047, 0.0111, 0.0075, 0.0137, 0.0310], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0124, 0.0102, 0.0077, 0.0103, 0.0115, 0.0095, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:29:18,489 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248204.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:29:21,286 INFO [finetune.py:992] (1/2) Epoch 13, batch 1400, loss[loss=0.1431, simple_loss=0.2298, pruned_loss=0.02821, over 12191.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2554, pruned_loss=0.03812, over 2383285.27 frames. ], batch size: 31, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:29:22,774 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248210.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:29:29,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 21:29:31,469 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5453, 2.6453, 3.2014, 4.3592, 2.3646, 4.3572, 4.5475, 4.5778], device='cuda:1'), covar=tensor([0.0124, 0.1162, 0.0512, 0.0153, 0.1357, 0.0279, 0.0144, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0200, 0.0180, 0.0116, 0.0186, 0.0175, 0.0172, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:29:36,828 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248229.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:29:58,092 INFO [finetune.py:992] (1/2) Epoch 13, batch 1450, loss[loss=0.1698, simple_loss=0.2634, pruned_loss=0.03804, over 12190.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2564, pruned_loss=0.0384, over 2373544.77 frames. ], batch size: 35, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:29:58,171 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248258.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:30:10,280 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.647e+02 2.996e+02 3.630e+02 5.731e+02, threshold=5.992e+02, percent-clipped=1.0 2023-05-16 21:30:25,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.31 vs. limit=5.0 2023-05-16 21:30:29,177 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9847, 4.8644, 4.7499, 4.7631, 4.4934, 4.9732, 4.9127, 5.1920], device='cuda:1'), covar=tensor([0.0209, 0.0161, 0.0193, 0.0356, 0.0830, 0.0301, 0.0167, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0197, 0.0190, 0.0247, 0.0242, 0.0217, 0.0178, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 21:30:34,009 INFO [finetune.py:992] (1/2) Epoch 13, batch 1500, loss[loss=0.1606, simple_loss=0.2436, pruned_loss=0.03876, over 12347.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2561, pruned_loss=0.03834, over 2379761.71 frames. ], batch size: 31, lr: 3.73e-03, grad_scale: 8.0 2023-05-16 21:30:51,353 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248332.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:30:57,007 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248340.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:31:05,607 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248352.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:31:09,758 INFO [finetune.py:992] (1/2) Epoch 13, batch 1550, loss[loss=0.1887, simple_loss=0.2721, pruned_loss=0.05268, over 7789.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2561, pruned_loss=0.03837, over 2375095.24 frames. ], batch size: 98, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:31:22,536 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 2.825e+02 3.179e+02 3.763e+02 8.904e+02, threshold=6.357e+02, percent-clipped=2.0 2023-05-16 21:31:28,522 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248383.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:31:36,320 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248393.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:31:46,863 INFO [finetune.py:992] (1/2) Epoch 13, batch 1600, loss[loss=0.1527, simple_loss=0.2357, pruned_loss=0.03481, over 11989.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.256, pruned_loss=0.03859, over 2372133.20 frames. ], batch size: 28, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:32:02,384 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6509, 3.1938, 5.0977, 2.4934, 2.9092, 3.7918, 3.1801, 3.9364], device='cuda:1'), covar=tensor([0.0596, 0.1317, 0.0351, 0.1387, 0.1980, 0.1773, 0.1420, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0238, 0.0250, 0.0186, 0.0241, 0.0297, 0.0226, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:32:08,620 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248438.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:32:10,083 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248440.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:32:13,062 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248444.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:32:22,793 INFO [finetune.py:992] (1/2) Epoch 13, batch 1650, loss[loss=0.1582, simple_loss=0.2295, pruned_loss=0.04339, over 12278.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2558, pruned_loss=0.03859, over 2382368.78 frames. ], batch size: 28, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:32:35,088 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.657e+02 3.152e+02 3.883e+02 1.000e+03, threshold=6.303e+02, percent-clipped=1.0 2023-05-16 21:32:44,439 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248488.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:32:58,425 INFO [finetune.py:992] (1/2) Epoch 13, batch 1700, loss[loss=0.1538, simple_loss=0.2323, pruned_loss=0.03766, over 12346.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2552, pruned_loss=0.03875, over 2383465.45 frames. ], batch size: 30, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:33:35,239 INFO [finetune.py:992] (1/2) Epoch 13, batch 1750, loss[loss=0.1785, simple_loss=0.2634, pruned_loss=0.04678, over 12292.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2549, pruned_loss=0.03863, over 2383343.90 frames. ], batch size: 33, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:33:47,398 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.588e+02 3.157e+02 3.850e+02 6.228e+02, threshold=6.313e+02, percent-clipped=0.0 2023-05-16 21:33:54,187 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-05-16 21:34:01,232 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2417, 4.0037, 4.0640, 4.4876, 2.8744, 3.9746, 2.7218, 4.0973], device='cuda:1'), covar=tensor([0.1657, 0.0751, 0.0828, 0.0559, 0.1254, 0.0598, 0.1769, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0262, 0.0291, 0.0355, 0.0239, 0.0241, 0.0259, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:34:10,937 INFO [finetune.py:992] (1/2) Epoch 13, batch 1800, loss[loss=0.1908, simple_loss=0.2738, pruned_loss=0.05392, over 12097.00 frames. ], tot_loss[loss=0.167, simple_loss=0.256, pruned_loss=0.03896, over 2390580.50 frames. ], batch size: 33, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:34:13,885 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9707, 5.7977, 5.4249, 5.3322, 5.9317, 5.0859, 5.3637, 5.3959], device='cuda:1'), covar=tensor([0.1625, 0.1054, 0.1063, 0.2073, 0.0951, 0.2331, 0.1982, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0492, 0.0393, 0.0444, 0.0462, 0.0435, 0.0392, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:34:33,806 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248640.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:34:35,191 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9926, 4.8438, 4.7735, 4.7692, 4.4617, 4.9377, 4.9310, 5.1666], device='cuda:1'), covar=tensor([0.0253, 0.0179, 0.0191, 0.0425, 0.0857, 0.0314, 0.0166, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0197, 0.0191, 0.0248, 0.0243, 0.0220, 0.0179, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 21:34:40,158 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0069, 4.6835, 4.7447, 4.8929, 4.8140, 4.8610, 4.7791, 2.5981], device='cuda:1'), covar=tensor([0.0096, 0.0073, 0.0100, 0.0072, 0.0047, 0.0110, 0.0079, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0079, 0.0082, 0.0073, 0.0060, 0.0092, 0.0082, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:34:42,274 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248652.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:34:46,421 INFO [finetune.py:992] (1/2) Epoch 13, batch 1850, loss[loss=0.2389, simple_loss=0.3149, pruned_loss=0.08147, over 8208.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.257, pruned_loss=0.03913, over 2384719.67 frames. ], batch size: 98, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:34:59,282 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.662e+02 3.210e+02 3.660e+02 5.824e+02, threshold=6.421e+02, percent-clipped=0.0 2023-05-16 21:35:08,717 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:08,733 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248688.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:10,982 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248691.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:17,800 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248700.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:23,539 INFO [finetune.py:992] (1/2) Epoch 13, batch 1900, loss[loss=0.1676, simple_loss=0.2649, pruned_loss=0.03517, over 12309.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2568, pruned_loss=0.03887, over 2384395.30 frames. ], batch size: 34, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:35:28,718 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248715.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:44,901 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248738.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:45,595 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248739.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:54,926 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248752.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:35:58,865 INFO [finetune.py:992] (1/2) Epoch 13, batch 1950, loss[loss=0.1679, simple_loss=0.249, pruned_loss=0.04338, over 12337.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2568, pruned_loss=0.03896, over 2380345.11 frames. ], batch size: 31, lr: 3.72e-03, grad_scale: 8.0 2023-05-16 21:36:11,032 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.728e+02 3.157e+02 3.801e+02 8.869e+02, threshold=6.314e+02, percent-clipped=3.0 2023-05-16 21:36:11,991 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248776.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:36:17,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 21:36:18,859 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248786.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:36:22,482 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9590, 5.9197, 5.6787, 5.2300, 5.1015, 5.8576, 5.4690, 5.2492], device='cuda:1'), covar=tensor([0.0725, 0.0899, 0.0744, 0.1541, 0.0703, 0.0712, 0.1511, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0563, 0.0522, 0.0639, 0.0417, 0.0721, 0.0779, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 21:36:34,927 INFO [finetune.py:992] (1/2) Epoch 13, batch 2000, loss[loss=0.1559, simple_loss=0.2464, pruned_loss=0.03269, over 11621.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2566, pruned_loss=0.03933, over 2383171.60 frames. ], batch size: 48, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:36:44,356 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4483, 2.4358, 3.1530, 4.2948, 2.2142, 4.3375, 4.5239, 4.5063], device='cuda:1'), covar=tensor([0.0149, 0.1275, 0.0474, 0.0168, 0.1408, 0.0235, 0.0137, 0.0113], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0181, 0.0116, 0.0186, 0.0175, 0.0171, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:36:51,584 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7398, 2.7895, 3.3689, 4.5826, 2.7625, 4.5653, 4.7946, 4.7775], device='cuda:1'), covar=tensor([0.0157, 0.1161, 0.0445, 0.0162, 0.1207, 0.0226, 0.0129, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0181, 0.0116, 0.0186, 0.0175, 0.0171, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:37:06,365 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8728, 4.4955, 4.8666, 4.2706, 4.5225, 4.3953, 4.8615, 4.5043], device='cuda:1'), covar=tensor([0.0291, 0.0430, 0.0297, 0.0291, 0.0424, 0.0350, 0.0255, 0.0531], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0265, 0.0290, 0.0262, 0.0263, 0.0264, 0.0239, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:37:11,907 INFO [finetune.py:992] (1/2) Epoch 13, batch 2050, loss[loss=0.1592, simple_loss=0.248, pruned_loss=0.03521, over 12147.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.257, pruned_loss=0.03922, over 2382520.04 frames. ], batch size: 36, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:37:16,356 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5272, 2.4046, 3.2294, 4.4366, 2.4153, 4.4456, 4.5830, 4.5566], device='cuda:1'), covar=tensor([0.0158, 0.1434, 0.0497, 0.0178, 0.1424, 0.0269, 0.0132, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0200, 0.0182, 0.0117, 0.0187, 0.0176, 0.0172, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:37:24,168 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.873e+02 2.825e+02 3.221e+02 3.921e+02 8.647e+02, threshold=6.442e+02, percent-clipped=3.0 2023-05-16 21:37:45,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 21:37:47,436 INFO [finetune.py:992] (1/2) Epoch 13, batch 2100, loss[loss=0.1748, simple_loss=0.2652, pruned_loss=0.04219, over 10724.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2569, pruned_loss=0.03938, over 2375922.35 frames. ], batch size: 68, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:38:23,482 INFO [finetune.py:992] (1/2) Epoch 13, batch 2150, loss[loss=0.2478, simple_loss=0.3249, pruned_loss=0.08536, over 8337.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2573, pruned_loss=0.03969, over 2356751.55 frames. ], batch size: 98, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:38:35,675 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.706e+02 3.066e+02 3.846e+02 6.335e+02, threshold=6.132e+02, percent-clipped=0.0 2023-05-16 21:38:41,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7214, 3.3685, 5.1295, 2.6548, 2.7453, 3.6824, 3.2557, 3.7553], device='cuda:1'), covar=tensor([0.0464, 0.1168, 0.0291, 0.1149, 0.2069, 0.1552, 0.1399, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0238, 0.0250, 0.0185, 0.0240, 0.0296, 0.0226, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:38:45,967 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248988.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:38:57,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.05 vs. limit=5.0 2023-05-16 21:38:58,919 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5320, 5.5031, 5.3138, 4.8628, 4.8423, 5.3876, 5.0365, 4.8567], device='cuda:1'), covar=tensor([0.0801, 0.0914, 0.0718, 0.1525, 0.0855, 0.0760, 0.1523, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0563, 0.0523, 0.0638, 0.0417, 0.0723, 0.0783, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 21:39:00,284 INFO [finetune.py:992] (1/2) Epoch 13, batch 2200, loss[loss=0.1476, simple_loss=0.2393, pruned_loss=0.02793, over 12363.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2569, pruned_loss=0.03953, over 2358081.88 frames. ], batch size: 30, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:39:08,440 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7332, 2.7135, 3.3365, 4.5568, 2.6303, 4.5164, 4.7752, 4.7631], device='cuda:1'), covar=tensor([0.0144, 0.1155, 0.0458, 0.0140, 0.1218, 0.0286, 0.0106, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0202, 0.0183, 0.0118, 0.0188, 0.0178, 0.0173, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:39:11,387 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6442, 2.7825, 4.6074, 4.6010, 2.6982, 2.5934, 2.7566, 2.0481], device='cuda:1'), covar=tensor([0.1687, 0.3103, 0.0428, 0.0491, 0.1511, 0.2454, 0.3060, 0.4368], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0391, 0.0276, 0.0301, 0.0272, 0.0309, 0.0387, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:39:20,422 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249036.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:39:22,675 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249039.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:39:28,318 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249047.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:39:35,911 INFO [finetune.py:992] (1/2) Epoch 13, batch 2250, loss[loss=0.1956, simple_loss=0.2926, pruned_loss=0.04929, over 10266.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.257, pruned_loss=0.03935, over 2362490.52 frames. ], batch size: 68, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:39:45,115 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249071.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:39:47,826 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.541e+02 3.006e+02 3.631e+02 8.139e+02, threshold=6.012e+02, percent-clipped=2.0 2023-05-16 21:39:56,340 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249087.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:40:11,557 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8406, 5.8721, 5.5821, 5.0761, 5.0309, 5.7034, 5.3292, 5.1193], device='cuda:1'), covar=tensor([0.0829, 0.0907, 0.0795, 0.1638, 0.0755, 0.0829, 0.1762, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0560, 0.0522, 0.0636, 0.0417, 0.0721, 0.0783, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 21:40:12,161 INFO [finetune.py:992] (1/2) Epoch 13, batch 2300, loss[loss=0.2003, simple_loss=0.2921, pruned_loss=0.05422, over 12015.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2574, pruned_loss=0.03955, over 2371502.01 frames. ], batch size: 40, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:40:36,999 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3106, 2.9090, 3.8185, 3.2879, 3.6100, 3.3959, 2.7159, 3.6218], device='cuda:1'), covar=tensor([0.0129, 0.0291, 0.0160, 0.0236, 0.0169, 0.0192, 0.0347, 0.0156], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0205, 0.0187, 0.0187, 0.0217, 0.0164, 0.0199, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:40:48,114 INFO [finetune.py:992] (1/2) Epoch 13, batch 2350, loss[loss=0.1642, simple_loss=0.2418, pruned_loss=0.04334, over 12251.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2574, pruned_loss=0.03933, over 2377973.39 frames. ], batch size: 32, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:41:00,165 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.825e+02 2.710e+02 3.210e+02 3.953e+02 6.628e+02, threshold=6.419e+02, percent-clipped=3.0 2023-05-16 21:41:24,143 INFO [finetune.py:992] (1/2) Epoch 13, batch 2400, loss[loss=0.148, simple_loss=0.227, pruned_loss=0.03449, over 12311.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2573, pruned_loss=0.03937, over 2368312.43 frames. ], batch size: 28, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:41:25,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 21:41:29,280 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249215.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:42:00,321 INFO [finetune.py:992] (1/2) Epoch 13, batch 2450, loss[loss=0.1453, simple_loss=0.2397, pruned_loss=0.02542, over 12298.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2566, pruned_loss=0.03887, over 2372819.91 frames. ], batch size: 34, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:42:13,064 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 2.619e+02 3.067e+02 3.670e+02 7.694e+02, threshold=6.133e+02, percent-clipped=1.0 2023-05-16 21:42:14,048 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249276.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:42:14,751 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6559, 5.4158, 5.5172, 5.6043, 5.2230, 5.2330, 5.0542, 5.5640], device='cuda:1'), covar=tensor([0.0538, 0.0572, 0.0651, 0.0533, 0.1744, 0.1213, 0.0488, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0712, 0.0625, 0.0644, 0.0856, 0.0758, 0.0564, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:42:36,734 INFO [finetune.py:992] (1/2) Epoch 13, batch 2500, loss[loss=0.1718, simple_loss=0.2734, pruned_loss=0.03512, over 12153.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2571, pruned_loss=0.03895, over 2377744.63 frames. ], batch size: 34, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:42:43,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 21:42:59,874 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5407, 4.8540, 3.2695, 2.8105, 4.1051, 2.7557, 4.1153, 3.4103], device='cuda:1'), covar=tensor([0.0613, 0.0426, 0.1010, 0.1471, 0.0355, 0.1228, 0.0452, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0257, 0.0177, 0.0202, 0.0143, 0.0182, 0.0198, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:43:04,665 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249347.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:43:12,509 INFO [finetune.py:992] (1/2) Epoch 13, batch 2550, loss[loss=0.191, simple_loss=0.282, pruned_loss=0.05005, over 12049.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2558, pruned_loss=0.0384, over 2384556.68 frames. ], batch size: 37, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:43:21,872 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249371.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:43:24,531 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.612e+02 3.049e+02 3.624e+02 1.436e+03, threshold=6.098e+02, percent-clipped=1.0 2023-05-16 21:43:37,051 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9297, 2.3248, 3.4415, 2.8756, 3.2360, 3.0250, 2.3066, 3.3089], device='cuda:1'), covar=tensor([0.0140, 0.0401, 0.0163, 0.0238, 0.0165, 0.0188, 0.0382, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0204, 0.0188, 0.0186, 0.0216, 0.0164, 0.0198, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:43:39,013 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249395.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:43:46,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-05-16 21:43:49,934 INFO [finetune.py:992] (1/2) Epoch 13, batch 2600, loss[loss=0.1878, simple_loss=0.2635, pruned_loss=0.05603, over 12115.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2557, pruned_loss=0.03852, over 2389729.96 frames. ], batch size: 39, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:43:57,712 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249419.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:44:25,249 INFO [finetune.py:992] (1/2) Epoch 13, batch 2650, loss[loss=0.1716, simple_loss=0.2649, pruned_loss=0.03912, over 12360.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2553, pruned_loss=0.03869, over 2384366.33 frames. ], batch size: 36, lr: 3.72e-03, grad_scale: 16.0 2023-05-16 21:44:37,131 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.791e+02 3.288e+02 3.836e+02 9.634e+02, threshold=6.576e+02, percent-clipped=1.0 2023-05-16 21:44:57,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 21:45:00,946 INFO [finetune.py:992] (1/2) Epoch 13, batch 2700, loss[loss=0.1599, simple_loss=0.2512, pruned_loss=0.03423, over 12253.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2561, pruned_loss=0.0391, over 2377122.50 frames. ], batch size: 32, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:45:37,645 INFO [finetune.py:992] (1/2) Epoch 13, batch 2750, loss[loss=0.1666, simple_loss=0.2588, pruned_loss=0.03718, over 11252.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2565, pruned_loss=0.03917, over 2367204.18 frames. ], batch size: 55, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:45:46,840 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249571.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:45:49,606 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.689e+02 3.252e+02 4.031e+02 1.393e+03, threshold=6.504e+02, percent-clipped=2.0 2023-05-16 21:46:13,561 INFO [finetune.py:992] (1/2) Epoch 13, batch 2800, loss[loss=0.1549, simple_loss=0.243, pruned_loss=0.03341, over 12292.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2556, pruned_loss=0.03856, over 2371283.12 frames. ], batch size: 33, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:46:24,448 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2366, 4.6932, 3.0026, 2.8471, 3.9930, 2.6502, 3.9873, 3.2374], device='cuda:1'), covar=tensor([0.0774, 0.0706, 0.1131, 0.1462, 0.0350, 0.1337, 0.0475, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0256, 0.0177, 0.0201, 0.0142, 0.0181, 0.0197, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:46:25,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3963, 5.0136, 5.4169, 4.7190, 4.9983, 4.9087, 5.4658, 5.1223], device='cuda:1'), covar=tensor([0.0265, 0.0361, 0.0255, 0.0255, 0.0374, 0.0310, 0.0173, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0266, 0.0291, 0.0263, 0.0264, 0.0265, 0.0240, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:46:48,995 INFO [finetune.py:992] (1/2) Epoch 13, batch 2850, loss[loss=0.1702, simple_loss=0.2715, pruned_loss=0.03441, over 12345.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2551, pruned_loss=0.03821, over 2384689.48 frames. ], batch size: 36, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:47:00,883 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 2.639e+02 3.024e+02 3.770e+02 5.817e+02, threshold=6.047e+02, percent-clipped=0.0 2023-05-16 21:47:16,922 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2632, 5.0748, 5.1762, 5.2443, 4.9045, 4.9236, 4.7101, 5.1993], device='cuda:1'), covar=tensor([0.0755, 0.0655, 0.0849, 0.0666, 0.1983, 0.1320, 0.0570, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0712, 0.0625, 0.0642, 0.0856, 0.0756, 0.0561, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:47:26,084 INFO [finetune.py:992] (1/2) Epoch 13, batch 2900, loss[loss=0.1557, simple_loss=0.2496, pruned_loss=0.03089, over 12111.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2547, pruned_loss=0.03799, over 2376404.29 frames. ], batch size: 33, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:48:01,924 INFO [finetune.py:992] (1/2) Epoch 13, batch 2950, loss[loss=0.1576, simple_loss=0.2453, pruned_loss=0.03494, over 12088.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2536, pruned_loss=0.0377, over 2378960.41 frames. ], batch size: 32, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:48:10,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-16 21:48:13,835 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 2.615e+02 3.019e+02 3.612e+02 8.318e+02, threshold=6.038e+02, percent-clipped=3.0 2023-05-16 21:48:37,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5834, 5.3548, 5.4391, 5.5156, 5.1659, 5.1562, 4.9588, 5.4423], device='cuda:1'), covar=tensor([0.0715, 0.0657, 0.0909, 0.0742, 0.2009, 0.1378, 0.0585, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0551, 0.0713, 0.0628, 0.0645, 0.0858, 0.0759, 0.0563, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:48:37,665 INFO [finetune.py:992] (1/2) Epoch 13, batch 3000, loss[loss=0.145, simple_loss=0.2251, pruned_loss=0.03246, over 12181.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.253, pruned_loss=0.03733, over 2381055.04 frames. ], batch size: 29, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:48:37,665 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 21:48:49,424 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1464, 2.1458, 2.9609, 4.0076, 2.2525, 4.1056, 3.9672, 4.1720], device='cuda:1'), covar=tensor([0.0109, 0.1298, 0.0488, 0.0139, 0.1238, 0.0230, 0.0237, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0180, 0.0118, 0.0186, 0.0176, 0.0171, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:48:54,430 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6117, 4.5319, 4.4593, 3.9184, 4.3312, 4.5197, 4.2356, 4.1422], device='cuda:1'), covar=tensor([0.0700, 0.0901, 0.0643, 0.1797, 0.0709, 0.0848, 0.1506, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0557, 0.0516, 0.0630, 0.0414, 0.0716, 0.0774, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 21:48:56,560 INFO [finetune.py:1026] (1/2) Epoch 13, validation: loss=0.3128, simple_loss=0.3917, pruned_loss=0.117, over 1020973.00 frames. 2023-05-16 21:48:56,560 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 21:49:01,033 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249814.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:49:07,290 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9542, 5.9081, 5.6564, 5.2102, 5.0687, 5.8384, 5.4512, 5.2133], device='cuda:1'), covar=tensor([0.0824, 0.0998, 0.0692, 0.1602, 0.0707, 0.0732, 0.1537, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0558, 0.0517, 0.0631, 0.0415, 0.0716, 0.0776, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-16 21:49:15,249 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2449, 2.5897, 3.8403, 3.1336, 3.6280, 3.2601, 2.5158, 3.6657], device='cuda:1'), covar=tensor([0.0145, 0.0331, 0.0110, 0.0208, 0.0135, 0.0168, 0.0343, 0.0130], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0204, 0.0187, 0.0187, 0.0217, 0.0164, 0.0197, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:49:17,535 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-05-16 21:49:31,973 INFO [finetune.py:992] (1/2) Epoch 13, batch 3050, loss[loss=0.1989, simple_loss=0.2786, pruned_loss=0.05959, over 8481.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2537, pruned_loss=0.03794, over 2374476.92 frames. ], batch size: 101, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:49:34,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2547, 6.0879, 5.6448, 5.6244, 6.1805, 5.4609, 5.6077, 5.6247], device='cuda:1'), covar=tensor([0.1515, 0.0871, 0.1272, 0.2124, 0.0883, 0.2361, 0.1791, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0500, 0.0399, 0.0449, 0.0468, 0.0437, 0.0397, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:49:41,332 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249871.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:49:44,043 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.733e+02 3.086e+02 3.787e+02 9.762e+02, threshold=6.172e+02, percent-clipped=2.0 2023-05-16 21:49:44,276 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249875.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:49:45,598 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0954, 4.7489, 5.1123, 4.4658, 4.7396, 4.6283, 5.1259, 4.7957], device='cuda:1'), covar=tensor([0.0280, 0.0383, 0.0284, 0.0276, 0.0415, 0.0301, 0.0201, 0.0374], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0266, 0.0291, 0.0263, 0.0263, 0.0264, 0.0238, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:50:07,197 INFO [finetune.py:992] (1/2) Epoch 13, batch 3100, loss[loss=0.1515, simple_loss=0.2398, pruned_loss=0.03162, over 12272.00 frames. ], tot_loss[loss=0.166, simple_loss=0.255, pruned_loss=0.03849, over 2370500.61 frames. ], batch size: 32, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:50:15,290 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249919.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:50:23,780 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9816, 3.8129, 3.9556, 3.5872, 3.8195, 3.6709, 3.9700, 3.5502], device='cuda:1'), covar=tensor([0.0410, 0.0381, 0.0372, 0.0281, 0.0406, 0.0341, 0.0276, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0267, 0.0292, 0.0264, 0.0264, 0.0265, 0.0239, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:50:41,573 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9929, 4.7110, 4.8807, 4.9725, 4.8966, 5.0414, 4.7873, 2.5326], device='cuda:1'), covar=tensor([0.0145, 0.0099, 0.0109, 0.0079, 0.0047, 0.0112, 0.0149, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0079, 0.0082, 0.0074, 0.0060, 0.0093, 0.0083, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:50:41,596 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249955.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 21:50:44,104 INFO [finetune.py:992] (1/2) Epoch 13, batch 3150, loss[loss=0.168, simple_loss=0.257, pruned_loss=0.03953, over 12192.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2542, pruned_loss=0.03815, over 2376615.89 frames. ], batch size: 35, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:50:56,310 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.680e+02 3.172e+02 3.678e+02 6.003e+02, threshold=6.344e+02, percent-clipped=0.0 2023-05-16 21:50:57,277 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249976.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:51:04,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-16 21:51:23,491 INFO [finetune.py:992] (1/2) Epoch 13, batch 3200, loss[loss=0.154, simple_loss=0.2368, pruned_loss=0.0356, over 11979.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2542, pruned_loss=0.03832, over 2372663.00 frames. ], batch size: 28, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:51:27,869 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6146, 4.4598, 4.5792, 4.6264, 4.3125, 4.3327, 4.1648, 4.5005], device='cuda:1'), covar=tensor([0.0808, 0.0715, 0.0890, 0.0702, 0.1941, 0.1486, 0.0572, 0.1215], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0717, 0.0629, 0.0646, 0.0863, 0.0765, 0.0567, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 21:51:29,403 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250016.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 21:51:32,263 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2820, 6.1558, 5.7206, 5.7440, 6.2663, 5.4570, 5.7101, 5.6472], device='cuda:1'), covar=tensor([0.1471, 0.0936, 0.0977, 0.1790, 0.0824, 0.1940, 0.1722, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0499, 0.0398, 0.0449, 0.0469, 0.0437, 0.0395, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:51:44,572 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250037.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:51:59,374 INFO [finetune.py:992] (1/2) Epoch 13, batch 3250, loss[loss=0.1574, simple_loss=0.2409, pruned_loss=0.03691, over 12151.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2529, pruned_loss=0.03822, over 2367721.86 frames. ], batch size: 29, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:52:09,507 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1662, 2.2374, 2.7593, 3.1627, 2.2501, 3.2740, 3.2281, 3.3329], device='cuda:1'), covar=tensor([0.0171, 0.1154, 0.0500, 0.0213, 0.1168, 0.0337, 0.0304, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0196, 0.0176, 0.0115, 0.0184, 0.0173, 0.0169, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:52:11,469 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 2.740e+02 3.037e+02 3.497e+02 7.472e+02, threshold=6.074e+02, percent-clipped=1.0 2023-05-16 21:52:35,808 INFO [finetune.py:992] (1/2) Epoch 13, batch 3300, loss[loss=0.1791, simple_loss=0.2688, pruned_loss=0.04475, over 12132.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2531, pruned_loss=0.03789, over 2376381.53 frames. ], batch size: 39, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:52:37,488 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 21:52:51,484 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4502, 4.8318, 4.1666, 4.9959, 4.6140, 3.1004, 4.3395, 3.2472], device='cuda:1'), covar=tensor([0.0711, 0.0748, 0.1374, 0.0593, 0.1006, 0.1558, 0.1024, 0.3158], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0381, 0.0360, 0.0314, 0.0368, 0.0274, 0.0346, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 21:53:11,478 INFO [finetune.py:992] (1/2) Epoch 13, batch 3350, loss[loss=0.191, simple_loss=0.2766, pruned_loss=0.05277, over 11213.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2526, pruned_loss=0.03783, over 2379903.76 frames. ], batch size: 55, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:53:20,021 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250170.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:53:23,280 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.069e+02 2.870e+02 3.367e+02 3.931e+02 8.641e+02, threshold=6.735e+02, percent-clipped=4.0 2023-05-16 21:53:47,315 INFO [finetune.py:992] (1/2) Epoch 13, batch 3400, loss[loss=0.1749, simple_loss=0.254, pruned_loss=0.04788, over 8109.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2531, pruned_loss=0.03806, over 2377450.24 frames. ], batch size: 98, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:54:24,370 INFO [finetune.py:992] (1/2) Epoch 13, batch 3450, loss[loss=0.1818, simple_loss=0.273, pruned_loss=0.04527, over 11839.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2527, pruned_loss=0.03768, over 2382726.34 frames. ], batch size: 44, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:54:36,531 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.009e+02 2.724e+02 3.154e+02 3.692e+02 7.544e+02, threshold=6.308e+02, percent-clipped=1.0 2023-05-16 21:54:59,965 INFO [finetune.py:992] (1/2) Epoch 13, batch 3500, loss[loss=0.148, simple_loss=0.2368, pruned_loss=0.02957, over 12350.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.253, pruned_loss=0.03767, over 2388658.69 frames. ], batch size: 31, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:55:02,175 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250311.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 21:55:17,432 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250332.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:55:35,833 INFO [finetune.py:992] (1/2) Epoch 13, batch 3550, loss[loss=0.1703, simple_loss=0.2664, pruned_loss=0.03707, over 12157.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2538, pruned_loss=0.03783, over 2394977.92 frames. ], batch size: 36, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:55:47,978 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.648e+02 3.157e+02 3.792e+02 7.894e+02, threshold=6.315e+02, percent-clipped=2.0 2023-05-16 21:56:13,476 INFO [finetune.py:992] (1/2) Epoch 13, batch 3600, loss[loss=0.1448, simple_loss=0.2317, pruned_loss=0.02894, over 12354.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2535, pruned_loss=0.03757, over 2393829.13 frames. ], batch size: 30, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:56:49,023 INFO [finetune.py:992] (1/2) Epoch 13, batch 3650, loss[loss=0.1655, simple_loss=0.2526, pruned_loss=0.0392, over 12245.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2534, pruned_loss=0.03733, over 2392652.72 frames. ], batch size: 32, lr: 3.71e-03, grad_scale: 16.0 2023-05-16 21:56:57,794 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250470.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:57:01,129 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.465e+02 2.932e+02 3.493e+02 5.778e+02, threshold=5.863e+02, percent-clipped=0.0 2023-05-16 21:57:24,636 INFO [finetune.py:992] (1/2) Epoch 13, batch 3700, loss[loss=0.1685, simple_loss=0.2623, pruned_loss=0.03739, over 12356.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2535, pruned_loss=0.03765, over 2387465.30 frames. ], batch size: 38, lr: 3.71e-03, grad_scale: 8.0 2023-05-16 21:57:31,731 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250518.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:57:58,376 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4293, 2.3732, 3.6509, 4.3341, 3.7000, 4.2661, 3.6561, 3.0832], device='cuda:1'), covar=tensor([0.0052, 0.0417, 0.0133, 0.0048, 0.0134, 0.0085, 0.0139, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0121, 0.0102, 0.0077, 0.0102, 0.0114, 0.0095, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 21:58:01,073 INFO [finetune.py:992] (1/2) Epoch 13, batch 3750, loss[loss=0.1609, simple_loss=0.2568, pruned_loss=0.03247, over 12367.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.253, pruned_loss=0.03776, over 2382227.56 frames. ], batch size: 35, lr: 3.71e-03, grad_scale: 8.0 2023-05-16 21:58:13,887 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.851e+02 2.934e+02 3.432e+02 4.063e+02 1.664e+03, threshold=6.864e+02, percent-clipped=5.0 2023-05-16 21:58:37,138 INFO [finetune.py:992] (1/2) Epoch 13, batch 3800, loss[loss=0.1715, simple_loss=0.2644, pruned_loss=0.03925, over 12191.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2536, pruned_loss=0.03792, over 2380502.02 frames. ], batch size: 35, lr: 3.71e-03, grad_scale: 8.0 2023-05-16 21:58:39,335 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250611.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 21:58:54,323 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250632.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:59:12,574 INFO [finetune.py:992] (1/2) Epoch 13, batch 3850, loss[loss=0.1739, simple_loss=0.2647, pruned_loss=0.04162, over 10557.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2538, pruned_loss=0.03826, over 2377146.48 frames. ], batch size: 68, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 21:59:13,335 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250659.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 21:59:15,515 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250662.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:59:25,642 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.662e+02 3.077e+02 3.752e+02 6.691e+02, threshold=6.154e+02, percent-clipped=0.0 2023-05-16 21:59:29,040 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250680.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 21:59:30,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 21:59:30,999 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 21:59:49,228 INFO [finetune.py:992] (1/2) Epoch 13, batch 3900, loss[loss=0.201, simple_loss=0.279, pruned_loss=0.06149, over 7819.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2546, pruned_loss=0.03858, over 2363075.13 frames. ], batch size: 97, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 21:59:50,110 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2137, 4.9226, 5.2102, 4.5522, 4.9592, 4.6039, 5.1802, 4.9339], device='cuda:1'), covar=tensor([0.0355, 0.0430, 0.0388, 0.0320, 0.0409, 0.0380, 0.0340, 0.0325], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0269, 0.0292, 0.0266, 0.0266, 0.0266, 0.0240, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 21:59:59,970 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250723.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:00:24,739 INFO [finetune.py:992] (1/2) Epoch 13, batch 3950, loss[loss=0.1762, simple_loss=0.2528, pruned_loss=0.04981, over 12305.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2541, pruned_loss=0.03828, over 2367658.10 frames. ], batch size: 28, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:00:31,568 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8326, 2.8499, 4.7526, 4.9132, 3.0597, 2.5827, 3.0009, 2.1979], device='cuda:1'), covar=tensor([0.1616, 0.3294, 0.0465, 0.0407, 0.1250, 0.2513, 0.2829, 0.4242], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0389, 0.0274, 0.0303, 0.0273, 0.0309, 0.0386, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:00:37,726 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.733e+02 3.127e+02 3.686e+02 8.040e+02, threshold=6.254e+02, percent-clipped=1.0 2023-05-16 22:00:43,590 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4114, 5.1822, 5.3145, 5.3913, 5.0145, 5.0789, 4.7857, 5.3244], device='cuda:1'), covar=tensor([0.0663, 0.0644, 0.0931, 0.0671, 0.1896, 0.1246, 0.0641, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0711, 0.0631, 0.0643, 0.0862, 0.0756, 0.0566, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 22:01:01,041 INFO [finetune.py:992] (1/2) Epoch 13, batch 4000, loss[loss=0.1833, simple_loss=0.2669, pruned_loss=0.04981, over 12370.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2551, pruned_loss=0.03867, over 2369253.45 frames. ], batch size: 38, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:01:01,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2024, 5.1217, 5.0032, 5.0720, 4.7901, 5.1827, 5.1923, 5.3882], device='cuda:1'), covar=tensor([0.0179, 0.0138, 0.0165, 0.0287, 0.0623, 0.0284, 0.0158, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0197, 0.0192, 0.0248, 0.0243, 0.0221, 0.0180, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 22:01:19,671 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250833.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:01:28,846 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250845.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:01:37,947 INFO [finetune.py:992] (1/2) Epoch 13, batch 4050, loss[loss=0.1585, simple_loss=0.2488, pruned_loss=0.0341, over 12368.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2553, pruned_loss=0.03852, over 2377677.64 frames. ], batch size: 36, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:01:50,782 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 2.811e+02 3.319e+02 3.784e+02 9.660e+02, threshold=6.638e+02, percent-clipped=5.0 2023-05-16 22:02:03,594 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250894.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:02:12,256 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250906.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 22:02:13,348 INFO [finetune.py:992] (1/2) Epoch 13, batch 4100, loss[loss=0.1863, simple_loss=0.2707, pruned_loss=0.05092, over 11605.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2553, pruned_loss=0.03896, over 2377467.81 frames. ], batch size: 48, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:02:48,839 INFO [finetune.py:992] (1/2) Epoch 13, batch 4150, loss[loss=0.2251, simple_loss=0.3018, pruned_loss=0.07421, over 7702.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2545, pruned_loss=0.0386, over 2377286.02 frames. ], batch size: 97, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:03:02,404 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 2.619e+02 3.090e+02 3.951e+02 3.498e+03, threshold=6.180e+02, percent-clipped=5.0 2023-05-16 22:03:19,369 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2961, 4.5079, 2.8328, 2.5834, 3.8642, 2.5164, 3.7988, 3.0155], device='cuda:1'), covar=tensor([0.0708, 0.0517, 0.1042, 0.1519, 0.0299, 0.1307, 0.0548, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0257, 0.0178, 0.0201, 0.0142, 0.0181, 0.0198, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:03:26,019 INFO [finetune.py:992] (1/2) Epoch 13, batch 4200, loss[loss=0.1568, simple_loss=0.2428, pruned_loss=0.03534, over 12279.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2559, pruned_loss=0.03892, over 2376955.67 frames. ], batch size: 28, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:03:33,345 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251018.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:04:01,505 INFO [finetune.py:992] (1/2) Epoch 13, batch 4250, loss[loss=0.1913, simple_loss=0.2784, pruned_loss=0.05205, over 12038.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2557, pruned_loss=0.03874, over 2373940.88 frames. ], batch size: 40, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:04:14,453 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.557e+02 2.991e+02 3.941e+02 6.668e+02, threshold=5.982e+02, percent-clipped=2.0 2023-05-16 22:04:27,592 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1452, 2.3897, 3.6724, 3.0712, 3.5153, 3.1844, 2.5625, 3.5343], device='cuda:1'), covar=tensor([0.0157, 0.0419, 0.0138, 0.0264, 0.0155, 0.0207, 0.0381, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0208, 0.0192, 0.0190, 0.0221, 0.0167, 0.0201, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:04:37,354 INFO [finetune.py:992] (1/2) Epoch 13, batch 4300, loss[loss=0.1679, simple_loss=0.2602, pruned_loss=0.03783, over 12361.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2552, pruned_loss=0.0388, over 2370782.77 frames. ], batch size: 36, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:04:56,067 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6504, 4.3226, 4.6068, 4.0869, 4.3715, 4.1698, 4.6324, 4.3686], device='cuda:1'), covar=tensor([0.0289, 0.0417, 0.0333, 0.0308, 0.0393, 0.0338, 0.0242, 0.0471], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0270, 0.0292, 0.0267, 0.0267, 0.0267, 0.0240, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:05:14,502 INFO [finetune.py:992] (1/2) Epoch 13, batch 4350, loss[loss=0.1884, simple_loss=0.2693, pruned_loss=0.0538, over 8182.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2557, pruned_loss=0.03885, over 2368987.43 frames. ], batch size: 99, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:05:27,215 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.611e+02 3.160e+02 3.927e+02 7.515e+02, threshold=6.319e+02, percent-clipped=3.0 2023-05-16 22:05:27,492 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2428, 4.6414, 4.2519, 4.9005, 4.4497, 2.5020, 4.1320, 3.0212], device='cuda:1'), covar=tensor([0.0772, 0.0816, 0.1191, 0.0557, 0.1069, 0.1981, 0.1133, 0.3324], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0374, 0.0353, 0.0308, 0.0363, 0.0271, 0.0340, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:05:36,414 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251189.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:05:42,941 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251198.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:05:45,310 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251201.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 22:05:50,025 INFO [finetune.py:992] (1/2) Epoch 13, batch 4400, loss[loss=0.2152, simple_loss=0.315, pruned_loss=0.05769, over 12130.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2567, pruned_loss=0.03923, over 2366199.62 frames. ], batch size: 39, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:06:25,234 INFO [finetune.py:992] (1/2) Epoch 13, batch 4450, loss[loss=0.2048, simple_loss=0.2815, pruned_loss=0.06408, over 7964.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.257, pruned_loss=0.03918, over 2372736.61 frames. ], batch size: 98, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:06:26,934 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251259.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:06:39,511 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.058e+02 2.761e+02 3.162e+02 4.023e+02 1.114e+03, threshold=6.324e+02, percent-clipped=6.0 2023-05-16 22:06:56,796 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6469, 3.6089, 3.3161, 3.1729, 2.9583, 2.7751, 3.7107, 2.2608], device='cuda:1'), covar=tensor([0.0360, 0.0169, 0.0210, 0.0225, 0.0404, 0.0378, 0.0150, 0.0477], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0162, 0.0166, 0.0191, 0.0206, 0.0201, 0.0173, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:06:57,457 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251301.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:06:58,962 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5430, 5.1109, 5.5056, 4.7945, 5.1794, 4.8608, 5.5562, 5.1951], device='cuda:1'), covar=tensor([0.0267, 0.0378, 0.0234, 0.0260, 0.0324, 0.0350, 0.0166, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0269, 0.0292, 0.0267, 0.0266, 0.0266, 0.0239, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:07:02,421 INFO [finetune.py:992] (1/2) Epoch 13, batch 4500, loss[loss=0.1523, simple_loss=0.2429, pruned_loss=0.03089, over 12335.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2559, pruned_loss=0.03874, over 2374310.70 frames. ], batch size: 31, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:07:09,841 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251318.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:07:37,819 INFO [finetune.py:992] (1/2) Epoch 13, batch 4550, loss[loss=0.1751, simple_loss=0.2637, pruned_loss=0.04325, over 12039.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2553, pruned_loss=0.03879, over 2366999.02 frames. ], batch size: 37, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:07:40,762 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251362.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:07:43,381 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251366.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:07:50,412 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.632e+02 3.087e+02 3.733e+02 6.616e+02, threshold=6.173e+02, percent-clipped=1.0 2023-05-16 22:07:57,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-16 22:08:10,832 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251404.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:08:14,072 INFO [finetune.py:992] (1/2) Epoch 13, batch 4600, loss[loss=0.1797, simple_loss=0.2778, pruned_loss=0.04083, over 12354.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2557, pruned_loss=0.03862, over 2373621.91 frames. ], batch size: 36, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:08:35,883 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-05-16 22:08:40,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 22:08:50,342 INFO [finetune.py:992] (1/2) Epoch 13, batch 4650, loss[loss=0.1855, simple_loss=0.2709, pruned_loss=0.05001, over 10674.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2558, pruned_loss=0.03865, over 2376638.22 frames. ], batch size: 69, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:08:52,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 22:08:55,495 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251465.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:08:58,274 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6832, 5.4813, 5.5583, 5.6650, 5.3425, 5.3189, 5.1737, 5.5022], device='cuda:1'), covar=tensor([0.0606, 0.0528, 0.0769, 0.0531, 0.1564, 0.1325, 0.0497, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0708, 0.0631, 0.0643, 0.0861, 0.0755, 0.0566, 0.0489], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:08:59,166 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3661, 4.7344, 4.2724, 5.0315, 4.6519, 3.0079, 4.3529, 3.1391], device='cuda:1'), covar=tensor([0.0788, 0.0734, 0.1360, 0.0438, 0.0920, 0.1606, 0.0911, 0.3179], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0381, 0.0360, 0.0315, 0.0369, 0.0276, 0.0346, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:09:03,007 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.631e+02 2.993e+02 3.545e+02 6.646e+02, threshold=5.986e+02, percent-clipped=1.0 2023-05-16 22:09:12,343 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251489.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:09:20,916 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251501.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 22:09:25,573 INFO [finetune.py:992] (1/2) Epoch 13, batch 4700, loss[loss=0.1949, simple_loss=0.2793, pruned_loss=0.05519, over 7792.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2559, pruned_loss=0.03903, over 2375514.65 frames. ], batch size: 98, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:09:46,481 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251537.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:09:55,182 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251549.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:09:58,325 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6577, 3.3177, 5.0598, 2.6560, 2.6225, 3.6897, 3.1262, 3.6244], device='cuda:1'), covar=tensor([0.0423, 0.1079, 0.0294, 0.1152, 0.2062, 0.1528, 0.1422, 0.1276], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0238, 0.0254, 0.0185, 0.0240, 0.0298, 0.0227, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:09:58,902 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251554.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:10:02,301 INFO [finetune.py:992] (1/2) Epoch 13, batch 4750, loss[loss=0.1809, simple_loss=0.2746, pruned_loss=0.04357, over 12131.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2561, pruned_loss=0.03891, over 2378403.97 frames. ], batch size: 39, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:10:15,785 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 2.680e+02 3.120e+02 3.758e+02 6.071e+02, threshold=6.239e+02, percent-clipped=1.0 2023-05-16 22:10:38,635 INFO [finetune.py:992] (1/2) Epoch 13, batch 4800, loss[loss=0.174, simple_loss=0.2652, pruned_loss=0.04143, over 11647.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2558, pruned_loss=0.03877, over 2376244.50 frames. ], batch size: 48, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:10:48,253 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7853, 2.8574, 4.4912, 4.6651, 3.0389, 2.6219, 2.9577, 2.1615], device='cuda:1'), covar=tensor([0.1538, 0.3003, 0.0494, 0.0399, 0.1251, 0.2385, 0.2796, 0.4168], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0391, 0.0275, 0.0304, 0.0274, 0.0309, 0.0386, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:11:01,980 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251641.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:11:13,332 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251657.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:11:13,949 INFO [finetune.py:992] (1/2) Epoch 13, batch 4850, loss[loss=0.1616, simple_loss=0.2627, pruned_loss=0.03025, over 12157.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2558, pruned_loss=0.03885, over 2371693.63 frames. ], batch size: 39, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:11:26,848 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.707e+02 3.022e+02 3.497e+02 7.230e+02, threshold=6.045e+02, percent-clipped=1.0 2023-05-16 22:11:46,146 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251702.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:11:50,249 INFO [finetune.py:992] (1/2) Epoch 13, batch 4900, loss[loss=0.1867, simple_loss=0.28, pruned_loss=0.04674, over 12274.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2556, pruned_loss=0.03883, over 2376666.01 frames. ], batch size: 37, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:11:51,229 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251709.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:11:59,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 22:12:26,819 INFO [finetune.py:992] (1/2) Epoch 13, batch 4950, loss[loss=0.1659, simple_loss=0.2569, pruned_loss=0.03749, over 12191.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2554, pruned_loss=0.03855, over 2372589.15 frames. ], batch size: 35, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:12:28,371 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251760.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:12:35,694 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251770.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:12:39,747 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.575e+02 3.026e+02 3.736e+02 6.475e+02, threshold=6.052e+02, percent-clipped=1.0 2023-05-16 22:12:53,645 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5571, 4.8898, 3.3230, 2.7823, 4.1481, 2.6309, 4.0180, 3.5189], device='cuda:1'), covar=tensor([0.0689, 0.0546, 0.0949, 0.1523, 0.0312, 0.1382, 0.0510, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0260, 0.0179, 0.0203, 0.0143, 0.0183, 0.0200, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:13:02,872 INFO [finetune.py:992] (1/2) Epoch 13, batch 5000, loss[loss=0.1778, simple_loss=0.2654, pruned_loss=0.04506, over 12272.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2555, pruned_loss=0.03885, over 2376414.93 frames. ], batch size: 37, lr: 3.70e-03, grad_scale: 8.0 2023-05-16 22:13:35,873 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251854.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:13:38,552 INFO [finetune.py:992] (1/2) Epoch 13, batch 5050, loss[loss=0.1449, simple_loss=0.2339, pruned_loss=0.02801, over 12183.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2557, pruned_loss=0.039, over 2372355.96 frames. ], batch size: 31, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:13:51,911 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 2.734e+02 3.058e+02 3.779e+02 7.472e+02, threshold=6.116e+02, percent-clipped=2.0 2023-05-16 22:14:03,362 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5325, 5.3601, 5.4450, 5.5370, 5.1756, 5.1678, 4.9565, 5.4011], device='cuda:1'), covar=tensor([0.0650, 0.0576, 0.0766, 0.0538, 0.1713, 0.1277, 0.0519, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0701, 0.0623, 0.0636, 0.0852, 0.0745, 0.0559, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:14:10,389 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251902.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:14:14,636 INFO [finetune.py:992] (1/2) Epoch 13, batch 5100, loss[loss=0.1864, simple_loss=0.2691, pruned_loss=0.05186, over 12119.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.256, pruned_loss=0.03905, over 2378662.60 frames. ], batch size: 33, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:14:49,365 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251957.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:14:49,895 INFO [finetune.py:992] (1/2) Epoch 13, batch 5150, loss[loss=0.1936, simple_loss=0.2904, pruned_loss=0.04839, over 12038.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2565, pruned_loss=0.03902, over 2375143.71 frames. ], batch size: 42, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:15:02,694 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.613e+02 3.156e+02 3.619e+02 6.427e+02, threshold=6.311e+02, percent-clipped=2.0 2023-05-16 22:15:18,344 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251997.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:15:27,920 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252005.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:15:30,036 INFO [finetune.py:992] (1/2) Epoch 13, batch 5200, loss[loss=0.1705, simple_loss=0.2627, pruned_loss=0.03912, over 10578.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2562, pruned_loss=0.03843, over 2381091.25 frames. ], batch size: 68, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:16:05,472 INFO [finetune.py:992] (1/2) Epoch 13, batch 5250, loss[loss=0.1771, simple_loss=0.2718, pruned_loss=0.0412, over 12343.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2558, pruned_loss=0.03834, over 2376002.79 frames. ], batch size: 36, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:16:07,107 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252060.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:16:10,383 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252065.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:16:16,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-16 22:16:18,038 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.604e+02 3.008e+02 3.541e+02 6.263e+02, threshold=6.016e+02, percent-clipped=0.0 2023-05-16 22:16:40,727 INFO [finetune.py:992] (1/2) Epoch 13, batch 5300, loss[loss=0.1602, simple_loss=0.2463, pruned_loss=0.03707, over 12414.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.256, pruned_loss=0.03842, over 2375627.78 frames. ], batch size: 32, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:16:40,794 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252108.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:17:09,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-16 22:17:17,356 INFO [finetune.py:992] (1/2) Epoch 13, batch 5350, loss[loss=0.16, simple_loss=0.2559, pruned_loss=0.03204, over 12279.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2563, pruned_loss=0.03858, over 2365164.68 frames. ], batch size: 37, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:17:30,034 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.749e+02 3.226e+02 3.819e+02 7.797e+02, threshold=6.451e+02, percent-clipped=2.0 2023-05-16 22:17:53,052 INFO [finetune.py:992] (1/2) Epoch 13, batch 5400, loss[loss=0.1875, simple_loss=0.2712, pruned_loss=0.05191, over 12133.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2564, pruned_loss=0.03892, over 2355944.95 frames. ], batch size: 39, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:18:00,406 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1234, 2.5038, 3.6342, 3.0497, 3.4590, 3.1425, 2.4514, 3.4658], device='cuda:1'), covar=tensor([0.0137, 0.0371, 0.0134, 0.0267, 0.0153, 0.0189, 0.0399, 0.0143], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0208, 0.0193, 0.0192, 0.0222, 0.0167, 0.0202, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:18:16,994 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8317, 3.3829, 5.2071, 2.7366, 2.9619, 3.9510, 3.3057, 3.9004], device='cuda:1'), covar=tensor([0.0442, 0.1127, 0.0313, 0.1218, 0.1876, 0.1470, 0.1353, 0.1198], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0238, 0.0255, 0.0185, 0.0239, 0.0297, 0.0227, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:18:28,923 INFO [finetune.py:992] (1/2) Epoch 13, batch 5450, loss[loss=0.1703, simple_loss=0.26, pruned_loss=0.04026, over 12187.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2559, pruned_loss=0.03848, over 2360656.80 frames. ], batch size: 35, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:18:42,285 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.623e+02 2.970e+02 3.795e+02 7.378e+02, threshold=5.940e+02, percent-clipped=2.0 2023-05-16 22:18:46,001 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6164, 5.2118, 5.5633, 4.9108, 5.2283, 5.0047, 5.6325, 5.1750], device='cuda:1'), covar=tensor([0.0206, 0.0301, 0.0250, 0.0211, 0.0324, 0.0280, 0.0182, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0271, 0.0294, 0.0268, 0.0267, 0.0267, 0.0239, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:18:49,687 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7362, 3.6252, 3.3504, 3.2727, 3.0115, 2.9169, 3.7800, 2.3737], device='cuda:1'), covar=tensor([0.0356, 0.0135, 0.0195, 0.0192, 0.0347, 0.0314, 0.0113, 0.0474], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0163, 0.0165, 0.0190, 0.0205, 0.0201, 0.0173, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:18:57,165 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252297.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:19:05,256 INFO [finetune.py:992] (1/2) Epoch 13, batch 5500, loss[loss=0.1872, simple_loss=0.2782, pruned_loss=0.04808, over 12079.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2558, pruned_loss=0.03857, over 2368740.64 frames. ], batch size: 42, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:19:31,846 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252345.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:19:41,086 INFO [finetune.py:992] (1/2) Epoch 13, batch 5550, loss[loss=0.1656, simple_loss=0.2411, pruned_loss=0.04509, over 12279.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2555, pruned_loss=0.03858, over 2375669.69 frames. ], batch size: 28, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:19:46,322 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252365.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:19:54,193 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.998e+02 2.832e+02 3.242e+02 3.996e+02 1.149e+03, threshold=6.485e+02, percent-clipped=5.0 2023-05-16 22:19:55,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-05-16 22:20:16,890 INFO [finetune.py:992] (1/2) Epoch 13, batch 5600, loss[loss=0.2471, simple_loss=0.312, pruned_loss=0.09108, over 7943.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2564, pruned_loss=0.03892, over 2367955.36 frames. ], batch size: 98, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:20:21,305 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252413.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:20:24,261 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252417.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:20:36,336 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252434.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:20:53,866 INFO [finetune.py:992] (1/2) Epoch 13, batch 5650, loss[loss=0.1599, simple_loss=0.2545, pruned_loss=0.03262, over 12080.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2571, pruned_loss=0.03921, over 2363040.16 frames. ], batch size: 32, lr: 3.69e-03, grad_scale: 8.0 2023-05-16 22:21:04,636 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0286, 4.6729, 4.9711, 4.3708, 4.6809, 4.4498, 5.0234, 4.6507], device='cuda:1'), covar=tensor([0.0258, 0.0354, 0.0279, 0.0273, 0.0395, 0.0350, 0.0200, 0.0435], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0271, 0.0295, 0.0268, 0.0269, 0.0267, 0.0240, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:21:06,630 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 2.706e+02 3.163e+02 3.828e+02 6.619e+02, threshold=6.327e+02, percent-clipped=2.0 2023-05-16 22:21:08,292 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252478.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:21:20,540 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252495.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:21:29,485 INFO [finetune.py:992] (1/2) Epoch 13, batch 5700, loss[loss=0.1702, simple_loss=0.2613, pruned_loss=0.03954, over 12376.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2568, pruned_loss=0.03917, over 2372597.45 frames. ], batch size: 38, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:21:46,141 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9941, 4.4516, 4.0168, 4.7574, 4.2448, 2.7844, 3.9880, 2.9781], device='cuda:1'), covar=tensor([0.0867, 0.0851, 0.1254, 0.0621, 0.1160, 0.1699, 0.1120, 0.3252], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0378, 0.0357, 0.0313, 0.0368, 0.0271, 0.0344, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:22:05,113 INFO [finetune.py:992] (1/2) Epoch 13, batch 5750, loss[loss=0.1579, simple_loss=0.2515, pruned_loss=0.03214, over 12364.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2561, pruned_loss=0.03871, over 2379050.71 frames. ], batch size: 36, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:22:18,527 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.695e+02 3.008e+02 3.623e+02 6.938e+02, threshold=6.016e+02, percent-clipped=1.0 2023-05-16 22:22:29,906 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252592.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:22:42,019 INFO [finetune.py:992] (1/2) Epoch 13, batch 5800, loss[loss=0.1502, simple_loss=0.2373, pruned_loss=0.03151, over 12092.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2558, pruned_loss=0.03875, over 2377455.92 frames. ], batch size: 32, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:22:45,804 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252613.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:23:09,436 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5800, 2.9854, 3.8721, 2.3827, 2.4917, 3.1072, 2.9608, 3.2745], device='cuda:1'), covar=tensor([0.0613, 0.1099, 0.0508, 0.1203, 0.1913, 0.1314, 0.1342, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0236, 0.0253, 0.0182, 0.0237, 0.0292, 0.0226, 0.0263], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:23:14,366 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252653.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:23:17,676 INFO [finetune.py:992] (1/2) Epoch 13, batch 5850, loss[loss=0.1695, simple_loss=0.2584, pruned_loss=0.04032, over 12191.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2555, pruned_loss=0.03868, over 2375954.32 frames. ], batch size: 31, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:23:25,052 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252668.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:23:29,323 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252674.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:23:30,484 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 2.713e+02 3.174e+02 3.869e+02 5.476e+02, threshold=6.348e+02, percent-clipped=0.0 2023-05-16 22:23:53,395 INFO [finetune.py:992] (1/2) Epoch 13, batch 5900, loss[loss=0.1654, simple_loss=0.2618, pruned_loss=0.03455, over 12256.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2562, pruned_loss=0.0387, over 2378593.09 frames. ], batch size: 32, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:24:09,183 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252729.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:24:30,083 INFO [finetune.py:992] (1/2) Epoch 13, batch 5950, loss[loss=0.1606, simple_loss=0.2509, pruned_loss=0.03512, over 12039.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2557, pruned_loss=0.03865, over 2378435.32 frames. ], batch size: 31, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:24:33,047 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3035, 5.2070, 5.1439, 5.2251, 4.7111, 5.3099, 5.2531, 5.4596], device='cuda:1'), covar=tensor([0.0199, 0.0137, 0.0150, 0.0275, 0.0769, 0.0250, 0.0157, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0201, 0.0195, 0.0251, 0.0248, 0.0222, 0.0183, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1') 2023-05-16 22:24:40,869 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252773.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:24:42,851 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.511e+02 2.995e+02 3.688e+02 9.486e+02, threshold=5.990e+02, percent-clipped=3.0 2023-05-16 22:24:52,799 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252790.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:25:05,592 INFO [finetune.py:992] (1/2) Epoch 13, batch 6000, loss[loss=0.154, simple_loss=0.2361, pruned_loss=0.03592, over 12173.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2554, pruned_loss=0.03871, over 2376011.64 frames. ], batch size: 31, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:25:05,593 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 22:25:14,340 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0524, 4.5840, 4.9382, 4.4159, 4.6369, 4.5632, 4.9456, 4.9928], device='cuda:1'), covar=tensor([0.0226, 0.0327, 0.0225, 0.0262, 0.0342, 0.0286, 0.0234, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0272, 0.0295, 0.0269, 0.0269, 0.0267, 0.0241, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:25:24,020 INFO [finetune.py:1026] (1/2) Epoch 13, validation: loss=0.3121, simple_loss=0.3891, pruned_loss=0.1176, over 1020973.00 frames. 2023-05-16 22:25:24,021 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 22:25:30,441 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9039, 3.8968, 3.8370, 3.9983, 3.7706, 3.7790, 3.6588, 3.8891], device='cuda:1'), covar=tensor([0.1284, 0.0644, 0.1641, 0.0738, 0.1678, 0.1348, 0.0615, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0712, 0.0630, 0.0645, 0.0870, 0.0762, 0.0568, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-16 22:25:55,719 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4790, 5.1135, 5.4469, 4.8020, 5.1107, 4.8722, 5.4619, 5.0857], device='cuda:1'), covar=tensor([0.0230, 0.0275, 0.0227, 0.0240, 0.0337, 0.0276, 0.0184, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0271, 0.0295, 0.0269, 0.0269, 0.0268, 0.0241, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:26:00,010 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252857.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:26:00,532 INFO [finetune.py:992] (1/2) Epoch 13, batch 6050, loss[loss=0.1638, simple_loss=0.2423, pruned_loss=0.04262, over 11771.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2555, pruned_loss=0.0389, over 2374108.42 frames. ], batch size: 26, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:26:03,934 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-16 22:26:06,577 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7782, 2.9589, 4.5296, 4.7016, 3.0482, 2.6949, 3.1355, 2.2251], device='cuda:1'), covar=tensor([0.1587, 0.2956, 0.0509, 0.0442, 0.1253, 0.2272, 0.2525, 0.3949], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0386, 0.0274, 0.0299, 0.0272, 0.0305, 0.0383, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:26:09,783 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 22:26:13,485 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.667e+02 3.146e+02 3.999e+02 6.169e+02, threshold=6.293e+02, percent-clipped=2.0 2023-05-16 22:26:36,471 INFO [finetune.py:992] (1/2) Epoch 13, batch 6100, loss[loss=0.1611, simple_loss=0.2515, pruned_loss=0.03533, over 12295.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2559, pruned_loss=0.0394, over 2366396.69 frames. ], batch size: 34, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:26:43,707 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252918.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:26:56,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.91 vs. limit=5.0 2023-05-16 22:27:04,759 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252948.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:27:11,731 INFO [finetune.py:992] (1/2) Epoch 13, batch 6150, loss[loss=0.1527, simple_loss=0.2507, pruned_loss=0.02737, over 12269.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2568, pruned_loss=0.03948, over 2370714.70 frames. ], batch size: 37, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:27:14,129 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6510, 2.9150, 4.6567, 4.8595, 3.0602, 2.7780, 3.0227, 2.2399], device='cuda:1'), covar=tensor([0.1630, 0.2810, 0.0437, 0.0375, 0.1265, 0.2218, 0.2741, 0.3923], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0388, 0.0275, 0.0300, 0.0273, 0.0307, 0.0385, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:27:20,318 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252969.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:27:25,177 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.798e+02 2.819e+02 3.351e+02 3.995e+02 6.387e+02, threshold=6.701e+02, percent-clipped=1.0 2023-05-16 22:27:38,304 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252993.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:27:49,134 INFO [finetune.py:992] (1/2) Epoch 13, batch 6200, loss[loss=0.1666, simple_loss=0.2631, pruned_loss=0.03508, over 12366.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2566, pruned_loss=0.03942, over 2371400.37 frames. ], batch size: 35, lr: 3.69e-03, grad_scale: 16.0 2023-05-16 22:28:00,660 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253024.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:28:04,468 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4279, 3.5405, 3.2978, 3.1402, 2.8297, 2.7160, 3.6087, 2.2311], device='cuda:1'), covar=tensor([0.0452, 0.0131, 0.0187, 0.0222, 0.0415, 0.0368, 0.0145, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0168, 0.0168, 0.0195, 0.0210, 0.0206, 0.0177, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:28:06,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 22:28:22,180 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253054.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:28:24,734 INFO [finetune.py:992] (1/2) Epoch 13, batch 6250, loss[loss=0.1499, simple_loss=0.2363, pruned_loss=0.03178, over 12345.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.257, pruned_loss=0.03944, over 2363081.62 frames. ], batch size: 31, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:28:35,754 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253073.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:28:37,688 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.515e+02 3.088e+02 3.662e+02 5.589e+02, threshold=6.175e+02, percent-clipped=0.0 2023-05-16 22:28:48,003 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253090.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:28:52,961 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6565, 3.0418, 3.8952, 4.6001, 3.9401, 4.6763, 3.9065, 3.6995], device='cuda:1'), covar=tensor([0.0040, 0.0303, 0.0126, 0.0041, 0.0117, 0.0049, 0.0122, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0122, 0.0103, 0.0078, 0.0103, 0.0115, 0.0095, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:29:01,283 INFO [finetune.py:992] (1/2) Epoch 13, batch 6300, loss[loss=0.1494, simple_loss=0.2392, pruned_loss=0.02978, over 12288.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2557, pruned_loss=0.03897, over 2367800.63 frames. ], batch size: 33, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:29:02,147 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5741, 5.4131, 5.5081, 5.5712, 5.2032, 5.2107, 4.9694, 5.4646], device='cuda:1'), covar=tensor([0.0753, 0.0546, 0.0833, 0.0635, 0.1844, 0.1312, 0.0550, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0702, 0.0623, 0.0636, 0.0857, 0.0750, 0.0559, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:29:10,663 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253121.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:29:23,159 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253138.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:29:27,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-05-16 22:29:36,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-05-16 22:29:37,370 INFO [finetune.py:992] (1/2) Epoch 13, batch 6350, loss[loss=0.1624, simple_loss=0.2644, pruned_loss=0.03017, over 11303.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2563, pruned_loss=0.03897, over 2372318.79 frames. ], batch size: 55, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:29:47,557 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9919, 5.9279, 5.5561, 5.4076, 5.9964, 5.1817, 5.3143, 5.4689], device='cuda:1'), covar=tensor([0.1625, 0.0945, 0.1224, 0.2127, 0.0940, 0.2336, 0.2331, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0507, 0.0405, 0.0457, 0.0474, 0.0446, 0.0402, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:29:50,180 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.572e+02 2.972e+02 3.425e+02 6.380e+02, threshold=5.944e+02, percent-clipped=2.0 2023-05-16 22:30:02,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 22:30:13,166 INFO [finetune.py:992] (1/2) Epoch 13, batch 6400, loss[loss=0.1708, simple_loss=0.2622, pruned_loss=0.0397, over 11840.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2564, pruned_loss=0.03906, over 2379165.20 frames. ], batch size: 44, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:30:16,837 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253213.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:30:16,896 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0364, 4.8314, 4.9869, 5.0149, 4.6792, 4.6724, 4.4066, 4.8918], device='cuda:1'), covar=tensor([0.0674, 0.0604, 0.0873, 0.0599, 0.1709, 0.1365, 0.0580, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0703, 0.0623, 0.0637, 0.0854, 0.0752, 0.0559, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:30:24,087 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5874, 3.2680, 4.9923, 2.6935, 2.5902, 3.7431, 2.9525, 3.8194], device='cuda:1'), covar=tensor([0.0416, 0.1230, 0.0348, 0.1107, 0.2065, 0.1434, 0.1522, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0238, 0.0254, 0.0183, 0.0239, 0.0294, 0.0226, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:30:42,195 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253248.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:30:45,796 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2120, 2.7319, 3.8208, 3.2229, 3.6322, 3.4072, 2.7166, 3.7165], device='cuda:1'), covar=tensor([0.0135, 0.0322, 0.0121, 0.0238, 0.0139, 0.0143, 0.0331, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0207, 0.0193, 0.0191, 0.0223, 0.0167, 0.0201, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:30:49,254 INFO [finetune.py:992] (1/2) Epoch 13, batch 6450, loss[loss=0.1725, simple_loss=0.2667, pruned_loss=0.03919, over 11573.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2575, pruned_loss=0.03963, over 2368562.18 frames. ], batch size: 48, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:30:57,227 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253269.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:31:02,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.693e+02 3.152e+02 3.812e+02 6.020e+02, threshold=6.304e+02, percent-clipped=1.0 2023-05-16 22:31:17,220 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253296.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:31:25,727 INFO [finetune.py:992] (1/2) Epoch 13, batch 6500, loss[loss=0.1916, simple_loss=0.2859, pruned_loss=0.04867, over 11241.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2571, pruned_loss=0.03936, over 2363615.34 frames. ], batch size: 55, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:31:27,539 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 22:31:32,084 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253317.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:31:37,085 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253324.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:31:40,838 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253329.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:31:55,032 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253349.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:31:55,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-16 22:32:01,403 INFO [finetune.py:992] (1/2) Epoch 13, batch 6550, loss[loss=0.2475, simple_loss=0.3231, pruned_loss=0.08597, over 8021.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2581, pruned_loss=0.03962, over 2361800.12 frames. ], batch size: 97, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:32:04,382 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6780, 2.7395, 3.9280, 4.6282, 3.9755, 4.6570, 4.1270, 3.4466], device='cuda:1'), covar=tensor([0.0042, 0.0388, 0.0144, 0.0039, 0.0144, 0.0073, 0.0109, 0.0328], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0122, 0.0104, 0.0078, 0.0103, 0.0114, 0.0095, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:32:04,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 22:32:11,548 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253372.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:32:14,397 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.653e+02 3.031e+02 3.652e+02 7.428e+02, threshold=6.061e+02, percent-clipped=2.0 2023-05-16 22:32:18,112 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253380.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:32:25,001 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253390.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:32:37,844 INFO [finetune.py:992] (1/2) Epoch 13, batch 6600, loss[loss=0.1699, simple_loss=0.2608, pruned_loss=0.03945, over 12259.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2585, pruned_loss=0.03977, over 2365650.12 frames. ], batch size: 37, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:32:46,527 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253419.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:32:53,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 22:33:02,120 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253441.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:33:04,856 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0051, 5.9214, 5.5501, 5.5526, 6.0545, 5.2368, 5.5730, 5.4277], device='cuda:1'), covar=tensor([0.1556, 0.1012, 0.1210, 0.1707, 0.0854, 0.2424, 0.1787, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0508, 0.0403, 0.0458, 0.0475, 0.0446, 0.0401, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:33:07,496 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 22:33:14,104 INFO [finetune.py:992] (1/2) Epoch 13, batch 6650, loss[loss=0.1624, simple_loss=0.2579, pruned_loss=0.03345, over 12105.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2588, pruned_loss=0.04001, over 2362114.76 frames. ], batch size: 39, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:33:26,877 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.716e+02 3.254e+02 3.771e+02 1.116e+03, threshold=6.508e+02, percent-clipped=6.0 2023-05-16 22:33:30,020 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253480.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:33:49,771 INFO [finetune.py:992] (1/2) Epoch 13, batch 6700, loss[loss=0.1498, simple_loss=0.2345, pruned_loss=0.03256, over 12189.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2578, pruned_loss=0.03973, over 2360155.29 frames. ], batch size: 31, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:33:53,437 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253513.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:33:59,819 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253521.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:34:08,916 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3398, 5.1164, 5.2667, 5.2972, 4.7959, 4.8218, 4.6898, 5.1568], device='cuda:1'), covar=tensor([0.0902, 0.0855, 0.1082, 0.0887, 0.2696, 0.1818, 0.0732, 0.1533], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0706, 0.0625, 0.0642, 0.0858, 0.0755, 0.0563, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:34:13,488 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-05-16 22:34:26,706 INFO [finetune.py:992] (1/2) Epoch 13, batch 6750, loss[loss=0.1408, simple_loss=0.2298, pruned_loss=0.02597, over 12337.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2579, pruned_loss=0.03947, over 2353636.05 frames. ], batch size: 31, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:34:28,889 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253561.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:34:32,646 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4723, 4.7923, 2.9703, 2.7269, 3.9638, 2.6005, 4.0047, 3.4294], device='cuda:1'), covar=tensor([0.0688, 0.0461, 0.1228, 0.1608, 0.0331, 0.1448, 0.0508, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0261, 0.0178, 0.0204, 0.0144, 0.0182, 0.0201, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:34:33,283 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253567.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:34:39,475 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.798e+02 3.178e+02 3.722e+02 7.084e+02, threshold=6.356e+02, percent-clipped=0.0 2023-05-16 22:34:43,969 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253582.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:35:00,224 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2473, 1.9942, 2.4529, 2.2227, 2.3565, 2.4315, 1.9061, 2.4090], device='cuda:1'), covar=tensor([0.0162, 0.0387, 0.0320, 0.0245, 0.0230, 0.0215, 0.0353, 0.0227], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0208, 0.0193, 0.0191, 0.0222, 0.0168, 0.0201, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:35:02,856 INFO [finetune.py:992] (1/2) Epoch 13, batch 6800, loss[loss=0.1527, simple_loss=0.2377, pruned_loss=0.03388, over 12160.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2581, pruned_loss=0.0399, over 2354029.47 frames. ], batch size: 29, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:35:13,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 22:35:17,238 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253628.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:35:32,302 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253649.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:35:32,989 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253650.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:35:33,632 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5542, 6.2973, 5.8306, 5.8811, 6.3381, 5.7180, 5.9625, 5.8490], device='cuda:1'), covar=tensor([0.1362, 0.0794, 0.1030, 0.1919, 0.0804, 0.1975, 0.1462, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0504, 0.0398, 0.0453, 0.0472, 0.0442, 0.0398, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:35:38,507 INFO [finetune.py:992] (1/2) Epoch 13, batch 6850, loss[loss=0.1515, simple_loss=0.2299, pruned_loss=0.03651, over 12271.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2573, pruned_loss=0.03993, over 2360088.40 frames. ], batch size: 28, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:35:52,197 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.532e+02 3.056e+02 3.723e+02 6.224e+02, threshold=6.112e+02, percent-clipped=1.0 2023-05-16 22:35:58,653 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253685.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:36:07,857 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253697.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:36:15,505 INFO [finetune.py:992] (1/2) Epoch 13, batch 6900, loss[loss=0.1566, simple_loss=0.2436, pruned_loss=0.0348, over 12139.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2574, pruned_loss=0.04001, over 2359537.79 frames. ], batch size: 34, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:36:17,746 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253711.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:36:35,350 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253736.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:36:51,110 INFO [finetune.py:992] (1/2) Epoch 13, batch 6950, loss[loss=0.1502, simple_loss=0.2347, pruned_loss=0.03282, over 12150.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2572, pruned_loss=0.03975, over 2365155.70 frames. ], batch size: 30, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:37:03,241 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253775.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:37:03,824 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.804e+02 3.198e+02 4.094e+02 1.309e+03, threshold=6.395e+02, percent-clipped=4.0 2023-05-16 22:37:27,392 INFO [finetune.py:992] (1/2) Epoch 13, batch 7000, loss[loss=0.1932, simple_loss=0.2851, pruned_loss=0.05064, over 12098.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2571, pruned_loss=0.03967, over 2366922.52 frames. ], batch size: 42, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:37:39,106 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 22:37:43,036 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6862, 2.7684, 3.8251, 4.5885, 3.9753, 4.6065, 3.9686, 3.3589], device='cuda:1'), covar=tensor([0.0045, 0.0406, 0.0148, 0.0036, 0.0124, 0.0077, 0.0124, 0.0370], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0123, 0.0104, 0.0078, 0.0104, 0.0115, 0.0096, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:38:03,344 INFO [finetune.py:992] (1/2) Epoch 13, batch 7050, loss[loss=0.1395, simple_loss=0.2251, pruned_loss=0.02694, over 12168.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2577, pruned_loss=0.03982, over 2365102.56 frames. ], batch size: 31, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:38:16,172 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.719e+02 3.082e+02 3.666e+02 6.671e+02, threshold=6.164e+02, percent-clipped=1.0 2023-05-16 22:38:17,000 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253877.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:38:33,278 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 22:38:39,272 INFO [finetune.py:992] (1/2) Epoch 13, batch 7100, loss[loss=0.1703, simple_loss=0.2618, pruned_loss=0.03943, over 12138.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2571, pruned_loss=0.03937, over 2372199.32 frames. ], batch size: 34, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:38:49,884 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253923.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:39:08,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3345, 2.4540, 3.1345, 4.1650, 2.2027, 4.2762, 4.3344, 4.3504], device='cuda:1'), covar=tensor([0.0133, 0.1287, 0.0537, 0.0180, 0.1344, 0.0207, 0.0137, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0183, 0.0120, 0.0188, 0.0179, 0.0174, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:39:15,631 INFO [finetune.py:992] (1/2) Epoch 13, batch 7150, loss[loss=0.1624, simple_loss=0.2587, pruned_loss=0.03304, over 12366.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2561, pruned_loss=0.03894, over 2373732.30 frames. ], batch size: 36, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:39:28,437 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.775e+02 3.126e+02 3.691e+02 5.581e+02, threshold=6.252e+02, percent-clipped=0.0 2023-05-16 22:39:35,070 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253985.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:39:53,947 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254006.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:39:55,380 INFO [finetune.py:992] (1/2) Epoch 13, batch 7200, loss[loss=0.173, simple_loss=0.272, pruned_loss=0.037, over 12026.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2562, pruned_loss=0.03892, over 2378163.58 frames. ], batch size: 40, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:40:13,116 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254033.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:40:14,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-05-16 22:40:15,449 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254036.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:40:30,712 INFO [finetune.py:992] (1/2) Epoch 13, batch 7250, loss[loss=0.178, simple_loss=0.2732, pruned_loss=0.04136, over 12112.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.256, pruned_loss=0.03859, over 2386158.18 frames. ], batch size: 38, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:40:43,001 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254075.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:40:43,614 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.970e+02 2.641e+02 3.030e+02 3.732e+02 8.374e+02, threshold=6.060e+02, percent-clipped=2.0 2023-05-16 22:40:49,330 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254084.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:40:58,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.01 vs. limit=5.0 2023-05-16 22:41:06,367 INFO [finetune.py:992] (1/2) Epoch 13, batch 7300, loss[loss=0.1594, simple_loss=0.2436, pruned_loss=0.03758, over 12251.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2566, pruned_loss=0.03884, over 2391876.53 frames. ], batch size: 32, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:41:17,589 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254123.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:41:43,115 INFO [finetune.py:992] (1/2) Epoch 13, batch 7350, loss[loss=0.1649, simple_loss=0.258, pruned_loss=0.03588, over 11858.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2571, pruned_loss=0.03898, over 2383974.16 frames. ], batch size: 44, lr: 3.68e-03, grad_scale: 16.0 2023-05-16 22:41:55,834 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.809e+02 3.297e+02 3.980e+02 6.738e+02, threshold=6.593e+02, percent-clipped=2.0 2023-05-16 22:41:56,654 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254177.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:42:18,908 INFO [finetune.py:992] (1/2) Epoch 13, batch 7400, loss[loss=0.1869, simple_loss=0.2791, pruned_loss=0.04735, over 12098.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2573, pruned_loss=0.03909, over 2377566.01 frames. ], batch size: 42, lr: 3.67e-03, grad_scale: 16.0 2023-05-16 22:42:29,741 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254223.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:42:31,078 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254225.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:42:55,234 INFO [finetune.py:992] (1/2) Epoch 13, batch 7450, loss[loss=0.156, simple_loss=0.2479, pruned_loss=0.03209, over 12267.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2572, pruned_loss=0.03941, over 2378244.75 frames. ], batch size: 33, lr: 3.67e-03, grad_scale: 16.0 2023-05-16 22:42:57,304 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 2023-05-16 22:42:57,786 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1572, 3.5310, 3.5613, 3.9853, 2.8636, 3.4156, 2.5842, 3.3762], device='cuda:1'), covar=tensor([0.1691, 0.0849, 0.0958, 0.0671, 0.1116, 0.0742, 0.1753, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0261, 0.0295, 0.0358, 0.0242, 0.0242, 0.0258, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:43:04,669 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254271.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:43:08,182 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 2.663e+02 2.998e+02 3.648e+02 6.249e+02, threshold=5.996e+02, percent-clipped=0.0 2023-05-16 22:43:18,910 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0575, 2.3363, 3.5878, 3.0714, 3.5025, 3.2027, 2.5342, 3.5356], device='cuda:1'), covar=tensor([0.0135, 0.0393, 0.0151, 0.0247, 0.0160, 0.0169, 0.0377, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0193, 0.0190, 0.0220, 0.0166, 0.0200, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:43:30,057 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254306.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:43:31,348 INFO [finetune.py:992] (1/2) Epoch 13, batch 7500, loss[loss=0.1498, simple_loss=0.236, pruned_loss=0.03176, over 12344.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2567, pruned_loss=0.03948, over 2370446.37 frames. ], batch size: 31, lr: 3.67e-03, grad_scale: 16.0 2023-05-16 22:44:04,338 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254354.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:44:07,116 INFO [finetune.py:992] (1/2) Epoch 13, batch 7550, loss[loss=0.1664, simple_loss=0.2684, pruned_loss=0.03221, over 12140.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2568, pruned_loss=0.03942, over 2376802.78 frames. ], batch size: 36, lr: 3.67e-03, grad_scale: 16.0 2023-05-16 22:44:19,948 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.642e+02 3.202e+02 3.788e+02 1.184e+03, threshold=6.403e+02, percent-clipped=7.0 2023-05-16 22:44:43,828 INFO [finetune.py:992] (1/2) Epoch 13, batch 7600, loss[loss=0.1655, simple_loss=0.2591, pruned_loss=0.036, over 12111.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2565, pruned_loss=0.0394, over 2369565.05 frames. ], batch size: 39, lr: 3.67e-03, grad_scale: 16.0 2023-05-16 22:45:06,978 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1300, 5.9365, 5.5614, 5.4841, 6.0158, 5.2678, 5.4262, 5.5180], device='cuda:1'), covar=tensor([0.1408, 0.0855, 0.1158, 0.1700, 0.0867, 0.2122, 0.2106, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0503, 0.0401, 0.0457, 0.0474, 0.0444, 0.0403, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:45:13,294 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1261, 6.0794, 5.8435, 5.2329, 5.2487, 5.9678, 5.5944, 5.3984], device='cuda:1'), covar=tensor([0.0660, 0.0900, 0.0639, 0.1631, 0.0660, 0.0778, 0.1439, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0636, 0.0574, 0.0535, 0.0649, 0.0425, 0.0741, 0.0807, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 22:45:19,552 INFO [finetune.py:992] (1/2) Epoch 13, batch 7650, loss[loss=0.1496, simple_loss=0.2342, pruned_loss=0.0325, over 12350.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2564, pruned_loss=0.03947, over 2365140.04 frames. ], batch size: 30, lr: 3.67e-03, grad_scale: 16.0 2023-05-16 22:45:19,756 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1371, 2.3191, 3.1186, 4.0032, 2.2342, 4.1005, 4.1027, 4.1950], device='cuda:1'), covar=tensor([0.0141, 0.1387, 0.0519, 0.0165, 0.1383, 0.0263, 0.0209, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0183, 0.0120, 0.0189, 0.0180, 0.0174, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:45:32,326 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.660e+02 3.239e+02 3.711e+02 7.626e+02, threshold=6.478e+02, percent-clipped=3.0 2023-05-16 22:45:54,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4755, 4.7952, 4.0756, 5.0412, 4.6816, 3.1286, 4.2836, 3.1555], device='cuda:1'), covar=tensor([0.0758, 0.0771, 0.1597, 0.0524, 0.1127, 0.1547, 0.1002, 0.3278], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0376, 0.0358, 0.0317, 0.0367, 0.0270, 0.0345, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:45:55,171 INFO [finetune.py:992] (1/2) Epoch 13, batch 7700, loss[loss=0.1587, simple_loss=0.2548, pruned_loss=0.03132, over 12298.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2556, pruned_loss=0.03906, over 2375589.12 frames. ], batch size: 34, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:46:15,336 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254536.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:46:31,514 INFO [finetune.py:992] (1/2) Epoch 13, batch 7750, loss[loss=0.1462, simple_loss=0.229, pruned_loss=0.03172, over 12380.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2558, pruned_loss=0.03901, over 2379859.62 frames. ], batch size: 30, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:46:44,058 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254574.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:46:45,219 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.670e+02 3.101e+02 3.752e+02 9.239e+02, threshold=6.202e+02, percent-clipped=1.0 2023-05-16 22:46:53,337 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2583, 2.5993, 3.8188, 3.2398, 3.6862, 3.3363, 2.7347, 3.7281], device='cuda:1'), covar=tensor([0.0137, 0.0352, 0.0137, 0.0224, 0.0151, 0.0172, 0.0338, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0208, 0.0194, 0.0190, 0.0221, 0.0166, 0.0201, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:47:00,544 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254597.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:47:08,421 INFO [finetune.py:992] (1/2) Epoch 13, batch 7800, loss[loss=0.1612, simple_loss=0.2507, pruned_loss=0.03585, over 12284.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2556, pruned_loss=0.03907, over 2376128.19 frames. ], batch size: 33, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:47:28,071 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254635.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:47:44,202 INFO [finetune.py:992] (1/2) Epoch 13, batch 7850, loss[loss=0.1542, simple_loss=0.2503, pruned_loss=0.02902, over 12261.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2554, pruned_loss=0.0388, over 2381306.00 frames. ], batch size: 37, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:47:55,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.87 vs. limit=5.0 2023-05-16 22:47:57,226 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.782e+02 3.189e+02 3.657e+02 7.453e+02, threshold=6.378e+02, percent-clipped=3.0 2023-05-16 22:48:21,221 INFO [finetune.py:992] (1/2) Epoch 13, batch 7900, loss[loss=0.1669, simple_loss=0.2629, pruned_loss=0.03545, over 11219.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2554, pruned_loss=0.0387, over 2379436.95 frames. ], batch size: 55, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:48:54,189 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254755.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:48:56,126 INFO [finetune.py:992] (1/2) Epoch 13, batch 7950, loss[loss=0.1657, simple_loss=0.2629, pruned_loss=0.03422, over 10701.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2558, pruned_loss=0.03893, over 2374443.21 frames. ], batch size: 69, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:48:59,322 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1382, 4.5493, 4.0436, 4.8666, 4.4936, 2.8700, 4.1584, 2.9959], device='cuda:1'), covar=tensor([0.0882, 0.0875, 0.1506, 0.0617, 0.1121, 0.1756, 0.1253, 0.3445], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0377, 0.0358, 0.0318, 0.0369, 0.0271, 0.0344, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:49:08,892 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.549e+02 3.114e+02 3.760e+02 7.466e+02, threshold=6.228e+02, percent-clipped=3.0 2023-05-16 22:49:24,020 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254797.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:49:31,911 INFO [finetune.py:992] (1/2) Epoch 13, batch 8000, loss[loss=0.1866, simple_loss=0.2753, pruned_loss=0.04897, over 12040.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.256, pruned_loss=0.03923, over 2376535.85 frames. ], batch size: 40, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:49:37,731 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254816.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:49:47,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1229, 5.8846, 5.4969, 5.3821, 6.0232, 5.3188, 5.4381, 5.4278], device='cuda:1'), covar=tensor([0.1573, 0.0917, 0.0966, 0.1964, 0.0858, 0.1961, 0.1821, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0501, 0.0398, 0.0452, 0.0470, 0.0440, 0.0400, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:49:47,662 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7809, 3.3234, 5.1617, 2.7633, 3.1876, 3.9230, 3.3416, 3.8148], device='cuda:1'), covar=tensor([0.0442, 0.1176, 0.0323, 0.1082, 0.1565, 0.1345, 0.1272, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0241, 0.0258, 0.0185, 0.0240, 0.0298, 0.0228, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:49:50,494 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0278, 4.8960, 4.8423, 4.8363, 4.5378, 4.9990, 5.0063, 5.1399], device='cuda:1'), covar=tensor([0.0205, 0.0159, 0.0198, 0.0349, 0.0744, 0.0372, 0.0155, 0.0227], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0200, 0.0194, 0.0250, 0.0248, 0.0223, 0.0179, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 22:50:03,307 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254851.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:50:08,785 INFO [finetune.py:992] (1/2) Epoch 13, batch 8050, loss[loss=0.1689, simple_loss=0.2646, pruned_loss=0.03664, over 12271.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2561, pruned_loss=0.03951, over 2375995.36 frames. ], batch size: 37, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:50:09,036 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254858.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:50:21,576 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.618e+02 3.061e+02 3.754e+02 7.407e+02, threshold=6.122e+02, percent-clipped=2.0 2023-05-16 22:50:27,476 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6228, 2.1565, 3.0260, 2.6857, 2.9377, 2.8620, 2.2202, 2.9912], device='cuda:1'), covar=tensor([0.0141, 0.0391, 0.0202, 0.0256, 0.0170, 0.0172, 0.0359, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0209, 0.0195, 0.0190, 0.0222, 0.0167, 0.0201, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:50:32,924 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254892.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:50:44,221 INFO [finetune.py:992] (1/2) Epoch 13, batch 8100, loss[loss=0.1462, simple_loss=0.2388, pruned_loss=0.02683, over 12256.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2556, pruned_loss=0.03939, over 2372565.01 frames. ], batch size: 32, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:50:47,182 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254912.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:50:59,716 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254930.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:51:16,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 22:51:19,555 INFO [finetune.py:992] (1/2) Epoch 13, batch 8150, loss[loss=0.1473, simple_loss=0.2245, pruned_loss=0.03502, over 12023.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2567, pruned_loss=0.04024, over 2356775.12 frames. ], batch size: 28, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:51:24,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-16 22:51:33,175 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 2.598e+02 3.051e+02 3.699e+02 6.433e+02, threshold=6.102e+02, percent-clipped=1.0 2023-05-16 22:51:34,092 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8867, 4.6494, 4.8128, 4.8448, 4.7621, 4.9587, 4.7193, 2.4801], device='cuda:1'), covar=tensor([0.0117, 0.0087, 0.0106, 0.0064, 0.0051, 0.0098, 0.0090, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0075, 0.0061, 0.0094, 0.0083, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:51:52,777 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0580, 5.8839, 5.5119, 5.4638, 5.9928, 5.3047, 5.5303, 5.4685], device='cuda:1'), covar=tensor([0.1437, 0.0888, 0.1051, 0.1801, 0.0847, 0.1987, 0.1761, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0497, 0.0396, 0.0449, 0.0468, 0.0436, 0.0397, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:51:55,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-05-16 22:51:56,903 INFO [finetune.py:992] (1/2) Epoch 13, batch 8200, loss[loss=0.1582, simple_loss=0.2429, pruned_loss=0.03674, over 12353.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2562, pruned_loss=0.03974, over 2369574.18 frames. ], batch size: 31, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:52:13,445 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2630, 5.1080, 5.1896, 5.2505, 4.6838, 4.6894, 4.6887, 5.0190], device='cuda:1'), covar=tensor([0.0898, 0.0702, 0.0974, 0.0714, 0.2436, 0.1869, 0.0629, 0.1421], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0698, 0.0621, 0.0635, 0.0855, 0.0747, 0.0563, 0.0482], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:52:32,421 INFO [finetune.py:992] (1/2) Epoch 13, batch 8250, loss[loss=0.1528, simple_loss=0.2362, pruned_loss=0.03471, over 12366.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2569, pruned_loss=0.03971, over 2375114.31 frames. ], batch size: 30, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:52:45,117 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 2.808e+02 3.143e+02 3.527e+02 5.785e+02, threshold=6.286e+02, percent-clipped=0.0 2023-05-16 22:52:52,397 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0846, 6.0661, 5.7793, 5.2736, 5.2133, 5.9948, 5.6156, 5.2944], device='cuda:1'), covar=tensor([0.0637, 0.0831, 0.0661, 0.1715, 0.0625, 0.0710, 0.1398, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0631, 0.0570, 0.0533, 0.0647, 0.0422, 0.0733, 0.0796, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 22:53:04,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2538, 2.0395, 2.4003, 2.2588, 2.3610, 2.4246, 1.8922, 2.4222], device='cuda:1'), covar=tensor([0.0107, 0.0289, 0.0163, 0.0179, 0.0145, 0.0146, 0.0264, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0195, 0.0189, 0.0222, 0.0167, 0.0200, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:53:08,029 INFO [finetune.py:992] (1/2) Epoch 13, batch 8300, loss[loss=0.1628, simple_loss=0.2576, pruned_loss=0.03401, over 12321.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2571, pruned_loss=0.03987, over 2366312.41 frames. ], batch size: 34, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:53:10,230 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255111.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:53:25,803 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3890, 4.9870, 5.3121, 4.6498, 4.9949, 4.7813, 5.3578, 5.0493], device='cuda:1'), covar=tensor([0.0240, 0.0321, 0.0273, 0.0260, 0.0385, 0.0333, 0.0200, 0.0282], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0270, 0.0292, 0.0265, 0.0268, 0.0268, 0.0240, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 22:53:36,104 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7535, 3.4150, 5.1653, 2.5838, 2.7867, 3.8360, 3.3158, 3.8163], device='cuda:1'), covar=tensor([0.0478, 0.1144, 0.0431, 0.1224, 0.1918, 0.1415, 0.1307, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0240, 0.0257, 0.0185, 0.0240, 0.0297, 0.0228, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:53:41,739 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255153.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:53:45,157 INFO [finetune.py:992] (1/2) Epoch 13, batch 8350, loss[loss=0.2579, simple_loss=0.3281, pruned_loss=0.09382, over 8063.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2576, pruned_loss=0.04033, over 2354225.82 frames. ], batch size: 98, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:53:57,869 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 2.635e+02 2.966e+02 3.892e+02 6.227e+02, threshold=5.932e+02, percent-clipped=0.0 2023-05-16 22:54:09,502 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255192.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:54:20,387 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255207.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:54:20,966 INFO [finetune.py:992] (1/2) Epoch 13, batch 8400, loss[loss=0.1839, simple_loss=0.2759, pruned_loss=0.04597, over 12050.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2568, pruned_loss=0.03987, over 2361830.42 frames. ], batch size: 37, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:54:36,713 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255230.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:54:43,622 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255240.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:54:57,119 INFO [finetune.py:992] (1/2) Epoch 13, batch 8450, loss[loss=0.1607, simple_loss=0.2488, pruned_loss=0.03627, over 12358.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2571, pruned_loss=0.03991, over 2366672.37 frames. ], batch size: 31, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:55:10,082 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.590e+02 2.974e+02 3.616e+02 6.901e+02, threshold=5.947e+02, percent-clipped=2.0 2023-05-16 22:55:11,584 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255278.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:55:33,487 INFO [finetune.py:992] (1/2) Epoch 13, batch 8500, loss[loss=0.1783, simple_loss=0.2672, pruned_loss=0.04471, over 11863.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2578, pruned_loss=0.04035, over 2360435.56 frames. ], batch size: 44, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:55:43,868 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-16 22:56:09,131 INFO [finetune.py:992] (1/2) Epoch 13, batch 8550, loss[loss=0.1543, simple_loss=0.2388, pruned_loss=0.03493, over 12289.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2574, pruned_loss=0.04021, over 2363597.50 frames. ], batch size: 33, lr: 3.67e-03, grad_scale: 32.0 2023-05-16 22:56:09,354 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6199, 3.6715, 3.3820, 3.1820, 2.8787, 2.7669, 3.7578, 2.4338], device='cuda:1'), covar=tensor([0.0413, 0.0147, 0.0178, 0.0231, 0.0457, 0.0374, 0.0130, 0.0515], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0168, 0.0169, 0.0196, 0.0209, 0.0208, 0.0177, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:56:10,710 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1265, 6.0767, 5.8163, 5.3442, 5.2021, 6.0379, 5.6096, 5.4063], device='cuda:1'), covar=tensor([0.0656, 0.0922, 0.0724, 0.1435, 0.0668, 0.0733, 0.1593, 0.0965], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0571, 0.0533, 0.0645, 0.0423, 0.0738, 0.0796, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 22:56:21,833 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.802e+02 3.178e+02 3.829e+02 1.090e+03, threshold=6.355e+02, percent-clipped=3.0 2023-05-16 22:56:30,553 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255388.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:56:45,526 INFO [finetune.py:992] (1/2) Epoch 13, batch 8600, loss[loss=0.1549, simple_loss=0.2383, pruned_loss=0.03579, over 12013.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2573, pruned_loss=0.04012, over 2364517.55 frames. ], batch size: 28, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 22:56:47,737 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255411.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:56:49,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-16 22:57:15,788 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255449.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:57:18,530 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255453.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:57:21,902 INFO [finetune.py:992] (1/2) Epoch 13, batch 8650, loss[loss=0.1522, simple_loss=0.2446, pruned_loss=0.02987, over 12182.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2568, pruned_loss=0.03968, over 2367214.55 frames. ], batch size: 35, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 22:57:22,655 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255459.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:57:34,825 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.718e+02 3.189e+02 3.747e+02 9.115e+02, threshold=6.379e+02, percent-clipped=2.0 2023-05-16 22:57:52,688 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255501.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:57:57,103 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255507.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:57:57,632 INFO [finetune.py:992] (1/2) Epoch 13, batch 8700, loss[loss=0.1635, simple_loss=0.2522, pruned_loss=0.03741, over 12304.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2564, pruned_loss=0.03927, over 2371817.88 frames. ], batch size: 33, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 22:58:06,936 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3825, 4.1045, 4.2100, 4.2815, 4.1703, 4.3231, 4.2023, 2.4110], device='cuda:1'), covar=tensor([0.0122, 0.0091, 0.0098, 0.0083, 0.0065, 0.0109, 0.0156, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0075, 0.0061, 0.0094, 0.0083, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:58:21,971 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7055, 3.2333, 5.2130, 2.7338, 2.8771, 3.8587, 2.9985, 3.7702], device='cuda:1'), covar=tensor([0.0459, 0.1220, 0.0257, 0.1103, 0.1884, 0.1589, 0.1583, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0239, 0.0257, 0.0184, 0.0239, 0.0297, 0.0227, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 22:58:27,664 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8775, 3.0157, 4.8824, 4.9154, 2.8525, 2.7428, 3.1013, 2.3153], device='cuda:1'), covar=tensor([0.1550, 0.2811, 0.0355, 0.0399, 0.1364, 0.2291, 0.2575, 0.3792], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0386, 0.0273, 0.0299, 0.0271, 0.0306, 0.0383, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 22:58:28,191 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1058, 6.0651, 5.8097, 5.3201, 5.1986, 5.9857, 5.6020, 5.3897], device='cuda:1'), covar=tensor([0.0668, 0.0904, 0.0687, 0.1579, 0.0699, 0.0735, 0.1436, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0570, 0.0531, 0.0643, 0.0423, 0.0740, 0.0790, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 22:58:31,703 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255555.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 22:58:33,622 INFO [finetune.py:992] (1/2) Epoch 13, batch 8750, loss[loss=0.1775, simple_loss=0.2707, pruned_loss=0.04212, over 12301.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2564, pruned_loss=0.03876, over 2378468.48 frames. ], batch size: 34, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 22:58:46,346 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.685e+02 3.096e+02 3.779e+02 7.101e+02, threshold=6.191e+02, percent-clipped=1.0 2023-05-16 22:58:46,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-16 22:59:09,925 INFO [finetune.py:992] (1/2) Epoch 13, batch 8800, loss[loss=0.1418, simple_loss=0.2227, pruned_loss=0.03047, over 12342.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2572, pruned_loss=0.03924, over 2368508.42 frames. ], batch size: 31, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 22:59:26,560 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8450, 4.7158, 4.6579, 4.6851, 4.3474, 4.8458, 4.8400, 4.9891], device='cuda:1'), covar=tensor([0.0214, 0.0151, 0.0189, 0.0371, 0.0768, 0.0339, 0.0158, 0.0196], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0199, 0.0193, 0.0249, 0.0246, 0.0223, 0.0180, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 22:59:45,643 INFO [finetune.py:992] (1/2) Epoch 13, batch 8850, loss[loss=0.1346, simple_loss=0.2131, pruned_loss=0.02807, over 12184.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2566, pruned_loss=0.03897, over 2371615.71 frames. ], batch size: 29, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 22:59:58,468 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.704e+02 3.244e+02 3.773e+02 9.691e+02, threshold=6.489e+02, percent-clipped=3.0 2023-05-16 23:00:16,282 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1519, 5.9246, 5.5444, 5.4830, 6.0129, 5.2050, 5.4483, 5.4650], device='cuda:1'), covar=tensor([0.1427, 0.1039, 0.1103, 0.1913, 0.0965, 0.2241, 0.2015, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0502, 0.0400, 0.0459, 0.0474, 0.0441, 0.0400, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:00:21,996 INFO [finetune.py:992] (1/2) Epoch 13, batch 8900, loss[loss=0.2408, simple_loss=0.3143, pruned_loss=0.08361, over 8171.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2563, pruned_loss=0.03892, over 2375685.88 frames. ], batch size: 98, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:00:48,224 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255744.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:00:58,026 INFO [finetune.py:992] (1/2) Epoch 13, batch 8950, loss[loss=0.1507, simple_loss=0.2335, pruned_loss=0.03398, over 11999.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2564, pruned_loss=0.03903, over 2378856.89 frames. ], batch size: 28, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:01:10,950 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.075e+02 2.765e+02 3.126e+02 3.642e+02 6.981e+02, threshold=6.253e+02, percent-clipped=2.0 2023-05-16 23:01:34,143 INFO [finetune.py:992] (1/2) Epoch 13, batch 9000, loss[loss=0.1518, simple_loss=0.232, pruned_loss=0.03581, over 11994.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2562, pruned_loss=0.0393, over 2370453.18 frames. ], batch size: 28, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:01:34,143 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 23:01:49,897 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9354, 5.5039, 5.3055, 5.0638, 5.6259, 4.9037, 4.8719, 5.0821], device='cuda:1'), covar=tensor([0.1461, 0.0942, 0.0912, 0.1584, 0.0781, 0.2403, 0.2171, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0501, 0.0399, 0.0456, 0.0472, 0.0438, 0.0397, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:01:51,094 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.8863, 3.4205, 3.7971, 4.1820, 2.6810, 3.6824, 2.5226, 3.5435], device='cuda:1'), covar=tensor([0.1945, 0.1191, 0.0962, 0.0539, 0.1435, 0.0820, 0.2105, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0263, 0.0295, 0.0356, 0.0243, 0.0241, 0.0260, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:01:52,586 INFO [finetune.py:1026] (1/2) Epoch 13, validation: loss=0.3238, simple_loss=0.3958, pruned_loss=0.1259, over 1020973.00 frames. 2023-05-16 23:01:52,587 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 23:02:28,940 INFO [finetune.py:992] (1/2) Epoch 13, batch 9050, loss[loss=0.1999, simple_loss=0.2875, pruned_loss=0.05614, over 12041.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2557, pruned_loss=0.03876, over 2379763.70 frames. ], batch size: 40, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:02:41,567 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.848e+02 3.184e+02 4.017e+02 7.938e+02, threshold=6.367e+02, percent-clipped=2.0 2023-05-16 23:03:04,323 INFO [finetune.py:992] (1/2) Epoch 13, batch 9100, loss[loss=0.1885, simple_loss=0.2761, pruned_loss=0.05039, over 11842.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2563, pruned_loss=0.03897, over 2374851.47 frames. ], batch size: 44, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:03:28,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-05-16 23:03:39,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255956.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:03:40,612 INFO [finetune.py:992] (1/2) Epoch 13, batch 9150, loss[loss=0.2325, simple_loss=0.2977, pruned_loss=0.08366, over 7995.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2568, pruned_loss=0.03953, over 2364468.28 frames. ], batch size: 99, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:03:53,465 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.219e+02 2.849e+02 3.314e+02 4.165e+02 8.053e+02, threshold=6.628e+02, percent-clipped=4.0 2023-05-16 23:04:20,282 INFO [finetune.py:992] (1/2) Epoch 13, batch 9200, loss[loss=0.1552, simple_loss=0.2479, pruned_loss=0.03119, over 12253.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2564, pruned_loss=0.0393, over 2368051.24 frames. ], batch size: 32, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:04:26,868 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256017.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:04:31,075 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1769, 6.0265, 5.6667, 5.6111, 6.1357, 5.4094, 5.6240, 5.5290], device='cuda:1'), covar=tensor([0.1485, 0.0928, 0.1111, 0.1849, 0.0914, 0.2133, 0.1784, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0499, 0.0397, 0.0454, 0.0470, 0.0435, 0.0395, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:04:42,149 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256038.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:04:46,197 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256044.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:04:56,099 INFO [finetune.py:992] (1/2) Epoch 13, batch 9250, loss[loss=0.1864, simple_loss=0.2695, pruned_loss=0.05161, over 11690.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2555, pruned_loss=0.03902, over 2369867.49 frames. ], batch size: 48, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:05:08,910 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 2.567e+02 3.035e+02 3.603e+02 9.664e+02, threshold=6.070e+02, percent-clipped=1.0 2023-05-16 23:05:12,615 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256081.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:05:15,389 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256084.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:05:20,901 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256092.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:05:26,086 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256099.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:05:32,135 INFO [finetune.py:992] (1/2) Epoch 13, batch 9300, loss[loss=0.1896, simple_loss=0.2817, pruned_loss=0.04878, over 11781.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2555, pruned_loss=0.03872, over 2369149.76 frames. ], batch size: 44, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:05:49,869 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256132.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:05:56,808 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256142.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:05:58,968 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256145.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:06:07,904 INFO [finetune.py:992] (1/2) Epoch 13, batch 9350, loss[loss=0.19, simple_loss=0.2763, pruned_loss=0.0518, over 12363.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2564, pruned_loss=0.03926, over 2369412.12 frames. ], batch size: 35, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:06:20,558 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.667e+02 3.151e+02 3.931e+02 7.899e+02, threshold=6.301e+02, percent-clipped=2.0 2023-05-16 23:06:32,713 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256193.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:06:43,622 INFO [finetune.py:992] (1/2) Epoch 13, batch 9400, loss[loss=0.1643, simple_loss=0.2512, pruned_loss=0.03871, over 12249.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2556, pruned_loss=0.03901, over 2373767.41 frames. ], batch size: 32, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:07:19,728 INFO [finetune.py:992] (1/2) Epoch 13, batch 9450, loss[loss=0.1629, simple_loss=0.2549, pruned_loss=0.03544, over 12152.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2569, pruned_loss=0.03927, over 2367540.57 frames. ], batch size: 34, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:07:33,399 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.790e+02 3.151e+02 3.840e+02 8.265e+02, threshold=6.302e+02, percent-clipped=1.0 2023-05-16 23:07:39,955 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256285.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:07:56,468 INFO [finetune.py:992] (1/2) Epoch 13, batch 9500, loss[loss=0.1674, simple_loss=0.2659, pruned_loss=0.03447, over 12361.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2575, pruned_loss=0.03941, over 2359098.31 frames. ], batch size: 36, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:07:59,438 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256312.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:08:22,379 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8002, 2.4758, 2.8919, 3.7162, 2.2141, 3.7324, 3.7074, 3.8920], device='cuda:1'), covar=tensor([0.0154, 0.1148, 0.0536, 0.0186, 0.1334, 0.0364, 0.0223, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0206, 0.0187, 0.0121, 0.0192, 0.0183, 0.0179, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:08:23,846 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256346.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:08:32,183 INFO [finetune.py:992] (1/2) Epoch 13, batch 9550, loss[loss=0.1713, simple_loss=0.27, pruned_loss=0.03625, over 12192.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2568, pruned_loss=0.03919, over 2358120.86 frames. ], batch size: 35, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:08:42,929 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256372.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:08:45,574 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 2.580e+02 3.048e+02 3.870e+02 8.344e+02, threshold=6.096e+02, percent-clipped=3.0 2023-05-16 23:08:58,365 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256394.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:09:05,574 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2005, 4.8809, 5.1811, 4.4933, 4.9327, 4.6189, 5.1810, 4.9412], device='cuda:1'), covar=tensor([0.0339, 0.0413, 0.0346, 0.0295, 0.0379, 0.0377, 0.0273, 0.0360], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0269, 0.0292, 0.0264, 0.0265, 0.0266, 0.0237, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:09:07,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-05-16 23:09:08,882 INFO [finetune.py:992] (1/2) Epoch 13, batch 9600, loss[loss=0.1604, simple_loss=0.2532, pruned_loss=0.03378, over 12352.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2575, pruned_loss=0.03927, over 2358519.07 frames. ], batch size: 35, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:09:24,095 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1922, 2.5957, 3.6876, 3.1125, 3.5122, 3.2293, 2.6849, 3.5918], device='cuda:1'), covar=tensor([0.0125, 0.0339, 0.0126, 0.0237, 0.0155, 0.0175, 0.0346, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0195, 0.0190, 0.0221, 0.0169, 0.0201, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:09:26,959 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256433.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:09:29,649 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256437.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:09:31,804 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256440.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:09:44,586 INFO [finetune.py:992] (1/2) Epoch 13, batch 9650, loss[loss=0.1736, simple_loss=0.2679, pruned_loss=0.03966, over 12347.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2572, pruned_loss=0.03918, over 2355293.79 frames. ], batch size: 36, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:09:54,081 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256471.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:09:57,457 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.607e+02 3.105e+02 3.769e+02 8.630e+02, threshold=6.211e+02, percent-clipped=3.0 2023-05-16 23:10:06,124 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256488.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:10:14,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-16 23:10:20,831 INFO [finetune.py:992] (1/2) Epoch 13, batch 9700, loss[loss=0.1806, simple_loss=0.2692, pruned_loss=0.04594, over 11221.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2569, pruned_loss=0.03906, over 2360479.17 frames. ], batch size: 55, lr: 3.66e-03, grad_scale: 64.0 2023-05-16 23:10:31,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-16 23:10:38,078 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256532.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:10:50,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0946, 6.1197, 5.8094, 5.3968, 5.2120, 5.9883, 5.6046, 5.4088], device='cuda:1'), covar=tensor([0.0709, 0.0807, 0.0623, 0.1438, 0.0640, 0.0711, 0.1513, 0.0976], device='cuda:1'), in_proj_covar=tensor([0.0634, 0.0574, 0.0533, 0.0650, 0.0428, 0.0747, 0.0798, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-16 23:10:56,424 INFO [finetune.py:992] (1/2) Epoch 13, batch 9750, loss[loss=0.1427, simple_loss=0.2308, pruned_loss=0.02731, over 12184.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2567, pruned_loss=0.0392, over 2360626.42 frames. ], batch size: 31, lr: 3.66e-03, grad_scale: 32.0 2023-05-16 23:11:10,586 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.672e+02 3.266e+02 4.073e+02 7.510e+02, threshold=6.532e+02, percent-clipped=3.0 2023-05-16 23:11:22,107 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6349, 4.4788, 4.4831, 4.4611, 4.1859, 4.5810, 4.6065, 4.7714], device='cuda:1'), covar=tensor([0.0233, 0.0195, 0.0199, 0.0457, 0.0718, 0.0488, 0.0194, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0201, 0.0195, 0.0252, 0.0248, 0.0226, 0.0182, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-16 23:11:32,948 INFO [finetune.py:992] (1/2) Epoch 13, batch 9800, loss[loss=0.1589, simple_loss=0.2495, pruned_loss=0.03414, over 12023.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2568, pruned_loss=0.03922, over 2357654.42 frames. ], batch size: 31, lr: 3.65e-03, grad_scale: 32.0 2023-05-16 23:11:35,990 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256612.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:11:56,527 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256641.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:12:02,169 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2334, 4.8677, 5.2375, 4.5442, 4.8944, 4.6971, 5.2070, 4.9678], device='cuda:1'), covar=tensor([0.0260, 0.0385, 0.0281, 0.0256, 0.0374, 0.0309, 0.0218, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0268, 0.0291, 0.0263, 0.0265, 0.0266, 0.0236, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:12:08,508 INFO [finetune.py:992] (1/2) Epoch 13, batch 9850, loss[loss=0.1604, simple_loss=0.2564, pruned_loss=0.03215, over 12190.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2561, pruned_loss=0.03861, over 2368885.68 frames. ], batch size: 35, lr: 3.65e-03, grad_scale: 32.0 2023-05-16 23:12:09,968 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256660.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:12:19,903 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1092, 3.8740, 4.0212, 4.3733, 3.0013, 3.7926, 2.6431, 4.0757], device='cuda:1'), covar=tensor([0.1788, 0.0806, 0.1034, 0.0636, 0.1227, 0.0666, 0.1882, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0265, 0.0298, 0.0361, 0.0244, 0.0243, 0.0262, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:12:22,418 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.747e+02 3.196e+02 3.714e+02 8.972e+02, threshold=6.391e+02, percent-clipped=3.0 2023-05-16 23:12:34,622 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256694.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:12:44,393 INFO [finetune.py:992] (1/2) Epoch 13, batch 9900, loss[loss=0.18, simple_loss=0.2761, pruned_loss=0.04192, over 11983.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2565, pruned_loss=0.03905, over 2363483.06 frames. ], batch size: 40, lr: 3.65e-03, grad_scale: 32.0 2023-05-16 23:12:59,667 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256728.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:05,919 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256737.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:07,970 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256740.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:09,277 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256742.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:15,056 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8412, 3.0683, 4.7543, 4.8980, 3.0586, 2.7756, 3.0866, 2.4402], device='cuda:1'), covar=tensor([0.1523, 0.2892, 0.0411, 0.0364, 0.1243, 0.2274, 0.2684, 0.3641], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0387, 0.0274, 0.0300, 0.0271, 0.0308, 0.0385, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:13:19,172 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7774, 3.0019, 4.5934, 4.7637, 2.9560, 2.6962, 2.9823, 2.2924], device='cuda:1'), covar=tensor([0.1495, 0.2796, 0.0444, 0.0390, 0.1280, 0.2279, 0.2731, 0.3833], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0387, 0.0274, 0.0300, 0.0271, 0.0308, 0.0385, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:13:20,312 INFO [finetune.py:992] (1/2) Epoch 13, batch 9950, loss[loss=0.1538, simple_loss=0.2439, pruned_loss=0.03188, over 12316.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2567, pruned_loss=0.03917, over 2367279.92 frames. ], batch size: 34, lr: 3.65e-03, grad_scale: 32.0 2023-05-16 23:13:23,477 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9117, 3.4565, 5.2521, 2.6983, 2.9858, 3.9723, 3.5321, 3.9419], device='cuda:1'), covar=tensor([0.0386, 0.1098, 0.0342, 0.1153, 0.1825, 0.1458, 0.1142, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0235, 0.0256, 0.0183, 0.0237, 0.0295, 0.0226, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:13:34,445 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.747e+02 3.108e+02 3.691e+02 6.204e+02, threshold=6.215e+02, percent-clipped=0.0 2023-05-16 23:13:39,342 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256785.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:41,486 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256788.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:41,612 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256788.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:13:46,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-16 23:13:56,569 INFO [finetune.py:992] (1/2) Epoch 13, batch 10000, loss[loss=0.1921, simple_loss=0.2816, pruned_loss=0.0513, over 12009.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.256, pruned_loss=0.03894, over 2365405.79 frames. ], batch size: 42, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:13:57,484 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2234, 2.7772, 3.8298, 3.2170, 3.6097, 3.3522, 2.9049, 3.6651], device='cuda:1'), covar=tensor([0.0146, 0.0334, 0.0168, 0.0244, 0.0158, 0.0175, 0.0284, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0211, 0.0198, 0.0193, 0.0225, 0.0171, 0.0205, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:14:10,205 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256827.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:14:10,341 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256827.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:14:13,210 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3874, 2.4194, 3.1754, 4.2625, 2.0833, 4.2208, 4.3790, 4.4517], device='cuda:1'), covar=tensor([0.0155, 0.1370, 0.0512, 0.0193, 0.1470, 0.0250, 0.0174, 0.0103], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0207, 0.0188, 0.0122, 0.0193, 0.0185, 0.0181, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:14:16,517 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256836.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:14:32,907 INFO [finetune.py:992] (1/2) Epoch 13, batch 10050, loss[loss=0.1503, simple_loss=0.2334, pruned_loss=0.0336, over 12407.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2568, pruned_loss=0.03941, over 2366300.65 frames. ], batch size: 32, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:14:46,917 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.663e+02 3.175e+02 3.942e+02 8.687e+02, threshold=6.350e+02, percent-clipped=3.0 2023-05-16 23:14:50,668 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256883.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:14:54,226 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256888.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:15:08,018 INFO [finetune.py:992] (1/2) Epoch 13, batch 10100, loss[loss=0.1776, simple_loss=0.2691, pruned_loss=0.04311, over 12194.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2571, pruned_loss=0.03924, over 2369010.22 frames. ], batch size: 35, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:15:31,683 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256941.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:15:33,933 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256944.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:15:44,627 INFO [finetune.py:992] (1/2) Epoch 13, batch 10150, loss[loss=0.1882, simple_loss=0.2803, pruned_loss=0.04804, over 12284.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2571, pruned_loss=0.03904, over 2369034.96 frames. ], batch size: 37, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:15:44,787 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0427, 3.8816, 4.0301, 3.6258, 3.8683, 3.7103, 4.0142, 3.6677], device='cuda:1'), covar=tensor([0.0341, 0.0336, 0.0304, 0.0240, 0.0349, 0.0310, 0.0273, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0270, 0.0293, 0.0265, 0.0267, 0.0269, 0.0239, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:15:58,633 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.565e+02 3.029e+02 3.805e+02 7.887e+02, threshold=6.058e+02, percent-clipped=0.0 2023-05-16 23:16:01,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.20 vs. limit=5.0 2023-05-16 23:16:06,504 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256989.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:16:08,701 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1103, 5.9617, 5.5454, 5.5159, 5.9781, 5.2532, 5.4954, 5.4931], device='cuda:1'), covar=tensor([0.1370, 0.0836, 0.1088, 0.1822, 0.0852, 0.1841, 0.1589, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0494, 0.0392, 0.0446, 0.0464, 0.0430, 0.0390, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:16:20,978 INFO [finetune.py:992] (1/2) Epoch 13, batch 10200, loss[loss=0.1455, simple_loss=0.2328, pruned_loss=0.02905, over 12371.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2573, pruned_loss=0.03955, over 2371105.75 frames. ], batch size: 31, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:16:24,513 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257013.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:16:34,946 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257028.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:16:55,844 INFO [finetune.py:992] (1/2) Epoch 13, batch 10250, loss[loss=0.1893, simple_loss=0.2903, pruned_loss=0.04416, over 11238.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2579, pruned_loss=0.03991, over 2361927.61 frames. ], batch size: 55, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:16:57,481 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2845, 4.7726, 3.0010, 2.9497, 3.9794, 2.6255, 3.9709, 3.3364], device='cuda:1'), covar=tensor([0.0748, 0.0433, 0.1092, 0.1340, 0.0330, 0.1339, 0.0460, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0260, 0.0178, 0.0202, 0.0145, 0.0181, 0.0200, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:17:07,409 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257074.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:17:08,709 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257076.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:17:10,108 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 2.697e+02 3.195e+02 3.744e+02 8.707e+02, threshold=6.389e+02, percent-clipped=4.0 2023-05-16 23:17:22,219 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3667, 4.8402, 3.0484, 2.9384, 4.0461, 2.6370, 4.0211, 3.3937], device='cuda:1'), covar=tensor([0.0656, 0.0501, 0.1011, 0.1359, 0.0261, 0.1234, 0.0430, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0259, 0.0177, 0.0201, 0.0144, 0.0180, 0.0199, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:17:22,900 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4009, 4.8200, 2.9797, 2.7922, 4.0067, 2.7169, 3.9856, 3.4620], device='cuda:1'), covar=tensor([0.0622, 0.0374, 0.1097, 0.1412, 0.0291, 0.1187, 0.0476, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0259, 0.0177, 0.0201, 0.0144, 0.0180, 0.0199, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:17:32,051 INFO [finetune.py:992] (1/2) Epoch 13, batch 10300, loss[loss=0.1503, simple_loss=0.2325, pruned_loss=0.03406, over 11771.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2576, pruned_loss=0.03976, over 2361867.39 frames. ], batch size: 26, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:17:45,885 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257127.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:17:56,489 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4098, 4.8536, 3.1648, 2.9565, 4.0359, 2.6412, 4.0291, 3.5350], device='cuda:1'), covar=tensor([0.0728, 0.0493, 0.1062, 0.1413, 0.0362, 0.1332, 0.0480, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0259, 0.0177, 0.0202, 0.0144, 0.0181, 0.0199, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:18:08,401 INFO [finetune.py:992] (1/2) Epoch 13, batch 10350, loss[loss=0.1453, simple_loss=0.2355, pruned_loss=0.02753, over 12409.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2566, pruned_loss=0.03947, over 2362839.91 frames. ], batch size: 32, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:18:13,628 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257165.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:18:20,725 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257175.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:18:22,863 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.630e+02 3.099e+02 3.655e+02 9.989e+02, threshold=6.198e+02, percent-clipped=1.0 2023-05-16 23:18:26,412 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257183.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:18:44,321 INFO [finetune.py:992] (1/2) Epoch 13, batch 10400, loss[loss=0.1671, simple_loss=0.2648, pruned_loss=0.0347, over 12160.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2561, pruned_loss=0.03914, over 2371677.28 frames. ], batch size: 36, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:18:54,531 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257222.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:18:57,230 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257226.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:19:07,156 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257239.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:19:08,580 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257241.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:19:20,100 INFO [finetune.py:992] (1/2) Epoch 13, batch 10450, loss[loss=0.2395, simple_loss=0.3112, pruned_loss=0.08394, over 7998.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2566, pruned_loss=0.0392, over 2374434.75 frames. ], batch size: 99, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:19:34,111 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.638e+02 3.175e+02 3.855e+02 7.127e+02, threshold=6.350e+02, percent-clipped=2.0 2023-05-16 23:19:38,585 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257283.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:19:52,297 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257302.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:19:56,472 INFO [finetune.py:992] (1/2) Epoch 13, batch 10500, loss[loss=0.168, simple_loss=0.2633, pruned_loss=0.03637, over 12304.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2565, pruned_loss=0.03934, over 2372720.45 frames. ], batch size: 34, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:20:32,019 INFO [finetune.py:992] (1/2) Epoch 13, batch 10550, loss[loss=0.1891, simple_loss=0.2744, pruned_loss=0.05189, over 11673.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2565, pruned_loss=0.03917, over 2371913.32 frames. ], batch size: 48, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:20:39,930 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257369.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:20:41,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3305, 3.4288, 3.1047, 3.0458, 2.6795, 2.6240, 3.5382, 2.1945], device='cuda:1'), covar=tensor([0.0430, 0.0173, 0.0200, 0.0214, 0.0441, 0.0384, 0.0142, 0.0528], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0164, 0.0166, 0.0190, 0.0201, 0.0201, 0.0172, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:20:46,425 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.617e+02 2.916e+02 3.833e+02 7.684e+02, threshold=5.832e+02, percent-clipped=1.0 2023-05-16 23:21:04,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-05-16 23:21:08,919 INFO [finetune.py:992] (1/2) Epoch 13, batch 10600, loss[loss=0.1551, simple_loss=0.2432, pruned_loss=0.03351, over 12155.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2567, pruned_loss=0.03921, over 2370566.73 frames. ], batch size: 34, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:21:38,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-16 23:21:45,025 INFO [finetune.py:992] (1/2) Epoch 13, batch 10650, loss[loss=0.1445, simple_loss=0.238, pruned_loss=0.02547, over 12272.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2569, pruned_loss=0.03903, over 2370661.24 frames. ], batch size: 37, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:21:51,525 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257467.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:21:59,304 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.794e+02 3.204e+02 3.744e+02 1.956e+03, threshold=6.408e+02, percent-clipped=5.0 2023-05-16 23:22:03,130 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257483.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:22:20,610 INFO [finetune.py:992] (1/2) Epoch 13, batch 10700, loss[loss=0.2033, simple_loss=0.2853, pruned_loss=0.06069, over 12049.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2574, pruned_loss=0.03948, over 2358424.29 frames. ], batch size: 37, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:22:29,851 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257521.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:22:35,645 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257528.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:22:37,545 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257531.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:22:43,330 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257539.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:22:56,853 INFO [finetune.py:992] (1/2) Epoch 13, batch 10750, loss[loss=0.1659, simple_loss=0.2631, pruned_loss=0.0344, over 12140.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2569, pruned_loss=0.03935, over 2365166.10 frames. ], batch size: 36, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:23:11,960 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.827e+02 2.658e+02 3.256e+02 3.758e+02 5.619e+02, threshold=6.512e+02, percent-clipped=0.0 2023-05-16 23:23:12,062 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257578.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:23:12,109 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257578.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:23:18,424 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257587.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:23:25,393 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257597.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:23:33,280 INFO [finetune.py:992] (1/2) Epoch 13, batch 10800, loss[loss=0.1538, simple_loss=0.25, pruned_loss=0.02876, over 12342.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.257, pruned_loss=0.03943, over 2363790.65 frames. ], batch size: 36, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:23:45,381 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1069, 3.9151, 2.5093, 2.2906, 3.4531, 2.4528, 3.5543, 2.9503], device='cuda:1'), covar=tensor([0.0715, 0.0709, 0.1165, 0.1624, 0.0307, 0.1252, 0.0502, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0260, 0.0177, 0.0201, 0.0143, 0.0180, 0.0199, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:23:55,460 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257639.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:24:09,014 INFO [finetune.py:992] (1/2) Epoch 13, batch 10850, loss[loss=0.177, simple_loss=0.2672, pruned_loss=0.04343, over 12340.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2564, pruned_loss=0.03915, over 2367894.54 frames. ], batch size: 36, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:24:16,935 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257669.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:24:23,610 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 2.635e+02 3.079e+02 3.670e+02 7.869e+02, threshold=6.158e+02, percent-clipped=2.0 2023-05-16 23:24:43,574 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6581, 3.7302, 3.3838, 3.2567, 2.8838, 2.8140, 3.7549, 2.4510], device='cuda:1'), covar=tensor([0.0367, 0.0154, 0.0174, 0.0212, 0.0413, 0.0374, 0.0154, 0.0472], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0166, 0.0167, 0.0192, 0.0205, 0.0205, 0.0174, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:24:46,269 INFO [finetune.py:992] (1/2) Epoch 13, batch 10900, loss[loss=0.1469, simple_loss=0.2365, pruned_loss=0.02868, over 12337.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2557, pruned_loss=0.03893, over 2373254.96 frames. ], batch size: 30, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:24:48,588 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8142, 4.3819, 4.4463, 4.5836, 4.5657, 4.6132, 4.4387, 2.1369], device='cuda:1'), covar=tensor([0.0157, 0.0126, 0.0163, 0.0115, 0.0083, 0.0206, 0.0153, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0075, 0.0061, 0.0094, 0.0083, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:24:52,562 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257717.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:25:08,561 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1429, 2.5710, 3.7560, 3.0916, 3.5305, 3.2224, 2.6986, 3.6176], device='cuda:1'), covar=tensor([0.0142, 0.0393, 0.0136, 0.0271, 0.0168, 0.0194, 0.0391, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0208, 0.0195, 0.0191, 0.0224, 0.0169, 0.0202, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:25:22,539 INFO [finetune.py:992] (1/2) Epoch 13, batch 10950, loss[loss=0.1653, simple_loss=0.249, pruned_loss=0.04073, over 12252.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2566, pruned_loss=0.0389, over 2375165.77 frames. ], batch size: 32, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:25:36,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.625e+02 3.127e+02 3.912e+02 6.509e+02, threshold=6.254e+02, percent-clipped=2.0 2023-05-16 23:25:41,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-16 23:25:58,129 INFO [finetune.py:992] (1/2) Epoch 13, batch 11000, loss[loss=0.2365, simple_loss=0.3203, pruned_loss=0.07632, over 10229.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2587, pruned_loss=0.04015, over 2358453.13 frames. ], batch size: 68, lr: 3.65e-03, grad_scale: 16.0 2023-05-16 23:26:08,339 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257821.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:26:09,735 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257823.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:26:14,150 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5136, 3.6179, 3.2721, 3.2397, 2.8543, 2.8569, 3.6688, 2.3257], device='cuda:1'), covar=tensor([0.0406, 0.0156, 0.0247, 0.0204, 0.0438, 0.0369, 0.0134, 0.0504], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0165, 0.0166, 0.0190, 0.0203, 0.0203, 0.0173, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:26:34,546 INFO [finetune.py:992] (1/2) Epoch 13, batch 11050, loss[loss=0.1677, simple_loss=0.256, pruned_loss=0.03971, over 12177.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2621, pruned_loss=0.04176, over 2334803.49 frames. ], batch size: 31, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:26:35,472 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257859.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:26:42,251 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257869.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:26:48,436 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.191e+02 3.085e+02 3.639e+02 4.467e+02 8.268e+02, threshold=7.277e+02, percent-clipped=4.0 2023-05-16 23:26:48,671 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257878.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:26:52,261 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257883.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:27:00,541 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2243, 5.1048, 4.9816, 5.1029, 4.7440, 5.1956, 5.2572, 5.3959], device='cuda:1'), covar=tensor([0.0179, 0.0131, 0.0171, 0.0340, 0.0745, 0.0242, 0.0140, 0.0164], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0199, 0.0195, 0.0250, 0.0247, 0.0224, 0.0179, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-16 23:27:01,888 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257897.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:27:09,264 INFO [finetune.py:992] (1/2) Epoch 13, batch 11100, loss[loss=0.2796, simple_loss=0.336, pruned_loss=0.1116, over 7907.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2648, pruned_loss=0.04363, over 2291378.44 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:27:18,711 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257920.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:27:22,676 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257926.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:27:28,944 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257934.0, num_to_drop=1, layers_to_drop={3} 2023-05-16 23:27:35,843 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257944.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:27:36,349 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257945.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:27:40,033 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257950.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:27:45,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-05-16 23:27:45,419 INFO [finetune.py:992] (1/2) Epoch 13, batch 11150, loss[loss=0.1986, simple_loss=0.287, pruned_loss=0.05514, over 11581.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2705, pruned_loss=0.04691, over 2237125.48 frames. ], batch size: 48, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:27:58,268 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4619, 3.6117, 3.3405, 3.1667, 2.8574, 2.6871, 3.5880, 2.1811], device='cuda:1'), covar=tensor([0.0421, 0.0153, 0.0151, 0.0223, 0.0365, 0.0378, 0.0145, 0.0553], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0163, 0.0164, 0.0189, 0.0199, 0.0201, 0.0170, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:27:59,421 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.323e+02 3.230e+02 4.017e+02 4.830e+02 1.321e+03, threshold=8.034e+02, percent-clipped=3.0 2023-05-16 23:28:24,879 INFO [finetune.py:992] (1/2) Epoch 13, batch 11200, loss[loss=0.3571, simple_loss=0.3942, pruned_loss=0.16, over 6868.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2778, pruned_loss=0.0517, over 2166966.07 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:28:27,236 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258011.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:28:37,713 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4135, 2.1614, 3.1396, 4.2741, 2.1083, 4.2986, 4.4465, 4.4997], device='cuda:1'), covar=tensor([0.0151, 0.1556, 0.0593, 0.0190, 0.1528, 0.0242, 0.0151, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0206, 0.0186, 0.0120, 0.0189, 0.0181, 0.0177, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:28:59,654 INFO [finetune.py:992] (1/2) Epoch 13, batch 11250, loss[loss=0.2075, simple_loss=0.2964, pruned_loss=0.05929, over 10267.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2838, pruned_loss=0.05569, over 2116189.85 frames. ], batch size: 68, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:29:02,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-16 23:29:14,068 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.627e+02 3.593e+02 4.439e+02 5.299e+02 8.794e+02, threshold=8.879e+02, percent-clipped=2.0 2023-05-16 23:29:20,211 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6714, 2.1353, 3.0628, 3.7135, 2.1377, 3.6865, 3.5359, 3.7984], device='cuda:1'), covar=tensor([0.0147, 0.1421, 0.0422, 0.0143, 0.1415, 0.0191, 0.0439, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0206, 0.0186, 0.0120, 0.0190, 0.0181, 0.0177, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:29:35,212 INFO [finetune.py:992] (1/2) Epoch 13, batch 11300, loss[loss=0.1682, simple_loss=0.2647, pruned_loss=0.03582, over 12351.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2889, pruned_loss=0.05858, over 2077084.41 frames. ], batch size: 30, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:29:45,484 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258123.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:29:51,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7556, 3.0458, 2.3394, 2.1937, 2.7286, 2.2591, 2.9520, 2.5489], device='cuda:1'), covar=tensor([0.0492, 0.0639, 0.0893, 0.1379, 0.0265, 0.1165, 0.0448, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0254, 0.0174, 0.0197, 0.0139, 0.0178, 0.0195, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:30:04,579 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1464, 6.0443, 5.8436, 5.3732, 5.2448, 6.0678, 5.7240, 5.4866], device='cuda:1'), covar=tensor([0.0658, 0.1118, 0.0744, 0.1897, 0.0642, 0.0740, 0.1418, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0553, 0.0513, 0.0624, 0.0414, 0.0720, 0.0763, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-16 23:30:10,571 INFO [finetune.py:992] (1/2) Epoch 13, batch 11350, loss[loss=0.2685, simple_loss=0.3421, pruned_loss=0.0975, over 6654.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2941, pruned_loss=0.06163, over 2021950.18 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:30:19,369 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258171.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:30:23,983 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.439e+02 3.322e+02 4.233e+02 4.928e+02 1.163e+03, threshold=8.466e+02, percent-clipped=1.0 2023-05-16 23:30:36,259 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6951, 2.7596, 4.0375, 4.2159, 3.0190, 2.6404, 2.7641, 2.1335], device='cuda:1'), covar=tensor([0.1424, 0.2510, 0.0490, 0.0376, 0.1012, 0.2156, 0.2509, 0.4117], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0381, 0.0272, 0.0297, 0.0270, 0.0305, 0.0382, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:30:44,442 INFO [finetune.py:992] (1/2) Epoch 13, batch 11400, loss[loss=0.2638, simple_loss=0.3447, pruned_loss=0.09143, over 7078.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2984, pruned_loss=0.06484, over 1964169.82 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:30:50,543 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258215.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:30:51,344 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9182, 2.1574, 2.8437, 2.7194, 3.0115, 3.0200, 2.8969, 2.4051], device='cuda:1'), covar=tensor([0.0086, 0.0424, 0.0188, 0.0112, 0.0109, 0.0116, 0.0146, 0.0416], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0126, 0.0107, 0.0080, 0.0107, 0.0118, 0.0099, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:31:03,636 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258234.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:31:05,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-05-16 23:31:06,905 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258239.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:31:19,666 INFO [finetune.py:992] (1/2) Epoch 13, batch 11450, loss[loss=0.2133, simple_loss=0.3002, pruned_loss=0.06319, over 11157.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.3026, pruned_loss=0.0686, over 1900695.35 frames. ], batch size: 55, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:31:34,359 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.211e+02 3.543e+02 4.076e+02 4.810e+02 8.439e+02, threshold=8.152e+02, percent-clipped=0.0 2023-05-16 23:31:37,135 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258282.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:31:43,961 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8944, 3.7605, 3.9003, 3.5950, 3.7789, 3.5929, 3.8575, 3.5610], device='cuda:1'), covar=tensor([0.0389, 0.0388, 0.0319, 0.0279, 0.0431, 0.0335, 0.0374, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0261, 0.0283, 0.0257, 0.0258, 0.0259, 0.0234, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:31:45,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-16 23:31:53,356 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258306.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:31:54,640 INFO [finetune.py:992] (1/2) Epoch 13, batch 11500, loss[loss=0.2251, simple_loss=0.3019, pruned_loss=0.07421, over 11539.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3052, pruned_loss=0.07097, over 1853822.14 frames. ], batch size: 48, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:32:30,657 INFO [finetune.py:992] (1/2) Epoch 13, batch 11550, loss[loss=0.2261, simple_loss=0.2983, pruned_loss=0.077, over 6848.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3076, pruned_loss=0.07322, over 1818617.82 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:32:32,957 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8098, 3.8004, 3.8088, 3.8942, 3.7059, 3.7490, 3.6482, 3.7924], device='cuda:1'), covar=tensor([0.1046, 0.0591, 0.1274, 0.0641, 0.1427, 0.1029, 0.0535, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0665, 0.0585, 0.0600, 0.0805, 0.0710, 0.0532, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:32:44,068 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.695e+02 3.630e+02 4.010e+02 4.837e+02 8.435e+02, threshold=8.019e+02, percent-clipped=2.0 2023-05-16 23:32:55,748 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3193, 5.2544, 5.1183, 4.6810, 4.6889, 5.2542, 4.9397, 4.8066], device='cuda:1'), covar=tensor([0.0658, 0.0905, 0.0633, 0.1622, 0.1127, 0.0668, 0.1338, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0546, 0.0505, 0.0616, 0.0408, 0.0706, 0.0750, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-16 23:33:03,764 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-16 23:33:06,086 INFO [finetune.py:992] (1/2) Epoch 13, batch 11600, loss[loss=0.216, simple_loss=0.2999, pruned_loss=0.06605, over 10338.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3091, pruned_loss=0.07443, over 1805263.37 frames. ], batch size: 69, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:33:39,086 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258454.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:33:41,834 INFO [finetune.py:992] (1/2) Epoch 13, batch 11650, loss[loss=0.1895, simple_loss=0.2793, pruned_loss=0.04984, over 10395.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3078, pruned_loss=0.07417, over 1788311.86 frames. ], batch size: 68, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:33:52,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-16 23:33:57,121 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.268e+02 3.530e+02 3.883e+02 4.638e+02 7.401e+02, threshold=7.767e+02, percent-clipped=0.0 2023-05-16 23:34:17,020 INFO [finetune.py:992] (1/2) Epoch 13, batch 11700, loss[loss=0.2636, simple_loss=0.3195, pruned_loss=0.1038, over 6842.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3083, pruned_loss=0.07571, over 1759136.96 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:34:19,041 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258511.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:34:21,779 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258515.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:34:21,819 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258515.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:34:39,258 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258539.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:34:51,849 INFO [finetune.py:992] (1/2) Epoch 13, batch 11750, loss[loss=0.2286, simple_loss=0.3128, pruned_loss=0.07217, over 10268.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3091, pruned_loss=0.07672, over 1736722.46 frames. ], batch size: 68, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:34:55,258 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258563.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:35:01,723 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258572.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:35:05,453 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.510e+02 3.510e+02 4.130e+02 4.908e+02 7.908e+02, threshold=8.259e+02, percent-clipped=1.0 2023-05-16 23:35:12,900 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258587.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:35:24,026 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7858, 3.7725, 3.7836, 3.8912, 3.6882, 3.7315, 3.6287, 3.7937], device='cuda:1'), covar=tensor([0.1232, 0.0707, 0.1604, 0.0650, 0.1517, 0.1216, 0.0568, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0657, 0.0580, 0.0591, 0.0794, 0.0700, 0.0526, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:35:26,136 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258606.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:35:27,320 INFO [finetune.py:992] (1/2) Epoch 13, batch 11800, loss[loss=0.1961, simple_loss=0.2954, pruned_loss=0.04837, over 10233.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3112, pruned_loss=0.07847, over 1697431.30 frames. ], batch size: 68, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:35:49,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.57 vs. limit=5.0 2023-05-16 23:36:00,270 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258654.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:36:02,920 INFO [finetune.py:992] (1/2) Epoch 13, batch 11850, loss[loss=0.2037, simple_loss=0.3007, pruned_loss=0.05331, over 11086.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3123, pruned_loss=0.07876, over 1677236.13 frames. ], batch size: 55, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:36:16,527 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.736e+02 3.563e+02 4.265e+02 5.101e+02 7.301e+02, threshold=8.530e+02, percent-clipped=0.0 2023-05-16 23:36:24,094 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0102, 4.3674, 3.8910, 4.6215, 4.1116, 2.5929, 3.9950, 2.9549], device='cuda:1'), covar=tensor([0.0893, 0.0837, 0.1477, 0.0549, 0.1367, 0.2078, 0.1172, 0.3886], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0360, 0.0342, 0.0299, 0.0354, 0.0262, 0.0331, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:36:37,164 INFO [finetune.py:992] (1/2) Epoch 13, batch 11900, loss[loss=0.2463, simple_loss=0.3175, pruned_loss=0.08756, over 7007.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3123, pruned_loss=0.07849, over 1655586.04 frames. ], batch size: 98, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:37:10,433 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8722, 3.8462, 3.8240, 3.9343, 3.7497, 3.7578, 3.6710, 3.8424], device='cuda:1'), covar=tensor([0.1144, 0.0695, 0.1406, 0.0660, 0.1599, 0.1160, 0.0552, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0644, 0.0567, 0.0580, 0.0777, 0.0687, 0.0516, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:37:11,601 INFO [finetune.py:992] (1/2) Epoch 13, batch 11950, loss[loss=0.23, simple_loss=0.3099, pruned_loss=0.07507, over 12140.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3085, pruned_loss=0.07528, over 1677154.15 frames. ], batch size: 34, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:37:26,344 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.223e+02 3.058e+02 3.451e+02 4.238e+02 1.115e+03, threshold=6.902e+02, percent-clipped=1.0 2023-05-16 23:37:36,284 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7861, 2.4039, 3.5703, 3.6056, 2.8829, 2.6630, 2.6366, 2.4302], device='cuda:1'), covar=tensor([0.1315, 0.2879, 0.0603, 0.0526, 0.0976, 0.2120, 0.2718, 0.3863], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0369, 0.0263, 0.0287, 0.0260, 0.0296, 0.0372, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:37:46,003 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0912, 2.2977, 2.3144, 2.2309, 2.0145, 1.8688, 2.2174, 1.7158], device='cuda:1'), covar=tensor([0.0274, 0.0202, 0.0189, 0.0213, 0.0316, 0.0272, 0.0177, 0.0389], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0155, 0.0156, 0.0181, 0.0191, 0.0192, 0.0163, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:37:47,176 INFO [finetune.py:992] (1/2) Epoch 13, batch 12000, loss[loss=0.1888, simple_loss=0.2877, pruned_loss=0.04502, over 10322.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.304, pruned_loss=0.0716, over 1684080.59 frames. ], batch size: 68, lr: 3.64e-03, grad_scale: 32.0 2023-05-16 23:37:47,176 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 23:38:03,652 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6440, 4.1373, 3.7123, 4.5131, 3.8657, 2.3975, 3.8174, 2.7076], device='cuda:1'), covar=tensor([0.1209, 0.1050, 0.1458, 0.0403, 0.1543, 0.2542, 0.1329, 0.4737], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0358, 0.0340, 0.0296, 0.0352, 0.0260, 0.0329, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:38:05,259 INFO [finetune.py:1026] (1/2) Epoch 13, validation: loss=0.2836, simple_loss=0.3609, pruned_loss=0.1031, over 1020973.00 frames. 2023-05-16 23:38:05,260 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 23:38:07,435 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258810.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:38:39,387 INFO [finetune.py:992] (1/2) Epoch 13, batch 12050, loss[loss=0.2062, simple_loss=0.2948, pruned_loss=0.05875, over 7466.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2994, pruned_loss=0.06828, over 1686324.83 frames. ], batch size: 97, lr: 3.64e-03, grad_scale: 32.0 2023-05-16 23:38:46,316 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258867.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:38:53,929 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 3.018e+02 3.545e+02 4.227e+02 7.592e+02, threshold=7.089e+02, percent-clipped=1.0 2023-05-16 23:38:55,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-16 23:39:12,977 INFO [finetune.py:992] (1/2) Epoch 13, batch 12100, loss[loss=0.2209, simple_loss=0.3103, pruned_loss=0.06579, over 11625.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.298, pruned_loss=0.0673, over 1676609.08 frames. ], batch size: 48, lr: 3.64e-03, grad_scale: 32.0 2023-05-16 23:39:29,565 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-16 23:39:45,267 INFO [finetune.py:992] (1/2) Epoch 13, batch 12150, loss[loss=0.2377, simple_loss=0.305, pruned_loss=0.08521, over 6882.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2989, pruned_loss=0.06748, over 1687070.68 frames. ], batch size: 99, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:39:49,860 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258965.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:39:58,683 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 3.395e+02 3.927e+02 4.540e+02 8.822e+02, threshold=7.853e+02, percent-clipped=2.0 2023-05-16 23:40:17,267 INFO [finetune.py:992] (1/2) Epoch 13, batch 12200, loss[loss=0.2279, simple_loss=0.3026, pruned_loss=0.07664, over 6635.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2995, pruned_loss=0.06804, over 1664329.13 frames. ], batch size: 99, lr: 3.64e-03, grad_scale: 16.0 2023-05-16 23:40:28,891 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259026.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:41:03,243 INFO [finetune.py:992] (1/2) Epoch 14, batch 0, loss[loss=0.2036, simple_loss=0.2835, pruned_loss=0.06186, over 12053.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2835, pruned_loss=0.06186, over 12053.00 frames. ], batch size: 37, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:41:03,244 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-16 23:41:18,034 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9061, 2.0696, 2.5169, 2.9462, 2.1657, 2.9888, 2.8548, 3.0327], device='cuda:1'), covar=tensor([0.0178, 0.1199, 0.0504, 0.0197, 0.1094, 0.0274, 0.0326, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0197, 0.0176, 0.0114, 0.0183, 0.0170, 0.0166, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:41:20,625 INFO [finetune.py:1026] (1/2) Epoch 14, validation: loss=0.2862, simple_loss=0.3615, pruned_loss=0.1055, over 1020973.00 frames. 2023-05-16 23:41:20,625 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-16 23:41:47,001 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.004e+02 2.906e+02 3.519e+02 4.209e+02 7.031e+02, threshold=7.039e+02, percent-clipped=0.0 2023-05-16 23:41:57,096 INFO [finetune.py:992] (1/2) Epoch 14, batch 50, loss[loss=0.1637, simple_loss=0.2549, pruned_loss=0.03623, over 12110.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2626, pruned_loss=0.04179, over 541665.11 frames. ], batch size: 33, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:42:10,632 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259110.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:42:33,356 INFO [finetune.py:992] (1/2) Epoch 14, batch 100, loss[loss=0.1526, simple_loss=0.2314, pruned_loss=0.0369, over 11326.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.261, pruned_loss=0.04118, over 951314.53 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:42:35,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 2023-05-16 23:42:44,741 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259158.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:42:51,167 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259167.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:42:59,573 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.838e+02 2.720e+02 3.079e+02 3.578e+02 6.303e+02, threshold=6.158e+02, percent-clipped=0.0 2023-05-16 23:43:08,604 INFO [finetune.py:992] (1/2) Epoch 14, batch 150, loss[loss=0.1582, simple_loss=0.2521, pruned_loss=0.03211, over 12305.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2637, pruned_loss=0.0418, over 1272102.56 frames. ], batch size: 33, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:43:25,192 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259215.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:43:31,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-16 23:43:33,231 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259226.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:43:39,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.63 vs. limit=5.0 2023-05-16 23:43:44,978 INFO [finetune.py:992] (1/2) Epoch 14, batch 200, loss[loss=0.1836, simple_loss=0.2711, pruned_loss=0.048, over 12354.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2615, pruned_loss=0.0411, over 1516421.20 frames. ], batch size: 36, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:44:01,339 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259264.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:44:11,575 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.966e+02 3.574e+02 4.503e+02 1.050e+03, threshold=7.149e+02, percent-clipped=8.0 2023-05-16 23:44:17,618 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259287.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:44:18,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-05-16 23:44:20,919 INFO [finetune.py:992] (1/2) Epoch 14, batch 250, loss[loss=0.1862, simple_loss=0.2672, pruned_loss=0.05262, over 12303.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2602, pruned_loss=0.041, over 1699181.41 frames. ], batch size: 34, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:44:41,511 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259321.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:44:44,436 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259325.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:44:56,105 INFO [finetune.py:992] (1/2) Epoch 14, batch 300, loss[loss=0.1488, simple_loss=0.2292, pruned_loss=0.03422, over 11994.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2609, pruned_loss=0.04119, over 1849388.66 frames. ], batch size: 28, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:45:22,644 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.710e+02 3.145e+02 3.769e+02 6.486e+02, threshold=6.290e+02, percent-clipped=0.0 2023-05-16 23:45:31,829 INFO [finetune.py:992] (1/2) Epoch 14, batch 350, loss[loss=0.1472, simple_loss=0.2409, pruned_loss=0.02676, over 12313.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2604, pruned_loss=0.04061, over 1964984.97 frames. ], batch size: 33, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:45:35,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-16 23:45:48,631 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259414.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:46:08,712 INFO [finetune.py:992] (1/2) Epoch 14, batch 400, loss[loss=0.1592, simple_loss=0.242, pruned_loss=0.03825, over 12288.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2596, pruned_loss=0.04047, over 2053391.23 frames. ], batch size: 33, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:46:32,186 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259475.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:46:34,825 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.661e+02 3.375e+02 3.946e+02 9.512e+02, threshold=6.750e+02, percent-clipped=4.0 2023-05-16 23:46:44,072 INFO [finetune.py:992] (1/2) Epoch 14, batch 450, loss[loss=0.1362, simple_loss=0.2237, pruned_loss=0.02439, over 12293.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2593, pruned_loss=0.04033, over 2120803.22 frames. ], batch size: 28, lr: 3.63e-03, grad_scale: 16.0 2023-05-16 23:47:20,175 INFO [finetune.py:992] (1/2) Epoch 14, batch 500, loss[loss=0.1736, simple_loss=0.2597, pruned_loss=0.04377, over 12087.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.259, pruned_loss=0.04037, over 2171936.27 frames. ], batch size: 32, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:47:47,789 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.764e+02 3.286e+02 4.293e+02 4.287e+03, threshold=6.572e+02, percent-clipped=4.0 2023-05-16 23:47:49,406 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259582.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:47:56,280 INFO [finetune.py:992] (1/2) Epoch 14, batch 550, loss[loss=0.1697, simple_loss=0.2622, pruned_loss=0.03856, over 12249.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2585, pruned_loss=0.03995, over 2219359.60 frames. ], batch size: 32, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:48:16,387 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259620.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:48:17,186 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259621.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:48:32,077 INFO [finetune.py:992] (1/2) Epoch 14, batch 600, loss[loss=0.1711, simple_loss=0.2657, pruned_loss=0.03826, over 12141.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2585, pruned_loss=0.03971, over 2249301.90 frames. ], batch size: 39, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:48:51,463 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259669.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:48:59,224 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.593e+02 3.062e+02 3.666e+02 6.023e+02, threshold=6.124e+02, percent-clipped=0.0 2023-05-16 23:49:07,551 INFO [finetune.py:992] (1/2) Epoch 14, batch 650, loss[loss=0.1478, simple_loss=0.2301, pruned_loss=0.03281, over 12281.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2575, pruned_loss=0.03927, over 2282592.09 frames. ], batch size: 28, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:49:43,308 INFO [finetune.py:992] (1/2) Epoch 14, batch 700, loss[loss=0.1407, simple_loss=0.2258, pruned_loss=0.02779, over 12367.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2577, pruned_loss=0.0391, over 2304968.93 frames. ], batch size: 30, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:49:47,609 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4504, 2.2775, 3.1201, 4.3647, 2.2148, 4.2851, 4.4605, 4.5078], device='cuda:1'), covar=tensor([0.0151, 0.1543, 0.0577, 0.0151, 0.1548, 0.0265, 0.0164, 0.0090], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0197, 0.0176, 0.0115, 0.0184, 0.0172, 0.0168, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:49:48,258 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259749.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:49:54,324 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.95 vs. limit=5.0 2023-05-16 23:50:03,065 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259770.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:50:10,191 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 2.727e+02 3.180e+02 3.776e+02 6.703e+02, threshold=6.360e+02, percent-clipped=3.0 2023-05-16 23:50:11,795 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259782.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:50:18,743 INFO [finetune.py:992] (1/2) Epoch 14, batch 750, loss[loss=0.1731, simple_loss=0.2684, pruned_loss=0.03885, over 12321.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2567, pruned_loss=0.03894, over 2313859.04 frames. ], batch size: 34, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:50:32,045 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259810.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:50:52,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-05-16 23:50:55,686 INFO [finetune.py:992] (1/2) Epoch 14, batch 800, loss[loss=0.1977, simple_loss=0.2899, pruned_loss=0.05271, over 12295.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2569, pruned_loss=0.03889, over 2327369.19 frames. ], batch size: 37, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:50:56,577 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259843.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:51:06,348 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4446, 5.0576, 5.4526, 4.6888, 5.1162, 4.8682, 5.4897, 5.1379], device='cuda:1'), covar=tensor([0.0241, 0.0384, 0.0250, 0.0271, 0.0353, 0.0293, 0.0187, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0260, 0.0283, 0.0259, 0.0259, 0.0259, 0.0234, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:51:12,745 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259866.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:51:14,144 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1669, 6.1316, 5.8943, 5.4329, 5.2190, 6.0663, 5.7053, 5.4054], device='cuda:1'), covar=tensor([0.0732, 0.1034, 0.0719, 0.1664, 0.0762, 0.0702, 0.1480, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0614, 0.0557, 0.0513, 0.0627, 0.0411, 0.0714, 0.0769, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-16 23:51:22,556 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.584e+02 2.997e+02 3.509e+02 7.772e+02, threshold=5.993e+02, percent-clipped=1.0 2023-05-16 23:51:24,146 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259882.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:51:31,072 INFO [finetune.py:992] (1/2) Epoch 14, batch 850, loss[loss=0.1784, simple_loss=0.2825, pruned_loss=0.03721, over 11829.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2572, pruned_loss=0.03909, over 2338888.44 frames. ], batch size: 44, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:51:37,046 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3119, 4.6224, 4.0195, 4.9963, 4.4728, 3.0397, 4.2463, 2.8882], device='cuda:1'), covar=tensor([0.0863, 0.0851, 0.1525, 0.0509, 0.1235, 0.1652, 0.1076, 0.3828], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0376, 0.0356, 0.0311, 0.0366, 0.0272, 0.0342, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:51:50,750 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259920.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:51:55,599 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259927.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:51:57,577 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259930.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:52:06,159 INFO [finetune.py:992] (1/2) Epoch 14, batch 900, loss[loss=0.1383, simple_loss=0.2196, pruned_loss=0.0285, over 11861.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2569, pruned_loss=0.03897, over 2352156.15 frames. ], batch size: 26, lr: 3.63e-03, grad_scale: 8.0 2023-05-16 23:52:25,062 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259968.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:52:33,417 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.713e+02 3.274e+02 3.740e+02 6.362e+02, threshold=6.548e+02, percent-clipped=1.0 2023-05-16 23:52:37,727 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1477, 4.9739, 5.0927, 5.1089, 4.8115, 4.8148, 4.5239, 5.0358], device='cuda:1'), covar=tensor([0.0673, 0.0606, 0.0788, 0.0599, 0.1820, 0.1507, 0.0566, 0.1005], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0677, 0.0596, 0.0610, 0.0816, 0.0723, 0.0542, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:52:41,906 INFO [finetune.py:992] (1/2) Epoch 14, batch 950, loss[loss=0.1464, simple_loss=0.2308, pruned_loss=0.03097, over 12137.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2569, pruned_loss=0.03901, over 2357558.23 frames. ], batch size: 30, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:52:46,983 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2513, 2.5727, 3.7494, 3.1133, 3.5734, 3.2313, 2.5737, 3.6484], device='cuda:1'), covar=tensor([0.0100, 0.0348, 0.0141, 0.0226, 0.0129, 0.0183, 0.0341, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0203, 0.0188, 0.0185, 0.0214, 0.0164, 0.0197, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:53:05,623 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-16 23:53:13,206 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260030.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:53:21,473 INFO [finetune.py:992] (1/2) Epoch 14, batch 1000, loss[loss=0.1514, simple_loss=0.2425, pruned_loss=0.03012, over 12339.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2574, pruned_loss=0.03887, over 2366516.38 frames. ], batch size: 35, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:53:40,890 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4024, 2.8413, 3.8513, 3.2922, 3.7587, 3.3443, 2.7527, 3.7916], device='cuda:1'), covar=tensor([0.0137, 0.0330, 0.0146, 0.0255, 0.0138, 0.0187, 0.0358, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0203, 0.0188, 0.0185, 0.0214, 0.0164, 0.0197, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:53:41,488 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260070.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:53:48,608 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.539e+02 3.055e+02 3.631e+02 6.681e+02, threshold=6.111e+02, percent-clipped=1.0 2023-05-16 23:53:50,477 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-05-16 23:53:54,686 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1213, 3.0921, 4.4248, 2.4825, 2.6856, 3.3679, 3.0278, 3.4188], device='cuda:1'), covar=tensor([0.0555, 0.1203, 0.0450, 0.1245, 0.1892, 0.1541, 0.1370, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0234, 0.0248, 0.0183, 0.0235, 0.0289, 0.0224, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:53:56,658 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260091.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:53:57,146 INFO [finetune.py:992] (1/2) Epoch 14, batch 1050, loss[loss=0.1686, simple_loss=0.2572, pruned_loss=0.03997, over 12339.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2561, pruned_loss=0.03819, over 2370668.26 frames. ], batch size: 31, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:53:57,347 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3290, 3.5489, 3.3076, 3.2083, 2.7640, 2.6604, 3.6742, 2.1972], device='cuda:1'), covar=tensor([0.0433, 0.0166, 0.0194, 0.0183, 0.0449, 0.0376, 0.0141, 0.0547], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0161, 0.0162, 0.0186, 0.0199, 0.0198, 0.0169, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:54:06,275 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260105.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:54:15,380 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260118.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:54:20,533 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3482, 5.1245, 5.3160, 5.3175, 4.9164, 4.9981, 4.7082, 5.2208], device='cuda:1'), covar=tensor([0.0714, 0.0682, 0.0851, 0.0617, 0.2023, 0.1523, 0.0636, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0684, 0.0602, 0.0617, 0.0824, 0.0730, 0.0547, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-16 23:54:31,331 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260138.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:54:33,971 INFO [finetune.py:992] (1/2) Epoch 14, batch 1100, loss[loss=0.1644, simple_loss=0.2497, pruned_loss=0.0396, over 12262.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2551, pruned_loss=0.03811, over 2376829.34 frames. ], batch size: 32, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:55:01,283 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.617e+02 3.016e+02 3.739e+02 7.886e+02, threshold=6.033e+02, percent-clipped=2.0 2023-05-16 23:55:07,111 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2428, 4.1848, 4.1287, 4.4364, 3.0070, 3.9616, 2.8215, 4.1403], device='cuda:1'), covar=tensor([0.1708, 0.0662, 0.0855, 0.0609, 0.1197, 0.0605, 0.1699, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0264, 0.0295, 0.0358, 0.0242, 0.0241, 0.0262, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:55:09,765 INFO [finetune.py:992] (1/2) Epoch 14, batch 1150, loss[loss=0.1814, simple_loss=0.2858, pruned_loss=0.03853, over 11536.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.255, pruned_loss=0.03802, over 2380032.30 frames. ], batch size: 48, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:55:15,053 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1037, 2.4671, 3.5599, 3.0619, 3.4063, 3.1265, 2.4742, 3.5374], device='cuda:1'), covar=tensor([0.0150, 0.0384, 0.0170, 0.0255, 0.0153, 0.0186, 0.0362, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0203, 0.0188, 0.0185, 0.0214, 0.0163, 0.0197, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:55:31,797 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260222.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:55:33,370 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260224.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:55:45,481 INFO [finetune.py:992] (1/2) Epoch 14, batch 1200, loss[loss=0.2254, simple_loss=0.3094, pruned_loss=0.07074, over 12384.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2553, pruned_loss=0.03848, over 2377035.11 frames. ], batch size: 38, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:56:02,077 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1320, 2.4146, 3.6583, 3.1026, 3.4943, 3.1644, 2.4936, 3.5584], device='cuda:1'), covar=tensor([0.0133, 0.0373, 0.0168, 0.0244, 0.0132, 0.0175, 0.0366, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0203, 0.0187, 0.0184, 0.0214, 0.0163, 0.0197, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:56:12,510 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.690e+02 3.229e+02 3.657e+02 1.170e+03, threshold=6.458e+02, percent-clipped=4.0 2023-05-16 23:56:17,481 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260285.0, num_to_drop=1, layers_to_drop={2} 2023-05-16 23:56:22,329 INFO [finetune.py:992] (1/2) Epoch 14, batch 1250, loss[loss=0.1667, simple_loss=0.258, pruned_loss=0.03769, over 12142.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2551, pruned_loss=0.03855, over 2370906.78 frames. ], batch size: 34, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:56:50,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.60 vs. limit=5.0 2023-05-16 23:56:57,484 INFO [finetune.py:992] (1/2) Epoch 14, batch 1300, loss[loss=0.1976, simple_loss=0.2967, pruned_loss=0.04929, over 11382.00 frames. ], tot_loss[loss=0.166, simple_loss=0.255, pruned_loss=0.0385, over 2378389.06 frames. ], batch size: 55, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:57:24,492 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 2.744e+02 3.203e+02 3.844e+02 2.936e+03, threshold=6.405e+02, percent-clipped=4.0 2023-05-16 23:57:28,889 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260386.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:57:29,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6679, 3.1594, 3.6003, 4.5910, 4.0840, 4.7437, 4.0311, 3.5856], device='cuda:1'), covar=tensor([0.0045, 0.0330, 0.0170, 0.0042, 0.0100, 0.0058, 0.0118, 0.0285], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0124, 0.0105, 0.0079, 0.0104, 0.0116, 0.0097, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:57:33,188 INFO [finetune.py:992] (1/2) Epoch 14, batch 1350, loss[loss=0.1618, simple_loss=0.2547, pruned_loss=0.03448, over 12060.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2542, pruned_loss=0.03819, over 2378819.37 frames. ], batch size: 40, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:57:42,858 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260405.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:58:08,115 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260438.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:58:10,852 INFO [finetune.py:992] (1/2) Epoch 14, batch 1400, loss[loss=0.1777, simple_loss=0.2745, pruned_loss=0.04046, over 12271.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2538, pruned_loss=0.03788, over 2383713.98 frames. ], batch size: 37, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:58:18,572 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260453.0, num_to_drop=1, layers_to_drop={1} 2023-05-16 23:58:38,137 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.620e+02 3.097e+02 3.656e+02 5.501e+02, threshold=6.194e+02, percent-clipped=0.0 2023-05-16 23:58:42,433 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260486.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:58:46,524 INFO [finetune.py:992] (1/2) Epoch 14, batch 1450, loss[loss=0.1459, simple_loss=0.2243, pruned_loss=0.03369, over 12192.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2541, pruned_loss=0.03773, over 2392154.45 frames. ], batch size: 29, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:58:47,368 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6188, 2.8982, 3.5954, 4.5143, 3.9860, 4.5931, 3.8807, 3.3129], device='cuda:1'), covar=tensor([0.0038, 0.0370, 0.0165, 0.0041, 0.0130, 0.0069, 0.0141, 0.0367], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0126, 0.0106, 0.0079, 0.0105, 0.0117, 0.0099, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-16 23:58:56,764 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6877, 2.7173, 3.4329, 4.6109, 2.6269, 4.4809, 4.6543, 4.7159], device='cuda:1'), covar=tensor([0.0134, 0.1275, 0.0476, 0.0137, 0.1299, 0.0261, 0.0142, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0204, 0.0181, 0.0118, 0.0189, 0.0178, 0.0173, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:58:57,583 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6966, 2.7309, 4.4923, 4.5042, 2.7971, 2.5739, 2.8675, 2.1654], device='cuda:1'), covar=tensor([0.1678, 0.3292, 0.0478, 0.0477, 0.1395, 0.2570, 0.3051, 0.4228], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0385, 0.0271, 0.0299, 0.0271, 0.0309, 0.0386, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-16 23:59:08,120 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260522.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:59:22,354 INFO [finetune.py:992] (1/2) Epoch 14, batch 1500, loss[loss=0.1696, simple_loss=0.2673, pruned_loss=0.03593, over 11618.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2542, pruned_loss=0.03768, over 2394110.79 frames. ], batch size: 48, lr: 3.62e-03, grad_scale: 8.0 2023-05-16 23:59:42,610 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260570.0, num_to_drop=0, layers_to_drop=set() 2023-05-16 23:59:49,693 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.561e+02 3.079e+02 3.709e+02 5.375e+02, threshold=6.157e+02, percent-clipped=0.0 2023-05-16 23:59:49,802 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260580.0, num_to_drop=1, layers_to_drop={0} 2023-05-16 23:59:59,539 INFO [finetune.py:992] (1/2) Epoch 14, batch 1550, loss[loss=0.1496, simple_loss=0.2322, pruned_loss=0.03346, over 12024.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2539, pruned_loss=0.03788, over 2377132.89 frames. ], batch size: 31, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:00:34,929 INFO [finetune.py:992] (1/2) Epoch 14, batch 1600, loss[loss=0.1493, simple_loss=0.2235, pruned_loss=0.03753, over 12302.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2543, pruned_loss=0.03804, over 2379967.65 frames. ], batch size: 28, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:01:01,894 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.613e+02 3.003e+02 3.667e+02 8.089e+02, threshold=6.007e+02, percent-clipped=3.0 2023-05-17 00:01:06,384 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260686.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:01:10,563 INFO [finetune.py:992] (1/2) Epoch 14, batch 1650, loss[loss=0.1585, simple_loss=0.2452, pruned_loss=0.03591, over 12294.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2544, pruned_loss=0.03787, over 2382842.66 frames. ], batch size: 34, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:01:12,443 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.52 vs. limit=5.0 2023-05-17 00:01:17,057 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4407, 2.4950, 3.2376, 4.3642, 2.2838, 4.2939, 4.4157, 4.4318], device='cuda:1'), covar=tensor([0.0119, 0.1313, 0.0508, 0.0126, 0.1429, 0.0262, 0.0135, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0181, 0.0118, 0.0189, 0.0177, 0.0173, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:01:37,933 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260730.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:01:40,728 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260734.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:01:46,408 INFO [finetune.py:992] (1/2) Epoch 14, batch 1700, loss[loss=0.2488, simple_loss=0.3257, pruned_loss=0.08597, over 7973.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2541, pruned_loss=0.03787, over 2379191.83 frames. ], batch size: 98, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:01:54,241 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260751.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:02:14,603 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.628e+02 3.189e+02 3.651e+02 5.680e+02, threshold=6.378e+02, percent-clipped=0.0 2023-05-17 00:02:22,683 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260791.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:02:23,151 INFO [finetune.py:992] (1/2) Epoch 14, batch 1750, loss[loss=0.1721, simple_loss=0.2593, pruned_loss=0.04242, over 12275.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2552, pruned_loss=0.03823, over 2372325.73 frames. ], batch size: 33, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:02:33,578 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8289, 2.9379, 4.8318, 4.9902, 3.1287, 2.8224, 3.0039, 2.1489], device='cuda:1'), covar=tensor([0.1520, 0.3008, 0.0423, 0.0378, 0.1157, 0.2297, 0.2720, 0.4145], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0385, 0.0272, 0.0300, 0.0271, 0.0308, 0.0386, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:02:37,872 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260812.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:02:58,989 INFO [finetune.py:992] (1/2) Epoch 14, batch 1800, loss[loss=0.1463, simple_loss=0.2292, pruned_loss=0.03171, over 12245.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2549, pruned_loss=0.03823, over 2379728.65 frames. ], batch size: 32, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:03:25,986 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 2.814e+02 3.306e+02 4.275e+02 7.612e+02, threshold=6.612e+02, percent-clipped=3.0 2023-05-17 00:03:26,127 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260880.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 00:03:26,794 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260881.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:03:35,660 INFO [finetune.py:992] (1/2) Epoch 14, batch 1850, loss[loss=0.1656, simple_loss=0.2563, pruned_loss=0.0375, over 12298.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2554, pruned_loss=0.03835, over 2378450.69 frames. ], batch size: 34, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:04:01,744 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260928.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 00:04:11,575 INFO [finetune.py:992] (1/2) Epoch 14, batch 1900, loss[loss=0.1684, simple_loss=0.2586, pruned_loss=0.03913, over 12142.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2551, pruned_loss=0.03816, over 2377644.05 frames. ], batch size: 34, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:04:11,751 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260942.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:04:16,773 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3167, 4.7205, 4.0699, 4.9279, 4.4772, 2.8664, 4.1767, 3.0290], device='cuda:1'), covar=tensor([0.0819, 0.0673, 0.1435, 0.0451, 0.1117, 0.1788, 0.1186, 0.3303], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0379, 0.0360, 0.0315, 0.0368, 0.0272, 0.0343, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:04:38,628 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.025e+02 2.640e+02 3.082e+02 3.439e+02 6.112e+02, threshold=6.164e+02, percent-clipped=0.0 2023-05-17 00:04:47,261 INFO [finetune.py:992] (1/2) Epoch 14, batch 1950, loss[loss=0.1901, simple_loss=0.2767, pruned_loss=0.05174, over 11606.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2553, pruned_loss=0.03797, over 2380339.67 frames. ], batch size: 48, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:05:08,619 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1011, 4.6892, 5.0973, 4.4733, 4.7878, 4.5926, 5.1035, 4.7781], device='cuda:1'), covar=tensor([0.0281, 0.0463, 0.0295, 0.0275, 0.0410, 0.0347, 0.0223, 0.0368], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0268, 0.0290, 0.0267, 0.0267, 0.0267, 0.0241, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:05:23,745 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4894, 4.7656, 4.2194, 5.0891, 4.6783, 3.1019, 4.5030, 3.2668], device='cuda:1'), covar=tensor([0.0744, 0.0842, 0.1530, 0.0557, 0.1117, 0.1582, 0.0985, 0.3254], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0383, 0.0364, 0.0318, 0.0372, 0.0274, 0.0346, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:05:24,380 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5757, 2.3437, 3.4183, 4.5642, 2.3031, 4.5851, 4.6218, 4.7135], device='cuda:1'), covar=tensor([0.0173, 0.1315, 0.0410, 0.0139, 0.1412, 0.0166, 0.0120, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0205, 0.0183, 0.0120, 0.0191, 0.0179, 0.0175, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:05:24,883 INFO [finetune.py:992] (1/2) Epoch 14, batch 2000, loss[loss=0.1479, simple_loss=0.2283, pruned_loss=0.03379, over 12324.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2556, pruned_loss=0.03805, over 2365994.42 frames. ], batch size: 28, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:05:52,005 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.570e+02 2.979e+02 3.561e+02 5.774e+02, threshold=5.957e+02, percent-clipped=0.0 2023-05-17 00:05:56,492 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261086.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:06:00,466 INFO [finetune.py:992] (1/2) Epoch 14, batch 2050, loss[loss=0.1687, simple_loss=0.2473, pruned_loss=0.04505, over 12038.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2559, pruned_loss=0.03803, over 2368818.00 frames. ], batch size: 28, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:06:11,441 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261107.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:06:31,841 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261136.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:06:35,850 INFO [finetune.py:992] (1/2) Epoch 14, batch 2100, loss[loss=0.1752, simple_loss=0.2622, pruned_loss=0.0441, over 11626.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2565, pruned_loss=0.03804, over 2372259.82 frames. ], batch size: 48, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:07:04,084 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.170e+02 2.873e+02 3.381e+02 4.160e+02 8.406e+02, threshold=6.762e+02, percent-clipped=4.0 2023-05-17 00:07:12,699 INFO [finetune.py:992] (1/2) Epoch 14, batch 2150, loss[loss=0.1686, simple_loss=0.2614, pruned_loss=0.03791, over 12328.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2565, pruned_loss=0.03822, over 2375965.70 frames. ], batch size: 34, lr: 3.62e-03, grad_scale: 8.0 2023-05-17 00:07:16,514 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261197.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:07:44,986 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261237.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:07:48,475 INFO [finetune.py:992] (1/2) Epoch 14, batch 2200, loss[loss=0.1805, simple_loss=0.2687, pruned_loss=0.04611, over 12029.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2569, pruned_loss=0.03839, over 2374520.01 frames. ], batch size: 40, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:08:15,594 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.577e+02 3.065e+02 3.609e+02 7.867e+02, threshold=6.130e+02, percent-clipped=1.0 2023-05-17 00:08:24,237 INFO [finetune.py:992] (1/2) Epoch 14, batch 2250, loss[loss=0.147, simple_loss=0.2337, pruned_loss=0.0302, over 11758.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2566, pruned_loss=0.03843, over 2374004.55 frames. ], batch size: 26, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:09:01,047 INFO [finetune.py:992] (1/2) Epoch 14, batch 2300, loss[loss=0.1819, simple_loss=0.2714, pruned_loss=0.04613, over 11293.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2563, pruned_loss=0.03834, over 2375673.94 frames. ], batch size: 55, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:09:24,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-17 00:09:28,110 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.585e+02 3.048e+02 3.546e+02 8.125e+02, threshold=6.097e+02, percent-clipped=1.0 2023-05-17 00:09:32,682 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261386.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:09:36,716 INFO [finetune.py:992] (1/2) Epoch 14, batch 2350, loss[loss=0.157, simple_loss=0.246, pruned_loss=0.03402, over 12130.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2557, pruned_loss=0.03819, over 2374692.37 frames. ], batch size: 39, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:09:39,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9575, 4.8230, 4.6856, 4.7726, 4.3799, 4.8315, 4.9024, 5.0944], device='cuda:1'), covar=tensor([0.0388, 0.0180, 0.0249, 0.0425, 0.0873, 0.0376, 0.0179, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0200, 0.0193, 0.0247, 0.0244, 0.0223, 0.0178, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 00:09:42,992 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7331, 3.2135, 5.0853, 2.7840, 2.9373, 3.9944, 3.2232, 3.7504], device='cuda:1'), covar=tensor([0.0510, 0.1262, 0.0396, 0.1172, 0.1923, 0.1439, 0.1417, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0237, 0.0254, 0.0184, 0.0239, 0.0294, 0.0227, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:09:47,834 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261407.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:09:48,641 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9072, 3.3183, 5.2778, 2.8143, 3.0563, 4.0872, 3.3444, 3.8094], device='cuda:1'), covar=tensor([0.0432, 0.1207, 0.0333, 0.1062, 0.1742, 0.1268, 0.1289, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0237, 0.0254, 0.0184, 0.0239, 0.0295, 0.0227, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:10:06,848 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261434.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:10:12,301 INFO [finetune.py:992] (1/2) Epoch 14, batch 2400, loss[loss=0.1831, simple_loss=0.2749, pruned_loss=0.04571, over 11803.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2565, pruned_loss=0.03852, over 2384582.00 frames. ], batch size: 44, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:10:17,704 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4371, 4.8213, 3.0933, 2.7216, 4.1480, 2.4729, 3.9453, 3.3914], device='cuda:1'), covar=tensor([0.0703, 0.0481, 0.1123, 0.1534, 0.0261, 0.1482, 0.0503, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0261, 0.0180, 0.0205, 0.0145, 0.0186, 0.0201, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:10:20,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-17 00:10:21,902 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261455.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:10:22,742 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261456.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:10:40,723 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.793e+02 3.197e+02 3.922e+02 8.062e+02, threshold=6.395e+02, percent-clipped=2.0 2023-05-17 00:10:44,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-17 00:10:49,443 INFO [finetune.py:992] (1/2) Epoch 14, batch 2450, loss[loss=0.1521, simple_loss=0.242, pruned_loss=0.03105, over 12143.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2566, pruned_loss=0.03852, over 2385499.46 frames. ], batch size: 30, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:10:49,520 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261492.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 00:10:56,653 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3235, 5.1106, 5.2690, 5.3189, 4.9181, 5.0078, 4.7404, 5.2372], device='cuda:1'), covar=tensor([0.0690, 0.0704, 0.0898, 0.0589, 0.2102, 0.1475, 0.0582, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0702, 0.0624, 0.0630, 0.0850, 0.0748, 0.0559, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:11:07,602 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261517.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:11:21,802 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261537.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:11:25,237 INFO [finetune.py:992] (1/2) Epoch 14, batch 2500, loss[loss=0.1871, simple_loss=0.2756, pruned_loss=0.04924, over 12112.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2565, pruned_loss=0.03821, over 2386823.42 frames. ], batch size: 39, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:11:52,005 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.621e+02 3.073e+02 3.822e+02 6.755e+02, threshold=6.147e+02, percent-clipped=2.0 2023-05-17 00:11:55,607 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261585.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:12:00,713 INFO [finetune.py:992] (1/2) Epoch 14, batch 2550, loss[loss=0.1723, simple_loss=0.2656, pruned_loss=0.03949, over 12159.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2555, pruned_loss=0.03807, over 2390784.82 frames. ], batch size: 34, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:12:07,004 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4659, 4.7484, 4.1331, 4.9349, 4.5580, 2.9833, 4.2983, 3.1353], device='cuda:1'), covar=tensor([0.0722, 0.0745, 0.1384, 0.0486, 0.1175, 0.1581, 0.0984, 0.3246], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0379, 0.0358, 0.0317, 0.0367, 0.0271, 0.0342, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:12:37,788 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-05-17 00:12:37,881 INFO [finetune.py:992] (1/2) Epoch 14, batch 2600, loss[loss=0.162, simple_loss=0.2575, pruned_loss=0.03323, over 10636.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2547, pruned_loss=0.03752, over 2386693.95 frames. ], batch size: 69, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:13:05,194 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.688e+02 3.200e+02 3.722e+02 1.106e+03, threshold=6.400e+02, percent-clipped=4.0 2023-05-17 00:13:11,756 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3930, 3.6476, 3.2352, 3.1143, 2.7640, 2.6588, 3.6931, 2.2582], device='cuda:1'), covar=tensor([0.0421, 0.0143, 0.0235, 0.0228, 0.0420, 0.0411, 0.0124, 0.0523], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0162, 0.0167, 0.0187, 0.0201, 0.0199, 0.0171, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:13:13,655 INFO [finetune.py:992] (1/2) Epoch 14, batch 2650, loss[loss=0.2044, simple_loss=0.2831, pruned_loss=0.06292, over 7728.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2554, pruned_loss=0.03805, over 2372476.24 frames. ], batch size: 97, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:13:49,368 INFO [finetune.py:992] (1/2) Epoch 14, batch 2700, loss[loss=0.1542, simple_loss=0.2521, pruned_loss=0.02816, over 12345.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2555, pruned_loss=0.03816, over 2369358.18 frames. ], batch size: 36, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:14:04,881 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9666, 5.8501, 5.4190, 5.3911, 5.9534, 5.1973, 5.4549, 5.4422], device='cuda:1'), covar=tensor([0.1517, 0.0974, 0.1118, 0.1874, 0.1002, 0.2091, 0.1949, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0498, 0.0401, 0.0452, 0.0472, 0.0437, 0.0397, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:14:17,596 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.851e+02 2.853e+02 3.281e+02 3.725e+02 8.071e+02, threshold=6.562e+02, percent-clipped=2.0 2023-05-17 00:14:26,026 INFO [finetune.py:992] (1/2) Epoch 14, batch 2750, loss[loss=0.1757, simple_loss=0.2706, pruned_loss=0.04046, over 11827.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.256, pruned_loss=0.0386, over 2366249.44 frames. ], batch size: 44, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:14:26,188 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261792.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:14:40,947 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261812.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:14:49,675 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261824.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:14:56,216 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261833.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:15:01,232 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261840.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:15:02,517 INFO [finetune.py:992] (1/2) Epoch 14, batch 2800, loss[loss=0.1624, simple_loss=0.2568, pruned_loss=0.03401, over 12261.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.255, pruned_loss=0.03836, over 2355301.23 frames. ], batch size: 37, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:15:11,208 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2897, 6.1927, 5.7737, 5.7636, 6.2514, 5.5782, 5.7044, 5.7614], device='cuda:1'), covar=tensor([0.1592, 0.1014, 0.1190, 0.2013, 0.0954, 0.2215, 0.1930, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0491, 0.0397, 0.0447, 0.0467, 0.0431, 0.0393, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:15:29,553 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 2.566e+02 2.904e+02 3.440e+02 8.546e+02, threshold=5.808e+02, percent-clipped=1.0 2023-05-17 00:15:33,402 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261885.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:15:38,253 INFO [finetune.py:992] (1/2) Epoch 14, batch 2850, loss[loss=0.1856, simple_loss=0.2841, pruned_loss=0.04352, over 11765.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2547, pruned_loss=0.03827, over 2351711.57 frames. ], batch size: 44, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:15:40,572 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261894.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:15:46,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.12 vs. limit=5.0 2023-05-17 00:15:55,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-17 00:16:02,878 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.68 vs. limit=5.0 2023-05-17 00:16:07,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-05-17 00:16:15,242 INFO [finetune.py:992] (1/2) Epoch 14, batch 2900, loss[loss=0.1464, simple_loss=0.2307, pruned_loss=0.03106, over 12020.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2554, pruned_loss=0.03862, over 2352150.74 frames. ], batch size: 31, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:16:40,602 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261977.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:16:42,506 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.710e+02 3.195e+02 4.108e+02 8.461e+02, threshold=6.390e+02, percent-clipped=1.0 2023-05-17 00:16:50,991 INFO [finetune.py:992] (1/2) Epoch 14, batch 2950, loss[loss=0.2014, simple_loss=0.2811, pruned_loss=0.06089, over 8075.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2546, pruned_loss=0.0382, over 2351508.95 frames. ], batch size: 98, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:17:27,987 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262038.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:17:30,631 INFO [finetune.py:992] (1/2) Epoch 14, batch 3000, loss[loss=0.1582, simple_loss=0.2507, pruned_loss=0.03283, over 12078.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2545, pruned_loss=0.0382, over 2351268.15 frames. ], batch size: 42, lr: 3.61e-03, grad_scale: 16.0 2023-05-17 00:17:30,632 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 00:17:36,709 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9004, 5.3234, 5.1217, 4.8328, 5.4330, 4.8094, 4.6160, 4.8912], device='cuda:1'), covar=tensor([0.1203, 0.0929, 0.1348, 0.1853, 0.0847, 0.2029, 0.2344, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0493, 0.0398, 0.0451, 0.0465, 0.0434, 0.0395, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:17:50,069 INFO [finetune.py:1026] (1/2) Epoch 14, validation: loss=0.315, simple_loss=0.3917, pruned_loss=0.1191, over 1020973.00 frames. 2023-05-17 00:17:50,069 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 00:17:52,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-05-17 00:17:56,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8552, 3.5480, 3.5918, 3.7481, 3.7165, 3.7137, 3.7023, 2.6173], device='cuda:1'), covar=tensor([0.0097, 0.0100, 0.0129, 0.0087, 0.0070, 0.0126, 0.0096, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0074, 0.0062, 0.0094, 0.0083, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:18:17,814 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.674e+02 3.142e+02 3.828e+02 1.600e+03, threshold=6.284e+02, percent-clipped=3.0 2023-05-17 00:18:22,228 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2713, 6.2780, 6.0310, 5.5833, 5.2506, 6.1947, 5.8182, 5.4795], device='cuda:1'), covar=tensor([0.0617, 0.0820, 0.0647, 0.1684, 0.0703, 0.0701, 0.1429, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0626, 0.0565, 0.0521, 0.0647, 0.0423, 0.0736, 0.0796, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:18:25,573 INFO [finetune.py:992] (1/2) Epoch 14, batch 3050, loss[loss=0.1727, simple_loss=0.2601, pruned_loss=0.04259, over 12184.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2555, pruned_loss=0.03883, over 2357743.31 frames. ], batch size: 31, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:18:25,730 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2929, 5.1248, 5.2262, 5.3078, 4.9041, 5.0040, 4.7091, 5.2084], device='cuda:1'), covar=tensor([0.0823, 0.0595, 0.0963, 0.0616, 0.2047, 0.1336, 0.0596, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0702, 0.0628, 0.0634, 0.0854, 0.0750, 0.0562, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:18:26,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-17 00:18:39,621 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262112.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:19:01,602 INFO [finetune.py:992] (1/2) Epoch 14, batch 3100, loss[loss=0.1435, simple_loss=0.2288, pruned_loss=0.02906, over 11801.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2553, pruned_loss=0.03875, over 2359673.16 frames. ], batch size: 26, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:19:14,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.06 vs. limit=5.0 2023-05-17 00:19:15,136 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262160.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:19:29,388 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262180.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:19:30,006 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.766e+02 3.239e+02 4.149e+02 8.563e+02, threshold=6.478e+02, percent-clipped=3.0 2023-05-17 00:19:35,797 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262189.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:19:37,779 INFO [finetune.py:992] (1/2) Epoch 14, batch 3150, loss[loss=0.1572, simple_loss=0.2538, pruned_loss=0.03033, over 12268.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.255, pruned_loss=0.0386, over 2365605.48 frames. ], batch size: 37, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:20:13,803 INFO [finetune.py:992] (1/2) Epoch 14, batch 3200, loss[loss=0.1493, simple_loss=0.2328, pruned_loss=0.03292, over 12371.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2552, pruned_loss=0.03853, over 2372159.96 frames. ], batch size: 30, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:20:14,715 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3915, 4.9261, 5.3538, 4.6700, 4.9577, 4.7422, 5.3804, 5.0061], device='cuda:1'), covar=tensor([0.0261, 0.0401, 0.0271, 0.0257, 0.0396, 0.0392, 0.0183, 0.0312], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0271, 0.0295, 0.0271, 0.0270, 0.0273, 0.0245, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:20:29,818 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1685, 6.0699, 5.9166, 5.3378, 5.2395, 6.0738, 5.6539, 5.4025], device='cuda:1'), covar=tensor([0.0636, 0.1107, 0.0653, 0.1935, 0.0665, 0.0749, 0.1571, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0569, 0.0523, 0.0651, 0.0425, 0.0740, 0.0799, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:20:37,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 00:20:41,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2023-05-17 00:20:42,291 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.540e+02 3.178e+02 3.757e+02 7.638e+02, threshold=6.357e+02, percent-clipped=1.0 2023-05-17 00:20:50,198 INFO [finetune.py:992] (1/2) Epoch 14, batch 3250, loss[loss=0.1837, simple_loss=0.2748, pruned_loss=0.04628, over 12377.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2551, pruned_loss=0.03854, over 2374865.62 frames. ], batch size: 38, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:21:17,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-17 00:21:20,113 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262333.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:21:26,391 INFO [finetune.py:992] (1/2) Epoch 14, batch 3300, loss[loss=0.1449, simple_loss=0.2268, pruned_loss=0.03151, over 11800.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2553, pruned_loss=0.03888, over 2367920.98 frames. ], batch size: 26, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:21:53,887 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 2.682e+02 3.147e+02 3.843e+02 7.432e+02, threshold=6.293e+02, percent-clipped=1.0 2023-05-17 00:22:01,883 INFO [finetune.py:992] (1/2) Epoch 14, batch 3350, loss[loss=0.1552, simple_loss=0.2507, pruned_loss=0.02982, over 10457.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2551, pruned_loss=0.03843, over 2370474.17 frames. ], batch size: 68, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:22:07,090 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5821, 5.3449, 5.4709, 5.5279, 5.1201, 5.1867, 4.9832, 5.4676], device='cuda:1'), covar=tensor([0.0576, 0.0604, 0.0733, 0.0531, 0.1897, 0.1316, 0.0556, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0698, 0.0622, 0.0631, 0.0848, 0.0746, 0.0559, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:22:31,015 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7194, 2.5556, 4.7153, 5.0361, 3.2191, 2.5424, 3.0068, 2.0882], device='cuda:1'), covar=tensor([0.1758, 0.3790, 0.0435, 0.0358, 0.1119, 0.2846, 0.3010, 0.5053], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0389, 0.0276, 0.0303, 0.0274, 0.0311, 0.0388, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:22:39,356 INFO [finetune.py:992] (1/2) Epoch 14, batch 3400, loss[loss=0.1705, simple_loss=0.2618, pruned_loss=0.03963, over 10714.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2541, pruned_loss=0.03792, over 2371847.57 frames. ], batch size: 68, lr: 3.61e-03, grad_scale: 8.0 2023-05-17 00:22:57,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-05-17 00:23:06,288 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262480.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:23:06,788 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.107e+02 2.615e+02 3.016e+02 3.667e+02 7.524e+02, threshold=6.033e+02, percent-clipped=3.0 2023-05-17 00:23:12,422 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262489.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:23:14,529 INFO [finetune.py:992] (1/2) Epoch 14, batch 3450, loss[loss=0.1634, simple_loss=0.2559, pruned_loss=0.03543, over 12176.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2542, pruned_loss=0.03795, over 2374029.52 frames. ], batch size: 31, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:23:40,175 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262528.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:23:40,377 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7508, 4.1092, 3.5957, 4.2480, 3.8517, 2.7772, 3.7161, 2.8920], device='cuda:1'), covar=tensor([0.0947, 0.0854, 0.1457, 0.0649, 0.1307, 0.1666, 0.1161, 0.3293], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0383, 0.0361, 0.0321, 0.0372, 0.0274, 0.0345, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:23:46,695 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262537.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:23:50,305 INFO [finetune.py:992] (1/2) Epoch 14, batch 3500, loss[loss=0.1807, simple_loss=0.2754, pruned_loss=0.04299, over 11562.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2543, pruned_loss=0.03788, over 2374605.04 frames. ], batch size: 48, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:24:08,114 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3824, 2.5027, 3.1755, 4.2814, 2.3884, 4.3647, 4.3453, 4.3714], device='cuda:1'), covar=tensor([0.0133, 0.1240, 0.0484, 0.0160, 0.1264, 0.0227, 0.0146, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0202, 0.0182, 0.0120, 0.0190, 0.0179, 0.0175, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:24:19,094 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.813e+02 3.162e+02 3.676e+02 6.941e+02, threshold=6.323e+02, percent-clipped=3.0 2023-05-17 00:24:22,996 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0717, 2.4004, 3.7647, 3.0929, 3.4696, 3.1088, 2.6184, 3.5189], device='cuda:1'), covar=tensor([0.0150, 0.0368, 0.0156, 0.0231, 0.0178, 0.0208, 0.0344, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0206, 0.0192, 0.0190, 0.0221, 0.0169, 0.0202, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:24:27,029 INFO [finetune.py:992] (1/2) Epoch 14, batch 3550, loss[loss=0.1615, simple_loss=0.2467, pruned_loss=0.03813, over 12293.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2552, pruned_loss=0.03826, over 2362119.00 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:24:56,455 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262633.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:25:02,391 INFO [finetune.py:992] (1/2) Epoch 14, batch 3600, loss[loss=0.1687, simple_loss=0.2636, pruned_loss=0.03688, over 12193.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2543, pruned_loss=0.03793, over 2365573.81 frames. ], batch size: 35, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:25:13,341 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5060, 5.0720, 5.4494, 4.7558, 5.0881, 4.8296, 5.4971, 5.0664], device='cuda:1'), covar=tensor([0.0252, 0.0388, 0.0276, 0.0271, 0.0409, 0.0388, 0.0207, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0269, 0.0295, 0.0270, 0.0270, 0.0272, 0.0244, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:25:14,127 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1567, 4.7930, 5.0258, 5.0002, 4.8177, 4.9174, 4.9260, 2.8126], device='cuda:1'), covar=tensor([0.0110, 0.0077, 0.0072, 0.0057, 0.0048, 0.0113, 0.0083, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0081, 0.0084, 0.0075, 0.0062, 0.0094, 0.0084, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:25:30,092 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.801e+02 3.265e+02 4.111e+02 6.652e+02, threshold=6.530e+02, percent-clipped=1.0 2023-05-17 00:25:30,174 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262681.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:25:38,085 INFO [finetune.py:992] (1/2) Epoch 14, batch 3650, loss[loss=0.1571, simple_loss=0.2479, pruned_loss=0.03314, over 12274.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2551, pruned_loss=0.03803, over 2370739.08 frames. ], batch size: 37, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:26:07,458 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3753, 4.6819, 4.0106, 4.9765, 4.5246, 2.9014, 4.3173, 3.2269], device='cuda:1'), covar=tensor([0.0766, 0.0877, 0.1567, 0.0485, 0.1177, 0.1756, 0.1145, 0.3121], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0384, 0.0362, 0.0321, 0.0371, 0.0274, 0.0346, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:26:08,077 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2522, 5.0756, 5.2064, 5.2343, 4.8656, 4.9457, 4.6661, 5.1185], device='cuda:1'), covar=tensor([0.0745, 0.0672, 0.0832, 0.0565, 0.1911, 0.1330, 0.0636, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0708, 0.0628, 0.0636, 0.0861, 0.0752, 0.0567, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:26:14,944 INFO [finetune.py:992] (1/2) Epoch 14, batch 3700, loss[loss=0.1577, simple_loss=0.2396, pruned_loss=0.0379, over 12155.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2551, pruned_loss=0.03815, over 2368056.64 frames. ], batch size: 29, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:26:20,136 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262749.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:26:42,895 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.711e+02 3.026e+02 3.699e+02 9.052e+02, threshold=6.051e+02, percent-clipped=1.0 2023-05-17 00:26:43,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-17 00:26:49,508 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7001, 2.6400, 4.3379, 4.3891, 2.7047, 2.5200, 2.8721, 2.1887], device='cuda:1'), covar=tensor([0.1632, 0.3134, 0.0518, 0.0474, 0.1394, 0.2500, 0.2756, 0.3836], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0392, 0.0279, 0.0306, 0.0277, 0.0312, 0.0393, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:26:50,661 INFO [finetune.py:992] (1/2) Epoch 14, batch 3750, loss[loss=0.1803, simple_loss=0.2655, pruned_loss=0.04759, over 12036.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2547, pruned_loss=0.03803, over 2371496.60 frames. ], batch size: 42, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:26:52,403 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6841, 2.8273, 4.5269, 4.6224, 2.6566, 2.5793, 2.9635, 2.2296], device='cuda:1'), covar=tensor([0.1623, 0.2952, 0.0468, 0.0447, 0.1438, 0.2459, 0.2757, 0.3710], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0392, 0.0280, 0.0306, 0.0277, 0.0313, 0.0393, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:26:59,501 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3273, 4.8651, 3.9667, 4.9792, 4.3977, 2.4960, 4.1050, 2.9386], device='cuda:1'), covar=tensor([0.0741, 0.0576, 0.1366, 0.0484, 0.1150, 0.1899, 0.1121, 0.3285], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0381, 0.0360, 0.0319, 0.0369, 0.0272, 0.0344, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:27:03,652 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262810.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:27:26,128 INFO [finetune.py:992] (1/2) Epoch 14, batch 3800, loss[loss=0.153, simple_loss=0.2504, pruned_loss=0.02784, over 10519.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2539, pruned_loss=0.03776, over 2381107.21 frames. ], batch size: 68, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:27:41,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0533, 6.0916, 5.8294, 5.3215, 5.2362, 6.0000, 5.6519, 5.3713], device='cuda:1'), covar=tensor([0.0728, 0.0848, 0.0572, 0.1677, 0.0655, 0.0650, 0.1412, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0621, 0.0562, 0.0514, 0.0638, 0.0420, 0.0731, 0.0788, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:27:55,341 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 2.445e+02 2.946e+02 3.863e+02 7.594e+02, threshold=5.892e+02, percent-clipped=4.0 2023-05-17 00:28:03,247 INFO [finetune.py:992] (1/2) Epoch 14, batch 3850, loss[loss=0.1611, simple_loss=0.2449, pruned_loss=0.03866, over 12021.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03776, over 2381345.04 frames. ], batch size: 31, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:28:38,951 INFO [finetune.py:992] (1/2) Epoch 14, batch 3900, loss[loss=0.1549, simple_loss=0.2405, pruned_loss=0.03465, over 12279.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2551, pruned_loss=0.03836, over 2370913.84 frames. ], batch size: 33, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:28:41,180 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9621, 4.8393, 4.7514, 4.8626, 4.3606, 4.9688, 4.9225, 5.1239], device='cuda:1'), covar=tensor([0.0250, 0.0175, 0.0202, 0.0341, 0.0880, 0.0334, 0.0187, 0.0184], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0203, 0.0195, 0.0251, 0.0247, 0.0224, 0.0181, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 00:29:06,457 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.531e+02 3.065e+02 3.595e+02 1.862e+03, threshold=6.130e+02, percent-clipped=2.0 2023-05-17 00:29:14,284 INFO [finetune.py:992] (1/2) Epoch 14, batch 3950, loss[loss=0.1696, simple_loss=0.2638, pruned_loss=0.0377, over 12197.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2552, pruned_loss=0.0385, over 2370550.33 frames. ], batch size: 35, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:29:21,160 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5292, 5.3022, 5.4587, 5.4640, 5.0522, 5.1681, 4.9078, 5.3566], device='cuda:1'), covar=tensor([0.0650, 0.0625, 0.0778, 0.0554, 0.1922, 0.1290, 0.0571, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0708, 0.0627, 0.0637, 0.0860, 0.0755, 0.0567, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:29:30,382 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5432, 2.3272, 3.3586, 4.3792, 2.2221, 4.4544, 4.4620, 4.5013], device='cuda:1'), covar=tensor([0.0125, 0.1358, 0.0495, 0.0165, 0.1324, 0.0183, 0.0143, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0205, 0.0184, 0.0121, 0.0192, 0.0181, 0.0177, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:29:39,810 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7366, 2.8354, 4.7652, 4.8528, 2.8098, 2.5892, 3.0754, 2.2590], device='cuda:1'), covar=tensor([0.1624, 0.3188, 0.0400, 0.0387, 0.1357, 0.2521, 0.2672, 0.4016], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0391, 0.0279, 0.0305, 0.0276, 0.0313, 0.0392, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:29:51,630 INFO [finetune.py:992] (1/2) Epoch 14, batch 4000, loss[loss=0.1433, simple_loss=0.2208, pruned_loss=0.03295, over 11999.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2547, pruned_loss=0.03851, over 2367103.77 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:30:16,027 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-05-17 00:30:19,048 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.610e+02 2.958e+02 3.557e+02 5.837e+02, threshold=5.916e+02, percent-clipped=0.0 2023-05-17 00:30:25,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 00:30:27,032 INFO [finetune.py:992] (1/2) Epoch 14, batch 4050, loss[loss=0.1486, simple_loss=0.2357, pruned_loss=0.03078, over 12083.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2546, pruned_loss=0.03857, over 2374850.82 frames. ], batch size: 32, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:30:29,362 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263095.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:30:36,207 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263105.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:30:40,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-17 00:30:40,652 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263111.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:30:53,512 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 00:30:55,552 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0341, 2.4621, 3.5145, 2.9624, 3.3774, 3.0420, 2.4570, 3.4255], device='cuda:1'), covar=tensor([0.0149, 0.0346, 0.0167, 0.0244, 0.0149, 0.0188, 0.0384, 0.0140], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0204, 0.0192, 0.0189, 0.0219, 0.0168, 0.0201, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:31:01,230 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263140.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:31:02,453 INFO [finetune.py:992] (1/2) Epoch 14, batch 4100, loss[loss=0.1703, simple_loss=0.2622, pruned_loss=0.03915, over 12163.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2546, pruned_loss=0.03811, over 2381747.42 frames. ], batch size: 36, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:31:13,878 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263156.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:31:16,741 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-05-17 00:31:25,123 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263172.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:31:31,294 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.679e+02 3.227e+02 3.947e+02 7.741e+02, threshold=6.455e+02, percent-clipped=5.0 2023-05-17 00:31:39,192 INFO [finetune.py:992] (1/2) Epoch 14, batch 4150, loss[loss=0.1579, simple_loss=0.2484, pruned_loss=0.03369, over 12207.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2545, pruned_loss=0.03815, over 2382451.78 frames. ], batch size: 35, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:31:46,206 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263201.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:31:54,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-17 00:31:55,284 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263214.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:32:11,635 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6267, 5.3748, 5.5128, 5.5706, 5.1014, 5.2288, 4.9571, 5.4564], device='cuda:1'), covar=tensor([0.0613, 0.0655, 0.0929, 0.0557, 0.2145, 0.1554, 0.0624, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0715, 0.0634, 0.0647, 0.0874, 0.0768, 0.0573, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:32:15,091 INFO [finetune.py:992] (1/2) Epoch 14, batch 4200, loss[loss=0.1572, simple_loss=0.2572, pruned_loss=0.0286, over 12360.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2549, pruned_loss=0.03844, over 2386780.45 frames. ], batch size: 35, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:32:18,099 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263246.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:32:31,164 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3287, 4.8089, 4.2039, 4.9640, 4.6050, 2.9352, 4.3662, 3.1813], device='cuda:1'), covar=tensor([0.0910, 0.0762, 0.1480, 0.0527, 0.1139, 0.1764, 0.1181, 0.3416], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0382, 0.0362, 0.0321, 0.0370, 0.0274, 0.0348, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:32:31,871 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8223, 2.5857, 3.5599, 3.6511, 2.8368, 2.6993, 2.7245, 2.4231], device='cuda:1'), covar=tensor([0.1341, 0.2687, 0.0707, 0.0577, 0.1112, 0.2114, 0.2496, 0.3413], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0391, 0.0279, 0.0304, 0.0276, 0.0312, 0.0392, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:32:38,929 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263275.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:32:42,995 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.543e+02 3.024e+02 3.575e+02 6.631e+02, threshold=6.048e+02, percent-clipped=1.0 2023-05-17 00:32:51,484 INFO [finetune.py:992] (1/2) Epoch 14, batch 4250, loss[loss=0.2039, simple_loss=0.2849, pruned_loss=0.0614, over 10714.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2557, pruned_loss=0.03873, over 2377791.44 frames. ], batch size: 68, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:32:59,926 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5957, 2.2392, 2.9817, 2.5673, 2.8321, 2.7866, 2.2033, 2.9430], device='cuda:1'), covar=tensor([0.0136, 0.0350, 0.0180, 0.0252, 0.0171, 0.0194, 0.0330, 0.0155], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0205, 0.0192, 0.0189, 0.0220, 0.0168, 0.0201, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:33:02,785 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263307.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:33:27,715 INFO [finetune.py:992] (1/2) Epoch 14, batch 4300, loss[loss=0.1732, simple_loss=0.2671, pruned_loss=0.03961, over 11486.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2555, pruned_loss=0.03844, over 2377217.46 frames. ], batch size: 55, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:33:37,141 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1099, 4.6935, 5.0570, 4.4220, 4.7174, 4.4944, 5.0859, 4.7298], device='cuda:1'), covar=tensor([0.0253, 0.0450, 0.0337, 0.0299, 0.0435, 0.0385, 0.0242, 0.0457], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0273, 0.0298, 0.0273, 0.0272, 0.0274, 0.0247, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:33:55,194 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.623e+02 2.984e+02 3.575e+02 6.098e+02, threshold=5.968e+02, percent-clipped=1.0 2023-05-17 00:34:01,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-05-17 00:34:02,857 INFO [finetune.py:992] (1/2) Epoch 14, batch 4350, loss[loss=0.1772, simple_loss=0.2685, pruned_loss=0.04301, over 12123.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2555, pruned_loss=0.03841, over 2378060.94 frames. ], batch size: 38, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:34:12,420 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263405.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:34:39,786 INFO [finetune.py:992] (1/2) Epoch 14, batch 4400, loss[loss=0.1696, simple_loss=0.2555, pruned_loss=0.04187, over 12240.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2556, pruned_loss=0.03856, over 2373722.79 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:34:45,839 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263451.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:34:47,349 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263453.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:34:57,323 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263467.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:35:07,231 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.773e+02 2.616e+02 3.252e+02 4.018e+02 9.612e+02, threshold=6.504e+02, percent-clipped=5.0 2023-05-17 00:35:15,002 INFO [finetune.py:992] (1/2) Epoch 14, batch 4450, loss[loss=0.1595, simple_loss=0.2387, pruned_loss=0.04014, over 12329.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2557, pruned_loss=0.03867, over 2372711.61 frames. ], batch size: 31, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:35:17,828 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263496.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:35:27,356 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1596, 4.5910, 4.0829, 4.8465, 4.4386, 2.9198, 4.2297, 2.8733], device='cuda:1'), covar=tensor([0.0900, 0.0760, 0.1386, 0.0532, 0.1059, 0.1678, 0.1050, 0.3669], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0385, 0.0365, 0.0324, 0.0373, 0.0276, 0.0351, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:35:38,697 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2395, 4.0078, 4.1292, 4.4153, 3.0840, 4.0864, 2.6849, 4.0896], device='cuda:1'), covar=tensor([0.1677, 0.0798, 0.0831, 0.0674, 0.1152, 0.0607, 0.1824, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0268, 0.0298, 0.0364, 0.0243, 0.0244, 0.0263, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:35:50,404 INFO [finetune.py:992] (1/2) Epoch 14, batch 4500, loss[loss=0.153, simple_loss=0.2281, pruned_loss=0.03895, over 12305.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2553, pruned_loss=0.03849, over 2372433.32 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:36:10,616 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263570.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:36:18,693 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3306, 4.8739, 5.0714, 5.1712, 4.8666, 5.1255, 5.0292, 3.1233], device='cuda:1'), covar=tensor([0.0072, 0.0066, 0.0070, 0.0052, 0.0051, 0.0087, 0.0067, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0081, 0.0085, 0.0076, 0.0063, 0.0096, 0.0085, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:36:19,219 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.705e+02 3.261e+02 3.794e+02 6.330e+02, threshold=6.522e+02, percent-clipped=0.0 2023-05-17 00:36:27,819 INFO [finetune.py:992] (1/2) Epoch 14, batch 4550, loss[loss=0.1783, simple_loss=0.2711, pruned_loss=0.04278, over 12284.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2547, pruned_loss=0.03795, over 2381184.14 frames. ], batch size: 37, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:36:29,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-17 00:36:35,128 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263602.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:36:44,789 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1018, 4.6435, 4.8830, 4.9478, 4.6971, 4.9048, 4.8129, 2.8836], device='cuda:1'), covar=tensor([0.0096, 0.0069, 0.0079, 0.0064, 0.0051, 0.0097, 0.0086, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0081, 0.0085, 0.0076, 0.0063, 0.0096, 0.0085, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:37:03,756 INFO [finetune.py:992] (1/2) Epoch 14, batch 4600, loss[loss=0.1298, simple_loss=0.2125, pruned_loss=0.02352, over 12008.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2549, pruned_loss=0.03809, over 2372441.04 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:37:06,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-05-17 00:37:09,393 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1520, 6.0762, 5.9478, 5.3586, 5.1812, 6.0534, 5.6821, 5.4439], device='cuda:1'), covar=tensor([0.0655, 0.1032, 0.0646, 0.1530, 0.0668, 0.0722, 0.1323, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0570, 0.0518, 0.0648, 0.0427, 0.0739, 0.0798, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:37:31,549 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.517e+02 3.041e+02 3.636e+02 5.915e+02, threshold=6.082e+02, percent-clipped=0.0 2023-05-17 00:37:39,338 INFO [finetune.py:992] (1/2) Epoch 14, batch 4650, loss[loss=0.1702, simple_loss=0.2647, pruned_loss=0.03782, over 12155.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2548, pruned_loss=0.03796, over 2377058.47 frames. ], batch size: 34, lr: 3.60e-03, grad_scale: 8.0 2023-05-17 00:38:16,071 INFO [finetune.py:992] (1/2) Epoch 14, batch 4700, loss[loss=0.1404, simple_loss=0.2279, pruned_loss=0.02647, over 12165.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2557, pruned_loss=0.03855, over 2374424.86 frames. ], batch size: 29, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:38:22,865 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263751.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:38:34,150 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263767.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:38:43,922 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.488e+02 3.045e+02 3.623e+02 8.469e+02, threshold=6.090e+02, percent-clipped=2.0 2023-05-17 00:38:51,485 INFO [finetune.py:992] (1/2) Epoch 14, batch 4750, loss[loss=0.1661, simple_loss=0.2633, pruned_loss=0.0345, over 12350.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2552, pruned_loss=0.03865, over 2379485.66 frames. ], batch size: 36, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:38:54,301 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263796.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:38:56,366 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263799.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:38:58,982 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6294, 2.8972, 4.4985, 4.6870, 2.8676, 2.6052, 2.9360, 2.1223], device='cuda:1'), covar=tensor([0.1662, 0.2933, 0.0533, 0.0408, 0.1367, 0.2420, 0.2743, 0.4145], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0388, 0.0275, 0.0301, 0.0273, 0.0309, 0.0389, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:39:07,958 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263815.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:39:20,915 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263833.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 00:39:26,663 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9619, 4.5235, 3.9438, 4.6838, 4.2468, 2.6960, 4.0250, 2.8549], device='cuda:1'), covar=tensor([0.0980, 0.0751, 0.1363, 0.0561, 0.1136, 0.1777, 0.1127, 0.3544], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0384, 0.0363, 0.0322, 0.0372, 0.0275, 0.0351, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:39:27,093 INFO [finetune.py:992] (1/2) Epoch 14, batch 4800, loss[loss=0.1843, simple_loss=0.2745, pruned_loss=0.04707, over 12046.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2558, pruned_loss=0.03883, over 2383137.20 frames. ], batch size: 42, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:39:28,502 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263844.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:39:47,103 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263870.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:39:48,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-05-17 00:39:55,297 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.609e+02 3.113e+02 3.835e+02 5.503e+02, threshold=6.225e+02, percent-clipped=0.0 2023-05-17 00:40:03,765 INFO [finetune.py:992] (1/2) Epoch 14, batch 4850, loss[loss=0.1503, simple_loss=0.2407, pruned_loss=0.02997, over 12105.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2549, pruned_loss=0.03846, over 2379979.43 frames. ], batch size: 32, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:40:05,280 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263894.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 00:40:10,808 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263902.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:40:22,298 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263918.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:40:39,482 INFO [finetune.py:992] (1/2) Epoch 14, batch 4900, loss[loss=0.1593, simple_loss=0.2477, pruned_loss=0.03543, over 12353.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2545, pruned_loss=0.0382, over 2376395.91 frames. ], batch size: 31, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:40:45,303 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263950.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:40:46,864 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6023, 5.3999, 5.5208, 5.5518, 5.1756, 5.2829, 4.9994, 5.4829], device='cuda:1'), covar=tensor([0.0684, 0.0615, 0.0871, 0.0640, 0.2017, 0.1303, 0.0542, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0711, 0.0633, 0.0644, 0.0869, 0.0762, 0.0571, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:40:50,386 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5685, 5.3242, 5.4705, 5.5070, 5.0991, 5.2233, 4.9126, 5.4387], device='cuda:1'), covar=tensor([0.0616, 0.0659, 0.0822, 0.0611, 0.2140, 0.1281, 0.0595, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0711, 0.0633, 0.0644, 0.0870, 0.0762, 0.0571, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:40:52,656 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263960.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:41:07,269 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.685e+02 3.434e+02 4.193e+02 1.143e+03, threshold=6.868e+02, percent-clipped=7.0 2023-05-17 00:41:15,122 INFO [finetune.py:992] (1/2) Epoch 14, batch 4950, loss[loss=0.1566, simple_loss=0.247, pruned_loss=0.03305, over 12034.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2546, pruned_loss=0.03809, over 2373807.09 frames. ], batch size: 31, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:41:39,737 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264021.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:41:55,104 INFO [finetune.py:992] (1/2) Epoch 14, batch 5000, loss[loss=0.1372, simple_loss=0.2179, pruned_loss=0.02828, over 12255.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2557, pruned_loss=0.03876, over 2358177.77 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 8.0 2023-05-17 00:41:56,759 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2002, 3.8892, 3.9527, 4.3746, 2.5729, 3.8617, 2.4649, 3.7317], device='cuda:1'), covar=tensor([0.1730, 0.0809, 0.0974, 0.0658, 0.1548, 0.0686, 0.1974, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0267, 0.0299, 0.0363, 0.0243, 0.0243, 0.0261, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:42:23,094 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 2.696e+02 3.369e+02 3.930e+02 9.340e+02, threshold=6.738e+02, percent-clipped=2.0 2023-05-17 00:42:30,646 INFO [finetune.py:992] (1/2) Epoch 14, batch 5050, loss[loss=0.1729, simple_loss=0.2645, pruned_loss=0.0406, over 12308.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2561, pruned_loss=0.03886, over 2365644.37 frames. ], batch size: 34, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:42:42,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.39 vs. limit=5.0 2023-05-17 00:42:59,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.29 vs. limit=5.0 2023-05-17 00:43:02,659 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9445, 4.5317, 4.6810, 4.8295, 4.5764, 4.8232, 4.5844, 2.5731], device='cuda:1'), covar=tensor([0.0112, 0.0086, 0.0101, 0.0077, 0.0065, 0.0112, 0.0094, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0081, 0.0084, 0.0076, 0.0063, 0.0096, 0.0084, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:43:06,007 INFO [finetune.py:992] (1/2) Epoch 14, batch 5100, loss[loss=0.1618, simple_loss=0.2504, pruned_loss=0.03656, over 12282.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.257, pruned_loss=0.03921, over 2366237.06 frames. ], batch size: 33, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:43:19,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.18 vs. limit=5.0 2023-05-17 00:43:34,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.640e+02 3.176e+02 3.824e+02 9.093e+02, threshold=6.353e+02, percent-clipped=2.0 2023-05-17 00:43:40,585 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264189.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:43:42,572 INFO [finetune.py:992] (1/2) Epoch 14, batch 5150, loss[loss=0.1795, simple_loss=0.2721, pruned_loss=0.04344, over 11832.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2567, pruned_loss=0.03906, over 2372295.09 frames. ], batch size: 44, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:43:44,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 00:44:12,795 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264234.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:44:18,386 INFO [finetune.py:992] (1/2) Epoch 14, batch 5200, loss[loss=0.1815, simple_loss=0.2772, pruned_loss=0.04296, over 11542.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2563, pruned_loss=0.03877, over 2372488.58 frames. ], batch size: 48, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:44:46,092 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.687e+02 3.109e+02 3.729e+02 7.458e+02, threshold=6.219e+02, percent-clipped=1.0 2023-05-17 00:44:53,718 INFO [finetune.py:992] (1/2) Epoch 14, batch 5250, loss[loss=0.2126, simple_loss=0.2903, pruned_loss=0.06746, over 8205.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2565, pruned_loss=0.03889, over 2369329.97 frames. ], batch size: 98, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:44:55,979 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264295.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:45:12,364 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264316.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:45:24,833 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0750, 6.0223, 5.8528, 5.2912, 5.1935, 5.9534, 5.5915, 5.3092], device='cuda:1'), covar=tensor([0.0676, 0.0870, 0.0645, 0.1748, 0.0654, 0.0721, 0.1528, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0574, 0.0522, 0.0656, 0.0430, 0.0744, 0.0803, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:45:30,406 INFO [finetune.py:992] (1/2) Epoch 14, batch 5300, loss[loss=0.1503, simple_loss=0.2355, pruned_loss=0.03258, over 12117.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2565, pruned_loss=0.03889, over 2367992.87 frames. ], batch size: 30, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:45:31,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.15 vs. limit=5.0 2023-05-17 00:45:34,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-17 00:45:36,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5893, 5.3731, 5.5071, 5.5762, 5.1852, 5.1816, 5.0073, 5.4145], device='cuda:1'), covar=tensor([0.0645, 0.0558, 0.0750, 0.0498, 0.1562, 0.1414, 0.0547, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0708, 0.0628, 0.0646, 0.0870, 0.0764, 0.0571, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:45:58,266 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.985e+02 3.539e+02 4.100e+02 7.031e+02, threshold=7.077e+02, percent-clipped=2.0 2023-05-17 00:46:06,208 INFO [finetune.py:992] (1/2) Epoch 14, batch 5350, loss[loss=0.1964, simple_loss=0.2846, pruned_loss=0.05411, over 11424.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2557, pruned_loss=0.03884, over 2363983.14 frames. ], batch size: 55, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:46:12,975 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1543, 5.9134, 5.5003, 5.4708, 6.0009, 5.2441, 5.4561, 5.5619], device='cuda:1'), covar=tensor([0.1472, 0.1005, 0.0968, 0.1805, 0.0853, 0.2302, 0.1958, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0508, 0.0404, 0.0454, 0.0477, 0.0442, 0.0401, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:46:42,443 INFO [finetune.py:992] (1/2) Epoch 14, batch 5400, loss[loss=0.1399, simple_loss=0.219, pruned_loss=0.03041, over 12015.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2558, pruned_loss=0.03879, over 2369491.35 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:47:09,397 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264479.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:47:10,571 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.548e+02 2.961e+02 3.629e+02 7.539e+02, threshold=5.921e+02, percent-clipped=1.0 2023-05-17 00:47:15,685 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9576, 3.8931, 3.9467, 4.0302, 3.7726, 3.7957, 3.6889, 3.9542], device='cuda:1'), covar=tensor([0.1244, 0.0809, 0.1605, 0.0755, 0.1896, 0.1507, 0.0646, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0709, 0.0629, 0.0647, 0.0872, 0.0764, 0.0573, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:47:16,380 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264489.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:47:18,405 INFO [finetune.py:992] (1/2) Epoch 14, batch 5450, loss[loss=0.1537, simple_loss=0.2396, pruned_loss=0.03387, over 12359.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2553, pruned_loss=0.03863, over 2362282.20 frames. ], batch size: 31, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:47:24,149 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264500.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 00:47:50,398 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264537.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:47:52,626 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264540.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:47:53,786 INFO [finetune.py:992] (1/2) Epoch 14, batch 5500, loss[loss=0.2477, simple_loss=0.3166, pruned_loss=0.08938, over 7936.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2554, pruned_loss=0.03869, over 2366900.79 frames. ], batch size: 98, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:48:07,535 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264561.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 00:48:20,385 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6380, 2.9694, 3.8293, 4.5971, 4.0241, 4.6995, 4.0517, 3.1900], device='cuda:1'), covar=tensor([0.0034, 0.0336, 0.0133, 0.0034, 0.0104, 0.0075, 0.0114, 0.0349], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0124, 0.0105, 0.0079, 0.0104, 0.0116, 0.0097, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:48:21,550 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.628e+02 3.099e+02 3.637e+02 7.816e+02, threshold=6.199e+02, percent-clipped=1.0 2023-05-17 00:48:26,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-17 00:48:28,809 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264590.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:48:30,152 INFO [finetune.py:992] (1/2) Epoch 14, batch 5550, loss[loss=0.1729, simple_loss=0.2673, pruned_loss=0.03925, over 11818.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2549, pruned_loss=0.03839, over 2375723.87 frames. ], batch size: 44, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:48:31,745 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9354, 5.8577, 5.6669, 5.1423, 5.0729, 5.8494, 5.4332, 5.2137], device='cuda:1'), covar=tensor([0.0809, 0.1120, 0.0759, 0.1914, 0.0840, 0.0769, 0.1699, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0571, 0.0521, 0.0651, 0.0427, 0.0740, 0.0801, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 00:48:48,566 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264616.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:49:06,847 INFO [finetune.py:992] (1/2) Epoch 14, batch 5600, loss[loss=0.1877, simple_loss=0.2773, pruned_loss=0.049, over 12105.00 frames. ], tot_loss[loss=0.166, simple_loss=0.255, pruned_loss=0.03854, over 2377002.07 frames. ], batch size: 38, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:49:22,150 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-17 00:49:22,513 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264664.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:49:34,635 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.679e+02 3.341e+02 3.846e+02 7.061e+02, threshold=6.682e+02, percent-clipped=3.0 2023-05-17 00:49:42,213 INFO [finetune.py:992] (1/2) Epoch 14, batch 5650, loss[loss=0.1497, simple_loss=0.2461, pruned_loss=0.02662, over 12268.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2549, pruned_loss=0.03835, over 2383135.02 frames. ], batch size: 37, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:50:04,493 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3047, 2.6844, 3.8851, 3.1457, 3.6795, 3.2775, 2.6795, 3.7477], device='cuda:1'), covar=tensor([0.0140, 0.0367, 0.0160, 0.0279, 0.0133, 0.0198, 0.0363, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0209, 0.0197, 0.0193, 0.0223, 0.0171, 0.0204, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:50:18,494 INFO [finetune.py:992] (1/2) Epoch 14, batch 5700, loss[loss=0.1738, simple_loss=0.2704, pruned_loss=0.03863, over 12159.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2549, pruned_loss=0.03859, over 2378538.17 frames. ], batch size: 36, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:50:32,021 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7072, 2.7500, 4.4134, 4.5121, 2.6989, 2.5800, 2.9475, 2.0992], device='cuda:1'), covar=tensor([0.1638, 0.3179, 0.0517, 0.0448, 0.1432, 0.2392, 0.2758, 0.4329], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0393, 0.0279, 0.0303, 0.0276, 0.0313, 0.0394, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:50:34,075 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4118, 2.4196, 3.6062, 4.2765, 3.8136, 4.4977, 3.7455, 3.1177], device='cuda:1'), covar=tensor([0.0037, 0.0443, 0.0164, 0.0062, 0.0113, 0.0070, 0.0136, 0.0357], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0125, 0.0106, 0.0079, 0.0104, 0.0116, 0.0098, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:50:46,732 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 2.696e+02 3.141e+02 3.814e+02 8.222e+02, threshold=6.283e+02, percent-clipped=3.0 2023-05-17 00:50:50,935 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-17 00:50:54,495 INFO [finetune.py:992] (1/2) Epoch 14, batch 5750, loss[loss=0.1791, simple_loss=0.2718, pruned_loss=0.04317, over 12150.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.255, pruned_loss=0.03874, over 2379113.55 frames. ], batch size: 36, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:51:03,410 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264804.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:51:09,078 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0118, 5.8406, 5.3768, 5.4193, 5.9601, 5.2281, 5.4151, 5.4763], device='cuda:1'), covar=tensor([0.1603, 0.1057, 0.1212, 0.2021, 0.0896, 0.2429, 0.1793, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0512, 0.0410, 0.0458, 0.0480, 0.0446, 0.0408, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:51:25,693 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264835.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:51:30,560 INFO [finetune.py:992] (1/2) Epoch 14, batch 5800, loss[loss=0.1357, simple_loss=0.2181, pruned_loss=0.02659, over 12348.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.255, pruned_loss=0.03875, over 2376285.62 frames. ], batch size: 30, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:51:40,282 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264856.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 00:51:46,783 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264865.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:51:57,900 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.752e+02 3.123e+02 3.647e+02 6.224e+02, threshold=6.246e+02, percent-clipped=0.0 2023-05-17 00:52:04,960 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264890.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:52:06,259 INFO [finetune.py:992] (1/2) Epoch 14, batch 5850, loss[loss=0.1619, simple_loss=0.2522, pruned_loss=0.03584, over 12099.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2566, pruned_loss=0.03922, over 2371816.40 frames. ], batch size: 32, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:52:24,061 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2407, 4.9157, 5.2497, 4.6056, 4.8785, 4.7135, 5.2732, 4.9646], device='cuda:1'), covar=tensor([0.0299, 0.0374, 0.0277, 0.0277, 0.0434, 0.0333, 0.0210, 0.0270], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0274, 0.0299, 0.0274, 0.0274, 0.0275, 0.0249, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:52:39,351 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264938.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:52:42,209 INFO [finetune.py:992] (1/2) Epoch 14, batch 5900, loss[loss=0.1762, simple_loss=0.2693, pruned_loss=0.04154, over 12205.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2561, pruned_loss=0.03906, over 2373953.74 frames. ], batch size: 35, lr: 3.59e-03, grad_scale: 16.0 2023-05-17 00:53:10,020 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.718e+02 3.176e+02 3.638e+02 7.396e+02, threshold=6.352e+02, percent-clipped=1.0 2023-05-17 00:53:17,779 INFO [finetune.py:992] (1/2) Epoch 14, batch 5950, loss[loss=0.1465, simple_loss=0.2368, pruned_loss=0.02811, over 12294.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2566, pruned_loss=0.03935, over 2378074.80 frames. ], batch size: 34, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:53:24,772 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5484, 2.6735, 3.7959, 4.4337, 3.9702, 4.6127, 3.8084, 3.4218], device='cuda:1'), covar=tensor([0.0043, 0.0397, 0.0127, 0.0051, 0.0108, 0.0061, 0.0137, 0.0301], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0124, 0.0105, 0.0079, 0.0104, 0.0116, 0.0098, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:53:54,410 INFO [finetune.py:992] (1/2) Epoch 14, batch 6000, loss[loss=0.1412, simple_loss=0.2268, pruned_loss=0.0278, over 12013.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2553, pruned_loss=0.03879, over 2380756.69 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:53:54,410 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 00:54:05,024 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9131, 2.7788, 3.2671, 2.2660, 2.5260, 2.9820, 2.5976, 2.9853], device='cuda:1'), covar=tensor([0.0589, 0.0970, 0.0352, 0.1172, 0.1446, 0.1088, 0.1160, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0237, 0.0252, 0.0183, 0.0237, 0.0295, 0.0225, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:54:06,380 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9466, 2.0271, 2.9445, 3.9292, 2.1803, 4.0265, 3.4695, 3.9089], device='cuda:1'), covar=tensor([0.0155, 0.1385, 0.0505, 0.0127, 0.1367, 0.0179, 0.0354, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0206, 0.0185, 0.0123, 0.0195, 0.0182, 0.0179, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:54:12,932 INFO [finetune.py:1026] (1/2) Epoch 14, validation: loss=0.3136, simple_loss=0.3909, pruned_loss=0.1181, over 1020973.00 frames. 2023-05-17 00:54:12,933 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 00:54:40,684 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.610e+02 3.085e+02 3.720e+02 7.946e+02, threshold=6.170e+02, percent-clipped=2.0 2023-05-17 00:54:48,375 INFO [finetune.py:992] (1/2) Epoch 14, batch 6050, loss[loss=0.1621, simple_loss=0.2593, pruned_loss=0.03241, over 12159.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2555, pruned_loss=0.03862, over 2384770.20 frames. ], batch size: 36, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:55:19,425 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265135.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:55:24,130 INFO [finetune.py:992] (1/2) Epoch 14, batch 6100, loss[loss=0.1506, simple_loss=0.241, pruned_loss=0.03013, over 12346.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2565, pruned_loss=0.03925, over 2378427.85 frames. ], batch size: 31, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:55:34,369 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265156.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 00:55:37,253 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265160.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:55:52,810 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.534e+02 2.916e+02 4.072e+02 6.898e+02, threshold=5.833e+02, percent-clipped=2.0 2023-05-17 00:55:54,322 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265183.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:56:00,678 INFO [finetune.py:992] (1/2) Epoch 14, batch 6150, loss[loss=0.1771, simple_loss=0.2684, pruned_loss=0.04289, over 11163.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2571, pruned_loss=0.03928, over 2374151.16 frames. ], batch size: 55, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:56:09,304 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265204.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 00:56:10,103 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3392, 2.6358, 3.6627, 4.2853, 3.8360, 4.4171, 3.6821, 2.9744], device='cuda:1'), covar=tensor([0.0043, 0.0385, 0.0150, 0.0052, 0.0117, 0.0061, 0.0159, 0.0392], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0124, 0.0105, 0.0079, 0.0104, 0.0116, 0.0098, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:56:12,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-05-17 00:56:27,300 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7274, 2.9063, 4.5395, 4.6565, 2.8711, 2.6742, 2.9918, 2.1854], device='cuda:1'), covar=tensor([0.1627, 0.3070, 0.0468, 0.0473, 0.1375, 0.2444, 0.2743, 0.4344], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0390, 0.0276, 0.0301, 0.0272, 0.0310, 0.0389, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 00:56:33,741 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8515, 3.5491, 5.2535, 2.7532, 2.9465, 3.8785, 3.2513, 3.8905], device='cuda:1'), covar=tensor([0.0433, 0.1067, 0.0289, 0.1215, 0.1865, 0.1467, 0.1416, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0237, 0.0252, 0.0184, 0.0238, 0.0296, 0.0226, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:56:36,206 INFO [finetune.py:992] (1/2) Epoch 14, batch 6200, loss[loss=0.1444, simple_loss=0.2233, pruned_loss=0.03279, over 12014.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2567, pruned_loss=0.03914, over 2380168.28 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:57:01,759 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265278.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:57:04,384 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.684e+02 3.070e+02 3.874e+02 6.926e+02, threshold=6.140e+02, percent-clipped=2.0 2023-05-17 00:57:08,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2023-05-17 00:57:12,072 INFO [finetune.py:992] (1/2) Epoch 14, batch 6250, loss[loss=0.1721, simple_loss=0.2667, pruned_loss=0.03874, over 11276.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2575, pruned_loss=0.03914, over 2380264.95 frames. ], batch size: 55, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:57:18,000 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2644, 4.8830, 5.1049, 5.1150, 4.9040, 5.1181, 5.0819, 2.8586], device='cuda:1'), covar=tensor([0.0090, 0.0068, 0.0068, 0.0058, 0.0058, 0.0094, 0.0083, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0078, 0.0082, 0.0073, 0.0061, 0.0093, 0.0081, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 00:57:21,676 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3864, 4.7215, 3.0300, 2.7845, 4.0927, 2.7065, 4.0496, 3.3824], device='cuda:1'), covar=tensor([0.0674, 0.0595, 0.1085, 0.1453, 0.0269, 0.1234, 0.0509, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0262, 0.0179, 0.0203, 0.0144, 0.0183, 0.0201, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:57:46,034 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265339.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:57:48,003 INFO [finetune.py:992] (1/2) Epoch 14, batch 6300, loss[loss=0.1486, simple_loss=0.2359, pruned_loss=0.03065, over 12120.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2571, pruned_loss=0.03918, over 2382418.59 frames. ], batch size: 30, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:58:15,271 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0102, 4.8781, 4.8394, 4.9046, 4.4551, 4.9904, 4.9805, 5.1627], device='cuda:1'), covar=tensor([0.0245, 0.0167, 0.0179, 0.0316, 0.0820, 0.0294, 0.0156, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0202, 0.0195, 0.0252, 0.0246, 0.0224, 0.0181, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 00:58:15,803 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 2.676e+02 3.101e+02 3.558e+02 6.054e+02, threshold=6.203e+02, percent-clipped=0.0 2023-05-17 00:58:23,344 INFO [finetune.py:992] (1/2) Epoch 14, batch 6350, loss[loss=0.1936, simple_loss=0.2838, pruned_loss=0.05176, over 12204.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2575, pruned_loss=0.03933, over 2375864.38 frames. ], batch size: 35, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:58:57,040 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1482, 5.9619, 5.5670, 5.5300, 6.0996, 5.3606, 5.5451, 5.5547], device='cuda:1'), covar=tensor([0.1429, 0.0892, 0.0810, 0.1905, 0.0852, 0.2132, 0.1755, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0508, 0.0405, 0.0456, 0.0476, 0.0442, 0.0407, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 00:58:59,866 INFO [finetune.py:992] (1/2) Epoch 14, batch 6400, loss[loss=0.1861, simple_loss=0.2794, pruned_loss=0.04633, over 12069.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2571, pruned_loss=0.03941, over 2367252.85 frames. ], batch size: 42, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:59:04,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-05-17 00:59:12,750 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265460.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 00:59:28,015 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.064e+02 2.672e+02 3.215e+02 3.915e+02 6.084e+02, threshold=6.430e+02, percent-clipped=0.0 2023-05-17 00:59:35,787 INFO [finetune.py:992] (1/2) Epoch 14, batch 6450, loss[loss=0.1571, simple_loss=0.2453, pruned_loss=0.03447, over 12155.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2578, pruned_loss=0.03951, over 2373613.72 frames. ], batch size: 34, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 00:59:47,376 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265508.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:00:01,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4877, 2.3904, 3.6313, 4.3197, 3.8629, 4.4481, 3.7848, 2.9820], device='cuda:1'), covar=tensor([0.0035, 0.0452, 0.0146, 0.0064, 0.0135, 0.0072, 0.0135, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0126, 0.0105, 0.0080, 0.0105, 0.0117, 0.0098, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:00:02,939 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265530.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:00:04,971 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4143, 4.6454, 4.2679, 4.9766, 4.6714, 3.1509, 4.3063, 3.0584], device='cuda:1'), covar=tensor([0.0808, 0.0922, 0.1500, 0.0625, 0.1155, 0.1654, 0.1190, 0.3689], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0388, 0.0368, 0.0324, 0.0375, 0.0276, 0.0355, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:00:10,925 INFO [finetune.py:992] (1/2) Epoch 14, batch 6500, loss[loss=0.1538, simple_loss=0.2366, pruned_loss=0.03553, over 12119.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2573, pruned_loss=0.03896, over 2374233.68 frames. ], batch size: 30, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:00:11,038 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2476, 6.0699, 5.6737, 5.6671, 6.1560, 5.5401, 5.5318, 5.6792], device='cuda:1'), covar=tensor([0.1418, 0.0796, 0.0865, 0.1656, 0.0749, 0.1886, 0.1666, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0509, 0.0407, 0.0457, 0.0477, 0.0443, 0.0407, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:00:39,182 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.785e+02 3.332e+02 3.898e+02 5.510e+02, threshold=6.664e+02, percent-clipped=0.0 2023-05-17 01:00:46,669 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265591.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:00:47,227 INFO [finetune.py:992] (1/2) Epoch 14, batch 6550, loss[loss=0.1809, simple_loss=0.2692, pruned_loss=0.04633, over 12333.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2575, pruned_loss=0.03904, over 2368628.84 frames. ], batch size: 36, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:01:04,072 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265614.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:01:17,257 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7032, 2.7607, 5.2787, 2.5046, 2.6238, 4.1252, 3.0416, 3.9991], device='cuda:1'), covar=tensor([0.0605, 0.1815, 0.0307, 0.1443, 0.2282, 0.1452, 0.1771, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0236, 0.0252, 0.0183, 0.0236, 0.0295, 0.0225, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:01:18,581 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265634.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:01:24,298 INFO [finetune.py:992] (1/2) Epoch 14, batch 6600, loss[loss=0.1797, simple_loss=0.272, pruned_loss=0.04373, over 12344.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2573, pruned_loss=0.03907, over 2373364.05 frames. ], batch size: 31, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:01:28,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.88 vs. limit=5.0 2023-05-17 01:01:48,153 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265675.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 01:01:52,166 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.631e+02 2.986e+02 3.533e+02 9.316e+02, threshold=5.973e+02, percent-clipped=1.0 2023-05-17 01:02:00,138 INFO [finetune.py:992] (1/2) Epoch 14, batch 6650, loss[loss=0.1434, simple_loss=0.2319, pruned_loss=0.02752, over 12282.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2568, pruned_loss=0.03888, over 2372101.47 frames. ], batch size: 33, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:02:08,437 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1216, 2.5921, 3.6805, 3.0646, 3.4755, 3.2028, 2.6004, 3.5405], device='cuda:1'), covar=tensor([0.0153, 0.0372, 0.0131, 0.0253, 0.0156, 0.0180, 0.0341, 0.0130], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0209, 0.0197, 0.0193, 0.0222, 0.0171, 0.0203, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:02:16,978 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265715.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:02:36,491 INFO [finetune.py:992] (1/2) Epoch 14, batch 6700, loss[loss=0.1493, simple_loss=0.2376, pruned_loss=0.03055, over 12352.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2564, pruned_loss=0.03843, over 2380639.57 frames. ], batch size: 31, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:03:01,696 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265776.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:03:04,968 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.051e+02 2.711e+02 3.255e+02 3.641e+02 5.853e+02, threshold=6.511e+02, percent-clipped=0.0 2023-05-17 01:03:12,930 INFO [finetune.py:992] (1/2) Epoch 14, batch 6750, loss[loss=0.1552, simple_loss=0.2474, pruned_loss=0.03153, over 12117.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2559, pruned_loss=0.03839, over 2376597.04 frames. ], batch size: 33, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:03:48,588 INFO [finetune.py:992] (1/2) Epoch 14, batch 6800, loss[loss=0.1777, simple_loss=0.2714, pruned_loss=0.042, over 12037.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2567, pruned_loss=0.03877, over 2380157.74 frames. ], batch size: 40, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:04:10,713 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265872.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:04:11,546 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3757, 4.8038, 4.2654, 5.0264, 4.5867, 3.0639, 4.3421, 2.9564], device='cuda:1'), covar=tensor([0.0811, 0.0687, 0.1367, 0.0440, 0.1164, 0.1599, 0.1088, 0.3634], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0386, 0.0366, 0.0322, 0.0373, 0.0276, 0.0353, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:04:17,069 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 2.693e+02 3.222e+02 3.796e+02 7.708e+02, threshold=6.444e+02, percent-clipped=1.0 2023-05-17 01:04:20,609 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265886.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:04:25,563 INFO [finetune.py:992] (1/2) Epoch 14, batch 6850, loss[loss=0.1694, simple_loss=0.2613, pruned_loss=0.03872, over 12153.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.256, pruned_loss=0.03839, over 2379422.39 frames. ], batch size: 36, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:04:30,122 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9145, 2.4316, 3.5129, 2.8223, 3.3021, 3.1161, 2.4042, 3.4215], device='cuda:1'), covar=tensor([0.0173, 0.0385, 0.0158, 0.0294, 0.0158, 0.0189, 0.0409, 0.0129], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0209, 0.0199, 0.0194, 0.0224, 0.0172, 0.0204, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:04:43,614 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265917.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:04:55,018 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265933.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:04:55,707 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265934.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:05:01,298 INFO [finetune.py:992] (1/2) Epoch 14, batch 6900, loss[loss=0.1633, simple_loss=0.2538, pruned_loss=0.03637, over 12342.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2572, pruned_loss=0.03882, over 2371836.20 frames. ], batch size: 31, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:05:21,152 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265970.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 01:05:27,070 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265978.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:05:28,995 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.838e+02 2.809e+02 3.282e+02 3.911e+02 1.414e+03, threshold=6.565e+02, percent-clipped=6.0 2023-05-17 01:05:29,824 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265982.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:05:31,340 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7069, 5.6630, 5.4716, 5.1289, 5.0407, 5.6439, 5.2595, 5.0540], device='cuda:1'), covar=tensor([0.0623, 0.0865, 0.0669, 0.1679, 0.0888, 0.0783, 0.1764, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0568, 0.0522, 0.0646, 0.0424, 0.0738, 0.0798, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:05:36,844 INFO [finetune.py:992] (1/2) Epoch 14, batch 6950, loss[loss=0.1396, simple_loss=0.2191, pruned_loss=0.02999, over 12271.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2568, pruned_loss=0.03894, over 2367275.04 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:05:39,939 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7346, 2.8829, 4.6724, 4.7621, 2.9738, 2.6796, 3.0773, 2.2114], device='cuda:1'), covar=tensor([0.1736, 0.3132, 0.0432, 0.0413, 0.1298, 0.2522, 0.2611, 0.3997], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0389, 0.0276, 0.0299, 0.0272, 0.0309, 0.0388, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:06:01,609 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4048, 2.4294, 3.7949, 4.3368, 3.8186, 4.5020, 3.8103, 3.0841], device='cuda:1'), covar=tensor([0.0049, 0.0490, 0.0131, 0.0052, 0.0139, 0.0068, 0.0151, 0.0398], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0124, 0.0105, 0.0080, 0.0104, 0.0117, 0.0099, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:06:16,370 INFO [finetune.py:992] (1/2) Epoch 14, batch 7000, loss[loss=0.1554, simple_loss=0.2339, pruned_loss=0.03849, over 11996.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2573, pruned_loss=0.03924, over 2372651.42 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-05-17 01:06:25,213 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8713, 4.1848, 3.7892, 4.4746, 4.1314, 2.8466, 3.8852, 2.9641], device='cuda:1'), covar=tensor([0.1056, 0.0996, 0.1612, 0.0568, 0.1338, 0.1796, 0.1224, 0.3367], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0389, 0.0367, 0.0324, 0.0375, 0.0277, 0.0355, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:06:28,885 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-17 01:06:37,173 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266071.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:06:44,282 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.623e+02 3.097e+02 3.755e+02 1.095e+03, threshold=6.194e+02, percent-clipped=4.0 2023-05-17 01:06:47,440 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3896, 4.2108, 4.1619, 4.5713, 3.2912, 3.9315, 2.5462, 4.1470], device='cuda:1'), covar=tensor([0.1649, 0.0654, 0.0916, 0.0644, 0.1125, 0.0639, 0.1935, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0268, 0.0298, 0.0361, 0.0243, 0.0244, 0.0261, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:06:52,093 INFO [finetune.py:992] (1/2) Epoch 14, batch 7050, loss[loss=0.1679, simple_loss=0.2641, pruned_loss=0.03579, over 12359.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2569, pruned_loss=0.0389, over 2364899.62 frames. ], batch size: 35, lr: 3.58e-03, grad_scale: 32.0 2023-05-17 01:06:56,450 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0070, 5.8193, 5.3995, 5.4169, 5.9762, 5.2964, 5.4644, 5.4025], device='cuda:1'), covar=tensor([0.1331, 0.0999, 0.1091, 0.1777, 0.0902, 0.2030, 0.1957, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0511, 0.0408, 0.0461, 0.0478, 0.0446, 0.0412, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:07:27,801 INFO [finetune.py:992] (1/2) Epoch 14, batch 7100, loss[loss=0.1679, simple_loss=0.2754, pruned_loss=0.03015, over 11735.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2565, pruned_loss=0.03877, over 2373005.61 frames. ], batch size: 48, lr: 3.58e-03, grad_scale: 32.0 2023-05-17 01:07:56,841 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.947e+02 2.647e+02 3.091e+02 3.573e+02 6.537e+02, threshold=6.182e+02, percent-clipped=2.0 2023-05-17 01:08:00,471 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266186.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:08:04,183 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2815, 3.1926, 3.0194, 2.9635, 2.7199, 2.5207, 3.2220, 2.1357], device='cuda:1'), covar=tensor([0.0426, 0.0188, 0.0198, 0.0202, 0.0380, 0.0370, 0.0173, 0.0500], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0164, 0.0167, 0.0189, 0.0203, 0.0201, 0.0173, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:08:04,720 INFO [finetune.py:992] (1/2) Epoch 14, batch 7150, loss[loss=0.1743, simple_loss=0.2672, pruned_loss=0.0407, over 11680.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2564, pruned_loss=0.03877, over 2376760.18 frames. ], batch size: 48, lr: 3.58e-03, grad_scale: 32.0 2023-05-17 01:08:21,313 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266214.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:08:29,736 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266226.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:08:31,093 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266228.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:08:35,280 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266234.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:08:40,995 INFO [finetune.py:992] (1/2) Epoch 14, batch 7200, loss[loss=0.1926, simple_loss=0.2806, pruned_loss=0.05235, over 12137.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2565, pruned_loss=0.03874, over 2374973.14 frames. ], batch size: 34, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:09:01,288 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266270.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:09:03,380 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266273.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:09:05,009 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266275.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:09:09,044 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.507e+02 2.959e+02 3.672e+02 9.427e+02, threshold=5.917e+02, percent-clipped=2.0 2023-05-17 01:09:13,466 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266287.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 01:09:16,673 INFO [finetune.py:992] (1/2) Epoch 14, batch 7250, loss[loss=0.1557, simple_loss=0.2515, pruned_loss=0.02991, over 12119.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2564, pruned_loss=0.0387, over 2373632.54 frames. ], batch size: 33, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:09:28,110 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0771, 3.8841, 4.0386, 3.6787, 3.8742, 3.6970, 4.0406, 3.5281], device='cuda:1'), covar=tensor([0.0379, 0.0410, 0.0362, 0.0290, 0.0413, 0.0397, 0.0322, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0277, 0.0304, 0.0278, 0.0275, 0.0277, 0.0252, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:09:35,899 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266318.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:09:53,357 INFO [finetune.py:992] (1/2) Epoch 14, batch 7300, loss[loss=0.1718, simple_loss=0.2628, pruned_loss=0.04041, over 11648.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2583, pruned_loss=0.03972, over 2351248.15 frames. ], batch size: 48, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:09:56,283 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266346.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:10:07,708 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3468, 4.9501, 5.3197, 4.7066, 5.0002, 4.7295, 5.3753, 5.0642], device='cuda:1'), covar=tensor([0.0280, 0.0364, 0.0286, 0.0259, 0.0342, 0.0334, 0.0196, 0.0300], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0277, 0.0303, 0.0277, 0.0275, 0.0276, 0.0251, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:10:13,792 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266371.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:10:20,974 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.881e+02 3.416e+02 4.291e+02 1.034e+03, threshold=6.831e+02, percent-clipped=8.0 2023-05-17 01:10:21,106 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0014, 5.8300, 5.3683, 5.4340, 5.9330, 5.2639, 5.4192, 5.4087], device='cuda:1'), covar=tensor([0.1351, 0.0848, 0.0970, 0.1877, 0.0803, 0.2201, 0.1727, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0512, 0.0407, 0.0462, 0.0478, 0.0442, 0.0410, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:10:28,854 INFO [finetune.py:992] (1/2) Epoch 14, batch 7350, loss[loss=0.1622, simple_loss=0.2577, pruned_loss=0.03337, over 12055.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2571, pruned_loss=0.03911, over 2360105.58 frames. ], batch size: 40, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:10:33,438 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0915, 4.9019, 4.8723, 4.9486, 4.4757, 5.0722, 5.0443, 5.1902], device='cuda:1'), covar=tensor([0.0223, 0.0166, 0.0180, 0.0320, 0.0826, 0.0272, 0.0148, 0.0197], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0205, 0.0198, 0.0255, 0.0250, 0.0226, 0.0183, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 01:10:40,081 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266407.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:10:48,462 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266419.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:11:05,617 INFO [finetune.py:992] (1/2) Epoch 14, batch 7400, loss[loss=0.1713, simple_loss=0.26, pruned_loss=0.04135, over 11810.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2564, pruned_loss=0.03873, over 2372532.07 frames. ], batch size: 44, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:11:09,440 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6310, 2.9096, 4.4134, 4.4972, 2.8149, 2.5708, 2.9573, 2.1917], device='cuda:1'), covar=tensor([0.1735, 0.2926, 0.0501, 0.0462, 0.1391, 0.2553, 0.2625, 0.4122], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0393, 0.0279, 0.0304, 0.0275, 0.0313, 0.0392, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:11:21,612 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266464.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:11:34,131 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.530e+02 2.995e+02 3.621e+02 5.505e+02, threshold=5.991e+02, percent-clipped=0.0 2023-05-17 01:11:41,972 INFO [finetune.py:992] (1/2) Epoch 14, batch 7450, loss[loss=0.1573, simple_loss=0.254, pruned_loss=0.03025, over 12145.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2559, pruned_loss=0.03858, over 2375305.16 frames. ], batch size: 36, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:12:05,585 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266525.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:12:07,670 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266528.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:12:16,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-17 01:12:17,577 INFO [finetune.py:992] (1/2) Epoch 14, batch 7500, loss[loss=0.185, simple_loss=0.2706, pruned_loss=0.04969, over 12080.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2569, pruned_loss=0.0391, over 2370610.39 frames. ], batch size: 32, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:12:27,060 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4819, 5.2484, 5.4297, 5.4336, 4.9001, 4.9593, 4.9608, 5.2541], device='cuda:1'), covar=tensor([0.0889, 0.0736, 0.0964, 0.0783, 0.2579, 0.1761, 0.0627, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0709, 0.0630, 0.0650, 0.0874, 0.0760, 0.0578, 0.0489], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:12:37,905 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266570.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:12:40,132 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266573.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:12:42,904 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266576.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:12:46,355 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.645e+02 3.187e+02 3.673e+02 5.507e+02, threshold=6.374e+02, percent-clipped=0.0 2023-05-17 01:12:47,105 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266582.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:12:54,152 INFO [finetune.py:992] (1/2) Epoch 14, batch 7550, loss[loss=0.1796, simple_loss=0.2699, pruned_loss=0.04467, over 12055.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2561, pruned_loss=0.0389, over 2376642.89 frames. ], batch size: 37, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:13:15,489 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266621.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:13:26,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4033, 4.8363, 3.0013, 2.7174, 4.1608, 2.8082, 4.0972, 3.3228], device='cuda:1'), covar=tensor([0.0707, 0.0446, 0.1117, 0.1541, 0.0299, 0.1195, 0.0448, 0.0738], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0261, 0.0179, 0.0203, 0.0144, 0.0184, 0.0201, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:13:30,101 INFO [finetune.py:992] (1/2) Epoch 14, batch 7600, loss[loss=0.1543, simple_loss=0.249, pruned_loss=0.0298, over 12351.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.257, pruned_loss=0.03931, over 2375656.67 frames. ], batch size: 35, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:13:56,682 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266679.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:13:57,950 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.549e+02 3.060e+02 3.704e+02 6.506e+02, threshold=6.120e+02, percent-clipped=2.0 2023-05-17 01:14:05,915 INFO [finetune.py:992] (1/2) Epoch 14, batch 7650, loss[loss=0.1632, simple_loss=0.2556, pruned_loss=0.03533, over 12340.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2566, pruned_loss=0.03881, over 2378851.11 frames. ], batch size: 36, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:14:13,129 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266702.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:14:14,655 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266704.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:14:24,683 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4449, 3.1000, 4.9901, 2.4533, 2.7392, 3.6729, 3.0506, 3.6438], device='cuda:1'), covar=tensor([0.0574, 0.1382, 0.0335, 0.1365, 0.1970, 0.1545, 0.1523, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0240, 0.0257, 0.0185, 0.0240, 0.0299, 0.0229, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:14:41,121 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266740.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:14:42,291 INFO [finetune.py:992] (1/2) Epoch 14, batch 7700, loss[loss=0.2298, simple_loss=0.3059, pruned_loss=0.07684, over 10778.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2555, pruned_loss=0.03865, over 2372355.55 frames. ], batch size: 68, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:14:59,021 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266765.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:15:10,716 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.682e+02 3.206e+02 3.905e+02 6.001e+02, threshold=6.412e+02, percent-clipped=0.0 2023-05-17 01:15:18,624 INFO [finetune.py:992] (1/2) Epoch 14, batch 7750, loss[loss=0.149, simple_loss=0.2332, pruned_loss=0.0324, over 12180.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2556, pruned_loss=0.03889, over 2366245.73 frames. ], batch size: 29, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:15:38,962 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266820.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:15:41,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 01:15:54,839 INFO [finetune.py:992] (1/2) Epoch 14, batch 7800, loss[loss=0.1461, simple_loss=0.2242, pruned_loss=0.03405, over 11822.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2548, pruned_loss=0.03865, over 2365874.82 frames. ], batch size: 26, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:16:15,791 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266870.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:16:23,498 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.701e+02 2.578e+02 2.944e+02 3.565e+02 1.039e+03, threshold=5.888e+02, percent-clipped=3.0 2023-05-17 01:16:24,426 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266882.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:16:31,379 INFO [finetune.py:992] (1/2) Epoch 14, batch 7850, loss[loss=0.1619, simple_loss=0.2579, pruned_loss=0.03294, over 12283.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.255, pruned_loss=0.03869, over 2367726.99 frames. ], batch size: 37, lr: 3.57e-03, grad_scale: 32.0 2023-05-17 01:16:40,918 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4135, 3.0372, 4.7894, 2.4518, 2.5999, 3.5291, 2.9529, 3.5561], device='cuda:1'), covar=tensor([0.0486, 0.1309, 0.0348, 0.1283, 0.2010, 0.1655, 0.1519, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0239, 0.0256, 0.0184, 0.0239, 0.0298, 0.0228, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:16:45,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8178, 5.7998, 5.5555, 5.0091, 5.0593, 5.7003, 5.3495, 5.1103], device='cuda:1'), covar=tensor([0.0734, 0.0931, 0.0731, 0.1642, 0.0766, 0.0695, 0.1542, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0630, 0.0572, 0.0524, 0.0654, 0.0429, 0.0745, 0.0801, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:16:49,897 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266918.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:16:52,194 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266920.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:16:59,257 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266930.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 01:17:07,567 INFO [finetune.py:992] (1/2) Epoch 14, batch 7900, loss[loss=0.1495, simple_loss=0.239, pruned_loss=0.03002, over 12101.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2561, pruned_loss=0.03928, over 2358712.88 frames. ], batch size: 33, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:17:35,571 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266981.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:17:36,067 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.705e+02 3.150e+02 3.986e+02 7.942e+02, threshold=6.299e+02, percent-clipped=5.0 2023-05-17 01:17:43,130 INFO [finetune.py:992] (1/2) Epoch 14, batch 7950, loss[loss=0.2533, simple_loss=0.3195, pruned_loss=0.09356, over 8332.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2556, pruned_loss=0.03968, over 2357768.02 frames. ], batch size: 97, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:17:50,544 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267002.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:18:14,554 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267035.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:18:19,546 INFO [finetune.py:992] (1/2) Epoch 14, batch 8000, loss[loss=0.1577, simple_loss=0.2428, pruned_loss=0.03627, over 12338.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2559, pruned_loss=0.03945, over 2369462.66 frames. ], batch size: 31, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:18:25,335 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267050.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:18:32,315 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267060.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:18:48,579 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 2.680e+02 3.132e+02 3.791e+02 6.983e+02, threshold=6.264e+02, percent-clipped=3.0 2023-05-17 01:18:48,773 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267082.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:18:55,799 INFO [finetune.py:992] (1/2) Epoch 14, batch 8050, loss[loss=0.1777, simple_loss=0.2636, pruned_loss=0.04591, over 12156.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2567, pruned_loss=0.03958, over 2371744.42 frames. ], batch size: 36, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:19:00,313 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267098.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:19:03,196 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1158, 3.7311, 3.8859, 4.2814, 2.6371, 3.7235, 2.5140, 3.7396], device='cuda:1'), covar=tensor([0.1831, 0.0980, 0.0997, 0.0724, 0.1493, 0.0770, 0.2029, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0267, 0.0298, 0.0361, 0.0243, 0.0244, 0.0262, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:19:15,928 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267120.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:19:31,333 INFO [finetune.py:992] (1/2) Epoch 14, batch 8100, loss[loss=0.1446, simple_loss=0.2267, pruned_loss=0.03124, over 12352.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2564, pruned_loss=0.03932, over 2375202.27 frames. ], batch size: 31, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:19:32,309 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267143.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:19:44,516 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267159.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:19:50,786 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267168.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:19:58,182 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9582, 3.9085, 3.9078, 4.0068, 3.7667, 3.8298, 3.6875, 3.9153], device='cuda:1'), covar=tensor([0.1152, 0.0663, 0.1488, 0.0722, 0.1761, 0.1175, 0.0631, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0551, 0.0711, 0.0632, 0.0659, 0.0875, 0.0761, 0.0578, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:20:00,808 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.604e+02 3.178e+02 3.850e+02 8.583e+02, threshold=6.356e+02, percent-clipped=3.0 2023-05-17 01:20:07,922 INFO [finetune.py:992] (1/2) Epoch 14, batch 8150, loss[loss=0.1687, simple_loss=0.2713, pruned_loss=0.0331, over 12286.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2565, pruned_loss=0.03926, over 2374048.43 frames. ], batch size: 37, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:20:13,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-05-17 01:20:16,877 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1057, 4.6310, 5.0579, 4.4226, 4.7380, 4.4787, 5.1191, 4.7570], device='cuda:1'), covar=tensor([0.0287, 0.0480, 0.0338, 0.0300, 0.0397, 0.0383, 0.0219, 0.0413], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0275, 0.0300, 0.0275, 0.0272, 0.0273, 0.0249, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:20:19,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-17 01:20:28,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-05-17 01:20:34,867 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3044, 3.2653, 2.9938, 2.9136, 2.6600, 2.4560, 3.2046, 2.0796], device='cuda:1'), covar=tensor([0.0444, 0.0187, 0.0240, 0.0234, 0.0474, 0.0408, 0.0178, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0166, 0.0169, 0.0190, 0.0206, 0.0201, 0.0173, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:20:44,811 INFO [finetune.py:992] (1/2) Epoch 14, batch 8200, loss[loss=0.1698, simple_loss=0.2627, pruned_loss=0.03844, over 11661.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2569, pruned_loss=0.03908, over 2375152.87 frames. ], batch size: 48, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:20:59,570 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-17 01:21:08,998 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267276.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:21:13,151 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.546e+02 3.025e+02 3.497e+02 5.221e+02, threshold=6.051e+02, percent-clipped=0.0 2023-05-17 01:21:20,192 INFO [finetune.py:992] (1/2) Epoch 14, batch 8250, loss[loss=0.1922, simple_loss=0.2886, pruned_loss=0.04788, over 12146.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2566, pruned_loss=0.03908, over 2373290.79 frames. ], batch size: 36, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:21:51,614 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267335.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:21:56,314 INFO [finetune.py:992] (1/2) Epoch 14, batch 8300, loss[loss=0.1767, simple_loss=0.2754, pruned_loss=0.039, over 12282.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2567, pruned_loss=0.03896, over 2370348.20 frames. ], batch size: 37, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:22:07,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-17 01:22:10,036 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267360.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:22:25,568 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.047e+02 2.816e+02 3.145e+02 3.654e+02 7.270e+02, threshold=6.290e+02, percent-clipped=2.0 2023-05-17 01:22:26,379 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267383.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:22:32,162 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4517, 5.2787, 5.3881, 5.4210, 5.0742, 5.1239, 4.8245, 5.3724], device='cuda:1'), covar=tensor([0.0781, 0.0580, 0.0806, 0.0650, 0.1967, 0.1268, 0.0602, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0707, 0.0626, 0.0655, 0.0866, 0.0757, 0.0575, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:22:32,724 INFO [finetune.py:992] (1/2) Epoch 14, batch 8350, loss[loss=0.1712, simple_loss=0.2542, pruned_loss=0.04404, over 12100.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2567, pruned_loss=0.03905, over 2363946.37 frames. ], batch size: 32, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:22:44,707 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267408.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:23:04,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 01:23:05,955 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267438.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:23:08,748 INFO [finetune.py:992] (1/2) Epoch 14, batch 8400, loss[loss=0.179, simple_loss=0.2734, pruned_loss=0.04231, over 10504.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2571, pruned_loss=0.03943, over 2354576.79 frames. ], batch size: 68, lr: 3.57e-03, grad_scale: 16.0 2023-05-17 01:23:18,340 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267454.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:23:38,236 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.794e+02 3.288e+02 4.070e+02 9.921e+02, threshold=6.576e+02, percent-clipped=2.0 2023-05-17 01:23:44,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-17 01:23:45,277 INFO [finetune.py:992] (1/2) Epoch 14, batch 8450, loss[loss=0.1667, simple_loss=0.2597, pruned_loss=0.03686, over 12039.00 frames. ], tot_loss[loss=0.168, simple_loss=0.257, pruned_loss=0.03946, over 2359668.83 frames. ], batch size: 37, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:23:46,886 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267494.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:24:08,162 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1267, 6.0239, 5.8143, 5.3542, 5.2174, 6.0022, 5.5948, 5.3413], device='cuda:1'), covar=tensor([0.0722, 0.1020, 0.0674, 0.1718, 0.0655, 0.0670, 0.1471, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0631, 0.0569, 0.0525, 0.0651, 0.0427, 0.0742, 0.0798, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:24:21,399 INFO [finetune.py:992] (1/2) Epoch 14, batch 8500, loss[loss=0.1592, simple_loss=0.2454, pruned_loss=0.03649, over 12115.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2571, pruned_loss=0.03952, over 2361118.23 frames. ], batch size: 30, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:24:22,268 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0919, 6.0060, 5.7702, 5.2986, 5.2190, 5.9685, 5.5365, 5.3269], device='cuda:1'), covar=tensor([0.0744, 0.1075, 0.0714, 0.1694, 0.0672, 0.0656, 0.1546, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0570, 0.0526, 0.0652, 0.0428, 0.0745, 0.0800, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:24:30,926 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267555.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:24:37,453 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5256, 5.0492, 5.4740, 4.8282, 5.1181, 4.9292, 5.5269, 5.0943], device='cuda:1'), covar=tensor([0.0240, 0.0390, 0.0281, 0.0256, 0.0356, 0.0317, 0.0197, 0.0279], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0273, 0.0299, 0.0274, 0.0271, 0.0272, 0.0247, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:24:45,954 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267576.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:24:48,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2662, 5.2088, 5.0274, 4.5627, 4.7470, 5.2049, 4.8562, 4.6748], device='cuda:1'), covar=tensor([0.0754, 0.0950, 0.0703, 0.1625, 0.1167, 0.0745, 0.1521, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0565, 0.0521, 0.0648, 0.0426, 0.0739, 0.0795, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:24:50,065 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.604e+02 3.237e+02 3.803e+02 9.594e+02, threshold=6.474e+02, percent-clipped=1.0 2023-05-17 01:24:58,019 INFO [finetune.py:992] (1/2) Epoch 14, batch 8550, loss[loss=0.1611, simple_loss=0.2544, pruned_loss=0.03387, over 12015.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2559, pruned_loss=0.0391, over 2364424.03 frames. ], batch size: 42, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:25:20,932 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267624.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:25:33,771 INFO [finetune.py:992] (1/2) Epoch 14, batch 8600, loss[loss=0.141, simple_loss=0.2203, pruned_loss=0.03083, over 12190.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2558, pruned_loss=0.03899, over 2362339.27 frames. ], batch size: 29, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:25:40,305 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4242, 4.8271, 3.0454, 2.7728, 4.0918, 2.6395, 4.1227, 3.2553], device='cuda:1'), covar=tensor([0.0741, 0.0450, 0.1164, 0.1638, 0.0306, 0.1429, 0.0490, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0264, 0.0182, 0.0208, 0.0146, 0.0187, 0.0203, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:25:51,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 01:26:02,849 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.742e+02 3.259e+02 3.597e+02 7.182e+02, threshold=6.518e+02, percent-clipped=1.0 2023-05-17 01:26:09,985 INFO [finetune.py:992] (1/2) Epoch 14, batch 8650, loss[loss=0.1306, simple_loss=0.2159, pruned_loss=0.02265, over 12365.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2563, pruned_loss=0.03906, over 2367158.10 frames. ], batch size: 30, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:26:25,469 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8538, 3.2300, 2.4262, 2.1835, 2.8627, 2.2982, 3.1028, 2.6191], device='cuda:1'), covar=tensor([0.0716, 0.0729, 0.0914, 0.1453, 0.0326, 0.1179, 0.0575, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0262, 0.0181, 0.0206, 0.0146, 0.0186, 0.0202, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:26:43,795 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267738.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:26:46,473 INFO [finetune.py:992] (1/2) Epoch 14, batch 8700, loss[loss=0.1731, simple_loss=0.2621, pruned_loss=0.04204, over 12149.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2561, pruned_loss=0.03903, over 2367773.66 frames. ], batch size: 36, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:26:55,020 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267754.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:27:14,865 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.664e+02 3.049e+02 3.883e+02 6.983e+02, threshold=6.098e+02, percent-clipped=1.0 2023-05-17 01:27:17,698 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267786.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:27:21,745 INFO [finetune.py:992] (1/2) Epoch 14, batch 8750, loss[loss=0.1665, simple_loss=0.2554, pruned_loss=0.03882, over 12109.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2567, pruned_loss=0.03945, over 2364263.94 frames. ], batch size: 33, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:27:29,433 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267802.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:27:30,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-17 01:27:35,143 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6346, 2.7776, 3.6901, 4.5606, 3.9988, 4.7058, 3.9689, 3.3710], device='cuda:1'), covar=tensor([0.0048, 0.0368, 0.0155, 0.0049, 0.0111, 0.0068, 0.0120, 0.0326], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0124, 0.0106, 0.0080, 0.0104, 0.0118, 0.0097, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:27:58,321 INFO [finetune.py:992] (1/2) Epoch 14, batch 8800, loss[loss=0.1813, simple_loss=0.269, pruned_loss=0.04684, over 12310.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2571, pruned_loss=0.0399, over 2359859.55 frames. ], batch size: 34, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:28:04,139 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267850.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:28:21,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-05-17 01:28:27,482 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.878e+02 3.369e+02 4.005e+02 1.931e+03, threshold=6.738e+02, percent-clipped=8.0 2023-05-17 01:28:34,583 INFO [finetune.py:992] (1/2) Epoch 14, batch 8850, loss[loss=0.193, simple_loss=0.2817, pruned_loss=0.05214, over 12163.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2572, pruned_loss=0.03961, over 2362884.69 frames. ], batch size: 36, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:28:49,924 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3671, 3.5968, 3.2149, 3.1431, 2.8759, 2.6457, 3.6450, 2.2825], device='cuda:1'), covar=tensor([0.0472, 0.0150, 0.0223, 0.0204, 0.0453, 0.0425, 0.0140, 0.0552], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0167, 0.0169, 0.0191, 0.0206, 0.0203, 0.0175, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:29:10,542 INFO [finetune.py:992] (1/2) Epoch 14, batch 8900, loss[loss=0.1672, simple_loss=0.2565, pruned_loss=0.03894, over 12083.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2564, pruned_loss=0.03904, over 2365095.62 frames. ], batch size: 32, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:29:21,260 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9165, 5.8217, 5.5937, 5.3981, 5.9972, 5.3590, 5.4037, 5.3643], device='cuda:1'), covar=tensor([0.1688, 0.1023, 0.1205, 0.1968, 0.0850, 0.2198, 0.2021, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0506, 0.0404, 0.0454, 0.0472, 0.0436, 0.0405, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:29:38,322 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9420, 4.7702, 4.9506, 4.9102, 4.4773, 4.4348, 4.3754, 4.7946], device='cuda:1'), covar=tensor([0.1112, 0.0782, 0.1106, 0.0935, 0.2419, 0.1894, 0.0755, 0.1372], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0720, 0.0635, 0.0662, 0.0875, 0.0766, 0.0580, 0.0496], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:29:39,548 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.749e+02 3.348e+02 3.854e+02 9.928e+02, threshold=6.695e+02, percent-clipped=3.0 2023-05-17 01:29:46,008 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3308, 5.1467, 5.2893, 5.2821, 4.9619, 4.9212, 4.7394, 5.2451], device='cuda:1'), covar=tensor([0.0758, 0.0606, 0.0905, 0.0682, 0.1805, 0.1469, 0.0597, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0718, 0.0634, 0.0660, 0.0873, 0.0764, 0.0579, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 01:29:46,580 INFO [finetune.py:992] (1/2) Epoch 14, batch 8950, loss[loss=0.1545, simple_loss=0.2555, pruned_loss=0.02677, over 12169.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2572, pruned_loss=0.03924, over 2359884.23 frames. ], batch size: 36, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:30:21,442 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2545, 2.3919, 3.0068, 4.0505, 2.1951, 4.1685, 4.1658, 4.2829], device='cuda:1'), covar=tensor([0.0150, 0.1353, 0.0566, 0.0175, 0.1435, 0.0300, 0.0224, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0205, 0.0186, 0.0123, 0.0193, 0.0183, 0.0178, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:30:24,414 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9495, 3.5296, 5.2906, 2.8472, 3.0026, 3.7886, 3.2766, 3.8590], device='cuda:1'), covar=tensor([0.0404, 0.1068, 0.0307, 0.1083, 0.1876, 0.1617, 0.1374, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0238, 0.0256, 0.0183, 0.0238, 0.0298, 0.0225, 0.0267], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:30:26,259 INFO [finetune.py:992] (1/2) Epoch 14, batch 9000, loss[loss=0.1983, simple_loss=0.2867, pruned_loss=0.05492, over 10550.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2569, pruned_loss=0.0392, over 2358471.80 frames. ], batch size: 68, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:30:26,259 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 01:30:44,386 INFO [finetune.py:1026] (1/2) Epoch 14, validation: loss=0.3235, simple_loss=0.3964, pruned_loss=0.1253, over 1020973.00 frames. 2023-05-17 01:30:44,387 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 01:31:13,644 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.652e+02 3.054e+02 3.579e+02 6.577e+02, threshold=6.108e+02, percent-clipped=0.0 2023-05-17 01:31:20,737 INFO [finetune.py:992] (1/2) Epoch 14, batch 9050, loss[loss=0.1791, simple_loss=0.2757, pruned_loss=0.04132, over 12348.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2568, pruned_loss=0.03908, over 2369210.49 frames. ], batch size: 36, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:31:27,383 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0442, 4.0869, 3.9884, 4.3984, 3.0827, 4.0485, 2.8550, 4.0686], device='cuda:1'), covar=tensor([0.1699, 0.0656, 0.0876, 0.0616, 0.1114, 0.0555, 0.1592, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0269, 0.0302, 0.0366, 0.0245, 0.0245, 0.0264, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:31:56,834 INFO [finetune.py:992] (1/2) Epoch 14, batch 9100, loss[loss=0.1627, simple_loss=0.256, pruned_loss=0.03473, over 12372.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.257, pruned_loss=0.03929, over 2354107.82 frames. ], batch size: 38, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:32:02,662 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268150.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:32:02,699 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4612, 2.5113, 3.0289, 4.2696, 2.2387, 4.3535, 4.3880, 4.4957], device='cuda:1'), covar=tensor([0.0152, 0.1399, 0.0563, 0.0159, 0.1476, 0.0275, 0.0162, 0.0127], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0204, 0.0186, 0.0123, 0.0193, 0.0182, 0.0178, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:32:02,757 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6886, 3.3030, 5.0856, 2.6508, 2.7148, 3.6713, 3.0391, 3.8050], device='cuda:1'), covar=tensor([0.0492, 0.1237, 0.0323, 0.1174, 0.2090, 0.1746, 0.1535, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0238, 0.0256, 0.0183, 0.0238, 0.0298, 0.0226, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:32:18,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 01:32:25,038 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.134e+02 2.594e+02 3.160e+02 3.768e+02 7.595e+02, threshold=6.320e+02, percent-clipped=1.0 2023-05-17 01:32:32,966 INFO [finetune.py:992] (1/2) Epoch 14, batch 9150, loss[loss=0.1513, simple_loss=0.2417, pruned_loss=0.0305, over 12290.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2569, pruned_loss=0.03893, over 2355727.41 frames. ], batch size: 33, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:32:37,230 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268198.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:32:47,613 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268212.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:33:08,906 INFO [finetune.py:992] (1/2) Epoch 14, batch 9200, loss[loss=0.1568, simple_loss=0.229, pruned_loss=0.0423, over 12336.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2573, pruned_loss=0.03926, over 2354803.46 frames. ], batch size: 30, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:33:29,603 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1969, 2.0694, 2.4439, 2.2431, 2.3904, 2.4306, 1.9689, 2.4239], device='cuda:1'), covar=tensor([0.0124, 0.0293, 0.0185, 0.0182, 0.0126, 0.0165, 0.0264, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0211, 0.0199, 0.0194, 0.0224, 0.0173, 0.0205, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:33:31,717 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268273.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:33:33,052 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268275.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:33:37,946 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.774e+02 2.664e+02 3.128e+02 3.850e+02 8.698e+02, threshold=6.257e+02, percent-clipped=2.0 2023-05-17 01:33:45,153 INFO [finetune.py:992] (1/2) Epoch 14, batch 9250, loss[loss=0.2481, simple_loss=0.3132, pruned_loss=0.09153, over 8258.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.257, pruned_loss=0.03937, over 2354160.07 frames. ], batch size: 97, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:33:59,040 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2612, 4.6159, 3.9977, 4.8220, 4.4046, 2.8675, 4.2291, 3.0339], device='cuda:1'), covar=tensor([0.0775, 0.0751, 0.1389, 0.0614, 0.1209, 0.1698, 0.0978, 0.3144], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0384, 0.0363, 0.0325, 0.0373, 0.0275, 0.0354, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:34:17,509 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268336.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 01:34:21,590 INFO [finetune.py:992] (1/2) Epoch 14, batch 9300, loss[loss=0.1606, simple_loss=0.2621, pruned_loss=0.02952, over 12299.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2577, pruned_loss=0.03916, over 2357612.78 frames. ], batch size: 34, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:34:49,813 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.778e+02 3.151e+02 3.808e+02 5.509e+02, threshold=6.302e+02, percent-clipped=0.0 2023-05-17 01:34:52,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1156, 4.8082, 4.9420, 4.9831, 4.8533, 4.9898, 4.9266, 2.7164], device='cuda:1'), covar=tensor([0.0106, 0.0073, 0.0086, 0.0058, 0.0048, 0.0113, 0.0071, 0.0768], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0082, 0.0085, 0.0076, 0.0063, 0.0096, 0.0084, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:34:56,965 INFO [finetune.py:992] (1/2) Epoch 14, batch 9350, loss[loss=0.1748, simple_loss=0.2699, pruned_loss=0.0398, over 12265.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2582, pruned_loss=0.03918, over 2362617.06 frames. ], batch size: 32, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:35:25,897 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0855, 3.6799, 5.3757, 2.9600, 3.0652, 3.9917, 3.4593, 3.9589], device='cuda:1'), covar=tensor([0.0388, 0.1006, 0.0221, 0.1126, 0.1898, 0.1608, 0.1282, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0235, 0.0253, 0.0182, 0.0235, 0.0295, 0.0224, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:35:33,560 INFO [finetune.py:992] (1/2) Epoch 14, batch 9400, loss[loss=0.1557, simple_loss=0.247, pruned_loss=0.0322, over 12137.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2575, pruned_loss=0.0389, over 2369001.62 frames. ], batch size: 39, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:36:02,121 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.584e+02 2.948e+02 3.482e+02 7.106e+02, threshold=5.896e+02, percent-clipped=2.0 2023-05-17 01:36:09,260 INFO [finetune.py:992] (1/2) Epoch 14, batch 9450, loss[loss=0.1477, simple_loss=0.2365, pruned_loss=0.02941, over 12105.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2566, pruned_loss=0.03829, over 2375204.68 frames. ], batch size: 33, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:36:35,217 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6192, 2.5114, 3.2615, 4.4325, 2.4091, 4.5631, 4.5538, 4.6723], device='cuda:1'), covar=tensor([0.0120, 0.1275, 0.0462, 0.0152, 0.1310, 0.0201, 0.0175, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0203, 0.0184, 0.0122, 0.0191, 0.0180, 0.0177, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:36:45,614 INFO [finetune.py:992] (1/2) Epoch 14, batch 9500, loss[loss=0.1703, simple_loss=0.2571, pruned_loss=0.04181, over 11244.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.256, pruned_loss=0.0385, over 2371019.20 frames. ], batch size: 55, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:36:59,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.82 vs. limit=5.0 2023-05-17 01:37:04,953 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268568.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 01:37:14,969 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.701e+02 3.279e+02 3.811e+02 1.640e+03, threshold=6.558e+02, percent-clipped=2.0 2023-05-17 01:37:22,002 INFO [finetune.py:992] (1/2) Epoch 14, batch 9550, loss[loss=0.1616, simple_loss=0.2595, pruned_loss=0.03184, over 12351.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2555, pruned_loss=0.03831, over 2367998.02 frames. ], batch size: 36, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:37:28,780 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268601.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:37:49,829 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268631.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 01:37:58,107 INFO [finetune.py:992] (1/2) Epoch 14, batch 9600, loss[loss=0.1422, simple_loss=0.2281, pruned_loss=0.02812, over 12086.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2556, pruned_loss=0.03834, over 2374421.37 frames. ], batch size: 32, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:38:12,658 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268662.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:38:26,701 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 2.534e+02 3.172e+02 3.807e+02 1.348e+03, threshold=6.344e+02, percent-clipped=3.0 2023-05-17 01:38:33,751 INFO [finetune.py:992] (1/2) Epoch 14, batch 9650, loss[loss=0.1518, simple_loss=0.2408, pruned_loss=0.03139, over 12083.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2555, pruned_loss=0.03799, over 2376092.03 frames. ], batch size: 32, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:39:10,286 INFO [finetune.py:992] (1/2) Epoch 14, batch 9700, loss[loss=0.1531, simple_loss=0.2355, pruned_loss=0.03541, over 12243.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2563, pruned_loss=0.03848, over 2366308.81 frames. ], batch size: 32, lr: 3.56e-03, grad_scale: 16.0 2023-05-17 01:39:38,639 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.581e+02 3.284e+02 3.800e+02 7.476e+02, threshold=6.568e+02, percent-clipped=3.0 2023-05-17 01:39:46,392 INFO [finetune.py:992] (1/2) Epoch 14, batch 9750, loss[loss=0.1435, simple_loss=0.2302, pruned_loss=0.0284, over 11979.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2549, pruned_loss=0.03803, over 2368259.03 frames. ], batch size: 28, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:39:55,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-17 01:39:58,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.91 vs. limit=5.0 2023-05-17 01:40:06,079 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0010, 5.7801, 5.4193, 5.2766, 5.9177, 5.1794, 5.3337, 5.4169], device='cuda:1'), covar=tensor([0.1495, 0.0938, 0.1136, 0.2183, 0.0957, 0.2515, 0.1833, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0505, 0.0406, 0.0458, 0.0473, 0.0436, 0.0408, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:40:22,327 INFO [finetune.py:992] (1/2) Epoch 14, batch 9800, loss[loss=0.1853, simple_loss=0.2782, pruned_loss=0.0462, over 12355.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2558, pruned_loss=0.03788, over 2373463.82 frames. ], batch size: 36, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:40:23,856 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3012, 5.1505, 5.2848, 5.3016, 4.9229, 4.9342, 4.7080, 5.1998], device='cuda:1'), covar=tensor([0.0787, 0.0593, 0.0810, 0.0614, 0.1988, 0.1348, 0.0569, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0709, 0.0622, 0.0653, 0.0865, 0.0752, 0.0572, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:40:41,473 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268868.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 01:40:46,636 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1588, 5.0744, 5.0072, 5.0409, 4.5905, 5.0889, 5.0814, 5.2869], device='cuda:1'), covar=tensor([0.0233, 0.0149, 0.0156, 0.0313, 0.0787, 0.0344, 0.0186, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0203, 0.0196, 0.0254, 0.0247, 0.0223, 0.0183, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 01:40:51,363 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.816e+02 3.389e+02 4.188e+02 5.492e+02, threshold=6.779e+02, percent-clipped=0.0 2023-05-17 01:40:58,420 INFO [finetune.py:992] (1/2) Epoch 14, batch 9850, loss[loss=0.1585, simple_loss=0.2514, pruned_loss=0.03277, over 12307.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2558, pruned_loss=0.03812, over 2372551.90 frames. ], batch size: 34, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:41:00,100 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268894.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:41:11,360 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2955, 5.1198, 5.2670, 5.2684, 4.9086, 4.9740, 4.7089, 5.2221], device='cuda:1'), covar=tensor([0.0688, 0.0631, 0.0797, 0.0624, 0.1996, 0.1339, 0.0568, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0709, 0.0622, 0.0654, 0.0864, 0.0753, 0.0572, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:41:15,623 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268916.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:41:27,170 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268931.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:41:34,908 INFO [finetune.py:992] (1/2) Epoch 14, batch 9900, loss[loss=0.1448, simple_loss=0.2284, pruned_loss=0.03057, over 11753.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2556, pruned_loss=0.03781, over 2379746.71 frames. ], batch size: 26, lr: 3.55e-03, grad_scale: 32.0 2023-05-17 01:41:35,124 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3796, 2.3516, 3.1070, 4.2254, 2.2246, 4.2997, 4.3515, 4.3886], device='cuda:1'), covar=tensor([0.0135, 0.1310, 0.0536, 0.0144, 0.1382, 0.0250, 0.0159, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0206, 0.0186, 0.0123, 0.0193, 0.0182, 0.0178, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:41:44,379 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268955.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:41:45,600 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268957.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:42:00,982 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268979.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:42:03,056 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.743e+02 3.231e+02 3.791e+02 6.341e+02, threshold=6.462e+02, percent-clipped=0.0 2023-05-17 01:42:10,415 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268991.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:42:10,926 INFO [finetune.py:992] (1/2) Epoch 14, batch 9950, loss[loss=0.1545, simple_loss=0.2468, pruned_loss=0.03114, over 12159.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2563, pruned_loss=0.03823, over 2372573.86 frames. ], batch size: 34, lr: 3.55e-03, grad_scale: 32.0 2023-05-17 01:42:32,209 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0094, 4.9701, 4.8808, 4.9891, 4.5582, 5.0576, 4.9952, 5.2398], device='cuda:1'), covar=tensor([0.0255, 0.0158, 0.0175, 0.0312, 0.0773, 0.0360, 0.0183, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0205, 0.0198, 0.0256, 0.0248, 0.0224, 0.0185, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 01:42:47,284 INFO [finetune.py:992] (1/2) Epoch 14, batch 10000, loss[loss=0.128, simple_loss=0.213, pruned_loss=0.02147, over 12003.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2558, pruned_loss=0.03802, over 2373123.85 frames. ], batch size: 28, lr: 3.55e-03, grad_scale: 32.0 2023-05-17 01:42:54,608 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269052.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:43:16,323 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 2.649e+02 3.101e+02 3.484e+02 6.410e+02, threshold=6.202e+02, percent-clipped=0.0 2023-05-17 01:43:23,343 INFO [finetune.py:992] (1/2) Epoch 14, batch 10050, loss[loss=0.1641, simple_loss=0.2613, pruned_loss=0.03342, over 12342.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2554, pruned_loss=0.03776, over 2380317.64 frames. ], batch size: 36, lr: 3.55e-03, grad_scale: 32.0 2023-05-17 01:43:26,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-05-17 01:43:27,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2678, 4.2247, 4.0745, 4.4547, 2.9380, 4.0186, 2.6401, 4.1655], device='cuda:1'), covar=tensor([0.1617, 0.0655, 0.0953, 0.0518, 0.1198, 0.0596, 0.1930, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0270, 0.0304, 0.0365, 0.0245, 0.0247, 0.0265, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:43:40,615 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9694, 2.5864, 3.5045, 2.9564, 3.3331, 3.1248, 2.5282, 3.4154], device='cuda:1'), covar=tensor([0.0179, 0.0374, 0.0184, 0.0272, 0.0177, 0.0192, 0.0391, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0213, 0.0201, 0.0198, 0.0228, 0.0176, 0.0207, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:43:59,164 INFO [finetune.py:992] (1/2) Epoch 14, batch 10100, loss[loss=0.1781, simple_loss=0.2642, pruned_loss=0.04597, over 12278.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.255, pruned_loss=0.03764, over 2391068.46 frames. ], batch size: 37, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:44:15,573 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5269, 3.5851, 3.1666, 3.1810, 2.8962, 2.7687, 3.5794, 2.2053], device='cuda:1'), covar=tensor([0.0436, 0.0162, 0.0282, 0.0206, 0.0456, 0.0414, 0.0170, 0.0577], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0168, 0.0171, 0.0192, 0.0207, 0.0204, 0.0176, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:44:17,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9197, 4.6018, 4.7133, 4.8202, 4.6595, 4.8243, 4.7364, 2.3445], device='cuda:1'), covar=tensor([0.0141, 0.0079, 0.0096, 0.0067, 0.0057, 0.0110, 0.0111, 0.0953], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0081, 0.0085, 0.0075, 0.0062, 0.0095, 0.0084, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:44:28,214 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.305e+02 2.721e+02 3.314e+02 4.067e+02 1.296e+03, threshold=6.628e+02, percent-clipped=5.0 2023-05-17 01:44:34,833 INFO [finetune.py:992] (1/2) Epoch 14, batch 10150, loss[loss=0.1352, simple_loss=0.2219, pruned_loss=0.0243, over 12000.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2551, pruned_loss=0.03766, over 2387236.63 frames. ], batch size: 28, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:44:54,924 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-17 01:45:10,495 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269241.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:45:11,058 INFO [finetune.py:992] (1/2) Epoch 14, batch 10200, loss[loss=0.1357, simple_loss=0.2253, pruned_loss=0.02306, over 12257.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2549, pruned_loss=0.03767, over 2378421.85 frames. ], batch size: 32, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:45:13,444 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5656, 2.5968, 3.1013, 4.3929, 2.3514, 4.4423, 4.5723, 4.5988], device='cuda:1'), covar=tensor([0.0150, 0.1255, 0.0572, 0.0178, 0.1354, 0.0287, 0.0131, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0207, 0.0186, 0.0124, 0.0194, 0.0183, 0.0179, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:45:16,814 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269250.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:45:22,009 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269257.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:45:40,051 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.691e+02 3.133e+02 3.921e+02 6.696e+02, threshold=6.266e+02, percent-clipped=1.0 2023-05-17 01:45:47,106 INFO [finetune.py:992] (1/2) Epoch 14, batch 10250, loss[loss=0.1379, simple_loss=0.2314, pruned_loss=0.02217, over 12365.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2545, pruned_loss=0.03769, over 2378083.86 frames. ], batch size: 30, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:45:54,469 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269302.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:45:56,497 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269305.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:46:22,514 INFO [finetune.py:992] (1/2) Epoch 14, batch 10300, loss[loss=0.1685, simple_loss=0.2648, pruned_loss=0.03614, over 12282.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2554, pruned_loss=0.03825, over 2372525.71 frames. ], batch size: 33, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:46:26,215 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269347.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:46:27,824 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5010, 2.4524, 3.1510, 4.3650, 2.1541, 4.3664, 4.4656, 4.5234], device='cuda:1'), covar=tensor([0.0106, 0.1242, 0.0552, 0.0137, 0.1370, 0.0256, 0.0170, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0205, 0.0186, 0.0123, 0.0193, 0.0182, 0.0178, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:46:29,902 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269352.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:46:52,123 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.825e+02 3.136e+02 3.593e+02 1.295e+03, threshold=6.272e+02, percent-clipped=3.0 2023-05-17 01:46:58,677 INFO [finetune.py:992] (1/2) Epoch 14, batch 10350, loss[loss=0.194, simple_loss=0.283, pruned_loss=0.05249, over 11887.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2558, pruned_loss=0.03835, over 2377350.19 frames. ], batch size: 44, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:47:14,189 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269413.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:47:34,731 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269441.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:47:35,276 INFO [finetune.py:992] (1/2) Epoch 14, batch 10400, loss[loss=0.1611, simple_loss=0.2539, pruned_loss=0.03411, over 12056.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2563, pruned_loss=0.03854, over 2374259.35 frames. ], batch size: 40, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:48:04,620 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.686e+02 2.696e+02 3.077e+02 3.789e+02 7.478e+02, threshold=6.154e+02, percent-clipped=3.0 2023-05-17 01:48:09,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 01:48:11,119 INFO [finetune.py:992] (1/2) Epoch 14, batch 10450, loss[loss=0.1914, simple_loss=0.2826, pruned_loss=0.05012, over 8013.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2563, pruned_loss=0.03856, over 2369122.55 frames. ], batch size: 97, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:48:12,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-05-17 01:48:18,157 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269502.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:48:22,279 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1727, 4.6853, 5.1391, 4.4709, 4.7759, 4.5310, 5.1685, 4.8019], device='cuda:1'), covar=tensor([0.0273, 0.0462, 0.0310, 0.0261, 0.0444, 0.0402, 0.0230, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0274, 0.0301, 0.0273, 0.0273, 0.0273, 0.0249, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:48:47,285 INFO [finetune.py:992] (1/2) Epoch 14, batch 10500, loss[loss=0.1561, simple_loss=0.2499, pruned_loss=0.03116, over 12167.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2567, pruned_loss=0.03873, over 2365849.63 frames. ], batch size: 36, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:48:53,346 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269550.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:49:16,815 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9587, 3.5291, 5.3401, 2.5811, 2.9852, 3.8401, 3.3984, 3.7923], device='cuda:1'), covar=tensor([0.0373, 0.1042, 0.0286, 0.1199, 0.1747, 0.1453, 0.1312, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0239, 0.0259, 0.0186, 0.0239, 0.0300, 0.0229, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:49:17,246 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.582e+02 3.202e+02 3.705e+02 6.357e+02, threshold=6.404e+02, percent-clipped=1.0 2023-05-17 01:49:23,436 INFO [finetune.py:992] (1/2) Epoch 14, batch 10550, loss[loss=0.1724, simple_loss=0.2664, pruned_loss=0.03926, over 12167.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2577, pruned_loss=0.03951, over 2357984.83 frames. ], batch size: 36, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:49:27,106 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269597.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:49:27,810 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269598.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:49:58,826 INFO [finetune.py:992] (1/2) Epoch 14, batch 10600, loss[loss=0.1647, simple_loss=0.2449, pruned_loss=0.04219, over 12025.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2578, pruned_loss=0.03956, over 2366354.02 frames. ], batch size: 31, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:50:02,465 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269647.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:50:28,526 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.539e+02 3.181e+02 3.899e+02 5.844e+02, threshold=6.361e+02, percent-clipped=0.0 2023-05-17 01:50:35,004 INFO [finetune.py:992] (1/2) Epoch 14, batch 10650, loss[loss=0.171, simple_loss=0.261, pruned_loss=0.04052, over 12107.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2574, pruned_loss=0.03935, over 2368986.97 frames. ], batch size: 38, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:50:37,315 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269695.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:50:46,463 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269708.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:51:11,124 INFO [finetune.py:992] (1/2) Epoch 14, batch 10700, loss[loss=0.1505, simple_loss=0.2319, pruned_loss=0.03458, over 12339.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2565, pruned_loss=0.03889, over 2377326.19 frames. ], batch size: 30, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:51:25,843 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-17 01:51:33,364 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269773.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:51:40,220 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 2.824e+02 3.372e+02 3.895e+02 7.438e+02, threshold=6.744e+02, percent-clipped=2.0 2023-05-17 01:51:46,546 INFO [finetune.py:992] (1/2) Epoch 14, batch 10750, loss[loss=0.2305, simple_loss=0.3046, pruned_loss=0.07823, over 7492.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2558, pruned_loss=0.0387, over 2376489.81 frames. ], batch size: 97, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:51:50,138 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269797.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 01:52:04,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1338, 2.4697, 3.7293, 3.0341, 3.4951, 3.1656, 2.5747, 3.5913], device='cuda:1'), covar=tensor([0.0152, 0.0426, 0.0143, 0.0289, 0.0170, 0.0219, 0.0382, 0.0150], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0210, 0.0198, 0.0195, 0.0225, 0.0173, 0.0205, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:52:17,122 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269834.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 01:52:18,524 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269836.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:52:22,658 INFO [finetune.py:992] (1/2) Epoch 14, batch 10800, loss[loss=0.1706, simple_loss=0.2601, pruned_loss=0.04054, over 10486.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2566, pruned_loss=0.03923, over 2362733.91 frames. ], batch size: 68, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:52:52,311 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 2.900e+02 3.363e+02 3.885e+02 6.041e+02, threshold=6.726e+02, percent-clipped=0.0 2023-05-17 01:52:55,804 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0114, 5.8485, 5.5067, 5.3766, 5.9950, 5.3177, 5.3261, 5.4411], device='cuda:1'), covar=tensor([0.1480, 0.1092, 0.1123, 0.1932, 0.0945, 0.2180, 0.2287, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0512, 0.0411, 0.0467, 0.0480, 0.0446, 0.0411, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:52:58,604 INFO [finetune.py:992] (1/2) Epoch 14, batch 10850, loss[loss=0.1949, simple_loss=0.2772, pruned_loss=0.05628, over 12118.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2573, pruned_loss=0.03968, over 2363058.47 frames. ], batch size: 38, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:53:02,316 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269897.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:53:02,351 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269897.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:53:34,837 INFO [finetune.py:992] (1/2) Epoch 14, batch 10900, loss[loss=0.1662, simple_loss=0.2487, pruned_loss=0.04186, over 12182.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2571, pruned_loss=0.03939, over 2375212.23 frames. ], batch size: 31, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:53:37,036 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269945.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:53:45,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-17 01:54:04,646 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 2.784e+02 3.347e+02 3.934e+02 6.582e+02, threshold=6.694e+02, percent-clipped=0.0 2023-05-17 01:54:11,051 INFO [finetune.py:992] (1/2) Epoch 14, batch 10950, loss[loss=0.1745, simple_loss=0.2772, pruned_loss=0.03596, over 12299.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2583, pruned_loss=0.04032, over 2367514.19 frames. ], batch size: 34, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:54:14,976 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 01:54:20,810 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5503, 2.4177, 3.2136, 4.3510, 2.3151, 4.4708, 4.5820, 4.6180], device='cuda:1'), covar=tensor([0.0132, 0.1426, 0.0566, 0.0170, 0.1436, 0.0223, 0.0143, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0206, 0.0186, 0.0123, 0.0191, 0.0182, 0.0178, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:54:20,900 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7727, 2.8817, 4.7000, 4.7672, 2.9072, 2.7113, 2.9740, 2.1812], device='cuda:1'), covar=tensor([0.1581, 0.2981, 0.0457, 0.0418, 0.1358, 0.2414, 0.2660, 0.4030], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0394, 0.0281, 0.0308, 0.0277, 0.0316, 0.0396, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:54:25,662 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270008.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:54:49,719 INFO [finetune.py:992] (1/2) Epoch 14, batch 11000, loss[loss=0.1491, simple_loss=0.2332, pruned_loss=0.03249, over 12080.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2609, pruned_loss=0.04224, over 2323252.77 frames. ], batch size: 32, lr: 3.55e-03, grad_scale: 16.0 2023-05-17 01:54:51,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-17 01:54:59,731 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270056.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:55:02,006 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4490, 2.3350, 3.0618, 4.2693, 2.1875, 4.3671, 4.4400, 4.4751], device='cuda:1'), covar=tensor([0.0148, 0.1463, 0.0606, 0.0164, 0.1557, 0.0244, 0.0188, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0204, 0.0184, 0.0123, 0.0190, 0.0181, 0.0177, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 01:55:10,384 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1318, 2.1940, 2.8694, 2.9938, 3.0814, 3.1854, 2.9078, 2.3793], device='cuda:1'), covar=tensor([0.0086, 0.0393, 0.0215, 0.0098, 0.0135, 0.0093, 0.0150, 0.0409], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0127, 0.0108, 0.0080, 0.0106, 0.0119, 0.0100, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:55:19,150 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.921e+02 3.787e+02 4.330e+02 1.308e+03, threshold=7.573e+02, percent-clipped=3.0 2023-05-17 01:55:22,726 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3640, 5.2059, 5.2995, 5.3232, 5.0089, 4.9966, 4.8107, 5.2487], device='cuda:1'), covar=tensor([0.0661, 0.0545, 0.0866, 0.0571, 0.1783, 0.1369, 0.0525, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0706, 0.0619, 0.0645, 0.0852, 0.0743, 0.0570, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:55:25,266 INFO [finetune.py:992] (1/2) Epoch 14, batch 11050, loss[loss=0.228, simple_loss=0.3175, pruned_loss=0.06924, over 11799.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2636, pruned_loss=0.04322, over 2305468.90 frames. ], batch size: 44, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:55:28,936 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270097.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:55:51,973 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270129.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 01:56:00,741 INFO [finetune.py:992] (1/2) Epoch 14, batch 11100, loss[loss=0.2624, simple_loss=0.3439, pruned_loss=0.09041, over 10471.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2656, pruned_loss=0.04421, over 2269834.77 frames. ], batch size: 68, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:56:02,990 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270145.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:56:28,430 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-05-17 01:56:29,925 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 3.413e+02 4.097e+02 4.880e+02 1.017e+03, threshold=8.194e+02, percent-clipped=2.0 2023-05-17 01:56:30,911 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1137, 3.9405, 2.6975, 2.3890, 3.4786, 2.4653, 3.6673, 2.8065], device='cuda:1'), covar=tensor([0.0701, 0.0456, 0.1065, 0.1644, 0.0264, 0.1332, 0.0437, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0261, 0.0180, 0.0203, 0.0146, 0.0185, 0.0201, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 01:56:36,315 INFO [finetune.py:992] (1/2) Epoch 14, batch 11150, loss[loss=0.1675, simple_loss=0.259, pruned_loss=0.03796, over 11401.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2711, pruned_loss=0.04788, over 2199888.23 frames. ], batch size: 25, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:56:36,423 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270192.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:57:12,070 INFO [finetune.py:992] (1/2) Epoch 14, batch 11200, loss[loss=0.3312, simple_loss=0.3972, pruned_loss=0.1326, over 6916.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2782, pruned_loss=0.05263, over 2118251.10 frames. ], batch size: 98, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:57:40,362 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270280.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 01:57:42,321 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.206e+02 3.467e+02 4.075e+02 5.309e+02 8.859e+02, threshold=8.151e+02, percent-clipped=3.0 2023-05-17 01:57:48,255 INFO [finetune.py:992] (1/2) Epoch 14, batch 11250, loss[loss=0.2983, simple_loss=0.3614, pruned_loss=0.1176, over 6876.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2858, pruned_loss=0.05747, over 2063792.87 frames. ], batch size: 98, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:58:22,972 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270341.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 01:58:23,472 INFO [finetune.py:992] (1/2) Epoch 14, batch 11300, loss[loss=0.2373, simple_loss=0.3286, pruned_loss=0.07304, over 10294.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2921, pruned_loss=0.06114, over 2015168.18 frames. ], batch size: 68, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:58:52,088 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.411e+02 3.330e+02 3.992e+02 4.745e+02 8.371e+02, threshold=7.984e+02, percent-clipped=1.0 2023-05-17 01:58:57,993 INFO [finetune.py:992] (1/2) Epoch 14, batch 11350, loss[loss=0.2614, simple_loss=0.3336, pruned_loss=0.0946, over 7121.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2965, pruned_loss=0.06444, over 1950793.67 frames. ], batch size: 98, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:59:01,676 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270396.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:59:24,490 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270429.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 01:59:31,329 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270439.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:59:33,138 INFO [finetune.py:992] (1/2) Epoch 14, batch 11400, loss[loss=0.206, simple_loss=0.2972, pruned_loss=0.05733, over 11224.00 frames. ], tot_loss[loss=0.218, simple_loss=0.3009, pruned_loss=0.06759, over 1888895.51 frames. ], batch size: 55, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 01:59:43,646 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270457.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 01:59:49,826 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2610, 3.1208, 3.0386, 3.3209, 2.5911, 3.1538, 2.5866, 2.6976], device='cuda:1'), covar=tensor([0.1605, 0.0887, 0.0769, 0.0470, 0.0990, 0.0752, 0.1641, 0.0533], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0260, 0.0291, 0.0348, 0.0236, 0.0239, 0.0255, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 01:59:58,596 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270477.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:00:02,438 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.325e+02 3.507e+02 4.046e+02 4.865e+02 1.035e+03, threshold=8.092e+02, percent-clipped=3.0 2023-05-17 02:00:08,532 INFO [finetune.py:992] (1/2) Epoch 14, batch 11450, loss[loss=0.2141, simple_loss=0.2972, pruned_loss=0.06555, over 12126.00 frames. ], tot_loss[loss=0.223, simple_loss=0.305, pruned_loss=0.07054, over 1849845.07 frames. ], batch size: 38, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:00:08,659 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270492.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:00:14,197 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270500.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:00:42,140 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270540.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:00:43,340 INFO [finetune.py:992] (1/2) Epoch 14, batch 11500, loss[loss=0.2907, simple_loss=0.3483, pruned_loss=0.1165, over 7136.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3061, pruned_loss=0.0716, over 1835366.47 frames. ], batch size: 99, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:00:45,551 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1529, 2.5188, 3.6714, 4.3092, 4.0199, 4.2398, 3.9517, 2.8006], device='cuda:1'), covar=tensor([0.0053, 0.0445, 0.0130, 0.0045, 0.0105, 0.0091, 0.0102, 0.0499], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0123, 0.0104, 0.0077, 0.0102, 0.0114, 0.0097, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:00:58,922 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7712, 3.5232, 3.6435, 3.7761, 3.2901, 3.7188, 3.8463, 3.8122], device='cuda:1'), covar=tensor([0.0225, 0.0216, 0.0207, 0.0368, 0.0861, 0.0442, 0.0190, 0.0338], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0189, 0.0183, 0.0236, 0.0231, 0.0208, 0.0171, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 02:01:10,859 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.541e+02 3.383e+02 4.011e+02 4.670e+02 1.056e+03, threshold=8.021e+02, percent-clipped=3.0 2023-05-17 02:01:11,833 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8704, 3.7121, 3.8592, 3.6201, 3.7491, 3.6459, 3.7981, 3.5298], device='cuda:1'), covar=tensor([0.0364, 0.0399, 0.0380, 0.0263, 0.0397, 0.0330, 0.0414, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0262, 0.0287, 0.0262, 0.0262, 0.0260, 0.0238, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:01:17,531 INFO [finetune.py:992] (1/2) Epoch 14, batch 11550, loss[loss=0.2883, simple_loss=0.353, pruned_loss=0.1118, over 6587.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3084, pruned_loss=0.07372, over 1802199.61 frames. ], batch size: 97, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:01:32,775 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270613.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:01:34,103 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9933, 2.0673, 2.5711, 3.0154, 2.1804, 3.1255, 3.0157, 3.1488], device='cuda:1'), covar=tensor([0.0207, 0.1347, 0.0521, 0.0211, 0.1224, 0.0308, 0.0358, 0.0155], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0202, 0.0180, 0.0121, 0.0188, 0.0177, 0.0172, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:01:36,723 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7912, 3.0798, 2.4177, 2.2040, 2.7486, 2.2342, 2.9959, 2.5909], device='cuda:1'), covar=tensor([0.0535, 0.0370, 0.0864, 0.1284, 0.0216, 0.1124, 0.0403, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0254, 0.0178, 0.0201, 0.0143, 0.0184, 0.0197, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:01:47,829 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270636.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 02:01:48,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-05-17 02:01:51,872 INFO [finetune.py:992] (1/2) Epoch 14, batch 11600, loss[loss=0.2226, simple_loss=0.3081, pruned_loss=0.06857, over 11782.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3096, pruned_loss=0.07498, over 1775681.36 frames. ], batch size: 44, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:02:11,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 02:02:15,484 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270674.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:02:21,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.195e+02 3.446e+02 3.907e+02 4.532e+02 7.259e+02, threshold=7.814e+02, percent-clipped=0.0 2023-05-17 02:02:21,927 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4376, 4.7007, 2.9161, 2.0591, 4.1598, 2.2047, 4.0808, 2.9614], device='cuda:1'), covar=tensor([0.0679, 0.0445, 0.1289, 0.2624, 0.0248, 0.2050, 0.0380, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0254, 0.0178, 0.0202, 0.0143, 0.0183, 0.0197, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:02:28,184 INFO [finetune.py:992] (1/2) Epoch 14, batch 11650, loss[loss=0.2089, simple_loss=0.2961, pruned_loss=0.06083, over 11600.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.309, pruned_loss=0.07562, over 1756138.31 frames. ], batch size: 48, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:02:38,444 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8992, 4.3663, 3.8520, 4.7496, 4.1637, 2.8490, 4.0759, 2.8471], device='cuda:1'), covar=tensor([0.1028, 0.0911, 0.1420, 0.0459, 0.1323, 0.1922, 0.1122, 0.3648], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0360, 0.0340, 0.0300, 0.0349, 0.0261, 0.0330, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:02:53,053 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270726.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:03:03,571 INFO [finetune.py:992] (1/2) Epoch 14, batch 11700, loss[loss=0.195, simple_loss=0.2894, pruned_loss=0.05024, over 11121.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3083, pruned_loss=0.07602, over 1737210.77 frames. ], batch size: 55, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:03:07,671 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7205, 4.4012, 4.0811, 4.1132, 4.4696, 3.8144, 4.0922, 3.9619], device='cuda:1'), covar=tensor([0.1689, 0.1129, 0.1534, 0.1790, 0.1131, 0.2264, 0.1701, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0494, 0.0399, 0.0447, 0.0463, 0.0427, 0.0390, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:03:10,410 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270752.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:03:32,506 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.544e+02 3.420e+02 4.029e+02 4.880e+02 9.418e+02, threshold=8.057e+02, percent-clipped=2.0 2023-05-17 02:03:35,488 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270787.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:03:38,565 INFO [finetune.py:992] (1/2) Epoch 14, batch 11750, loss[loss=0.2415, simple_loss=0.3034, pruned_loss=0.08986, over 6541.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3086, pruned_loss=0.07628, over 1736104.69 frames. ], batch size: 97, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:03:40,734 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270795.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:04:14,131 INFO [finetune.py:992] (1/2) Epoch 14, batch 11800, loss[loss=0.2331, simple_loss=0.3118, pruned_loss=0.07718, over 6600.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3113, pruned_loss=0.07867, over 1703696.53 frames. ], batch size: 100, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:04:42,103 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 3.576e+02 4.134e+02 5.028e+02 8.984e+02, threshold=8.267e+02, percent-clipped=2.0 2023-05-17 02:04:48,186 INFO [finetune.py:992] (1/2) Epoch 14, batch 11850, loss[loss=0.2275, simple_loss=0.299, pruned_loss=0.07802, over 6945.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3122, pruned_loss=0.07888, over 1688820.78 frames. ], batch size: 104, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:05:06,738 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270917.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:05:19,565 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270936.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 02:05:20,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9346, 2.2561, 2.1363, 2.1320, 1.8882, 2.0049, 2.1377, 1.6061], device='cuda:1'), covar=tensor([0.0368, 0.0223, 0.0234, 0.0214, 0.0374, 0.0298, 0.0207, 0.0461], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0161, 0.0163, 0.0187, 0.0198, 0.0197, 0.0170, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:05:23,439 INFO [finetune.py:992] (1/2) Epoch 14, batch 11900, loss[loss=0.1846, simple_loss=0.2851, pruned_loss=0.04201, over 10223.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3114, pruned_loss=0.07781, over 1674146.24 frames. ], batch size: 68, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:05:26,654 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-17 02:05:43,130 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270969.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:05:49,353 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270978.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:05:52,084 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8159, 3.1379, 2.3675, 2.1612, 2.7975, 2.2712, 2.9519, 2.5951], device='cuda:1'), covar=tensor([0.0620, 0.0628, 0.1101, 0.1646, 0.0269, 0.1295, 0.0538, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0248, 0.0175, 0.0199, 0.0139, 0.0181, 0.0193, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:05:52,499 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.421e+02 3.376e+02 3.838e+02 4.530e+02 1.360e+03, threshold=7.675e+02, percent-clipped=2.0 2023-05-17 02:05:53,291 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270984.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 02:05:58,457 INFO [finetune.py:992] (1/2) Epoch 14, batch 11950, loss[loss=0.2008, simple_loss=0.2808, pruned_loss=0.06042, over 7036.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3087, pruned_loss=0.07517, over 1666584.07 frames. ], batch size: 98, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:06:22,787 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1022, 3.1474, 4.4649, 2.4881, 2.6217, 3.4037, 3.1590, 3.3909], device='cuda:1'), covar=tensor([0.0457, 0.1207, 0.0262, 0.1354, 0.2050, 0.1562, 0.1300, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0231, 0.0241, 0.0180, 0.0230, 0.0284, 0.0218, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:06:34,149 INFO [finetune.py:992] (1/2) Epoch 14, batch 12000, loss[loss=0.1943, simple_loss=0.2808, pruned_loss=0.05389, over 7038.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3034, pruned_loss=0.07102, over 1669286.84 frames. ], batch size: 98, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:06:34,149 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 02:06:52,783 INFO [finetune.py:1026] (1/2) Epoch 14, validation: loss=0.2855, simple_loss=0.361, pruned_loss=0.105, over 1020973.00 frames. 2023-05-17 02:06:52,784 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 02:06:59,505 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271052.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:07:17,708 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9473, 3.7586, 3.9239, 3.6489, 3.7974, 3.6765, 3.8962, 3.5933], device='cuda:1'), covar=tensor([0.0413, 0.0430, 0.0380, 0.0285, 0.0419, 0.0326, 0.0348, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0255, 0.0277, 0.0254, 0.0255, 0.0252, 0.0231, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:07:20,946 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271082.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:07:21,517 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.853e+02 3.352e+02 4.102e+02 7.378e+02, threshold=6.704e+02, percent-clipped=0.0 2023-05-17 02:07:27,502 INFO [finetune.py:992] (1/2) Epoch 14, batch 12050, loss[loss=0.207, simple_loss=0.2978, pruned_loss=0.05807, over 10596.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2994, pruned_loss=0.06804, over 1684534.61 frames. ], batch size: 70, lr: 3.54e-03, grad_scale: 16.0 2023-05-17 02:07:29,598 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271095.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:07:32,750 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271100.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:07:40,570 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7161, 3.4422, 3.6220, 3.7888, 3.4700, 3.7863, 3.8307, 3.8163], device='cuda:1'), covar=tensor([0.0210, 0.0184, 0.0174, 0.0270, 0.0482, 0.0322, 0.0164, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0180, 0.0174, 0.0224, 0.0220, 0.0199, 0.0163, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 02:08:00,811 INFO [finetune.py:992] (1/2) Epoch 14, batch 12100, loss[loss=0.197, simple_loss=0.2924, pruned_loss=0.05079, over 10323.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.298, pruned_loss=0.06623, over 1694390.86 frames. ], batch size: 68, lr: 3.54e-03, grad_scale: 32.0 2023-05-17 02:08:01,556 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271143.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:08:11,063 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-05-17 02:08:27,436 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.397e+02 3.236e+02 3.719e+02 4.278e+02 7.447e+02, threshold=7.437e+02, percent-clipped=1.0 2023-05-17 02:08:33,301 INFO [finetune.py:992] (1/2) Epoch 14, batch 12150, loss[loss=0.2128, simple_loss=0.3008, pruned_loss=0.06235, over 11783.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.298, pruned_loss=0.06637, over 1698532.17 frames. ], batch size: 44, lr: 3.54e-03, grad_scale: 32.0 2023-05-17 02:08:49,666 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5720, 4.8939, 3.0291, 2.6871, 4.3367, 2.7004, 4.2208, 3.3087], device='cuda:1'), covar=tensor([0.0667, 0.0399, 0.1194, 0.1692, 0.0210, 0.1417, 0.0369, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0242, 0.0172, 0.0196, 0.0136, 0.0178, 0.0189, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:09:05,325 INFO [finetune.py:992] (1/2) Epoch 14, batch 12200, loss[loss=0.1994, simple_loss=0.2762, pruned_loss=0.06127, over 7044.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2989, pruned_loss=0.06693, over 1689137.64 frames. ], batch size: 100, lr: 3.54e-03, grad_scale: 32.0 2023-05-17 02:09:08,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7328, 3.5270, 3.5607, 3.6241, 3.6705, 3.7720, 3.6640, 2.5795], device='cuda:1'), covar=tensor([0.0098, 0.0105, 0.0147, 0.0096, 0.0073, 0.0114, 0.0099, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0077, 0.0081, 0.0072, 0.0060, 0.0090, 0.0079, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:09:17,058 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271261.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:09:21,819 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271269.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:09:24,267 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271273.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:09:49,993 INFO [finetune.py:992] (1/2) Epoch 15, batch 0, loss[loss=0.1664, simple_loss=0.2566, pruned_loss=0.03805, over 12095.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2566, pruned_loss=0.03805, over 12095.00 frames. ], batch size: 32, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:09:49,993 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 02:10:00,842 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8248, 2.3845, 3.4556, 3.9845, 3.6663, 3.8618, 3.6818, 2.4363], device='cuda:1'), covar=tensor([0.0061, 0.0432, 0.0148, 0.0049, 0.0105, 0.0096, 0.0119, 0.0525], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0123, 0.0103, 0.0077, 0.0101, 0.0114, 0.0096, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:10:07,278 INFO [finetune.py:1026] (1/2) Epoch 15, validation: loss=0.287, simple_loss=0.3614, pruned_loss=0.1063, over 1020973.00 frames. 2023-05-17 02:10:07,279 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 02:10:11,515 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 3.235e+02 3.835e+02 4.619e+02 1.041e+03, threshold=7.669e+02, percent-clipped=4.0 2023-05-17 02:10:23,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-17 02:10:35,719 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271317.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:10:39,548 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271322.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:10:42,938 INFO [finetune.py:992] (1/2) Epoch 15, batch 50, loss[loss=0.1718, simple_loss=0.2641, pruned_loss=0.03972, over 12370.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2665, pruned_loss=0.04328, over 545495.90 frames. ], batch size: 35, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:11:02,651 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2298, 4.7762, 5.2097, 4.5289, 4.9089, 4.6447, 5.2343, 4.8964], device='cuda:1'), covar=tensor([0.0310, 0.0482, 0.0340, 0.0306, 0.0405, 0.0349, 0.0266, 0.0308], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0253, 0.0275, 0.0252, 0.0253, 0.0251, 0.0229, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:11:12,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-17 02:11:16,729 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271375.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:11:18,000 INFO [finetune.py:992] (1/2) Epoch 15, batch 100, loss[loss=0.1558, simple_loss=0.2485, pruned_loss=0.0316, over 12353.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2666, pruned_loss=0.0426, over 954918.82 frames. ], batch size: 30, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:11:22,266 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271382.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:11:22,824 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.781e+02 3.225e+02 3.711e+02 7.745e+02, threshold=6.450e+02, percent-clipped=1.0 2023-05-17 02:11:45,494 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7339, 2.9498, 4.7489, 4.9304, 2.8750, 2.5937, 3.0005, 2.1479], device='cuda:1'), covar=tensor([0.1763, 0.3443, 0.0459, 0.0428, 0.1461, 0.2808, 0.3041, 0.4672], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0385, 0.0272, 0.0297, 0.0269, 0.0310, 0.0391, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:11:54,551 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271426.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:11:55,131 INFO [finetune.py:992] (1/2) Epoch 15, batch 150, loss[loss=0.1835, simple_loss=0.2761, pruned_loss=0.04547, over 10482.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2627, pruned_loss=0.04123, over 1272820.56 frames. ], batch size: 69, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:11:57,425 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271430.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:12:01,755 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271436.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:12:24,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-05-17 02:12:30,907 INFO [finetune.py:992] (1/2) Epoch 15, batch 200, loss[loss=0.1742, simple_loss=0.2679, pruned_loss=0.04026, over 12048.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2597, pruned_loss=0.04046, over 1512332.50 frames. ], batch size: 40, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:12:35,269 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.755e+02 3.140e+02 3.726e+02 5.198e+02, threshold=6.280e+02, percent-clipped=0.0 2023-05-17 02:12:38,359 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271487.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:13:06,814 INFO [finetune.py:992] (1/2) Epoch 15, batch 250, loss[loss=0.1595, simple_loss=0.246, pruned_loss=0.03652, over 12102.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2582, pruned_loss=0.03943, over 1706184.65 frames. ], batch size: 32, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:13:20,540 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7291, 4.1174, 3.9357, 4.3767, 3.2908, 4.2127, 2.6793, 4.3842], device='cuda:1'), covar=tensor([0.1216, 0.0681, 0.1224, 0.0894, 0.1007, 0.0479, 0.1765, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0264, 0.0295, 0.0350, 0.0241, 0.0243, 0.0261, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:13:39,722 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1652, 2.3218, 3.6246, 3.1304, 3.4786, 3.1710, 2.4767, 3.5199], device='cuda:1'), covar=tensor([0.0165, 0.0478, 0.0195, 0.0287, 0.0187, 0.0231, 0.0492, 0.0172], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0203, 0.0186, 0.0184, 0.0213, 0.0165, 0.0196, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:13:41,084 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271573.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:13:42,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-17 02:13:43,946 INFO [finetune.py:992] (1/2) Epoch 15, batch 300, loss[loss=0.1822, simple_loss=0.2787, pruned_loss=0.04283, over 12151.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2573, pruned_loss=0.0389, over 1859056.02 frames. ], batch size: 36, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:13:48,293 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.692e+02 3.140e+02 3.693e+02 6.015e+02, threshold=6.279e+02, percent-clipped=0.0 2023-05-17 02:14:12,951 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271617.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:14:15,899 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271621.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:14:20,118 INFO [finetune.py:992] (1/2) Epoch 15, batch 350, loss[loss=0.1426, simple_loss=0.2264, pruned_loss=0.02941, over 12184.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2578, pruned_loss=0.03885, over 1975075.17 frames. ], batch size: 31, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:14:42,820 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6253, 3.3598, 5.0174, 2.5822, 2.7666, 3.7398, 3.2490, 3.7018], device='cuda:1'), covar=tensor([0.0438, 0.1189, 0.0367, 0.1302, 0.1976, 0.1595, 0.1325, 0.1307], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0236, 0.0247, 0.0184, 0.0235, 0.0291, 0.0222, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:14:46,358 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2424, 2.6477, 3.7842, 3.2029, 3.6166, 3.2612, 2.6928, 3.7087], device='cuda:1'), covar=tensor([0.0150, 0.0370, 0.0135, 0.0277, 0.0141, 0.0209, 0.0387, 0.0152], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0202, 0.0186, 0.0184, 0.0213, 0.0165, 0.0196, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:14:55,434 INFO [finetune.py:992] (1/2) Epoch 15, batch 400, loss[loss=0.1369, simple_loss=0.2226, pruned_loss=0.02562, over 12179.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2585, pruned_loss=0.03936, over 2062307.74 frames. ], batch size: 29, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:14:59,682 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.614e+02 3.141e+02 3.533e+02 5.574e+02, threshold=6.283e+02, percent-clipped=0.0 2023-05-17 02:15:26,041 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0083, 5.7816, 5.4032, 5.2911, 5.8499, 5.0840, 5.2943, 5.3227], device='cuda:1'), covar=tensor([0.1694, 0.0969, 0.1165, 0.2037, 0.0980, 0.2269, 0.1979, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0496, 0.0400, 0.0449, 0.0461, 0.0426, 0.0389, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:15:32,306 INFO [finetune.py:992] (1/2) Epoch 15, batch 450, loss[loss=0.1609, simple_loss=0.2392, pruned_loss=0.04127, over 12350.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2591, pruned_loss=0.0398, over 2118653.77 frames. ], batch size: 30, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:15:35,305 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271731.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:15:56,230 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1467, 4.5492, 4.0049, 4.8658, 4.3346, 2.8006, 4.1226, 2.9705], device='cuda:1'), covar=tensor([0.0935, 0.0742, 0.1600, 0.0489, 0.1206, 0.1897, 0.1111, 0.3295], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0375, 0.0357, 0.0314, 0.0364, 0.0274, 0.0345, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:15:58,962 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4480, 5.2965, 5.3902, 5.4226, 5.0713, 5.1129, 4.8666, 5.3656], device='cuda:1'), covar=tensor([0.0882, 0.0609, 0.0804, 0.0667, 0.2058, 0.1369, 0.0535, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0689, 0.0608, 0.0623, 0.0827, 0.0731, 0.0558, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:16:07,215 INFO [finetune.py:992] (1/2) Epoch 15, batch 500, loss[loss=0.174, simple_loss=0.2757, pruned_loss=0.03614, over 12349.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2586, pruned_loss=0.03915, over 2180559.15 frames. ], batch size: 36, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:16:08,184 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1960, 2.5552, 3.7812, 3.2446, 3.6057, 3.2633, 2.6435, 3.6214], device='cuda:1'), covar=tensor([0.0150, 0.0387, 0.0160, 0.0286, 0.0166, 0.0211, 0.0423, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0206, 0.0189, 0.0187, 0.0217, 0.0168, 0.0200, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:16:10,996 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271782.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:16:11,644 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.663e+02 3.335e+02 3.915e+02 7.207e+02, threshold=6.671e+02, percent-clipped=2.0 2023-05-17 02:16:31,266 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0341, 5.9537, 5.7514, 5.2897, 5.1393, 5.9063, 5.5002, 5.2502], device='cuda:1'), covar=tensor([0.0783, 0.1002, 0.0666, 0.1532, 0.0855, 0.0721, 0.1449, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0618, 0.0551, 0.0507, 0.0627, 0.0411, 0.0712, 0.0764, 0.0562], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-17 02:16:43,425 INFO [finetune.py:992] (1/2) Epoch 15, batch 550, loss[loss=0.1722, simple_loss=0.2653, pruned_loss=0.03953, over 12151.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2578, pruned_loss=0.03864, over 2232224.51 frames. ], batch size: 34, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:16:48,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 02:17:04,147 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271854.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:17:11,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 02:17:20,559 INFO [finetune.py:992] (1/2) Epoch 15, batch 600, loss[loss=0.1711, simple_loss=0.2619, pruned_loss=0.04013, over 12354.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2572, pruned_loss=0.03853, over 2271082.42 frames. ], batch size: 36, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:17:24,787 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.643e+02 2.962e+02 3.585e+02 7.418e+02, threshold=5.924e+02, percent-clipped=1.0 2023-05-17 02:17:47,580 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271915.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:17:48,972 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271917.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:17:56,063 INFO [finetune.py:992] (1/2) Epoch 15, batch 650, loss[loss=0.1665, simple_loss=0.2598, pruned_loss=0.03659, over 12347.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2573, pruned_loss=0.03883, over 2295106.71 frames. ], batch size: 35, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:18:01,206 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271934.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:18:13,246 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6078, 2.5398, 4.6110, 4.8398, 2.9858, 2.5224, 2.7773, 1.8629], device='cuda:1'), covar=tensor([0.1718, 0.3749, 0.0464, 0.0340, 0.1200, 0.2653, 0.3426, 0.5137], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0388, 0.0274, 0.0300, 0.0271, 0.0311, 0.0391, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:18:16,672 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2437, 4.5640, 2.8112, 2.7304, 4.0306, 2.5281, 3.8892, 3.1172], device='cuda:1'), covar=tensor([0.0832, 0.0596, 0.1243, 0.1565, 0.0268, 0.1536, 0.0574, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0255, 0.0180, 0.0203, 0.0142, 0.0185, 0.0198, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:18:22,925 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271965.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:18:31,322 INFO [finetune.py:992] (1/2) Epoch 15, batch 700, loss[loss=0.1463, simple_loss=0.2323, pruned_loss=0.03018, over 12350.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2572, pruned_loss=0.03885, over 2312330.78 frames. ], batch size: 30, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:18:36,820 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 2.488e+02 2.987e+02 3.609e+02 6.450e+02, threshold=5.974e+02, percent-clipped=2.0 2023-05-17 02:18:45,587 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271995.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:18:59,575 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1081, 4.5691, 3.9825, 4.8991, 4.2669, 2.5189, 4.0230, 2.9417], device='cuda:1'), covar=tensor([0.0908, 0.0782, 0.1584, 0.0452, 0.1350, 0.2051, 0.1156, 0.3370], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0377, 0.0360, 0.0317, 0.0366, 0.0276, 0.0348, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:19:11,342 INFO [finetune.py:992] (1/2) Epoch 15, batch 750, loss[loss=0.16, simple_loss=0.2536, pruned_loss=0.03318, over 12010.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2566, pruned_loss=0.0388, over 2320700.33 frames. ], batch size: 40, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:19:14,368 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272031.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:19:43,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.60 vs. limit=5.0 2023-05-17 02:19:47,241 INFO [finetune.py:992] (1/2) Epoch 15, batch 800, loss[loss=0.1601, simple_loss=0.2454, pruned_loss=0.03737, over 12417.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.256, pruned_loss=0.0384, over 2340071.51 frames. ], batch size: 32, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:19:48,791 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272079.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:19:50,953 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272082.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:19:51,537 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.501e+02 2.779e+02 3.240e+02 5.466e+02, threshold=5.558e+02, percent-clipped=0.0 2023-05-17 02:20:18,027 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2370, 3.7095, 3.6827, 4.1759, 2.7927, 3.7166, 2.4403, 3.7255], device='cuda:1'), covar=tensor([0.1686, 0.0811, 0.1014, 0.0753, 0.1286, 0.0645, 0.2031, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0270, 0.0300, 0.0359, 0.0246, 0.0247, 0.0264, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:20:23,528 INFO [finetune.py:992] (1/2) Epoch 15, batch 850, loss[loss=0.1839, simple_loss=0.2702, pruned_loss=0.04884, over 12146.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2555, pruned_loss=0.03796, over 2352149.43 frames. ], batch size: 39, lr: 3.53e-03, grad_scale: 32.0 2023-05-17 02:20:23,728 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272127.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:20:25,651 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272130.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:21:00,058 INFO [finetune.py:992] (1/2) Epoch 15, batch 900, loss[loss=0.1794, simple_loss=0.2697, pruned_loss=0.04458, over 12356.00 frames. ], tot_loss[loss=0.165, simple_loss=0.255, pruned_loss=0.03749, over 2364622.78 frames. ], batch size: 38, lr: 3.53e-03, grad_scale: 16.0 2023-05-17 02:21:02,982 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272181.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 02:21:04,680 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 02:21:04,983 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.001e+02 2.661e+02 3.135e+02 3.668e+02 6.921e+02, threshold=6.270e+02, percent-clipped=3.0 2023-05-17 02:21:08,147 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272188.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:21:17,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-17 02:21:24,220 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272210.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:21:33,542 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272223.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:21:36,198 INFO [finetune.py:992] (1/2) Epoch 15, batch 950, loss[loss=0.1868, simple_loss=0.2809, pruned_loss=0.04637, over 12069.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2556, pruned_loss=0.03767, over 2365451.92 frames. ], batch size: 42, lr: 3.53e-03, grad_scale: 16.0 2023-05-17 02:21:47,232 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272242.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 02:21:50,580 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9146, 4.7982, 4.6324, 4.7283, 4.4143, 4.8808, 4.8903, 5.0950], device='cuda:1'), covar=tensor([0.0289, 0.0176, 0.0249, 0.0369, 0.0815, 0.0317, 0.0161, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0195, 0.0189, 0.0243, 0.0238, 0.0215, 0.0175, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:21:50,629 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272247.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:22:08,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-17 02:22:12,398 INFO [finetune.py:992] (1/2) Epoch 15, batch 1000, loss[loss=0.172, simple_loss=0.2643, pruned_loss=0.03987, over 12182.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2556, pruned_loss=0.03778, over 2363612.69 frames. ], batch size: 31, lr: 3.53e-03, grad_scale: 16.0 2023-05-17 02:22:18,069 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.601e+02 3.157e+02 3.834e+02 6.545e+02, threshold=6.314e+02, percent-clipped=1.0 2023-05-17 02:22:18,269 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272284.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:22:22,476 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272290.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:22:35,490 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272308.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:22:46,896 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3384, 4.5716, 2.9597, 2.7485, 3.9885, 2.6673, 3.8464, 3.1681], device='cuda:1'), covar=tensor([0.0694, 0.0461, 0.1080, 0.1526, 0.0280, 0.1348, 0.0565, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0256, 0.0180, 0.0203, 0.0142, 0.0186, 0.0200, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:22:48,898 INFO [finetune.py:992] (1/2) Epoch 15, batch 1050, loss[loss=0.1631, simple_loss=0.2557, pruned_loss=0.0352, over 11601.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2541, pruned_loss=0.03738, over 2367951.79 frames. ], batch size: 48, lr: 3.53e-03, grad_scale: 16.0 2023-05-17 02:23:21,835 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272373.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:23:24,535 INFO [finetune.py:992] (1/2) Epoch 15, batch 1100, loss[loss=0.1734, simple_loss=0.2615, pruned_loss=0.04264, over 12293.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2543, pruned_loss=0.03747, over 2375069.02 frames. ], batch size: 33, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:23:29,570 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.645e+02 3.121e+02 3.561e+02 5.261e+02, threshold=6.242e+02, percent-clipped=0.0 2023-05-17 02:23:47,092 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7234, 5.7318, 5.4787, 4.9808, 5.0108, 5.6284, 5.2924, 5.0611], device='cuda:1'), covar=tensor([0.1035, 0.1051, 0.0783, 0.1908, 0.0820, 0.0846, 0.1693, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0562, 0.0513, 0.0635, 0.0418, 0.0721, 0.0774, 0.0568], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-17 02:24:01,152 INFO [finetune.py:992] (1/2) Epoch 15, batch 1150, loss[loss=0.1844, simple_loss=0.2735, pruned_loss=0.04768, over 12100.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2546, pruned_loss=0.03751, over 2373209.14 frames. ], batch size: 39, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:24:07,047 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272434.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:24:35,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 02:24:37,840 INFO [finetune.py:992] (1/2) Epoch 15, batch 1200, loss[loss=0.1813, simple_loss=0.2767, pruned_loss=0.04291, over 11874.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2543, pruned_loss=0.03738, over 2376335.94 frames. ], batch size: 44, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:24:42,205 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272483.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:24:42,806 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.776e+02 3.345e+02 4.010e+02 1.303e+03, threshold=6.691e+02, percent-clipped=6.0 2023-05-17 02:25:01,755 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272510.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:25:13,705 INFO [finetune.py:992] (1/2) Epoch 15, batch 1250, loss[loss=0.1591, simple_loss=0.2595, pruned_loss=0.02931, over 12068.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2552, pruned_loss=0.03738, over 2379084.73 frames. ], batch size: 42, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:25:20,937 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272537.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 02:25:33,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-17 02:25:35,706 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272558.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:25:39,467 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2494, 4.6965, 4.2151, 4.9671, 4.5961, 2.9113, 4.2308, 3.0629], device='cuda:1'), covar=tensor([0.0864, 0.0820, 0.1332, 0.0515, 0.1059, 0.1730, 0.0987, 0.3561], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0380, 0.0359, 0.0319, 0.0368, 0.0276, 0.0349, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:25:50,016 INFO [finetune.py:992] (1/2) Epoch 15, batch 1300, loss[loss=0.1638, simple_loss=0.263, pruned_loss=0.0323, over 12278.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2553, pruned_loss=0.03744, over 2379958.17 frames. ], batch size: 37, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:25:51,604 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272579.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:25:55,666 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.581e+02 3.049e+02 3.397e+02 7.875e+02, threshold=6.098e+02, percent-clipped=1.0 2023-05-17 02:26:00,034 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272590.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:26:09,219 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272603.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:26:16,398 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272613.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:26:24,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-17 02:26:26,350 INFO [finetune.py:992] (1/2) Epoch 15, batch 1350, loss[loss=0.1891, simple_loss=0.2811, pruned_loss=0.04852, over 12121.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2559, pruned_loss=0.03763, over 2383744.94 frames. ], batch size: 38, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:26:31,492 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2944, 4.6618, 4.0493, 4.9745, 4.4451, 2.8884, 4.2136, 2.9977], device='cuda:1'), covar=tensor([0.0876, 0.0883, 0.1507, 0.0495, 0.1325, 0.1751, 0.1171, 0.3419], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0380, 0.0359, 0.0320, 0.0369, 0.0276, 0.0349, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:26:34,273 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272638.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:26:37,121 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3982, 6.1300, 5.7306, 5.7250, 6.2977, 5.7632, 5.7784, 5.7662], device='cuda:1'), covar=tensor([0.1646, 0.1059, 0.1178, 0.1986, 0.0931, 0.2010, 0.1789, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0498, 0.0400, 0.0451, 0.0463, 0.0428, 0.0395, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:27:00,026 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272674.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:27:01,943 INFO [finetune.py:992] (1/2) Epoch 15, batch 1400, loss[loss=0.1765, simple_loss=0.2632, pruned_loss=0.0449, over 12154.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2558, pruned_loss=0.03753, over 2379381.94 frames. ], batch size: 36, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:27:07,032 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.607e+02 3.137e+02 3.946e+02 6.819e+02, threshold=6.274e+02, percent-clipped=2.0 2023-05-17 02:27:20,040 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5974, 4.4312, 4.3498, 4.5048, 4.1109, 4.6246, 4.5655, 4.7552], device='cuda:1'), covar=tensor([0.0322, 0.0206, 0.0261, 0.0375, 0.0916, 0.0429, 0.0200, 0.0222], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0198, 0.0193, 0.0247, 0.0242, 0.0219, 0.0178, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:27:38,492 INFO [finetune.py:992] (1/2) Epoch 15, batch 1450, loss[loss=0.1606, simple_loss=0.2564, pruned_loss=0.03235, over 12146.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2563, pruned_loss=0.0378, over 2380127.64 frames. ], batch size: 36, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:27:40,077 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272729.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:28:15,193 INFO [finetune.py:992] (1/2) Epoch 15, batch 1500, loss[loss=0.1665, simple_loss=0.2657, pruned_loss=0.03366, over 12378.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2556, pruned_loss=0.03775, over 2376972.34 frames. ], batch size: 36, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:28:19,521 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272783.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:28:20,071 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.683e+02 3.061e+02 3.641e+02 1.079e+03, threshold=6.122e+02, percent-clipped=1.0 2023-05-17 02:28:49,406 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4682, 4.7252, 3.0638, 2.7436, 4.2235, 2.5839, 4.0438, 3.4248], device='cuda:1'), covar=tensor([0.0772, 0.0728, 0.1178, 0.1628, 0.0276, 0.1409, 0.0504, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0262, 0.0183, 0.0206, 0.0145, 0.0187, 0.0203, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:28:51,261 INFO [finetune.py:992] (1/2) Epoch 15, batch 1550, loss[loss=0.1873, simple_loss=0.2848, pruned_loss=0.04489, over 12048.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2563, pruned_loss=0.0381, over 2379216.20 frames. ], batch size: 42, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:28:53,009 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6580, 2.7418, 4.6334, 4.7740, 2.8425, 2.6353, 2.9768, 2.1757], device='cuda:1'), covar=tensor([0.1811, 0.3701, 0.0472, 0.0451, 0.1393, 0.2541, 0.2944, 0.4247], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0390, 0.0275, 0.0302, 0.0273, 0.0312, 0.0392, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:28:54,175 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272831.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:28:58,381 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272837.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 02:29:04,196 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5381, 3.5140, 3.2003, 3.0804, 2.8447, 2.7389, 3.5303, 2.2885], device='cuda:1'), covar=tensor([0.0445, 0.0173, 0.0217, 0.0241, 0.0451, 0.0392, 0.0183, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0162, 0.0164, 0.0189, 0.0201, 0.0198, 0.0172, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:29:27,292 INFO [finetune.py:992] (1/2) Epoch 15, batch 1600, loss[loss=0.161, simple_loss=0.2475, pruned_loss=0.03724, over 12091.00 frames. ], tot_loss[loss=0.166, simple_loss=0.256, pruned_loss=0.038, over 2384071.97 frames. ], batch size: 32, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:29:28,892 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272879.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:29:32,361 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 2.828e+02 3.239e+02 3.866e+02 7.933e+02, threshold=6.477e+02, percent-clipped=1.0 2023-05-17 02:29:33,841 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272885.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 02:29:46,827 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272903.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:30:03,980 INFO [finetune.py:992] (1/2) Epoch 15, batch 1650, loss[loss=0.1678, simple_loss=0.2583, pruned_loss=0.03861, over 12373.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2566, pruned_loss=0.03836, over 2380031.77 frames. ], batch size: 38, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:30:04,050 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272927.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:30:16,405 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6159, 4.3471, 4.6171, 4.0793, 4.3956, 4.1243, 4.6095, 4.2628], device='cuda:1'), covar=tensor([0.0295, 0.0398, 0.0299, 0.0291, 0.0399, 0.0371, 0.0249, 0.0693], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0270, 0.0296, 0.0268, 0.0271, 0.0271, 0.0244, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:30:21,355 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272951.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:30:28,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-05-17 02:30:34,193 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272969.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:30:39,924 INFO [finetune.py:992] (1/2) Epoch 15, batch 1700, loss[loss=0.1828, simple_loss=0.2757, pruned_loss=0.04492, over 12136.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2556, pruned_loss=0.03795, over 2382482.77 frames. ], batch size: 39, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:30:44,983 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.674e+02 3.132e+02 3.940e+02 7.043e+02, threshold=6.265e+02, percent-clipped=2.0 2023-05-17 02:31:16,985 INFO [finetune.py:992] (1/2) Epoch 15, batch 1750, loss[loss=0.177, simple_loss=0.2676, pruned_loss=0.04322, over 12068.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2556, pruned_loss=0.03797, over 2382690.02 frames. ], batch size: 42, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:31:18,617 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273029.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:31:27,921 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0965, 5.8296, 5.4790, 5.4030, 5.9562, 5.2660, 5.3209, 5.4434], device='cuda:1'), covar=tensor([0.1563, 0.0935, 0.1153, 0.1853, 0.0911, 0.2139, 0.2017, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0503, 0.0401, 0.0454, 0.0466, 0.0436, 0.0398, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:31:53,343 INFO [finetune.py:992] (1/2) Epoch 15, batch 1800, loss[loss=0.1751, simple_loss=0.2664, pruned_loss=0.04189, over 12275.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2554, pruned_loss=0.03793, over 2379353.81 frames. ], batch size: 37, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:31:53,411 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273077.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:31:58,072 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.669e+02 3.049e+02 3.861e+02 7.728e+02, threshold=6.099e+02, percent-clipped=2.0 2023-05-17 02:32:28,571 INFO [finetune.py:992] (1/2) Epoch 15, batch 1850, loss[loss=0.1457, simple_loss=0.2291, pruned_loss=0.03115, over 12006.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2555, pruned_loss=0.03792, over 2388954.01 frames. ], batch size: 28, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:32:36,648 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1897, 2.1401, 2.6119, 3.1723, 2.2069, 3.2392, 3.1509, 3.2693], device='cuda:1'), covar=tensor([0.0198, 0.1304, 0.0622, 0.0191, 0.1228, 0.0362, 0.0403, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0207, 0.0185, 0.0121, 0.0192, 0.0182, 0.0178, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:33:04,609 INFO [finetune.py:992] (1/2) Epoch 15, batch 1900, loss[loss=0.1439, simple_loss=0.2301, pruned_loss=0.02887, over 12339.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2559, pruned_loss=0.0384, over 2374681.28 frames. ], batch size: 30, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:33:09,695 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0748, 2.4250, 2.3480, 2.2932, 2.0916, 2.0953, 2.3218, 1.7078], device='cuda:1'), covar=tensor([0.0312, 0.0193, 0.0226, 0.0193, 0.0370, 0.0285, 0.0207, 0.0422], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0162, 0.0164, 0.0189, 0.0202, 0.0199, 0.0172, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:33:10,120 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 2.728e+02 3.072e+02 3.546e+02 6.596e+02, threshold=6.144e+02, percent-clipped=0.0 2023-05-17 02:33:41,221 INFO [finetune.py:992] (1/2) Epoch 15, batch 1950, loss[loss=0.168, simple_loss=0.2665, pruned_loss=0.03473, over 11358.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2562, pruned_loss=0.0389, over 2359402.92 frames. ], batch size: 55, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:33:47,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-05-17 02:33:57,168 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4192, 3.5016, 3.2254, 3.0399, 2.7707, 2.6468, 3.6603, 2.0688], device='cuda:1'), covar=tensor([0.0491, 0.0181, 0.0222, 0.0247, 0.0516, 0.0509, 0.0141, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0163, 0.0165, 0.0189, 0.0203, 0.0200, 0.0173, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:34:01,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-17 02:34:05,405 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3458, 4.9945, 5.3671, 4.6308, 5.0173, 4.7290, 5.3696, 5.0830], device='cuda:1'), covar=tensor([0.0295, 0.0353, 0.0285, 0.0268, 0.0416, 0.0337, 0.0230, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0270, 0.0296, 0.0268, 0.0272, 0.0271, 0.0244, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:34:11,157 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273269.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:34:16,719 INFO [finetune.py:992] (1/2) Epoch 15, batch 2000, loss[loss=0.1624, simple_loss=0.2538, pruned_loss=0.03549, over 12107.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2559, pruned_loss=0.03836, over 2370326.38 frames. ], batch size: 39, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:34:21,753 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.866e+02 2.695e+02 3.081e+02 3.617e+02 1.680e+03, threshold=6.163e+02, percent-clipped=6.0 2023-05-17 02:34:42,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-05-17 02:34:45,402 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3313, 4.9359, 5.2365, 5.1471, 5.0830, 5.2656, 5.1803, 3.0432], device='cuda:1'), covar=tensor([0.0096, 0.0064, 0.0060, 0.0055, 0.0034, 0.0081, 0.0058, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0080, 0.0084, 0.0074, 0.0061, 0.0093, 0.0083, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:34:46,009 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273317.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:34:53,121 INFO [finetune.py:992] (1/2) Epoch 15, batch 2050, loss[loss=0.1794, simple_loss=0.271, pruned_loss=0.04394, over 12141.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2561, pruned_loss=0.03829, over 2376221.87 frames. ], batch size: 39, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:35:29,954 INFO [finetune.py:992] (1/2) Epoch 15, batch 2100, loss[loss=0.163, simple_loss=0.2625, pruned_loss=0.03175, over 12199.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2553, pruned_loss=0.03808, over 2381205.25 frames. ], batch size: 35, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:35:34,746 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.558e+02 2.830e+02 3.304e+02 5.961e+02, threshold=5.659e+02, percent-clipped=0.0 2023-05-17 02:35:52,862 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273409.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:36:05,473 INFO [finetune.py:992] (1/2) Epoch 15, batch 2150, loss[loss=0.1747, simple_loss=0.2589, pruned_loss=0.04521, over 12156.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2551, pruned_loss=0.03794, over 2384799.21 frames. ], batch size: 36, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:36:06,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 2023-05-17 02:36:08,683 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3541, 3.3952, 3.1214, 3.0327, 2.7239, 2.6767, 3.4807, 2.0825], device='cuda:1'), covar=tensor([0.0457, 0.0220, 0.0206, 0.0230, 0.0452, 0.0431, 0.0136, 0.0596], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0163, 0.0165, 0.0190, 0.0202, 0.0200, 0.0172, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:36:11,547 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273435.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:36:37,314 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273470.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:36:42,248 INFO [finetune.py:992] (1/2) Epoch 15, batch 2200, loss[loss=0.1562, simple_loss=0.2381, pruned_loss=0.03718, over 12141.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2541, pruned_loss=0.0374, over 2386316.24 frames. ], batch size: 30, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:36:47,157 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.612e+02 3.130e+02 3.605e+02 1.162e+03, threshold=6.261e+02, percent-clipped=3.0 2023-05-17 02:36:56,557 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273496.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:37:18,401 INFO [finetune.py:992] (1/2) Epoch 15, batch 2250, loss[loss=0.164, simple_loss=0.2579, pruned_loss=0.03506, over 12174.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2532, pruned_loss=0.03703, over 2388520.91 frames. ], batch size: 35, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:37:30,627 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5956, 4.4488, 4.4761, 4.5040, 4.1230, 4.6452, 4.6137, 4.7250], device='cuda:1'), covar=tensor([0.0284, 0.0193, 0.0216, 0.0399, 0.0852, 0.0350, 0.0196, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0201, 0.0195, 0.0252, 0.0245, 0.0223, 0.0182, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:37:35,662 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3320, 4.6400, 2.9017, 2.5855, 3.9992, 2.5413, 4.0022, 3.2302], device='cuda:1'), covar=tensor([0.0773, 0.0473, 0.1191, 0.1556, 0.0285, 0.1367, 0.0443, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0264, 0.0183, 0.0207, 0.0146, 0.0188, 0.0203, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:37:38,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 02:37:49,032 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2382, 5.2388, 5.0658, 4.6075, 4.6663, 5.1830, 4.8328, 4.6935], device='cuda:1'), covar=tensor([0.0835, 0.0978, 0.0669, 0.1540, 0.1389, 0.0796, 0.1583, 0.1048], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0569, 0.0524, 0.0647, 0.0427, 0.0740, 0.0798, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 02:37:53,939 INFO [finetune.py:992] (1/2) Epoch 15, batch 2300, loss[loss=0.139, simple_loss=0.2158, pruned_loss=0.03111, over 12338.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2538, pruned_loss=0.03723, over 2388293.12 frames. ], batch size: 30, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:37:58,828 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.471e+02 2.860e+02 3.398e+02 1.097e+03, threshold=5.720e+02, percent-clipped=1.0 2023-05-17 02:38:15,954 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 02:38:18,477 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9495, 4.8126, 4.8067, 4.8335, 4.4924, 4.9931, 4.9102, 5.1206], device='cuda:1'), covar=tensor([0.0243, 0.0173, 0.0175, 0.0400, 0.0772, 0.0273, 0.0164, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0201, 0.0195, 0.0253, 0.0245, 0.0223, 0.0182, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:38:30,515 INFO [finetune.py:992] (1/2) Epoch 15, batch 2350, loss[loss=0.1852, simple_loss=0.273, pruned_loss=0.04873, over 10560.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03775, over 2376339.79 frames. ], batch size: 69, lr: 3.52e-03, grad_scale: 16.0 2023-05-17 02:38:55,270 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4742, 5.2388, 5.3613, 5.4035, 5.0026, 5.0800, 4.8819, 5.4207], device='cuda:1'), covar=tensor([0.0768, 0.0713, 0.0839, 0.0735, 0.2050, 0.1277, 0.0529, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0720, 0.0630, 0.0651, 0.0871, 0.0765, 0.0577, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 02:39:06,587 INFO [finetune.py:992] (1/2) Epoch 15, batch 2400, loss[loss=0.1521, simple_loss=0.2383, pruned_loss=0.03293, over 12099.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2539, pruned_loss=0.03767, over 2381767.09 frames. ], batch size: 32, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:39:11,644 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.980e+02 2.708e+02 3.183e+02 3.628e+02 6.615e+02, threshold=6.367e+02, percent-clipped=2.0 2023-05-17 02:39:42,470 INFO [finetune.py:992] (1/2) Epoch 15, batch 2450, loss[loss=0.1759, simple_loss=0.2731, pruned_loss=0.03937, over 12348.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2542, pruned_loss=0.03749, over 2381677.52 frames. ], batch size: 35, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:40:10,207 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273765.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:40:18,793 INFO [finetune.py:992] (1/2) Epoch 15, batch 2500, loss[loss=0.1831, simple_loss=0.273, pruned_loss=0.04663, over 12051.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2545, pruned_loss=0.03754, over 2373219.20 frames. ], batch size: 42, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:40:24,380 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.648e+02 3.049e+02 3.649e+02 6.935e+02, threshold=6.097e+02, percent-clipped=1.0 2023-05-17 02:40:29,446 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273791.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:40:35,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 02:40:55,336 INFO [finetune.py:992] (1/2) Epoch 15, batch 2550, loss[loss=0.1578, simple_loss=0.2471, pruned_loss=0.03426, over 12388.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.254, pruned_loss=0.03717, over 2372950.13 frames. ], batch size: 38, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:41:20,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.05 vs. limit=5.0 2023-05-17 02:41:31,341 INFO [finetune.py:992] (1/2) Epoch 15, batch 2600, loss[loss=0.1817, simple_loss=0.2703, pruned_loss=0.04652, over 11225.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2535, pruned_loss=0.03721, over 2371510.95 frames. ], batch size: 55, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:41:36,387 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.651e+02 3.107e+02 3.567e+02 6.473e+02, threshold=6.214e+02, percent-clipped=1.0 2023-05-17 02:41:36,560 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3934, 5.2471, 5.3511, 5.3994, 5.0004, 5.0609, 4.8215, 5.3875], device='cuda:1'), covar=tensor([0.0788, 0.0630, 0.0879, 0.0602, 0.1875, 0.1249, 0.0509, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0558, 0.0723, 0.0636, 0.0655, 0.0876, 0.0770, 0.0582, 0.0496], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 02:42:07,964 INFO [finetune.py:992] (1/2) Epoch 15, batch 2650, loss[loss=0.1753, simple_loss=0.2635, pruned_loss=0.04349, over 11345.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2533, pruned_loss=0.0369, over 2374844.18 frames. ], batch size: 55, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:42:29,118 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2703, 4.9298, 5.3272, 4.6195, 5.0057, 4.6752, 5.3410, 4.9376], device='cuda:1'), covar=tensor([0.0308, 0.0367, 0.0247, 0.0272, 0.0376, 0.0385, 0.0198, 0.0318], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0270, 0.0296, 0.0269, 0.0272, 0.0271, 0.0244, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:42:43,087 INFO [finetune.py:992] (1/2) Epoch 15, batch 2700, loss[loss=0.1634, simple_loss=0.2459, pruned_loss=0.04042, over 12347.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2543, pruned_loss=0.03736, over 2370463.02 frames. ], batch size: 31, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:42:48,119 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 2.625e+02 3.117e+02 3.846e+02 1.086e+03, threshold=6.234e+02, percent-clipped=6.0 2023-05-17 02:43:03,565 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274001.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:43:03,839 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 02:43:15,581 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.04 vs. limit=5.0 2023-05-17 02:43:21,525 INFO [finetune.py:992] (1/2) Epoch 15, batch 2750, loss[loss=0.1661, simple_loss=0.261, pruned_loss=0.03553, over 10560.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2539, pruned_loss=0.03759, over 2365041.85 frames. ], batch size: 68, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:43:38,084 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1612, 5.0322, 4.9887, 5.0705, 4.3666, 5.1392, 5.0779, 5.3223], device='cuda:1'), covar=tensor([0.0250, 0.0181, 0.0187, 0.0364, 0.1179, 0.0287, 0.0174, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0202, 0.0195, 0.0253, 0.0246, 0.0223, 0.0182, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:43:45,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-17 02:43:47,425 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274062.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:43:48,924 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9179, 3.6509, 5.2981, 2.7733, 2.9790, 3.8947, 3.3990, 3.8848], device='cuda:1'), covar=tensor([0.0398, 0.1026, 0.0293, 0.1184, 0.1858, 0.1537, 0.1257, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0240, 0.0256, 0.0187, 0.0239, 0.0297, 0.0227, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:43:49,565 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274065.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:43:58,701 INFO [finetune.py:992] (1/2) Epoch 15, batch 2800, loss[loss=0.1701, simple_loss=0.2708, pruned_loss=0.03468, over 12183.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2532, pruned_loss=0.03728, over 2374887.54 frames. ], batch size: 35, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:44:03,735 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.628e+02 3.055e+02 3.704e+02 7.126e+02, threshold=6.111e+02, percent-clipped=1.0 2023-05-17 02:44:08,895 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274091.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:44:24,285 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274113.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:44:34,253 INFO [finetune.py:992] (1/2) Epoch 15, batch 2850, loss[loss=0.1432, simple_loss=0.2449, pruned_loss=0.02077, over 12148.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2525, pruned_loss=0.0369, over 2371519.02 frames. ], batch size: 34, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:44:36,653 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5652, 2.2192, 2.9963, 2.6609, 2.8483, 2.7600, 2.2643, 2.9517], device='cuda:1'), covar=tensor([0.0139, 0.0329, 0.0169, 0.0210, 0.0160, 0.0186, 0.0311, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0208, 0.0196, 0.0189, 0.0222, 0.0170, 0.0202, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:44:37,283 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274131.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:44:42,783 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274139.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:45:09,683 INFO [finetune.py:992] (1/2) Epoch 15, batch 2900, loss[loss=0.1537, simple_loss=0.2509, pruned_loss=0.02826, over 12178.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2532, pruned_loss=0.03692, over 2377011.86 frames. ], batch size: 35, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:45:14,749 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.580e+02 2.952e+02 3.474e+02 5.749e+02, threshold=5.904e+02, percent-clipped=0.0 2023-05-17 02:45:20,584 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274192.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:45:21,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-17 02:45:47,211 INFO [finetune.py:992] (1/2) Epoch 15, batch 2950, loss[loss=0.1713, simple_loss=0.2586, pruned_loss=0.04203, over 12335.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2533, pruned_loss=0.03707, over 2379346.06 frames. ], batch size: 31, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:46:16,934 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274269.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:46:22,459 INFO [finetune.py:992] (1/2) Epoch 15, batch 3000, loss[loss=0.1536, simple_loss=0.2264, pruned_loss=0.0404, over 12265.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2535, pruned_loss=0.03726, over 2372147.71 frames. ], batch size: 28, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:46:22,459 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 02:46:40,669 INFO [finetune.py:1026] (1/2) Epoch 15, validation: loss=0.3142, simple_loss=0.3917, pruned_loss=0.1184, over 1020973.00 frames. 2023-05-17 02:46:40,669 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 02:46:45,562 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.680e+02 3.036e+02 3.627e+02 7.186e+02, threshold=6.071e+02, percent-clipped=2.0 2023-05-17 02:46:52,129 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1714, 2.4418, 3.0256, 4.0650, 2.1869, 4.1209, 4.0916, 4.2168], device='cuda:1'), covar=tensor([0.0176, 0.1354, 0.0548, 0.0150, 0.1466, 0.0281, 0.0192, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0208, 0.0186, 0.0122, 0.0193, 0.0183, 0.0180, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:46:59,914 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274303.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:47:17,337 INFO [finetune.py:992] (1/2) Epoch 15, batch 3050, loss[loss=0.1657, simple_loss=0.255, pruned_loss=0.03819, over 12029.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03771, over 2367112.39 frames. ], batch size: 31, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:47:19,677 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274330.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:47:39,043 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274357.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:47:44,045 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274364.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:47:53,127 INFO [finetune.py:992] (1/2) Epoch 15, batch 3100, loss[loss=0.1785, simple_loss=0.2713, pruned_loss=0.04284, over 12269.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2546, pruned_loss=0.03786, over 2361767.28 frames. ], batch size: 37, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:47:58,034 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.566e+02 2.991e+02 3.624e+02 6.819e+02, threshold=5.983e+02, percent-clipped=2.0 2023-05-17 02:48:28,799 INFO [finetune.py:992] (1/2) Epoch 15, batch 3150, loss[loss=0.1772, simple_loss=0.2733, pruned_loss=0.0406, over 11701.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2537, pruned_loss=0.03751, over 2369854.76 frames. ], batch size: 48, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:48:48,846 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5649, 4.7658, 3.1630, 2.7315, 4.0864, 2.6589, 4.0040, 3.2126], device='cuda:1'), covar=tensor([0.0671, 0.0608, 0.0992, 0.1557, 0.0353, 0.1333, 0.0533, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0266, 0.0183, 0.0206, 0.0147, 0.0189, 0.0204, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:49:05,476 INFO [finetune.py:992] (1/2) Epoch 15, batch 3200, loss[loss=0.1562, simple_loss=0.2542, pruned_loss=0.02913, over 12364.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2548, pruned_loss=0.03822, over 2362154.65 frames. ], batch size: 36, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:49:10,432 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 2.855e+02 3.190e+02 3.704e+02 5.907e+02, threshold=6.379e+02, percent-clipped=0.0 2023-05-17 02:49:12,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2023-05-17 02:49:12,638 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274487.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:49:16,360 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0829, 4.4673, 3.8135, 4.6961, 4.2361, 2.7044, 4.0198, 2.8410], device='cuda:1'), covar=tensor([0.0876, 0.0794, 0.1507, 0.0563, 0.1281, 0.1843, 0.1082, 0.3594], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0385, 0.0366, 0.0327, 0.0375, 0.0280, 0.0353, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:49:17,673 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2421, 2.9023, 2.7713, 2.7656, 2.4938, 2.4955, 2.8960, 1.9365], device='cuda:1'), covar=tensor([0.0423, 0.0199, 0.0218, 0.0193, 0.0409, 0.0309, 0.0166, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0165, 0.0168, 0.0191, 0.0205, 0.0201, 0.0174, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:49:40,960 INFO [finetune.py:992] (1/2) Epoch 15, batch 3250, loss[loss=0.174, simple_loss=0.2676, pruned_loss=0.04024, over 12194.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2545, pruned_loss=0.03787, over 2364352.87 frames. ], batch size: 35, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:49:50,310 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274540.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:49:56,773 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9203, 3.0012, 4.7904, 5.0152, 2.9736, 2.8151, 3.1280, 2.3786], device='cuda:1'), covar=tensor([0.1592, 0.2951, 0.0432, 0.0377, 0.1310, 0.2401, 0.2648, 0.3805], device='cuda:1'), in_proj_covar=tensor([0.0300, 0.0391, 0.0278, 0.0304, 0.0276, 0.0314, 0.0393, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:50:17,318 INFO [finetune.py:992] (1/2) Epoch 15, batch 3300, loss[loss=0.1863, simple_loss=0.281, pruned_loss=0.04583, over 12142.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2546, pruned_loss=0.03783, over 2360438.82 frames. ], batch size: 36, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:50:22,372 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.627e+02 3.016e+02 3.479e+02 6.691e+02, threshold=6.031e+02, percent-clipped=1.0 2023-05-17 02:50:34,870 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274601.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:50:35,598 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4187, 3.5007, 3.2197, 3.1687, 2.7658, 2.6847, 3.5259, 2.3077], device='cuda:1'), covar=tensor([0.0431, 0.0175, 0.0189, 0.0196, 0.0441, 0.0397, 0.0143, 0.0507], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0167, 0.0169, 0.0191, 0.0206, 0.0203, 0.0175, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:50:52,259 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274625.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:50:53,645 INFO [finetune.py:992] (1/2) Epoch 15, batch 3350, loss[loss=0.147, simple_loss=0.2464, pruned_loss=0.02379, over 12350.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2551, pruned_loss=0.03767, over 2361574.97 frames. ], batch size: 35, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:50:59,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 02:51:15,442 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274657.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:51:16,824 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274659.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:51:29,375 INFO [finetune.py:992] (1/2) Epoch 15, batch 3400, loss[loss=0.1616, simple_loss=0.2552, pruned_loss=0.03393, over 12365.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2548, pruned_loss=0.03788, over 2361725.35 frames. ], batch size: 35, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:51:34,339 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.752e+02 3.196e+02 3.746e+02 7.387e+02, threshold=6.392e+02, percent-clipped=3.0 2023-05-17 02:51:49,561 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274705.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:52:05,807 INFO [finetune.py:992] (1/2) Epoch 15, batch 3450, loss[loss=0.1685, simple_loss=0.2641, pruned_loss=0.03651, over 12105.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2546, pruned_loss=0.03778, over 2367079.31 frames. ], batch size: 33, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:52:41,748 INFO [finetune.py:992] (1/2) Epoch 15, batch 3500, loss[loss=0.158, simple_loss=0.2583, pruned_loss=0.02883, over 12301.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2544, pruned_loss=0.03735, over 2370283.98 frames. ], batch size: 34, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:52:46,313 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-17 02:52:46,624 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.661e+02 2.952e+02 3.432e+02 7.718e+02, threshold=5.904e+02, percent-clipped=1.0 2023-05-17 02:52:48,914 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274787.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:53:02,584 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2410, 5.1535, 5.0975, 5.1467, 4.8061, 5.2335, 5.1803, 5.4791], device='cuda:1'), covar=tensor([0.0233, 0.0138, 0.0168, 0.0316, 0.0699, 0.0313, 0.0165, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0204, 0.0197, 0.0256, 0.0249, 0.0226, 0.0184, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:53:09,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-05-17 02:53:17,160 INFO [finetune.py:992] (1/2) Epoch 15, batch 3550, loss[loss=0.1929, simple_loss=0.2933, pruned_loss=0.04619, over 11671.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2548, pruned_loss=0.03763, over 2367160.68 frames. ], batch size: 48, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:53:22,879 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274835.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:53:31,069 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3599, 4.7864, 4.1914, 4.9132, 4.5780, 3.1373, 4.3290, 3.2260], device='cuda:1'), covar=tensor([0.0774, 0.0791, 0.1357, 0.0603, 0.1068, 0.1579, 0.0994, 0.3101], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0383, 0.0364, 0.0325, 0.0373, 0.0278, 0.0351, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:53:34,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-05-17 02:53:53,488 INFO [finetune.py:992] (1/2) Epoch 15, batch 3600, loss[loss=0.1619, simple_loss=0.2424, pruned_loss=0.04066, over 12344.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2546, pruned_loss=0.03765, over 2370184.57 frames. ], batch size: 31, lr: 3.51e-03, grad_scale: 32.0 2023-05-17 02:53:57,177 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8768, 4.7977, 4.7256, 4.7559, 4.4043, 4.8720, 4.8479, 5.0836], device='cuda:1'), covar=tensor([0.0275, 0.0167, 0.0228, 0.0392, 0.0810, 0.0459, 0.0214, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0203, 0.0197, 0.0256, 0.0248, 0.0226, 0.0184, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 02:53:58,374 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.835e+02 3.190e+02 3.643e+02 6.667e+02, threshold=6.380e+02, percent-clipped=3.0 2023-05-17 02:54:05,710 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274894.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:54:07,098 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274896.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:54:08,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6002, 2.6816, 4.4457, 4.5514, 2.7138, 2.4127, 2.7497, 2.1841], device='cuda:1'), covar=tensor([0.1712, 0.3157, 0.0471, 0.0418, 0.1412, 0.2742, 0.2985, 0.4180], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0395, 0.0280, 0.0308, 0.0279, 0.0318, 0.0397, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:54:28,504 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274925.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:54:29,715 INFO [finetune.py:992] (1/2) Epoch 15, batch 3650, loss[loss=0.1852, simple_loss=0.2816, pruned_loss=0.04435, over 11557.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2545, pruned_loss=0.0374, over 2378935.50 frames. ], batch size: 48, lr: 3.51e-03, grad_scale: 16.0 2023-05-17 02:54:35,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-05-17 02:54:43,519 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8055, 2.9136, 4.7111, 4.8925, 2.7156, 2.6438, 3.0476, 2.3742], device='cuda:1'), covar=tensor([0.1668, 0.3070, 0.0429, 0.0361, 0.1448, 0.2522, 0.2878, 0.3809], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0393, 0.0279, 0.0307, 0.0278, 0.0316, 0.0395, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:54:49,927 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274955.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:54:52,723 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274959.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:55:02,669 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274973.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:55:05,403 INFO [finetune.py:992] (1/2) Epoch 15, batch 3700, loss[loss=0.1616, simple_loss=0.2521, pruned_loss=0.03558, over 12295.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2544, pruned_loss=0.03712, over 2384348.04 frames. ], batch size: 33, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:55:07,424 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2023-05-17 02:55:11,054 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.616e+02 2.961e+02 3.537e+02 1.203e+03, threshold=5.922e+02, percent-clipped=3.0 2023-05-17 02:55:27,128 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275007.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:55:41,938 INFO [finetune.py:992] (1/2) Epoch 15, batch 3750, loss[loss=0.1639, simple_loss=0.2576, pruned_loss=0.03513, over 11233.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2543, pruned_loss=0.03731, over 2370643.02 frames. ], batch size: 55, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:56:17,605 INFO [finetune.py:992] (1/2) Epoch 15, batch 3800, loss[loss=0.1671, simple_loss=0.2672, pruned_loss=0.03353, over 12357.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2551, pruned_loss=0.03764, over 2370550.28 frames. ], batch size: 36, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:56:23,260 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.098e+02 2.835e+02 3.309e+02 3.845e+02 7.855e+02, threshold=6.617e+02, percent-clipped=6.0 2023-05-17 02:56:27,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1650, 4.8329, 4.8863, 4.9741, 4.9181, 5.0394, 4.8867, 2.6568], device='cuda:1'), covar=tensor([0.0080, 0.0088, 0.0095, 0.0061, 0.0042, 0.0092, 0.0144, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0081, 0.0085, 0.0076, 0.0062, 0.0095, 0.0084, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:56:52,558 INFO [finetune.py:992] (1/2) Epoch 15, batch 3850, loss[loss=0.1451, simple_loss=0.2313, pruned_loss=0.02943, over 12287.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2546, pruned_loss=0.03758, over 2368020.82 frames. ], batch size: 33, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:57:04,943 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4370, 2.3775, 3.6909, 4.3061, 3.8010, 4.3486, 3.9025, 3.2356], device='cuda:1'), covar=tensor([0.0042, 0.0422, 0.0133, 0.0060, 0.0150, 0.0078, 0.0134, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0124, 0.0105, 0.0079, 0.0107, 0.0116, 0.0100, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:57:05,954 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-05-17 02:57:08,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 02:57:15,581 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275158.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:57:28,945 INFO [finetune.py:992] (1/2) Epoch 15, batch 3900, loss[loss=0.1892, simple_loss=0.2776, pruned_loss=0.05044, over 11356.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2544, pruned_loss=0.03765, over 2374155.35 frames. ], batch size: 55, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:57:34,722 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.712e+02 3.194e+02 3.668e+02 1.051e+03, threshold=6.388e+02, percent-clipped=2.0 2023-05-17 02:57:42,878 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275196.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:58:00,337 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275219.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:58:05,750 INFO [finetune.py:992] (1/2) Epoch 15, batch 3950, loss[loss=0.1609, simple_loss=0.2551, pruned_loss=0.03331, over 10599.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.254, pruned_loss=0.03761, over 2368558.99 frames. ], batch size: 68, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:58:12,913 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3086, 6.2166, 5.7448, 5.7228, 6.2757, 5.6674, 5.8411, 5.7240], device='cuda:1'), covar=tensor([0.1674, 0.0921, 0.1192, 0.2124, 0.0955, 0.2024, 0.1692, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0505, 0.0407, 0.0458, 0.0470, 0.0442, 0.0403, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 02:58:17,890 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275244.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:58:22,282 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275250.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:58:30,452 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9978, 3.4489, 5.2405, 2.9657, 2.9218, 3.9759, 3.4607, 3.9011], device='cuda:1'), covar=tensor([0.0406, 0.1151, 0.0350, 0.1159, 0.2014, 0.1491, 0.1299, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0243, 0.0260, 0.0189, 0.0242, 0.0301, 0.0230, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 02:58:41,637 INFO [finetune.py:992] (1/2) Epoch 15, batch 4000, loss[loss=0.1537, simple_loss=0.2469, pruned_loss=0.0303, over 12207.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2539, pruned_loss=0.0377, over 2370592.11 frames. ], batch size: 35, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:58:47,335 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.716e+02 3.114e+02 3.647e+02 6.314e+02, threshold=6.229e+02, percent-clipped=0.0 2023-05-17 02:58:55,666 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.49 vs. limit=5.0 2023-05-17 02:59:03,934 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275307.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:59:18,112 INFO [finetune.py:992] (1/2) Epoch 15, batch 4050, loss[loss=0.1539, simple_loss=0.2388, pruned_loss=0.03448, over 12172.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2537, pruned_loss=0.03763, over 2368797.65 frames. ], batch size: 29, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 02:59:28,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-17 02:59:32,662 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275347.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 02:59:39,182 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4596, 2.7221, 3.2537, 4.3525, 2.5090, 4.4372, 4.5139, 4.5115], device='cuda:1'), covar=tensor([0.0192, 0.1228, 0.0498, 0.0179, 0.1253, 0.0281, 0.0152, 0.0141], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0205, 0.0184, 0.0122, 0.0191, 0.0183, 0.0178, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 02:59:48,568 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275368.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 02:59:54,693 INFO [finetune.py:992] (1/2) Epoch 15, batch 4100, loss[loss=0.1676, simple_loss=0.2588, pruned_loss=0.03824, over 12165.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2542, pruned_loss=0.03767, over 2373620.09 frames. ], batch size: 36, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:00:00,458 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.590e+02 3.056e+02 3.712e+02 6.933e+02, threshold=6.111e+02, percent-clipped=2.0 2023-05-17 03:00:00,655 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275385.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:00:17,327 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275408.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 03:00:23,624 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6127, 4.4943, 4.5883, 4.6290, 4.2738, 4.3523, 4.2156, 4.4829], device='cuda:1'), covar=tensor([0.0770, 0.0643, 0.0937, 0.0654, 0.1969, 0.1313, 0.0586, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0723, 0.0630, 0.0650, 0.0869, 0.0766, 0.0581, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:00:30,582 INFO [finetune.py:992] (1/2) Epoch 15, batch 4150, loss[loss=0.1648, simple_loss=0.2572, pruned_loss=0.0362, over 11589.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2551, pruned_loss=0.0379, over 2373670.25 frames. ], batch size: 48, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:00:45,025 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275446.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:00:45,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-17 03:00:58,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 03:01:04,522 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5029, 2.6085, 3.1549, 4.3151, 2.2156, 4.4189, 4.4712, 4.5756], device='cuda:1'), covar=tensor([0.0123, 0.1294, 0.0565, 0.0162, 0.1474, 0.0244, 0.0122, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0207, 0.0185, 0.0123, 0.0192, 0.0184, 0.0178, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:01:06,524 INFO [finetune.py:992] (1/2) Epoch 15, batch 4200, loss[loss=0.1767, simple_loss=0.2731, pruned_loss=0.04015, over 12300.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2557, pruned_loss=0.03784, over 2372542.74 frames. ], batch size: 34, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:01:10,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 03:01:12,130 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.645e+02 3.044e+02 3.736e+02 5.843e+02, threshold=6.088e+02, percent-clipped=0.0 2023-05-17 03:01:33,384 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275514.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:01:42,482 INFO [finetune.py:992] (1/2) Epoch 15, batch 4250, loss[loss=0.1608, simple_loss=0.258, pruned_loss=0.03174, over 12146.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2556, pruned_loss=0.03799, over 2367563.59 frames. ], batch size: 34, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:01:44,096 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6178, 4.3661, 4.6229, 4.0752, 4.3598, 4.1252, 4.5811, 4.2899], device='cuda:1'), covar=tensor([0.0327, 0.0377, 0.0311, 0.0309, 0.0437, 0.0365, 0.0298, 0.0599], device='cuda:1'), in_proj_covar=tensor([0.0269, 0.0272, 0.0299, 0.0270, 0.0275, 0.0271, 0.0245, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:01:44,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-05-17 03:01:59,013 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275550.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:02:15,818 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0502, 4.7533, 4.8154, 4.9561, 4.8111, 4.9890, 4.8470, 2.4024], device='cuda:1'), covar=tensor([0.0100, 0.0072, 0.0092, 0.0058, 0.0047, 0.0089, 0.0076, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0081, 0.0084, 0.0075, 0.0062, 0.0094, 0.0084, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:02:17,796 INFO [finetune.py:992] (1/2) Epoch 15, batch 4300, loss[loss=0.1413, simple_loss=0.2294, pruned_loss=0.02658, over 12170.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2564, pruned_loss=0.03851, over 2356550.24 frames. ], batch size: 29, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:02:24,273 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.684e+02 3.230e+02 3.791e+02 8.652e+02, threshold=6.460e+02, percent-clipped=1.0 2023-05-17 03:02:33,655 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275598.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:02:41,267 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1051, 3.7254, 3.8518, 4.2412, 2.8169, 3.7296, 2.5799, 3.9093], device='cuda:1'), covar=tensor([0.1741, 0.0844, 0.0974, 0.0737, 0.1237, 0.0688, 0.1832, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0270, 0.0301, 0.0363, 0.0244, 0.0248, 0.0264, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:02:54,701 INFO [finetune.py:992] (1/2) Epoch 15, batch 4350, loss[loss=0.1629, simple_loss=0.2502, pruned_loss=0.03781, over 12019.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2555, pruned_loss=0.03788, over 2370897.27 frames. ], batch size: 31, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:03:02,805 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6098, 4.6028, 4.4459, 4.0669, 4.1767, 4.5673, 4.3018, 4.1501], device='cuda:1'), covar=tensor([0.0896, 0.0944, 0.0747, 0.1587, 0.2202, 0.0888, 0.1482, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0576, 0.0532, 0.0657, 0.0431, 0.0745, 0.0803, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:03:21,603 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275663.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:03:31,333 INFO [finetune.py:992] (1/2) Epoch 15, batch 4400, loss[loss=0.1422, simple_loss=0.2207, pruned_loss=0.03184, over 12013.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2546, pruned_loss=0.03733, over 2369284.02 frames. ], batch size: 28, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:03:36,991 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.030e+02 2.679e+02 3.186e+02 3.610e+02 8.620e+02, threshold=6.372e+02, percent-clipped=1.0 2023-05-17 03:03:49,472 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4756, 3.5930, 3.3000, 3.0995, 2.9484, 2.6912, 3.6597, 2.3437], device='cuda:1'), covar=tensor([0.0444, 0.0164, 0.0197, 0.0213, 0.0389, 0.0408, 0.0143, 0.0512], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0165, 0.0168, 0.0190, 0.0205, 0.0203, 0.0176, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:03:50,045 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275703.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 03:04:06,890 INFO [finetune.py:992] (1/2) Epoch 15, batch 4450, loss[loss=0.1529, simple_loss=0.2444, pruned_loss=0.03072, over 12102.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2547, pruned_loss=0.03742, over 2372869.23 frames. ], batch size: 32, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:04:10,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 03:04:17,558 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275741.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:04:35,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-17 03:04:43,216 INFO [finetune.py:992] (1/2) Epoch 15, batch 4500, loss[loss=0.1607, simple_loss=0.2497, pruned_loss=0.03588, over 12089.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2538, pruned_loss=0.03703, over 2377210.56 frames. ], batch size: 33, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:04:48,942 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.591e+02 3.128e+02 3.902e+02 6.297e+02, threshold=6.256e+02, percent-clipped=0.0 2023-05-17 03:05:11,170 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8611, 5.7657, 5.3598, 5.2779, 5.7888, 4.9673, 5.2529, 5.2549], device='cuda:1'), covar=tensor([0.1636, 0.1014, 0.1250, 0.1943, 0.0960, 0.2562, 0.1934, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0509, 0.0412, 0.0460, 0.0475, 0.0444, 0.0404, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:05:11,248 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275814.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:05:20,521 INFO [finetune.py:992] (1/2) Epoch 15, batch 4550, loss[loss=0.1548, simple_loss=0.2545, pruned_loss=0.02754, over 12346.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2525, pruned_loss=0.03642, over 2387764.54 frames. ], batch size: 35, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:05:44,095 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275860.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:05:45,345 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275862.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:05:55,992 INFO [finetune.py:992] (1/2) Epoch 15, batch 4600, loss[loss=0.1716, simple_loss=0.2671, pruned_loss=0.03805, over 10731.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2527, pruned_loss=0.03666, over 2387442.82 frames. ], batch size: 68, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:06:00,335 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7961, 5.7360, 5.4995, 5.0904, 5.0414, 5.7175, 5.3515, 5.0662], device='cuda:1'), covar=tensor([0.0711, 0.0942, 0.0701, 0.1673, 0.0755, 0.0678, 0.1421, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0643, 0.0574, 0.0531, 0.0652, 0.0429, 0.0742, 0.0800, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:06:02,263 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.630e+02 3.077e+02 3.861e+02 7.958e+02, threshold=6.155e+02, percent-clipped=3.0 2023-05-17 03:06:28,052 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275921.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:06:32,368 INFO [finetune.py:992] (1/2) Epoch 15, batch 4650, loss[loss=0.151, simple_loss=0.2501, pruned_loss=0.0259, over 11098.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.253, pruned_loss=0.03687, over 2385405.19 frames. ], batch size: 55, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:06:51,235 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275952.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:06:58,857 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275963.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:07:08,577 INFO [finetune.py:992] (1/2) Epoch 15, batch 4700, loss[loss=0.1798, simple_loss=0.2714, pruned_loss=0.04412, over 12190.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2537, pruned_loss=0.0371, over 2386973.33 frames. ], batch size: 35, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:07:14,349 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.857e+02 3.354e+02 4.097e+02 8.987e+02, threshold=6.708e+02, percent-clipped=2.0 2023-05-17 03:07:18,893 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0883, 4.7458, 4.8936, 4.9571, 4.7876, 4.9743, 4.8486, 2.7269], device='cuda:1'), covar=tensor([0.0081, 0.0076, 0.0086, 0.0079, 0.0062, 0.0089, 0.0126, 0.0712], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0084, 0.0075, 0.0062, 0.0094, 0.0083, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:07:31,047 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276003.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 03:07:36,434 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276011.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:07:38,057 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276013.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:07:48,316 INFO [finetune.py:992] (1/2) Epoch 15, batch 4750, loss[loss=0.1787, simple_loss=0.2641, pruned_loss=0.04666, over 10274.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2542, pruned_loss=0.03759, over 2380432.28 frames. ], batch size: 68, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:07:58,519 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276041.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:08:05,529 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276051.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 03:08:20,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 03:08:23,975 INFO [finetune.py:992] (1/2) Epoch 15, batch 4800, loss[loss=0.1434, simple_loss=0.2306, pruned_loss=0.02813, over 12340.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2546, pruned_loss=0.03777, over 2373270.16 frames. ], batch size: 30, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:08:30,242 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.861e+02 3.478e+02 4.262e+02 1.522e+03, threshold=6.956e+02, percent-clipped=2.0 2023-05-17 03:08:33,191 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276089.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:08:56,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 03:09:00,430 INFO [finetune.py:992] (1/2) Epoch 15, batch 4850, loss[loss=0.1402, simple_loss=0.2298, pruned_loss=0.02526, over 12358.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2546, pruned_loss=0.03795, over 2365247.09 frames. ], batch size: 30, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:09:36,887 INFO [finetune.py:992] (1/2) Epoch 15, batch 4900, loss[loss=0.1696, simple_loss=0.2548, pruned_loss=0.04222, over 12197.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2539, pruned_loss=0.03758, over 2369149.07 frames. ], batch size: 35, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:09:42,530 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.073e+02 2.675e+02 3.087e+02 3.649e+02 6.820e+02, threshold=6.174e+02, percent-clipped=0.0 2023-05-17 03:10:05,208 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276216.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:10:13,648 INFO [finetune.py:992] (1/2) Epoch 15, batch 4950, loss[loss=0.1964, simple_loss=0.284, pruned_loss=0.05443, over 12281.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2554, pruned_loss=0.03836, over 2357372.89 frames. ], batch size: 37, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:10:23,617 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6250, 3.6094, 3.2741, 3.0643, 2.8931, 2.7196, 3.6711, 2.3970], device='cuda:1'), covar=tensor([0.0386, 0.0146, 0.0217, 0.0248, 0.0457, 0.0419, 0.0146, 0.0523], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0165, 0.0168, 0.0191, 0.0205, 0.0202, 0.0177, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:10:29,952 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0357, 4.9072, 4.8241, 4.8629, 4.5457, 5.0070, 5.0427, 5.1589], device='cuda:1'), covar=tensor([0.0318, 0.0171, 0.0226, 0.0385, 0.0846, 0.0364, 0.0183, 0.0222], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0204, 0.0198, 0.0258, 0.0250, 0.0227, 0.0184, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 03:10:49,086 INFO [finetune.py:992] (1/2) Epoch 15, batch 5000, loss[loss=0.1547, simple_loss=0.2391, pruned_loss=0.03518, over 12016.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2543, pruned_loss=0.03786, over 2364091.87 frames. ], batch size: 31, lr: 3.50e-03, grad_scale: 16.0 2023-05-17 03:10:54,696 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.604e+02 3.078e+02 3.726e+02 1.471e+03, threshold=6.157e+02, percent-clipped=5.0 2023-05-17 03:11:11,879 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276308.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:11:18,221 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276317.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:11:24,527 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0625, 5.9469, 5.5872, 5.3908, 6.0020, 5.3970, 5.4767, 5.4518], device='cuda:1'), covar=tensor([0.1721, 0.0928, 0.1233, 0.2230, 0.0946, 0.2364, 0.1971, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0513, 0.0413, 0.0464, 0.0480, 0.0449, 0.0407, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:11:25,116 INFO [finetune.py:992] (1/2) Epoch 15, batch 5050, loss[loss=0.1631, simple_loss=0.252, pruned_loss=0.03711, over 12153.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.254, pruned_loss=0.03732, over 2371003.98 frames. ], batch size: 34, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:11:34,783 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0711, 4.4059, 3.9758, 4.5451, 4.2309, 2.7538, 3.8826, 2.9204], device='cuda:1'), covar=tensor([0.0880, 0.0794, 0.1382, 0.0683, 0.1253, 0.1756, 0.1197, 0.3619], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0382, 0.0361, 0.0325, 0.0369, 0.0276, 0.0348, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:12:01,393 INFO [finetune.py:992] (1/2) Epoch 15, batch 5100, loss[loss=0.1471, simple_loss=0.2316, pruned_loss=0.03129, over 12255.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.0371, over 2377340.62 frames. ], batch size: 32, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:12:02,276 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276378.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:12:06,898 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.657e+02 3.057e+02 3.698e+02 1.370e+03, threshold=6.113e+02, percent-clipped=2.0 2023-05-17 03:12:37,395 INFO [finetune.py:992] (1/2) Epoch 15, batch 5150, loss[loss=0.1468, simple_loss=0.243, pruned_loss=0.0253, over 12150.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2527, pruned_loss=0.03703, over 2382231.79 frames. ], batch size: 34, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:12:56,343 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2236, 2.3750, 3.6163, 4.1686, 3.7650, 4.2190, 3.7792, 2.9098], device='cuda:1'), covar=tensor([0.0041, 0.0412, 0.0129, 0.0044, 0.0116, 0.0078, 0.0127, 0.0386], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0122, 0.0104, 0.0078, 0.0107, 0.0114, 0.0100, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:13:09,808 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9577, 5.9845, 5.7445, 5.2518, 5.2432, 5.9051, 5.4963, 5.2748], device='cuda:1'), covar=tensor([0.0863, 0.0889, 0.0742, 0.1710, 0.0705, 0.0719, 0.1656, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0652, 0.0581, 0.0535, 0.0658, 0.0433, 0.0749, 0.0805, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:13:13,940 INFO [finetune.py:992] (1/2) Epoch 15, batch 5200, loss[loss=0.1996, simple_loss=0.281, pruned_loss=0.05911, over 12109.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2534, pruned_loss=0.03742, over 2374895.89 frames. ], batch size: 39, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:13:16,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-05-17 03:13:19,581 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.618e+02 2.916e+02 3.289e+02 6.315e+02, threshold=5.832e+02, percent-clipped=1.0 2023-05-17 03:13:35,036 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1049, 3.6716, 5.3975, 2.8883, 3.0753, 4.1004, 3.5474, 4.0481], device='cuda:1'), covar=tensor([0.0349, 0.1001, 0.0329, 0.1138, 0.1783, 0.1407, 0.1189, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0239, 0.0257, 0.0185, 0.0239, 0.0296, 0.0227, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:13:42,088 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276516.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:13:50,504 INFO [finetune.py:992] (1/2) Epoch 15, batch 5250, loss[loss=0.1362, simple_loss=0.2128, pruned_loss=0.02983, over 11801.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2523, pruned_loss=0.03692, over 2376239.21 frames. ], batch size: 26, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:14:02,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-17 03:14:04,486 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9762, 5.8869, 5.5631, 5.4071, 5.9487, 5.2842, 5.4264, 5.3658], device='cuda:1'), covar=tensor([0.1345, 0.0923, 0.0994, 0.1812, 0.0929, 0.1964, 0.1733, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0509, 0.0410, 0.0456, 0.0475, 0.0443, 0.0403, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:14:13,040 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5024, 5.2778, 5.4348, 5.4424, 5.0332, 5.1810, 4.9409, 5.4026], device='cuda:1'), covar=tensor([0.0643, 0.0626, 0.0739, 0.0587, 0.1981, 0.1191, 0.0503, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0728, 0.0635, 0.0654, 0.0876, 0.0771, 0.0587, 0.0499], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:14:16,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-17 03:14:16,617 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276564.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:14:26,022 INFO [finetune.py:992] (1/2) Epoch 15, batch 5300, loss[loss=0.149, simple_loss=0.2429, pruned_loss=0.02759, over 12244.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2531, pruned_loss=0.03726, over 2370814.40 frames. ], batch size: 32, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:14:31,810 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.612e+02 3.140e+02 3.525e+02 6.336e+02, threshold=6.281e+02, percent-clipped=1.0 2023-05-17 03:14:49,511 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276608.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:14:51,758 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276611.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:15:02,950 INFO [finetune.py:992] (1/2) Epoch 15, batch 5350, loss[loss=0.1832, simple_loss=0.2623, pruned_loss=0.05204, over 12030.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2529, pruned_loss=0.03717, over 2365486.84 frames. ], batch size: 31, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:15:10,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.93 vs. limit=5.0 2023-05-17 03:15:17,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 2023-05-17 03:15:23,785 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276656.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:15:33,361 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276668.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:15:36,173 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276672.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:15:36,692 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276673.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:15:39,301 INFO [finetune.py:992] (1/2) Epoch 15, batch 5400, loss[loss=0.1792, simple_loss=0.2761, pruned_loss=0.04117, over 12371.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2532, pruned_loss=0.03699, over 2368263.97 frames. ], batch size: 38, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:15:44,947 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.736e+02 3.127e+02 3.839e+02 6.677e+02, threshold=6.254e+02, percent-clipped=1.0 2023-05-17 03:16:05,607 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8662, 5.5910, 5.1651, 5.1912, 5.7148, 5.0770, 5.0913, 5.0697], device='cuda:1'), covar=tensor([0.1456, 0.1001, 0.1193, 0.1951, 0.0982, 0.2186, 0.2012, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0510, 0.0412, 0.0457, 0.0475, 0.0445, 0.0405, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:16:14,459 INFO [finetune.py:992] (1/2) Epoch 15, batch 5450, loss[loss=0.1832, simple_loss=0.2799, pruned_loss=0.04329, over 12138.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2538, pruned_loss=0.03733, over 2374349.43 frames. ], batch size: 39, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:16:16,038 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276729.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:16:33,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-17 03:16:43,415 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0688, 4.9538, 4.8409, 4.8933, 4.6060, 5.0627, 5.0514, 5.1739], device='cuda:1'), covar=tensor([0.0209, 0.0152, 0.0189, 0.0343, 0.0733, 0.0275, 0.0140, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0206, 0.0199, 0.0258, 0.0251, 0.0228, 0.0185, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 03:16:50,367 INFO [finetune.py:992] (1/2) Epoch 15, batch 5500, loss[loss=0.1972, simple_loss=0.2976, pruned_loss=0.04836, over 11241.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2544, pruned_loss=0.03742, over 2367733.98 frames. ], batch size: 55, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:16:56,031 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.745e+02 3.154e+02 3.777e+02 7.064e+02, threshold=6.307e+02, percent-clipped=2.0 2023-05-17 03:16:59,870 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276790.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:17:07,206 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1405, 2.1711, 2.9542, 3.1142, 3.0274, 3.1850, 3.0096, 2.4867], device='cuda:1'), covar=tensor([0.0093, 0.0434, 0.0181, 0.0088, 0.0170, 0.0106, 0.0148, 0.0381], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0123, 0.0105, 0.0079, 0.0108, 0.0115, 0.0099, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:17:26,677 INFO [finetune.py:992] (1/2) Epoch 15, batch 5550, loss[loss=0.2273, simple_loss=0.2944, pruned_loss=0.08009, over 7841.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2547, pruned_loss=0.03761, over 2361628.60 frames. ], batch size: 97, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:17:31,064 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0478, 6.0305, 5.7683, 5.2929, 5.1897, 5.9574, 5.5775, 5.3228], device='cuda:1'), covar=tensor([0.0694, 0.0826, 0.0694, 0.1638, 0.0670, 0.0646, 0.1320, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0649, 0.0580, 0.0533, 0.0656, 0.0431, 0.0747, 0.0801, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:17:35,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-17 03:17:37,415 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1851, 4.5570, 3.9644, 4.9640, 4.3770, 2.9527, 4.1940, 2.9804], device='cuda:1'), covar=tensor([0.0843, 0.0789, 0.1342, 0.0489, 0.1178, 0.1586, 0.1076, 0.3392], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0384, 0.0363, 0.0327, 0.0371, 0.0276, 0.0350, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:17:43,120 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1382, 2.5099, 3.7413, 3.1516, 3.5844, 3.2307, 2.5866, 3.6906], device='cuda:1'), covar=tensor([0.0146, 0.0343, 0.0180, 0.0238, 0.0169, 0.0182, 0.0346, 0.0141], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0207, 0.0195, 0.0191, 0.0223, 0.0169, 0.0201, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:17:43,795 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276851.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:17:52,558 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276863.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:17:58,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.94 vs. limit=5.0 2023-05-17 03:18:02,417 INFO [finetune.py:992] (1/2) Epoch 15, batch 5600, loss[loss=0.1862, simple_loss=0.2742, pruned_loss=0.04909, over 8444.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2538, pruned_loss=0.03714, over 2365463.40 frames. ], batch size: 98, lr: 3.49e-03, grad_scale: 16.0 2023-05-17 03:18:08,761 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.770e+02 3.218e+02 3.855e+02 6.712e+02, threshold=6.436e+02, percent-clipped=1.0 2023-05-17 03:18:37,011 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276924.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:18:38,816 INFO [finetune.py:992] (1/2) Epoch 15, batch 5650, loss[loss=0.1602, simple_loss=0.2517, pruned_loss=0.03437, over 12353.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2534, pruned_loss=0.03712, over 2363346.41 frames. ], batch size: 35, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:18:40,350 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2939, 6.2426, 5.9783, 5.4306, 5.2847, 6.1743, 5.8487, 5.5145], device='cuda:1'), covar=tensor([0.0566, 0.0714, 0.0626, 0.1588, 0.0695, 0.0652, 0.1242, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0577, 0.0532, 0.0653, 0.0429, 0.0742, 0.0797, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:18:41,728 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276931.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:18:50,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2429, 4.8480, 4.9940, 5.0894, 4.8909, 5.1194, 5.0048, 2.7587], device='cuda:1'), covar=tensor([0.0088, 0.0066, 0.0080, 0.0056, 0.0047, 0.0080, 0.0073, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0086, 0.0076, 0.0063, 0.0096, 0.0086, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:18:56,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-17 03:19:07,756 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276967.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:19:12,015 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276973.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:19:14,636 INFO [finetune.py:992] (1/2) Epoch 15, batch 5700, loss[loss=0.1761, simple_loss=0.269, pruned_loss=0.04158, over 11564.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2529, pruned_loss=0.03679, over 2374926.50 frames. ], batch size: 48, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:19:20,246 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.920e+02 2.620e+02 3.127e+02 4.121e+02 6.231e+02, threshold=6.254e+02, percent-clipped=0.0 2023-05-17 03:19:25,525 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276992.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 03:19:32,328 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9579, 2.4463, 3.5907, 3.0005, 3.4380, 3.0839, 2.3858, 3.5563], device='cuda:1'), covar=tensor([0.0150, 0.0416, 0.0182, 0.0288, 0.0146, 0.0244, 0.0487, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0208, 0.0196, 0.0192, 0.0223, 0.0170, 0.0202, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:19:39,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277011.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:19:46,441 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277021.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:19:48,479 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277024.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:19:50,579 INFO [finetune.py:992] (1/2) Epoch 15, batch 5750, loss[loss=0.1648, simple_loss=0.2506, pruned_loss=0.03951, over 12086.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2528, pruned_loss=0.03674, over 2377561.02 frames. ], batch size: 32, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:19:50,829 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6716, 3.0695, 3.3927, 4.5369, 2.2989, 4.6483, 4.6856, 4.7216], device='cuda:1'), covar=tensor([0.0132, 0.1076, 0.0468, 0.0189, 0.1410, 0.0217, 0.0142, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0210, 0.0188, 0.0125, 0.0196, 0.0187, 0.0182, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:20:23,175 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277072.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:20:26,350 INFO [finetune.py:992] (1/2) Epoch 15, batch 5800, loss[loss=0.2401, simple_loss=0.313, pruned_loss=0.08354, over 7915.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.03703, over 2377635.44 frames. ], batch size: 98, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:20:31,931 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.608e+02 3.112e+02 3.705e+02 8.677e+02, threshold=6.225e+02, percent-clipped=5.0 2023-05-17 03:20:42,684 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1378, 2.6800, 3.7925, 3.1906, 3.5790, 3.3153, 2.6321, 3.6657], device='cuda:1'), covar=tensor([0.0159, 0.0374, 0.0168, 0.0272, 0.0167, 0.0210, 0.0380, 0.0153], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0207, 0.0196, 0.0191, 0.0222, 0.0170, 0.0201, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:21:02,533 INFO [finetune.py:992] (1/2) Epoch 15, batch 5850, loss[loss=0.1793, simple_loss=0.273, pruned_loss=0.04279, over 12057.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2537, pruned_loss=0.03722, over 2375346.53 frames. ], batch size: 42, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:21:09,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1203, 2.8580, 2.6979, 2.7139, 2.4858, 2.3854, 2.8311, 2.1053], device='cuda:1'), covar=tensor([0.0421, 0.0214, 0.0245, 0.0212, 0.0434, 0.0407, 0.0213, 0.0446], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0161, 0.0164, 0.0186, 0.0197, 0.0195, 0.0171, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:21:16,044 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277146.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:21:21,132 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3745, 4.8424, 4.1026, 4.9674, 4.5168, 3.0343, 4.2654, 3.0544], device='cuda:1'), covar=tensor([0.0838, 0.0638, 0.1563, 0.0585, 0.1144, 0.1700, 0.1108, 0.3763], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0386, 0.0365, 0.0330, 0.0374, 0.0277, 0.0352, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:21:26,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 03:21:38,767 INFO [finetune.py:992] (1/2) Epoch 15, batch 5900, loss[loss=0.1713, simple_loss=0.2643, pruned_loss=0.03912, over 11255.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2535, pruned_loss=0.03725, over 2367163.81 frames. ], batch size: 55, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:21:38,971 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277177.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:21:44,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-05-17 03:21:44,302 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.580e+02 3.106e+02 3.557e+02 8.858e+02, threshold=6.212e+02, percent-clipped=2.0 2023-05-17 03:22:05,151 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5016, 2.5222, 3.7739, 4.4567, 3.8258, 4.5124, 3.9369, 2.9445], device='cuda:1'), covar=tensor([0.0039, 0.0430, 0.0130, 0.0046, 0.0121, 0.0071, 0.0120, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0126, 0.0108, 0.0080, 0.0110, 0.0118, 0.0102, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:22:09,490 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277219.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:22:09,558 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277219.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:22:15,090 INFO [finetune.py:992] (1/2) Epoch 15, batch 5950, loss[loss=0.1788, simple_loss=0.2685, pruned_loss=0.04457, over 11617.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2532, pruned_loss=0.03686, over 2366406.00 frames. ], batch size: 48, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:22:16,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2023-05-17 03:22:21,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.05 vs. limit=5.0 2023-05-17 03:22:23,054 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277238.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:22:44,501 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277267.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:22:51,375 INFO [finetune.py:992] (1/2) Epoch 15, batch 6000, loss[loss=0.1651, simple_loss=0.265, pruned_loss=0.03256, over 12362.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2529, pruned_loss=0.03648, over 2377161.77 frames. ], batch size: 36, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:22:51,376 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 03:23:09,833 INFO [finetune.py:1026] (1/2) Epoch 15, validation: loss=0.3187, simple_loss=0.3936, pruned_loss=0.1219, over 1020973.00 frames. 2023-05-17 03:23:09,834 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 03:23:12,083 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277280.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:23:15,954 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.528e+02 2.885e+02 3.414e+02 8.880e+02, threshold=5.771e+02, percent-clipped=2.0 2023-05-17 03:23:17,466 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277287.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 03:23:30,629 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277305.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:23:37,656 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277315.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:23:44,203 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277324.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:23:46,202 INFO [finetune.py:992] (1/2) Epoch 15, batch 6050, loss[loss=0.1728, simple_loss=0.2678, pruned_loss=0.03891, over 12098.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2529, pruned_loss=0.03662, over 2377061.07 frames. ], batch size: 40, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:24:14,353 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277366.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:24:14,917 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277367.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:24:18,480 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277372.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:24:21,950 INFO [finetune.py:992] (1/2) Epoch 15, batch 6100, loss[loss=0.1571, simple_loss=0.2521, pruned_loss=0.03106, over 12299.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2534, pruned_loss=0.03697, over 2379056.25 frames. ], batch size: 34, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:24:27,650 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.661e+02 3.167e+02 3.859e+02 7.157e+02, threshold=6.333e+02, percent-clipped=3.0 2023-05-17 03:24:54,638 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5870, 2.6881, 3.2209, 4.4513, 2.2257, 4.4953, 4.5530, 4.6278], device='cuda:1'), covar=tensor([0.0131, 0.1276, 0.0505, 0.0147, 0.1458, 0.0246, 0.0153, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0208, 0.0187, 0.0124, 0.0195, 0.0186, 0.0181, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:24:58,114 INFO [finetune.py:992] (1/2) Epoch 15, batch 6150, loss[loss=0.158, simple_loss=0.2504, pruned_loss=0.03279, over 12304.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2542, pruned_loss=0.03722, over 2375075.93 frames. ], batch size: 34, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:25:12,030 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277446.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:25:15,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 2023-05-17 03:25:33,881 INFO [finetune.py:992] (1/2) Epoch 15, batch 6200, loss[loss=0.1606, simple_loss=0.2524, pruned_loss=0.0344, over 12337.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03772, over 2379266.65 frames. ], batch size: 38, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:25:39,637 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.720e+02 3.197e+02 3.558e+02 7.007e+02, threshold=6.393e+02, percent-clipped=2.0 2023-05-17 03:25:46,791 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277494.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:25:46,941 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4409, 2.4826, 3.1218, 4.2316, 2.1781, 4.3547, 4.4128, 4.5093], device='cuda:1'), covar=tensor([0.0141, 0.1377, 0.0555, 0.0201, 0.1392, 0.0274, 0.0212, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0207, 0.0186, 0.0124, 0.0194, 0.0185, 0.0180, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:26:04,886 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277519.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:26:07,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2582, 6.1520, 5.6290, 5.7150, 6.2140, 5.4759, 5.5315, 5.6638], device='cuda:1'), covar=tensor([0.1559, 0.0979, 0.1089, 0.1621, 0.0936, 0.2279, 0.2100, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0509, 0.0410, 0.0458, 0.0474, 0.0444, 0.0408, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:26:10,498 INFO [finetune.py:992] (1/2) Epoch 15, batch 6250, loss[loss=0.1727, simple_loss=0.2762, pruned_loss=0.03456, over 12278.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.254, pruned_loss=0.03776, over 2375215.41 frames. ], batch size: 37, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:26:14,855 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277533.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:26:29,900 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277554.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:26:39,170 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277567.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:26:41,488 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277570.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:26:44,805 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277575.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:26:46,190 INFO [finetune.py:992] (1/2) Epoch 15, batch 6300, loss[loss=0.1512, simple_loss=0.2434, pruned_loss=0.02945, over 12327.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2539, pruned_loss=0.03761, over 2375709.56 frames. ], batch size: 31, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:26:52,621 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.629e+02 3.057e+02 3.498e+02 6.273e+02, threshold=6.115e+02, percent-clipped=0.0 2023-05-17 03:26:54,320 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277587.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 03:27:14,331 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277615.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:27:22,494 INFO [finetune.py:992] (1/2) Epoch 15, batch 6350, loss[loss=0.176, simple_loss=0.268, pruned_loss=0.04202, over 12047.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2539, pruned_loss=0.03762, over 2374280.96 frames. ], batch size: 37, lr: 3.49e-03, grad_scale: 32.0 2023-05-17 03:27:26,090 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277631.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:27:28,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 03:27:28,802 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277635.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:27:47,790 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277661.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:27:52,182 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277667.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:27:59,291 INFO [finetune.py:992] (1/2) Epoch 15, batch 6400, loss[loss=0.1553, simple_loss=0.2312, pruned_loss=0.03974, over 12019.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2537, pruned_loss=0.03758, over 2368417.76 frames. ], batch size: 28, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:28:04,862 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.566e+02 2.970e+02 3.495e+02 7.369e+02, threshold=5.939e+02, percent-clipped=2.0 2023-05-17 03:28:09,255 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277691.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:28:24,361 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3060, 3.5350, 3.1544, 3.0825, 2.7679, 2.5479, 3.5021, 2.2568], device='cuda:1'), covar=tensor([0.0493, 0.0154, 0.0220, 0.0253, 0.0476, 0.0428, 0.0159, 0.0561], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0165, 0.0168, 0.0191, 0.0203, 0.0200, 0.0176, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:28:26,408 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277715.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:28:35,616 INFO [finetune.py:992] (1/2) Epoch 15, batch 6450, loss[loss=0.1893, simple_loss=0.2814, pruned_loss=0.04865, over 12048.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2553, pruned_loss=0.03803, over 2365933.03 frames. ], batch size: 42, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:28:46,487 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5408, 5.4030, 5.5345, 5.5472, 5.1849, 5.2015, 4.9839, 5.4819], device='cuda:1'), covar=tensor([0.0794, 0.0686, 0.0899, 0.0678, 0.1828, 0.1423, 0.0540, 0.0976], device='cuda:1'), in_proj_covar=tensor([0.0569, 0.0734, 0.0642, 0.0662, 0.0887, 0.0779, 0.0593, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:28:52,994 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3872, 3.4983, 3.1333, 3.1049, 2.7828, 2.6066, 3.5299, 2.2448], device='cuda:1'), covar=tensor([0.0449, 0.0171, 0.0241, 0.0225, 0.0460, 0.0457, 0.0162, 0.0546], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0165, 0.0168, 0.0190, 0.0202, 0.0200, 0.0176, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:28:53,722 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277752.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 03:29:12,051 INFO [finetune.py:992] (1/2) Epoch 15, batch 6500, loss[loss=0.1794, simple_loss=0.2703, pruned_loss=0.04425, over 12150.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2554, pruned_loss=0.03839, over 2362823.72 frames. ], batch size: 34, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:29:17,667 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.648e+02 3.126e+02 3.804e+02 5.907e+02, threshold=6.252e+02, percent-clipped=0.0 2023-05-17 03:29:47,789 INFO [finetune.py:992] (1/2) Epoch 15, batch 6550, loss[loss=0.1522, simple_loss=0.2379, pruned_loss=0.03324, over 12191.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2562, pruned_loss=0.0386, over 2364145.21 frames. ], batch size: 29, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:29:50,074 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2329, 4.2412, 4.1915, 4.5649, 3.0229, 4.0826, 2.6485, 4.2190], device='cuda:1'), covar=tensor([0.1582, 0.0624, 0.0825, 0.0519, 0.1225, 0.0564, 0.1815, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0266, 0.0298, 0.0360, 0.0243, 0.0246, 0.0262, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:29:52,114 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277833.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:30:18,693 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6137, 3.7298, 3.3311, 3.2717, 2.9579, 2.7804, 3.7559, 2.5191], device='cuda:1'), covar=tensor([0.0368, 0.0132, 0.0174, 0.0196, 0.0380, 0.0390, 0.0116, 0.0430], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0165, 0.0168, 0.0190, 0.0203, 0.0201, 0.0175, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:30:22,694 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277875.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:30:23,961 INFO [finetune.py:992] (1/2) Epoch 15, batch 6600, loss[loss=0.1489, simple_loss=0.2412, pruned_loss=0.0283, over 12120.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2558, pruned_loss=0.03871, over 2358607.50 frames. ], batch size: 39, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:30:26,950 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277881.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:30:29,648 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.688e+02 3.224e+02 3.921e+02 8.533e+02, threshold=6.448e+02, percent-clipped=2.0 2023-05-17 03:30:33,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 03:30:47,651 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277910.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:30:48,582 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1112, 2.5351, 3.6202, 3.0585, 3.5407, 3.1629, 2.5842, 3.5356], device='cuda:1'), covar=tensor([0.0136, 0.0364, 0.0154, 0.0240, 0.0143, 0.0202, 0.0356, 0.0141], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0210, 0.0197, 0.0192, 0.0224, 0.0171, 0.0202, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:30:49,968 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4944, 2.8862, 3.7853, 4.5558, 3.8852, 4.6368, 4.0579, 3.5342], device='cuda:1'), covar=tensor([0.0047, 0.0376, 0.0156, 0.0040, 0.0140, 0.0068, 0.0115, 0.0302], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0124, 0.0107, 0.0080, 0.0109, 0.0117, 0.0100, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:30:57,480 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277923.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:30:59,577 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277926.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:31:00,253 INFO [finetune.py:992] (1/2) Epoch 15, batch 6650, loss[loss=0.1296, simple_loss=0.2095, pruned_loss=0.02485, over 12185.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2549, pruned_loss=0.03824, over 2357026.53 frames. ], batch size: 29, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:31:24,587 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277961.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:31:31,696 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2522, 4.8256, 5.2715, 4.6082, 4.9289, 4.6507, 5.2659, 4.9978], device='cuda:1'), covar=tensor([0.0280, 0.0401, 0.0280, 0.0276, 0.0393, 0.0361, 0.0229, 0.0294], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0274, 0.0297, 0.0267, 0.0271, 0.0269, 0.0244, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:31:31,763 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277971.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:31:33,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.72 vs. limit=5.0 2023-05-17 03:31:35,921 INFO [finetune.py:992] (1/2) Epoch 15, batch 6700, loss[loss=0.1401, simple_loss=0.2324, pruned_loss=0.02384, over 12342.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2547, pruned_loss=0.03804, over 2362587.67 frames. ], batch size: 30, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:31:41,607 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.527e+02 3.046e+02 3.825e+02 1.241e+03, threshold=6.091e+02, percent-clipped=2.0 2023-05-17 03:31:59,851 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8639, 4.6331, 4.2113, 4.1471, 4.6734, 4.0494, 4.2142, 4.0208], device='cuda:1'), covar=tensor([0.1879, 0.1206, 0.1564, 0.2216, 0.1283, 0.2424, 0.1934, 0.1552], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0509, 0.0410, 0.0461, 0.0474, 0.0443, 0.0408, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:32:02,669 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278009.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:32:15,937 INFO [finetune.py:992] (1/2) Epoch 15, batch 6750, loss[loss=0.1912, simple_loss=0.2741, pruned_loss=0.05411, over 12305.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2545, pruned_loss=0.03783, over 2372209.70 frames. ], batch size: 34, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:32:19,658 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278032.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:32:30,263 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278047.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 03:32:52,491 INFO [finetune.py:992] (1/2) Epoch 15, batch 6800, loss[loss=0.1641, simple_loss=0.2577, pruned_loss=0.03527, over 12160.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2546, pruned_loss=0.03789, over 2376692.84 frames. ], batch size: 36, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:32:58,186 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.542e+02 3.281e+02 4.073e+02 8.715e+02, threshold=6.562e+02, percent-clipped=2.0 2023-05-17 03:33:27,877 INFO [finetune.py:992] (1/2) Epoch 15, batch 6850, loss[loss=0.1709, simple_loss=0.2602, pruned_loss=0.04086, over 12109.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2563, pruned_loss=0.03865, over 2364030.27 frames. ], batch size: 33, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:33:31,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-05-17 03:33:41,765 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0859, 5.9784, 5.4385, 5.4518, 6.0387, 5.3532, 5.5002, 5.4829], device='cuda:1'), covar=tensor([0.1523, 0.0927, 0.1156, 0.1981, 0.0806, 0.2130, 0.1927, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0514, 0.0412, 0.0465, 0.0476, 0.0445, 0.0411, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:34:04,485 INFO [finetune.py:992] (1/2) Epoch 15, batch 6900, loss[loss=0.1498, simple_loss=0.2314, pruned_loss=0.03411, over 12170.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2555, pruned_loss=0.03849, over 2365485.50 frames. ], batch size: 29, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:34:10,166 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.747e+02 3.327e+02 4.069e+02 6.730e+02, threshold=6.654e+02, percent-clipped=1.0 2023-05-17 03:34:29,064 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278210.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:34:40,201 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278226.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:34:40,753 INFO [finetune.py:992] (1/2) Epoch 15, batch 6950, loss[loss=0.1899, simple_loss=0.2761, pruned_loss=0.05182, over 12101.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2559, pruned_loss=0.03839, over 2372040.08 frames. ], batch size: 33, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:35:02,920 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278258.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:35:06,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1400, 2.5955, 3.7252, 3.0556, 3.5550, 3.2225, 2.5426, 3.5945], device='cuda:1'), covar=tensor([0.0161, 0.0374, 0.0154, 0.0268, 0.0152, 0.0213, 0.0410, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0211, 0.0198, 0.0192, 0.0225, 0.0172, 0.0202, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:35:14,325 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278274.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:35:16,401 INFO [finetune.py:992] (1/2) Epoch 15, batch 7000, loss[loss=0.1366, simple_loss=0.2219, pruned_loss=0.02564, over 11756.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2563, pruned_loss=0.03887, over 2357094.41 frames. ], batch size: 26, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:35:22,219 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.518e+02 3.074e+02 3.919e+02 6.846e+02, threshold=6.148e+02, percent-clipped=1.0 2023-05-17 03:35:24,615 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4211, 2.6547, 3.7087, 4.4211, 3.6661, 4.5160, 3.8536, 3.2776], device='cuda:1'), covar=tensor([0.0055, 0.0381, 0.0155, 0.0044, 0.0174, 0.0068, 0.0110, 0.0324], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0125, 0.0108, 0.0081, 0.0110, 0.0117, 0.0102, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:35:50,423 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278323.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:35:53,083 INFO [finetune.py:992] (1/2) Epoch 15, batch 7050, loss[loss=0.1614, simple_loss=0.2601, pruned_loss=0.03131, over 12343.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2555, pruned_loss=0.03823, over 2363002.58 frames. ], batch size: 35, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:35:53,165 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278327.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:36:05,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-05-17 03:36:08,136 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278347.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 03:36:27,325 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278374.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:36:29,296 INFO [finetune.py:992] (1/2) Epoch 15, batch 7100, loss[loss=0.1438, simple_loss=0.2371, pruned_loss=0.02528, over 12364.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2557, pruned_loss=0.03839, over 2361825.08 frames. ], batch size: 30, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:36:34,500 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278384.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:36:34,965 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.567e+02 2.960e+02 3.611e+02 5.466e+02, threshold=5.920e+02, percent-clipped=0.0 2023-05-17 03:36:42,059 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278395.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:36:44,866 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4002, 4.8152, 3.0602, 2.8114, 4.0840, 2.8054, 4.0249, 3.2738], device='cuda:1'), covar=tensor([0.0666, 0.0434, 0.1060, 0.1402, 0.0349, 0.1190, 0.0497, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0260, 0.0179, 0.0202, 0.0144, 0.0185, 0.0200, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:36:49,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9476, 4.4240, 4.0384, 4.7147, 4.2515, 2.8591, 4.0631, 2.8735], device='cuda:1'), covar=tensor([0.1074, 0.0902, 0.1477, 0.0528, 0.1278, 0.1818, 0.1652, 0.3763], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0389, 0.0367, 0.0333, 0.0377, 0.0280, 0.0353, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:37:04,848 INFO [finetune.py:992] (1/2) Epoch 15, batch 7150, loss[loss=0.1857, simple_loss=0.2808, pruned_loss=0.04532, over 12089.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2559, pruned_loss=0.03837, over 2373287.57 frames. ], batch size: 40, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:37:10,782 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278435.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:37:28,411 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5095, 5.2850, 5.4710, 5.4950, 5.0975, 5.1771, 4.9089, 5.4103], device='cuda:1'), covar=tensor([0.0732, 0.0699, 0.0938, 0.0584, 0.1935, 0.1224, 0.0598, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0734, 0.0644, 0.0663, 0.0884, 0.0777, 0.0594, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:37:36,371 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1476, 4.7978, 4.9951, 5.0122, 4.7731, 5.0726, 5.0298, 2.6951], device='cuda:1'), covar=tensor([0.0101, 0.0070, 0.0082, 0.0063, 0.0050, 0.0100, 0.0069, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0085, 0.0075, 0.0062, 0.0094, 0.0084, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:37:40,934 INFO [finetune.py:992] (1/2) Epoch 15, batch 7200, loss[loss=0.1757, simple_loss=0.2694, pruned_loss=0.04101, over 12142.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2555, pruned_loss=0.03834, over 2366525.40 frames. ], batch size: 39, lr: 3.48e-03, grad_scale: 32.0 2023-05-17 03:37:46,543 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.745e+02 3.138e+02 3.930e+02 1.052e+03, threshold=6.276e+02, percent-clipped=4.0 2023-05-17 03:37:58,801 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8511, 2.4302, 3.2657, 2.8324, 3.1953, 3.0359, 2.2752, 3.2726], device='cuda:1'), covar=tensor([0.0155, 0.0350, 0.0203, 0.0270, 0.0185, 0.0202, 0.0424, 0.0146], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0213, 0.0201, 0.0195, 0.0227, 0.0174, 0.0205, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:38:05,961 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278511.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:38:17,165 INFO [finetune.py:992] (1/2) Epoch 15, batch 7250, loss[loss=0.1532, simple_loss=0.2275, pruned_loss=0.03943, over 12174.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2551, pruned_loss=0.03805, over 2373240.24 frames. ], batch size: 29, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:38:17,556 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-17 03:38:34,036 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7770, 2.9105, 4.6467, 4.8536, 2.8526, 2.7389, 3.0673, 2.3215], device='cuda:1'), covar=tensor([0.1648, 0.3042, 0.0552, 0.0420, 0.1382, 0.2428, 0.2844, 0.4030], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0391, 0.0282, 0.0304, 0.0275, 0.0316, 0.0393, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:38:49,783 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278572.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:38:53,119 INFO [finetune.py:992] (1/2) Epoch 15, batch 7300, loss[loss=0.1736, simple_loss=0.2674, pruned_loss=0.03992, over 12186.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2555, pruned_loss=0.03805, over 2371432.82 frames. ], batch size: 35, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:38:59,563 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.602e+02 2.987e+02 3.460e+02 6.466e+02, threshold=5.973e+02, percent-clipped=1.0 2023-05-17 03:39:19,473 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7395, 4.0261, 3.5817, 4.2013, 3.8473, 2.7001, 3.7166, 2.8581], device='cuda:1'), covar=tensor([0.0913, 0.0942, 0.1513, 0.0662, 0.1188, 0.1720, 0.1047, 0.3231], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0389, 0.0368, 0.0332, 0.0375, 0.0280, 0.0353, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:39:30,111 INFO [finetune.py:992] (1/2) Epoch 15, batch 7350, loss[loss=0.1456, simple_loss=0.2317, pruned_loss=0.02978, over 12177.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2551, pruned_loss=0.03778, over 2369759.82 frames. ], batch size: 31, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:39:30,225 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278627.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:40:04,782 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278675.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:40:06,107 INFO [finetune.py:992] (1/2) Epoch 15, batch 7400, loss[loss=0.1997, simple_loss=0.2878, pruned_loss=0.05584, over 10545.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2553, pruned_loss=0.0379, over 2370874.82 frames. ], batch size: 68, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:40:07,645 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278679.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:40:12,469 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.552e+02 3.075e+02 3.617e+02 5.572e+02, threshold=6.150e+02, percent-clipped=0.0 2023-05-17 03:40:41,669 INFO [finetune.py:992] (1/2) Epoch 15, batch 7450, loss[loss=0.1868, simple_loss=0.2785, pruned_loss=0.04757, over 10481.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.255, pruned_loss=0.03785, over 2375809.12 frames. ], batch size: 68, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:40:43,914 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278730.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:41:06,963 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6347, 3.1713, 3.9101, 4.6964, 4.0224, 4.8158, 4.1291, 3.5568], device='cuda:1'), covar=tensor([0.0044, 0.0336, 0.0133, 0.0045, 0.0120, 0.0073, 0.0118, 0.0325], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0124, 0.0107, 0.0081, 0.0109, 0.0118, 0.0101, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:41:18,049 INFO [finetune.py:992] (1/2) Epoch 15, batch 7500, loss[loss=0.1746, simple_loss=0.28, pruned_loss=0.03455, over 12150.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2544, pruned_loss=0.03784, over 2371832.15 frames. ], batch size: 34, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:41:24,596 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.649e+02 3.038e+02 3.624e+02 6.631e+02, threshold=6.076e+02, percent-clipped=2.0 2023-05-17 03:41:54,363 INFO [finetune.py:992] (1/2) Epoch 15, batch 7550, loss[loss=0.1682, simple_loss=0.2597, pruned_loss=0.03838, over 11103.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2548, pruned_loss=0.03808, over 2372209.35 frames. ], batch size: 55, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:42:22,788 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278867.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:42:23,794 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-17 03:42:24,968 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7985, 5.6994, 5.7084, 5.0197, 5.0896, 5.9179, 5.0490, 5.2554], device='cuda:1'), covar=tensor([0.1530, 0.1986, 0.1218, 0.3288, 0.1219, 0.1394, 0.3665, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.0654, 0.0584, 0.0537, 0.0664, 0.0434, 0.0753, 0.0812, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:42:29,700 INFO [finetune.py:992] (1/2) Epoch 15, batch 7600, loss[loss=0.1464, simple_loss=0.2292, pruned_loss=0.03184, over 12358.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2543, pruned_loss=0.0379, over 2373798.75 frames. ], batch size: 31, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:42:36,834 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.704e+02 3.260e+02 3.972e+02 8.600e+02, threshold=6.520e+02, percent-clipped=2.0 2023-05-17 03:42:44,925 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2212, 5.0787, 4.9646, 5.1014, 4.7200, 5.1306, 5.2371, 5.3644], device='cuda:1'), covar=tensor([0.0245, 0.0139, 0.0195, 0.0301, 0.0709, 0.0278, 0.0120, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0205, 0.0199, 0.0258, 0.0250, 0.0229, 0.0185, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 03:42:50,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 03:42:55,378 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 03:42:58,650 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.7053, 5.7289, 5.5153, 5.0877, 4.9889, 5.6559, 5.2310, 5.1048], device='cuda:1'), covar=tensor([0.0902, 0.0971, 0.0701, 0.1701, 0.0819, 0.0732, 0.1666, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0582, 0.0535, 0.0661, 0.0433, 0.0750, 0.0810, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:43:06,455 INFO [finetune.py:992] (1/2) Epoch 15, batch 7650, loss[loss=0.1758, simple_loss=0.2675, pruned_loss=0.04207, over 12311.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2555, pruned_loss=0.03813, over 2375120.62 frames. ], batch size: 34, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:43:12,823 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2598, 6.2177, 5.8439, 5.7335, 6.2161, 5.5271, 5.6730, 5.7575], device='cuda:1'), covar=tensor([0.1285, 0.0841, 0.1142, 0.1716, 0.0916, 0.2241, 0.2103, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0503, 0.0408, 0.0459, 0.0473, 0.0441, 0.0407, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:43:42,940 INFO [finetune.py:992] (1/2) Epoch 15, batch 7700, loss[loss=0.1541, simple_loss=0.2441, pruned_loss=0.03201, over 12266.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2553, pruned_loss=0.03801, over 2372183.09 frames. ], batch size: 32, lr: 3.48e-03, grad_scale: 16.0 2023-05-17 03:43:44,367 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278979.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:43:48,982 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.736e+02 3.229e+02 3.704e+02 6.260e+02, threshold=6.459e+02, percent-clipped=0.0 2023-05-17 03:44:14,289 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1210, 3.2486, 3.4408, 3.8301, 2.7208, 3.2068, 2.3150, 3.1705], device='cuda:1'), covar=tensor([0.1799, 0.1037, 0.1024, 0.0714, 0.1299, 0.0912, 0.2104, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0270, 0.0302, 0.0364, 0.0246, 0.0249, 0.0264, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:44:14,532 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 03:44:18,301 INFO [finetune.py:992] (1/2) Epoch 15, batch 7750, loss[loss=0.1702, simple_loss=0.2649, pruned_loss=0.0377, over 12149.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2558, pruned_loss=0.03839, over 2372612.48 frames. ], batch size: 36, lr: 3.47e-03, grad_scale: 16.0 2023-05-17 03:44:18,359 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279027.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:44:21,245 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279030.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:44:40,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.46 vs. limit=5.0 2023-05-17 03:44:45,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 03:44:52,286 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7894, 2.8273, 4.5713, 4.6570, 2.7614, 2.5508, 2.8495, 2.1748], device='cuda:1'), covar=tensor([0.1557, 0.3069, 0.0452, 0.0438, 0.1334, 0.2439, 0.2949, 0.4111], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0394, 0.0284, 0.0307, 0.0278, 0.0319, 0.0397, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:44:54,128 INFO [finetune.py:992] (1/2) Epoch 15, batch 7800, loss[loss=0.1566, simple_loss=0.2335, pruned_loss=0.03979, over 11808.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2555, pruned_loss=0.03813, over 2373103.85 frames. ], batch size: 26, lr: 3.47e-03, grad_scale: 16.0 2023-05-17 03:44:55,528 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279078.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:45:01,253 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.663e+02 3.247e+02 3.788e+02 7.442e+02, threshold=6.494e+02, percent-clipped=4.0 2023-05-17 03:45:27,354 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1621, 4.4569, 3.9857, 4.8985, 4.4563, 2.7404, 4.2072, 2.9147], device='cuda:1'), covar=tensor([0.0887, 0.0990, 0.1567, 0.0566, 0.1214, 0.1871, 0.1040, 0.3542], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0387, 0.0365, 0.0330, 0.0372, 0.0277, 0.0350, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:45:28,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-05-17 03:45:30,594 INFO [finetune.py:992] (1/2) Epoch 15, batch 7850, loss[loss=0.1625, simple_loss=0.2552, pruned_loss=0.03491, over 12118.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2549, pruned_loss=0.03774, over 2373550.40 frames. ], batch size: 38, lr: 3.47e-03, grad_scale: 16.0 2023-05-17 03:45:31,544 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6498, 3.7484, 3.3976, 3.2326, 3.0107, 2.8639, 3.6857, 2.4132], device='cuda:1'), covar=tensor([0.0400, 0.0152, 0.0202, 0.0218, 0.0446, 0.0387, 0.0163, 0.0502], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0167, 0.0170, 0.0192, 0.0207, 0.0202, 0.0178, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:45:33,538 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279131.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:45:45,072 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6176, 3.0249, 3.3276, 4.5052, 2.4456, 4.5004, 4.6424, 4.6547], device='cuda:1'), covar=tensor([0.0143, 0.1111, 0.0473, 0.0171, 0.1358, 0.0212, 0.0145, 0.0129], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0209, 0.0187, 0.0124, 0.0195, 0.0186, 0.0182, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:45:59,256 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279167.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:46:06,375 INFO [finetune.py:992] (1/2) Epoch 15, batch 7900, loss[loss=0.1825, simple_loss=0.274, pruned_loss=0.04553, over 11312.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2537, pruned_loss=0.0373, over 2378362.56 frames. ], batch size: 55, lr: 3.47e-03, grad_scale: 16.0 2023-05-17 03:46:13,340 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 2.661e+02 3.094e+02 3.803e+02 8.182e+02, threshold=6.187e+02, percent-clipped=2.0 2023-05-17 03:46:18,025 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279192.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:46:19,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-17 03:46:23,266 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-17 03:46:34,679 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279215.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:46:43,839 INFO [finetune.py:992] (1/2) Epoch 15, batch 7950, loss[loss=0.1563, simple_loss=0.2465, pruned_loss=0.03306, over 12103.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.254, pruned_loss=0.03733, over 2377861.15 frames. ], batch size: 32, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:47:02,626 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6353, 2.6775, 3.7815, 4.6136, 3.9290, 4.6455, 3.9642, 3.4240], device='cuda:1'), covar=tensor([0.0042, 0.0445, 0.0147, 0.0059, 0.0150, 0.0081, 0.0122, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0123, 0.0106, 0.0080, 0.0107, 0.0116, 0.0101, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:47:19,691 INFO [finetune.py:992] (1/2) Epoch 15, batch 8000, loss[loss=0.1955, simple_loss=0.2877, pruned_loss=0.05162, over 12060.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2541, pruned_loss=0.03782, over 2375013.19 frames. ], batch size: 42, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:47:26,734 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 2.777e+02 3.188e+02 3.959e+02 2.988e+03, threshold=6.375e+02, percent-clipped=3.0 2023-05-17 03:47:30,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.91 vs. limit=5.0 2023-05-17 03:47:40,535 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5027, 5.0583, 5.5145, 4.8489, 5.1536, 4.9043, 5.5495, 5.2258], device='cuda:1'), covar=tensor([0.0273, 0.0395, 0.0270, 0.0233, 0.0399, 0.0327, 0.0197, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0277, 0.0301, 0.0269, 0.0273, 0.0273, 0.0247, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:47:44,816 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3580, 2.5312, 3.1435, 4.2396, 2.1881, 4.3116, 4.3469, 4.4368], device='cuda:1'), covar=tensor([0.0145, 0.1336, 0.0517, 0.0157, 0.1433, 0.0234, 0.0158, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0208, 0.0186, 0.0125, 0.0194, 0.0185, 0.0182, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:47:55,591 INFO [finetune.py:992] (1/2) Epoch 15, batch 8050, loss[loss=0.197, simple_loss=0.2863, pruned_loss=0.05386, over 11258.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2551, pruned_loss=0.03813, over 2373613.77 frames. ], batch size: 56, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:48:00,889 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9181, 3.7761, 5.3001, 2.9279, 2.9076, 3.9225, 3.3760, 3.9216], device='cuda:1'), covar=tensor([0.0420, 0.0931, 0.0274, 0.1048, 0.1897, 0.1492, 0.1267, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0240, 0.0259, 0.0185, 0.0239, 0.0296, 0.0227, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:48:19,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-05-17 03:48:29,971 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7119, 3.3669, 5.1238, 2.5372, 2.6562, 3.7435, 3.1489, 3.7843], device='cuda:1'), covar=tensor([0.0492, 0.1158, 0.0309, 0.1278, 0.2092, 0.1508, 0.1457, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0240, 0.0260, 0.0185, 0.0240, 0.0297, 0.0227, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:48:31,829 INFO [finetune.py:992] (1/2) Epoch 15, batch 8100, loss[loss=0.1691, simple_loss=0.2617, pruned_loss=0.03819, over 12152.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2565, pruned_loss=0.03951, over 2353898.84 frames. ], batch size: 36, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:48:39,011 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 2.813e+02 3.387e+02 3.883e+02 1.132e+03, threshold=6.774e+02, percent-clipped=3.0 2023-05-17 03:48:45,670 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0505, 6.0239, 5.8137, 5.3797, 5.0989, 5.9276, 5.5500, 5.3209], device='cuda:1'), covar=tensor([0.0712, 0.0959, 0.0615, 0.1514, 0.0731, 0.0717, 0.1515, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0648, 0.0578, 0.0531, 0.0654, 0.0429, 0.0745, 0.0802, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:48:51,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-05-17 03:48:53,292 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8750, 5.6911, 5.7588, 5.1648, 5.0978, 5.9125, 4.9938, 5.2503], device='cuda:1'), covar=tensor([0.1274, 0.2036, 0.1254, 0.3065, 0.1099, 0.1787, 0.4350, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.0647, 0.0578, 0.0531, 0.0654, 0.0428, 0.0744, 0.0801, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:49:07,748 INFO [finetune.py:992] (1/2) Epoch 15, batch 8150, loss[loss=0.1749, simple_loss=0.2666, pruned_loss=0.04154, over 11838.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2569, pruned_loss=0.03963, over 2356719.44 frames. ], batch size: 44, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:49:34,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-17 03:49:44,223 INFO [finetune.py:992] (1/2) Epoch 15, batch 8200, loss[loss=0.1861, simple_loss=0.2764, pruned_loss=0.04788, over 12139.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2563, pruned_loss=0.03907, over 2363019.76 frames. ], batch size: 39, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:49:51,246 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.721e+02 3.143e+02 3.749e+02 8.958e+02, threshold=6.285e+02, percent-clipped=2.0 2023-05-17 03:49:51,363 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279487.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:49:53,490 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7518, 5.4692, 5.0187, 5.1006, 5.5485, 4.9274, 4.9798, 5.0925], device='cuda:1'), covar=tensor([0.1429, 0.0940, 0.1245, 0.1817, 0.0980, 0.2154, 0.1830, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0507, 0.0407, 0.0459, 0.0474, 0.0441, 0.0404, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:50:20,374 INFO [finetune.py:992] (1/2) Epoch 15, batch 8250, loss[loss=0.1265, simple_loss=0.2135, pruned_loss=0.01972, over 12012.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2568, pruned_loss=0.03921, over 2362923.34 frames. ], batch size: 28, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:50:21,942 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279529.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:50:31,169 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279542.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:50:39,717 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9081, 5.8911, 5.6359, 5.1813, 5.1133, 5.7874, 5.4074, 5.1709], device='cuda:1'), covar=tensor([0.0763, 0.0851, 0.0621, 0.1725, 0.0750, 0.0741, 0.1474, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0644, 0.0576, 0.0528, 0.0652, 0.0428, 0.0742, 0.0799, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:50:40,773 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 03:50:54,650 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0314, 4.7939, 4.7500, 4.8387, 4.6925, 4.9260, 4.8462, 2.5309], device='cuda:1'), covar=tensor([0.0127, 0.0066, 0.0109, 0.0080, 0.0051, 0.0095, 0.0089, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0085, 0.0076, 0.0062, 0.0096, 0.0085, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:50:55,722 INFO [finetune.py:992] (1/2) Epoch 15, batch 8300, loss[loss=0.1793, simple_loss=0.2708, pruned_loss=0.04388, over 12197.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2571, pruned_loss=0.03913, over 2363272.30 frames. ], batch size: 35, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:51:02,926 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.526e+02 2.895e+02 3.782e+02 1.118e+03, threshold=5.789e+02, percent-clipped=4.0 2023-05-17 03:51:05,286 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279590.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:51:14,739 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279603.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:51:32,059 INFO [finetune.py:992] (1/2) Epoch 15, batch 8350, loss[loss=0.147, simple_loss=0.245, pruned_loss=0.02452, over 12182.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2562, pruned_loss=0.03873, over 2372656.61 frames. ], batch size: 31, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:52:08,314 INFO [finetune.py:992] (1/2) Epoch 15, batch 8400, loss[loss=0.1627, simple_loss=0.2584, pruned_loss=0.03348, over 12309.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2549, pruned_loss=0.03834, over 2377982.11 frames. ], batch size: 34, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:52:15,368 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.699e+02 3.158e+02 3.706e+02 8.753e+02, threshold=6.316e+02, percent-clipped=3.0 2023-05-17 03:52:30,049 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7339, 2.3015, 2.9154, 3.6834, 2.0616, 3.7411, 3.6764, 3.8328], device='cuda:1'), covar=tensor([0.0167, 0.1404, 0.0519, 0.0170, 0.1476, 0.0318, 0.0268, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0206, 0.0185, 0.0124, 0.0193, 0.0184, 0.0181, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:52:43,821 INFO [finetune.py:992] (1/2) Epoch 15, batch 8450, loss[loss=0.1664, simple_loss=0.2632, pruned_loss=0.03481, over 11584.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2555, pruned_loss=0.0386, over 2377163.89 frames. ], batch size: 48, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:53:01,447 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0287, 5.9827, 5.7410, 5.3127, 5.1340, 5.9096, 5.4921, 5.2504], device='cuda:1'), covar=tensor([0.0742, 0.1010, 0.0682, 0.1899, 0.0792, 0.0737, 0.1740, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0644, 0.0575, 0.0528, 0.0651, 0.0429, 0.0742, 0.0797, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:53:20,538 INFO [finetune.py:992] (1/2) Epoch 15, batch 8500, loss[loss=0.16, simple_loss=0.2555, pruned_loss=0.03224, over 12349.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2557, pruned_loss=0.03826, over 2379586.33 frames. ], batch size: 36, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:53:27,637 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.155e+02 2.697e+02 3.107e+02 3.683e+02 6.037e+02, threshold=6.214e+02, percent-clipped=0.0 2023-05-17 03:53:27,820 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279787.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:53:39,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-05-17 03:53:41,855 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9522, 2.4134, 3.4754, 2.8766, 3.2699, 3.0715, 2.3478, 3.3633], device='cuda:1'), covar=tensor([0.0155, 0.0362, 0.0150, 0.0273, 0.0182, 0.0179, 0.0370, 0.0138], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0210, 0.0198, 0.0192, 0.0223, 0.0171, 0.0202, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:53:57,229 INFO [finetune.py:992] (1/2) Epoch 15, batch 8550, loss[loss=0.1537, simple_loss=0.2423, pruned_loss=0.03255, over 12033.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2556, pruned_loss=0.03833, over 2375033.06 frames. ], batch size: 31, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:54:02,843 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279835.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:54:32,563 INFO [finetune.py:992] (1/2) Epoch 15, batch 8600, loss[loss=0.1746, simple_loss=0.2722, pruned_loss=0.03844, over 12136.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2556, pruned_loss=0.03838, over 2373865.24 frames. ], batch size: 38, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:54:38,527 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279885.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:54:39,877 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.994e+02 2.663e+02 3.176e+02 3.689e+02 7.221e+02, threshold=6.352e+02, percent-clipped=3.0 2023-05-17 03:54:47,715 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279898.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:55:09,153 INFO [finetune.py:992] (1/2) Epoch 15, batch 8650, loss[loss=0.1634, simple_loss=0.2587, pruned_loss=0.03406, over 12356.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.255, pruned_loss=0.03833, over 2369658.83 frames. ], batch size: 36, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:55:45,474 INFO [finetune.py:992] (1/2) Epoch 15, batch 8700, loss[loss=0.1507, simple_loss=0.2399, pruned_loss=0.03079, over 12126.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2555, pruned_loss=0.03844, over 2361487.07 frames. ], batch size: 30, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:55:52,660 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 2.577e+02 3.050e+02 3.655e+02 6.083e+02, threshold=6.099e+02, percent-clipped=0.0 2023-05-17 03:56:21,889 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3435, 3.1225, 2.8822, 2.8844, 2.6800, 2.5670, 3.1546, 2.0803], device='cuda:1'), covar=tensor([0.0450, 0.0202, 0.0260, 0.0241, 0.0432, 0.0356, 0.0177, 0.0572], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0166, 0.0170, 0.0191, 0.0204, 0.0201, 0.0175, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 03:56:23,861 INFO [finetune.py:992] (1/2) Epoch 15, batch 8750, loss[loss=0.2015, simple_loss=0.2947, pruned_loss=0.05417, over 10370.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.256, pruned_loss=0.03877, over 2357234.51 frames. ], batch size: 68, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:56:39,504 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2447, 6.1993, 6.0238, 5.4850, 5.2937, 6.1405, 5.7531, 5.5431], device='cuda:1'), covar=tensor([0.0658, 0.0860, 0.0593, 0.1800, 0.0641, 0.0776, 0.1548, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0640, 0.0569, 0.0525, 0.0646, 0.0425, 0.0736, 0.0791, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 03:56:41,728 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280051.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:56:59,747 INFO [finetune.py:992] (1/2) Epoch 15, batch 8800, loss[loss=0.1903, simple_loss=0.2901, pruned_loss=0.0453, over 12191.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2561, pruned_loss=0.03895, over 2360434.31 frames. ], batch size: 35, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:57:02,813 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4149, 2.5521, 3.6563, 4.4481, 3.8199, 4.4563, 3.9197, 3.2329], device='cuda:1'), covar=tensor([0.0045, 0.0427, 0.0168, 0.0045, 0.0122, 0.0071, 0.0116, 0.0380], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0124, 0.0108, 0.0081, 0.0108, 0.0119, 0.0101, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 03:57:07,449 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 2.782e+02 3.180e+02 3.650e+02 1.240e+03, threshold=6.359e+02, percent-clipped=2.0 2023-05-17 03:57:08,391 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280088.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:57:25,374 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280112.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:57:31,613 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7703, 4.5566, 4.2432, 4.1239, 4.6430, 4.0614, 4.2140, 3.9509], device='cuda:1'), covar=tensor([0.1795, 0.1217, 0.1533, 0.2143, 0.1157, 0.2393, 0.1923, 0.1763], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0507, 0.0407, 0.0457, 0.0470, 0.0441, 0.0404, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:57:32,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.15 vs. limit=5.0 2023-05-17 03:57:35,702 INFO [finetune.py:992] (1/2) Epoch 15, batch 8850, loss[loss=0.1507, simple_loss=0.2431, pruned_loss=0.02916, over 12158.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2552, pruned_loss=0.03869, over 2359502.15 frames. ], batch size: 36, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:57:51,394 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280149.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:58:11,396 INFO [finetune.py:992] (1/2) Epoch 15, batch 8900, loss[loss=0.178, simple_loss=0.2744, pruned_loss=0.04078, over 12350.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2554, pruned_loss=0.03861, over 2368998.59 frames. ], batch size: 35, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:58:17,220 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280185.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:58:18,480 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.939e+02 2.621e+02 3.017e+02 3.657e+02 5.872e+02, threshold=6.035e+02, percent-clipped=0.0 2023-05-17 03:58:27,124 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280198.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:58:48,063 INFO [finetune.py:992] (1/2) Epoch 15, batch 8950, loss[loss=0.1669, simple_loss=0.2609, pruned_loss=0.0365, over 12261.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2551, pruned_loss=0.03841, over 2374046.04 frames. ], batch size: 37, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:58:52,988 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280233.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:58:57,197 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0003, 5.9291, 5.5919, 5.4029, 5.9756, 5.2923, 5.3944, 5.4929], device='cuda:1'), covar=tensor([0.1629, 0.0947, 0.0905, 0.2199, 0.0936, 0.2253, 0.1902, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0507, 0.0406, 0.0457, 0.0467, 0.0440, 0.0403, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 03:59:02,067 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280246.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 03:59:11,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-17 03:59:24,219 INFO [finetune.py:992] (1/2) Epoch 15, batch 9000, loss[loss=0.1555, simple_loss=0.2298, pruned_loss=0.04063, over 12276.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2554, pruned_loss=0.03847, over 2377636.16 frames. ], batch size: 28, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 03:59:24,219 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 03:59:42,467 INFO [finetune.py:1026] (1/2) Epoch 15, validation: loss=0.3191, simple_loss=0.3935, pruned_loss=0.1223, over 1020973.00 frames. 2023-05-17 03:59:42,468 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 03:59:49,614 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.626e+02 3.045e+02 3.488e+02 5.350e+02, threshold=6.090e+02, percent-clipped=0.0 2023-05-17 04:00:18,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 04:00:18,858 INFO [finetune.py:992] (1/2) Epoch 15, batch 9050, loss[loss=0.1511, simple_loss=0.2291, pruned_loss=0.03658, over 12174.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2548, pruned_loss=0.03808, over 2374650.17 frames. ], batch size: 29, lr: 3.47e-03, grad_scale: 8.0 2023-05-17 04:00:19,013 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280327.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:00:21,596 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-17 04:00:30,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 04:00:54,714 INFO [finetune.py:992] (1/2) Epoch 15, batch 9100, loss[loss=0.1747, simple_loss=0.2674, pruned_loss=0.04097, over 12310.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2558, pruned_loss=0.03821, over 2380032.69 frames. ], batch size: 34, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:01:01,653 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.720e+02 3.183e+02 3.780e+02 5.897e+02, threshold=6.365e+02, percent-clipped=0.0 2023-05-17 04:01:02,625 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280388.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:01:14,469 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2187, 3.0311, 4.6436, 2.5380, 2.5920, 3.5603, 2.9239, 3.6609], device='cuda:1'), covar=tensor([0.0574, 0.1451, 0.0437, 0.1291, 0.2177, 0.1443, 0.1682, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0244, 0.0261, 0.0186, 0.0242, 0.0300, 0.0229, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:01:16,443 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280407.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:01:30,553 INFO [finetune.py:992] (1/2) Epoch 15, batch 9150, loss[loss=0.1992, simple_loss=0.2831, pruned_loss=0.0576, over 12125.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2564, pruned_loss=0.03873, over 2365036.37 frames. ], batch size: 39, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:01:43,426 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280444.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:01:55,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-17 04:02:06,334 INFO [finetune.py:992] (1/2) Epoch 15, batch 9200, loss[loss=0.1509, simple_loss=0.2434, pruned_loss=0.02919, over 12163.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2552, pruned_loss=0.03821, over 2372419.54 frames. ], batch size: 34, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:02:14,208 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.595e+02 3.020e+02 3.681e+02 7.637e+02, threshold=6.041e+02, percent-clipped=2.0 2023-05-17 04:02:33,555 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0118, 2.3916, 3.4767, 2.9554, 3.3687, 3.1425, 2.3739, 3.4178], device='cuda:1'), covar=tensor([0.0142, 0.0374, 0.0141, 0.0266, 0.0140, 0.0166, 0.0350, 0.0127], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0208, 0.0197, 0.0190, 0.0220, 0.0170, 0.0199, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:02:33,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 04:02:36,238 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280518.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:02:42,324 INFO [finetune.py:992] (1/2) Epoch 15, batch 9250, loss[loss=0.1808, simple_loss=0.2716, pruned_loss=0.04498, over 12106.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2549, pruned_loss=0.03767, over 2376837.76 frames. ], batch size: 38, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:02:46,658 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280533.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:02:53,050 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6860, 2.2602, 2.9585, 2.6329, 2.8475, 2.7962, 2.1615, 2.8857], device='cuda:1'), covar=tensor([0.0142, 0.0386, 0.0179, 0.0256, 0.0167, 0.0191, 0.0337, 0.0179], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0208, 0.0198, 0.0191, 0.0221, 0.0171, 0.0200, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:02:56,439 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280547.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:03:01,594 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6335, 3.2599, 5.0776, 2.5596, 2.8542, 3.8116, 3.1379, 3.8424], device='cuda:1'), covar=tensor([0.0439, 0.1163, 0.0273, 0.1181, 0.1815, 0.1361, 0.1366, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0244, 0.0262, 0.0186, 0.0242, 0.0301, 0.0229, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:03:03,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-05-17 04:03:17,700 INFO [finetune.py:992] (1/2) Epoch 15, batch 9300, loss[loss=0.1933, simple_loss=0.2777, pruned_loss=0.05447, over 12132.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.255, pruned_loss=0.03777, over 2381876.65 frames. ], batch size: 38, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:03:19,301 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280579.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:03:24,705 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.780e+02 3.170e+02 3.668e+02 7.596e+02, threshold=6.340e+02, percent-clipped=1.0 2023-05-17 04:03:30,508 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280594.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:03:40,696 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280608.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:03:44,990 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280614.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:03:54,232 INFO [finetune.py:992] (1/2) Epoch 15, batch 9350, loss[loss=0.2762, simple_loss=0.3384, pruned_loss=0.107, over 7810.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2557, pruned_loss=0.0381, over 2369905.90 frames. ], batch size: 99, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:04:25,104 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8208, 2.9306, 4.5609, 4.8512, 2.8898, 2.6760, 3.0658, 2.1808], device='cuda:1'), covar=tensor([0.1645, 0.2921, 0.0520, 0.0385, 0.1391, 0.2535, 0.2751, 0.4150], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0393, 0.0284, 0.0308, 0.0277, 0.0317, 0.0397, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:04:25,799 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5606, 2.5563, 3.2741, 4.4625, 2.4922, 4.5685, 4.5900, 4.6083], device='cuda:1'), covar=tensor([0.0157, 0.1402, 0.0556, 0.0142, 0.1388, 0.0188, 0.0165, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0207, 0.0186, 0.0125, 0.0194, 0.0184, 0.0180, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:04:29,260 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280675.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:04:30,486 INFO [finetune.py:992] (1/2) Epoch 15, batch 9400, loss[loss=0.1591, simple_loss=0.2463, pruned_loss=0.03596, over 12076.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2556, pruned_loss=0.03821, over 2369079.63 frames. ], batch size: 32, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:04:34,530 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280683.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:04:37,244 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.825e+02 3.289e+02 3.761e+02 8.250e+02, threshold=6.578e+02, percent-clipped=2.0 2023-05-17 04:04:47,990 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280702.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:04:51,503 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280707.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:05:04,917 INFO [finetune.py:992] (1/2) Epoch 15, batch 9450, loss[loss=0.1608, simple_loss=0.2477, pruned_loss=0.03689, over 12169.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2555, pruned_loss=0.03825, over 2370731.35 frames. ], batch size: 31, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:05:17,827 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280744.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:05:24,202 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4380, 2.3682, 3.6508, 4.4206, 3.9119, 4.4464, 3.8435, 3.4542], device='cuda:1'), covar=tensor([0.0042, 0.0417, 0.0158, 0.0041, 0.0119, 0.0062, 0.0145, 0.0287], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0123, 0.0106, 0.0080, 0.0106, 0.0117, 0.0100, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:05:25,458 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280755.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:05:29,006 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4954, 5.0685, 5.4670, 4.8767, 5.1846, 4.9212, 5.5263, 5.0834], device='cuda:1'), covar=tensor([0.0256, 0.0328, 0.0241, 0.0208, 0.0312, 0.0270, 0.0165, 0.0277], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0278, 0.0301, 0.0271, 0.0273, 0.0274, 0.0248, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:05:31,214 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280763.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:05:34,860 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4309, 3.4281, 3.1333, 3.1393, 2.8167, 2.6069, 3.4399, 2.2961], device='cuda:1'), covar=tensor([0.0460, 0.0163, 0.0219, 0.0209, 0.0440, 0.0437, 0.0157, 0.0530], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0167, 0.0170, 0.0192, 0.0205, 0.0203, 0.0177, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:05:41,016 INFO [finetune.py:992] (1/2) Epoch 15, batch 9500, loss[loss=0.2397, simple_loss=0.3157, pruned_loss=0.08184, over 8532.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.256, pruned_loss=0.03887, over 2355970.60 frames. ], batch size: 99, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:05:48,703 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.589e+02 3.024e+02 3.554e+02 7.083e+02, threshold=6.048e+02, percent-clipped=2.0 2023-05-17 04:05:52,258 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280792.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:06:17,002 INFO [finetune.py:992] (1/2) Epoch 15, batch 9550, loss[loss=0.1502, simple_loss=0.2374, pruned_loss=0.03145, over 12363.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2565, pruned_loss=0.03888, over 2355039.03 frames. ], batch size: 30, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:06:22,043 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1039, 3.6367, 5.4944, 2.9334, 3.1002, 3.9149, 3.4611, 3.9347], device='cuda:1'), covar=tensor([0.0393, 0.1146, 0.0251, 0.1195, 0.1919, 0.1632, 0.1388, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0244, 0.0262, 0.0186, 0.0242, 0.0300, 0.0229, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:06:50,982 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280874.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:06:52,930 INFO [finetune.py:992] (1/2) Epoch 15, batch 9600, loss[loss=0.1711, simple_loss=0.2621, pruned_loss=0.04003, over 12180.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.256, pruned_loss=0.03865, over 2360658.57 frames. ], batch size: 29, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:07:00,195 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.610e+02 3.063e+02 3.778e+02 8.298e+02, threshold=6.127e+02, percent-clipped=2.0 2023-05-17 04:07:01,777 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280889.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:07:04,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0 2023-05-17 04:07:11,650 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280903.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:07:29,233 INFO [finetune.py:992] (1/2) Epoch 15, batch 9650, loss[loss=0.1628, simple_loss=0.2546, pruned_loss=0.0355, over 12308.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2558, pruned_loss=0.03851, over 2372962.32 frames. ], batch size: 34, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:07:51,954 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-17 04:08:00,058 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280970.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:08:05,058 INFO [finetune.py:992] (1/2) Epoch 15, batch 9700, loss[loss=0.1737, simple_loss=0.2558, pruned_loss=0.04577, over 12328.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2551, pruned_loss=0.03808, over 2377080.15 frames. ], batch size: 31, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:08:09,446 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280983.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:08:12,217 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.705e+02 3.092e+02 3.654e+02 5.516e+02, threshold=6.184e+02, percent-clipped=0.0 2023-05-17 04:08:38,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-05-17 04:08:41,605 INFO [finetune.py:992] (1/2) Epoch 15, batch 9750, loss[loss=0.152, simple_loss=0.2492, pruned_loss=0.02743, over 12185.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2553, pruned_loss=0.03816, over 2362066.97 frames. ], batch size: 35, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:08:44,396 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281031.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:08:57,463 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281049.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 04:09:03,901 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281058.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:09:17,866 INFO [finetune.py:992] (1/2) Epoch 15, batch 9800, loss[loss=0.15, simple_loss=0.2354, pruned_loss=0.03233, over 12185.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2551, pruned_loss=0.03837, over 2365536.65 frames. ], batch size: 31, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:09:25,046 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.638e+02 3.044e+02 3.630e+02 1.190e+03, threshold=6.087e+02, percent-clipped=4.0 2023-05-17 04:09:36,484 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4988, 5.2925, 5.4264, 5.4465, 5.0526, 5.1477, 4.8936, 5.3458], device='cuda:1'), covar=tensor([0.0668, 0.0529, 0.0861, 0.0571, 0.1876, 0.1253, 0.0564, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0734, 0.0646, 0.0666, 0.0893, 0.0773, 0.0598, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:09:41,519 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281110.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 04:09:53,393 INFO [finetune.py:992] (1/2) Epoch 15, batch 9850, loss[loss=0.1635, simple_loss=0.2446, pruned_loss=0.04114, over 12256.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2544, pruned_loss=0.03783, over 2371092.17 frames. ], batch size: 32, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:10:01,225 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0691, 4.9047, 4.8710, 4.9685, 4.5844, 5.0424, 5.0286, 5.2855], device='cuda:1'), covar=tensor([0.0316, 0.0177, 0.0204, 0.0296, 0.0770, 0.0299, 0.0173, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0203, 0.0197, 0.0256, 0.0248, 0.0227, 0.0182, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 04:10:03,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.91 vs. limit=5.0 2023-05-17 04:10:20,513 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8463, 3.7023, 3.8175, 3.8333, 3.2963, 3.3595, 3.4709, 3.6579], device='cuda:1'), covar=tensor([0.1392, 0.1255, 0.1521, 0.1002, 0.2843, 0.2149, 0.0937, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0727, 0.0641, 0.0660, 0.0887, 0.0767, 0.0593, 0.0500], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:10:23,932 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281169.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:10:27,321 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281174.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:10:29,385 INFO [finetune.py:992] (1/2) Epoch 15, batch 9900, loss[loss=0.1728, simple_loss=0.2639, pruned_loss=0.04091, over 12195.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2551, pruned_loss=0.03817, over 2365892.85 frames. ], batch size: 29, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:10:34,989 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8731, 3.4366, 5.2592, 2.9556, 2.8988, 3.7966, 3.2117, 3.7546], device='cuda:1'), covar=tensor([0.0413, 0.1147, 0.0263, 0.1089, 0.2049, 0.1502, 0.1431, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0244, 0.0262, 0.0186, 0.0241, 0.0301, 0.0229, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:10:36,014 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.700e+02 3.215e+02 3.866e+02 9.804e+02, threshold=6.430e+02, percent-clipped=3.0 2023-05-17 04:10:37,680 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281189.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:10:43,278 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5150, 2.6773, 3.7153, 4.4241, 3.9718, 4.4912, 3.8842, 3.3428], device='cuda:1'), covar=tensor([0.0042, 0.0393, 0.0153, 0.0054, 0.0119, 0.0070, 0.0137, 0.0354], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0126, 0.0108, 0.0082, 0.0108, 0.0120, 0.0102, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:10:47,833 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281203.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:11:01,820 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281222.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:11:05,378 INFO [finetune.py:992] (1/2) Epoch 15, batch 9950, loss[loss=0.1748, simple_loss=0.2625, pruned_loss=0.04351, over 12098.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2551, pruned_loss=0.0382, over 2372281.14 frames. ], batch size: 32, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:11:07,664 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281230.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 04:11:12,382 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281237.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:11:22,418 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281251.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:11:35,959 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281270.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:11:40,902 INFO [finetune.py:992] (1/2) Epoch 15, batch 10000, loss[loss=0.1764, simple_loss=0.2653, pruned_loss=0.04378, over 12122.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2545, pruned_loss=0.03796, over 2381646.14 frames. ], batch size: 39, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:11:47,836 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.734e+02 3.185e+02 3.967e+02 7.095e+02, threshold=6.371e+02, percent-clipped=3.0 2023-05-17 04:12:10,698 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281318.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:12:16,990 INFO [finetune.py:992] (1/2) Epoch 15, batch 10050, loss[loss=0.161, simple_loss=0.2601, pruned_loss=0.03098, over 12186.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2543, pruned_loss=0.03783, over 2375811.57 frames. ], batch size: 35, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:12:20,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-17 04:12:29,258 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3057, 4.7403, 4.1293, 5.0467, 4.4315, 2.9116, 4.2606, 3.0093], device='cuda:1'), covar=tensor([0.0815, 0.0791, 0.1440, 0.0523, 0.1212, 0.1711, 0.1046, 0.3333], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0386, 0.0365, 0.0329, 0.0375, 0.0279, 0.0351, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:12:38,950 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281358.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:12:53,027 INFO [finetune.py:992] (1/2) Epoch 15, batch 10100, loss[loss=0.1649, simple_loss=0.2606, pruned_loss=0.03459, over 11813.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2544, pruned_loss=0.03762, over 2378345.97 frames. ], batch size: 44, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:13:00,141 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 2.617e+02 3.084e+02 3.596e+02 6.290e+02, threshold=6.168e+02, percent-clipped=0.0 2023-05-17 04:13:02,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-17 04:13:13,232 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281405.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:13:13,934 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281406.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:13:22,743 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1800, 2.7214, 3.7082, 3.1700, 3.5382, 3.2673, 2.7881, 3.5901], device='cuda:1'), covar=tensor([0.0168, 0.0335, 0.0165, 0.0264, 0.0195, 0.0203, 0.0343, 0.0173], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0212, 0.0201, 0.0193, 0.0225, 0.0173, 0.0203, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:13:28,424 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281426.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:13:28,946 INFO [finetune.py:992] (1/2) Epoch 15, batch 10150, loss[loss=0.1658, simple_loss=0.2628, pruned_loss=0.03441, over 12195.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2546, pruned_loss=0.03754, over 2383583.65 frames. ], batch size: 35, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:13:42,678 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5671, 4.9304, 4.3564, 5.1514, 4.7623, 2.7533, 4.2834, 3.1788], device='cuda:1'), covar=tensor([0.0777, 0.0648, 0.1293, 0.0574, 0.1127, 0.1834, 0.1134, 0.2985], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0386, 0.0364, 0.0329, 0.0374, 0.0278, 0.0351, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:13:49,609 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281456.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:14:04,822 INFO [finetune.py:992] (1/2) Epoch 15, batch 10200, loss[loss=0.1492, simple_loss=0.2432, pruned_loss=0.02763, over 11777.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2543, pruned_loss=0.03746, over 2386840.27 frames. ], batch size: 44, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:14:11,941 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.679e+02 3.131e+02 3.626e+02 7.226e+02, threshold=6.261e+02, percent-clipped=3.0 2023-05-17 04:14:12,149 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281487.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:14:34,096 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281517.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:14:39,713 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281525.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 04:14:40,869 INFO [finetune.py:992] (1/2) Epoch 15, batch 10250, loss[loss=0.1536, simple_loss=0.2326, pruned_loss=0.03732, over 11994.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2543, pruned_loss=0.03759, over 2378338.85 frames. ], batch size: 28, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:15:03,070 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5956, 5.1180, 5.5833, 4.9650, 5.2230, 5.0607, 5.6127, 5.2345], device='cuda:1'), covar=tensor([0.0268, 0.0377, 0.0227, 0.0230, 0.0392, 0.0291, 0.0184, 0.0261], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0281, 0.0301, 0.0272, 0.0276, 0.0275, 0.0249, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:15:16,389 INFO [finetune.py:992] (1/2) Epoch 15, batch 10300, loss[loss=0.1516, simple_loss=0.2436, pruned_loss=0.02977, over 12358.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2547, pruned_loss=0.0378, over 2374452.70 frames. ], batch size: 35, lr: 3.46e-03, grad_scale: 16.0 2023-05-17 04:15:23,592 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.676e+02 3.123e+02 3.776e+02 1.681e+03, threshold=6.246e+02, percent-clipped=5.0 2023-05-17 04:15:23,907 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4580, 4.8743, 4.2203, 5.1276, 4.5803, 3.0439, 4.3809, 3.0300], device='cuda:1'), covar=tensor([0.0725, 0.0725, 0.1466, 0.0429, 0.1154, 0.1581, 0.0980, 0.3230], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0383, 0.0362, 0.0327, 0.0372, 0.0276, 0.0348, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:15:53,068 INFO [finetune.py:992] (1/2) Epoch 15, batch 10350, loss[loss=0.1805, simple_loss=0.2769, pruned_loss=0.04199, over 11926.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2541, pruned_loss=0.03742, over 2381644.17 frames. ], batch size: 44, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:16:27,333 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1068, 5.8618, 5.4353, 5.3979, 5.9538, 5.2761, 5.4641, 5.5106], device='cuda:1'), covar=tensor([0.1314, 0.0940, 0.1113, 0.1826, 0.0864, 0.1975, 0.1765, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0514, 0.0409, 0.0464, 0.0475, 0.0444, 0.0410, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:16:29,396 INFO [finetune.py:992] (1/2) Epoch 15, batch 10400, loss[loss=0.1371, simple_loss=0.2289, pruned_loss=0.02267, over 12150.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2534, pruned_loss=0.03693, over 2380280.88 frames. ], batch size: 30, lr: 3.46e-03, grad_scale: 8.0 2023-05-17 04:16:37,414 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.484e+02 2.960e+02 3.585e+02 5.106e+02, threshold=5.920e+02, percent-clipped=0.0 2023-05-17 04:16:49,576 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281705.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:16:53,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-05-17 04:16:55,415 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2174, 3.6588, 3.8589, 4.2330, 3.0562, 3.5984, 2.4401, 3.8137], device='cuda:1'), covar=tensor([0.1604, 0.0844, 0.0910, 0.0693, 0.1053, 0.0712, 0.1900, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0266, 0.0300, 0.0359, 0.0242, 0.0245, 0.0261, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:17:05,132 INFO [finetune.py:992] (1/2) Epoch 15, batch 10450, loss[loss=0.1435, simple_loss=0.2327, pruned_loss=0.02717, over 12283.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2539, pruned_loss=0.03759, over 2367915.58 frames. ], batch size: 33, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:17:14,342 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0980, 4.8890, 5.2075, 5.0947, 4.2711, 4.4323, 4.5202, 4.8584], device='cuda:1'), covar=tensor([0.0931, 0.1016, 0.0832, 0.0795, 0.3097, 0.2106, 0.0687, 0.1492], device='cuda:1'), in_proj_covar=tensor([0.0560, 0.0722, 0.0631, 0.0656, 0.0878, 0.0762, 0.0586, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:17:23,906 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281753.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:17:40,764 INFO [finetune.py:992] (1/2) Epoch 15, batch 10500, loss[loss=0.1883, simple_loss=0.2771, pruned_loss=0.04976, over 10308.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2551, pruned_loss=0.03838, over 2356433.78 frames. ], batch size: 68, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:17:44,520 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281782.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:17:48,815 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.691e+02 3.266e+02 3.969e+02 1.128e+03, threshold=6.532e+02, percent-clipped=4.0 2023-05-17 04:18:07,141 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281812.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:18:16,341 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281825.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:18:17,608 INFO [finetune.py:992] (1/2) Epoch 15, batch 10550, loss[loss=0.1458, simple_loss=0.242, pruned_loss=0.02476, over 12190.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2542, pruned_loss=0.03777, over 2370633.65 frames. ], batch size: 35, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:18:35,722 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281852.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:18:50,610 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281873.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:18:53,547 INFO [finetune.py:992] (1/2) Epoch 15, batch 10600, loss[loss=0.1648, simple_loss=0.2496, pruned_loss=0.03996, over 12035.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2533, pruned_loss=0.03717, over 2380040.15 frames. ], batch size: 31, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:19:01,391 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.591e+02 2.933e+02 3.502e+02 6.423e+02, threshold=5.866e+02, percent-clipped=0.0 2023-05-17 04:19:09,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 04:19:18,098 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281910.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:19:20,433 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281913.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:19:29,925 INFO [finetune.py:992] (1/2) Epoch 15, batch 10650, loss[loss=0.2258, simple_loss=0.3045, pruned_loss=0.07354, over 8063.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2538, pruned_loss=0.03739, over 2373216.87 frames. ], batch size: 98, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:19:32,972 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6265, 3.5014, 3.1052, 3.1548, 2.7309, 2.6395, 3.5788, 2.2812], device='cuda:1'), covar=tensor([0.0404, 0.0147, 0.0249, 0.0229, 0.0486, 0.0444, 0.0150, 0.0594], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0168, 0.0173, 0.0194, 0.0207, 0.0206, 0.0178, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:19:43,609 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281945.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:20:02,195 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281971.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 04:20:06,166 INFO [finetune.py:992] (1/2) Epoch 15, batch 10700, loss[loss=0.1579, simple_loss=0.2474, pruned_loss=0.03418, over 12287.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2534, pruned_loss=0.03735, over 2373607.44 frames. ], batch size: 33, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:20:14,068 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.695e+02 3.139e+02 3.686e+02 1.665e+03, threshold=6.278e+02, percent-clipped=5.0 2023-05-17 04:20:30,548 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282006.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:20:45,556 INFO [finetune.py:992] (1/2) Epoch 15, batch 10750, loss[loss=0.1841, simple_loss=0.2755, pruned_loss=0.04636, over 12256.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2539, pruned_loss=0.03757, over 2370352.51 frames. ], batch size: 37, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:20:55,015 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5953, 2.1067, 2.9001, 3.5738, 1.9905, 3.6817, 3.5600, 3.7580], device='cuda:1'), covar=tensor([0.0205, 0.1576, 0.0571, 0.0230, 0.1563, 0.0332, 0.0294, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0207, 0.0186, 0.0125, 0.0193, 0.0185, 0.0181, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:21:21,464 INFO [finetune.py:992] (1/2) Epoch 15, batch 10800, loss[loss=0.1587, simple_loss=0.2367, pruned_loss=0.04036, over 12033.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2546, pruned_loss=0.0378, over 2372514.00 frames. ], batch size: 28, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:21:25,233 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282082.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:21:30,152 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.997e+02 2.673e+02 3.156e+02 3.969e+02 9.157e+02, threshold=6.313e+02, percent-clipped=4.0 2023-05-17 04:21:35,931 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282096.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:21:47,240 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282112.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:21:57,546 INFO [finetune.py:992] (1/2) Epoch 15, batch 10850, loss[loss=0.1922, simple_loss=0.284, pruned_loss=0.05022, over 10811.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2545, pruned_loss=0.03764, over 2375611.34 frames. ], batch size: 69, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:21:59,793 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282130.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:22:19,640 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282157.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:22:21,662 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282160.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:22:34,242 INFO [finetune.py:992] (1/2) Epoch 15, batch 10900, loss[loss=0.1794, simple_loss=0.2731, pruned_loss=0.04279, over 12351.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2552, pruned_loss=0.03821, over 2370229.72 frames. ], batch size: 35, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:22:42,559 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.563e+02 2.975e+02 3.620e+02 1.018e+03, threshold=5.950e+02, percent-clipped=3.0 2023-05-17 04:22:57,245 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282208.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:23:10,294 INFO [finetune.py:992] (1/2) Epoch 15, batch 10950, loss[loss=0.1953, simple_loss=0.2793, pruned_loss=0.05565, over 11769.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2567, pruned_loss=0.03929, over 2353458.01 frames. ], batch size: 44, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:23:20,482 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2701, 4.6779, 4.1912, 5.0914, 4.6212, 3.1231, 4.2219, 3.1092], device='cuda:1'), covar=tensor([0.0938, 0.0862, 0.1491, 0.0421, 0.1157, 0.1663, 0.1129, 0.3507], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0388, 0.0367, 0.0333, 0.0375, 0.0278, 0.0353, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:23:20,506 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8266, 3.0334, 4.7633, 4.9553, 2.8962, 2.7959, 3.1351, 2.2773], device='cuda:1'), covar=tensor([0.1637, 0.2853, 0.0400, 0.0389, 0.1322, 0.2397, 0.2570, 0.4120], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0394, 0.0284, 0.0307, 0.0277, 0.0317, 0.0397, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:23:38,146 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8737, 3.6319, 3.7450, 3.8370, 3.5268, 3.9124, 3.9117, 3.9829], device='cuda:1'), covar=tensor([0.0283, 0.0249, 0.0222, 0.0472, 0.0667, 0.0544, 0.0220, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0203, 0.0198, 0.0256, 0.0249, 0.0228, 0.0182, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 04:23:38,844 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282266.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 04:23:46,256 INFO [finetune.py:992] (1/2) Epoch 15, batch 11000, loss[loss=0.1582, simple_loss=0.2473, pruned_loss=0.03448, over 12179.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.258, pruned_loss=0.04006, over 2343874.22 frames. ], batch size: 31, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:23:54,115 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.897e+02 3.408e+02 4.046e+02 8.329e+02, threshold=6.815e+02, percent-clipped=5.0 2023-05-17 04:24:03,461 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282301.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:24:08,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-17 04:24:22,505 INFO [finetune.py:992] (1/2) Epoch 15, batch 11050, loss[loss=0.1886, simple_loss=0.2767, pruned_loss=0.05024, over 12282.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2608, pruned_loss=0.04144, over 2306643.88 frames. ], batch size: 37, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:24:28,837 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0036, 5.9904, 5.7386, 5.2820, 5.1919, 5.9267, 5.5600, 5.3155], device='cuda:1'), covar=tensor([0.0794, 0.0896, 0.0651, 0.1556, 0.0727, 0.0659, 0.1425, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0642, 0.0569, 0.0521, 0.0644, 0.0421, 0.0735, 0.0791, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:24:35,876 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4192, 4.8637, 3.1366, 2.8651, 4.1404, 2.8128, 4.1164, 3.5251], device='cuda:1'), covar=tensor([0.0771, 0.0562, 0.1078, 0.1440, 0.0326, 0.1282, 0.0478, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0266, 0.0181, 0.0206, 0.0147, 0.0187, 0.0202, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:24:58,051 INFO [finetune.py:992] (1/2) Epoch 15, batch 11100, loss[loss=0.2328, simple_loss=0.3226, pruned_loss=0.07156, over 7879.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2633, pruned_loss=0.04273, over 2277996.25 frames. ], batch size: 99, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:25:05,859 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.138e+02 2.930e+02 3.609e+02 4.369e+02 7.372e+02, threshold=7.218e+02, percent-clipped=3.0 2023-05-17 04:25:18,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 04:25:21,485 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6219, 4.3606, 4.6101, 4.1932, 4.3632, 4.1691, 4.6344, 4.2312], device='cuda:1'), covar=tensor([0.0325, 0.0401, 0.0309, 0.0273, 0.0468, 0.0387, 0.0238, 0.0628], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0278, 0.0298, 0.0270, 0.0275, 0.0273, 0.0248, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:25:33,896 INFO [finetune.py:992] (1/2) Epoch 15, batch 11150, loss[loss=0.1323, simple_loss=0.2173, pruned_loss=0.02367, over 12130.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2684, pruned_loss=0.04584, over 2228362.22 frames. ], batch size: 30, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:25:49,770 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8002, 5.4093, 5.1178, 5.0740, 5.5446, 4.8664, 5.0545, 5.0366], device='cuda:1'), covar=tensor([0.1440, 0.1073, 0.1264, 0.1829, 0.0977, 0.2292, 0.1853, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0509, 0.0404, 0.0455, 0.0468, 0.0440, 0.0404, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:25:51,916 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282452.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:26:08,905 INFO [finetune.py:992] (1/2) Epoch 15, batch 11200, loss[loss=0.2234, simple_loss=0.3225, pruned_loss=0.06215, over 10445.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.275, pruned_loss=0.05023, over 2163729.59 frames. ], batch size: 69, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:26:17,124 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 3.308e+02 3.948e+02 4.776e+02 9.685e+02, threshold=7.896e+02, percent-clipped=3.0 2023-05-17 04:26:17,984 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282489.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:26:23,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-17 04:26:24,216 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4148, 3.4953, 3.1860, 3.0913, 2.8688, 2.7021, 3.3133, 2.2869], device='cuda:1'), covar=tensor([0.0463, 0.0190, 0.0218, 0.0220, 0.0425, 0.0382, 0.0228, 0.0578], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0168, 0.0172, 0.0193, 0.0205, 0.0205, 0.0179, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:26:31,743 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282508.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:26:41,616 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1924, 2.9916, 2.8590, 2.7744, 2.6179, 2.4615, 2.6691, 1.9730], device='cuda:1'), covar=tensor([0.0424, 0.0140, 0.0184, 0.0200, 0.0340, 0.0272, 0.0240, 0.0567], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0167, 0.0171, 0.0193, 0.0204, 0.0204, 0.0178, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:26:44,878 INFO [finetune.py:992] (1/2) Epoch 15, batch 11250, loss[loss=0.1787, simple_loss=0.2713, pruned_loss=0.04308, over 12289.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2824, pruned_loss=0.05541, over 2087290.55 frames. ], batch size: 33, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:27:01,257 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282550.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:27:05,971 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282556.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:27:12,963 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282566.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:27:20,975 INFO [finetune.py:992] (1/2) Epoch 15, batch 11300, loss[loss=0.1943, simple_loss=0.2883, pruned_loss=0.05017, over 12035.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2875, pruned_loss=0.05832, over 2052272.86 frames. ], batch size: 37, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:27:28,508 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.112e+02 3.356e+02 3.951e+02 4.972e+02 8.345e+02, threshold=7.901e+02, percent-clipped=2.0 2023-05-17 04:27:37,914 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282601.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:27:38,720 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1306, 2.5026, 3.7083, 4.1066, 4.0555, 4.1235, 3.9478, 2.8095], device='cuda:1'), covar=tensor([0.0049, 0.0416, 0.0120, 0.0076, 0.0090, 0.0090, 0.0107, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0121, 0.0104, 0.0078, 0.0103, 0.0115, 0.0099, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:27:46,828 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282614.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:27:55,934 INFO [finetune.py:992] (1/2) Epoch 15, batch 11350, loss[loss=0.2588, simple_loss=0.3296, pruned_loss=0.09405, over 6279.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2923, pruned_loss=0.06142, over 1990792.46 frames. ], batch size: 98, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:28:11,638 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282649.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:28:14,443 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1340, 3.0166, 2.9203, 2.8084, 2.6423, 2.5000, 2.7138, 2.0118], device='cuda:1'), covar=tensor([0.0463, 0.0153, 0.0179, 0.0213, 0.0309, 0.0282, 0.0241, 0.0551], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0165, 0.0169, 0.0191, 0.0203, 0.0202, 0.0177, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:28:17,900 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282658.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:28:30,528 INFO [finetune.py:992] (1/2) Epoch 15, batch 11400, loss[loss=0.1897, simple_loss=0.2758, pruned_loss=0.05178, over 11006.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2972, pruned_loss=0.06471, over 1945043.81 frames. ], batch size: 55, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:28:30,694 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282677.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:28:37,839 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.754e+02 3.597e+02 4.130e+02 4.941e+02 1.183e+03, threshold=8.259e+02, percent-clipped=2.0 2023-05-17 04:28:45,445 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282698.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 04:28:46,818 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7610, 3.1256, 2.4205, 2.1821, 2.8149, 2.2815, 2.9903, 2.5543], device='cuda:1'), covar=tensor([0.0646, 0.0474, 0.0965, 0.1525, 0.0267, 0.1176, 0.0503, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0256, 0.0177, 0.0201, 0.0143, 0.0182, 0.0196, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:28:56,878 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 04:29:00,742 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282719.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:29:05,962 INFO [finetune.py:992] (1/2) Epoch 15, batch 11450, loss[loss=0.2532, simple_loss=0.3259, pruned_loss=0.09019, over 6883.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.3001, pruned_loss=0.06716, over 1900969.50 frames. ], batch size: 99, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:29:13,526 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282738.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:29:22,631 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282752.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:29:27,882 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282759.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 04:29:40,755 INFO [finetune.py:992] (1/2) Epoch 15, batch 11500, loss[loss=0.272, simple_loss=0.3371, pruned_loss=0.1035, over 6407.00 frames. ], tot_loss[loss=0.221, simple_loss=0.3034, pruned_loss=0.0693, over 1874564.06 frames. ], batch size: 98, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:29:48,011 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.245e+02 3.457e+02 4.104e+02 4.955e+02 2.185e+03, threshold=8.208e+02, percent-clipped=6.0 2023-05-17 04:29:56,377 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282800.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:29:59,998 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282805.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:30:15,597 INFO [finetune.py:992] (1/2) Epoch 15, batch 11550, loss[loss=0.2038, simple_loss=0.283, pruned_loss=0.06235, over 11259.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3049, pruned_loss=0.07059, over 1854786.38 frames. ], batch size: 55, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:30:27,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 04:30:28,546 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282845.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:30:42,783 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282866.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:30:49,820 INFO [finetune.py:992] (1/2) Epoch 15, batch 11600, loss[loss=0.2465, simple_loss=0.3181, pruned_loss=0.08743, over 7068.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3066, pruned_loss=0.07251, over 1816989.10 frames. ], batch size: 99, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:30:58,117 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.453e+02 3.489e+02 4.044e+02 4.459e+02 7.552e+02, threshold=8.088e+02, percent-clipped=0.0 2023-05-17 04:31:27,094 INFO [finetune.py:992] (1/2) Epoch 15, batch 11650, loss[loss=0.2368, simple_loss=0.3064, pruned_loss=0.08364, over 7062.00 frames. ], tot_loss[loss=0.227, simple_loss=0.307, pruned_loss=0.07351, over 1786773.25 frames. ], batch size: 98, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:31:30,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-17 04:32:02,322 INFO [finetune.py:992] (1/2) Epoch 15, batch 11700, loss[loss=0.2117, simple_loss=0.3019, pruned_loss=0.06073, over 10957.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3068, pruned_loss=0.07415, over 1751510.03 frames. ], batch size: 55, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:32:09,579 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.623e+02 3.290e+02 3.801e+02 4.370e+02 9.229e+02, threshold=7.602e+02, percent-clipped=4.0 2023-05-17 04:32:28,146 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283014.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:32:37,614 INFO [finetune.py:992] (1/2) Epoch 15, batch 11750, loss[loss=0.1985, simple_loss=0.2964, pruned_loss=0.05026, over 10391.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3075, pruned_loss=0.07498, over 1722121.72 frames. ], batch size: 68, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:32:41,897 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283033.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:32:55,932 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283054.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 04:33:12,090 INFO [finetune.py:992] (1/2) Epoch 15, batch 11800, loss[loss=0.2259, simple_loss=0.3163, pruned_loss=0.06776, over 10534.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3088, pruned_loss=0.07561, over 1728982.23 frames. ], batch size: 68, lr: 3.45e-03, grad_scale: 8.0 2023-05-17 04:33:18,452 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9253, 2.1453, 2.1917, 2.1337, 1.9316, 2.0514, 1.9625, 1.6646], device='cuda:1'), covar=tensor([0.0340, 0.0201, 0.0198, 0.0234, 0.0329, 0.0257, 0.0236, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0162, 0.0165, 0.0188, 0.0199, 0.0199, 0.0174, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:33:20,193 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.485e+02 3.503e+02 4.210e+02 5.100e+02 7.921e+02, threshold=8.420e+02, percent-clipped=1.0 2023-05-17 04:33:29,707 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9385, 2.3198, 3.3472, 2.9539, 3.2965, 3.2035, 2.3121, 3.3968], device='cuda:1'), covar=tensor([0.0143, 0.0421, 0.0121, 0.0258, 0.0129, 0.0166, 0.0430, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0200, 0.0185, 0.0181, 0.0209, 0.0161, 0.0191, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:33:35,800 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8929, 4.7268, 4.8576, 4.8824, 4.5677, 4.6197, 4.4620, 4.7769], device='cuda:1'), covar=tensor([0.0760, 0.0611, 0.0863, 0.0657, 0.1784, 0.1325, 0.0546, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0665, 0.0584, 0.0605, 0.0796, 0.0704, 0.0543, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:33:37,274 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3240, 3.1677, 3.0742, 3.3435, 2.6676, 3.1105, 2.6798, 2.7273], device='cuda:1'), covar=tensor([0.1518, 0.0868, 0.0805, 0.0538, 0.0985, 0.0780, 0.1597, 0.0630], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0267, 0.0296, 0.0353, 0.0240, 0.0244, 0.0261, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:33:38,576 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283115.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:33:40,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-05-17 04:33:47,131 INFO [finetune.py:992] (1/2) Epoch 15, batch 11850, loss[loss=0.258, simple_loss=0.3253, pruned_loss=0.09533, over 6620.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3097, pruned_loss=0.07595, over 1706087.38 frames. ], batch size: 97, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:33:58,998 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283144.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:34:00,313 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283145.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:34:11,026 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283161.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:34:11,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-05-17 04:34:21,602 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283176.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:34:22,047 INFO [finetune.py:992] (1/2) Epoch 15, batch 11900, loss[loss=0.2048, simple_loss=0.2966, pruned_loss=0.05651, over 11251.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3089, pruned_loss=0.07452, over 1715087.46 frames. ], batch size: 55, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:34:22,944 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7939, 3.5960, 3.6861, 3.7748, 3.4578, 3.8583, 3.8152, 3.8733], device='cuda:1'), covar=tensor([0.0282, 0.0186, 0.0221, 0.0324, 0.0610, 0.0375, 0.0232, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0181, 0.0178, 0.0228, 0.0221, 0.0204, 0.0164, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 04:34:30,118 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.255e+02 3.323e+02 3.837e+02 4.368e+02 2.807e+03, threshold=7.674e+02, percent-clipped=2.0 2023-05-17 04:34:33,651 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283193.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:34:34,521 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8949, 2.4875, 3.4949, 3.5820, 2.8008, 2.6505, 2.6642, 2.3973], device='cuda:1'), covar=tensor([0.1227, 0.2770, 0.0616, 0.0494, 0.0994, 0.2196, 0.2496, 0.3560], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0382, 0.0273, 0.0296, 0.0270, 0.0310, 0.0388, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:34:42,729 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283205.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:34:57,387 INFO [finetune.py:992] (1/2) Epoch 15, batch 11950, loss[loss=0.2142, simple_loss=0.2907, pruned_loss=0.06881, over 7020.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3054, pruned_loss=0.0719, over 1690145.76 frames. ], batch size: 98, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:35:08,513 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2136, 2.8179, 2.8777, 2.7759, 2.4996, 2.2793, 2.7570, 1.9522], device='cuda:1'), covar=tensor([0.0467, 0.0239, 0.0205, 0.0264, 0.0412, 0.0394, 0.0199, 0.0613], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0161, 0.0165, 0.0187, 0.0199, 0.0199, 0.0173, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:35:32,769 INFO [finetune.py:992] (1/2) Epoch 15, batch 12000, loss[loss=0.1867, simple_loss=0.2728, pruned_loss=0.05027, over 6952.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.3018, pruned_loss=0.06918, over 1690193.07 frames. ], batch size: 98, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:35:32,769 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 04:35:51,766 INFO [finetune.py:1026] (1/2) Epoch 15, validation: loss=0.2887, simple_loss=0.3627, pruned_loss=0.1073, over 1020973.00 frames. 2023-05-17 04:35:51,767 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 04:35:56,562 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283283.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:35:59,708 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.961e+02 3.359e+02 3.925e+02 1.066e+03, threshold=6.717e+02, percent-clipped=1.0 2023-05-17 04:36:00,623 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3581, 3.5325, 3.2024, 3.5841, 3.3503, 2.6549, 3.2507, 2.8673], device='cuda:1'), covar=tensor([0.0941, 0.1127, 0.1736, 0.0701, 0.1366, 0.1749, 0.1311, 0.3043], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0359, 0.0340, 0.0301, 0.0349, 0.0262, 0.0329, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:36:13,443 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8220, 3.0456, 2.4284, 2.2448, 2.8463, 2.4140, 2.9853, 2.6603], device='cuda:1'), covar=tensor([0.0648, 0.0615, 0.1024, 0.1530, 0.0323, 0.1174, 0.0612, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0245, 0.0172, 0.0196, 0.0137, 0.0178, 0.0188, 0.0167], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:36:18,019 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283314.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:36:26,863 INFO [finetune.py:992] (1/2) Epoch 15, batch 12050, loss[loss=0.196, simple_loss=0.2925, pruned_loss=0.04973, over 10458.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2976, pruned_loss=0.06639, over 1688086.54 frames. ], batch size: 68, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:36:30,984 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283333.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:36:34,472 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8047, 3.0485, 4.6095, 4.7578, 2.8807, 2.6980, 3.2214, 1.9367], device='cuda:1'), covar=tensor([0.1564, 0.2708, 0.0432, 0.0403, 0.1428, 0.2645, 0.2502, 0.4911], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0381, 0.0272, 0.0296, 0.0270, 0.0310, 0.0388, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:36:38,343 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283344.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:36:45,174 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283354.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 04:36:50,271 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283362.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:36:59,659 INFO [finetune.py:992] (1/2) Epoch 15, batch 12100, loss[loss=0.2227, simple_loss=0.2992, pruned_loss=0.07313, over 7406.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2961, pruned_loss=0.06535, over 1683698.92 frames. ], batch size: 98, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:37:02,294 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283381.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:37:06,554 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 3.015e+02 3.368e+02 3.745e+02 9.475e+02, threshold=6.736e+02, percent-clipped=5.0 2023-05-17 04:37:15,727 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283402.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:37:31,936 INFO [finetune.py:992] (1/2) Epoch 15, batch 12150, loss[loss=0.2482, simple_loss=0.3173, pruned_loss=0.08957, over 6847.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2969, pruned_loss=0.06551, over 1693236.24 frames. ], batch size: 99, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:37:44,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-05-17 04:37:53,382 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283461.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:37:59,570 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283471.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:38:03,183 INFO [finetune.py:992] (1/2) Epoch 15, batch 12200, loss[loss=0.2426, simple_loss=0.3064, pruned_loss=0.08936, over 7036.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2985, pruned_loss=0.06705, over 1664650.06 frames. ], batch size: 98, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:38:10,056 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.356e+02 3.265e+02 3.898e+02 4.626e+02 7.846e+02, threshold=7.795e+02, percent-clipped=3.0 2023-05-17 04:38:17,607 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283500.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:38:23,169 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283509.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:38:48,075 INFO [finetune.py:992] (1/2) Epoch 16, batch 0, loss[loss=0.1971, simple_loss=0.2771, pruned_loss=0.05851, over 8365.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2771, pruned_loss=0.05851, over 8365.00 frames. ], batch size: 98, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:38:48,075 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 04:39:03,412 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9932, 2.2825, 3.4561, 2.9790, 3.3062, 3.2007, 2.3950, 3.4166], device='cuda:1'), covar=tensor([0.0145, 0.0494, 0.0090, 0.0298, 0.0143, 0.0175, 0.0434, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0198, 0.0180, 0.0177, 0.0205, 0.0159, 0.0189, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:39:04,705 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2873, 2.4005, 3.8523, 3.2881, 3.5594, 3.4563, 2.6684, 3.6918], device='cuda:1'), covar=tensor([0.0098, 0.0420, 0.0046, 0.0214, 0.0138, 0.0127, 0.0382, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0198, 0.0180, 0.0177, 0.0205, 0.0159, 0.0189, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:39:06,183 INFO [finetune.py:1026] (1/2) Epoch 16, validation: loss=0.2859, simple_loss=0.3611, pruned_loss=0.1054, over 1020973.00 frames. 2023-05-17 04:39:06,183 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 04:39:29,605 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283544.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:39:41,603 INFO [finetune.py:992] (1/2) Epoch 16, batch 50, loss[loss=0.1567, simple_loss=0.2539, pruned_loss=0.02979, over 12300.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.264, pruned_loss=0.04128, over 539867.58 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:40:01,188 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 2.943e+02 3.465e+02 4.160e+02 7.410e+02, threshold=6.930e+02, percent-clipped=0.0 2023-05-17 04:40:14,084 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283605.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 04:40:18,286 INFO [finetune.py:992] (1/2) Epoch 16, batch 100, loss[loss=0.1757, simple_loss=0.2652, pruned_loss=0.04305, over 12098.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2602, pruned_loss=0.04018, over 943399.63 frames. ], batch size: 32, lr: 3.44e-03, grad_scale: 16.0 2023-05-17 04:40:38,317 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283639.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:40:45,576 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5123, 2.7388, 3.1643, 4.2213, 2.1256, 4.3020, 4.4443, 4.4941], device='cuda:1'), covar=tensor([0.0117, 0.1257, 0.0534, 0.0197, 0.1516, 0.0257, 0.0153, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0201, 0.0179, 0.0119, 0.0187, 0.0175, 0.0170, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:40:53,908 INFO [finetune.py:992] (1/2) Epoch 16, batch 150, loss[loss=0.1707, simple_loss=0.256, pruned_loss=0.04271, over 12040.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2585, pruned_loss=0.03902, over 1266991.92 frames. ], batch size: 31, lr: 3.44e-03, grad_scale: 16.0 2023-05-17 04:41:13,466 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2159, 4.6556, 4.0281, 4.9251, 4.5335, 2.9560, 4.2768, 2.9205], device='cuda:1'), covar=tensor([0.0861, 0.0842, 0.1676, 0.0591, 0.1048, 0.1734, 0.1088, 0.3942], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0370, 0.0350, 0.0309, 0.0359, 0.0268, 0.0338, 0.0357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:41:13,870 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.118e+02 2.555e+02 2.911e+02 3.555e+02 5.516e+02, threshold=5.823e+02, percent-clipped=0.0 2023-05-17 04:41:30,408 INFO [finetune.py:992] (1/2) Epoch 16, batch 200, loss[loss=0.1523, simple_loss=0.2386, pruned_loss=0.03298, over 12319.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2588, pruned_loss=0.03936, over 1509454.83 frames. ], batch size: 31, lr: 3.44e-03, grad_scale: 16.0 2023-05-17 04:42:06,339 INFO [finetune.py:992] (1/2) Epoch 16, batch 250, loss[loss=0.1623, simple_loss=0.2576, pruned_loss=0.03352, over 12139.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2587, pruned_loss=0.03918, over 1711836.60 frames. ], batch size: 36, lr: 3.44e-03, grad_scale: 16.0 2023-05-17 04:42:10,869 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0033, 4.6660, 4.7626, 4.8875, 4.7722, 4.8753, 4.8718, 2.6351], device='cuda:1'), covar=tensor([0.0112, 0.0080, 0.0102, 0.0064, 0.0056, 0.0114, 0.0075, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0078, 0.0082, 0.0073, 0.0060, 0.0092, 0.0081, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:42:13,736 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283771.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:42:16,853 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8917, 4.2111, 3.7585, 4.5648, 4.0966, 2.7450, 4.0340, 2.8772], device='cuda:1'), covar=tensor([0.1013, 0.1107, 0.1672, 0.0733, 0.1416, 0.1980, 0.1151, 0.3748], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0375, 0.0354, 0.0314, 0.0364, 0.0273, 0.0343, 0.0363], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:42:25,829 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.578e+02 3.087e+02 3.629e+02 5.392e+02, threshold=6.174e+02, percent-clipped=0.0 2023-05-17 04:42:26,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 2023-05-17 04:42:34,753 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283800.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:42:42,612 INFO [finetune.py:992] (1/2) Epoch 16, batch 300, loss[loss=0.1582, simple_loss=0.251, pruned_loss=0.03265, over 12300.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2588, pruned_loss=0.03924, over 1863134.56 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 16.0 2023-05-17 04:42:48,459 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283819.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:43:05,572 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283842.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:43:09,856 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283848.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:43:19,235 INFO [finetune.py:992] (1/2) Epoch 16, batch 350, loss[loss=0.1717, simple_loss=0.2625, pruned_loss=0.04038, over 10549.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2575, pruned_loss=0.03877, over 1972300.16 frames. ], batch size: 68, lr: 3.44e-03, grad_scale: 16.0 2023-05-17 04:43:39,051 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.698e+02 3.166e+02 3.611e+02 6.102e+02, threshold=6.331e+02, percent-clipped=0.0 2023-05-17 04:43:47,862 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283900.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 04:43:50,011 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283903.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:43:52,103 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6431, 5.5847, 5.4515, 4.9393, 4.9472, 5.5644, 5.2035, 4.9564], device='cuda:1'), covar=tensor([0.0889, 0.1120, 0.0791, 0.1765, 0.0893, 0.0762, 0.1656, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0562, 0.0510, 0.0627, 0.0414, 0.0711, 0.0765, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-17 04:43:55,564 INFO [finetune.py:992] (1/2) Epoch 16, batch 400, loss[loss=0.1398, simple_loss=0.2206, pruned_loss=0.02954, over 12271.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2563, pruned_loss=0.03851, over 2065650.50 frames. ], batch size: 28, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:44:08,446 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6858, 2.7649, 3.4388, 4.4540, 2.3003, 4.4432, 4.5786, 4.7785], device='cuda:1'), covar=tensor([0.0101, 0.1174, 0.0408, 0.0133, 0.1332, 0.0178, 0.0132, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0202, 0.0181, 0.0120, 0.0189, 0.0176, 0.0172, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:44:11,797 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5119, 2.6459, 3.7318, 4.5307, 3.9359, 4.4928, 3.7940, 3.2003], device='cuda:1'), covar=tensor([0.0041, 0.0399, 0.0125, 0.0042, 0.0124, 0.0073, 0.0137, 0.0364], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0121, 0.0102, 0.0076, 0.0101, 0.0113, 0.0098, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:44:15,455 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283939.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:44:31,097 INFO [finetune.py:992] (1/2) Epoch 16, batch 450, loss[loss=0.1567, simple_loss=0.2417, pruned_loss=0.0359, over 12356.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2562, pruned_loss=0.0385, over 2131808.27 frames. ], batch size: 30, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:44:34,841 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283966.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:44:50,463 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283987.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:44:51,824 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 2.511e+02 3.056e+02 3.598e+02 7.738e+02, threshold=6.112e+02, percent-clipped=2.0 2023-05-17 04:45:10,894 INFO [finetune.py:992] (1/2) Epoch 16, batch 500, loss[loss=0.1813, simple_loss=0.2717, pruned_loss=0.04545, over 12118.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2559, pruned_loss=0.03831, over 2191518.51 frames. ], batch size: 39, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:45:11,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 04:45:22,503 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7664, 2.9878, 3.3870, 4.5792, 2.8225, 4.5629, 4.6667, 4.7894], device='cuda:1'), covar=tensor([0.0094, 0.1062, 0.0479, 0.0137, 0.1134, 0.0211, 0.0132, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0201, 0.0180, 0.0120, 0.0188, 0.0175, 0.0171, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:45:22,520 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284027.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:45:43,691 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2722, 6.0751, 5.7243, 5.6479, 6.1496, 5.4813, 5.5561, 5.6337], device='cuda:1'), covar=tensor([0.1656, 0.0997, 0.1030, 0.2322, 0.0992, 0.2486, 0.2012, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0498, 0.0403, 0.0451, 0.0463, 0.0434, 0.0396, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:45:46,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-17 04:45:47,229 INFO [finetune.py:992] (1/2) Epoch 16, batch 550, loss[loss=0.1423, simple_loss=0.2242, pruned_loss=0.03017, over 12349.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2549, pruned_loss=0.03799, over 2235321.27 frames. ], batch size: 30, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:46:03,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-17 04:46:07,264 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.018e+02 2.524e+02 2.956e+02 3.674e+02 6.026e+02, threshold=5.913e+02, percent-clipped=0.0 2023-05-17 04:46:23,015 INFO [finetune.py:992] (1/2) Epoch 16, batch 600, loss[loss=0.1747, simple_loss=0.2711, pruned_loss=0.03914, over 12355.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2546, pruned_loss=0.03795, over 2268213.79 frames. ], batch size: 36, lr: 3.44e-03, grad_scale: 8.0 2023-05-17 04:46:26,042 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2009, 3.8765, 4.0243, 4.3572, 2.9190, 3.8687, 2.4246, 3.8104], device='cuda:1'), covar=tensor([0.2021, 0.0933, 0.1051, 0.0764, 0.1423, 0.0723, 0.2403, 0.1491], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0273, 0.0305, 0.0363, 0.0248, 0.0249, 0.0268, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:46:52,751 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0 2023-05-17 04:46:59,390 INFO [finetune.py:992] (1/2) Epoch 16, batch 650, loss[loss=0.1313, simple_loss=0.2224, pruned_loss=0.02013, over 12135.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2552, pruned_loss=0.03828, over 2285341.30 frames. ], batch size: 30, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:47:07,467 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-05-17 04:47:19,450 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.652e+02 3.123e+02 3.791e+02 1.719e+03, threshold=6.246e+02, percent-clipped=4.0 2023-05-17 04:47:25,960 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284198.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:47:27,790 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284200.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:47:35,675 INFO [finetune.py:992] (1/2) Epoch 16, batch 700, loss[loss=0.16, simple_loss=0.2539, pruned_loss=0.03307, over 12282.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.255, pruned_loss=0.038, over 2309148.81 frames. ], batch size: 37, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:47:42,081 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8191, 3.5501, 3.5449, 3.6886, 3.7218, 3.7687, 3.7064, 2.5333], device='cuda:1'), covar=tensor([0.0106, 0.0110, 0.0173, 0.0095, 0.0076, 0.0144, 0.0104, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0084, 0.0074, 0.0061, 0.0094, 0.0082, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:47:42,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-17 04:47:47,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-05-17 04:48:02,000 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284248.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:48:11,144 INFO [finetune.py:992] (1/2) Epoch 16, batch 750, loss[loss=0.1918, simple_loss=0.2862, pruned_loss=0.0487, over 12095.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2556, pruned_loss=0.03811, over 2325000.45 frames. ], batch size: 33, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:48:31,963 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.420e+02 3.067e+02 4.052e+02 9.430e+02, threshold=6.135e+02, percent-clipped=7.0 2023-05-17 04:48:47,803 INFO [finetune.py:992] (1/2) Epoch 16, batch 800, loss[loss=0.1348, simple_loss=0.214, pruned_loss=0.02778, over 11843.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.256, pruned_loss=0.03841, over 2337640.56 frames. ], batch size: 26, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:48:55,913 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284322.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:48:59,624 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3490, 4.9565, 5.3514, 4.7184, 5.0021, 4.7618, 5.3984, 5.0360], device='cuda:1'), covar=tensor([0.0289, 0.0369, 0.0264, 0.0270, 0.0443, 0.0352, 0.0216, 0.0315], device='cuda:1'), in_proj_covar=tensor([0.0270, 0.0275, 0.0296, 0.0268, 0.0272, 0.0272, 0.0247, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:49:20,741 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0829, 5.9601, 5.5088, 5.3989, 6.0000, 5.3370, 5.5271, 5.4692], device='cuda:1'), covar=tensor([0.1566, 0.0943, 0.1074, 0.2043, 0.0865, 0.2354, 0.1776, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0499, 0.0406, 0.0451, 0.0464, 0.0435, 0.0397, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:49:24,321 INFO [finetune.py:992] (1/2) Epoch 16, batch 850, loss[loss=0.1741, simple_loss=0.2705, pruned_loss=0.03885, over 10541.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2557, pruned_loss=0.03819, over 2339746.15 frames. ], batch size: 68, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:49:31,662 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6327, 5.3962, 5.5608, 5.5884, 5.2329, 5.2931, 5.0921, 5.4571], device='cuda:1'), covar=tensor([0.0636, 0.0600, 0.0783, 0.0583, 0.1730, 0.1259, 0.0519, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0697, 0.0612, 0.0629, 0.0839, 0.0741, 0.0568, 0.0480], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:49:43,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-05-17 04:49:44,520 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 2.671e+02 2.987e+02 3.565e+02 5.859e+02, threshold=5.974e+02, percent-clipped=0.0 2023-05-17 04:49:46,124 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1763, 4.7842, 4.8496, 4.9558, 4.8588, 4.9551, 4.9666, 2.3643], device='cuda:1'), covar=tensor([0.0100, 0.0063, 0.0094, 0.0061, 0.0055, 0.0094, 0.0073, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0074, 0.0061, 0.0094, 0.0082, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:50:00,397 INFO [finetune.py:992] (1/2) Epoch 16, batch 900, loss[loss=0.1286, simple_loss=0.2119, pruned_loss=0.02265, over 12278.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2544, pruned_loss=0.03735, over 2356552.81 frames. ], batch size: 28, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:50:24,160 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284443.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:50:29,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.23 vs. limit=5.0 2023-05-17 04:50:36,956 INFO [finetune.py:992] (1/2) Epoch 16, batch 950, loss[loss=0.2112, simple_loss=0.2904, pruned_loss=0.06599, over 7951.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2546, pruned_loss=0.03726, over 2359671.84 frames. ], batch size: 98, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:50:40,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-17 04:50:57,334 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.691e+02 3.092e+02 3.404e+02 8.207e+02, threshold=6.184e+02, percent-clipped=1.0 2023-05-17 04:51:03,883 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284498.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:51:08,142 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284504.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:51:12,825 INFO [finetune.py:992] (1/2) Epoch 16, batch 1000, loss[loss=0.1721, simple_loss=0.2718, pruned_loss=0.03624, over 11830.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2549, pruned_loss=0.03744, over 2358107.55 frames. ], batch size: 44, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:51:13,733 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9949, 5.9651, 5.6988, 5.2699, 5.1931, 5.8539, 5.5264, 5.2331], device='cuda:1'), covar=tensor([0.0648, 0.0777, 0.0677, 0.1638, 0.0753, 0.0744, 0.1440, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0634, 0.0571, 0.0519, 0.0641, 0.0421, 0.0729, 0.0782, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-17 04:51:18,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 04:51:37,809 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284546.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:51:48,471 INFO [finetune.py:992] (1/2) Epoch 16, batch 1050, loss[loss=0.1669, simple_loss=0.2587, pruned_loss=0.0375, over 12035.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2553, pruned_loss=0.03752, over 2354232.16 frames. ], batch size: 37, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:52:09,087 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.650e+02 3.060e+02 3.658e+02 8.576e+02, threshold=6.119e+02, percent-clipped=2.0 2023-05-17 04:52:11,299 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0310, 5.8569, 5.5022, 5.3902, 5.9468, 5.3148, 5.4472, 5.3700], device='cuda:1'), covar=tensor([0.1735, 0.0981, 0.1183, 0.2059, 0.0985, 0.2493, 0.2041, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0506, 0.0410, 0.0457, 0.0472, 0.0439, 0.0404, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:52:25,102 INFO [finetune.py:992] (1/2) Epoch 16, batch 1100, loss[loss=0.1652, simple_loss=0.2691, pruned_loss=0.03069, over 12161.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.255, pruned_loss=0.03738, over 2365070.91 frames. ], batch size: 36, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:52:33,709 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284622.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:52:43,908 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4267, 2.8255, 3.6141, 4.4557, 3.9396, 4.3836, 3.8294, 3.1059], device='cuda:1'), covar=tensor([0.0035, 0.0353, 0.0147, 0.0039, 0.0122, 0.0078, 0.0122, 0.0367], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0121, 0.0103, 0.0077, 0.0103, 0.0115, 0.0099, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:53:01,495 INFO [finetune.py:992] (1/2) Epoch 16, batch 1150, loss[loss=0.1652, simple_loss=0.2585, pruned_loss=0.03593, over 10533.00 frames. ], tot_loss[loss=0.164, simple_loss=0.254, pruned_loss=0.03699, over 2368093.10 frames. ], batch size: 69, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:53:08,044 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284670.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:53:08,079 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0471, 5.9336, 5.6494, 5.4658, 6.0058, 5.4518, 5.5578, 5.4318], device='cuda:1'), covar=tensor([0.1575, 0.0994, 0.1198, 0.1971, 0.0897, 0.2037, 0.1790, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0510, 0.0413, 0.0461, 0.0474, 0.0442, 0.0407, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 04:53:08,920 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5598, 3.6489, 3.2202, 3.1856, 2.9393, 2.7551, 3.6527, 2.2771], device='cuda:1'), covar=tensor([0.0429, 0.0156, 0.0243, 0.0229, 0.0424, 0.0434, 0.0164, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0165, 0.0170, 0.0192, 0.0204, 0.0203, 0.0178, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:53:21,775 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.636e+02 3.102e+02 3.453e+02 5.251e+02, threshold=6.203e+02, percent-clipped=0.0 2023-05-17 04:53:37,880 INFO [finetune.py:992] (1/2) Epoch 16, batch 1200, loss[loss=0.1389, simple_loss=0.2311, pruned_loss=0.02333, over 12181.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2536, pruned_loss=0.0368, over 2364798.40 frames. ], batch size: 31, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:54:14,009 INFO [finetune.py:992] (1/2) Epoch 16, batch 1250, loss[loss=0.1459, simple_loss=0.2336, pruned_loss=0.02908, over 12109.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2548, pruned_loss=0.03714, over 2361061.31 frames. ], batch size: 33, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:54:20,801 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1076, 4.4498, 3.9386, 4.6780, 4.3095, 2.8252, 4.0143, 2.8500], device='cuda:1'), covar=tensor([0.0805, 0.0860, 0.1512, 0.0602, 0.1156, 0.1822, 0.1136, 0.3575], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0383, 0.0362, 0.0324, 0.0372, 0.0277, 0.0349, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:54:34,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.693e+02 3.233e+02 3.746e+02 7.896e+02, threshold=6.466e+02, percent-clipped=1.0 2023-05-17 04:54:36,468 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0448, 4.7264, 4.8984, 4.9506, 4.7270, 4.9544, 4.8705, 2.5783], device='cuda:1'), covar=tensor([0.0106, 0.0068, 0.0084, 0.0059, 0.0055, 0.0092, 0.0076, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0074, 0.0061, 0.0093, 0.0082, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:54:42,101 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284799.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:54:50,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-17 04:54:50,940 INFO [finetune.py:992] (1/2) Epoch 16, batch 1300, loss[loss=0.1453, simple_loss=0.2282, pruned_loss=0.03121, over 12122.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2544, pruned_loss=0.03691, over 2372559.89 frames. ], batch size: 30, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:54:54,759 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284816.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:55:01,234 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5486, 4.9500, 4.2064, 5.0639, 4.7248, 2.8383, 4.1325, 3.2273], device='cuda:1'), covar=tensor([0.0681, 0.0659, 0.1453, 0.0488, 0.1076, 0.1866, 0.1231, 0.2944], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0384, 0.0363, 0.0325, 0.0373, 0.0277, 0.0351, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:55:09,023 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7373, 2.3667, 4.0203, 4.2037, 3.1021, 2.5230, 2.6511, 2.1210], device='cuda:1'), covar=tensor([0.1695, 0.3654, 0.0585, 0.0441, 0.1107, 0.2788, 0.3058, 0.4967], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0398, 0.0282, 0.0306, 0.0281, 0.0323, 0.0402, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 04:55:26,493 INFO [finetune.py:992] (1/2) Epoch 16, batch 1350, loss[loss=0.1512, simple_loss=0.2352, pruned_loss=0.03356, over 12180.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2551, pruned_loss=0.03725, over 2371939.51 frames. ], batch size: 29, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:55:38,563 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284877.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:55:46,813 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.652e+02 3.047e+02 3.454e+02 8.344e+02, threshold=6.094e+02, percent-clipped=2.0 2023-05-17 04:56:02,640 INFO [finetune.py:992] (1/2) Epoch 16, batch 1400, loss[loss=0.1625, simple_loss=0.2638, pruned_loss=0.03065, over 12276.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2545, pruned_loss=0.03692, over 2368333.52 frames. ], batch size: 37, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:56:04,860 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3720, 4.2208, 4.2022, 4.2213, 3.9366, 4.4032, 4.3002, 4.5300], device='cuda:1'), covar=tensor([0.0293, 0.0216, 0.0245, 0.0465, 0.0845, 0.0444, 0.0243, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0197, 0.0194, 0.0249, 0.0241, 0.0224, 0.0179, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 04:56:19,065 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.8959, 5.8919, 5.6357, 5.2428, 5.1606, 5.8061, 5.3566, 5.1954], device='cuda:1'), covar=tensor([0.0843, 0.0953, 0.0794, 0.1631, 0.0776, 0.0767, 0.1725, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0572, 0.0520, 0.0646, 0.0422, 0.0731, 0.0789, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:56:23,537 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284939.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:56:39,287 INFO [finetune.py:992] (1/2) Epoch 16, batch 1450, loss[loss=0.1512, simple_loss=0.2382, pruned_loss=0.0321, over 12329.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2533, pruned_loss=0.03671, over 2375625.29 frames. ], batch size: 30, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:56:52,942 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2776, 6.2235, 5.9897, 5.5065, 5.3790, 6.1598, 5.7852, 5.4940], device='cuda:1'), covar=tensor([0.0709, 0.0964, 0.0632, 0.1619, 0.0676, 0.0716, 0.1460, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0573, 0.0521, 0.0648, 0.0422, 0.0732, 0.0789, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:56:59,198 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 2.503e+02 2.956e+02 3.648e+02 7.736e+02, threshold=5.913e+02, percent-clipped=3.0 2023-05-17 04:57:04,471 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284996.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:57:07,817 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285000.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:57:15,272 INFO [finetune.py:992] (1/2) Epoch 16, batch 1500, loss[loss=0.1558, simple_loss=0.2514, pruned_loss=0.03012, over 12263.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2536, pruned_loss=0.03687, over 2371755.34 frames. ], batch size: 37, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:57:20,470 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4768, 5.2446, 5.4452, 5.4329, 5.0431, 5.1525, 4.8584, 5.3469], device='cuda:1'), covar=tensor([0.0653, 0.0597, 0.0776, 0.0548, 0.1946, 0.1229, 0.0548, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0711, 0.0624, 0.0642, 0.0859, 0.0755, 0.0578, 0.0489], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 04:57:26,123 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2862, 2.5993, 3.5913, 4.3611, 3.8294, 4.2585, 3.7117, 2.9460], device='cuda:1'), covar=tensor([0.0052, 0.0414, 0.0164, 0.0038, 0.0144, 0.0092, 0.0147, 0.0408], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0123, 0.0105, 0.0078, 0.0104, 0.0116, 0.0100, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 04:57:38,224 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285042.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 04:57:48,922 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285057.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:57:51,532 INFO [finetune.py:992] (1/2) Epoch 16, batch 1550, loss[loss=0.1434, simple_loss=0.2231, pruned_loss=0.0318, over 12011.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2541, pruned_loss=0.03711, over 2380962.96 frames. ], batch size: 28, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:58:11,929 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 2.594e+02 3.142e+02 3.617e+02 6.812e+02, threshold=6.284e+02, percent-clipped=2.0 2023-05-17 04:58:19,259 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285099.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:58:22,222 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285103.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 04:58:27,693 INFO [finetune.py:992] (1/2) Epoch 16, batch 1600, loss[loss=0.1855, simple_loss=0.2683, pruned_loss=0.0513, over 12030.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2542, pruned_loss=0.03706, over 2378567.56 frames. ], batch size: 42, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:58:53,376 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285147.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:59:03,413 INFO [finetune.py:992] (1/2) Epoch 16, batch 1650, loss[loss=0.139, simple_loss=0.241, pruned_loss=0.0185, over 12073.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2542, pruned_loss=0.03706, over 2368943.49 frames. ], batch size: 32, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 04:59:11,344 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285172.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 04:59:23,744 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.631e+02 3.056e+02 3.621e+02 1.098e+03, threshold=6.112e+02, percent-clipped=1.0 2023-05-17 04:59:40,070 INFO [finetune.py:992] (1/2) Epoch 16, batch 1700, loss[loss=0.1584, simple_loss=0.2557, pruned_loss=0.03059, over 12149.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2536, pruned_loss=0.03705, over 2372489.43 frames. ], batch size: 34, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:00:15,633 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5238, 5.4434, 5.2298, 4.7599, 4.8504, 5.4131, 5.0374, 4.8493], device='cuda:1'), covar=tensor([0.0798, 0.1050, 0.0745, 0.1627, 0.1144, 0.0814, 0.1677, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0574, 0.0523, 0.0648, 0.0422, 0.0733, 0.0791, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 05:00:16,848 INFO [finetune.py:992] (1/2) Epoch 16, batch 1750, loss[loss=0.1447, simple_loss=0.2331, pruned_loss=0.02816, over 12115.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2539, pruned_loss=0.03694, over 2377788.61 frames. ], batch size: 30, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:00:36,546 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.639e+02 3.105e+02 3.707e+02 1.479e+03, threshold=6.210e+02, percent-clipped=4.0 2023-05-17 05:00:40,982 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285295.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:00:52,204 INFO [finetune.py:992] (1/2) Epoch 16, batch 1800, loss[loss=0.1844, simple_loss=0.2763, pruned_loss=0.04618, over 12133.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2545, pruned_loss=0.03728, over 2372853.64 frames. ], batch size: 39, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:00:58,959 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7719, 2.6792, 4.2573, 4.4316, 2.6905, 2.5098, 2.8151, 2.1410], device='cuda:1'), covar=tensor([0.1655, 0.3460, 0.0563, 0.0489, 0.1520, 0.2693, 0.3097, 0.4412], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0395, 0.0283, 0.0304, 0.0279, 0.0321, 0.0399, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:01:21,810 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5968, 2.8314, 3.8191, 4.6622, 4.2137, 4.6696, 4.0330, 3.5231], device='cuda:1'), covar=tensor([0.0040, 0.0370, 0.0136, 0.0042, 0.0090, 0.0068, 0.0105, 0.0303], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0122, 0.0104, 0.0078, 0.0103, 0.0115, 0.0098, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:01:22,437 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285352.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:01:29,101 INFO [finetune.py:992] (1/2) Epoch 16, batch 1850, loss[loss=0.1633, simple_loss=0.2537, pruned_loss=0.03648, over 12012.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2547, pruned_loss=0.03731, over 2362115.80 frames. ], batch size: 28, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:01:49,237 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 2.598e+02 2.984e+02 3.373e+02 7.020e+02, threshold=5.967e+02, percent-clipped=3.0 2023-05-17 05:01:55,498 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285398.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:02:04,953 INFO [finetune.py:992] (1/2) Epoch 16, batch 1900, loss[loss=0.1597, simple_loss=0.2522, pruned_loss=0.03359, over 12345.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2541, pruned_loss=0.03721, over 2372732.50 frames. ], batch size: 31, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:02:05,981 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6929, 3.1847, 5.1008, 2.6926, 2.9665, 3.8857, 3.2715, 3.9087], device='cuda:1'), covar=tensor([0.0478, 0.1344, 0.0415, 0.1292, 0.2044, 0.1499, 0.1425, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0243, 0.0259, 0.0187, 0.0242, 0.0297, 0.0229, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:02:35,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-17 05:02:40,369 INFO [finetune.py:992] (1/2) Epoch 16, batch 1950, loss[loss=0.168, simple_loss=0.2642, pruned_loss=0.03585, over 12194.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2542, pruned_loss=0.03732, over 2365395.60 frames. ], batch size: 35, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:02:49,039 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285472.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:02:50,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 05:02:57,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-05-17 05:03:00,827 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.602e+02 3.218e+02 3.856e+02 6.883e+02, threshold=6.436e+02, percent-clipped=4.0 2023-05-17 05:03:17,176 INFO [finetune.py:992] (1/2) Epoch 16, batch 2000, loss[loss=0.1631, simple_loss=0.2627, pruned_loss=0.03174, over 11366.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2539, pruned_loss=0.03711, over 2367153.87 frames. ], batch size: 55, lr: 3.43e-03, grad_scale: 8.0 2023-05-17 05:03:23,757 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285520.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:03:53,040 INFO [finetune.py:992] (1/2) Epoch 16, batch 2050, loss[loss=0.1512, simple_loss=0.236, pruned_loss=0.03323, over 12289.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2547, pruned_loss=0.03714, over 2378044.20 frames. ], batch size: 28, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:04:13,133 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.449e+02 2.928e+02 3.605e+02 5.021e+02, threshold=5.856e+02, percent-clipped=0.0 2023-05-17 05:04:17,504 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285595.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:04:29,155 INFO [finetune.py:992] (1/2) Epoch 16, batch 2100, loss[loss=0.1618, simple_loss=0.2599, pruned_loss=0.03187, over 12179.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2539, pruned_loss=0.03689, over 2382243.04 frames. ], batch size: 35, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:04:37,708 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285622.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:04:49,744 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5791, 2.9620, 3.7947, 4.5206, 4.0837, 4.6027, 3.9563, 3.4521], device='cuda:1'), covar=tensor([0.0037, 0.0325, 0.0154, 0.0045, 0.0115, 0.0086, 0.0114, 0.0321], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0120, 0.0103, 0.0078, 0.0102, 0.0114, 0.0097, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:04:52,436 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285643.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:04:53,286 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0311, 4.5398, 4.6867, 4.8014, 4.5993, 4.8266, 4.7433, 2.5626], device='cuda:1'), covar=tensor([0.0110, 0.0077, 0.0102, 0.0069, 0.0071, 0.0104, 0.0081, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0083, 0.0074, 0.0061, 0.0093, 0.0083, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:04:58,825 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285652.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:05:05,868 INFO [finetune.py:992] (1/2) Epoch 16, batch 2150, loss[loss=0.1642, simple_loss=0.2517, pruned_loss=0.03836, over 12279.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2549, pruned_loss=0.03741, over 2384841.85 frames. ], batch size: 33, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:05:21,837 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285683.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:05:25,717 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.738e+02 3.209e+02 3.836e+02 6.080e+02, threshold=6.417e+02, percent-clipped=1.0 2023-05-17 05:05:32,436 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285698.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:05:33,808 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285700.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:05:41,649 INFO [finetune.py:992] (1/2) Epoch 16, batch 2200, loss[loss=0.1375, simple_loss=0.2165, pruned_loss=0.0292, over 12003.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2545, pruned_loss=0.03751, over 2378281.76 frames. ], batch size: 28, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:06:06,587 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285746.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:06:17,891 INFO [finetune.py:992] (1/2) Epoch 16, batch 2250, loss[loss=0.161, simple_loss=0.2537, pruned_loss=0.03415, over 12009.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2555, pruned_loss=0.0381, over 2372773.75 frames. ], batch size: 40, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:06:37,890 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.712e+02 3.166e+02 3.755e+02 7.301e+02, threshold=6.332e+02, percent-clipped=2.0 2023-05-17 05:06:54,483 INFO [finetune.py:992] (1/2) Epoch 16, batch 2300, loss[loss=0.1749, simple_loss=0.2695, pruned_loss=0.04015, over 12032.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.255, pruned_loss=0.03761, over 2378456.33 frames. ], batch size: 40, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:07:30,539 INFO [finetune.py:992] (1/2) Epoch 16, batch 2350, loss[loss=0.1655, simple_loss=0.2659, pruned_loss=0.03249, over 11689.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.254, pruned_loss=0.03721, over 2377102.38 frames. ], batch size: 48, lr: 3.42e-03, grad_scale: 8.0 2023-05-17 05:07:45,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2023-05-17 05:07:50,642 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.624e+02 3.039e+02 3.508e+02 6.519e+02, threshold=6.078e+02, percent-clipped=2.0 2023-05-17 05:07:56,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 05:08:07,067 INFO [finetune.py:992] (1/2) Epoch 16, batch 2400, loss[loss=0.1535, simple_loss=0.2391, pruned_loss=0.03398, over 12181.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2532, pruned_loss=0.03674, over 2380508.50 frames. ], batch size: 31, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:08:26,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-05-17 05:08:43,456 INFO [finetune.py:992] (1/2) Epoch 16, batch 2450, loss[loss=0.1656, simple_loss=0.2431, pruned_loss=0.04401, over 12180.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2538, pruned_loss=0.03744, over 2369349.96 frames. ], batch size: 31, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:08:55,802 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285978.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:09:03,569 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.739e+02 3.151e+02 3.666e+02 5.852e+02, threshold=6.301e+02, percent-clipped=0.0 2023-05-17 05:09:22,628 INFO [finetune.py:992] (1/2) Epoch 16, batch 2500, loss[loss=0.1539, simple_loss=0.246, pruned_loss=0.03092, over 12279.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2531, pruned_loss=0.03717, over 2362868.11 frames. ], batch size: 33, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:09:23,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3155, 5.1002, 5.2723, 5.2767, 4.8370, 4.9899, 4.7573, 5.2170], device='cuda:1'), covar=tensor([0.0731, 0.0683, 0.0936, 0.0603, 0.2155, 0.1447, 0.0623, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0718, 0.0626, 0.0641, 0.0865, 0.0759, 0.0578, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:09:30,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 05:09:34,197 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0823, 5.7730, 5.3621, 5.2756, 5.9016, 5.1684, 5.2455, 5.2480], device='cuda:1'), covar=tensor([0.1465, 0.0984, 0.1208, 0.1829, 0.0960, 0.2332, 0.2418, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0509, 0.0411, 0.0456, 0.0476, 0.0441, 0.0403, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:09:54,973 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7554, 4.0565, 3.5740, 4.1778, 3.8521, 2.7109, 3.6538, 2.8849], device='cuda:1'), covar=tensor([0.0890, 0.0927, 0.1614, 0.0721, 0.1351, 0.1809, 0.1308, 0.3407], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0382, 0.0362, 0.0329, 0.0374, 0.0276, 0.0349, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:09:58,902 INFO [finetune.py:992] (1/2) Epoch 16, batch 2550, loss[loss=0.1592, simple_loss=0.2589, pruned_loss=0.02972, over 12342.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2544, pruned_loss=0.03758, over 2361741.41 frames. ], batch size: 36, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:10:12,061 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2267, 3.8533, 4.0057, 4.2540, 2.7177, 3.7332, 2.5333, 3.7683], device='cuda:1'), covar=tensor([0.1677, 0.0882, 0.0875, 0.0689, 0.1463, 0.0774, 0.2126, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0266, 0.0296, 0.0356, 0.0241, 0.0241, 0.0261, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:10:19,543 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.621e+02 2.966e+02 3.355e+02 6.018e+02, threshold=5.932e+02, percent-clipped=0.0 2023-05-17 05:10:35,294 INFO [finetune.py:992] (1/2) Epoch 16, batch 2600, loss[loss=0.1719, simple_loss=0.2662, pruned_loss=0.0388, over 12287.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2533, pruned_loss=0.03718, over 2365345.25 frames. ], batch size: 37, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:10:43,997 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286123.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:10:48,085 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286129.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:11:10,661 INFO [finetune.py:992] (1/2) Epoch 16, batch 2650, loss[loss=0.1686, simple_loss=0.2604, pruned_loss=0.03837, over 12012.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2542, pruned_loss=0.03731, over 2368192.05 frames. ], batch size: 40, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:11:26,972 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8320, 5.5928, 5.2462, 5.2230, 5.7334, 4.9613, 5.0974, 5.1376], device='cuda:1'), covar=tensor([0.1542, 0.1042, 0.1062, 0.1789, 0.1054, 0.2366, 0.2043, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0510, 0.0410, 0.0456, 0.0477, 0.0441, 0.0403, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:11:27,792 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286184.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:11:29,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.59 vs. limit=5.0 2023-05-17 05:11:31,123 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.530e+02 2.932e+02 3.499e+02 1.091e+03, threshold=5.863e+02, percent-clipped=1.0 2023-05-17 05:11:32,178 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286190.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:11:47,726 INFO [finetune.py:992] (1/2) Epoch 16, batch 2700, loss[loss=0.2035, simple_loss=0.2805, pruned_loss=0.06324, over 8136.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2546, pruned_loss=0.03769, over 2362511.85 frames. ], batch size: 99, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:12:24,235 INFO [finetune.py:992] (1/2) Epoch 16, batch 2750, loss[loss=0.1657, simple_loss=0.2585, pruned_loss=0.03641, over 12160.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2548, pruned_loss=0.0374, over 2367964.66 frames. ], batch size: 36, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:12:36,638 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286278.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:12:44,318 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.654e+02 3.141e+02 3.640e+02 5.193e+02, threshold=6.282e+02, percent-clipped=0.0 2023-05-17 05:12:48,953 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286295.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:13:00,143 INFO [finetune.py:992] (1/2) Epoch 16, batch 2800, loss[loss=0.1583, simple_loss=0.2563, pruned_loss=0.03016, over 12192.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2539, pruned_loss=0.03695, over 2361820.43 frames. ], batch size: 35, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:13:10,136 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4131, 4.9966, 5.4475, 4.7237, 5.0363, 4.8318, 5.4555, 5.0797], device='cuda:1'), covar=tensor([0.0277, 0.0402, 0.0261, 0.0271, 0.0440, 0.0337, 0.0215, 0.0257], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0280, 0.0298, 0.0273, 0.0275, 0.0274, 0.0249, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:13:10,765 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286326.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:13:24,500 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286344.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:13:30,301 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9590, 5.8843, 5.6143, 5.2356, 5.0658, 5.8289, 5.4501, 5.1794], device='cuda:1'), covar=tensor([0.0712, 0.0959, 0.0766, 0.1720, 0.0903, 0.0777, 0.1565, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0578, 0.0530, 0.0655, 0.0427, 0.0742, 0.0795, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 05:13:33,177 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286356.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:13:36,297 INFO [finetune.py:992] (1/2) Epoch 16, batch 2850, loss[loss=0.1777, simple_loss=0.2639, pruned_loss=0.04578, over 12059.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2543, pruned_loss=0.03718, over 2363573.52 frames. ], batch size: 42, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:13:52,120 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1554, 6.1028, 5.7985, 5.4093, 5.1766, 5.9916, 5.6520, 5.3400], device='cuda:1'), covar=tensor([0.0655, 0.0857, 0.0717, 0.1826, 0.0665, 0.0830, 0.1598, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0635, 0.0576, 0.0529, 0.0653, 0.0426, 0.0741, 0.0792, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 05:13:52,951 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3273, 4.7519, 2.8717, 2.6659, 4.0256, 2.6700, 3.9803, 3.1962], device='cuda:1'), covar=tensor([0.0729, 0.0636, 0.1166, 0.1582, 0.0333, 0.1339, 0.0475, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0258, 0.0176, 0.0203, 0.0143, 0.0183, 0.0198, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:13:56,931 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.626e+02 3.103e+02 3.601e+02 5.239e+02, threshold=6.207e+02, percent-clipped=0.0 2023-05-17 05:14:08,999 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-17 05:14:09,234 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286405.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:14:13,108 INFO [finetune.py:992] (1/2) Epoch 16, batch 2900, loss[loss=0.1741, simple_loss=0.2625, pruned_loss=0.04289, over 11050.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2528, pruned_loss=0.03663, over 2371301.43 frames. ], batch size: 55, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:14:31,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-17 05:14:48,805 INFO [finetune.py:992] (1/2) Epoch 16, batch 2950, loss[loss=0.1901, simple_loss=0.2817, pruned_loss=0.04924, over 12364.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2531, pruned_loss=0.03681, over 2373957.59 frames. ], batch size: 38, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:15:01,568 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286479.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:15:02,399 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5948, 2.6436, 3.7228, 4.5766, 4.1582, 4.5489, 3.9650, 3.4859], device='cuda:1'), covar=tensor([0.0039, 0.0430, 0.0159, 0.0044, 0.0100, 0.0088, 0.0120, 0.0329], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0123, 0.0105, 0.0079, 0.0104, 0.0117, 0.0100, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:15:06,589 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286485.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:15:09,211 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.711e+02 3.100e+02 3.818e+02 6.057e+02, threshold=6.200e+02, percent-clipped=0.0 2023-05-17 05:15:11,996 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-17 05:15:24,608 INFO [finetune.py:992] (1/2) Epoch 16, batch 3000, loss[loss=0.1649, simple_loss=0.2573, pruned_loss=0.03628, over 12295.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2521, pruned_loss=0.03659, over 2374294.86 frames. ], batch size: 34, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:15:24,609 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 05:15:35,138 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0167, 3.1552, 2.9732, 3.3369, 3.1778, 2.5265, 2.9563, 2.6537], device='cuda:1'), covar=tensor([0.0965, 0.1233, 0.1413, 0.0822, 0.1298, 0.1561, 0.1264, 0.2763], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0385, 0.0362, 0.0330, 0.0375, 0.0277, 0.0351, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:15:43,070 INFO [finetune.py:1026] (1/2) Epoch 16, validation: loss=0.3133, simple_loss=0.39, pruned_loss=0.1183, over 1020973.00 frames. 2023-05-17 05:15:43,071 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 05:16:18,967 INFO [finetune.py:992] (1/2) Epoch 16, batch 3050, loss[loss=0.1524, simple_loss=0.2445, pruned_loss=0.03011, over 12173.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2522, pruned_loss=0.03661, over 2374663.81 frames. ], batch size: 31, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:16:31,842 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3499, 4.9707, 5.3478, 4.6431, 4.9938, 4.7159, 5.3260, 4.9459], device='cuda:1'), covar=tensor([0.0308, 0.0378, 0.0288, 0.0302, 0.0480, 0.0391, 0.0315, 0.0374], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0282, 0.0298, 0.0274, 0.0275, 0.0275, 0.0250, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:16:39,066 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286588.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:16:39,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.577e+02 3.122e+02 3.634e+02 7.326e+02, threshold=6.244e+02, percent-clipped=2.0 2023-05-17 05:16:55,653 INFO [finetune.py:992] (1/2) Epoch 16, batch 3100, loss[loss=0.1585, simple_loss=0.2545, pruned_loss=0.03123, over 12358.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.2521, pruned_loss=0.03634, over 2377869.53 frames. ], batch size: 36, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:17:24,224 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286649.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:17:25,493 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286651.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:17:32,344 INFO [finetune.py:992] (1/2) Epoch 16, batch 3150, loss[loss=0.1589, simple_loss=0.2621, pruned_loss=0.02785, over 12178.00 frames. ], tot_loss[loss=0.1622, simple_loss=0.2517, pruned_loss=0.03634, over 2382024.34 frames. ], batch size: 35, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:17:39,776 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286671.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:17:52,329 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.049e+02 2.642e+02 3.020e+02 3.724e+02 5.646e+02, threshold=6.040e+02, percent-clipped=0.0 2023-05-17 05:18:00,302 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286700.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:18:07,771 INFO [finetune.py:992] (1/2) Epoch 16, batch 3200, loss[loss=0.1835, simple_loss=0.2769, pruned_loss=0.04505, over 12062.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2524, pruned_loss=0.03672, over 2381549.76 frames. ], batch size: 37, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:18:11,981 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0879, 5.9200, 5.5703, 5.4988, 6.0218, 5.2644, 5.3556, 5.5104], device='cuda:1'), covar=tensor([0.1662, 0.1112, 0.1047, 0.2076, 0.0945, 0.2473, 0.2191, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0505, 0.0406, 0.0455, 0.0474, 0.0438, 0.0401, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:18:23,404 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286732.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:18:43,770 INFO [finetune.py:992] (1/2) Epoch 16, batch 3250, loss[loss=0.1593, simple_loss=0.2468, pruned_loss=0.03592, over 11484.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.253, pruned_loss=0.03687, over 2382197.66 frames. ], batch size: 48, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:18:47,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3654, 4.0148, 3.9961, 4.4209, 2.7900, 3.7804, 2.5502, 4.0175], device='cuda:1'), covar=tensor([0.1495, 0.0733, 0.0889, 0.0552, 0.1336, 0.0668, 0.1955, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0267, 0.0298, 0.0358, 0.0242, 0.0243, 0.0263, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:18:51,078 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286770.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:18:57,299 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286779.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:19:01,457 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286785.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:19:04,247 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.645e+02 3.195e+02 3.737e+02 6.329e+02, threshold=6.389e+02, percent-clipped=3.0 2023-05-17 05:19:20,209 INFO [finetune.py:992] (1/2) Epoch 16, batch 3300, loss[loss=0.1758, simple_loss=0.2634, pruned_loss=0.04406, over 12060.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2539, pruned_loss=0.03715, over 2378133.00 frames. ], batch size: 42, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:19:31,431 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286827.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:19:34,256 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286831.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:19:35,609 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286833.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:19:38,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6235, 3.6728, 3.3640, 3.1893, 3.0515, 2.8282, 3.6964, 2.3857], device='cuda:1'), covar=tensor([0.0416, 0.0174, 0.0191, 0.0213, 0.0378, 0.0394, 0.0161, 0.0564], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0168, 0.0172, 0.0194, 0.0208, 0.0203, 0.0178, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:19:40,045 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8694, 4.6793, 4.6183, 4.6501, 4.3718, 4.8013, 4.7353, 4.9346], device='cuda:1'), covar=tensor([0.0251, 0.0209, 0.0202, 0.0462, 0.0788, 0.0462, 0.0198, 0.0213], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0203, 0.0198, 0.0256, 0.0248, 0.0230, 0.0184, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 05:19:42,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 05:19:55,547 INFO [finetune.py:992] (1/2) Epoch 16, batch 3350, loss[loss=0.1646, simple_loss=0.2599, pruned_loss=0.03469, over 10480.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.254, pruned_loss=0.03723, over 2378305.34 frames. ], batch size: 68, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:20:15,937 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.653e+02 2.997e+02 3.607e+02 6.484e+02, threshold=5.995e+02, percent-clipped=1.0 2023-05-17 05:20:31,630 INFO [finetune.py:992] (1/2) Epoch 16, batch 3400, loss[loss=0.1503, simple_loss=0.2301, pruned_loss=0.03524, over 12198.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2541, pruned_loss=0.03725, over 2376285.95 frames. ], batch size: 29, lr: 3.42e-03, grad_scale: 16.0 2023-05-17 05:20:55,575 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286944.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:21:00,668 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286951.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:21:07,604 INFO [finetune.py:992] (1/2) Epoch 16, batch 3450, loss[loss=0.1559, simple_loss=0.2494, pruned_loss=0.0312, over 12297.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2544, pruned_loss=0.03747, over 2365337.38 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:21:27,923 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 2.606e+02 3.061e+02 3.584e+02 6.458e+02, threshold=6.122e+02, percent-clipped=1.0 2023-05-17 05:21:35,158 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286999.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:21:36,294 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287000.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:21:36,335 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2540, 2.5290, 3.4725, 4.2329, 3.7803, 4.1942, 3.4604, 3.0851], device='cuda:1'), covar=tensor([0.0049, 0.0419, 0.0182, 0.0055, 0.0140, 0.0087, 0.0161, 0.0363], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0124, 0.0106, 0.0080, 0.0105, 0.0118, 0.0101, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:21:43,904 INFO [finetune.py:992] (1/2) Epoch 16, batch 3500, loss[loss=0.1465, simple_loss=0.2383, pruned_loss=0.02738, over 12308.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2542, pruned_loss=0.03736, over 2367440.65 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:21:55,436 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287027.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:22:11,117 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287048.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:22:20,989 INFO [finetune.py:992] (1/2) Epoch 16, batch 3550, loss[loss=0.1591, simple_loss=0.2447, pruned_loss=0.03675, over 12247.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2531, pruned_loss=0.03727, over 2367224.96 frames. ], batch size: 32, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:22:36,205 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7428, 2.9992, 4.6517, 4.8220, 2.9797, 2.5937, 3.1208, 2.1865], device='cuda:1'), covar=tensor([0.1742, 0.3082, 0.0538, 0.0432, 0.1367, 0.2686, 0.2817, 0.4412], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0400, 0.0286, 0.0309, 0.0284, 0.0326, 0.0405, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:22:41,036 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.536e+02 2.993e+02 3.580e+02 5.373e+02, threshold=5.986e+02, percent-clipped=0.0 2023-05-17 05:22:56,519 INFO [finetune.py:992] (1/2) Epoch 16, batch 3600, loss[loss=0.2174, simple_loss=0.3049, pruned_loss=0.06498, over 8031.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2537, pruned_loss=0.03753, over 2360804.72 frames. ], batch size: 97, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:23:05,117 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2504, 2.8234, 3.8902, 3.2433, 3.6221, 3.4108, 2.8108, 3.7301], device='cuda:1'), covar=tensor([0.0160, 0.0325, 0.0160, 0.0263, 0.0266, 0.0188, 0.0359, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0213, 0.0202, 0.0196, 0.0227, 0.0174, 0.0206, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:23:07,167 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287126.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:23:27,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 05:23:28,790 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8730, 3.0844, 4.7738, 4.9238, 2.9913, 2.7749, 3.2000, 2.2965], device='cuda:1'), covar=tensor([0.1683, 0.2965, 0.0473, 0.0409, 0.1422, 0.2534, 0.2746, 0.4326], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0399, 0.0285, 0.0308, 0.0283, 0.0324, 0.0404, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:23:32,144 INFO [finetune.py:992] (1/2) Epoch 16, batch 3650, loss[loss=0.1437, simple_loss=0.2266, pruned_loss=0.03036, over 12359.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2535, pruned_loss=0.03719, over 2364114.32 frames. ], batch size: 30, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:23:52,648 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.588e+02 2.548e+02 3.085e+02 3.772e+02 8.104e+02, threshold=6.169e+02, percent-clipped=4.0 2023-05-17 05:24:09,426 INFO [finetune.py:992] (1/2) Epoch 16, batch 3700, loss[loss=0.1362, simple_loss=0.2312, pruned_loss=0.02059, over 12283.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2532, pruned_loss=0.03682, over 2366823.25 frames. ], batch size: 33, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:24:31,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-05-17 05:24:33,380 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287244.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:24:43,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-05-17 05:24:45,407 INFO [finetune.py:992] (1/2) Epoch 16, batch 3750, loss[loss=0.1771, simple_loss=0.2663, pruned_loss=0.04394, over 12147.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.253, pruned_loss=0.03684, over 2367358.79 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:24:49,928 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9620, 4.8337, 4.7285, 4.7458, 4.4671, 4.9409, 4.8271, 5.0771], device='cuda:1'), covar=tensor([0.0314, 0.0178, 0.0230, 0.0459, 0.0847, 0.0367, 0.0190, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0202, 0.0198, 0.0256, 0.0247, 0.0229, 0.0183, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 05:25:05,615 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.523e+02 2.915e+02 3.511e+02 5.231e+02, threshold=5.830e+02, percent-clipped=0.0 2023-05-17 05:25:07,706 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287292.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:25:16,259 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2596, 6.1942, 5.7863, 5.7865, 6.2459, 5.5044, 5.6355, 5.7604], device='cuda:1'), covar=tensor([0.1531, 0.0825, 0.1238, 0.1734, 0.0831, 0.2087, 0.1767, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0502, 0.0408, 0.0456, 0.0476, 0.0438, 0.0403, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:25:21,151 INFO [finetune.py:992] (1/2) Epoch 16, batch 3800, loss[loss=0.1493, simple_loss=0.2437, pruned_loss=0.0274, over 12267.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2534, pruned_loss=0.037, over 2372877.74 frames. ], batch size: 32, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:25:33,330 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287327.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:25:46,012 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5153, 5.0534, 5.5130, 4.8403, 5.1409, 4.9214, 5.5325, 5.1342], device='cuda:1'), covar=tensor([0.0238, 0.0405, 0.0245, 0.0256, 0.0403, 0.0313, 0.0201, 0.0261], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0283, 0.0299, 0.0273, 0.0276, 0.0275, 0.0250, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:25:54,740 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1973, 4.9060, 5.0828, 5.0736, 4.9756, 5.0695, 5.0805, 2.7729], device='cuda:1'), covar=tensor([0.0099, 0.0062, 0.0064, 0.0058, 0.0038, 0.0101, 0.0057, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0080, 0.0082, 0.0074, 0.0061, 0.0094, 0.0082, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:25:58,068 INFO [finetune.py:992] (1/2) Epoch 16, batch 3850, loss[loss=0.1481, simple_loss=0.2411, pruned_loss=0.02753, over 12326.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.037, over 2364787.70 frames. ], batch size: 30, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:26:02,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0 2023-05-17 05:26:08,176 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287375.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:26:18,138 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.775e+02 3.174e+02 3.684e+02 7.449e+02, threshold=6.349e+02, percent-clipped=3.0 2023-05-17 05:26:30,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-05-17 05:26:33,966 INFO [finetune.py:992] (1/2) Epoch 16, batch 3900, loss[loss=0.194, simple_loss=0.2816, pruned_loss=0.05321, over 12041.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2533, pruned_loss=0.03718, over 2373144.12 frames. ], batch size: 37, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:26:34,130 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287411.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:26:44,668 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287426.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:27:09,357 INFO [finetune.py:992] (1/2) Epoch 16, batch 3950, loss[loss=0.1613, simple_loss=0.2548, pruned_loss=0.03385, over 12191.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2538, pruned_loss=0.03755, over 2376289.59 frames. ], batch size: 35, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:27:18,116 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287472.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:27:19,371 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287474.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:27:30,784 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.806e+02 3.106e+02 3.714e+02 5.718e+02, threshold=6.213e+02, percent-clipped=0.0 2023-05-17 05:27:46,595 INFO [finetune.py:992] (1/2) Epoch 16, batch 4000, loss[loss=0.1519, simple_loss=0.2386, pruned_loss=0.03259, over 12169.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2541, pruned_loss=0.03729, over 2378695.96 frames. ], batch size: 29, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:28:04,620 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287536.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:28:14,052 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 05:28:21,978 INFO [finetune.py:992] (1/2) Epoch 16, batch 4050, loss[loss=0.1696, simple_loss=0.2592, pruned_loss=0.04002, over 11909.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2543, pruned_loss=0.03764, over 2374231.18 frames. ], batch size: 44, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:28:22,772 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287562.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:28:34,243 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3738, 4.8544, 2.9408, 2.7882, 4.2122, 2.6409, 4.0070, 3.3435], device='cuda:1'), covar=tensor([0.0781, 0.0512, 0.1253, 0.1594, 0.0253, 0.1419, 0.0491, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0264, 0.0180, 0.0205, 0.0146, 0.0187, 0.0201, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:28:37,819 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287583.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:28:41,933 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.640e+02 3.108e+02 3.631e+02 7.122e+02, threshold=6.216e+02, percent-clipped=2.0 2023-05-17 05:28:47,786 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287597.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:28:58,551 INFO [finetune.py:992] (1/2) Epoch 16, batch 4100, loss[loss=0.1862, simple_loss=0.2786, pruned_loss=0.04688, over 11215.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2535, pruned_loss=0.03734, over 2368649.46 frames. ], batch size: 55, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:29:07,367 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287623.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:29:23,109 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287644.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:29:35,005 INFO [finetune.py:992] (1/2) Epoch 16, batch 4150, loss[loss=0.151, simple_loss=0.2456, pruned_loss=0.02822, over 12151.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.03704, over 2378817.61 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 2023-05-17 05:29:54,849 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1362, 6.0366, 5.5333, 5.6105, 6.0927, 5.4309, 5.4369, 5.5890], device='cuda:1'), covar=tensor([0.1665, 0.0847, 0.0946, 0.1652, 0.0811, 0.2044, 0.1877, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0501, 0.0406, 0.0454, 0.0475, 0.0438, 0.0403, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:29:55,483 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.445e+02 3.091e+02 3.464e+02 5.875e+02, threshold=6.183e+02, percent-clipped=0.0 2023-05-17 05:30:04,795 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2334, 4.0029, 4.2600, 4.4255, 3.0584, 3.9406, 2.5416, 4.2324], device='cuda:1'), covar=tensor([0.1679, 0.0826, 0.0863, 0.0745, 0.1249, 0.0690, 0.2008, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0272, 0.0303, 0.0362, 0.0247, 0.0248, 0.0265, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:30:10,885 INFO [finetune.py:992] (1/2) Epoch 16, batch 4200, loss[loss=0.1731, simple_loss=0.2698, pruned_loss=0.0382, over 11675.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2526, pruned_loss=0.03661, over 2380397.05 frames. ], batch size: 48, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:30:17,292 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287720.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:30:46,909 INFO [finetune.py:992] (1/2) Epoch 16, batch 4250, loss[loss=0.1642, simple_loss=0.2545, pruned_loss=0.03692, over 10661.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2525, pruned_loss=0.03689, over 2377515.49 frames. ], batch size: 68, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:30:51,449 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287767.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:30:56,004 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287773.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:31:01,705 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287781.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:31:05,283 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2535, 4.5034, 4.1087, 4.9201, 4.5379, 2.7668, 4.2756, 3.0268], device='cuda:1'), covar=tensor([0.0815, 0.0935, 0.1461, 0.0542, 0.1096, 0.1874, 0.1112, 0.3471], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0385, 0.0365, 0.0333, 0.0376, 0.0279, 0.0353, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:31:08,557 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.769e+02 3.131e+02 3.644e+02 2.115e+03, threshold=6.263e+02, percent-clipped=1.0 2023-05-17 05:31:09,387 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8775, 4.5819, 4.1398, 4.1239, 4.6429, 4.0897, 4.1577, 3.9668], device='cuda:1'), covar=tensor([0.1832, 0.1124, 0.1512, 0.2152, 0.1121, 0.1994, 0.1996, 0.1708], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0507, 0.0410, 0.0460, 0.0480, 0.0443, 0.0407, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:31:23,788 INFO [finetune.py:992] (1/2) Epoch 16, batch 4300, loss[loss=0.1704, simple_loss=0.2621, pruned_loss=0.03935, over 11883.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2523, pruned_loss=0.03699, over 2365431.97 frames. ], batch size: 44, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:31:40,647 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287834.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:31:59,322 INFO [finetune.py:992] (1/2) Epoch 16, batch 4350, loss[loss=0.1712, simple_loss=0.2665, pruned_loss=0.03789, over 12006.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2529, pruned_loss=0.03695, over 2369142.63 frames. ], batch size: 40, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:32:06,639 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1570, 2.6253, 3.7592, 3.1367, 3.4525, 3.2560, 2.7175, 3.5970], device='cuda:1'), covar=tensor([0.0163, 0.0366, 0.0159, 0.0302, 0.0208, 0.0215, 0.0398, 0.0183], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0212, 0.0201, 0.0195, 0.0227, 0.0175, 0.0204, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:32:19,829 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.910e+02 3.370e+02 3.789e+02 2.398e+03, threshold=6.741e+02, percent-clipped=1.0 2023-05-17 05:32:21,378 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287892.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:32:35,338 INFO [finetune.py:992] (1/2) Epoch 16, batch 4400, loss[loss=0.1591, simple_loss=0.2526, pruned_loss=0.0328, over 12129.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2533, pruned_loss=0.03716, over 2368619.08 frames. ], batch size: 30, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:32:40,448 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287918.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:32:52,716 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7741, 3.0970, 3.4539, 4.7208, 2.6111, 4.6330, 4.7789, 4.8960], device='cuda:1'), covar=tensor([0.0153, 0.1069, 0.0480, 0.0125, 0.1278, 0.0250, 0.0126, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0203, 0.0183, 0.0123, 0.0191, 0.0182, 0.0177, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:32:56,155 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287939.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:33:11,646 INFO [finetune.py:992] (1/2) Epoch 16, batch 4450, loss[loss=0.1503, simple_loss=0.2349, pruned_loss=0.03279, over 12181.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2536, pruned_loss=0.03756, over 2356250.70 frames. ], batch size: 29, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:33:32,222 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.609e+02 3.072e+02 3.662e+02 6.835e+02, threshold=6.144e+02, percent-clipped=1.0 2023-05-17 05:33:45,283 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288003.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:33:49,017 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4119, 2.4749, 3.1851, 4.3722, 2.0740, 4.3469, 4.4881, 4.4946], device='cuda:1'), covar=tensor([0.0150, 0.1379, 0.0551, 0.0151, 0.1521, 0.0228, 0.0149, 0.0136], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0203, 0.0183, 0.0123, 0.0191, 0.0182, 0.0177, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:33:50,916 INFO [finetune.py:992] (1/2) Epoch 16, batch 4500, loss[loss=0.1911, simple_loss=0.2699, pruned_loss=0.05617, over 7956.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2536, pruned_loss=0.03727, over 2356962.94 frames. ], batch size: 98, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:34:07,496 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1225, 4.9878, 4.9360, 5.0613, 3.8838, 5.2411, 5.1674, 5.2247], device='cuda:1'), covar=tensor([0.0308, 0.0283, 0.0246, 0.0397, 0.1446, 0.0414, 0.0202, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0202, 0.0198, 0.0256, 0.0246, 0.0228, 0.0183, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 05:34:27,053 INFO [finetune.py:992] (1/2) Epoch 16, batch 4550, loss[loss=0.1852, simple_loss=0.2767, pruned_loss=0.0468, over 12034.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2545, pruned_loss=0.03734, over 2360815.61 frames. ], batch size: 42, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:34:29,344 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288064.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:34:31,426 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288067.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:34:38,562 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288076.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:34:48,453 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.623e+02 3.177e+02 3.615e+02 5.730e+02, threshold=6.355e+02, percent-clipped=0.0 2023-05-17 05:34:59,859 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288106.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:35:03,220 INFO [finetune.py:992] (1/2) Epoch 16, batch 4600, loss[loss=0.1689, simple_loss=0.2701, pruned_loss=0.03385, over 11792.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2538, pruned_loss=0.03676, over 2368308.17 frames. ], batch size: 44, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:35:06,213 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288115.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:35:16,257 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288129.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:35:32,577 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288152.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:35:38,565 INFO [finetune.py:992] (1/2) Epoch 16, batch 4650, loss[loss=0.1613, simple_loss=0.2483, pruned_loss=0.03718, over 12345.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2539, pruned_loss=0.03673, over 2372978.45 frames. ], batch size: 31, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:35:43,061 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288167.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 05:36:00,055 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.786e+02 3.203e+02 3.745e+02 6.992e+02, threshold=6.407e+02, percent-clipped=1.0 2023-05-17 05:36:01,476 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288192.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:36:15,076 INFO [finetune.py:992] (1/2) Epoch 16, batch 4700, loss[loss=0.1818, simple_loss=0.2663, pruned_loss=0.0487, over 12129.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2542, pruned_loss=0.03691, over 2374225.14 frames. ], batch size: 39, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:36:16,702 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288213.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:36:20,196 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288218.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:36:27,448 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7044, 2.8535, 4.6953, 4.7725, 2.9001, 2.6518, 3.0048, 2.3476], device='cuda:1'), covar=tensor([0.1750, 0.3007, 0.0475, 0.0421, 0.1415, 0.2467, 0.2954, 0.4123], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0398, 0.0285, 0.0308, 0.0282, 0.0323, 0.0402, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:36:35,803 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288239.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:36:36,395 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288240.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:36:51,213 INFO [finetune.py:992] (1/2) Epoch 16, batch 4750, loss[loss=0.1681, simple_loss=0.2547, pruned_loss=0.04073, over 12122.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.254, pruned_loss=0.03709, over 2375528.87 frames. ], batch size: 30, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:36:54,741 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288266.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:37:09,775 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288287.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:37:11,897 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.140e+02 2.896e+02 3.408e+02 4.008e+02 6.839e+02, threshold=6.815e+02, percent-clipped=1.0 2023-05-17 05:37:26,693 INFO [finetune.py:992] (1/2) Epoch 16, batch 4800, loss[loss=0.2007, simple_loss=0.2848, pruned_loss=0.05825, over 8173.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2545, pruned_loss=0.0373, over 2373661.91 frames. ], batch size: 98, lr: 3.41e-03, grad_scale: 8.0 2023-05-17 05:37:38,215 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288327.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:37:45,616 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 05:37:51,693 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288345.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:01,586 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288359.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:02,951 INFO [finetune.py:992] (1/2) Epoch 16, batch 4850, loss[loss=0.1772, simple_loss=0.273, pruned_loss=0.04072, over 11266.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2547, pruned_loss=0.03763, over 2375929.70 frames. ], batch size: 55, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:38:14,266 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288376.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:22,877 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288388.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:24,081 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 2.698e+02 3.119e+02 3.725e+02 7.730e+02, threshold=6.239e+02, percent-clipped=1.0 2023-05-17 05:38:28,636 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288396.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:35,860 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288406.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:39,182 INFO [finetune.py:992] (1/2) Epoch 16, batch 4900, loss[loss=0.1482, simple_loss=0.2292, pruned_loss=0.03364, over 12346.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2548, pruned_loss=0.03773, over 2377676.63 frames. ], batch size: 30, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:38:40,188 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9131, 3.5192, 5.2139, 2.7854, 2.9408, 3.7580, 3.3092, 3.7852], device='cuda:1'), covar=tensor([0.0369, 0.1044, 0.0274, 0.1089, 0.1819, 0.1567, 0.1314, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0240, 0.0243, 0.0261, 0.0186, 0.0241, 0.0297, 0.0229, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:38:48,491 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288424.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:38:52,219 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288429.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:39:12,018 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288457.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:39:14,524 INFO [finetune.py:992] (1/2) Epoch 16, batch 4950, loss[loss=0.1698, simple_loss=0.2631, pruned_loss=0.03826, over 12190.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2546, pruned_loss=0.03787, over 2364373.96 frames. ], batch size: 35, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:39:15,348 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288462.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:39:19,067 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-05-17 05:39:26,341 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288477.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:39:35,411 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.842e+02 2.739e+02 3.110e+02 3.721e+02 3.368e+03, threshold=6.220e+02, percent-clipped=4.0 2023-05-17 05:39:49,185 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288508.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:39:51,247 INFO [finetune.py:992] (1/2) Epoch 16, batch 5000, loss[loss=0.2152, simple_loss=0.2929, pruned_loss=0.06877, over 8383.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2553, pruned_loss=0.03801, over 2361284.22 frames. ], batch size: 97, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:40:14,150 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2858, 4.8501, 5.2551, 4.6239, 4.8843, 4.6840, 5.3064, 4.9963], device='cuda:1'), covar=tensor([0.0257, 0.0384, 0.0272, 0.0273, 0.0420, 0.0373, 0.0192, 0.0314], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0281, 0.0298, 0.0272, 0.0275, 0.0275, 0.0248, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:40:26,550 INFO [finetune.py:992] (1/2) Epoch 16, batch 5050, loss[loss=0.1603, simple_loss=0.2463, pruned_loss=0.03716, over 12110.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2547, pruned_loss=0.03773, over 2370155.88 frames. ], batch size: 33, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:40:47,126 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 2.542e+02 3.078e+02 3.605e+02 6.274e+02, threshold=6.155e+02, percent-clipped=1.0 2023-05-17 05:41:00,958 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288609.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:41:01,708 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5027, 3.5782, 3.2918, 3.1108, 2.8499, 2.7620, 3.5993, 2.4493], device='cuda:1'), covar=tensor([0.0444, 0.0175, 0.0234, 0.0230, 0.0425, 0.0421, 0.0183, 0.0500], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0166, 0.0170, 0.0193, 0.0203, 0.0203, 0.0176, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:41:02,188 INFO [finetune.py:992] (1/2) Epoch 16, batch 5100, loss[loss=0.1632, simple_loss=0.2557, pruned_loss=0.03529, over 12104.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2553, pruned_loss=0.038, over 2365931.85 frames. ], batch size: 33, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:41:20,667 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.6500, 5.4628, 5.5649, 5.5916, 5.2153, 5.2731, 5.0961, 5.4391], device='cuda:1'), covar=tensor([0.0573, 0.0582, 0.0794, 0.0586, 0.1965, 0.1362, 0.0558, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0730, 0.0642, 0.0662, 0.0885, 0.0782, 0.0593, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 05:41:37,773 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288659.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:41:39,083 INFO [finetune.py:992] (1/2) Epoch 16, batch 5150, loss[loss=0.1609, simple_loss=0.2564, pruned_loss=0.03264, over 12198.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2543, pruned_loss=0.03766, over 2369811.66 frames. ], batch size: 35, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:41:45,565 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288670.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:41:54,747 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288683.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:41:59,641 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.642e+02 3.043e+02 3.741e+02 6.453e+02, threshold=6.085e+02, percent-clipped=1.0 2023-05-17 05:42:00,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-17 05:42:07,205 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288701.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:42:11,326 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288707.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:42:14,093 INFO [finetune.py:992] (1/2) Epoch 16, batch 5200, loss[loss=0.1636, simple_loss=0.2598, pruned_loss=0.03363, over 12355.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2547, pruned_loss=0.03778, over 2375288.01 frames. ], batch size: 35, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:42:43,588 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288752.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:42:50,560 INFO [finetune.py:992] (1/2) Epoch 16, batch 5250, loss[loss=0.154, simple_loss=0.246, pruned_loss=0.03102, over 12160.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03769, over 2374207.62 frames. ], batch size: 34, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:42:51,416 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288762.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:43:11,941 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.670e+02 3.124e+02 3.667e+02 1.117e+03, threshold=6.248e+02, percent-clipped=2.0 2023-05-17 05:43:17,196 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288797.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:43:25,232 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288808.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:43:26,612 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288810.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:43:27,208 INFO [finetune.py:992] (1/2) Epoch 16, batch 5300, loss[loss=0.1826, simple_loss=0.2739, pruned_loss=0.04571, over 11547.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2547, pruned_loss=0.03738, over 2378331.13 frames. ], batch size: 48, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:43:40,954 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4901, 3.0039, 3.7708, 2.2509, 2.6639, 3.0401, 2.9291, 3.1885], device='cuda:1'), covar=tensor([0.0567, 0.1075, 0.0431, 0.1305, 0.1739, 0.1488, 0.1231, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0245, 0.0263, 0.0187, 0.0243, 0.0300, 0.0231, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:43:59,321 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288856.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:44:00,852 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288858.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:44:02,776 INFO [finetune.py:992] (1/2) Epoch 16, batch 5350, loss[loss=0.207, simple_loss=0.2901, pruned_loss=0.06193, over 8011.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2543, pruned_loss=0.03706, over 2377665.07 frames. ], batch size: 98, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:44:04,321 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3054, 5.0786, 5.2555, 5.2605, 4.8776, 4.9319, 4.6471, 5.1670], device='cuda:1'), covar=tensor([0.0639, 0.0608, 0.0897, 0.0568, 0.2003, 0.1416, 0.0624, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0725, 0.0636, 0.0657, 0.0878, 0.0777, 0.0588, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 05:44:23,248 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.696e+02 3.125e+02 3.859e+02 8.981e+02, threshold=6.250e+02, percent-clipped=2.0 2023-05-17 05:44:28,067 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-05-17 05:44:38,735 INFO [finetune.py:992] (1/2) Epoch 16, batch 5400, loss[loss=0.1511, simple_loss=0.2384, pruned_loss=0.03194, over 12134.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2538, pruned_loss=0.03699, over 2379866.24 frames. ], batch size: 30, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:45:07,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 05:45:14,464 INFO [finetune.py:992] (1/2) Epoch 16, batch 5450, loss[loss=0.1608, simple_loss=0.2566, pruned_loss=0.03251, over 12337.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2541, pruned_loss=0.03748, over 2378462.27 frames. ], batch size: 35, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:45:17,481 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288965.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:45:30,274 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288983.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:45:35,054 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.974e+02 2.608e+02 3.066e+02 3.634e+02 6.198e+02, threshold=6.131e+02, percent-clipped=0.0 2023-05-17 05:45:43,307 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289001.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:45:50,248 INFO [finetune.py:992] (1/2) Epoch 16, batch 5500, loss[loss=0.1481, simple_loss=0.2441, pruned_loss=0.02611, over 12184.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2536, pruned_loss=0.03727, over 2384558.76 frames. ], batch size: 31, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:46:04,657 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289031.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:46:15,337 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289046.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:46:18,325 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289049.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:46:20,442 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289052.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:46:26,842 INFO [finetune.py:992] (1/2) Epoch 16, batch 5550, loss[loss=0.1599, simple_loss=0.2488, pruned_loss=0.03553, over 12343.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2545, pruned_loss=0.03775, over 2372724.94 frames. ], batch size: 36, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:46:41,001 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-17 05:46:48,379 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.813e+02 2.627e+02 3.073e+02 3.584e+02 7.360e+02, threshold=6.145e+02, percent-clipped=6.0 2023-05-17 05:46:55,635 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289100.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:47:00,891 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289107.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 05:47:03,494 INFO [finetune.py:992] (1/2) Epoch 16, batch 5600, loss[loss=0.1336, simple_loss=0.2207, pruned_loss=0.02324, over 12275.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2531, pruned_loss=0.03716, over 2379473.60 frames. ], batch size: 28, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:47:09,244 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0003, 4.8513, 4.7182, 4.8216, 4.5176, 4.9952, 4.9073, 5.0738], device='cuda:1'), covar=tensor([0.0190, 0.0170, 0.0215, 0.0345, 0.0712, 0.0294, 0.0151, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0205, 0.0201, 0.0260, 0.0250, 0.0233, 0.0186, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 05:47:21,442 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289136.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:47:33,613 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289153.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:47:39,345 INFO [finetune.py:992] (1/2) Epoch 16, batch 5650, loss[loss=0.1634, simple_loss=0.2583, pruned_loss=0.03431, over 12148.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.253, pruned_loss=0.03721, over 2375322.09 frames. ], batch size: 36, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:47:49,317 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4795, 5.0374, 5.4705, 4.8267, 5.0609, 4.9029, 5.4964, 5.1332], device='cuda:1'), covar=tensor([0.0284, 0.0429, 0.0271, 0.0286, 0.0421, 0.0365, 0.0229, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0287, 0.0306, 0.0278, 0.0281, 0.0281, 0.0255, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:47:59,909 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.624e+02 3.112e+02 3.713e+02 1.702e+03, threshold=6.223e+02, percent-clipped=2.0 2023-05-17 05:48:05,860 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289197.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:48:16,125 INFO [finetune.py:992] (1/2) Epoch 16, batch 5700, loss[loss=0.1346, simple_loss=0.2246, pruned_loss=0.02233, over 11792.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2524, pruned_loss=0.03678, over 2379400.11 frames. ], batch size: 26, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:48:49,969 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1487, 4.0633, 4.0730, 4.3568, 2.9697, 3.7726, 2.5757, 4.1131], device='cuda:1'), covar=tensor([0.1675, 0.0747, 0.1017, 0.0699, 0.1191, 0.0757, 0.1986, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0269, 0.0301, 0.0362, 0.0246, 0.0246, 0.0265, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:48:52,409 INFO [finetune.py:992] (1/2) Epoch 16, batch 5750, loss[loss=0.1965, simple_loss=0.2813, pruned_loss=0.05587, over 11326.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2518, pruned_loss=0.03653, over 2378494.61 frames. ], batch size: 55, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:48:55,487 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289265.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:49:13,134 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.626e+02 2.935e+02 3.698e+02 1.071e+03, threshold=5.870e+02, percent-clipped=1.0 2023-05-17 05:49:28,336 INFO [finetune.py:992] (1/2) Epoch 16, batch 5800, loss[loss=0.1546, simple_loss=0.2446, pruned_loss=0.03226, over 12033.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2512, pruned_loss=0.03643, over 2376150.33 frames. ], batch size: 31, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:49:29,890 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289313.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:50:04,469 INFO [finetune.py:992] (1/2) Epoch 16, batch 5850, loss[loss=0.1584, simple_loss=0.2638, pruned_loss=0.0265, over 11828.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.2513, pruned_loss=0.03618, over 2385131.24 frames. ], batch size: 44, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:50:19,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 2023-05-17 05:50:25,900 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.733e+02 3.224e+02 3.691e+02 7.666e+02, threshold=6.448e+02, percent-clipped=6.0 2023-05-17 05:50:35,043 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289402.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:50:41,321 INFO [finetune.py:992] (1/2) Epoch 16, batch 5900, loss[loss=0.1511, simple_loss=0.2394, pruned_loss=0.03142, over 12246.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2522, pruned_loss=0.03669, over 2382142.44 frames. ], batch size: 32, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:50:45,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.50 vs. limit=5.0 2023-05-17 05:51:02,234 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9600, 5.6587, 5.3185, 5.2347, 5.7846, 5.1272, 5.1932, 5.2072], device='cuda:1'), covar=tensor([0.1423, 0.0926, 0.1202, 0.1856, 0.0920, 0.2013, 0.2031, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0506, 0.0404, 0.0456, 0.0472, 0.0434, 0.0398, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:51:11,570 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289453.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:51:17,110 INFO [finetune.py:992] (1/2) Epoch 16, batch 5950, loss[loss=0.131, simple_loss=0.2102, pruned_loss=0.02586, over 11992.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2525, pruned_loss=0.03654, over 2382180.65 frames. ], batch size: 28, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:51:26,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-05-17 05:51:37,577 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.502e+02 2.971e+02 3.587e+02 5.831e+02, threshold=5.943e+02, percent-clipped=0.0 2023-05-17 05:51:39,861 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289492.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:51:46,190 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289501.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:51:47,784 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289503.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:51:53,291 INFO [finetune.py:992] (1/2) Epoch 16, batch 6000, loss[loss=0.16, simple_loss=0.2383, pruned_loss=0.04087, over 12344.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2525, pruned_loss=0.03663, over 2387060.79 frames. ], batch size: 30, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:51:53,291 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 05:52:11,667 INFO [finetune.py:1026] (1/2) Epoch 16, validation: loss=0.3117, simple_loss=0.3889, pruned_loss=0.1172, over 1020973.00 frames. 2023-05-17 05:52:11,668 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 05:52:47,257 INFO [finetune.py:992] (1/2) Epoch 16, batch 6050, loss[loss=0.2456, simple_loss=0.3131, pruned_loss=0.08908, over 7738.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2539, pruned_loss=0.03727, over 2370943.04 frames. ], batch size: 98, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:52:49,512 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289564.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:53:07,813 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.709e+02 3.140e+02 3.726e+02 6.638e+02, threshold=6.280e+02, percent-clipped=1.0 2023-05-17 05:53:24,181 INFO [finetune.py:992] (1/2) Epoch 16, batch 6100, loss[loss=0.2375, simple_loss=0.3097, pruned_loss=0.08264, over 8377.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.255, pruned_loss=0.03785, over 2358199.66 frames. ], batch size: 99, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:53:29,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-17 05:54:00,126 INFO [finetune.py:992] (1/2) Epoch 16, batch 6150, loss[loss=0.2911, simple_loss=0.3471, pruned_loss=0.1176, over 7963.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2549, pruned_loss=0.03806, over 2355358.02 frames. ], batch size: 99, lr: 3.40e-03, grad_scale: 8.0 2023-05-17 05:54:20,929 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.594e+02 3.012e+02 3.792e+02 6.022e+02, threshold=6.025e+02, percent-clipped=0.0 2023-05-17 05:54:29,790 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289702.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:54:36,047 INFO [finetune.py:992] (1/2) Epoch 16, batch 6200, loss[loss=0.157, simple_loss=0.2493, pruned_loss=0.03237, over 12166.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2549, pruned_loss=0.03861, over 2347702.25 frames. ], batch size: 34, lr: 3.40e-03, grad_scale: 16.0 2023-05-17 05:54:51,361 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9960, 3.3052, 5.5476, 2.6242, 2.5575, 4.2390, 3.3265, 4.1682], device='cuda:1'), covar=tensor([0.0372, 0.1295, 0.0166, 0.1219, 0.2070, 0.1203, 0.1350, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0244, 0.0261, 0.0187, 0.0241, 0.0299, 0.0229, 0.0271], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 05:54:58,321 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3610, 5.0144, 5.3653, 4.6743, 4.9758, 4.7496, 5.3553, 5.0542], device='cuda:1'), covar=tensor([0.0281, 0.0358, 0.0289, 0.0292, 0.0418, 0.0325, 0.0246, 0.0330], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0281, 0.0299, 0.0274, 0.0274, 0.0275, 0.0249, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 05:55:04,770 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289750.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 05:55:12,967 INFO [finetune.py:992] (1/2) Epoch 16, batch 6250, loss[loss=0.1705, simple_loss=0.2578, pruned_loss=0.04156, over 12132.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2547, pruned_loss=0.0383, over 2352664.38 frames. ], batch size: 39, lr: 3.40e-03, grad_scale: 16.0 2023-05-17 05:55:33,878 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.669e+02 3.033e+02 3.557e+02 7.068e+02, threshold=6.066e+02, percent-clipped=2.0 2023-05-17 05:55:35,480 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289792.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:55:37,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-05-17 05:55:49,214 INFO [finetune.py:992] (1/2) Epoch 16, batch 6300, loss[loss=0.1692, simple_loss=0.2646, pruned_loss=0.03687, over 12136.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.254, pruned_loss=0.03795, over 2355444.43 frames. ], batch size: 36, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:56:10,325 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289840.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:56:14,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 2023-05-17 05:56:23,887 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289859.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:56:25,166 INFO [finetune.py:992] (1/2) Epoch 16, batch 6350, loss[loss=0.1367, simple_loss=0.2186, pruned_loss=0.02735, over 12120.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2536, pruned_loss=0.03778, over 2357202.90 frames. ], batch size: 30, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:56:46,422 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.684e+02 3.149e+02 3.789e+02 6.168e+02, threshold=6.298e+02, percent-clipped=1.0 2023-05-17 05:57:02,187 INFO [finetune.py:992] (1/2) Epoch 16, batch 6400, loss[loss=0.1953, simple_loss=0.2776, pruned_loss=0.05651, over 12081.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2538, pruned_loss=0.03737, over 2359069.44 frames. ], batch size: 42, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:57:08,665 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289920.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 05:57:37,816 INFO [finetune.py:992] (1/2) Epoch 16, batch 6450, loss[loss=0.1566, simple_loss=0.2502, pruned_loss=0.03155, over 12197.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2548, pruned_loss=0.03775, over 2361493.04 frames. ], batch size: 35, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:57:46,711 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3879, 4.7013, 4.2524, 5.0567, 4.6757, 2.9118, 4.2348, 3.0538], device='cuda:1'), covar=tensor([0.0744, 0.0809, 0.1302, 0.0413, 0.0946, 0.1737, 0.1080, 0.3548], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0387, 0.0367, 0.0336, 0.0378, 0.0279, 0.0356, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:57:52,389 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289981.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 05:57:58,791 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.638e+02 3.095e+02 3.629e+02 6.932e+02, threshold=6.190e+02, percent-clipped=4.0 2023-05-17 05:58:04,653 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289998.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:58:17,164 INFO [finetune.py:992] (1/2) Epoch 16, batch 6500, loss[loss=0.1441, simple_loss=0.2311, pruned_loss=0.02861, over 11994.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.254, pruned_loss=0.0371, over 2367979.18 frames. ], batch size: 28, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:58:23,790 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5081, 3.5854, 3.2549, 3.0986, 2.8340, 2.7306, 3.5865, 2.4112], device='cuda:1'), covar=tensor([0.0415, 0.0173, 0.0215, 0.0216, 0.0441, 0.0432, 0.0149, 0.0498], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0168, 0.0173, 0.0196, 0.0206, 0.0205, 0.0179, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 05:58:46,636 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290050.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:58:47,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 05:58:52,798 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290059.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 05:58:53,998 INFO [finetune.py:992] (1/2) Epoch 16, batch 6550, loss[loss=0.1815, simple_loss=0.2735, pruned_loss=0.04472, over 10618.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2539, pruned_loss=0.03719, over 2372934.67 frames. ], batch size: 68, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:59:06,982 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.9301, 5.8791, 5.6757, 5.2108, 5.2210, 5.8266, 5.4086, 5.2281], device='cuda:1'), covar=tensor([0.0759, 0.1027, 0.0732, 0.1585, 0.0751, 0.0734, 0.1525, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0647, 0.0586, 0.0540, 0.0660, 0.0428, 0.0759, 0.0808, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 05:59:14,845 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.611e+02 2.892e+02 3.613e+02 1.747e+03, threshold=5.784e+02, percent-clipped=1.0 2023-05-17 05:59:29,842 INFO [finetune.py:992] (1/2) Epoch 16, batch 6600, loss[loss=0.1718, simple_loss=0.267, pruned_loss=0.0383, over 12126.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2534, pruned_loss=0.03705, over 2376845.60 frames. ], batch size: 38, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 05:59:30,097 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290111.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:00:04,681 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290159.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:00:05,902 INFO [finetune.py:992] (1/2) Epoch 16, batch 6650, loss[loss=0.1807, simple_loss=0.2791, pruned_loss=0.04116, over 12341.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2532, pruned_loss=0.03694, over 2373048.11 frames. ], batch size: 36, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:00:27,287 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.630e+02 3.118e+02 3.870e+02 7.746e+02, threshold=6.235e+02, percent-clipped=1.0 2023-05-17 06:00:40,570 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290207.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:00:43,345 INFO [finetune.py:992] (1/2) Epoch 16, batch 6700, loss[loss=0.1908, simple_loss=0.2771, pruned_loss=0.05226, over 11362.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2538, pruned_loss=0.03723, over 2374305.36 frames. ], batch size: 55, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:01:19,146 INFO [finetune.py:992] (1/2) Epoch 16, batch 6750, loss[loss=0.1636, simple_loss=0.2674, pruned_loss=0.02987, over 12196.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2538, pruned_loss=0.03751, over 2358813.82 frames. ], batch size: 35, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:01:21,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-05-17 06:01:29,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-05-17 06:01:29,924 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290276.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 06:01:33,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 06:01:39,842 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.624e+02 3.096e+02 3.902e+02 1.134e+03, threshold=6.192e+02, percent-clipped=3.0 2023-05-17 06:01:55,101 INFO [finetune.py:992] (1/2) Epoch 16, batch 6800, loss[loss=0.1613, simple_loss=0.2514, pruned_loss=0.03564, over 12109.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2547, pruned_loss=0.03817, over 2363100.27 frames. ], batch size: 32, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:01:58,712 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3146, 4.8860, 5.2764, 4.6458, 4.8855, 4.7496, 5.2824, 4.9534], device='cuda:1'), covar=tensor([0.0274, 0.0398, 0.0305, 0.0291, 0.0462, 0.0344, 0.0212, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0283, 0.0304, 0.0276, 0.0277, 0.0278, 0.0251, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:02:00,194 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290318.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:02:00,230 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1185, 2.4914, 3.6835, 3.0830, 3.3361, 3.2270, 2.5660, 3.5290], device='cuda:1'), covar=tensor([0.0143, 0.0400, 0.0152, 0.0267, 0.0205, 0.0195, 0.0407, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0213, 0.0203, 0.0196, 0.0229, 0.0176, 0.0206, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:02:01,006 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6826, 2.5107, 4.5856, 4.8597, 3.1477, 2.5328, 2.8014, 2.0199], device='cuda:1'), covar=tensor([0.1822, 0.3961, 0.0469, 0.0399, 0.1170, 0.2802, 0.3268, 0.5117], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0395, 0.0280, 0.0304, 0.0279, 0.0319, 0.0397, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:02:20,685 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1491, 4.4515, 2.7988, 2.5814, 3.8708, 2.5317, 3.8644, 3.0824], device='cuda:1'), covar=tensor([0.0793, 0.0602, 0.1131, 0.1535, 0.0317, 0.1371, 0.0516, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0262, 0.0178, 0.0205, 0.0145, 0.0185, 0.0201, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:02:26,158 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290354.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:02:30,948 INFO [finetune.py:992] (1/2) Epoch 16, batch 6850, loss[loss=0.1711, simple_loss=0.2642, pruned_loss=0.03897, over 12159.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2547, pruned_loss=0.03794, over 2369861.27 frames. ], batch size: 36, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:02:32,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 06:02:44,048 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290379.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:02:51,662 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.699e+02 3.058e+02 3.912e+02 8.049e+02, threshold=6.116e+02, percent-clipped=2.0 2023-05-17 06:03:02,857 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7653, 3.5531, 5.2938, 2.8791, 2.9537, 3.8694, 3.2351, 3.9882], device='cuda:1'), covar=tensor([0.0506, 0.1160, 0.0327, 0.1196, 0.1956, 0.1647, 0.1506, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0245, 0.0261, 0.0187, 0.0242, 0.0300, 0.0230, 0.0273], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:03:03,390 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290406.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:03:06,722 INFO [finetune.py:992] (1/2) Epoch 16, batch 6900, loss[loss=0.2279, simple_loss=0.3052, pruned_loss=0.07536, over 8379.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2547, pruned_loss=0.03786, over 2370474.01 frames. ], batch size: 97, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:03:39,278 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5656, 2.6176, 3.2115, 4.3892, 2.3720, 4.4203, 4.5381, 4.5929], device='cuda:1'), covar=tensor([0.0177, 0.1389, 0.0587, 0.0270, 0.1411, 0.0266, 0.0196, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0210, 0.0188, 0.0127, 0.0194, 0.0187, 0.0183, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:03:43,378 INFO [finetune.py:992] (1/2) Epoch 16, batch 6950, loss[loss=0.1639, simple_loss=0.2535, pruned_loss=0.03716, over 11577.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.255, pruned_loss=0.03761, over 2365982.35 frames. ], batch size: 48, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:03:48,655 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2106, 4.5206, 4.1321, 4.8457, 4.4155, 2.9522, 4.2468, 3.0646], device='cuda:1'), covar=tensor([0.0815, 0.0840, 0.1423, 0.0827, 0.1328, 0.1704, 0.1082, 0.3253], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0385, 0.0366, 0.0335, 0.0376, 0.0279, 0.0354, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:04:04,624 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.638e+02 3.087e+02 3.748e+02 8.122e+02, threshold=6.174e+02, percent-clipped=3.0 2023-05-17 06:04:10,988 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-17 06:04:19,813 INFO [finetune.py:992] (1/2) Epoch 16, batch 7000, loss[loss=0.1361, simple_loss=0.222, pruned_loss=0.02514, over 12372.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2552, pruned_loss=0.03749, over 2376967.33 frames. ], batch size: 30, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:04:29,433 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2536, 4.6864, 2.8556, 2.7945, 3.9702, 2.7447, 3.9431, 3.2708], device='cuda:1'), covar=tensor([0.0739, 0.0490, 0.1308, 0.1472, 0.0329, 0.1273, 0.0523, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0260, 0.0177, 0.0203, 0.0144, 0.0183, 0.0199, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:04:35,961 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5108, 3.6235, 3.2015, 3.0690, 2.8122, 2.7501, 3.5753, 2.3578], device='cuda:1'), covar=tensor([0.0412, 0.0137, 0.0233, 0.0214, 0.0462, 0.0358, 0.0136, 0.0530], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0168, 0.0173, 0.0195, 0.0206, 0.0205, 0.0180, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:04:56,244 INFO [finetune.py:992] (1/2) Epoch 16, batch 7050, loss[loss=0.1273, simple_loss=0.2099, pruned_loss=0.02232, over 12253.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2544, pruned_loss=0.03721, over 2377357.28 frames. ], batch size: 28, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:05:06,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-05-17 06:05:07,175 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290576.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 06:05:16,976 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.502e+02 2.920e+02 3.536e+02 8.522e+02, threshold=5.839e+02, percent-clipped=1.0 2023-05-17 06:05:17,250 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2661, 2.5865, 3.8425, 3.2206, 3.6343, 3.3661, 2.7306, 3.6712], device='cuda:1'), covar=tensor([0.0138, 0.0402, 0.0123, 0.0241, 0.0150, 0.0175, 0.0348, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0217, 0.0206, 0.0199, 0.0233, 0.0179, 0.0209, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:05:26,024 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0145, 3.5437, 5.4119, 3.0323, 3.0225, 3.9218, 3.4011, 3.9256], device='cuda:1'), covar=tensor([0.0359, 0.1107, 0.0258, 0.1139, 0.1922, 0.1553, 0.1314, 0.1184], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0245, 0.0263, 0.0187, 0.0243, 0.0300, 0.0230, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:05:32,986 INFO [finetune.py:992] (1/2) Epoch 16, batch 7100, loss[loss=0.1528, simple_loss=0.2367, pruned_loss=0.03443, over 12316.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2548, pruned_loss=0.03763, over 2370172.46 frames. ], batch size: 30, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:05:42,870 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290624.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 06:06:03,872 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290654.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:06:08,613 INFO [finetune.py:992] (1/2) Epoch 16, batch 7150, loss[loss=0.1513, simple_loss=0.2469, pruned_loss=0.0278, over 12028.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2547, pruned_loss=0.03779, over 2370697.35 frames. ], batch size: 31, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:06:18,120 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290674.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:06:29,243 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.710e+02 3.160e+02 3.653e+02 5.571e+02, threshold=6.320e+02, percent-clipped=0.0 2023-05-17 06:06:37,798 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290702.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:06:40,684 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290706.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:06:44,190 INFO [finetune.py:992] (1/2) Epoch 16, batch 7200, loss[loss=0.1392, simple_loss=0.2271, pruned_loss=0.02567, over 12174.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2543, pruned_loss=0.03766, over 2375583.64 frames. ], batch size: 31, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:07:15,631 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290754.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:07:17,205 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4563, 2.9947, 4.0042, 3.4059, 3.8300, 3.4442, 2.9194, 3.8589], device='cuda:1'), covar=tensor([0.0145, 0.0309, 0.0134, 0.0252, 0.0156, 0.0197, 0.0315, 0.0140], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0217, 0.0205, 0.0199, 0.0231, 0.0178, 0.0208, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:07:20,455 INFO [finetune.py:992] (1/2) Epoch 16, batch 7250, loss[loss=0.1319, simple_loss=0.2188, pruned_loss=0.0225, over 12124.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2546, pruned_loss=0.038, over 2366225.81 frames. ], batch size: 33, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:07:22,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-05-17 06:07:26,261 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6568, 3.7908, 3.2926, 3.2196, 2.9250, 2.7833, 3.7213, 2.4039], device='cuda:1'), covar=tensor([0.0385, 0.0122, 0.0236, 0.0192, 0.0437, 0.0408, 0.0142, 0.0524], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0167, 0.0172, 0.0194, 0.0205, 0.0203, 0.0178, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:07:41,806 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.496e+02 3.019e+02 3.951e+02 6.788e+02, threshold=6.038e+02, percent-clipped=2.0 2023-05-17 06:07:57,143 INFO [finetune.py:992] (1/2) Epoch 16, batch 7300, loss[loss=0.1738, simple_loss=0.2604, pruned_loss=0.04364, over 12079.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.255, pruned_loss=0.03824, over 2367295.78 frames. ], batch size: 32, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:08:03,007 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9208, 5.7547, 5.4030, 5.2759, 5.8478, 5.1012, 5.2219, 5.2255], device='cuda:1'), covar=tensor([0.1569, 0.0987, 0.1150, 0.1976, 0.1027, 0.2315, 0.2440, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0501, 0.0401, 0.0450, 0.0472, 0.0434, 0.0402, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:08:04,631 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1007, 2.4403, 3.6782, 3.0587, 3.4646, 3.1019, 2.5607, 3.4975], device='cuda:1'), covar=tensor([0.0141, 0.0427, 0.0130, 0.0255, 0.0144, 0.0237, 0.0392, 0.0138], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0217, 0.0206, 0.0200, 0.0232, 0.0179, 0.0209, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:08:32,806 INFO [finetune.py:992] (1/2) Epoch 16, batch 7350, loss[loss=0.1782, simple_loss=0.2647, pruned_loss=0.04583, over 12184.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2549, pruned_loss=0.03791, over 2373766.29 frames. ], batch size: 35, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:08:44,661 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6456, 2.9911, 4.5794, 4.7105, 2.7946, 2.7637, 3.1124, 2.1855], device='cuda:1'), covar=tensor([0.1755, 0.3043, 0.0487, 0.0453, 0.1450, 0.2518, 0.2704, 0.4130], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0398, 0.0283, 0.0307, 0.0282, 0.0322, 0.0400, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:08:52,931 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2966, 4.7290, 4.1886, 4.9736, 4.5825, 2.8668, 4.1982, 3.0155], device='cuda:1'), covar=tensor([0.0810, 0.0748, 0.1254, 0.0539, 0.1091, 0.1826, 0.1071, 0.3416], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0387, 0.0369, 0.0336, 0.0379, 0.0281, 0.0357, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:08:54,041 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.777e+02 3.176e+02 3.962e+02 8.033e+02, threshold=6.352e+02, percent-clipped=3.0 2023-05-17 06:09:01,519 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0137, 2.4203, 3.0051, 3.8909, 2.2441, 3.9557, 3.9794, 4.1085], device='cuda:1'), covar=tensor([0.0156, 0.1201, 0.0540, 0.0185, 0.1301, 0.0308, 0.0204, 0.0151], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0212, 0.0190, 0.0128, 0.0196, 0.0189, 0.0184, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:09:02,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0549, 4.5315, 3.9636, 4.8148, 4.2919, 3.0390, 4.1254, 2.9778], device='cuda:1'), covar=tensor([0.0967, 0.0769, 0.1527, 0.0397, 0.1390, 0.1660, 0.1075, 0.3376], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0387, 0.0369, 0.0336, 0.0379, 0.0281, 0.0356, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:09:09,091 INFO [finetune.py:992] (1/2) Epoch 16, batch 7400, loss[loss=0.1456, simple_loss=0.2352, pruned_loss=0.02804, over 12184.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2537, pruned_loss=0.03768, over 2378864.82 frames. ], batch size: 31, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:09:13,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-17 06:09:40,787 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2111, 4.7203, 4.2484, 4.9983, 4.5829, 3.1113, 4.3786, 3.1097], device='cuda:1'), covar=tensor([0.0954, 0.0812, 0.1494, 0.0458, 0.1101, 0.1715, 0.0963, 0.3402], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0387, 0.0369, 0.0337, 0.0380, 0.0281, 0.0357, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:09:45,437 INFO [finetune.py:992] (1/2) Epoch 16, batch 7450, loss[loss=0.2282, simple_loss=0.2958, pruned_loss=0.08034, over 7696.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2538, pruned_loss=0.03781, over 2373860.04 frames. ], batch size: 100, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:09:53,903 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4218, 4.3439, 4.2015, 4.4719, 3.1978, 4.0764, 2.8523, 4.3316], device='cuda:1'), covar=tensor([0.1397, 0.0545, 0.0847, 0.0594, 0.1076, 0.0531, 0.1595, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0270, 0.0300, 0.0362, 0.0246, 0.0246, 0.0264, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:09:54,568 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290974.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:10:05,517 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.703e+02 3.035e+02 3.836e+02 7.685e+02, threshold=6.070e+02, percent-clipped=2.0 2023-05-17 06:10:20,234 INFO [finetune.py:992] (1/2) Epoch 16, batch 7500, loss[loss=0.1639, simple_loss=0.2596, pruned_loss=0.03412, over 12296.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2544, pruned_loss=0.0383, over 2365885.31 frames. ], batch size: 34, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:10:28,227 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=291022.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:10:28,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 2023-05-17 06:10:43,352 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3433, 4.6847, 4.1445, 4.9474, 4.5496, 2.8647, 4.3509, 2.9834], device='cuda:1'), covar=tensor([0.0802, 0.0826, 0.1480, 0.0460, 0.1172, 0.1796, 0.1016, 0.3530], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0389, 0.0370, 0.0337, 0.0380, 0.0282, 0.0357, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:10:53,121 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6056, 3.2636, 5.1072, 2.5579, 2.7681, 3.7044, 3.1480, 3.7993], device='cuda:1'), covar=tensor([0.0566, 0.1328, 0.0323, 0.1320, 0.2076, 0.1625, 0.1520, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0246, 0.0264, 0.0188, 0.0244, 0.0301, 0.0232, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:10:57,128 INFO [finetune.py:992] (1/2) Epoch 16, batch 7550, loss[loss=0.1505, simple_loss=0.2484, pruned_loss=0.02626, over 12200.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2537, pruned_loss=0.03785, over 2370419.55 frames. ], batch size: 35, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:11:02,953 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=291069.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:11:17,743 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.906e+02 3.450e+02 4.180e+02 9.677e+02, threshold=6.900e+02, percent-clipped=7.0 2023-05-17 06:11:32,979 INFO [finetune.py:992] (1/2) Epoch 16, batch 7600, loss[loss=0.1621, simple_loss=0.2595, pruned_loss=0.03233, over 12151.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2533, pruned_loss=0.0375, over 2378206.59 frames. ], batch size: 34, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:11:37,226 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3675, 2.4925, 3.7347, 4.3891, 3.8681, 4.4036, 3.8394, 2.9567], device='cuda:1'), covar=tensor([0.0050, 0.0487, 0.0144, 0.0051, 0.0129, 0.0077, 0.0154, 0.0453], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0124, 0.0105, 0.0080, 0.0106, 0.0117, 0.0101, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:11:46,528 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5536, 5.3891, 5.5302, 5.5839, 5.1996, 5.2528, 4.9731, 5.4198], device='cuda:1'), covar=tensor([0.0711, 0.0556, 0.0788, 0.0560, 0.2007, 0.1150, 0.0545, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0561, 0.0719, 0.0630, 0.0650, 0.0877, 0.0770, 0.0580, 0.0500], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:11:46,620 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=291130.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:12:08,462 INFO [finetune.py:992] (1/2) Epoch 16, batch 7650, loss[loss=0.1567, simple_loss=0.245, pruned_loss=0.03417, over 12131.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2531, pruned_loss=0.03761, over 2374191.32 frames. ], batch size: 30, lr: 3.39e-03, grad_scale: 16.0 2023-05-17 06:12:30,420 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.451e+02 2.997e+02 3.576e+02 6.341e+02, threshold=5.994e+02, percent-clipped=0.0 2023-05-17 06:12:40,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-17 06:12:45,710 INFO [finetune.py:992] (1/2) Epoch 16, batch 7700, loss[loss=0.1439, simple_loss=0.2326, pruned_loss=0.02766, over 12033.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2535, pruned_loss=0.03755, over 2375696.91 frames. ], batch size: 31, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:13:21,111 INFO [finetune.py:992] (1/2) Epoch 16, batch 7750, loss[loss=0.1666, simple_loss=0.2615, pruned_loss=0.03582, over 12279.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.254, pruned_loss=0.03788, over 2385546.41 frames. ], batch size: 37, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:13:41,738 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.245e+02 2.900e+02 3.369e+02 4.087e+02 1.776e+03, threshold=6.739e+02, percent-clipped=5.0 2023-05-17 06:13:57,203 INFO [finetune.py:992] (1/2) Epoch 16, batch 7800, loss[loss=0.1705, simple_loss=0.2604, pruned_loss=0.04035, over 11675.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2549, pruned_loss=0.03848, over 2372331.81 frames. ], batch size: 48, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:14:32,959 INFO [finetune.py:992] (1/2) Epoch 16, batch 7850, loss[loss=0.164, simple_loss=0.2572, pruned_loss=0.03538, over 12298.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2547, pruned_loss=0.03838, over 2376418.24 frames. ], batch size: 34, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:14:53,383 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 2.682e+02 3.106e+02 4.145e+02 1.209e+03, threshold=6.213e+02, percent-clipped=5.0 2023-05-17 06:15:08,705 INFO [finetune.py:992] (1/2) Epoch 16, batch 7900, loss[loss=0.1217, simple_loss=0.2062, pruned_loss=0.01859, over 12251.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2541, pruned_loss=0.03819, over 2367668.13 frames. ], batch size: 28, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:15:18,933 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=291425.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:15:45,475 INFO [finetune.py:992] (1/2) Epoch 16, batch 7950, loss[loss=0.1613, simple_loss=0.2502, pruned_loss=0.03616, over 12194.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2528, pruned_loss=0.0377, over 2376089.51 frames. ], batch size: 35, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:15:50,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-17 06:16:06,547 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.592e+02 3.141e+02 3.877e+02 1.057e+03, threshold=6.283e+02, percent-clipped=3.0 2023-05-17 06:16:20,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.16 vs. limit=5.0 2023-05-17 06:16:21,534 INFO [finetune.py:992] (1/2) Epoch 16, batch 8000, loss[loss=0.1921, simple_loss=0.2798, pruned_loss=0.05221, over 12062.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2524, pruned_loss=0.03753, over 2383751.70 frames. ], batch size: 45, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:16:53,147 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3290, 3.2817, 3.0669, 3.0234, 2.7367, 2.5879, 3.3335, 2.2245], device='cuda:1'), covar=tensor([0.0497, 0.0211, 0.0253, 0.0237, 0.0431, 0.0400, 0.0166, 0.0538], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0166, 0.0171, 0.0193, 0.0203, 0.0202, 0.0177, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:16:57,161 INFO [finetune.py:992] (1/2) Epoch 16, batch 8050, loss[loss=0.1511, simple_loss=0.2386, pruned_loss=0.03177, over 12005.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2527, pruned_loss=0.03758, over 2378833.75 frames. ], batch size: 28, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:17:17,797 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.054e+02 2.728e+02 3.105e+02 3.762e+02 7.004e+02, threshold=6.210e+02, percent-clipped=2.0 2023-05-17 06:17:26,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-05-17 06:17:33,165 INFO [finetune.py:992] (1/2) Epoch 16, batch 8100, loss[loss=0.1613, simple_loss=0.2456, pruned_loss=0.03846, over 12360.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2528, pruned_loss=0.03705, over 2386419.79 frames. ], batch size: 31, lr: 3.38e-03, grad_scale: 16.0 2023-05-17 06:18:09,655 INFO [finetune.py:992] (1/2) Epoch 16, batch 8150, loss[loss=0.1839, simple_loss=0.251, pruned_loss=0.05843, over 12004.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2534, pruned_loss=0.03721, over 2388763.01 frames. ], batch size: 28, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:18:30,832 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.696e+02 3.139e+02 3.881e+02 6.152e+02, threshold=6.277e+02, percent-clipped=0.0 2023-05-17 06:18:32,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-05-17 06:18:35,357 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1971, 3.5678, 3.5464, 3.9389, 2.8879, 3.4712, 2.4743, 3.4156], device='cuda:1'), covar=tensor([0.1686, 0.0858, 0.0992, 0.0636, 0.1151, 0.0736, 0.1883, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0272, 0.0304, 0.0365, 0.0247, 0.0247, 0.0266, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:18:45,006 INFO [finetune.py:992] (1/2) Epoch 16, batch 8200, loss[loss=0.1483, simple_loss=0.2408, pruned_loss=0.02786, over 12162.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2554, pruned_loss=0.03882, over 2362139.59 frames. ], batch size: 36, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:18:55,074 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=291725.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:19:20,307 INFO [finetune.py:992] (1/2) Epoch 16, batch 8250, loss[loss=0.1641, simple_loss=0.2613, pruned_loss=0.03352, over 12239.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2557, pruned_loss=0.03862, over 2367084.27 frames. ], batch size: 32, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:19:29,619 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=291773.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:19:31,863 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=291776.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:19:43,003 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.597e+02 2.979e+02 3.573e+02 8.446e+02, threshold=5.958e+02, percent-clipped=2.0 2023-05-17 06:19:48,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-17 06:19:54,296 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3516, 4.3866, 4.0966, 4.4867, 3.1260, 4.1384, 2.5126, 4.2668], device='cuda:1'), covar=tensor([0.1506, 0.0564, 0.0907, 0.0677, 0.1166, 0.0525, 0.1921, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0271, 0.0303, 0.0365, 0.0247, 0.0247, 0.0266, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:19:57,169 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9814, 2.3458, 2.2214, 2.2711, 2.0020, 2.0005, 2.1535, 1.7968], device='cuda:1'), covar=tensor([0.0335, 0.0198, 0.0237, 0.0204, 0.0332, 0.0260, 0.0215, 0.0375], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0167, 0.0173, 0.0195, 0.0206, 0.0203, 0.0178, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:19:57,575 INFO [finetune.py:992] (1/2) Epoch 16, batch 8300, loss[loss=0.2782, simple_loss=0.3323, pruned_loss=0.112, over 7791.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2561, pruned_loss=0.03858, over 2367025.28 frames. ], batch size: 98, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:20:10,469 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5418, 2.8735, 3.1846, 4.3078, 2.2666, 4.3101, 4.5442, 4.5223], device='cuda:1'), covar=tensor([0.0140, 0.1071, 0.0534, 0.0200, 0.1365, 0.0279, 0.0132, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0209, 0.0187, 0.0127, 0.0193, 0.0186, 0.0182, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:20:16,129 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=291837.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:20:33,135 INFO [finetune.py:992] (1/2) Epoch 16, batch 8350, loss[loss=0.1746, simple_loss=0.2641, pruned_loss=0.04258, over 11650.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2563, pruned_loss=0.03832, over 2375898.14 frames. ], batch size: 48, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:20:54,478 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.600e+02 3.057e+02 3.626e+02 1.055e+03, threshold=6.113e+02, percent-clipped=3.0 2023-05-17 06:21:08,758 INFO [finetune.py:992] (1/2) Epoch 16, batch 8400, loss[loss=0.1759, simple_loss=0.2657, pruned_loss=0.04305, over 12357.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2565, pruned_loss=0.03826, over 2372236.23 frames. ], batch size: 35, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:21:29,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2023-05-17 06:21:43,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-17 06:21:45,801 INFO [finetune.py:992] (1/2) Epoch 16, batch 8450, loss[loss=0.154, simple_loss=0.2476, pruned_loss=0.03019, over 12103.00 frames. ], tot_loss[loss=0.165, simple_loss=0.255, pruned_loss=0.03747, over 2384383.24 frames. ], batch size: 33, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:21:55,848 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0983, 5.9916, 5.6500, 5.5050, 6.0722, 5.4518, 5.6076, 5.5570], device='cuda:1'), covar=tensor([0.1399, 0.0933, 0.1082, 0.1981, 0.0902, 0.2079, 0.1719, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0516, 0.0410, 0.0466, 0.0485, 0.0445, 0.0412, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:22:06,990 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 2.756e+02 3.165e+02 3.955e+02 9.030e+02, threshold=6.330e+02, percent-clipped=2.0 2023-05-17 06:22:23,435 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1699, 4.6164, 3.9792, 4.7699, 4.3927, 2.8904, 4.1277, 2.9684], device='cuda:1'), covar=tensor([0.0889, 0.0767, 0.1540, 0.0542, 0.1218, 0.1816, 0.1196, 0.3618], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0383, 0.0364, 0.0332, 0.0374, 0.0277, 0.0351, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:22:24,605 INFO [finetune.py:992] (1/2) Epoch 16, batch 8500, loss[loss=0.1375, simple_loss=0.2265, pruned_loss=0.02421, over 12355.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2536, pruned_loss=0.03703, over 2388698.51 frames. ], batch size: 31, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:22:48,381 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2960, 3.9954, 4.1084, 4.5111, 3.1029, 3.9608, 2.4920, 4.1846], device='cuda:1'), covar=tensor([0.1671, 0.0827, 0.0950, 0.0577, 0.1263, 0.0638, 0.2061, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0273, 0.0304, 0.0366, 0.0248, 0.0248, 0.0267, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:22:56,135 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=292054.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:23:01,038 INFO [finetune.py:992] (1/2) Epoch 16, batch 8550, loss[loss=0.1513, simple_loss=0.2497, pruned_loss=0.02642, over 12355.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2538, pruned_loss=0.03754, over 2381886.38 frames. ], batch size: 36, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:23:22,691 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.666e+02 3.356e+02 3.786e+02 5.808e+02, threshold=6.713e+02, percent-clipped=0.0 2023-05-17 06:23:36,956 INFO [finetune.py:992] (1/2) Epoch 16, batch 8600, loss[loss=0.1791, simple_loss=0.26, pruned_loss=0.04908, over 11737.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2542, pruned_loss=0.03803, over 2369030.63 frames. ], batch size: 48, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:23:39,953 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=292115.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:23:46,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-17 06:23:52,031 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292132.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:24:12,745 INFO [finetune.py:992] (1/2) Epoch 16, batch 8650, loss[loss=0.1538, simple_loss=0.2462, pruned_loss=0.03071, over 12180.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2544, pruned_loss=0.03814, over 2366881.72 frames. ], batch size: 29, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:24:34,124 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.729e+02 3.150e+02 3.727e+02 5.480e+02, threshold=6.300e+02, percent-clipped=0.0 2023-05-17 06:24:42,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 06:24:49,466 INFO [finetune.py:992] (1/2) Epoch 16, batch 8700, loss[loss=0.1724, simple_loss=0.2634, pruned_loss=0.04073, over 12296.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2543, pruned_loss=0.0379, over 2369654.66 frames. ], batch size: 34, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:25:10,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-17 06:25:25,802 INFO [finetune.py:992] (1/2) Epoch 16, batch 8750, loss[loss=0.1595, simple_loss=0.2472, pruned_loss=0.03592, over 12029.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2537, pruned_loss=0.03759, over 2370070.99 frames. ], batch size: 31, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:25:29,127 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.40 vs. limit=5.0 2023-05-17 06:25:45,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 06:25:47,146 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.707e+02 3.123e+02 3.591e+02 5.330e+02, threshold=6.245e+02, percent-clipped=0.0 2023-05-17 06:26:01,307 INFO [finetune.py:992] (1/2) Epoch 16, batch 8800, loss[loss=0.1342, simple_loss=0.2207, pruned_loss=0.02379, over 12128.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2538, pruned_loss=0.03755, over 2377197.41 frames. ], batch size: 30, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:26:14,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.42 vs. limit=5.0 2023-05-17 06:26:19,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-17 06:26:37,628 INFO [finetune.py:992] (1/2) Epoch 16, batch 8850, loss[loss=0.2261, simple_loss=0.2995, pruned_loss=0.07639, over 8146.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2541, pruned_loss=0.03795, over 2366841.22 frames. ], batch size: 99, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:26:59,616 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.980e+02 2.756e+02 3.464e+02 4.372e+02 8.358e+02, threshold=6.928e+02, percent-clipped=4.0 2023-05-17 06:27:13,635 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292410.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:27:14,276 INFO [finetune.py:992] (1/2) Epoch 16, batch 8900, loss[loss=0.1757, simple_loss=0.2675, pruned_loss=0.04195, over 12038.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2544, pruned_loss=0.03818, over 2360273.44 frames. ], batch size: 40, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:27:29,252 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=292432.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:27:49,598 INFO [finetune.py:992] (1/2) Epoch 16, batch 8950, loss[loss=0.2622, simple_loss=0.3205, pruned_loss=0.1019, over 8277.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2553, pruned_loss=0.03873, over 2354395.72 frames. ], batch size: 99, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:28:03,312 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=292480.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:28:03,773 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-17 06:28:11,807 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.832e+02 3.435e+02 4.189e+02 1.010e+03, threshold=6.871e+02, percent-clipped=2.0 2023-05-17 06:28:14,660 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0754, 4.9519, 4.8349, 4.9264, 4.6101, 5.0709, 5.0212, 5.1745], device='cuda:1'), covar=tensor([0.0227, 0.0168, 0.0242, 0.0407, 0.0816, 0.0426, 0.0180, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0208, 0.0206, 0.0261, 0.0253, 0.0234, 0.0187, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 06:28:26,612 INFO [finetune.py:992] (1/2) Epoch 16, batch 9000, loss[loss=0.1757, simple_loss=0.271, pruned_loss=0.04024, over 12348.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2557, pruned_loss=0.03901, over 2348934.77 frames. ], batch size: 36, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:28:26,612 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 06:28:44,910 INFO [finetune.py:1026] (1/2) Epoch 16, validation: loss=0.3156, simple_loss=0.3914, pruned_loss=0.1199, over 1020973.00 frames. 2023-05-17 06:28:44,911 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 06:28:53,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-17 06:29:14,462 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-17 06:29:20,400 INFO [finetune.py:992] (1/2) Epoch 16, batch 9050, loss[loss=0.217, simple_loss=0.3008, pruned_loss=0.06658, over 10764.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2562, pruned_loss=0.03907, over 2352545.48 frames. ], batch size: 68, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:29:22,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.50 vs. limit=5.0 2023-05-17 06:29:22,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-17 06:29:34,007 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0608, 4.0271, 4.1111, 4.3618, 2.9226, 3.9150, 2.7627, 4.0809], device='cuda:1'), covar=tensor([0.1644, 0.0683, 0.0848, 0.0507, 0.1226, 0.0591, 0.1665, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0270, 0.0303, 0.0363, 0.0245, 0.0247, 0.0265, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:29:42,890 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.660e+02 3.096e+02 3.918e+02 9.243e+02, threshold=6.193e+02, percent-clipped=1.0 2023-05-17 06:29:57,492 INFO [finetune.py:992] (1/2) Epoch 16, batch 9100, loss[loss=0.1459, simple_loss=0.2368, pruned_loss=0.02755, over 12251.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2559, pruned_loss=0.03905, over 2346729.45 frames. ], batch size: 32, lr: 3.38e-03, grad_scale: 8.0 2023-05-17 06:30:33,341 INFO [finetune.py:992] (1/2) Epoch 16, batch 9150, loss[loss=0.1512, simple_loss=0.2422, pruned_loss=0.03008, over 12113.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2554, pruned_loss=0.03905, over 2351303.99 frames. ], batch size: 33, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:30:54,423 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.517e+02 2.822e+02 3.782e+02 1.055e+03, threshold=5.645e+02, percent-clipped=1.0 2023-05-17 06:30:55,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-17 06:31:08,261 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=292710.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:31:08,885 INFO [finetune.py:992] (1/2) Epoch 16, batch 9200, loss[loss=0.1357, simple_loss=0.2216, pruned_loss=0.02492, over 12028.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2547, pruned_loss=0.03857, over 2349265.46 frames. ], batch size: 28, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:31:37,471 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5015, 3.1009, 5.0471, 2.5391, 2.7109, 3.6653, 3.0515, 3.6749], device='cuda:1'), covar=tensor([0.0531, 0.1410, 0.0306, 0.1273, 0.2067, 0.1567, 0.1512, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0242, 0.0260, 0.0186, 0.0241, 0.0296, 0.0229, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:31:43,740 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=292758.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:31:45,807 INFO [finetune.py:992] (1/2) Epoch 16, batch 9250, loss[loss=0.1819, simple_loss=0.2788, pruned_loss=0.04248, over 12163.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2556, pruned_loss=0.03878, over 2356584.31 frames. ], batch size: 36, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:31:52,397 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3455, 3.5488, 3.2255, 3.0723, 2.8086, 2.6404, 3.5357, 2.2583], device='cuda:1'), covar=tensor([0.0455, 0.0146, 0.0235, 0.0220, 0.0442, 0.0412, 0.0168, 0.0560], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0169, 0.0174, 0.0198, 0.0208, 0.0207, 0.0182, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:32:07,222 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.575e+02 3.009e+02 3.644e+02 6.909e+02, threshold=6.018e+02, percent-clipped=2.0 2023-05-17 06:32:17,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-05-17 06:32:21,777 INFO [finetune.py:992] (1/2) Epoch 16, batch 9300, loss[loss=0.158, simple_loss=0.2418, pruned_loss=0.03712, over 12121.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2538, pruned_loss=0.0382, over 2364799.25 frames. ], batch size: 30, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:32:25,958 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5643, 3.7030, 3.3386, 3.1419, 2.9780, 2.8050, 3.7478, 2.4698], device='cuda:1'), covar=tensor([0.0450, 0.0154, 0.0227, 0.0244, 0.0395, 0.0382, 0.0134, 0.0500], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0169, 0.0175, 0.0199, 0.0209, 0.0207, 0.0182, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:32:57,233 INFO [finetune.py:992] (1/2) Epoch 16, batch 9350, loss[loss=0.1805, simple_loss=0.2714, pruned_loss=0.04483, over 10536.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2537, pruned_loss=0.03816, over 2372106.39 frames. ], batch size: 68, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:33:19,945 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.707e+02 2.513e+02 3.048e+02 3.850e+02 7.180e+02, threshold=6.096e+02, percent-clipped=3.0 2023-05-17 06:33:28,120 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7359, 3.0535, 4.5165, 4.7488, 3.0700, 2.7762, 3.0061, 2.3017], device='cuda:1'), covar=tensor([0.1598, 0.2772, 0.0485, 0.0475, 0.1249, 0.2416, 0.2782, 0.3992], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0394, 0.0279, 0.0305, 0.0280, 0.0321, 0.0399, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:33:34,197 INFO [finetune.py:992] (1/2) Epoch 16, batch 9400, loss[loss=0.1401, simple_loss=0.2342, pruned_loss=0.02303, over 12009.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2534, pruned_loss=0.03764, over 2376152.52 frames. ], batch size: 31, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:33:56,459 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7457, 2.8520, 4.5879, 4.6664, 2.7539, 2.6631, 2.9652, 2.2270], device='cuda:1'), covar=tensor([0.1648, 0.2957, 0.0472, 0.0506, 0.1452, 0.2422, 0.2895, 0.4072], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0394, 0.0280, 0.0305, 0.0280, 0.0321, 0.0400, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:34:09,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 06:34:09,544 INFO [finetune.py:992] (1/2) Epoch 16, batch 9450, loss[loss=0.1524, simple_loss=0.2387, pruned_loss=0.03305, over 12411.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2523, pruned_loss=0.03737, over 2382303.35 frames. ], batch size: 32, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:34:14,102 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=292967.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:34:30,773 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 2.666e+02 3.237e+02 3.941e+02 7.125e+02, threshold=6.474e+02, percent-clipped=3.0 2023-05-17 06:34:45,419 INFO [finetune.py:992] (1/2) Epoch 16, batch 9500, loss[loss=0.1835, simple_loss=0.2702, pruned_loss=0.04846, over 11159.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2529, pruned_loss=0.0375, over 2373445.64 frames. ], batch size: 55, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:34:58,760 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293028.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:35:22,148 INFO [finetune.py:992] (1/2) Epoch 16, batch 9550, loss[loss=0.163, simple_loss=0.2608, pruned_loss=0.03261, over 11808.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2525, pruned_loss=0.03724, over 2371812.41 frames. ], batch size: 44, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:35:36,637 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293081.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:35:43,590 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.048e+02 2.670e+02 3.195e+02 3.630e+02 6.304e+02, threshold=6.391e+02, percent-clipped=0.0 2023-05-17 06:35:56,036 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4484, 2.6350, 2.9511, 4.3098, 2.1873, 4.3227, 4.4717, 4.3843], device='cuda:1'), covar=tensor([0.0153, 0.1193, 0.0616, 0.0169, 0.1432, 0.0250, 0.0178, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0209, 0.0188, 0.0126, 0.0194, 0.0186, 0.0183, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:35:57,893 INFO [finetune.py:992] (1/2) Epoch 16, batch 9600, loss[loss=0.1774, simple_loss=0.2719, pruned_loss=0.04142, over 12066.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2532, pruned_loss=0.03729, over 2367562.77 frames. ], batch size: 40, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:36:02,316 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293117.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:36:04,562 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9416, 3.5652, 5.3575, 2.7700, 2.9598, 3.8511, 3.2777, 3.8871], device='cuda:1'), covar=tensor([0.0433, 0.1103, 0.0328, 0.1190, 0.1924, 0.1608, 0.1425, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0243, 0.0260, 0.0187, 0.0240, 0.0297, 0.0230, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:36:20,480 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293142.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:36:27,788 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6295, 4.1484, 4.2635, 4.4405, 4.2810, 4.5080, 4.3941, 2.5390], device='cuda:1'), covar=tensor([0.0107, 0.0097, 0.0124, 0.0074, 0.0071, 0.0118, 0.0095, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0083, 0.0085, 0.0076, 0.0062, 0.0096, 0.0085, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:36:33,921 INFO [finetune.py:992] (1/2) Epoch 16, batch 9650, loss[loss=0.1521, simple_loss=0.2445, pruned_loss=0.02984, over 12363.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2523, pruned_loss=0.03708, over 2372466.49 frames. ], batch size: 36, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:36:47,555 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293178.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:36:57,420 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.792e+02 3.150e+02 3.953e+02 6.811e+02, threshold=6.300e+02, percent-clipped=2.0 2023-05-17 06:37:07,612 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.1966, 6.0819, 5.9273, 5.4101, 5.3489, 6.0978, 5.6927, 5.4729], device='cuda:1'), covar=tensor([0.0673, 0.1067, 0.0700, 0.1605, 0.0597, 0.0706, 0.1460, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0649, 0.0586, 0.0541, 0.0649, 0.0431, 0.0755, 0.0800, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 06:37:09,047 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8422, 5.7186, 5.2313, 5.3247, 5.7886, 5.1345, 5.2312, 5.3302], device='cuda:1'), covar=tensor([0.1570, 0.0952, 0.1085, 0.1905, 0.0924, 0.2269, 0.2034, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0512, 0.0408, 0.0464, 0.0482, 0.0444, 0.0412, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:37:11,096 INFO [finetune.py:992] (1/2) Epoch 16, batch 9700, loss[loss=0.2158, simple_loss=0.2883, pruned_loss=0.07166, over 8015.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2524, pruned_loss=0.03724, over 2372183.11 frames. ], batch size: 98, lr: 3.37e-03, grad_scale: 4.0 2023-05-17 06:37:25,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.04 vs. limit=5.0 2023-05-17 06:37:46,889 INFO [finetune.py:992] (1/2) Epoch 16, batch 9750, loss[loss=0.1873, simple_loss=0.2818, pruned_loss=0.04642, over 12131.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2525, pruned_loss=0.03724, over 2370271.74 frames. ], batch size: 38, lr: 3.37e-03, grad_scale: 4.0 2023-05-17 06:37:48,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 06:37:51,164 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4452, 4.9895, 5.4124, 4.7403, 5.0720, 4.8587, 5.4480, 5.1238], device='cuda:1'), covar=tensor([0.0219, 0.0367, 0.0246, 0.0262, 0.0388, 0.0312, 0.0181, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0282, 0.0305, 0.0277, 0.0277, 0.0280, 0.0251, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:38:08,865 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.520e+02 2.905e+02 3.730e+02 7.475e+02, threshold=5.811e+02, percent-clipped=2.0 2023-05-17 06:38:18,978 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0197, 5.8199, 5.3977, 5.3858, 5.9710, 5.3193, 5.3293, 5.3966], device='cuda:1'), covar=tensor([0.1716, 0.1077, 0.1038, 0.2111, 0.0936, 0.2244, 0.2116, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0511, 0.0407, 0.0464, 0.0482, 0.0444, 0.0411, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:38:22,489 INFO [finetune.py:992] (1/2) Epoch 16, batch 9800, loss[loss=0.167, simple_loss=0.2602, pruned_loss=0.03692, over 12156.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.253, pruned_loss=0.03726, over 2372604.88 frames. ], batch size: 34, lr: 3.37e-03, grad_scale: 4.0 2023-05-17 06:38:31,933 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293323.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:38:39,282 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 06:38:58,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 06:38:59,105 INFO [finetune.py:992] (1/2) Epoch 16, batch 9850, loss[loss=0.1868, simple_loss=0.2787, pruned_loss=0.04743, over 10469.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2537, pruned_loss=0.03735, over 2371398.72 frames. ], batch size: 70, lr: 3.37e-03, grad_scale: 4.0 2023-05-17 06:38:59,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293361.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:39:01,337 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293364.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:39:05,656 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4676, 3.5432, 3.1916, 3.0644, 2.8179, 2.6174, 3.5619, 2.3670], device='cuda:1'), covar=tensor([0.0466, 0.0173, 0.0222, 0.0263, 0.0453, 0.0452, 0.0154, 0.0533], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0171, 0.0175, 0.0199, 0.0209, 0.0207, 0.0183, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:39:21,029 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.575e+02 3.091e+02 3.647e+02 7.324e+02, threshold=6.182e+02, percent-clipped=5.0 2023-05-17 06:39:34,879 INFO [finetune.py:992] (1/2) Epoch 16, batch 9900, loss[loss=0.1676, simple_loss=0.2518, pruned_loss=0.04176, over 12361.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2541, pruned_loss=0.03739, over 2375218.03 frames. ], batch size: 30, lr: 3.37e-03, grad_scale: 4.0 2023-05-17 06:39:43,072 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293422.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:39:45,200 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293425.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:39:53,686 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293437.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:40:10,789 INFO [finetune.py:992] (1/2) Epoch 16, batch 9950, loss[loss=0.1582, simple_loss=0.2503, pruned_loss=0.03301, over 12156.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2536, pruned_loss=0.03704, over 2370256.01 frames. ], batch size: 34, lr: 3.37e-03, grad_scale: 4.0 2023-05-17 06:40:20,695 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293473.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:40:28,705 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8210, 5.5431, 5.1255, 5.1194, 5.6366, 4.9566, 5.0136, 5.0920], device='cuda:1'), covar=tensor([0.1446, 0.1005, 0.1280, 0.1852, 0.0904, 0.2005, 0.2049, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0507, 0.0407, 0.0461, 0.0479, 0.0442, 0.0409, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:40:34,296 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.574e+02 3.007e+02 3.575e+02 1.265e+03, threshold=6.015e+02, percent-clipped=3.0 2023-05-17 06:40:36,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-05-17 06:40:43,078 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9902, 4.4483, 3.8385, 4.7031, 4.2116, 2.7003, 3.9678, 2.9220], device='cuda:1'), covar=tensor([0.0879, 0.0810, 0.1428, 0.0568, 0.1122, 0.1828, 0.1185, 0.3382], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0385, 0.0366, 0.0335, 0.0376, 0.0278, 0.0354, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:40:47,690 INFO [finetune.py:992] (1/2) Epoch 16, batch 10000, loss[loss=0.1467, simple_loss=0.2394, pruned_loss=0.02699, over 12343.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2545, pruned_loss=0.03719, over 2374668.66 frames. ], batch size: 35, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:41:23,664 INFO [finetune.py:992] (1/2) Epoch 16, batch 10050, loss[loss=0.1518, simple_loss=0.2323, pruned_loss=0.03561, over 12035.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2532, pruned_loss=0.03651, over 2382880.40 frames. ], batch size: 28, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:41:45,648 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.608e+02 3.078e+02 3.728e+02 5.894e+02, threshold=6.156e+02, percent-clipped=0.0 2023-05-17 06:41:51,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2023-05-17 06:41:59,642 INFO [finetune.py:992] (1/2) Epoch 16, batch 10100, loss[loss=0.1831, simple_loss=0.2748, pruned_loss=0.04571, over 12126.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.254, pruned_loss=0.03682, over 2383085.17 frames. ], batch size: 39, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:42:03,534 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6753, 2.7573, 4.2348, 4.3610, 2.7742, 2.5653, 2.8175, 2.1991], device='cuda:1'), covar=tensor([0.1677, 0.2854, 0.0529, 0.0532, 0.1380, 0.2488, 0.2699, 0.3911], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0395, 0.0280, 0.0306, 0.0281, 0.0322, 0.0400, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:42:09,563 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293623.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:42:36,650 INFO [finetune.py:992] (1/2) Epoch 16, batch 10150, loss[loss=0.1591, simple_loss=0.2521, pruned_loss=0.03308, over 12282.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2545, pruned_loss=0.03731, over 2365615.88 frames. ], batch size: 37, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:42:43,782 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293671.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:42:58,809 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.069e+02 2.676e+02 3.125e+02 3.829e+02 1.121e+03, threshold=6.250e+02, percent-clipped=3.0 2023-05-17 06:43:00,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-17 06:43:12,371 INFO [finetune.py:992] (1/2) Epoch 16, batch 10200, loss[loss=0.1656, simple_loss=0.261, pruned_loss=0.03505, over 12206.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2539, pruned_loss=0.03687, over 2370841.81 frames. ], batch size: 35, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:43:16,645 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293717.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:43:18,821 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293720.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:43:20,312 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1140, 4.9713, 4.9366, 4.9870, 4.6294, 5.1868, 5.1446, 5.2677], device='cuda:1'), covar=tensor([0.0180, 0.0147, 0.0197, 0.0289, 0.0710, 0.0274, 0.0136, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0206, 0.0202, 0.0257, 0.0252, 0.0231, 0.0186, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 06:43:31,123 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293737.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:43:49,496 INFO [finetune.py:992] (1/2) Epoch 16, batch 10250, loss[loss=0.1791, simple_loss=0.2657, pruned_loss=0.04622, over 12149.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2543, pruned_loss=0.03713, over 2371662.36 frames. ], batch size: 39, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:43:58,308 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293773.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:44:06,590 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293785.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:44:11,364 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 2.592e+02 2.928e+02 3.521e+02 7.905e+02, threshold=5.855e+02, percent-clipped=2.0 2023-05-17 06:44:15,819 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2723, 2.7888, 3.8370, 3.2312, 3.7405, 3.3266, 2.8279, 3.7722], device='cuda:1'), covar=tensor([0.0146, 0.0339, 0.0182, 0.0259, 0.0133, 0.0212, 0.0349, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0211, 0.0200, 0.0196, 0.0227, 0.0174, 0.0204, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:44:24,807 INFO [finetune.py:992] (1/2) Epoch 16, batch 10300, loss[loss=0.1673, simple_loss=0.2639, pruned_loss=0.03538, over 12347.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2544, pruned_loss=0.03758, over 2375878.20 frames. ], batch size: 36, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:44:31,787 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293821.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:45:00,322 INFO [finetune.py:992] (1/2) Epoch 16, batch 10350, loss[loss=0.1473, simple_loss=0.2346, pruned_loss=0.03007, over 11999.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2545, pruned_loss=0.03722, over 2384962.91 frames. ], batch size: 28, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:45:03,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-17 06:45:22,194 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.689e+02 3.056e+02 3.910e+02 6.982e+02, threshold=6.113e+02, percent-clipped=5.0 2023-05-17 06:45:28,832 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4057, 2.8877, 3.9672, 3.4110, 3.7955, 3.5161, 2.9491, 3.8796], device='cuda:1'), covar=tensor([0.0133, 0.0354, 0.0148, 0.0245, 0.0180, 0.0180, 0.0321, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0213, 0.0201, 0.0196, 0.0229, 0.0175, 0.0205, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:45:36,312 INFO [finetune.py:992] (1/2) Epoch 16, batch 10400, loss[loss=0.1494, simple_loss=0.2299, pruned_loss=0.03445, over 12337.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2549, pruned_loss=0.03735, over 2387458.34 frames. ], batch size: 30, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:45:37,944 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1923, 3.9194, 4.0517, 4.4159, 2.9374, 3.7902, 2.5925, 4.0450], device='cuda:1'), covar=tensor([0.1651, 0.0850, 0.0874, 0.0691, 0.1302, 0.0707, 0.1836, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0270, 0.0300, 0.0361, 0.0244, 0.0245, 0.0263, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:45:43,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-05-17 06:46:12,624 INFO [finetune.py:992] (1/2) Epoch 16, batch 10450, loss[loss=0.1493, simple_loss=0.2286, pruned_loss=0.03497, over 11995.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2546, pruned_loss=0.03715, over 2391367.43 frames. ], batch size: 28, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:46:26,278 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 2023-05-17 06:46:27,689 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7001, 2.8996, 4.4499, 4.5823, 2.9329, 2.5584, 2.9081, 2.2203], device='cuda:1'), covar=tensor([0.1659, 0.2755, 0.0472, 0.0468, 0.1267, 0.2674, 0.2825, 0.4038], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0398, 0.0283, 0.0310, 0.0283, 0.0325, 0.0404, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:46:31,120 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293987.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:46:34,542 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.724e+02 3.226e+02 3.643e+02 5.561e+02, threshold=6.451e+02, percent-clipped=0.0 2023-05-17 06:46:51,160 INFO [finetune.py:992] (1/2) Epoch 16, batch 10500, loss[loss=0.1352, simple_loss=0.2195, pruned_loss=0.02548, over 12179.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2542, pruned_loss=0.03709, over 2387109.87 frames. ], batch size: 29, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:46:55,625 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294017.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:46:57,865 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294020.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:47:18,441 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294048.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:47:27,948 INFO [finetune.py:992] (1/2) Epoch 16, batch 10550, loss[loss=0.1696, simple_loss=0.2641, pruned_loss=0.03761, over 12295.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2547, pruned_loss=0.03711, over 2384338.65 frames. ], batch size: 34, lr: 3.37e-03, grad_scale: 8.0 2023-05-17 06:47:30,800 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294065.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:47:32,955 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294068.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:47:50,101 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.579e+02 2.988e+02 3.523e+02 1.065e+03, threshold=5.977e+02, percent-clipped=6.0 2023-05-17 06:48:03,339 INFO [finetune.py:992] (1/2) Epoch 16, batch 10600, loss[loss=0.1685, simple_loss=0.2653, pruned_loss=0.03581, over 12115.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2549, pruned_loss=0.03724, over 2378408.85 frames. ], batch size: 33, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:48:38,728 INFO [finetune.py:992] (1/2) Epoch 16, batch 10650, loss[loss=0.1682, simple_loss=0.2492, pruned_loss=0.04363, over 11781.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2541, pruned_loss=0.03714, over 2386198.19 frames. ], batch size: 26, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:48:45,583 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294170.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:49:01,146 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.814e+02 2.732e+02 2.981e+02 3.538e+02 9.530e+02, threshold=5.962e+02, percent-clipped=5.0 2023-05-17 06:49:16,478 INFO [finetune.py:992] (1/2) Epoch 16, batch 10700, loss[loss=0.1547, simple_loss=0.2489, pruned_loss=0.03022, over 11566.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2539, pruned_loss=0.03675, over 2385821.16 frames. ], batch size: 48, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:49:18,048 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8246, 4.4554, 4.4566, 4.6692, 4.5997, 4.7257, 4.6502, 2.2139], device='cuda:1'), covar=tensor([0.0101, 0.0078, 0.0119, 0.0070, 0.0056, 0.0098, 0.0099, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0083, 0.0085, 0.0076, 0.0062, 0.0096, 0.0085, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:49:30,590 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294231.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:49:51,528 INFO [finetune.py:992] (1/2) Epoch 16, batch 10750, loss[loss=0.1401, simple_loss=0.2167, pruned_loss=0.03176, over 12015.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2541, pruned_loss=0.03721, over 2384997.56 frames. ], batch size: 28, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:50:13,689 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.614e+02 3.041e+02 3.851e+02 7.162e+02, threshold=6.082e+02, percent-clipped=2.0 2023-05-17 06:50:27,245 INFO [finetune.py:992] (1/2) Epoch 16, batch 10800, loss[loss=0.1854, simple_loss=0.288, pruned_loss=0.04136, over 11279.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2539, pruned_loss=0.03735, over 2380310.44 frames. ], batch size: 55, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:50:47,879 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294339.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:50:50,729 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294343.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:51:01,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 06:51:04,181 INFO [finetune.py:992] (1/2) Epoch 16, batch 10850, loss[loss=0.1575, simple_loss=0.2372, pruned_loss=0.03894, over 12131.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2529, pruned_loss=0.03678, over 2383395.19 frames. ], batch size: 30, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:51:27,117 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.525e+02 3.170e+02 3.668e+02 7.601e+02, threshold=6.341e+02, percent-clipped=1.0 2023-05-17 06:51:31,115 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4072, 4.8278, 3.2283, 2.8355, 4.1488, 2.7430, 4.1468, 3.3928], device='cuda:1'), covar=tensor([0.0766, 0.0584, 0.1082, 0.1516, 0.0300, 0.1375, 0.0470, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0267, 0.0180, 0.0205, 0.0147, 0.0186, 0.0204, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:51:33,569 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294400.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:51:41,259 INFO [finetune.py:992] (1/2) Epoch 16, batch 10900, loss[loss=0.1469, simple_loss=0.2354, pruned_loss=0.02918, over 12252.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2544, pruned_loss=0.03758, over 2368243.75 frames. ], batch size: 32, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:51:49,819 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294423.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:52:16,365 INFO [finetune.py:992] (1/2) Epoch 16, batch 10950, loss[loss=0.1466, simple_loss=0.2383, pruned_loss=0.0275, over 12336.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2551, pruned_loss=0.03777, over 2369213.08 frames. ], batch size: 31, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:52:33,529 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294484.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:52:37,023 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294488.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:52:39,755 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.050e+02 2.775e+02 3.300e+02 3.931e+02 7.532e+02, threshold=6.601e+02, percent-clipped=2.0 2023-05-17 06:52:53,138 INFO [finetune.py:992] (1/2) Epoch 16, batch 11000, loss[loss=0.1456, simple_loss=0.2305, pruned_loss=0.03034, over 12124.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2566, pruned_loss=0.03851, over 2346973.10 frames. ], batch size: 30, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:53:03,641 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294526.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:53:20,210 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294549.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:53:20,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1486, 6.0226, 5.6205, 5.5805, 6.0657, 5.3864, 5.5713, 5.6033], device='cuda:1'), covar=tensor([0.1495, 0.0874, 0.1006, 0.1674, 0.0810, 0.2227, 0.1689, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0504, 0.0403, 0.0459, 0.0478, 0.0442, 0.0408, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:53:28,481 INFO [finetune.py:992] (1/2) Epoch 16, batch 11050, loss[loss=0.1781, simple_loss=0.2784, pruned_loss=0.03891, over 11613.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2596, pruned_loss=0.03996, over 2321785.90 frames. ], batch size: 48, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:53:35,511 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294571.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:53:51,083 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4893, 3.1557, 4.9681, 2.6973, 2.6410, 3.9178, 3.1336, 3.8247], device='cuda:1'), covar=tensor([0.0543, 0.1345, 0.0320, 0.1142, 0.2023, 0.1315, 0.1457, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0243, 0.0260, 0.0186, 0.0241, 0.0299, 0.0227, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:53:51,505 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 2.846e+02 3.526e+02 4.373e+02 7.159e+02, threshold=7.053e+02, percent-clipped=1.0 2023-05-17 06:54:04,832 INFO [finetune.py:992] (1/2) Epoch 16, batch 11100, loss[loss=0.2367, simple_loss=0.3088, pruned_loss=0.08227, over 8046.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2643, pruned_loss=0.04269, over 2270407.61 frames. ], batch size: 98, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:54:19,791 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294632.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:54:20,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 06:54:27,491 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294643.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:54:39,434 INFO [finetune.py:992] (1/2) Epoch 16, batch 11150, loss[loss=0.2519, simple_loss=0.334, pruned_loss=0.08496, over 10554.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2713, pruned_loss=0.04721, over 2206134.37 frames. ], batch size: 69, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:54:48,734 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3402, 3.2058, 3.2113, 3.4374, 2.6849, 3.1869, 2.7277, 3.0040], device='cuda:1'), covar=tensor([0.1447, 0.0796, 0.0779, 0.0452, 0.0882, 0.0728, 0.1379, 0.0601], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0270, 0.0300, 0.0363, 0.0244, 0.0245, 0.0263, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:55:01,755 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294691.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:55:02,370 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.366e+02 3.566e+02 4.262e+02 5.350e+02 1.116e+03, threshold=8.524e+02, percent-clipped=9.0 2023-05-17 06:55:03,851 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4221, 4.2884, 4.3262, 4.3689, 3.9968, 4.5690, 4.4519, 4.5624], device='cuda:1'), covar=tensor([0.0249, 0.0174, 0.0190, 0.0467, 0.0705, 0.0333, 0.0195, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0202, 0.0198, 0.0252, 0.0247, 0.0227, 0.0182, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 06:55:04,437 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294695.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:55:15,157 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3650, 4.2165, 4.2778, 4.3272, 3.9491, 4.5127, 4.4202, 4.5461], device='cuda:1'), covar=tensor([0.0216, 0.0190, 0.0191, 0.0473, 0.0716, 0.0319, 0.0181, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0202, 0.0198, 0.0252, 0.0247, 0.0227, 0.0181, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 06:55:15,657 INFO [finetune.py:992] (1/2) Epoch 16, batch 11200, loss[loss=0.2154, simple_loss=0.3094, pruned_loss=0.06074, over 10318.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2772, pruned_loss=0.05107, over 2144969.56 frames. ], batch size: 69, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:55:50,019 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6531, 3.3697, 3.5177, 3.7049, 3.3772, 3.8302, 3.7561, 3.8192], device='cuda:1'), covar=tensor([0.0223, 0.0222, 0.0185, 0.0471, 0.0546, 0.0364, 0.0226, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0200, 0.0196, 0.0250, 0.0244, 0.0224, 0.0180, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 06:55:51,927 INFO [finetune.py:992] (1/2) Epoch 16, batch 11250, loss[loss=0.1755, simple_loss=0.2601, pruned_loss=0.04547, over 12185.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2854, pruned_loss=0.05644, over 2071701.10 frames. ], batch size: 31, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:55:54,788 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5877, 2.4017, 3.3752, 3.4310, 3.5666, 3.6743, 3.5209, 2.6030], device='cuda:1'), covar=tensor([0.0078, 0.0438, 0.0190, 0.0114, 0.0118, 0.0098, 0.0130, 0.0498], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0124, 0.0105, 0.0080, 0.0105, 0.0117, 0.0102, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:56:04,002 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294779.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:56:04,125 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.7882, 4.1024, 4.1210, 4.2816, 2.8180, 3.8886, 3.0699, 4.0556], device='cuda:1'), covar=tensor([0.1871, 0.0692, 0.0579, 0.0376, 0.1288, 0.0689, 0.1501, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0270, 0.0299, 0.0362, 0.0244, 0.0245, 0.0263, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 06:56:12,886 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 3.710e+02 4.466e+02 5.909e+02 1.059e+03, threshold=8.932e+02, percent-clipped=2.0 2023-05-17 06:56:26,105 INFO [finetune.py:992] (1/2) Epoch 16, batch 11300, loss[loss=0.2811, simple_loss=0.3493, pruned_loss=0.1064, over 7008.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2914, pruned_loss=0.06036, over 2015611.15 frames. ], batch size: 98, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:56:36,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-05-17 06:56:37,552 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294826.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:56:49,904 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294844.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:56:56,909 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7561, 3.7435, 3.8065, 3.8450, 3.6587, 3.7017, 3.5333, 3.7211], device='cuda:1'), covar=tensor([0.1382, 0.0569, 0.1251, 0.0659, 0.1325, 0.1150, 0.0521, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0695, 0.0612, 0.0629, 0.0843, 0.0747, 0.0564, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:57:00,804 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8928, 4.4296, 4.0743, 4.1573, 4.4787, 3.9784, 4.0649, 3.9612], device='cuda:1'), covar=tensor([0.1612, 0.1057, 0.1464, 0.1804, 0.1058, 0.1955, 0.1730, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0494, 0.0396, 0.0449, 0.0466, 0.0432, 0.0398, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:57:01,351 INFO [finetune.py:992] (1/2) Epoch 16, batch 11350, loss[loss=0.2559, simple_loss=0.3324, pruned_loss=0.08968, over 6542.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2949, pruned_loss=0.06251, over 1976391.76 frames. ], batch size: 98, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:57:09,997 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294874.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:57:21,351 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-17 06:57:23,763 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.315e+02 3.427e+02 4.075e+02 4.942e+02 2.146e+03, threshold=8.150e+02, percent-clipped=2.0 2023-05-17 06:57:28,887 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2059, 4.2129, 2.7133, 2.4012, 3.7584, 2.3073, 3.8920, 2.9072], device='cuda:1'), covar=tensor([0.0684, 0.0429, 0.1143, 0.1685, 0.0235, 0.1556, 0.0357, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0260, 0.0178, 0.0201, 0.0145, 0.0183, 0.0200, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:57:36,780 INFO [finetune.py:992] (1/2) Epoch 16, batch 11400, loss[loss=0.2895, simple_loss=0.3598, pruned_loss=0.1096, over 7497.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2979, pruned_loss=0.0646, over 1931283.24 frames. ], batch size: 98, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:57:42,394 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7455, 3.0408, 2.3960, 2.1914, 2.7443, 2.2635, 2.9855, 2.5841], device='cuda:1'), covar=tensor([0.0618, 0.0582, 0.0937, 0.1509, 0.0278, 0.1272, 0.0493, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0261, 0.0178, 0.0201, 0.0145, 0.0184, 0.0200, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 06:57:42,453 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2650, 3.4633, 3.1609, 3.5406, 3.3224, 2.6539, 3.2335, 2.8367], device='cuda:1'), covar=tensor([0.0974, 0.1056, 0.1572, 0.0841, 0.1412, 0.1655, 0.1293, 0.2757], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0369, 0.0352, 0.0319, 0.0364, 0.0267, 0.0340, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:57:47,540 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294927.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:58:04,488 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7008, 3.3268, 3.5149, 3.7452, 3.4663, 3.8473, 3.8161, 3.8533], device='cuda:1'), covar=tensor([0.0199, 0.0222, 0.0170, 0.0379, 0.0524, 0.0355, 0.0196, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0192, 0.0189, 0.0242, 0.0236, 0.0217, 0.0173, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 06:58:11,669 INFO [finetune.py:992] (1/2) Epoch 16, batch 11450, loss[loss=0.234, simple_loss=0.3164, pruned_loss=0.07583, over 7352.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.3012, pruned_loss=0.06756, over 1882891.35 frames. ], batch size: 98, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:58:30,608 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294989.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:58:32,476 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.574e+02 3.335e+02 3.934e+02 4.570e+02 8.629e+02, threshold=7.868e+02, percent-clipped=1.0 2023-05-17 06:58:34,704 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294995.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:58:46,063 INFO [finetune.py:992] (1/2) Epoch 16, batch 11500, loss[loss=0.3013, simple_loss=0.3472, pruned_loss=0.1277, over 6946.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.3026, pruned_loss=0.06896, over 1857341.76 frames. ], batch size: 101, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:58:47,712 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6850, 4.0745, 3.4655, 4.2361, 3.7561, 2.6654, 3.7747, 2.8701], device='cuda:1'), covar=tensor([0.1065, 0.0840, 0.1739, 0.0614, 0.1626, 0.2001, 0.1123, 0.3542], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0369, 0.0353, 0.0318, 0.0365, 0.0269, 0.0340, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 06:59:08,850 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295043.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:59:13,843 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295050.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 06:59:21,043 INFO [finetune.py:992] (1/2) Epoch 16, batch 11550, loss[loss=0.2268, simple_loss=0.3076, pruned_loss=0.07297, over 10212.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.3033, pruned_loss=0.06983, over 1843809.86 frames. ], batch size: 68, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 06:59:25,990 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295068.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:59:34,705 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295079.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 06:59:43,393 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.687e+02 3.365e+02 3.923e+02 4.459e+02 8.081e+02, threshold=7.845e+02, percent-clipped=1.0 2023-05-17 06:59:43,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-05-17 06:59:55,840 INFO [finetune.py:992] (1/2) Epoch 16, batch 11600, loss[loss=0.2793, simple_loss=0.3434, pruned_loss=0.1076, over 7096.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3056, pruned_loss=0.07152, over 1810945.85 frames. ], batch size: 98, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 07:00:06,705 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295127.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:00:08,148 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295129.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:00:20,505 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295144.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:00:32,616 INFO [finetune.py:992] (1/2) Epoch 16, batch 11650, loss[loss=0.2303, simple_loss=0.3044, pruned_loss=0.07807, over 12019.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3054, pruned_loss=0.07191, over 1800380.69 frames. ], batch size: 40, lr: 3.36e-03, grad_scale: 8.0 2023-05-17 07:00:53,737 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.481e+02 3.491e+02 3.968e+02 4.748e+02 1.073e+03, threshold=7.935e+02, percent-clipped=3.0 2023-05-17 07:00:53,850 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295192.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:01:08,322 INFO [finetune.py:992] (1/2) Epoch 16, batch 11700, loss[loss=0.2318, simple_loss=0.3181, pruned_loss=0.07271, over 10466.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3049, pruned_loss=0.07246, over 1773562.85 frames. ], batch size: 69, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:01:17,749 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295225.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:01:19,082 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295227.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:01:41,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-05-17 07:01:42,641 INFO [finetune.py:992] (1/2) Epoch 16, batch 11750, loss[loss=0.2363, simple_loss=0.313, pruned_loss=0.07977, over 6808.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3051, pruned_loss=0.07247, over 1760757.78 frames. ], batch size: 99, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:01:43,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-17 07:01:52,785 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295275.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:01:55,652 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295279.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:01:56,868 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8286, 4.4391, 4.0712, 4.1068, 4.5142, 3.9529, 4.1649, 3.9364], device='cuda:1'), covar=tensor([0.1712, 0.1091, 0.1487, 0.1958, 0.0952, 0.1995, 0.1616, 0.1431], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0494, 0.0399, 0.0449, 0.0467, 0.0434, 0.0397, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:02:00,331 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295286.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:02:04,212 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.194e+02 3.411e+02 3.937e+02 4.615e+02 7.621e+02, threshold=7.874e+02, percent-clipped=0.0 2023-05-17 07:02:16,901 INFO [finetune.py:992] (1/2) Epoch 16, batch 11800, loss[loss=0.2468, simple_loss=0.3136, pruned_loss=0.09005, over 6737.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3069, pruned_loss=0.07373, over 1742368.96 frames. ], batch size: 103, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:02:33,729 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295333.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:02:38,418 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295340.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:02:41,665 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295345.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:02:51,057 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295359.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:02:52,297 INFO [finetune.py:992] (1/2) Epoch 16, batch 11850, loss[loss=0.2064, simple_loss=0.3136, pruned_loss=0.04965, over 11031.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3095, pruned_loss=0.07574, over 1704672.51 frames. ], batch size: 55, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:02:52,503 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8386, 2.0800, 2.8540, 2.7892, 2.8854, 2.8586, 2.8641, 2.4127], device='cuda:1'), covar=tensor([0.0104, 0.0458, 0.0175, 0.0105, 0.0136, 0.0144, 0.0157, 0.0419], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0123, 0.0104, 0.0079, 0.0103, 0.0117, 0.0101, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:03:09,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-05-17 07:03:14,241 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 3.456e+02 4.054e+02 4.912e+02 9.762e+02, threshold=8.107e+02, percent-clipped=2.0 2023-05-17 07:03:15,835 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295394.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:03:27,561 INFO [finetune.py:992] (1/2) Epoch 16, batch 11900, loss[loss=0.2284, simple_loss=0.3115, pruned_loss=0.07265, over 7125.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3089, pruned_loss=0.0749, over 1694289.32 frames. ], batch size: 100, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:03:33,783 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295420.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:03:36,477 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295424.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:04:02,353 INFO [finetune.py:992] (1/2) Epoch 16, batch 11950, loss[loss=0.206, simple_loss=0.2897, pruned_loss=0.06118, over 11539.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3055, pruned_loss=0.07229, over 1690084.53 frames. ], batch size: 48, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:04:18,002 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9182, 3.7892, 3.8901, 3.6237, 3.7827, 3.6574, 3.8890, 3.5987], device='cuda:1'), covar=tensor([0.0395, 0.0365, 0.0312, 0.0288, 0.0423, 0.0338, 0.0317, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0261, 0.0281, 0.0258, 0.0259, 0.0258, 0.0234, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:04:23,240 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 3.064e+02 3.708e+02 4.359e+02 1.064e+03, threshold=7.417e+02, percent-clipped=1.0 2023-05-17 07:04:26,309 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.6259, 4.1090, 3.4867, 4.2679, 3.8471, 2.4848, 3.7173, 2.8906], device='cuda:1'), covar=tensor([0.1205, 0.0838, 0.1721, 0.0667, 0.1372, 0.2217, 0.1295, 0.3428], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0361, 0.0343, 0.0307, 0.0355, 0.0263, 0.0331, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:04:37,518 INFO [finetune.py:992] (1/2) Epoch 16, batch 12000, loss[loss=0.1902, simple_loss=0.2927, pruned_loss=0.04388, over 10357.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.3012, pruned_loss=0.0688, over 1689187.33 frames. ], batch size: 69, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:04:37,519 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 07:04:50,325 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4431, 5.3949, 5.4742, 5.5041, 5.2363, 5.1944, 5.1023, 5.3193], device='cuda:1'), covar=tensor([0.0740, 0.0430, 0.0724, 0.0536, 0.1292, 0.1324, 0.0485, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0663, 0.0585, 0.0598, 0.0799, 0.0710, 0.0539, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:04:55,572 INFO [finetune.py:1026] (1/2) Epoch 16, validation: loss=0.2865, simple_loss=0.3618, pruned_loss=0.1056, over 1020973.00 frames. 2023-05-17 07:04:55,572 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 07:05:07,822 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4069, 3.1496, 2.9750, 3.2335, 2.6218, 3.1043, 2.6008, 2.6980], device='cuda:1'), covar=tensor([0.1460, 0.0797, 0.0842, 0.0585, 0.1051, 0.0811, 0.1602, 0.0523], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0266, 0.0293, 0.0352, 0.0239, 0.0243, 0.0259, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:05:19,938 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9555, 3.8311, 3.9450, 3.6795, 3.8095, 3.7025, 3.9215, 3.6160], device='cuda:1'), covar=tensor([0.0377, 0.0337, 0.0329, 0.0277, 0.0414, 0.0341, 0.0305, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0261, 0.0280, 0.0258, 0.0259, 0.0258, 0.0234, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:05:26,739 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9657, 2.3873, 2.9980, 3.9629, 2.3173, 4.0004, 3.9765, 4.1050], device='cuda:1'), covar=tensor([0.0144, 0.1452, 0.0512, 0.0157, 0.1520, 0.0248, 0.0236, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0201, 0.0180, 0.0119, 0.0186, 0.0174, 0.0171, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:05:27,443 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295556.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:05:28,717 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295558.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:05:30,613 INFO [finetune.py:992] (1/2) Epoch 16, batch 12050, loss[loss=0.1987, simple_loss=0.2747, pruned_loss=0.06139, over 7083.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2967, pruned_loss=0.06554, over 1701889.54 frames. ], batch size: 99, lr: 3.36e-03, grad_scale: 16.0 2023-05-17 07:05:43,698 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295581.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:05:50,512 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.099e+02 2.927e+02 3.374e+02 4.273e+02 1.460e+03, threshold=6.749e+02, percent-clipped=2.0 2023-05-17 07:06:02,838 INFO [finetune.py:992] (1/2) Epoch 16, batch 12100, loss[loss=0.2402, simple_loss=0.3126, pruned_loss=0.08396, over 6941.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2954, pruned_loss=0.06418, over 1696156.71 frames. ], batch size: 100, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:06:06,820 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295617.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:06:08,024 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295619.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:06:14,564 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295629.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:06:18,216 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295635.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:06:24,508 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295645.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:06:29,470 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2200, 4.2583, 2.7948, 2.5352, 3.7444, 2.5890, 3.8931, 3.0229], device='cuda:1'), covar=tensor([0.0665, 0.0456, 0.1166, 0.1640, 0.0241, 0.1377, 0.0371, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0246, 0.0172, 0.0196, 0.0138, 0.0179, 0.0192, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:06:35,549 INFO [finetune.py:992] (1/2) Epoch 16, batch 12150, loss[loss=0.2222, simple_loss=0.3037, pruned_loss=0.07035, over 6816.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2953, pruned_loss=0.06391, over 1695671.34 frames. ], batch size: 102, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:06:49,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-05-17 07:06:50,778 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295685.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:06:53,242 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295689.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:06:53,987 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295690.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:06:55,025 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.036e+02 3.228e+02 3.632e+02 4.475e+02 1.263e+03, threshold=7.264e+02, percent-clipped=3.0 2023-05-17 07:06:55,839 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295693.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:07:07,176 INFO [finetune.py:992] (1/2) Epoch 16, batch 12200, loss[loss=0.2234, simple_loss=0.3017, pruned_loss=0.07254, over 7190.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2965, pruned_loss=0.06508, over 1670835.67 frames. ], batch size: 98, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:07:09,763 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295715.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:07:11,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7380, 3.5498, 3.6492, 3.7963, 3.5170, 3.8750, 3.8571, 3.8645], device='cuda:1'), covar=tensor([0.0274, 0.0181, 0.0190, 0.0304, 0.0535, 0.0336, 0.0218, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0180, 0.0176, 0.0224, 0.0219, 0.0200, 0.0161, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 07:07:15,384 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295724.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:07:51,914 INFO [finetune.py:992] (1/2) Epoch 17, batch 0, loss[loss=0.1558, simple_loss=0.2493, pruned_loss=0.03112, over 12144.00 frames. ], tot_loss[loss=0.1558, simple_loss=0.2493, pruned_loss=0.03112, over 12144.00 frames. ], batch size: 34, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:07:51,914 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 07:08:09,390 INFO [finetune.py:1026] (1/2) Epoch 17, validation: loss=0.2904, simple_loss=0.3634, pruned_loss=0.1087, over 1020973.00 frames. 2023-05-17 07:08:09,391 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 07:08:10,333 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295746.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:08:23,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.49 vs. limit=5.0 2023-05-17 07:08:28,637 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295772.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:08:36,775 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0720, 4.5339, 3.8780, 4.7797, 4.1839, 2.8026, 4.1659, 2.9456], device='cuda:1'), covar=tensor([0.1078, 0.0763, 0.1807, 0.0558, 0.1565, 0.2072, 0.1128, 0.4127], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0363, 0.0345, 0.0306, 0.0357, 0.0264, 0.0333, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:08:43,014 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 3.043e+02 3.475e+02 4.357e+02 6.788e+02, threshold=6.950e+02, percent-clipped=0.0 2023-05-17 07:08:45,102 INFO [finetune.py:992] (1/2) Epoch 17, batch 50, loss[loss=0.1454, simple_loss=0.2404, pruned_loss=0.02513, over 12259.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2639, pruned_loss=0.04112, over 533203.06 frames. ], batch size: 32, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:09:03,749 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295820.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:09:21,371 INFO [finetune.py:992] (1/2) Epoch 17, batch 100, loss[loss=0.1699, simple_loss=0.265, pruned_loss=0.03738, over 12349.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2633, pruned_loss=0.0402, over 939171.37 frames. ], batch size: 36, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:09:30,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7939, 2.1367, 3.2352, 2.7497, 3.1176, 2.9036, 2.2401, 3.1546], device='cuda:1'), covar=tensor([0.0184, 0.0487, 0.0224, 0.0303, 0.0196, 0.0245, 0.0498, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0199, 0.0185, 0.0182, 0.0210, 0.0161, 0.0192, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:09:40,720 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0102, 4.9605, 4.8195, 4.8701, 4.5558, 4.9443, 4.9506, 5.1124], device='cuda:1'), covar=tensor([0.0236, 0.0180, 0.0232, 0.0397, 0.0853, 0.0379, 0.0160, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0185, 0.0181, 0.0231, 0.0226, 0.0207, 0.0166, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 07:09:41,444 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295873.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:09:47,766 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295881.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:09:47,877 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295881.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:09:56,378 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 2.730e+02 3.190e+02 3.765e+02 6.955e+02, threshold=6.380e+02, percent-clipped=1.0 2023-05-17 07:09:58,562 INFO [finetune.py:992] (1/2) Epoch 17, batch 150, loss[loss=0.1811, simple_loss=0.2669, pruned_loss=0.04767, over 12149.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2607, pruned_loss=0.03912, over 1262224.76 frames. ], batch size: 39, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:10:10,898 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295912.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 07:10:12,375 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295914.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 07:10:17,623 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 2023-05-17 07:10:19,399 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5276, 5.4904, 5.3294, 4.8122, 4.9041, 5.4392, 5.0865, 4.9034], device='cuda:1'), covar=tensor([0.0854, 0.1049, 0.0731, 0.1789, 0.1016, 0.0848, 0.1710, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0554, 0.0510, 0.0617, 0.0410, 0.0708, 0.0753, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-05-17 07:10:22,930 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295929.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:10:26,698 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295934.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:10:27,395 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295935.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:10:34,450 INFO [finetune.py:992] (1/2) Epoch 17, batch 200, loss[loss=0.1403, simple_loss=0.2257, pruned_loss=0.02746, over 12174.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2593, pruned_loss=0.03877, over 1514275.09 frames. ], batch size: 29, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:11:01,737 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295983.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:11:03,254 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295985.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:11:06,057 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295989.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:11:06,174 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4609, 3.5309, 3.1630, 3.0891, 2.7754, 2.6558, 3.4701, 2.4008], device='cuda:1'), covar=tensor([0.0444, 0.0180, 0.0252, 0.0241, 0.0477, 0.0426, 0.0181, 0.0536], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0162, 0.0167, 0.0191, 0.0203, 0.0199, 0.0175, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:11:08,103 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.624e+02 3.123e+02 3.965e+02 6.314e+02, threshold=6.245e+02, percent-clipped=0.0 2023-05-17 07:11:10,234 INFO [finetune.py:992] (1/2) Epoch 17, batch 250, loss[loss=0.1722, simple_loss=0.2543, pruned_loss=0.04506, over 12293.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2589, pruned_loss=0.0389, over 1702568.19 frames. ], batch size: 28, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:11:28,094 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296015.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:11:44,870 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296037.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:11:47,734 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296041.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:11:50,547 INFO [finetune.py:992] (1/2) Epoch 17, batch 300, loss[loss=0.1544, simple_loss=0.2373, pruned_loss=0.03579, over 12098.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2573, pruned_loss=0.03841, over 1855872.75 frames. ], batch size: 32, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:11:53,671 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296049.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:12:03,473 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296063.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:12:23,839 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.686e+02 3.153e+02 3.747e+02 6.952e+02, threshold=6.306e+02, percent-clipped=1.0 2023-05-17 07:12:25,966 INFO [finetune.py:992] (1/2) Epoch 17, batch 350, loss[loss=0.2319, simple_loss=0.3098, pruned_loss=0.07702, over 8026.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2582, pruned_loss=0.03908, over 1953375.64 frames. ], batch size: 97, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:12:36,709 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296110.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:12:55,242 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9610, 2.3162, 3.5183, 2.8884, 3.3205, 3.1251, 2.4368, 3.3602], device='cuda:1'), covar=tensor([0.0150, 0.0438, 0.0155, 0.0306, 0.0183, 0.0197, 0.0393, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0201, 0.0188, 0.0184, 0.0213, 0.0163, 0.0194, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:13:01,362 INFO [finetune.py:992] (1/2) Epoch 17, batch 400, loss[loss=0.1662, simple_loss=0.2563, pruned_loss=0.03802, over 12090.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2569, pruned_loss=0.03824, over 2053589.37 frames. ], batch size: 39, lr: 3.35e-03, grad_scale: 16.0 2023-05-17 07:13:06,597 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.5349, 4.9189, 3.1186, 3.0340, 4.1404, 2.9675, 4.1113, 3.6197], device='cuda:1'), covar=tensor([0.0691, 0.0482, 0.1128, 0.1406, 0.0356, 0.1184, 0.0496, 0.0701], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0253, 0.0176, 0.0200, 0.0142, 0.0182, 0.0197, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:13:24,202 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296176.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:13:37,025 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.556e+02 3.066e+02 3.538e+02 2.337e+03, threshold=6.132e+02, percent-clipped=1.0 2023-05-17 07:13:38,569 INFO [finetune.py:992] (1/2) Epoch 17, batch 450, loss[loss=0.1402, simple_loss=0.2262, pruned_loss=0.02707, over 12288.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2564, pruned_loss=0.03814, over 2130290.52 frames. ], batch size: 28, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:13:49,311 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7445, 2.8199, 4.5077, 4.8127, 2.9216, 2.5882, 2.9411, 2.1088], device='cuda:1'), covar=tensor([0.1735, 0.3570, 0.0547, 0.0449, 0.1447, 0.2853, 0.3004, 0.4644], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0395, 0.0277, 0.0303, 0.0279, 0.0321, 0.0401, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:13:51,319 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296212.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:13:52,684 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296214.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:14:03,407 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296229.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:14:14,662 INFO [finetune.py:992] (1/2) Epoch 17, batch 500, loss[loss=0.1693, simple_loss=0.2612, pruned_loss=0.03872, over 12051.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2557, pruned_loss=0.03769, over 2193787.02 frames. ], batch size: 37, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:14:25,397 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296260.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:14:26,794 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296262.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:14:43,417 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296285.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:14:48,992 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.639e+02 3.014e+02 3.514e+02 6.174e+02, threshold=6.028e+02, percent-clipped=1.0 2023-05-17 07:14:50,409 INFO [finetune.py:992] (1/2) Epoch 17, batch 550, loss[loss=0.1506, simple_loss=0.2372, pruned_loss=0.03199, over 12185.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2554, pruned_loss=0.03766, over 2235661.73 frames. ], batch size: 31, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:14:52,639 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0952, 2.4523, 3.6699, 3.0400, 3.5057, 3.1836, 2.4930, 3.5176], device='cuda:1'), covar=tensor([0.0175, 0.0407, 0.0156, 0.0279, 0.0167, 0.0222, 0.0447, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0205, 0.0191, 0.0187, 0.0216, 0.0167, 0.0197, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:15:04,711 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296315.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:15:18,763 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296333.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:15:24,408 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296341.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:15:27,038 INFO [finetune.py:992] (1/2) Epoch 17, batch 600, loss[loss=0.1618, simple_loss=0.2601, pruned_loss=0.0318, over 12341.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2542, pruned_loss=0.0372, over 2261674.49 frames. ], batch size: 35, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:15:39,654 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.56 vs. limit=2.0 2023-05-17 07:15:49,549 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296376.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:15:58,521 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296389.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:16:01,310 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.646e+02 2.973e+02 3.749e+02 7.703e+02, threshold=5.946e+02, percent-clipped=3.0 2023-05-17 07:16:02,814 INFO [finetune.py:992] (1/2) Epoch 17, batch 650, loss[loss=0.1617, simple_loss=0.259, pruned_loss=0.03226, over 12352.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2548, pruned_loss=0.03773, over 2281528.42 frames. ], batch size: 35, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:16:10,570 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296405.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:16:31,576 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296434.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:16:39,015 INFO [finetune.py:992] (1/2) Epoch 17, batch 700, loss[loss=0.1734, simple_loss=0.2609, pruned_loss=0.04294, over 12024.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2547, pruned_loss=0.0375, over 2300080.91 frames. ], batch size: 40, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:17:02,173 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296476.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:17:14,294 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296492.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:17:14,812 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.527e+02 2.946e+02 3.308e+02 5.689e+02, threshold=5.892e+02, percent-clipped=0.0 2023-05-17 07:17:16,318 INFO [finetune.py:992] (1/2) Epoch 17, batch 750, loss[loss=0.1645, simple_loss=0.2406, pruned_loss=0.04418, over 11795.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2543, pruned_loss=0.03715, over 2318822.63 frames. ], batch size: 26, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:17:16,527 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296495.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:17:18,000 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296497.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:17:30,161 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296514.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:17:32,984 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2113, 5.0342, 5.1366, 5.1833, 4.8252, 4.8229, 4.5812, 5.0685], device='cuda:1'), covar=tensor([0.0829, 0.0652, 0.1000, 0.0645, 0.2013, 0.1605, 0.0647, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0698, 0.0612, 0.0623, 0.0834, 0.0749, 0.0563, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:17:37,188 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296524.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:17:40,845 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296529.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:17:52,169 INFO [finetune.py:992] (1/2) Epoch 17, batch 800, loss[loss=0.1902, simple_loss=0.2741, pruned_loss=0.05314, over 11569.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2539, pruned_loss=0.03684, over 2339516.43 frames. ], batch size: 48, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:17:58,050 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296553.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:18:01,399 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296558.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:18:13,596 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296575.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:18:14,879 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296577.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:18:26,281 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.820e+02 3.244e+02 3.783e+02 6.620e+02, threshold=6.487e+02, percent-clipped=2.0 2023-05-17 07:18:27,672 INFO [finetune.py:992] (1/2) Epoch 17, batch 850, loss[loss=0.1784, simple_loss=0.2661, pruned_loss=0.04532, over 12311.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2549, pruned_loss=0.03744, over 2345712.37 frames. ], batch size: 34, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:18:41,359 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296613.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:18:48,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-17 07:19:04,461 INFO [finetune.py:992] (1/2) Epoch 17, batch 900, loss[loss=0.1569, simple_loss=0.2393, pruned_loss=0.03723, over 12385.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.03704, over 2352332.90 frames. ], batch size: 30, lr: 3.35e-03, grad_scale: 8.0 2023-05-17 07:19:05,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-05-17 07:19:15,021 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-05-17 07:19:23,081 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296671.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:19:25,456 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4638, 4.8647, 3.2134, 2.8679, 4.0607, 2.8288, 4.0788, 3.5565], device='cuda:1'), covar=tensor([0.0654, 0.0502, 0.1106, 0.1545, 0.0319, 0.1340, 0.0518, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0259, 0.0178, 0.0203, 0.0144, 0.0186, 0.0201, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:19:25,474 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296674.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:19:35,334 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296688.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:19:38,586 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.605e+02 2.984e+02 3.568e+02 6.116e+02, threshold=5.969e+02, percent-clipped=0.0 2023-05-17 07:19:40,041 INFO [finetune.py:992] (1/2) Epoch 17, batch 950, loss[loss=0.1684, simple_loss=0.263, pruned_loss=0.03689, over 12347.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2541, pruned_loss=0.03689, over 2360552.90 frames. ], batch size: 35, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:19:47,179 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296705.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:19:57,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-05-17 07:20:10,201 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5375, 2.8006, 3.3397, 4.3828, 2.6748, 4.3774, 4.4792, 4.5428], device='cuda:1'), covar=tensor([0.0166, 0.1282, 0.0506, 0.0166, 0.1259, 0.0292, 0.0171, 0.0113], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0209, 0.0186, 0.0124, 0.0194, 0.0181, 0.0178, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:20:15,821 INFO [finetune.py:992] (1/2) Epoch 17, batch 1000, loss[loss=0.1684, simple_loss=0.2549, pruned_loss=0.04099, over 12123.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2533, pruned_loss=0.03659, over 2371844.46 frames. ], batch size: 38, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:20:18,960 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296749.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:20:21,782 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296753.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:20:43,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-17 07:20:45,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 2023-05-17 07:20:49,295 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296790.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:20:51,278 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.591e+02 3.027e+02 3.502e+02 6.154e+02, threshold=6.055e+02, percent-clipped=2.0 2023-05-17 07:20:52,714 INFO [finetune.py:992] (1/2) Epoch 17, batch 1050, loss[loss=0.1784, simple_loss=0.2673, pruned_loss=0.04474, over 11161.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2531, pruned_loss=0.03681, over 2363132.27 frames. ], batch size: 55, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:21:04,054 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5506, 3.7259, 3.2574, 3.1473, 2.8323, 2.7782, 3.7029, 2.4133], device='cuda:1'), covar=tensor([0.0377, 0.0151, 0.0227, 0.0226, 0.0492, 0.0381, 0.0144, 0.0488], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0162, 0.0168, 0.0192, 0.0203, 0.0200, 0.0176, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:21:29,016 INFO [finetune.py:992] (1/2) Epoch 17, batch 1100, loss[loss=0.159, simple_loss=0.2582, pruned_loss=0.02992, over 12159.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2537, pruned_loss=0.03681, over 2373831.06 frames. ], batch size: 34, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:21:31,236 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296848.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 07:21:34,910 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296853.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:21:47,050 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296870.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:21:49,320 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296873.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:21:59,839 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4368, 5.0454, 5.4227, 4.5946, 5.0077, 4.7333, 5.4062, 5.1180], device='cuda:1'), covar=tensor([0.0351, 0.0434, 0.0356, 0.0355, 0.0495, 0.0452, 0.0328, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0276, 0.0297, 0.0273, 0.0274, 0.0272, 0.0248, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:22:03,209 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.842e+02 2.567e+02 3.061e+02 3.490e+02 7.482e+02, threshold=6.122e+02, percent-clipped=1.0 2023-05-17 07:22:04,672 INFO [finetune.py:992] (1/2) Epoch 17, batch 1150, loss[loss=0.1913, simple_loss=0.2783, pruned_loss=0.05217, over 11825.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.254, pruned_loss=0.03721, over 2375614.55 frames. ], batch size: 44, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:22:26,509 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3822, 4.8783, 4.1494, 4.9734, 4.5366, 3.0763, 4.3102, 3.0722], device='cuda:1'), covar=tensor([0.0913, 0.0715, 0.1547, 0.0634, 0.1322, 0.1719, 0.1022, 0.3586], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0378, 0.0360, 0.0325, 0.0371, 0.0274, 0.0346, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:22:34,246 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296934.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:22:34,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-05-17 07:22:37,014 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296938.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:22:41,889 INFO [finetune.py:992] (1/2) Epoch 17, batch 1200, loss[loss=0.1793, simple_loss=0.271, pruned_loss=0.04383, over 12357.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2528, pruned_loss=0.03694, over 2377880.11 frames. ], batch size: 36, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:22:59,075 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296969.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:23:00,474 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296971.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:23:07,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 2023-05-17 07:23:07,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2023-05-17 07:23:15,926 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.560e+02 3.113e+02 3.575e+02 6.952e+02, threshold=6.225e+02, percent-clipped=1.0 2023-05-17 07:23:17,389 INFO [finetune.py:992] (1/2) Epoch 17, batch 1250, loss[loss=0.1607, simple_loss=0.2538, pruned_loss=0.03377, over 12286.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2537, pruned_loss=0.03705, over 2382333.80 frames. ], batch size: 37, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:23:20,554 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296999.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:23:34,838 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297019.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:23:52,751 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297044.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:23:53,385 INFO [finetune.py:992] (1/2) Epoch 17, batch 1300, loss[loss=0.1648, simple_loss=0.2543, pruned_loss=0.03767, over 12366.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2534, pruned_loss=0.03689, over 2380615.45 frames. ], batch size: 35, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:24:13,425 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-05-17 07:24:19,715 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0501, 4.6710, 4.7586, 4.9262, 4.7669, 4.9656, 4.8738, 2.4456], device='cuda:1'), covar=tensor([0.0124, 0.0081, 0.0105, 0.0082, 0.0056, 0.0099, 0.0091, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0081, 0.0085, 0.0075, 0.0062, 0.0095, 0.0083, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:24:26,823 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297090.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:24:28,822 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.588e+02 2.952e+02 3.635e+02 8.207e+02, threshold=5.903e+02, percent-clipped=3.0 2023-05-17 07:24:29,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 07:24:30,306 INFO [finetune.py:992] (1/2) Epoch 17, batch 1350, loss[loss=0.1774, simple_loss=0.2715, pruned_loss=0.04163, over 12060.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2535, pruned_loss=0.0367, over 2374475.30 frames. ], batch size: 40, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:24:47,511 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9279, 3.9118, 3.9546, 4.0203, 3.7962, 3.8055, 3.6509, 3.9137], device='cuda:1'), covar=tensor([0.1126, 0.0774, 0.1364, 0.0730, 0.1777, 0.1381, 0.0667, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0710, 0.0623, 0.0636, 0.0851, 0.0762, 0.0569, 0.0489], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:24:57,021 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9483, 4.5409, 4.6245, 4.8120, 4.6005, 4.8280, 4.7490, 2.4998], device='cuda:1'), covar=tensor([0.0111, 0.0083, 0.0112, 0.0074, 0.0065, 0.0098, 0.0096, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0081, 0.0085, 0.0076, 0.0062, 0.0096, 0.0084, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:25:00,378 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297138.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:25:05,295 INFO [finetune.py:992] (1/2) Epoch 17, batch 1400, loss[loss=0.1575, simple_loss=0.2501, pruned_loss=0.03245, over 12167.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.03702, over 2373375.07 frames. ], batch size: 36, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:25:07,547 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297148.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:25:09,779 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1870, 4.7624, 5.0094, 5.0649, 4.9397, 5.1071, 5.0382, 2.9402], device='cuda:1'), covar=tensor([0.0102, 0.0080, 0.0085, 0.0057, 0.0049, 0.0104, 0.0068, 0.0675], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0081, 0.0085, 0.0076, 0.0062, 0.0096, 0.0084, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:25:11,205 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297153.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:25:23,400 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297170.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:25:23,513 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3588, 3.6606, 3.2850, 3.0932, 2.8109, 2.7049, 3.5953, 2.2668], device='cuda:1'), covar=tensor([0.0445, 0.0156, 0.0210, 0.0249, 0.0430, 0.0399, 0.0173, 0.0556], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0164, 0.0169, 0.0194, 0.0205, 0.0203, 0.0179, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:25:39,579 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.670e+02 3.053e+02 3.744e+02 7.369e+02, threshold=6.105e+02, percent-clipped=2.0 2023-05-17 07:25:40,979 INFO [finetune.py:992] (1/2) Epoch 17, batch 1450, loss[loss=0.1729, simple_loss=0.2654, pruned_loss=0.04016, over 12382.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2532, pruned_loss=0.03695, over 2384385.09 frames. ], batch size: 38, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:25:41,772 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297196.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:25:45,703 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297201.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:25:59,059 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297218.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:26:01,860 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297222.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:26:06,842 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297229.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 07:26:18,056 INFO [finetune.py:992] (1/2) Epoch 17, batch 1500, loss[loss=0.14, simple_loss=0.2279, pruned_loss=0.02607, over 12111.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2533, pruned_loss=0.03725, over 2379580.24 frames. ], batch size: 32, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:26:35,638 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297269.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:26:38,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-17 07:26:45,603 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297283.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:26:52,593 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.625e+02 2.986e+02 3.551e+02 5.451e+02, threshold=5.973e+02, percent-clipped=0.0 2023-05-17 07:26:53,392 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297294.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:26:54,035 INFO [finetune.py:992] (1/2) Epoch 17, batch 1550, loss[loss=0.1336, simple_loss=0.2208, pruned_loss=0.02322, over 12036.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.252, pruned_loss=0.0367, over 2386764.45 frames. ], batch size: 31, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:27:09,872 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297317.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:27:29,118 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297344.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:27:29,715 INFO [finetune.py:992] (1/2) Epoch 17, batch 1600, loss[loss=0.1639, simple_loss=0.2519, pruned_loss=0.03792, over 12135.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.253, pruned_loss=0.03676, over 2382558.67 frames. ], batch size: 36, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:27:30,575 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3819, 4.9627, 5.3724, 4.6664, 5.0209, 4.7798, 5.4118, 5.0446], device='cuda:1'), covar=tensor([0.0332, 0.0424, 0.0304, 0.0304, 0.0506, 0.0364, 0.0217, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0279, 0.0301, 0.0276, 0.0277, 0.0276, 0.0252, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:27:30,653 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1113, 2.5874, 3.6838, 3.0515, 3.4903, 3.2158, 2.6548, 3.5631], device='cuda:1'), covar=tensor([0.0156, 0.0362, 0.0160, 0.0276, 0.0166, 0.0195, 0.0338, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0209, 0.0195, 0.0191, 0.0221, 0.0169, 0.0200, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:27:37,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.59 vs. limit=5.0 2023-05-17 07:28:04,620 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297392.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:28:05,253 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.500e+02 2.563e+02 3.041e+02 3.694e+02 7.011e+02, threshold=6.081e+02, percent-clipped=1.0 2023-05-17 07:28:06,713 INFO [finetune.py:992] (1/2) Epoch 17, batch 1650, loss[loss=0.1539, simple_loss=0.2406, pruned_loss=0.03364, over 12297.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2529, pruned_loss=0.03684, over 2382588.84 frames. ], batch size: 33, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:28:30,072 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1570, 2.3225, 3.6449, 3.0034, 3.5100, 3.1711, 2.4527, 3.4851], device='cuda:1'), covar=tensor([0.0152, 0.0479, 0.0188, 0.0306, 0.0182, 0.0252, 0.0480, 0.0172], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0209, 0.0195, 0.0191, 0.0222, 0.0170, 0.0201, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:28:42,804 INFO [finetune.py:992] (1/2) Epoch 17, batch 1700, loss[loss=0.1794, simple_loss=0.2668, pruned_loss=0.04602, over 12195.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2528, pruned_loss=0.0365, over 2389009.60 frames. ], batch size: 35, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:29:16,884 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.569e+02 2.942e+02 3.457e+02 5.618e+02, threshold=5.884e+02, percent-clipped=0.0 2023-05-17 07:29:18,384 INFO [finetune.py:992] (1/2) Epoch 17, batch 1750, loss[loss=0.1582, simple_loss=0.256, pruned_loss=0.03026, over 12193.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2528, pruned_loss=0.03626, over 2392720.50 frames. ], batch size: 35, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:29:43,684 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297529.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:29:51,563 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1847, 3.5158, 3.5706, 3.9378, 2.7849, 3.6188, 2.5592, 3.4406], device='cuda:1'), covar=tensor([0.1672, 0.0853, 0.0860, 0.0660, 0.1238, 0.0654, 0.1829, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0276, 0.0303, 0.0368, 0.0249, 0.0252, 0.0267, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:29:54,856 INFO [finetune.py:992] (1/2) Epoch 17, batch 1800, loss[loss=0.1787, simple_loss=0.2763, pruned_loss=0.04054, over 12132.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2538, pruned_loss=0.03649, over 2389283.18 frames. ], batch size: 38, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:30:17,840 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297577.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:30:18,602 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297578.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:30:29,049 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.682e+02 3.139e+02 3.795e+02 6.234e+02, threshold=6.278e+02, percent-clipped=1.0 2023-05-17 07:30:29,875 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297594.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:30:30,472 INFO [finetune.py:992] (1/2) Epoch 17, batch 1850, loss[loss=0.1594, simple_loss=0.2463, pruned_loss=0.03623, over 12089.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2536, pruned_loss=0.03642, over 2392635.62 frames. ], batch size: 32, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:31:04,646 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297642.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:31:06,618 INFO [finetune.py:992] (1/2) Epoch 17, batch 1900, loss[loss=0.1753, simple_loss=0.2682, pruned_loss=0.04126, over 10512.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2543, pruned_loss=0.03682, over 2386910.05 frames. ], batch size: 69, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:31:20,981 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4140, 4.9274, 5.3710, 4.7227, 5.0210, 4.7703, 5.3935, 5.1070], device='cuda:1'), covar=tensor([0.0287, 0.0455, 0.0297, 0.0274, 0.0442, 0.0360, 0.0239, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0279, 0.0301, 0.0276, 0.0277, 0.0277, 0.0251, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:31:42,145 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.617e+02 2.974e+02 3.412e+02 6.785e+02, threshold=5.949e+02, percent-clipped=1.0 2023-05-17 07:31:43,632 INFO [finetune.py:992] (1/2) Epoch 17, batch 1950, loss[loss=0.141, simple_loss=0.225, pruned_loss=0.02854, over 12343.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2541, pruned_loss=0.03691, over 2383957.06 frames. ], batch size: 30, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:31:45,343 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6902, 3.8025, 3.3662, 3.2195, 3.0212, 2.8529, 3.7796, 2.4630], device='cuda:1'), covar=tensor([0.0415, 0.0136, 0.0221, 0.0274, 0.0406, 0.0420, 0.0146, 0.0496], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0165, 0.0171, 0.0196, 0.0206, 0.0204, 0.0180, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:32:03,049 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.97 vs. limit=5.0 2023-05-17 07:32:19,266 INFO [finetune.py:992] (1/2) Epoch 17, batch 2000, loss[loss=0.1615, simple_loss=0.2509, pruned_loss=0.03611, over 12299.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2544, pruned_loss=0.03703, over 2378128.85 frames. ], batch size: 34, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:32:53,425 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.509e+02 3.015e+02 3.436e+02 6.650e+02, threshold=6.030e+02, percent-clipped=2.0 2023-05-17 07:32:54,840 INFO [finetune.py:992] (1/2) Epoch 17, batch 2050, loss[loss=0.1793, simple_loss=0.2706, pruned_loss=0.04403, over 12368.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2546, pruned_loss=0.03725, over 2372324.09 frames. ], batch size: 36, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:33:31,898 INFO [finetune.py:992] (1/2) Epoch 17, batch 2100, loss[loss=0.2057, simple_loss=0.2834, pruned_loss=0.06404, over 8320.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2547, pruned_loss=0.03747, over 2372794.57 frames. ], batch size: 98, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:33:55,307 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297878.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:34:05,844 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.706e+02 3.127e+02 3.743e+02 8.710e+02, threshold=6.254e+02, percent-clipped=2.0 2023-05-17 07:34:07,289 INFO [finetune.py:992] (1/2) Epoch 17, batch 2150, loss[loss=0.2093, simple_loss=0.2995, pruned_loss=0.0596, over 8198.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2542, pruned_loss=0.03727, over 2377542.55 frames. ], batch size: 97, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:34:29,304 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297926.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:34:42,804 INFO [finetune.py:992] (1/2) Epoch 17, batch 2200, loss[loss=0.1584, simple_loss=0.2489, pruned_loss=0.034, over 12253.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2538, pruned_loss=0.03702, over 2379090.79 frames. ], batch size: 32, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:34:52,198 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297957.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:35:03,384 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2962, 2.7143, 3.8885, 3.3417, 3.7719, 3.4050, 2.8461, 3.7557], device='cuda:1'), covar=tensor([0.0123, 0.0362, 0.0142, 0.0242, 0.0141, 0.0179, 0.0368, 0.0146], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0211, 0.0197, 0.0194, 0.0224, 0.0170, 0.0202, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:35:17,966 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.570e+02 3.028e+02 3.621e+02 1.131e+03, threshold=6.056e+02, percent-clipped=3.0 2023-05-17 07:35:19,381 INFO [finetune.py:992] (1/2) Epoch 17, batch 2250, loss[loss=0.1675, simple_loss=0.2568, pruned_loss=0.03914, over 12097.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2533, pruned_loss=0.03681, over 2384072.53 frames. ], batch size: 33, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:35:27,266 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6423, 2.8431, 3.7211, 4.6899, 4.0472, 4.7097, 3.9463, 3.5167], device='cuda:1'), covar=tensor([0.0043, 0.0420, 0.0162, 0.0039, 0.0122, 0.0076, 0.0154, 0.0348], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0126, 0.0106, 0.0082, 0.0106, 0.0118, 0.0103, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:35:39,259 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298018.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:35:58,215 INFO [finetune.py:992] (1/2) Epoch 17, batch 2300, loss[loss=0.1761, simple_loss=0.2708, pruned_loss=0.04069, over 12187.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2538, pruned_loss=0.03711, over 2381179.62 frames. ], batch size: 31, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:36:32,015 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.758e+02 3.155e+02 4.025e+02 7.996e+02, threshold=6.311e+02, percent-clipped=3.0 2023-05-17 07:36:33,450 INFO [finetune.py:992] (1/2) Epoch 17, batch 2350, loss[loss=0.1298, simple_loss=0.2172, pruned_loss=0.02122, over 12117.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2538, pruned_loss=0.03721, over 2379591.46 frames. ], batch size: 30, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:36:33,663 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2661, 4.6458, 2.8362, 2.4057, 4.0256, 2.4526, 3.9421, 3.2103], device='cuda:1'), covar=tensor([0.0845, 0.0633, 0.1322, 0.1998, 0.0352, 0.1660, 0.0616, 0.0881], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0262, 0.0179, 0.0204, 0.0144, 0.0186, 0.0203, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:36:47,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-17 07:36:58,107 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3549, 4.7539, 4.1332, 5.0310, 4.5368, 3.2491, 4.4321, 3.0914], device='cuda:1'), covar=tensor([0.0834, 0.0763, 0.1401, 0.0591, 0.1057, 0.1524, 0.0891, 0.3397], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0384, 0.0367, 0.0335, 0.0378, 0.0279, 0.0354, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:37:10,632 INFO [finetune.py:992] (1/2) Epoch 17, batch 2400, loss[loss=0.1286, simple_loss=0.218, pruned_loss=0.01961, over 12353.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2537, pruned_loss=0.0368, over 2383344.91 frames. ], batch size: 30, lr: 3.34e-03, grad_scale: 8.0 2023-05-17 07:37:44,952 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.507e+02 3.015e+02 3.482e+02 1.547e+03, threshold=6.029e+02, percent-clipped=2.0 2023-05-17 07:37:46,362 INFO [finetune.py:992] (1/2) Epoch 17, batch 2450, loss[loss=0.1768, simple_loss=0.2695, pruned_loss=0.04211, over 12035.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2536, pruned_loss=0.0369, over 2377687.91 frames. ], batch size: 42, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:38:22,192 INFO [finetune.py:992] (1/2) Epoch 17, batch 2500, loss[loss=0.1733, simple_loss=0.2581, pruned_loss=0.04428, over 11612.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2534, pruned_loss=0.03672, over 2387593.48 frames. ], batch size: 48, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:38:28,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 07:38:40,477 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1425, 4.9569, 4.9119, 4.9105, 4.6745, 5.1274, 5.1242, 5.2913], device='cuda:1'), covar=tensor([0.0321, 0.0259, 0.0253, 0.0430, 0.0784, 0.0578, 0.0200, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0206, 0.0198, 0.0253, 0.0248, 0.0229, 0.0182, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 07:38:57,554 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.660e+02 3.049e+02 3.576e+02 5.574e+02, threshold=6.099e+02, percent-clipped=0.0 2023-05-17 07:38:59,007 INFO [finetune.py:992] (1/2) Epoch 17, batch 2550, loss[loss=0.1449, simple_loss=0.2429, pruned_loss=0.0234, over 12153.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2529, pruned_loss=0.03645, over 2384445.78 frames. ], batch size: 36, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:39:04,083 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5423, 2.8558, 3.5782, 4.4762, 3.8438, 4.5719, 3.8114, 3.3867], device='cuda:1'), covar=tensor([0.0039, 0.0363, 0.0176, 0.0049, 0.0139, 0.0058, 0.0156, 0.0338], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0126, 0.0106, 0.0082, 0.0106, 0.0118, 0.0103, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:39:11,978 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298313.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:39:20,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-05-17 07:39:34,555 INFO [finetune.py:992] (1/2) Epoch 17, batch 2600, loss[loss=0.1628, simple_loss=0.252, pruned_loss=0.03676, over 12158.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2537, pruned_loss=0.03693, over 2379618.77 frames. ], batch size: 34, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:39:51,673 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1925, 2.5831, 3.4451, 4.1123, 3.7276, 4.1802, 3.6296, 3.0373], device='cuda:1'), covar=tensor([0.0049, 0.0411, 0.0165, 0.0059, 0.0138, 0.0085, 0.0145, 0.0385], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0126, 0.0106, 0.0082, 0.0106, 0.0119, 0.0103, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:40:09,182 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.526e+02 3.029e+02 3.726e+02 9.097e+02, threshold=6.058e+02, percent-clipped=2.0 2023-05-17 07:40:10,502 INFO [finetune.py:992] (1/2) Epoch 17, batch 2650, loss[loss=0.1438, simple_loss=0.236, pruned_loss=0.02575, over 12273.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2529, pruned_loss=0.03653, over 2381988.61 frames. ], batch size: 28, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:40:33,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.36 vs. limit=5.0 2023-05-17 07:40:47,091 INFO [finetune.py:992] (1/2) Epoch 17, batch 2700, loss[loss=0.1863, simple_loss=0.2799, pruned_loss=0.04641, over 11645.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2535, pruned_loss=0.03662, over 2378664.06 frames. ], batch size: 48, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:41:14,068 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298482.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:41:21,936 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 2.706e+02 3.064e+02 3.739e+02 6.020e+02, threshold=6.128e+02, percent-clipped=0.0 2023-05-17 07:41:23,399 INFO [finetune.py:992] (1/2) Epoch 17, batch 2750, loss[loss=0.1286, simple_loss=0.2158, pruned_loss=0.02074, over 12205.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2525, pruned_loss=0.03633, over 2376001.35 frames. ], batch size: 29, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:41:34,528 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7138, 2.4980, 4.6639, 4.9832, 2.9524, 2.4957, 2.7935, 1.9846], device='cuda:1'), covar=tensor([0.1888, 0.3860, 0.0486, 0.0331, 0.1285, 0.2876, 0.3579, 0.5520], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0395, 0.0276, 0.0303, 0.0279, 0.0321, 0.0402, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:41:49,147 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298530.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:41:55,001 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4347, 4.1661, 4.2336, 4.4976, 3.0714, 3.9867, 2.7070, 4.2037], device='cuda:1'), covar=tensor([0.1592, 0.0668, 0.0811, 0.0644, 0.1212, 0.0656, 0.1870, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0271, 0.0299, 0.0362, 0.0245, 0.0248, 0.0262, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:41:58,691 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298543.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:41:59,895 INFO [finetune.py:992] (1/2) Epoch 17, batch 2800, loss[loss=0.1497, simple_loss=0.2416, pruned_loss=0.02892, over 12285.00 frames. ], tot_loss[loss=0.1617, simple_loss=0.2516, pruned_loss=0.03593, over 2381343.31 frames. ], batch size: 33, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:42:33,439 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298591.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:42:34,704 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.094e+02 2.580e+02 2.945e+02 3.483e+02 6.486e+02, threshold=5.889e+02, percent-clipped=1.0 2023-05-17 07:42:36,183 INFO [finetune.py:992] (1/2) Epoch 17, batch 2850, loss[loss=0.1613, simple_loss=0.2524, pruned_loss=0.03516, over 10503.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2516, pruned_loss=0.03617, over 2383865.67 frames. ], batch size: 68, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:42:43,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-05-17 07:42:48,924 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=298613.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:43:00,450 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298629.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:43:11,502 INFO [finetune.py:992] (1/2) Epoch 17, batch 2900, loss[loss=0.1675, simple_loss=0.2628, pruned_loss=0.03606, over 12145.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2523, pruned_loss=0.03644, over 2375702.61 frames. ], batch size: 36, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:43:22,738 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=298661.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:43:27,011 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3353, 4.9506, 5.2317, 5.1756, 4.9689, 5.1683, 5.1391, 2.9912], device='cuda:1'), covar=tensor([0.0101, 0.0068, 0.0062, 0.0057, 0.0048, 0.0092, 0.0075, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0087, 0.0077, 0.0062, 0.0098, 0.0085, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:43:43,453 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298689.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:43:44,200 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298690.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:43:46,194 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 2.709e+02 3.170e+02 3.797e+02 7.098e+02, threshold=6.341e+02, percent-clipped=3.0 2023-05-17 07:43:47,689 INFO [finetune.py:992] (1/2) Epoch 17, batch 2950, loss[loss=0.1666, simple_loss=0.2627, pruned_loss=0.03521, over 11592.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.252, pruned_loss=0.03649, over 2361442.50 frames. ], batch size: 48, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:44:05,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-05-17 07:44:24,205 INFO [finetune.py:992] (1/2) Epoch 17, batch 3000, loss[loss=0.1529, simple_loss=0.2469, pruned_loss=0.02944, over 12293.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.2514, pruned_loss=0.03607, over 2369159.37 frames. ], batch size: 33, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:44:24,206 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 07:44:42,390 INFO [finetune.py:1026] (1/2) Epoch 17, validation: loss=0.3104, simple_loss=0.3873, pruned_loss=0.1167, over 1020973.00 frames. 2023-05-17 07:44:42,391 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 07:44:46,165 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298750.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:45:17,055 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.736e+02 3.128e+02 3.461e+02 8.040e+02, threshold=6.255e+02, percent-clipped=1.0 2023-05-17 07:45:18,416 INFO [finetune.py:992] (1/2) Epoch 17, batch 3050, loss[loss=0.1729, simple_loss=0.2624, pruned_loss=0.04172, over 12113.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2521, pruned_loss=0.03657, over 2364887.63 frames. ], batch size: 39, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:45:50,515 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298838.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:45:55,401 INFO [finetune.py:992] (1/2) Epoch 17, batch 3100, loss[loss=0.1624, simple_loss=0.2673, pruned_loss=0.02879, over 12269.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2526, pruned_loss=0.03663, over 2368227.71 frames. ], batch size: 37, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:46:08,341 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0786, 6.0568, 5.8091, 5.2936, 5.2680, 5.9477, 5.5195, 5.3029], device='cuda:1'), covar=tensor([0.0667, 0.0810, 0.0694, 0.1851, 0.0730, 0.0788, 0.1695, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0643, 0.0574, 0.0529, 0.0647, 0.0428, 0.0742, 0.0795, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-17 07:46:24,634 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0648, 5.8895, 5.5321, 5.4482, 6.0328, 5.2442, 5.3537, 5.4800], device='cuda:1'), covar=tensor([0.1560, 0.0912, 0.0929, 0.2072, 0.0913, 0.2343, 0.2100, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0500, 0.0405, 0.0451, 0.0473, 0.0441, 0.0407, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:46:24,645 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298886.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:46:29,431 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.477e+02 2.875e+02 3.677e+02 5.878e+02, threshold=5.751e+02, percent-clipped=0.0 2023-05-17 07:46:30,909 INFO [finetune.py:992] (1/2) Epoch 17, batch 3150, loss[loss=0.1551, simple_loss=0.2447, pruned_loss=0.0328, over 12101.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2527, pruned_loss=0.03661, over 2376525.01 frames. ], batch size: 32, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:46:47,635 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1607, 2.2391, 2.9675, 3.0354, 3.0326, 3.2021, 2.9894, 2.4542], device='cuda:1'), covar=tensor([0.0087, 0.0409, 0.0169, 0.0085, 0.0160, 0.0105, 0.0145, 0.0373], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0126, 0.0106, 0.0082, 0.0107, 0.0118, 0.0103, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:47:07,190 INFO [finetune.py:992] (1/2) Epoch 17, batch 3200, loss[loss=0.1673, simple_loss=0.2514, pruned_loss=0.04162, over 12190.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2525, pruned_loss=0.03663, over 2378507.77 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:47:36,546 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298985.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:47:40,102 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8351, 5.1248, 4.2909, 5.1817, 4.7528, 3.0116, 4.3065, 3.2334], device='cuda:1'), covar=tensor([0.0514, 0.0527, 0.1347, 0.0519, 0.0989, 0.1682, 0.1134, 0.3037], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0383, 0.0365, 0.0334, 0.0377, 0.0280, 0.0355, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:47:41,858 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.653e+02 3.106e+02 3.672e+02 8.833e+02, threshold=6.212e+02, percent-clipped=1.0 2023-05-17 07:47:43,242 INFO [finetune.py:992] (1/2) Epoch 17, batch 3250, loss[loss=0.1404, simple_loss=0.2248, pruned_loss=0.02803, over 11978.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2528, pruned_loss=0.03663, over 2385272.72 frames. ], batch size: 28, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:47:46,166 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2515, 4.9028, 5.2535, 5.1479, 4.3141, 4.5118, 4.5325, 4.9546], device='cuda:1'), covar=tensor([0.0949, 0.1153, 0.1092, 0.0944, 0.3561, 0.2196, 0.0869, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0713, 0.0630, 0.0647, 0.0862, 0.0767, 0.0572, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:47:48,598 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1733, 2.3586, 3.4251, 4.0957, 3.6821, 4.1337, 3.6649, 2.8618], device='cuda:1'), covar=tensor([0.0052, 0.0444, 0.0167, 0.0055, 0.0138, 0.0086, 0.0164, 0.0391], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0125, 0.0106, 0.0082, 0.0106, 0.0118, 0.0103, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 07:48:08,051 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3622, 3.4750, 3.1740, 3.0878, 2.7848, 2.6243, 3.4581, 2.2924], device='cuda:1'), covar=tensor([0.0444, 0.0136, 0.0213, 0.0247, 0.0447, 0.0432, 0.0147, 0.0547], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0166, 0.0173, 0.0197, 0.0209, 0.0206, 0.0181, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:48:19,324 INFO [finetune.py:992] (1/2) Epoch 17, batch 3300, loss[loss=0.1716, simple_loss=0.2617, pruned_loss=0.04079, over 12020.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2529, pruned_loss=0.03687, over 2384400.61 frames. ], batch size: 42, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:48:19,442 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299045.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:48:42,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 07:48:44,848 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1663, 5.9585, 5.5769, 5.5274, 6.0822, 5.4335, 5.5482, 5.5111], device='cuda:1'), covar=tensor([0.1489, 0.0985, 0.1209, 0.1878, 0.1034, 0.2168, 0.2092, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0505, 0.0409, 0.0457, 0.0477, 0.0445, 0.0410, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:48:54,067 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.781e+02 3.093e+02 3.666e+02 9.135e+02, threshold=6.186e+02, percent-clipped=3.0 2023-05-17 07:48:55,548 INFO [finetune.py:992] (1/2) Epoch 17, batch 3350, loss[loss=0.1567, simple_loss=0.2507, pruned_loss=0.03136, over 12145.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2535, pruned_loss=0.03726, over 2381461.65 frames. ], batch size: 36, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:48:56,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-05-17 07:49:05,225 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0872, 2.6531, 3.6805, 3.1040, 3.5362, 3.2742, 2.6802, 3.6327], device='cuda:1'), covar=tensor([0.0181, 0.0397, 0.0200, 0.0270, 0.0151, 0.0211, 0.0373, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0216, 0.0203, 0.0199, 0.0231, 0.0176, 0.0208, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:49:18,496 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299126.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:49:26,864 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299138.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:49:31,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-05-17 07:49:31,854 INFO [finetune.py:992] (1/2) Epoch 17, batch 3400, loss[loss=0.1414, simple_loss=0.2272, pruned_loss=0.02785, over 12191.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.253, pruned_loss=0.03721, over 2377647.81 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:49:51,769 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5973, 2.6855, 3.3417, 4.3670, 2.6494, 4.4094, 4.5864, 4.5888], device='cuda:1'), covar=tensor([0.0125, 0.1259, 0.0511, 0.0169, 0.1295, 0.0250, 0.0133, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0207, 0.0186, 0.0124, 0.0193, 0.0181, 0.0177, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:49:57,278 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4237, 5.2142, 5.3370, 5.3432, 4.9704, 5.0465, 4.7147, 5.3030], device='cuda:1'), covar=tensor([0.0695, 0.0615, 0.0950, 0.0656, 0.2104, 0.1452, 0.0626, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0715, 0.0631, 0.0652, 0.0864, 0.0770, 0.0574, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:50:00,760 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299186.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:50:00,872 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299186.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:50:01,650 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299187.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:50:04,442 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299191.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:50:05,617 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.674e+02 3.092e+02 3.690e+02 1.169e+03, threshold=6.184e+02, percent-clipped=3.0 2023-05-17 07:50:07,144 INFO [finetune.py:992] (1/2) Epoch 17, batch 3450, loss[loss=0.1686, simple_loss=0.2623, pruned_loss=0.03748, over 12196.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2525, pruned_loss=0.03682, over 2376550.60 frames. ], batch size: 35, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:50:08,849 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5889, 2.3576, 2.9893, 2.6431, 2.9327, 2.8657, 2.2727, 3.0277], device='cuda:1'), covar=tensor([0.0181, 0.0392, 0.0211, 0.0268, 0.0220, 0.0215, 0.0381, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0216, 0.0202, 0.0199, 0.0231, 0.0176, 0.0207, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:50:35,064 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5902, 3.8713, 3.4350, 3.4414, 3.2289, 3.1266, 3.8669, 2.4401], device='cuda:1'), covar=tensor([0.0453, 0.0147, 0.0218, 0.0222, 0.0331, 0.0332, 0.0115, 0.0532], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0167, 0.0174, 0.0197, 0.0209, 0.0206, 0.0181, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:50:36,354 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299234.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:50:42,806 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299243.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:50:44,035 INFO [finetune.py:992] (1/2) Epoch 17, batch 3500, loss[loss=0.1658, simple_loss=0.2574, pruned_loss=0.03708, over 12146.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2521, pruned_loss=0.03648, over 2375588.15 frames. ], batch size: 36, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:50:49,644 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299252.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:51:12,957 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299285.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 07:51:18,434 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.522e+02 2.567e+02 2.962e+02 3.764e+02 6.367e+02, threshold=5.924e+02, percent-clipped=1.0 2023-05-17 07:51:19,796 INFO [finetune.py:992] (1/2) Epoch 17, batch 3550, loss[loss=0.1695, simple_loss=0.262, pruned_loss=0.03849, over 12149.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2525, pruned_loss=0.0365, over 2373612.04 frames. ], batch size: 36, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:51:25,717 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0055, 6.0156, 5.7487, 5.2768, 5.1334, 5.8686, 5.4842, 5.2466], device='cuda:1'), covar=tensor([0.0900, 0.0914, 0.0783, 0.1596, 0.0865, 0.0783, 0.1728, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0653, 0.0582, 0.0537, 0.0656, 0.0437, 0.0754, 0.0809, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 07:51:26,534 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299304.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:51:40,247 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299323.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:51:47,279 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299333.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:51:49,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-05-17 07:51:55,724 INFO [finetune.py:992] (1/2) Epoch 17, batch 3600, loss[loss=0.1901, simple_loss=0.2738, pruned_loss=0.05319, over 11559.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.252, pruned_loss=0.03646, over 2377234.32 frames. ], batch size: 48, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:51:55,844 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299345.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:52:25,058 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299384.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:52:31,077 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.023e+02 2.632e+02 3.074e+02 3.664e+02 7.130e+02, threshold=6.147e+02, percent-clipped=1.0 2023-05-17 07:52:31,160 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299393.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:52:32,556 INFO [finetune.py:992] (1/2) Epoch 17, batch 3650, loss[loss=0.1549, simple_loss=0.2403, pruned_loss=0.03472, over 12027.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2522, pruned_loss=0.0365, over 2379345.87 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:52:51,399 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0321, 6.0715, 5.8180, 5.3310, 5.1366, 5.9283, 5.5262, 5.2360], device='cuda:1'), covar=tensor([0.0850, 0.0764, 0.0700, 0.1655, 0.0763, 0.0811, 0.1739, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0652, 0.0580, 0.0536, 0.0655, 0.0435, 0.0751, 0.0807, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 07:53:08,234 INFO [finetune.py:992] (1/2) Epoch 17, batch 3700, loss[loss=0.1869, simple_loss=0.2777, pruned_loss=0.04805, over 12372.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2531, pruned_loss=0.03661, over 2380701.96 frames. ], batch size: 38, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:53:32,973 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 07:53:34,724 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299482.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:53:39,116 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1875, 4.7411, 5.1807, 4.4552, 4.8079, 4.5559, 5.2264, 4.8042], device='cuda:1'), covar=tensor([0.0313, 0.0436, 0.0301, 0.0295, 0.0430, 0.0356, 0.0213, 0.0426], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0281, 0.0302, 0.0278, 0.0276, 0.0275, 0.0251, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:53:42,563 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.649e+02 2.993e+02 3.473e+02 6.479e+02, threshold=5.985e+02, percent-clipped=1.0 2023-05-17 07:53:44,058 INFO [finetune.py:992] (1/2) Epoch 17, batch 3750, loss[loss=0.1401, simple_loss=0.2329, pruned_loss=0.02368, over 12297.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2525, pruned_loss=0.03632, over 2389830.28 frames. ], batch size: 34, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:54:20,889 INFO [finetune.py:992] (1/2) Epoch 17, batch 3800, loss[loss=0.1572, simple_loss=0.2523, pruned_loss=0.0311, over 12352.00 frames. ], tot_loss[loss=0.1623, simple_loss=0.2523, pruned_loss=0.03611, over 2389632.50 frames. ], batch size: 35, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:54:22,398 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299547.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:54:55,135 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.650e+02 3.115e+02 3.689e+02 6.414e+02, threshold=6.230e+02, percent-clipped=3.0 2023-05-17 07:54:56,642 INFO [finetune.py:992] (1/2) Epoch 17, batch 3850, loss[loss=0.1619, simple_loss=0.2511, pruned_loss=0.03634, over 12173.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2526, pruned_loss=0.0364, over 2389572.34 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-05-17 07:54:59,477 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299599.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:55:32,513 INFO [finetune.py:992] (1/2) Epoch 17, batch 3900, loss[loss=0.1471, simple_loss=0.2329, pruned_loss=0.03059, over 12366.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.2524, pruned_loss=0.03625, over 2389822.82 frames. ], batch size: 30, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:55:44,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 07:55:57,585 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299679.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:56:03,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-05-17 07:56:08,118 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 2.706e+02 3.266e+02 3.999e+02 8.575e+02, threshold=6.532e+02, percent-clipped=2.0 2023-05-17 07:56:09,538 INFO [finetune.py:992] (1/2) Epoch 17, batch 3950, loss[loss=0.1729, simple_loss=0.2691, pruned_loss=0.03841, over 11542.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2529, pruned_loss=0.03654, over 2384895.10 frames. ], batch size: 48, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:56:17,704 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 07:56:44,711 INFO [finetune.py:992] (1/2) Epoch 17, batch 4000, loss[loss=0.176, simple_loss=0.2609, pruned_loss=0.04556, over 12065.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.253, pruned_loss=0.03662, over 2383030.91 frames. ], batch size: 42, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:57:01,698 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299769.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:57:02,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-05-17 07:57:10,664 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299782.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:57:18,495 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.576e+02 3.064e+02 3.696e+02 7.843e+02, threshold=6.128e+02, percent-clipped=2.0 2023-05-17 07:57:20,608 INFO [finetune.py:992] (1/2) Epoch 17, batch 4050, loss[loss=0.1622, simple_loss=0.2449, pruned_loss=0.03976, over 12322.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2534, pruned_loss=0.03702, over 2366481.59 frames. ], batch size: 31, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:57:22,994 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299798.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 07:57:47,055 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299830.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:57:47,186 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299830.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:57:57,498 INFO [finetune.py:992] (1/2) Epoch 17, batch 4100, loss[loss=0.1668, simple_loss=0.2654, pruned_loss=0.03405, over 12348.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2532, pruned_loss=0.03683, over 2377444.23 frames. ], batch size: 36, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:57:59,001 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299847.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:58:07,695 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299859.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 07:58:14,093 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3474, 5.1604, 5.2314, 5.2862, 4.8902, 4.9235, 4.6800, 5.2243], device='cuda:1'), covar=tensor([0.0713, 0.0611, 0.0928, 0.0602, 0.2066, 0.1510, 0.0640, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0723, 0.0644, 0.0668, 0.0878, 0.0781, 0.0583, 0.0504], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 07:58:31,673 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.564e+02 2.897e+02 3.446e+02 7.491e+02, threshold=5.795e+02, percent-clipped=2.0 2023-05-17 07:58:33,054 INFO [finetune.py:992] (1/2) Epoch 17, batch 4150, loss[loss=0.2302, simple_loss=0.3016, pruned_loss=0.07937, over 7513.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2532, pruned_loss=0.03681, over 2370365.53 frames. ], batch size: 97, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:58:33,127 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299895.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:58:35,921 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299899.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:58:56,840 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9155, 4.5362, 4.0794, 4.1930, 4.6583, 4.0322, 4.2163, 3.9045], device='cuda:1'), covar=tensor([0.1668, 0.1195, 0.1673, 0.1934, 0.1098, 0.2247, 0.1945, 0.1562], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0516, 0.0421, 0.0464, 0.0484, 0.0455, 0.0419, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 07:59:07,121 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-05-17 07:59:08,815 INFO [finetune.py:992] (1/2) Epoch 17, batch 4200, loss[loss=0.1286, simple_loss=0.2103, pruned_loss=0.02345, over 12186.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.253, pruned_loss=0.03638, over 2372207.14 frames. ], batch size: 29, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 07:59:11,031 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299947.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:59:29,828 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6925, 3.0026, 4.5772, 4.7880, 2.8702, 2.5914, 2.9473, 2.1947], device='cuda:1'), covar=tensor([0.1806, 0.3056, 0.0478, 0.0436, 0.1417, 0.2725, 0.2985, 0.4424], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0397, 0.0279, 0.0306, 0.0282, 0.0324, 0.0405, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:59:31,202 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6396, 2.4691, 3.2347, 4.5808, 2.5129, 4.5442, 4.6646, 4.6960], device='cuda:1'), covar=tensor([0.0156, 0.1365, 0.0496, 0.0148, 0.1335, 0.0216, 0.0145, 0.0110], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0207, 0.0186, 0.0122, 0.0192, 0.0181, 0.0177, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 07:59:34,758 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299979.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 07:59:44,515 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.549e+02 3.017e+02 3.536e+02 5.011e+02, threshold=6.035e+02, percent-clipped=0.0 2023-05-17 07:59:45,975 INFO [finetune.py:992] (1/2) Epoch 17, batch 4250, loss[loss=0.1526, simple_loss=0.2382, pruned_loss=0.03345, over 12126.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2527, pruned_loss=0.03636, over 2371300.59 frames. ], batch size: 33, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:00:12,173 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300027.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:00:24,949 INFO [finetune.py:992] (1/2) Epoch 17, batch 4300, loss[loss=0.1309, simple_loss=0.2068, pruned_loss=0.02752, over 11738.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2522, pruned_loss=0.03643, over 2368088.35 frames. ], batch size: 26, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:00:44,669 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300073.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:00:59,087 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.688e+02 3.373e+02 4.196e+02 1.008e+03, threshold=6.746e+02, percent-clipped=3.0 2023-05-17 08:01:00,502 INFO [finetune.py:992] (1/2) Epoch 17, batch 4350, loss[loss=0.1463, simple_loss=0.2358, pruned_loss=0.02839, over 12142.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2537, pruned_loss=0.03701, over 2373228.53 frames. ], batch size: 30, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:01:15,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-05-17 08:01:22,594 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300125.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:01:29,242 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300134.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:01:32,145 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0876, 4.6786, 4.7549, 5.0005, 4.7617, 4.9723, 4.8437, 2.5228], device='cuda:1'), covar=tensor([0.0112, 0.0081, 0.0114, 0.0058, 0.0054, 0.0108, 0.0093, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0073, 0.0083, 0.0088, 0.0078, 0.0064, 0.0099, 0.0086, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:01:36,912 INFO [finetune.py:992] (1/2) Epoch 17, batch 4400, loss[loss=0.1638, simple_loss=0.2617, pruned_loss=0.03297, over 12097.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2536, pruned_loss=0.0369, over 2367679.49 frames. ], batch size: 38, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:01:43,526 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300154.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:02:11,063 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.546e+02 3.053e+02 3.701e+02 7.801e+02, threshold=6.107e+02, percent-clipped=1.0 2023-05-17 08:02:12,541 INFO [finetune.py:992] (1/2) Epoch 17, batch 4450, loss[loss=0.1537, simple_loss=0.2548, pruned_loss=0.02629, over 12335.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2546, pruned_loss=0.03707, over 2369562.38 frames. ], batch size: 36, lr: 3.32e-03, grad_scale: 32.0 2023-05-17 08:02:12,725 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1324, 5.0012, 4.9653, 5.0277, 4.6961, 5.1281, 5.1374, 5.3469], device='cuda:1'), covar=tensor([0.0311, 0.0173, 0.0217, 0.0364, 0.0778, 0.0313, 0.0159, 0.0173], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0208, 0.0200, 0.0254, 0.0250, 0.0228, 0.0182, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 08:02:32,119 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5698, 2.5727, 3.1863, 4.4518, 2.5053, 4.4275, 4.6390, 4.6057], device='cuda:1'), covar=tensor([0.0187, 0.1371, 0.0577, 0.0191, 0.1407, 0.0263, 0.0139, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0206, 0.0185, 0.0122, 0.0191, 0.0181, 0.0177, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:02:48,861 INFO [finetune.py:992] (1/2) Epoch 17, batch 4500, loss[loss=0.1989, simple_loss=0.2789, pruned_loss=0.05945, over 7754.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2535, pruned_loss=0.03695, over 2364605.71 frames. ], batch size: 98, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:03:09,988 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6263, 2.3932, 3.3285, 4.5500, 2.3988, 4.5026, 4.6621, 4.7555], device='cuda:1'), covar=tensor([0.0157, 0.1477, 0.0521, 0.0165, 0.1424, 0.0195, 0.0128, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0206, 0.0186, 0.0122, 0.0191, 0.0182, 0.0178, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:03:10,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-17 08:03:13,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.13 vs. limit=5.0 2023-05-17 08:03:24,016 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.601e+02 3.141e+02 3.922e+02 6.717e+02, threshold=6.282e+02, percent-clipped=1.0 2023-05-17 08:03:24,751 INFO [finetune.py:992] (1/2) Epoch 17, batch 4550, loss[loss=0.177, simple_loss=0.2727, pruned_loss=0.0407, over 12058.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2536, pruned_loss=0.03701, over 2364715.62 frames. ], batch size: 40, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:03:40,980 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3426, 4.9976, 5.3019, 4.6040, 4.9612, 4.6753, 5.3777, 4.9561], device='cuda:1'), covar=tensor([0.0279, 0.0364, 0.0323, 0.0279, 0.0419, 0.0327, 0.0211, 0.0334], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0282, 0.0306, 0.0277, 0.0278, 0.0275, 0.0251, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:04:00,671 INFO [finetune.py:992] (1/2) Epoch 17, batch 4600, loss[loss=0.2006, simple_loss=0.2963, pruned_loss=0.05245, over 11739.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.254, pruned_loss=0.03711, over 2361924.58 frames. ], batch size: 44, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:04:04,499 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0080, 2.3479, 2.9942, 3.9339, 2.2411, 3.9748, 3.9917, 4.1218], device='cuda:1'), covar=tensor([0.0168, 0.1331, 0.0532, 0.0155, 0.1354, 0.0302, 0.0194, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0206, 0.0185, 0.0122, 0.0191, 0.0181, 0.0178, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:04:36,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-05-17 08:04:36,372 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.656e+02 3.013e+02 3.569e+02 7.028e+02, threshold=6.027e+02, percent-clipped=3.0 2023-05-17 08:04:37,146 INFO [finetune.py:992] (1/2) Epoch 17, batch 4650, loss[loss=0.147, simple_loss=0.2292, pruned_loss=0.03246, over 12290.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2526, pruned_loss=0.0367, over 2370616.06 frames. ], batch size: 28, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:04:59,099 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300425.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:05:01,688 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300429.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:05:12,814 INFO [finetune.py:992] (1/2) Epoch 17, batch 4700, loss[loss=0.181, simple_loss=0.2751, pruned_loss=0.04346, over 12345.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.253, pruned_loss=0.03702, over 2368063.22 frames. ], batch size: 36, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:05:13,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-05-17 08:05:19,382 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300454.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 08:05:32,653 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300473.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:05:33,496 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3969, 2.6361, 3.6101, 4.3077, 3.7476, 4.3906, 3.8036, 3.1183], device='cuda:1'), covar=tensor([0.0044, 0.0386, 0.0144, 0.0050, 0.0125, 0.0073, 0.0123, 0.0360], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0126, 0.0107, 0.0082, 0.0107, 0.0119, 0.0104, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:05:47,514 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.583e+02 3.059e+02 3.722e+02 6.746e+02, threshold=6.118e+02, percent-clipped=2.0 2023-05-17 08:05:48,236 INFO [finetune.py:992] (1/2) Epoch 17, batch 4750, loss[loss=0.1559, simple_loss=0.2492, pruned_loss=0.03135, over 12248.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2531, pruned_loss=0.03677, over 2373422.25 frames. ], batch size: 32, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:05:53,428 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300502.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 08:06:24,889 INFO [finetune.py:992] (1/2) Epoch 17, batch 4800, loss[loss=0.1596, simple_loss=0.2516, pruned_loss=0.03382, over 12155.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2543, pruned_loss=0.03715, over 2373416.19 frames. ], batch size: 34, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:06:34,289 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7372, 2.9193, 4.6838, 4.8378, 2.9098, 2.5358, 3.0566, 2.2045], device='cuda:1'), covar=tensor([0.1800, 0.3023, 0.0451, 0.0453, 0.1377, 0.2749, 0.2863, 0.4373], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0395, 0.0278, 0.0305, 0.0281, 0.0322, 0.0402, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:07:00,332 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.535e+02 3.170e+02 3.712e+02 6.173e+02, threshold=6.341e+02, percent-clipped=1.0 2023-05-17 08:07:01,078 INFO [finetune.py:992] (1/2) Epoch 17, batch 4850, loss[loss=0.153, simple_loss=0.2487, pruned_loss=0.02862, over 12366.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.254, pruned_loss=0.03676, over 2379483.18 frames. ], batch size: 35, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:07:17,300 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300617.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:07:22,309 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7367, 2.8302, 3.4630, 4.6403, 2.7792, 4.4982, 4.6752, 4.8142], device='cuda:1'), covar=tensor([0.0123, 0.1203, 0.0421, 0.0154, 0.1146, 0.0261, 0.0136, 0.0108], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0203, 0.0183, 0.0121, 0.0190, 0.0180, 0.0176, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:07:37,153 INFO [finetune.py:992] (1/2) Epoch 17, batch 4900, loss[loss=0.184, simple_loss=0.2729, pruned_loss=0.04757, over 12146.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2541, pruned_loss=0.03695, over 2362851.25 frames. ], batch size: 36, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:08:00,422 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9439, 3.5159, 5.3332, 2.8218, 2.9972, 3.9287, 3.4260, 3.9287], device='cuda:1'), covar=tensor([0.0379, 0.1082, 0.0281, 0.1184, 0.1915, 0.1587, 0.1274, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0240, 0.0260, 0.0187, 0.0242, 0.0299, 0.0228, 0.0270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:08:01,822 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300678.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:08:12,889 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.566e+02 3.041e+02 3.803e+02 9.854e+02, threshold=6.082e+02, percent-clipped=2.0 2023-05-17 08:08:13,650 INFO [finetune.py:992] (1/2) Epoch 17, batch 4950, loss[loss=0.1604, simple_loss=0.2558, pruned_loss=0.03253, over 12364.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2535, pruned_loss=0.03704, over 2355481.69 frames. ], batch size: 35, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:08:38,134 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300729.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:08:49,240 INFO [finetune.py:992] (1/2) Epoch 17, batch 5000, loss[loss=0.1546, simple_loss=0.2408, pruned_loss=0.03421, over 12351.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.254, pruned_loss=0.03749, over 2358633.79 frames. ], batch size: 30, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:09:10,585 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6400, 4.5149, 4.5737, 4.6456, 4.3507, 4.4190, 4.1809, 4.5400], device='cuda:1'), covar=tensor([0.0779, 0.0613, 0.1072, 0.0551, 0.1663, 0.1283, 0.0586, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0563, 0.0720, 0.0641, 0.0659, 0.0871, 0.0777, 0.0579, 0.0499], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:09:11,952 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300777.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:09:24,028 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.799e+02 3.234e+02 3.956e+02 8.201e+02, threshold=6.469e+02, percent-clipped=3.0 2023-05-17 08:09:24,774 INFO [finetune.py:992] (1/2) Epoch 17, batch 5050, loss[loss=0.1675, simple_loss=0.2613, pruned_loss=0.03681, over 12306.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2539, pruned_loss=0.03727, over 2361584.97 frames. ], batch size: 34, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:09:31,007 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300802.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:09:33,181 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7520, 3.8719, 3.4656, 3.3229, 3.1850, 2.9713, 3.8833, 2.5487], device='cuda:1'), covar=tensor([0.0399, 0.0139, 0.0200, 0.0217, 0.0364, 0.0363, 0.0124, 0.0479], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0165, 0.0172, 0.0195, 0.0207, 0.0203, 0.0179, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:09:57,658 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4368, 3.6094, 3.2320, 3.6747, 3.4335, 2.6857, 3.2095, 2.8101], device='cuda:1'), covar=tensor([0.0961, 0.1077, 0.1632, 0.0870, 0.1351, 0.1614, 0.1392, 0.2954], device='cuda:1'), in_proj_covar=tensor([0.0315, 0.0385, 0.0367, 0.0334, 0.0377, 0.0279, 0.0353, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:10:02,320 INFO [finetune.py:992] (1/2) Epoch 17, batch 5100, loss[loss=0.164, simple_loss=0.2569, pruned_loss=0.0356, over 12354.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2539, pruned_loss=0.03714, over 2360612.21 frames. ], batch size: 35, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:10:15,304 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300863.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:10:37,316 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.054e+02 2.723e+02 3.013e+02 3.762e+02 5.575e+02, threshold=6.027e+02, percent-clipped=0.0 2023-05-17 08:10:37,910 INFO [finetune.py:992] (1/2) Epoch 17, batch 5150, loss[loss=0.177, simple_loss=0.2681, pruned_loss=0.04293, over 12256.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2532, pruned_loss=0.03681, over 2367437.16 frames. ], batch size: 32, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:10:39,501 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9718, 4.7993, 4.8044, 4.9279, 3.8554, 5.0435, 4.9742, 5.1356], device='cuda:1'), covar=tensor([0.0270, 0.0221, 0.0242, 0.0386, 0.1269, 0.0477, 0.0214, 0.0251], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0208, 0.0201, 0.0255, 0.0252, 0.0229, 0.0184, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 08:11:07,636 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.4075, 4.7887, 3.1714, 2.8180, 4.1034, 2.6531, 3.9964, 3.2583], device='cuda:1'), covar=tensor([0.0758, 0.0534, 0.1104, 0.1534, 0.0302, 0.1476, 0.0578, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0261, 0.0178, 0.0202, 0.0142, 0.0185, 0.0202, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:11:13,885 INFO [finetune.py:992] (1/2) Epoch 17, batch 5200, loss[loss=0.1868, simple_loss=0.2766, pruned_loss=0.04847, over 11193.00 frames. ], tot_loss[loss=0.1623, simple_loss=0.252, pruned_loss=0.03632, over 2374815.60 frames. ], batch size: 55, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:11:28,287 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.5012, 5.3175, 5.4176, 5.5074, 5.0911, 5.1586, 4.8585, 5.3706], device='cuda:1'), covar=tensor([0.0747, 0.0661, 0.0916, 0.0567, 0.2079, 0.1393, 0.0567, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0723, 0.0643, 0.0659, 0.0874, 0.0781, 0.0580, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:11:34,749 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300973.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:11:40,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-05-17 08:11:50,239 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.582e+02 2.981e+02 3.629e+02 7.540e+02, threshold=5.961e+02, percent-clipped=3.0 2023-05-17 08:11:50,987 INFO [finetune.py:992] (1/2) Epoch 17, batch 5250, loss[loss=0.1875, simple_loss=0.2812, pruned_loss=0.04688, over 12271.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.2523, pruned_loss=0.03624, over 2370349.17 frames. ], batch size: 37, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:12:26,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-05-17 08:12:26,591 INFO [finetune.py:992] (1/2) Epoch 17, batch 5300, loss[loss=0.1697, simple_loss=0.2619, pruned_loss=0.03877, over 12192.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2526, pruned_loss=0.03679, over 2364986.34 frames. ], batch size: 35, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:13:01,960 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 2.644e+02 3.162e+02 3.755e+02 1.139e+03, threshold=6.323e+02, percent-clipped=5.0 2023-05-17 08:13:02,684 INFO [finetune.py:992] (1/2) Epoch 17, batch 5350, loss[loss=0.1722, simple_loss=0.2639, pruned_loss=0.04024, over 11643.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2526, pruned_loss=0.03661, over 2368462.02 frames. ], batch size: 48, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:13:10,915 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0638, 2.4162, 3.6641, 2.9619, 3.5078, 3.1184, 2.5295, 3.5895], device='cuda:1'), covar=tensor([0.0156, 0.0403, 0.0167, 0.0292, 0.0144, 0.0232, 0.0439, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0214, 0.0201, 0.0198, 0.0230, 0.0176, 0.0207, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:13:35,583 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301140.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:13:39,025 INFO [finetune.py:992] (1/2) Epoch 17, batch 5400, loss[loss=0.1228, simple_loss=0.2081, pruned_loss=0.01876, over 11996.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2529, pruned_loss=0.03689, over 2364164.91 frames. ], batch size: 28, lr: 3.32e-03, grad_scale: 16.0 2023-05-17 08:13:44,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6449, 2.7907, 3.8062, 4.6316, 3.9708, 4.7400, 3.8705, 3.4600], device='cuda:1'), covar=tensor([0.0040, 0.0405, 0.0150, 0.0053, 0.0133, 0.0069, 0.0159, 0.0329], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0127, 0.0107, 0.0082, 0.0108, 0.0119, 0.0105, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:13:47,181 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1579, 2.6240, 3.7735, 3.1426, 3.5621, 3.3146, 2.7467, 3.6925], device='cuda:1'), covar=tensor([0.0163, 0.0355, 0.0196, 0.0269, 0.0170, 0.0230, 0.0361, 0.0156], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0215, 0.0202, 0.0199, 0.0231, 0.0176, 0.0207, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:13:48,349 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301158.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:13:55,064 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0105, 2.4276, 3.6516, 2.9342, 3.3977, 3.1583, 2.5241, 3.5166], device='cuda:1'), covar=tensor([0.0173, 0.0402, 0.0160, 0.0296, 0.0172, 0.0206, 0.0414, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0216, 0.0203, 0.0199, 0.0231, 0.0177, 0.0208, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:14:00,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=3.88 vs. limit=5.0 2023-05-17 08:14:13,685 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.605e+02 3.084e+02 3.605e+02 6.767e+02, threshold=6.167e+02, percent-clipped=2.0 2023-05-17 08:14:14,362 INFO [finetune.py:992] (1/2) Epoch 17, batch 5450, loss[loss=0.1973, simple_loss=0.2857, pruned_loss=0.05451, over 12040.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2531, pruned_loss=0.03706, over 2361452.07 frames. ], batch size: 42, lr: 3.31e-03, grad_scale: 16.0 2023-05-17 08:14:19,093 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301201.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:14:50,774 INFO [finetune.py:992] (1/2) Epoch 17, batch 5500, loss[loss=0.1641, simple_loss=0.2523, pruned_loss=0.03796, over 10568.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2522, pruned_loss=0.03655, over 2364077.86 frames. ], batch size: 68, lr: 3.31e-03, grad_scale: 16.0 2023-05-17 08:14:52,963 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3859, 5.2042, 5.3235, 5.3760, 4.9804, 5.0502, 4.7316, 5.2756], device='cuda:1'), covar=tensor([0.0725, 0.0600, 0.0863, 0.0524, 0.1795, 0.1266, 0.0566, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0722, 0.0642, 0.0656, 0.0872, 0.0776, 0.0579, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:15:10,722 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301273.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:15:27,089 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.504e+02 2.789e+02 3.471e+02 7.035e+02, threshold=5.578e+02, percent-clipped=1.0 2023-05-17 08:15:27,108 INFO [finetune.py:992] (1/2) Epoch 17, batch 5550, loss[loss=0.1686, simple_loss=0.2574, pruned_loss=0.03995, over 12342.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2521, pruned_loss=0.03672, over 2373869.27 frames. ], batch size: 36, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:15:42,542 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4397, 3.5898, 3.1856, 3.0619, 2.8083, 2.6401, 3.5325, 2.3552], device='cuda:1'), covar=tensor([0.0461, 0.0134, 0.0242, 0.0275, 0.0452, 0.0473, 0.0165, 0.0555], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0166, 0.0173, 0.0197, 0.0207, 0.0204, 0.0180, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:15:45,944 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301321.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:15:53,886 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.1232, 2.5731, 3.7538, 3.0941, 3.5009, 3.2544, 2.6230, 3.6053], device='cuda:1'), covar=tensor([0.0174, 0.0381, 0.0200, 0.0265, 0.0157, 0.0218, 0.0395, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0216, 0.0203, 0.0200, 0.0231, 0.0177, 0.0208, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:16:02,990 INFO [finetune.py:992] (1/2) Epoch 17, batch 5600, loss[loss=0.1418, simple_loss=0.225, pruned_loss=0.02934, over 12189.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2524, pruned_loss=0.03681, over 2373610.09 frames. ], batch size: 29, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:16:24,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-05-17 08:16:38,972 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.655e+02 3.111e+02 3.833e+02 7.587e+02, threshold=6.222e+02, percent-clipped=4.0 2023-05-17 08:16:38,991 INFO [finetune.py:992] (1/2) Epoch 17, batch 5650, loss[loss=0.1737, simple_loss=0.2763, pruned_loss=0.03555, over 10371.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2535, pruned_loss=0.03698, over 2375420.56 frames. ], batch size: 68, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:16:54,498 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.2416, 6.2170, 5.9951, 5.5314, 5.3513, 6.1521, 5.7494, 5.4863], device='cuda:1'), covar=tensor([0.0736, 0.1007, 0.0706, 0.1684, 0.0649, 0.0799, 0.1557, 0.0994], device='cuda:1'), in_proj_covar=tensor([0.0653, 0.0586, 0.0536, 0.0660, 0.0439, 0.0755, 0.0809, 0.0588], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0003], device='cuda:1') 2023-05-17 08:17:05,246 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301430.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:17:14,202 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-05-17 08:17:15,908 INFO [finetune.py:992] (1/2) Epoch 17, batch 5700, loss[loss=0.2234, simple_loss=0.2995, pruned_loss=0.07364, over 8117.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2536, pruned_loss=0.037, over 2364648.56 frames. ], batch size: 98, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:17:25,541 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301458.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:17:38,396 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7095, 4.3891, 4.4805, 4.6145, 4.4188, 4.6028, 4.4578, 2.4102], device='cuda:1'), covar=tensor([0.0101, 0.0080, 0.0115, 0.0069, 0.0064, 0.0115, 0.0104, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0087, 0.0076, 0.0063, 0.0097, 0.0085, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:17:45,976 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 08:17:47,754 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4587, 3.0198, 4.9231, 2.7570, 2.8435, 3.8694, 3.1299, 3.6490], device='cuda:1'), covar=tensor([0.0574, 0.1341, 0.0401, 0.1228, 0.1989, 0.1466, 0.1568, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0241, 0.0261, 0.0188, 0.0243, 0.0300, 0.0230, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:17:49,044 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301491.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:17:51,657 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.682e+02 3.114e+02 3.562e+02 5.424e+02, threshold=6.228e+02, percent-clipped=0.0 2023-05-17 08:17:51,676 INFO [finetune.py:992] (1/2) Epoch 17, batch 5750, loss[loss=0.1579, simple_loss=0.2549, pruned_loss=0.03042, over 12282.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.254, pruned_loss=0.03732, over 2366684.14 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:17:52,446 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301496.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:17:59,735 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301506.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:18:19,253 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4413, 4.3223, 4.2987, 4.5323, 3.0916, 4.0276, 2.6641, 4.3530], device='cuda:1'), covar=tensor([0.1506, 0.0612, 0.0797, 0.0791, 0.1208, 0.0613, 0.1857, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0273, 0.0302, 0.0365, 0.0245, 0.0248, 0.0266, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:18:28,117 INFO [finetune.py:992] (1/2) Epoch 17, batch 5800, loss[loss=0.1516, simple_loss=0.2453, pruned_loss=0.0289, over 12106.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2545, pruned_loss=0.03762, over 2364665.25 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:18:42,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 08:18:43,181 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5319, 2.6426, 3.6722, 4.4939, 3.8122, 4.5294, 3.7576, 3.3875], device='cuda:1'), covar=tensor([0.0042, 0.0458, 0.0169, 0.0047, 0.0157, 0.0090, 0.0168, 0.0361], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0130, 0.0110, 0.0084, 0.0110, 0.0121, 0.0108, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:19:04,502 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.491e+02 2.825e+02 3.428e+02 1.107e+03, threshold=5.650e+02, percent-clipped=1.0 2023-05-17 08:19:04,521 INFO [finetune.py:992] (1/2) Epoch 17, batch 5850, loss[loss=0.1519, simple_loss=0.2434, pruned_loss=0.03018, over 11831.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2533, pruned_loss=0.03702, over 2365686.20 frames. ], batch size: 44, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:19:12,989 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2841, 4.0300, 4.1534, 4.3514, 3.0626, 3.9168, 2.6134, 4.1462], device='cuda:1'), covar=tensor([0.1588, 0.0736, 0.0857, 0.0737, 0.1179, 0.0653, 0.1889, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0272, 0.0301, 0.0364, 0.0244, 0.0247, 0.0265, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:19:27,819 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1835, 3.8370, 3.9474, 4.2706, 2.8595, 3.7736, 2.4263, 3.9146], device='cuda:1'), covar=tensor([0.1811, 0.0890, 0.0962, 0.0708, 0.1377, 0.0741, 0.2167, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0271, 0.0300, 0.0363, 0.0244, 0.0246, 0.0264, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:19:40,610 INFO [finetune.py:992] (1/2) Epoch 17, batch 5900, loss[loss=0.1728, simple_loss=0.2592, pruned_loss=0.04325, over 12112.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2538, pruned_loss=0.0373, over 2365300.77 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:19:53,589 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1912, 2.3684, 3.1819, 4.0891, 2.0852, 4.1142, 4.1628, 4.3028], device='cuda:1'), covar=tensor([0.0188, 0.1360, 0.0478, 0.0189, 0.1504, 0.0310, 0.0218, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0203, 0.0184, 0.0122, 0.0190, 0.0181, 0.0179, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:20:05,634 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8651, 4.5894, 4.1822, 4.0684, 4.6951, 4.0422, 4.1969, 4.0130], device='cuda:1'), covar=tensor([0.1775, 0.1193, 0.1658, 0.2397, 0.1183, 0.2185, 0.2166, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0508, 0.0413, 0.0457, 0.0476, 0.0449, 0.0412, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:20:08,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.50 vs. limit=5.0 2023-05-17 08:20:16,723 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.751e+02 3.117e+02 3.766e+02 7.477e+02, threshold=6.233e+02, percent-clipped=4.0 2023-05-17 08:20:16,742 INFO [finetune.py:992] (1/2) Epoch 17, batch 5950, loss[loss=0.1546, simple_loss=0.2384, pruned_loss=0.03537, over 12350.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2541, pruned_loss=0.03769, over 2359833.57 frames. ], batch size: 31, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:20:17,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-05-17 08:20:17,233 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 08:20:51,998 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301743.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:20:53,231 INFO [finetune.py:992] (1/2) Epoch 17, batch 6000, loss[loss=0.1788, simple_loss=0.2706, pruned_loss=0.04346, over 12349.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2546, pruned_loss=0.03813, over 2350595.35 frames. ], batch size: 36, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:20:53,231 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 08:21:05,346 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0910, 3.4901, 3.3153, 2.9808, 2.9292, 2.8915, 2.0018, 2.1481], device='cuda:1'), covar=tensor([0.0511, 0.0109, 0.0116, 0.0218, 0.0283, 0.0239, 0.0400, 0.0557], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0165, 0.0171, 0.0196, 0.0206, 0.0204, 0.0179, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:21:12,096 INFO [finetune.py:1026] (1/2) Epoch 17, validation: loss=0.3119, simple_loss=0.3879, pruned_loss=0.118, over 1020973.00 frames. 2023-05-17 08:21:12,097 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 08:21:32,496 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4867, 2.4417, 3.3289, 4.3904, 2.5602, 4.4127, 4.5253, 4.5520], device='cuda:1'), covar=tensor([0.0158, 0.1328, 0.0472, 0.0170, 0.1201, 0.0262, 0.0151, 0.0150], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0203, 0.0183, 0.0122, 0.0189, 0.0181, 0.0179, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:21:42,106 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301786.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:21:48,305 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.650e+02 3.139e+02 3.810e+02 6.475e+02, threshold=6.278e+02, percent-clipped=1.0 2023-05-17 08:21:48,324 INFO [finetune.py:992] (1/2) Epoch 17, batch 6050, loss[loss=0.1539, simple_loss=0.2458, pruned_loss=0.03101, over 12348.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2546, pruned_loss=0.03786, over 2359941.26 frames. ], batch size: 31, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:21:48,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-05-17 08:21:49,170 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301796.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:21:50,686 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3536, 4.7600, 3.0250, 2.6002, 4.0874, 2.8075, 4.0438, 3.4126], device='cuda:1'), covar=tensor([0.0714, 0.0634, 0.1178, 0.1511, 0.0350, 0.1244, 0.0507, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0264, 0.0178, 0.0204, 0.0143, 0.0186, 0.0204, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:21:51,399 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6485, 2.9682, 3.7198, 4.5781, 3.9161, 4.5224, 3.8125, 3.4454], device='cuda:1'), covar=tensor([0.0031, 0.0374, 0.0167, 0.0040, 0.0128, 0.0086, 0.0184, 0.0349], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0128, 0.0109, 0.0083, 0.0109, 0.0120, 0.0106, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:21:55,386 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301804.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:22:23,749 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301844.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:22:24,378 INFO [finetune.py:992] (1/2) Epoch 17, batch 6100, loss[loss=0.1513, simple_loss=0.239, pruned_loss=0.03181, over 12019.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03768, over 2355804.57 frames. ], batch size: 28, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:23:00,275 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.658e+02 2.989e+02 3.684e+02 9.440e+02, threshold=5.978e+02, percent-clipped=2.0 2023-05-17 08:23:00,295 INFO [finetune.py:992] (1/2) Epoch 17, batch 6150, loss[loss=0.1803, simple_loss=0.2723, pruned_loss=0.04416, over 12277.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2538, pruned_loss=0.03717, over 2366819.92 frames. ], batch size: 37, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:23:03,352 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301899.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:23:07,580 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301905.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:23:08,982 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3069, 4.7551, 3.0927, 2.6282, 4.2011, 2.8669, 4.0901, 3.3433], device='cuda:1'), covar=tensor([0.0797, 0.0647, 0.1116, 0.1548, 0.0298, 0.1235, 0.0496, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0265, 0.0179, 0.0205, 0.0144, 0.0187, 0.0205, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:23:29,002 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4766, 5.3205, 5.4139, 5.4301, 4.9113, 4.9888, 4.9162, 5.2101], device='cuda:1'), covar=tensor([0.1098, 0.0920, 0.1095, 0.1092, 0.3429, 0.2101, 0.0786, 0.2204], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0727, 0.0643, 0.0661, 0.0878, 0.0781, 0.0586, 0.0504], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-05-17 08:23:36,864 INFO [finetune.py:992] (1/2) Epoch 17, batch 6200, loss[loss=0.1386, simple_loss=0.2309, pruned_loss=0.02313, over 12167.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2539, pruned_loss=0.03714, over 2363890.93 frames. ], batch size: 31, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:23:48,297 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301960.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 08:23:49,612 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0678, 6.0633, 5.8342, 5.3588, 5.2067, 5.9555, 5.5546, 5.3764], device='cuda:1'), covar=tensor([0.0829, 0.0969, 0.0731, 0.1544, 0.0664, 0.0778, 0.1621, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0657, 0.0587, 0.0539, 0.0666, 0.0443, 0.0759, 0.0813, 0.0588], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 08:23:52,531 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301966.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:23:53,962 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0473, 2.4744, 3.6974, 3.0724, 3.4311, 3.2912, 2.5377, 3.5184], device='cuda:1'), covar=tensor([0.0154, 0.0406, 0.0168, 0.0261, 0.0184, 0.0194, 0.0414, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0216, 0.0203, 0.0198, 0.0231, 0.0176, 0.0207, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:24:12,864 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.627e+02 3.010e+02 3.487e+02 9.928e+02, threshold=6.019e+02, percent-clipped=2.0 2023-05-17 08:24:12,884 INFO [finetune.py:992] (1/2) Epoch 17, batch 6250, loss[loss=0.1517, simple_loss=0.2341, pruned_loss=0.03466, over 12183.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2542, pruned_loss=0.03699, over 2363766.13 frames. ], batch size: 31, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:24:33,673 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 08:24:51,787 INFO [finetune.py:992] (1/2) Epoch 17, batch 6300, loss[loss=0.1701, simple_loss=0.2627, pruned_loss=0.03876, over 12086.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2539, pruned_loss=0.03681, over 2372570.24 frames. ], batch size: 42, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:25:10,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 2023-05-17 08:25:21,811 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302086.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:25:27,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.24 vs. limit=5.0 2023-05-17 08:25:28,043 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.542e+02 3.035e+02 3.884e+02 6.196e+02, threshold=6.069e+02, percent-clipped=2.0 2023-05-17 08:25:28,062 INFO [finetune.py:992] (1/2) Epoch 17, batch 6350, loss[loss=0.1598, simple_loss=0.2553, pruned_loss=0.03219, over 12316.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.254, pruned_loss=0.0366, over 2374798.98 frames. ], batch size: 34, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:25:30,847 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302099.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:25:33,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-05-17 08:25:49,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5353, 3.7439, 3.2811, 3.1682, 2.9709, 2.8031, 3.6475, 2.5083], device='cuda:1'), covar=tensor([0.0445, 0.0132, 0.0251, 0.0211, 0.0402, 0.0370, 0.0150, 0.0470], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0167, 0.0173, 0.0199, 0.0208, 0.0206, 0.0180, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:25:56,210 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302134.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:26:04,108 INFO [finetune.py:992] (1/2) Epoch 17, batch 6400, loss[loss=0.1485, simple_loss=0.2504, pruned_loss=0.0233, over 12359.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2539, pruned_loss=0.03652, over 2382068.66 frames. ], batch size: 38, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:26:29,147 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5612, 2.7536, 3.7162, 4.5158, 3.9511, 4.5753, 3.8187, 3.1471], device='cuda:1'), covar=tensor([0.0038, 0.0431, 0.0147, 0.0050, 0.0128, 0.0074, 0.0179, 0.0399], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0127, 0.0108, 0.0083, 0.0108, 0.0120, 0.0106, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:26:39,312 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.608e+02 3.103e+02 3.901e+02 7.929e+02, threshold=6.206e+02, percent-clipped=1.0 2023-05-17 08:26:39,331 INFO [finetune.py:992] (1/2) Epoch 17, batch 6450, loss[loss=0.1573, simple_loss=0.2465, pruned_loss=0.03403, over 12115.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2537, pruned_loss=0.03659, over 2379494.78 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:26:43,372 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302200.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:27:16,170 INFO [finetune.py:992] (1/2) Epoch 17, batch 6500, loss[loss=0.1532, simple_loss=0.2264, pruned_loss=0.04003, over 11765.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2537, pruned_loss=0.03669, over 2378795.97 frames. ], batch size: 26, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:27:21,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 08:27:24,195 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302255.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:27:28,452 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302261.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:27:28,620 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302261.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:27:52,639 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.722e+02 3.140e+02 3.968e+02 8.913e+02, threshold=6.279e+02, percent-clipped=2.0 2023-05-17 08:27:52,659 INFO [finetune.py:992] (1/2) Epoch 17, batch 6550, loss[loss=0.1902, simple_loss=0.2783, pruned_loss=0.05109, over 12082.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.254, pruned_loss=0.03688, over 2375945.98 frames. ], batch size: 42, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:27:53,497 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302296.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:28:00,374 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3346, 2.3502, 3.6226, 4.3292, 3.8078, 4.2439, 3.7388, 2.8547], device='cuda:1'), covar=tensor([0.0046, 0.0494, 0.0148, 0.0045, 0.0126, 0.0102, 0.0186, 0.0434], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0127, 0.0108, 0.0083, 0.0108, 0.0119, 0.0106, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:28:02,418 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0374, 5.9719, 5.7606, 5.2189, 5.1517, 5.9149, 5.4850, 5.2806], device='cuda:1'), covar=tensor([0.0710, 0.1013, 0.0697, 0.1784, 0.0666, 0.0722, 0.1519, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0654, 0.0584, 0.0535, 0.0661, 0.0440, 0.0753, 0.0813, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-17 08:28:05,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-17 08:28:12,452 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2776, 2.6250, 3.5744, 4.1849, 3.7482, 4.1827, 3.6441, 3.0246], device='cuda:1'), covar=tensor([0.0039, 0.0423, 0.0153, 0.0058, 0.0124, 0.0100, 0.0160, 0.0413], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0126, 0.0108, 0.0083, 0.0107, 0.0119, 0.0106, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:28:27,652 INFO [finetune.py:992] (1/2) Epoch 17, batch 6600, loss[loss=0.1405, simple_loss=0.2187, pruned_loss=0.03115, over 12345.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2554, pruned_loss=0.03779, over 2366705.83 frames. ], batch size: 30, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:28:33,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 2023-05-17 08:28:36,806 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302357.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:28:38,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-05-17 08:28:58,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 08:28:59,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-05-17 08:29:03,992 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.638e+02 3.054e+02 3.796e+02 1.353e+03, threshold=6.108e+02, percent-clipped=2.0 2023-05-17 08:29:04,020 INFO [finetune.py:992] (1/2) Epoch 17, batch 6650, loss[loss=0.132, simple_loss=0.218, pruned_loss=0.02296, over 12367.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2544, pruned_loss=0.03742, over 2372821.41 frames. ], batch size: 30, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:29:07,695 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302399.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:29:20,722 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2851, 4.5233, 4.1345, 4.8803, 4.4675, 3.0100, 4.2229, 3.0196], device='cuda:1'), covar=tensor([0.0888, 0.0921, 0.1498, 0.0595, 0.1254, 0.1695, 0.1137, 0.3521], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0384, 0.0365, 0.0334, 0.0376, 0.0279, 0.0351, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:29:40,261 INFO [finetune.py:992] (1/2) Epoch 17, batch 6700, loss[loss=0.1727, simple_loss=0.2591, pruned_loss=0.04321, over 12368.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2547, pruned_loss=0.03762, over 2375997.40 frames. ], batch size: 35, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:29:41,747 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302447.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:30:03,264 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302477.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:30:16,290 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.118e+02 2.707e+02 3.206e+02 3.617e+02 8.520e+02, threshold=6.411e+02, percent-clipped=1.0 2023-05-17 08:30:16,309 INFO [finetune.py:992] (1/2) Epoch 17, batch 6750, loss[loss=0.1494, simple_loss=0.2438, pruned_loss=0.02753, over 12070.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.255, pruned_loss=0.0378, over 2367803.10 frames. ], batch size: 32, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:30:21,287 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0991, 6.0731, 5.8224, 5.4502, 5.2438, 6.0274, 5.6064, 5.3853], device='cuda:1'), covar=tensor([0.0768, 0.0921, 0.0660, 0.1590, 0.0688, 0.0704, 0.1648, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0652, 0.0583, 0.0534, 0.0659, 0.0439, 0.0750, 0.0812, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002], device='cuda:1') 2023-05-17 08:30:41,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-05-17 08:30:46,996 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302538.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:30:52,454 INFO [finetune.py:992] (1/2) Epoch 17, batch 6800, loss[loss=0.1611, simple_loss=0.2546, pruned_loss=0.03379, over 12358.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2541, pruned_loss=0.03755, over 2369512.33 frames. ], batch size: 35, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:30:59,604 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302555.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:31:00,368 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302556.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:31:03,903 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302561.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:31:28,298 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.614e+02 3.177e+02 3.949e+02 9.293e+02, threshold=6.353e+02, percent-clipped=5.0 2023-05-17 08:31:28,317 INFO [finetune.py:992] (1/2) Epoch 17, batch 6850, loss[loss=0.1459, simple_loss=0.2337, pruned_loss=0.02905, over 12015.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2532, pruned_loss=0.03726, over 2376163.76 frames. ], batch size: 31, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:31:34,239 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302603.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:31:34,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.44 vs. limit=2.0 2023-05-17 08:31:38,535 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302609.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:31:55,385 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302632.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:32:05,133 INFO [finetune.py:992] (1/2) Epoch 17, batch 6900, loss[loss=0.1921, simple_loss=0.2767, pruned_loss=0.05371, over 10671.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2527, pruned_loss=0.03706, over 2377477.91 frames. ], batch size: 68, lr: 3.31e-03, grad_scale: 8.0 2023-05-17 08:32:10,109 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302652.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:32:32,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 08:32:40,453 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302693.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 08:32:41,565 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.663e+02 3.007e+02 3.511e+02 7.299e+02, threshold=6.014e+02, percent-clipped=1.0 2023-05-17 08:32:41,584 INFO [finetune.py:992] (1/2) Epoch 17, batch 6950, loss[loss=0.1684, simple_loss=0.2663, pruned_loss=0.03526, over 11639.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2525, pruned_loss=0.0368, over 2378350.56 frames. ], batch size: 48, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:32:47,392 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0581, 4.9388, 4.8713, 4.9627, 4.5821, 5.1103, 4.9546, 5.2574], device='cuda:1'), covar=tensor([0.0287, 0.0165, 0.0194, 0.0412, 0.0767, 0.0301, 0.0189, 0.0175], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0210, 0.0202, 0.0257, 0.0252, 0.0232, 0.0185, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 08:32:58,687 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9337, 3.9181, 3.9037, 3.9937, 3.6523, 3.5736, 3.6220, 3.8152], device='cuda:1'), covar=tensor([0.1567, 0.0854, 0.1848, 0.0859, 0.2367, 0.2100, 0.0804, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0723, 0.0639, 0.0660, 0.0875, 0.0782, 0.0584, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:33:17,304 INFO [finetune.py:992] (1/2) Epoch 17, batch 7000, loss[loss=0.1592, simple_loss=0.2437, pruned_loss=0.03739, over 12247.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2529, pruned_loss=0.03656, over 2371118.25 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:33:32,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-05-17 08:33:38,467 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-17 08:33:53,473 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.681e+02 3.106e+02 3.603e+02 9.207e+02, threshold=6.212e+02, percent-clipped=1.0 2023-05-17 08:33:53,492 INFO [finetune.py:992] (1/2) Epoch 17, batch 7050, loss[loss=0.1714, simple_loss=0.2588, pruned_loss=0.04204, over 12057.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2535, pruned_loss=0.03665, over 2377065.53 frames. ], batch size: 42, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:34:10,472 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5327, 3.6388, 3.2298, 3.1076, 2.8340, 2.7415, 3.6221, 2.3519], device='cuda:1'), covar=tensor([0.0440, 0.0157, 0.0242, 0.0265, 0.0478, 0.0442, 0.0153, 0.0544], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0167, 0.0172, 0.0199, 0.0208, 0.0206, 0.0182, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:34:18,502 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7823, 2.8324, 4.4024, 4.5958, 2.8139, 2.6336, 2.9304, 2.0816], device='cuda:1'), covar=tensor([0.1609, 0.2966, 0.0530, 0.0474, 0.1396, 0.2594, 0.2796, 0.4358], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0398, 0.0282, 0.0307, 0.0284, 0.0325, 0.0407, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:34:21,867 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302833.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:34:30,452 INFO [finetune.py:992] (1/2) Epoch 17, batch 7100, loss[loss=0.1674, simple_loss=0.2603, pruned_loss=0.03724, over 12293.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2536, pruned_loss=0.03677, over 2380468.13 frames. ], batch size: 34, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:34:38,718 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302856.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:34:56,803 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302881.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:35:06,435 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.672e+02 3.103e+02 3.650e+02 5.427e+02, threshold=6.206e+02, percent-clipped=0.0 2023-05-17 08:35:06,455 INFO [finetune.py:992] (1/2) Epoch 17, batch 7150, loss[loss=0.1958, simple_loss=0.2814, pruned_loss=0.05517, over 8185.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2538, pruned_loss=0.03696, over 2372431.75 frames. ], batch size: 97, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:35:11,100 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4360, 4.9898, 5.4488, 4.7714, 5.0612, 4.8481, 5.4745, 5.0847], device='cuda:1'), covar=tensor([0.0280, 0.0434, 0.0258, 0.0274, 0.0400, 0.0326, 0.0191, 0.0285], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0294, 0.0318, 0.0288, 0.0287, 0.0285, 0.0262, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:35:13,177 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302904.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:35:28,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-05-17 08:35:41,178 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302942.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:35:43,089 INFO [finetune.py:992] (1/2) Epoch 17, batch 7200, loss[loss=0.1981, simple_loss=0.276, pruned_loss=0.06004, over 11197.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2531, pruned_loss=0.03685, over 2373657.28 frames. ], batch size: 55, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:35:48,027 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302952.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:36:13,747 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1809, 2.6378, 3.4681, 4.1942, 3.5761, 4.1619, 3.5847, 2.9064], device='cuda:1'), covar=tensor([0.0046, 0.0387, 0.0174, 0.0043, 0.0159, 0.0076, 0.0151, 0.0401], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0126, 0.0108, 0.0082, 0.0107, 0.0119, 0.0105, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:36:14,297 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302988.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 08:36:17,271 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2751, 4.6167, 4.1929, 4.8671, 4.5135, 2.9790, 4.3268, 3.1151], device='cuda:1'), covar=tensor([0.0876, 0.0820, 0.1339, 0.0640, 0.1256, 0.1764, 0.1191, 0.3366], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0389, 0.0369, 0.0339, 0.0381, 0.0282, 0.0355, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:36:19,191 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.698e+02 3.189e+02 3.662e+02 7.824e+02, threshold=6.378e+02, percent-clipped=3.0 2023-05-17 08:36:19,210 INFO [finetune.py:992] (1/2) Epoch 17, batch 7250, loss[loss=0.1593, simple_loss=0.2517, pruned_loss=0.03348, over 12251.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2531, pruned_loss=0.03685, over 2378641.28 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:36:22,987 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303000.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:36:54,921 INFO [finetune.py:992] (1/2) Epoch 17, batch 7300, loss[loss=0.1624, simple_loss=0.2481, pruned_loss=0.03839, over 12149.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.253, pruned_loss=0.03705, over 2374607.42 frames. ], batch size: 34, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:37:00,278 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303052.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:37:31,400 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.740e+02 3.135e+02 3.845e+02 6.476e+02, threshold=6.269e+02, percent-clipped=1.0 2023-05-17 08:37:31,419 INFO [finetune.py:992] (1/2) Epoch 17, batch 7350, loss[loss=0.1696, simple_loss=0.2675, pruned_loss=0.0359, over 12115.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2534, pruned_loss=0.03735, over 2379044.84 frames. ], batch size: 38, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:37:45,124 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303113.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:37:59,238 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303133.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:38:07,520 INFO [finetune.py:992] (1/2) Epoch 17, batch 7400, loss[loss=0.1785, simple_loss=0.2669, pruned_loss=0.04499, over 11724.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2526, pruned_loss=0.03692, over 2386805.00 frames. ], batch size: 44, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:38:33,309 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303181.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:38:38,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-05-17 08:38:43,009 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.618e+02 2.999e+02 3.673e+02 1.361e+03, threshold=5.997e+02, percent-clipped=1.0 2023-05-17 08:38:43,028 INFO [finetune.py:992] (1/2) Epoch 17, batch 7450, loss[loss=0.1663, simple_loss=0.2622, pruned_loss=0.0352, over 12081.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.253, pruned_loss=0.03711, over 2383942.25 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:38:55,359 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-05-17 08:39:13,942 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303237.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:39:19,438 INFO [finetune.py:992] (1/2) Epoch 17, batch 7500, loss[loss=0.1956, simple_loss=0.284, pruned_loss=0.0536, over 12151.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2533, pruned_loss=0.03707, over 2380080.41 frames. ], batch size: 34, lr: 3.30e-03, grad_scale: 8.0 2023-05-17 08:39:29,550 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303258.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:39:37,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-05-17 08:39:49,052 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.4678, 3.6326, 3.2721, 3.1472, 2.9070, 2.8168, 3.6293, 2.3353], device='cuda:1'), covar=tensor([0.0478, 0.0146, 0.0226, 0.0230, 0.0427, 0.0400, 0.0144, 0.0557], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0167, 0.0173, 0.0199, 0.0208, 0.0207, 0.0181, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:39:51,226 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303288.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:39:55,780 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.593e+02 3.048e+02 3.585e+02 8.586e+02, threshold=6.095e+02, percent-clipped=3.0 2023-05-17 08:39:55,800 INFO [finetune.py:992] (1/2) Epoch 17, batch 7550, loss[loss=0.17, simple_loss=0.2625, pruned_loss=0.03877, over 11665.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2526, pruned_loss=0.03682, over 2380525.08 frames. ], batch size: 48, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:40:04,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 08:40:06,628 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3958, 4.8281, 2.9813, 2.8903, 4.1627, 2.9816, 4.0686, 3.4909], device='cuda:1'), covar=tensor([0.0767, 0.0469, 0.1187, 0.1401, 0.0282, 0.1140, 0.0526, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0265, 0.0180, 0.0205, 0.0145, 0.0187, 0.0205, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:40:13,070 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303319.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:40:25,354 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303336.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:40:31,494 INFO [finetune.py:992] (1/2) Epoch 17, batch 7600, loss[loss=0.1696, simple_loss=0.2635, pruned_loss=0.03783, over 12133.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.2512, pruned_loss=0.03618, over 2387774.45 frames. ], batch size: 38, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:40:38,926 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5144, 2.3958, 3.7312, 4.3461, 3.8965, 4.3733, 3.9328, 3.0324], device='cuda:1'), covar=tensor([0.0044, 0.0564, 0.0150, 0.0064, 0.0133, 0.0098, 0.0169, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0125, 0.0107, 0.0082, 0.0106, 0.0118, 0.0105, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:40:54,655 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.1655, 3.9313, 3.9956, 4.3535, 2.6771, 3.8969, 2.6646, 4.0237], device='cuda:1'), covar=tensor([0.1615, 0.0759, 0.0879, 0.0663, 0.1314, 0.0620, 0.1800, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0235, 0.0273, 0.0303, 0.0368, 0.0247, 0.0250, 0.0267, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:40:59,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-05-17 08:41:08,086 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.479e+02 2.824e+02 3.319e+02 6.002e+02, threshold=5.649e+02, percent-clipped=0.0 2023-05-17 08:41:08,115 INFO [finetune.py:992] (1/2) Epoch 17, batch 7650, loss[loss=0.1616, simple_loss=0.2544, pruned_loss=0.03442, over 12361.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.2512, pruned_loss=0.03623, over 2391540.51 frames. ], batch size: 38, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:41:14,437 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303403.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:41:17,930 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303408.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:41:44,967 INFO [finetune.py:992] (1/2) Epoch 17, batch 7700, loss[loss=0.1502, simple_loss=0.232, pruned_loss=0.03421, over 12331.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.2504, pruned_loss=0.03607, over 2390862.63 frames. ], batch size: 31, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:41:58,623 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303464.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:42:20,225 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.621e+02 3.064e+02 3.825e+02 7.913e+02, threshold=6.128e+02, percent-clipped=5.0 2023-05-17 08:42:20,244 INFO [finetune.py:992] (1/2) Epoch 17, batch 7750, loss[loss=0.1671, simple_loss=0.2586, pruned_loss=0.0378, over 12059.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2521, pruned_loss=0.03687, over 2374863.27 frames. ], batch size: 37, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:42:25,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 08:42:50,713 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303537.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:42:56,409 INFO [finetune.py:992] (1/2) Epoch 17, batch 7800, loss[loss=0.1568, simple_loss=0.2379, pruned_loss=0.03782, over 12350.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2529, pruned_loss=0.03708, over 2374695.71 frames. ], batch size: 31, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:43:14,941 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0105, 4.5674, 4.7528, 4.8992, 4.7813, 4.8773, 4.8413, 2.6090], device='cuda:1'), covar=tensor([0.0116, 0.0083, 0.0101, 0.0054, 0.0049, 0.0096, 0.0083, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0087, 0.0076, 0.0063, 0.0098, 0.0085, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:43:17,845 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303574.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:43:25,443 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303585.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:43:30,571 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2611, 4.8482, 5.0741, 5.1621, 4.9819, 5.1782, 5.0704, 2.7670], device='cuda:1'), covar=tensor([0.0088, 0.0073, 0.0077, 0.0052, 0.0045, 0.0092, 0.0075, 0.0748], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0087, 0.0076, 0.0063, 0.0098, 0.0085, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:43:32,524 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.681e+02 3.081e+02 3.940e+02 6.716e+02, threshold=6.161e+02, percent-clipped=3.0 2023-05-17 08:43:32,552 INFO [finetune.py:992] (1/2) Epoch 17, batch 7850, loss[loss=0.1595, simple_loss=0.2472, pruned_loss=0.03592, over 12125.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2534, pruned_loss=0.03742, over 2367745.04 frames. ], batch size: 30, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:43:44,050 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-05-17 08:43:46,365 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303614.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:43:54,970 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7583, 2.6595, 3.9684, 4.1154, 2.8556, 2.5681, 2.7564, 2.2207], device='cuda:1'), covar=tensor([0.1599, 0.2996, 0.0581, 0.0527, 0.1360, 0.2638, 0.2899, 0.4057], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0395, 0.0282, 0.0306, 0.0283, 0.0323, 0.0404, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:44:01,268 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303635.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:44:08,063 INFO [finetune.py:992] (1/2) Epoch 17, batch 7900, loss[loss=0.1816, simple_loss=0.2698, pruned_loss=0.04667, over 12356.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2534, pruned_loss=0.03744, over 2374160.69 frames. ], batch size: 35, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:44:13,905 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2979, 5.1996, 5.1714, 5.2001, 4.9093, 5.2611, 5.3915, 5.5160], device='cuda:1'), covar=tensor([0.0222, 0.0151, 0.0167, 0.0363, 0.0665, 0.0427, 0.0119, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0212, 0.0203, 0.0259, 0.0256, 0.0233, 0.0186, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 08:44:40,104 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2345, 4.5212, 2.8256, 2.4139, 4.0420, 2.6559, 3.8211, 3.0040], device='cuda:1'), covar=tensor([0.0779, 0.0556, 0.1082, 0.1622, 0.0251, 0.1267, 0.0563, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0268, 0.0182, 0.0207, 0.0146, 0.0188, 0.0207, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:44:44,109 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 2.729e+02 3.278e+02 3.930e+02 6.379e+02, threshold=6.555e+02, percent-clipped=1.0 2023-05-17 08:44:44,128 INFO [finetune.py:992] (1/2) Epoch 17, batch 7950, loss[loss=0.1492, simple_loss=0.2385, pruned_loss=0.02998, over 12347.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2538, pruned_loss=0.03737, over 2377556.23 frames. ], batch size: 30, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:44:54,268 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303708.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:45:11,152 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2067, 6.0656, 5.6677, 5.5866, 6.1838, 5.4715, 5.4282, 5.6225], device='cuda:1'), covar=tensor([0.1632, 0.0931, 0.0942, 0.1820, 0.0824, 0.2235, 0.2278, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0519, 0.0421, 0.0468, 0.0485, 0.0462, 0.0422, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:45:20,509 INFO [finetune.py:992] (1/2) Epoch 17, batch 8000, loss[loss=0.1395, simple_loss=0.2271, pruned_loss=0.026, over 12238.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2542, pruned_loss=0.03767, over 2368645.08 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:45:28,478 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303756.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:45:30,553 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303759.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:45:55,804 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.710e+02 3.151e+02 3.848e+02 7.315e+02, threshold=6.302e+02, percent-clipped=1.0 2023-05-17 08:45:55,824 INFO [finetune.py:992] (1/2) Epoch 17, batch 8050, loss[loss=0.1811, simple_loss=0.2738, pruned_loss=0.0442, over 12366.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2539, pruned_loss=0.03776, over 2364709.87 frames. ], batch size: 35, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:46:07,839 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303810.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:46:32,988 INFO [finetune.py:992] (1/2) Epoch 17, batch 8100, loss[loss=0.2338, simple_loss=0.2975, pruned_loss=0.08505, over 7660.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2546, pruned_loss=0.03842, over 2358565.41 frames. ], batch size: 97, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:46:51,315 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303871.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:46:59,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-17 08:47:08,398 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.672e+02 3.103e+02 3.782e+02 1.395e+03, threshold=6.206e+02, percent-clipped=4.0 2023-05-17 08:47:08,417 INFO [finetune.py:992] (1/2) Epoch 17, batch 8150, loss[loss=0.1528, simple_loss=0.2443, pruned_loss=0.03062, over 12153.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2547, pruned_loss=0.03828, over 2361118.14 frames. ], batch size: 34, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:47:09,341 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303896.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:47:22,012 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303914.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:47:23,590 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5954, 4.1334, 4.2591, 4.6382, 3.2319, 4.1366, 2.7559, 4.3136], device='cuda:1'), covar=tensor([0.1429, 0.0680, 0.0765, 0.0529, 0.1174, 0.0582, 0.1785, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0271, 0.0302, 0.0365, 0.0246, 0.0249, 0.0266, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:47:24,245 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3932, 5.0228, 5.2449, 5.2944, 5.0840, 5.2818, 5.1696, 3.1107], device='cuda:1'), covar=tensor([0.0078, 0.0070, 0.0066, 0.0054, 0.0039, 0.0102, 0.0099, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0083, 0.0088, 0.0076, 0.0063, 0.0098, 0.0086, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:47:33,455 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303930.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:47:44,045 INFO [finetune.py:992] (1/2) Epoch 17, batch 8200, loss[loss=0.1526, simple_loss=0.237, pruned_loss=0.03413, over 12352.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.255, pruned_loss=0.03815, over 2362739.48 frames. ], batch size: 30, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:47:53,610 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303957.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:47:56,684 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-05-17 08:47:56,988 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303962.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:48:05,583 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303974.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:48:20,264 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.983e+02 2.564e+02 2.946e+02 3.521e+02 6.941e+02, threshold=5.893e+02, percent-clipped=1.0 2023-05-17 08:48:20,283 INFO [finetune.py:992] (1/2) Epoch 17, batch 8250, loss[loss=0.1655, simple_loss=0.2605, pruned_loss=0.03531, over 12277.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2545, pruned_loss=0.03805, over 2364334.70 frames. ], batch size: 37, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:48:50,975 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8158, 2.7792, 4.1748, 4.2752, 2.8857, 2.6224, 2.9960, 2.2048], device='cuda:1'), covar=tensor([0.1563, 0.2768, 0.0551, 0.0551, 0.1357, 0.2521, 0.2476, 0.4006], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0395, 0.0282, 0.0306, 0.0282, 0.0324, 0.0405, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:48:53,025 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304035.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:49:00,024 INFO [finetune.py:992] (1/2) Epoch 17, batch 8300, loss[loss=0.1634, simple_loss=0.2514, pruned_loss=0.03765, over 12353.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2544, pruned_loss=0.03778, over 2370167.20 frames. ], batch size: 35, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:49:10,193 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304059.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:49:36,077 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.634e+02 3.099e+02 3.547e+02 8.930e+02, threshold=6.198e+02, percent-clipped=2.0 2023-05-17 08:49:36,096 INFO [finetune.py:992] (1/2) Epoch 17, batch 8350, loss[loss=0.1754, simple_loss=0.2773, pruned_loss=0.03678, over 12350.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2543, pruned_loss=0.03777, over 2372709.07 frames. ], batch size: 35, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:49:37,616 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2088, 4.8098, 5.1064, 5.1197, 4.8945, 5.1462, 5.0094, 2.9039], device='cuda:1'), covar=tensor([0.0103, 0.0079, 0.0075, 0.0055, 0.0048, 0.0108, 0.0076, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0083, 0.0088, 0.0076, 0.0063, 0.0099, 0.0086, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:49:44,646 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304107.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:50:12,014 INFO [finetune.py:992] (1/2) Epoch 17, batch 8400, loss[loss=0.2398, simple_loss=0.3107, pruned_loss=0.08445, over 8137.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2551, pruned_loss=0.03841, over 2368210.99 frames. ], batch size: 97, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:50:27,004 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304166.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:50:47,443 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.079e+02 2.713e+02 3.189e+02 4.007e+02 8.492e+02, threshold=6.378e+02, percent-clipped=2.0 2023-05-17 08:50:47,462 INFO [finetune.py:992] (1/2) Epoch 17, batch 8450, loss[loss=0.147, simple_loss=0.2358, pruned_loss=0.02914, over 12172.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2552, pruned_loss=0.03845, over 2361928.91 frames. ], batch size: 29, lr: 3.30e-03, grad_scale: 16.0 2023-05-17 08:51:12,518 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304230.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:51:23,787 INFO [finetune.py:992] (1/2) Epoch 17, batch 8500, loss[loss=0.1741, simple_loss=0.2654, pruned_loss=0.0414, over 12106.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2548, pruned_loss=0.03821, over 2367283.86 frames. ], batch size: 38, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:51:28,770 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304252.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:51:47,825 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304278.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:51:59,812 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.720e+02 3.115e+02 3.611e+02 6.407e+02, threshold=6.230e+02, percent-clipped=1.0 2023-05-17 08:51:59,831 INFO [finetune.py:992] (1/2) Epoch 17, batch 8550, loss[loss=0.1667, simple_loss=0.2544, pruned_loss=0.03951, over 12105.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.255, pruned_loss=0.03835, over 2368806.78 frames. ], batch size: 32, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:52:24,666 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304330.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:52:34,689 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-05-17 08:52:34,916 INFO [finetune.py:992] (1/2) Epoch 17, batch 8600, loss[loss=0.1768, simple_loss=0.2682, pruned_loss=0.04271, over 12154.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2552, pruned_loss=0.03876, over 2355179.06 frames. ], batch size: 39, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:52:41,121 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-05-17 08:53:07,068 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2162, 5.1014, 5.1431, 4.6793, 5.1537, 4.6474, 5.0674, 5.0403], device='cuda:1'), covar=tensor([0.0672, 0.0567, 0.0832, 0.0397, 0.0482, 0.0553, 0.0769, 0.0360], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0289, 0.0312, 0.0283, 0.0283, 0.0281, 0.0257, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:53:09,193 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2002, 4.7929, 4.9842, 5.0826, 4.8453, 5.0418, 4.9506, 2.6677], device='cuda:1'), covar=tensor([0.0097, 0.0068, 0.0081, 0.0055, 0.0049, 0.0118, 0.0084, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0087, 0.0075, 0.0063, 0.0097, 0.0085, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:53:11,175 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.657e+02 3.103e+02 3.765e+02 6.079e+02, threshold=6.206e+02, percent-clipped=0.0 2023-05-17 08:53:11,205 INFO [finetune.py:992] (1/2) Epoch 17, batch 8650, loss[loss=0.1821, simple_loss=0.2604, pruned_loss=0.05191, over 12149.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2549, pruned_loss=0.0385, over 2359169.37 frames. ], batch size: 36, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:53:48,060 INFO [finetune.py:992] (1/2) Epoch 17, batch 8700, loss[loss=0.2221, simple_loss=0.2998, pruned_loss=0.07221, over 8444.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2535, pruned_loss=0.03777, over 2366384.73 frames. ], batch size: 98, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:53:55,509 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3293, 3.9735, 4.0639, 4.2900, 2.9804, 3.7972, 2.7117, 4.0816], device='cuda:1'), covar=tensor([0.1536, 0.0728, 0.0914, 0.0580, 0.1185, 0.0669, 0.1738, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0270, 0.0302, 0.0361, 0.0245, 0.0248, 0.0264, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:54:01,171 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.1027, 4.6600, 4.0249, 4.9132, 4.3247, 2.8852, 4.0547, 2.9492], device='cuda:1'), covar=tensor([0.0942, 0.0756, 0.1476, 0.0487, 0.1248, 0.1745, 0.1181, 0.3429], device='cuda:1'), in_proj_covar=tensor([0.0319, 0.0390, 0.0372, 0.0341, 0.0384, 0.0283, 0.0359, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:54:03,020 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304466.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:54:23,647 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.553e+02 2.955e+02 3.456e+02 6.024e+02, threshold=5.911e+02, percent-clipped=0.0 2023-05-17 08:54:23,667 INFO [finetune.py:992] (1/2) Epoch 17, batch 8750, loss[loss=0.1498, simple_loss=0.2404, pruned_loss=0.02964, over 12090.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2536, pruned_loss=0.03762, over 2368923.83 frames. ], batch size: 32, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:54:37,294 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304514.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:54:40,235 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304518.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:54:59,897 INFO [finetune.py:992] (1/2) Epoch 17, batch 8800, loss[loss=0.1989, simple_loss=0.2916, pruned_loss=0.05315, over 11791.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2541, pruned_loss=0.03783, over 2365145.85 frames. ], batch size: 44, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:55:03,610 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0659, 3.9063, 4.0138, 3.6969, 3.8735, 3.7323, 3.9888, 3.6764], device='cuda:1'), covar=tensor([0.0399, 0.0425, 0.0445, 0.0299, 0.0423, 0.0372, 0.0448, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0288, 0.0310, 0.0282, 0.0282, 0.0280, 0.0257, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:55:05,045 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304552.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:55:11,541 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1146, 5.9273, 5.5605, 5.5287, 6.0844, 5.4182, 5.5396, 5.5133], device='cuda:1'), covar=tensor([0.1483, 0.1036, 0.1127, 0.1956, 0.0970, 0.2352, 0.2022, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0516, 0.0418, 0.0466, 0.0480, 0.0457, 0.0418, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 08:55:25,104 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304579.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:55:36,326 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.569e+02 2.981e+02 3.638e+02 6.773e+02, threshold=5.961e+02, percent-clipped=1.0 2023-05-17 08:55:36,345 INFO [finetune.py:992] (1/2) Epoch 17, batch 8850, loss[loss=0.1788, simple_loss=0.2748, pruned_loss=0.04136, over 12104.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2539, pruned_loss=0.03758, over 2364173.35 frames. ], batch size: 38, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:55:40,251 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304600.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:55:51,584 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304616.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:55:55,844 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304622.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 08:56:01,670 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304630.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 08:56:11,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 08:56:12,330 INFO [finetune.py:992] (1/2) Epoch 17, batch 8900, loss[loss=0.2139, simple_loss=0.2855, pruned_loss=0.07112, over 7432.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2534, pruned_loss=0.03746, over 2366068.87 frames. ], batch size: 98, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:56:12,472 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304645.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:56:29,430 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304668.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:56:34,490 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304675.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:56:35,940 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304677.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:56:36,527 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304678.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 08:56:39,954 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304683.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:56:48,102 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 2.720e+02 3.219e+02 3.830e+02 1.038e+03, threshold=6.438e+02, percent-clipped=4.0 2023-05-17 08:56:48,122 INFO [finetune.py:992] (1/2) Epoch 17, batch 8950, loss[loss=0.1579, simple_loss=0.2496, pruned_loss=0.03314, over 12279.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2537, pruned_loss=0.03747, over 2368410.47 frames. ], batch size: 37, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:56:56,110 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304706.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:57:12,803 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304729.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:57:17,835 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304736.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:57:18,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-05-17 08:57:23,959 INFO [finetune.py:992] (1/2) Epoch 17, batch 9000, loss[loss=0.1546, simple_loss=0.2442, pruned_loss=0.03251, over 12189.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2535, pruned_loss=0.03781, over 2370880.07 frames. ], batch size: 31, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:57:23,959 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 08:57:43,042 INFO [finetune.py:1026] (1/2) Epoch 17, validation: loss=0.3201, simple_loss=0.3917, pruned_loss=0.1243, over 1020973.00 frames. 2023-05-17 08:57:43,042 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 08:58:19,171 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304794.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:58:19,665 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.754e+02 3.090e+02 3.592e+02 8.668e+02, threshold=6.181e+02, percent-clipped=1.0 2023-05-17 08:58:19,694 INFO [finetune.py:992] (1/2) Epoch 17, batch 9050, loss[loss=0.172, simple_loss=0.2651, pruned_loss=0.03949, over 12028.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2537, pruned_loss=0.03753, over 2378466.19 frames. ], batch size: 40, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:58:31,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-05-17 08:58:37,115 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2646, 4.8462, 5.0664, 5.0979, 4.9277, 5.1425, 5.0445, 3.0106], device='cuda:1'), covar=tensor([0.0074, 0.0073, 0.0081, 0.0055, 0.0046, 0.0089, 0.0094, 0.0649], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0082, 0.0087, 0.0076, 0.0063, 0.0098, 0.0086, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 08:58:56,320 INFO [finetune.py:992] (1/2) Epoch 17, batch 9100, loss[loss=0.1662, simple_loss=0.256, pruned_loss=0.0382, over 10503.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2546, pruned_loss=0.03769, over 2381044.63 frames. ], batch size: 68, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:59:03,668 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304855.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 08:59:17,260 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304874.0, num_to_drop=1, layers_to_drop={3} 2023-05-17 08:59:23,207 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3883, 4.8176, 4.2497, 5.0333, 4.5004, 3.0190, 4.3426, 3.0455], device='cuda:1'), covar=tensor([0.0809, 0.0737, 0.1408, 0.0457, 0.1168, 0.1754, 0.1071, 0.3535], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0388, 0.0371, 0.0340, 0.0385, 0.0283, 0.0359, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 08:59:32,222 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.375e+02 2.795e+02 3.469e+02 7.319e+02, threshold=5.590e+02, percent-clipped=2.0 2023-05-17 08:59:32,241 INFO [finetune.py:992] (1/2) Epoch 17, batch 9150, loss[loss=0.1524, simple_loss=0.2424, pruned_loss=0.0312, over 12062.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2536, pruned_loss=0.03703, over 2387462.93 frames. ], batch size: 32, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 08:59:52,260 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0962, 5.9421, 5.4890, 5.4952, 6.0538, 5.2683, 5.4628, 5.4835], device='cuda:1'), covar=tensor([0.1537, 0.0946, 0.1140, 0.1810, 0.0958, 0.2350, 0.1827, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0514, 0.0419, 0.0466, 0.0482, 0.0455, 0.0418, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:00:08,398 INFO [finetune.py:992] (1/2) Epoch 17, batch 9200, loss[loss=0.1734, simple_loss=0.2665, pruned_loss=0.04015, over 12129.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2533, pruned_loss=0.03642, over 2394347.50 frames. ], batch size: 38, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:00:28,178 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304972.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:00:32,317 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304978.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 09:00:44,332 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.586e+02 3.113e+02 3.849e+02 6.015e+02, threshold=6.226e+02, percent-clipped=1.0 2023-05-17 09:00:44,352 INFO [finetune.py:992] (1/2) Epoch 17, batch 9250, loss[loss=0.1795, simple_loss=0.2817, pruned_loss=0.03859, over 12063.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2536, pruned_loss=0.03664, over 2392090.28 frames. ], batch size: 42, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:00:48,881 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305001.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:01:05,375 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305024.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:01:10,319 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305031.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:01:20,153 INFO [finetune.py:992] (1/2) Epoch 17, batch 9300, loss[loss=0.1733, simple_loss=0.2666, pruned_loss=0.03998, over 12020.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2532, pruned_loss=0.03662, over 2390875.08 frames. ], batch size: 40, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:01:55,876 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.651e+02 3.225e+02 3.688e+02 6.047e+02, threshold=6.451e+02, percent-clipped=0.0 2023-05-17 09:01:55,895 INFO [finetune.py:992] (1/2) Epoch 17, batch 9350, loss[loss=0.161, simple_loss=0.2614, pruned_loss=0.0303, over 12150.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2532, pruned_loss=0.03647, over 2392824.85 frames. ], batch size: 36, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:02:20,539 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.0521, 4.6648, 5.0041, 4.4359, 4.7070, 4.4600, 5.0369, 4.6191], device='cuda:1'), covar=tensor([0.0277, 0.0437, 0.0309, 0.0281, 0.0402, 0.0371, 0.0226, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0287, 0.0311, 0.0281, 0.0280, 0.0280, 0.0256, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:02:32,086 INFO [finetune.py:992] (1/2) Epoch 17, batch 9400, loss[loss=0.1549, simple_loss=0.2524, pruned_loss=0.02874, over 12359.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2525, pruned_loss=0.03627, over 2391554.27 frames. ], batch size: 35, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:02:35,637 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305150.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:02:42,953 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3347, 4.7028, 2.9255, 2.7224, 4.0570, 2.6496, 3.9585, 3.1801], device='cuda:1'), covar=tensor([0.0726, 0.0508, 0.1133, 0.1467, 0.0293, 0.1296, 0.0516, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0263, 0.0179, 0.0204, 0.0145, 0.0185, 0.0205, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:02:46,458 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2768, 4.8158, 5.2252, 4.5933, 4.8915, 4.6442, 5.2906, 4.8213], device='cuda:1'), covar=tensor([0.0272, 0.0494, 0.0319, 0.0293, 0.0429, 0.0379, 0.0194, 0.0466], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0288, 0.0311, 0.0281, 0.0280, 0.0280, 0.0255, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:02:52,791 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.5927, 4.4824, 4.6050, 4.6774, 4.3473, 4.4080, 4.1629, 4.5837], device='cuda:1'), covar=tensor([0.0917, 0.0671, 0.1005, 0.0565, 0.1865, 0.1203, 0.0613, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0561, 0.0724, 0.0640, 0.0655, 0.0874, 0.0771, 0.0583, 0.0498], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:02:52,815 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305174.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 09:02:56,472 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3150, 2.6848, 3.8667, 3.2920, 3.7540, 3.4932, 2.8425, 3.7532], device='cuda:1'), covar=tensor([0.0134, 0.0385, 0.0175, 0.0258, 0.0173, 0.0174, 0.0391, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0217, 0.0206, 0.0198, 0.0233, 0.0178, 0.0209, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:02:59,496 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.0658, 4.3651, 3.9438, 4.6710, 4.2326, 2.8861, 3.9493, 2.9215], device='cuda:1'), covar=tensor([0.0902, 0.0939, 0.1435, 0.0664, 0.1344, 0.1789, 0.1197, 0.3496], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0387, 0.0370, 0.0340, 0.0383, 0.0282, 0.0358, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:03:07,803 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.616e+02 3.003e+02 3.663e+02 5.501e+02, threshold=6.006e+02, percent-clipped=0.0 2023-05-17 09:03:07,822 INFO [finetune.py:992] (1/2) Epoch 17, batch 9450, loss[loss=0.2539, simple_loss=0.3193, pruned_loss=0.09421, over 7601.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2527, pruned_loss=0.03664, over 2388684.29 frames. ], batch size: 99, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:03:12,492 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0578, 2.3951, 3.6984, 3.0473, 3.4403, 3.2131, 2.5363, 3.4948], device='cuda:1'), covar=tensor([0.0141, 0.0409, 0.0145, 0.0247, 0.0193, 0.0185, 0.0397, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0217, 0.0206, 0.0198, 0.0232, 0.0178, 0.0208, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:03:28,292 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305222.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 09:03:43,092 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305243.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:03:44,269 INFO [finetune.py:992] (1/2) Epoch 17, batch 9500, loss[loss=0.1484, simple_loss=0.2314, pruned_loss=0.03265, over 11847.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.2523, pruned_loss=0.03622, over 2386966.97 frames. ], batch size: 26, lr: 3.29e-03, grad_scale: 16.0 2023-05-17 09:04:04,224 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305272.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:08,556 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305278.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 09:04:20,294 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.458e+02 2.931e+02 3.477e+02 6.918e+02, threshold=5.861e+02, percent-clipped=2.0 2023-05-17 09:04:20,313 INFO [finetune.py:992] (1/2) Epoch 17, batch 9550, loss[loss=0.1562, simple_loss=0.2494, pruned_loss=0.03146, over 12190.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2534, pruned_loss=0.0368, over 2374136.11 frames. ], batch size: 35, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:04:22,020 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.0246, 2.3395, 2.2809, 2.2719, 2.1224, 2.0115, 2.1490, 1.7184], device='cuda:1'), covar=tensor([0.0373, 0.0213, 0.0231, 0.0211, 0.0377, 0.0297, 0.0241, 0.0486], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0168, 0.0174, 0.0200, 0.0210, 0.0208, 0.0183, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:04:24,920 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305301.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:27,149 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305304.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:35,536 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7023, 3.8450, 3.3458, 3.2931, 2.9627, 2.8645, 3.7883, 2.4879], device='cuda:1'), covar=tensor([0.0426, 0.0122, 0.0236, 0.0247, 0.0464, 0.0385, 0.0136, 0.0543], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0169, 0.0175, 0.0200, 0.0211, 0.0209, 0.0184, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:04:38,126 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305320.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:39,687 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305322.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:41,062 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305324.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:42,417 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305326.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 09:04:45,897 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305331.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:04:56,392 INFO [finetune.py:992] (1/2) Epoch 17, batch 9600, loss[loss=0.1588, simple_loss=0.2417, pruned_loss=0.0379, over 12349.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2543, pruned_loss=0.03743, over 2359819.66 frames. ], batch size: 30, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:04:58,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2023-05-17 09:04:59,213 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305349.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:05:15,811 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305372.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:05:20,907 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305379.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:05:23,972 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305383.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:05:24,961 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-05-17 09:05:28,856 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5135, 4.0768, 4.2544, 4.5011, 3.0964, 4.0561, 2.6919, 4.2304], device='cuda:1'), covar=tensor([0.1387, 0.0734, 0.0739, 0.0614, 0.1138, 0.0598, 0.1832, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0269, 0.0300, 0.0361, 0.0245, 0.0247, 0.0264, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:05:32,257 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.584e+02 3.111e+02 3.897e+02 1.039e+03, threshold=6.221e+02, percent-clipped=5.0 2023-05-17 09:05:32,276 INFO [finetune.py:992] (1/2) Epoch 17, batch 9650, loss[loss=0.1595, simple_loss=0.2446, pruned_loss=0.0372, over 12072.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2545, pruned_loss=0.0376, over 2353513.73 frames. ], batch size: 32, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:06:09,065 INFO [finetune.py:992] (1/2) Epoch 17, batch 9700, loss[loss=0.1343, simple_loss=0.2164, pruned_loss=0.02615, over 11980.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.254, pruned_loss=0.03742, over 2355095.51 frames. ], batch size: 28, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:06:12,858 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305450.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:06:44,924 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.649e+02 3.133e+02 3.869e+02 5.650e+02, threshold=6.266e+02, percent-clipped=0.0 2023-05-17 09:06:44,943 INFO [finetune.py:992] (1/2) Epoch 17, batch 9750, loss[loss=0.1537, simple_loss=0.2422, pruned_loss=0.03261, over 12098.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2535, pruned_loss=0.03733, over 2357138.84 frames. ], batch size: 32, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:06:47,092 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305498.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:07:20,621 INFO [finetune.py:992] (1/2) Epoch 17, batch 9800, loss[loss=0.1771, simple_loss=0.2694, pruned_loss=0.04239, over 12029.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2541, pruned_loss=0.03725, over 2368213.00 frames. ], batch size: 37, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:07:39,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-05-17 09:07:57,101 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.590e+02 2.906e+02 3.453e+02 6.733e+02, threshold=5.813e+02, percent-clipped=1.0 2023-05-17 09:07:57,120 INFO [finetune.py:992] (1/2) Epoch 17, batch 9850, loss[loss=0.1603, simple_loss=0.2513, pruned_loss=0.03464, over 12300.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2548, pruned_loss=0.03734, over 2368774.74 frames. ], batch size: 33, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:07:59,945 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305599.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:08:17,318 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305623.0, num_to_drop=1, layers_to_drop={1} 2023-05-17 09:08:28,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.44 vs. limit=5.0 2023-05-17 09:08:33,428 INFO [finetune.py:992] (1/2) Epoch 17, batch 9900, loss[loss=0.1944, simple_loss=0.2834, pruned_loss=0.05266, over 11817.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2546, pruned_loss=0.0373, over 2376543.89 frames. ], batch size: 44, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:08:33,698 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6491, 4.3730, 4.4537, 4.6497, 3.5270, 4.2257, 2.8092, 4.4785], device='cuda:1'), covar=tensor([0.1395, 0.0641, 0.0746, 0.0649, 0.0935, 0.0570, 0.1784, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0234, 0.0274, 0.0303, 0.0367, 0.0248, 0.0251, 0.0268, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:08:38,429 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305652.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:08:57,038 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305678.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:09:01,663 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305684.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 09:09:09,280 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.607e+02 3.012e+02 3.644e+02 9.148e+02, threshold=6.023e+02, percent-clipped=2.0 2023-05-17 09:09:09,308 INFO [finetune.py:992] (1/2) Epoch 17, batch 9950, loss[loss=0.172, simple_loss=0.2605, pruned_loss=0.04181, over 12134.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2547, pruned_loss=0.03707, over 2377608.42 frames. ], batch size: 39, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:09:22,226 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6697, 4.3878, 4.6449, 4.1637, 4.3928, 4.1955, 4.6486, 4.2147], device='cuda:1'), covar=tensor([0.0281, 0.0390, 0.0281, 0.0277, 0.0407, 0.0381, 0.0247, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0286, 0.0311, 0.0281, 0.0280, 0.0280, 0.0255, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:09:22,931 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305713.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:09:45,750 INFO [finetune.py:992] (1/2) Epoch 17, batch 10000, loss[loss=0.1585, simple_loss=0.2502, pruned_loss=0.0334, over 12354.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2545, pruned_loss=0.03728, over 2367400.90 frames. ], batch size: 35, lr: 3.29e-03, grad_scale: 32.0 2023-05-17 09:10:13,163 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6610, 2.7864, 3.3362, 4.3648, 2.6138, 4.4693, 4.6186, 4.6716], device='cuda:1'), covar=tensor([0.0122, 0.1166, 0.0510, 0.0171, 0.1282, 0.0280, 0.0175, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0208, 0.0189, 0.0127, 0.0194, 0.0187, 0.0183, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:10:22,157 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.074e+02 2.635e+02 3.117e+02 3.744e+02 7.189e+02, threshold=6.234e+02, percent-clipped=2.0 2023-05-17 09:10:22,186 INFO [finetune.py:992] (1/2) Epoch 17, batch 10050, loss[loss=0.2052, simple_loss=0.2987, pruned_loss=0.05586, over 10692.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2544, pruned_loss=0.03718, over 2364662.11 frames. ], batch size: 68, lr: 3.28e-03, grad_scale: 32.0 2023-05-17 09:10:59,048 INFO [finetune.py:992] (1/2) Epoch 17, batch 10100, loss[loss=0.1761, simple_loss=0.2641, pruned_loss=0.04404, over 12142.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2543, pruned_loss=0.03715, over 2361064.07 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:11:04,309 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305852.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:11:10,876 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6323, 3.3721, 5.1131, 2.6719, 2.9446, 3.7146, 3.2163, 3.8223], device='cuda:1'), covar=tensor([0.0474, 0.1208, 0.0330, 0.1223, 0.1995, 0.1601, 0.1477, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0241, 0.0263, 0.0186, 0.0242, 0.0300, 0.0230, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:11:34,769 INFO [finetune.py:992] (1/2) Epoch 17, batch 10150, loss[loss=0.1288, simple_loss=0.2132, pruned_loss=0.02219, over 12274.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2549, pruned_loss=0.03729, over 2369318.44 frames. ], batch size: 28, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:11:35,448 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.972e+02 2.729e+02 3.119e+02 3.676e+02 7.359e+02, threshold=6.237e+02, percent-clipped=2.0 2023-05-17 09:11:37,687 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305899.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:11:39,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-05-17 09:11:47,807 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305913.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:12:10,969 INFO [finetune.py:992] (1/2) Epoch 17, batch 10200, loss[loss=0.1736, simple_loss=0.2625, pruned_loss=0.04232, over 11618.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2549, pruned_loss=0.03738, over 2364533.49 frames. ], batch size: 48, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:12:12,364 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305947.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:12:34,681 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305978.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:12:35,332 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305979.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 09:12:47,291 INFO [finetune.py:992] (1/2) Epoch 17, batch 10250, loss[loss=0.1503, simple_loss=0.2347, pruned_loss=0.03296, over 12281.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2548, pruned_loss=0.03731, over 2367892.00 frames. ], batch size: 33, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:12:47,493 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0567, 2.5349, 3.6240, 3.0884, 3.4255, 3.1993, 2.6025, 3.4794], device='cuda:1'), covar=tensor([0.0161, 0.0386, 0.0181, 0.0264, 0.0195, 0.0200, 0.0380, 0.0147], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0216, 0.0204, 0.0197, 0.0231, 0.0177, 0.0207, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:12:47,937 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 2.699e+02 3.265e+02 4.214e+02 9.898e+02, threshold=6.531e+02, percent-clipped=5.0 2023-05-17 09:12:54,353 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.7566, 3.0175, 3.4430, 4.5449, 2.6229, 4.6387, 4.7691, 4.7750], device='cuda:1'), covar=tensor([0.0112, 0.1072, 0.0445, 0.0135, 0.1334, 0.0235, 0.0118, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0206, 0.0187, 0.0126, 0.0193, 0.0185, 0.0182, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:12:59,982 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306008.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:13:12,859 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306026.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:13:26,455 INFO [finetune.py:992] (1/2) Epoch 17, batch 10300, loss[loss=0.1692, simple_loss=0.2655, pruned_loss=0.03652, over 11318.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2533, pruned_loss=0.03682, over 2365987.79 frames. ], batch size: 55, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:14:02,905 INFO [finetune.py:992] (1/2) Epoch 17, batch 10350, loss[loss=0.1661, simple_loss=0.2639, pruned_loss=0.03415, over 12146.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2533, pruned_loss=0.03672, over 2368685.90 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:14:03,615 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.658e+02 3.080e+02 3.475e+02 5.774e+02, threshold=6.161e+02, percent-clipped=0.0 2023-05-17 09:14:17,737 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8013, 4.5265, 4.7871, 4.2204, 4.5775, 4.2308, 4.7840, 4.4722], device='cuda:1'), covar=tensor([0.0344, 0.0441, 0.0393, 0.0323, 0.0402, 0.0438, 0.0339, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0285, 0.0310, 0.0280, 0.0279, 0.0279, 0.0254, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:14:23,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-05-17 09:14:32,928 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-05-17 09:14:38,946 INFO [finetune.py:992] (1/2) Epoch 17, batch 10400, loss[loss=0.1651, simple_loss=0.2607, pruned_loss=0.03473, over 12358.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2545, pruned_loss=0.03729, over 2361907.59 frames. ], batch size: 35, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:15:13,739 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306194.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:15:14,234 INFO [finetune.py:992] (1/2) Epoch 17, batch 10450, loss[loss=0.1491, simple_loss=0.2284, pruned_loss=0.03487, over 12189.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2555, pruned_loss=0.03753, over 2370234.52 frames. ], batch size: 29, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:15:14,934 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 2.638e+02 3.099e+02 3.984e+02 6.407e+02, threshold=6.199e+02, percent-clipped=2.0 2023-05-17 09:15:18,881 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2154, 5.0841, 5.0045, 5.1058, 4.7937, 5.1520, 5.2223, 5.3912], device='cuda:1'), covar=tensor([0.0264, 0.0162, 0.0184, 0.0336, 0.0655, 0.0285, 0.0132, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0211, 0.0202, 0.0259, 0.0255, 0.0233, 0.0186, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0003, 0.0005], device='cuda:1') 2023-05-17 09:15:19,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-05-17 09:15:24,573 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306208.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:15:50,228 INFO [finetune.py:992] (1/2) Epoch 17, batch 10500, loss[loss=0.156, simple_loss=0.2393, pruned_loss=0.03637, over 12275.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2559, pruned_loss=0.03792, over 2370282.78 frames. ], batch size: 28, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:15:57,731 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306255.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:16:15,160 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306279.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 09:16:26,439 INFO [finetune.py:992] (1/2) Epoch 17, batch 10550, loss[loss=0.1628, simple_loss=0.2556, pruned_loss=0.03504, over 12053.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2555, pruned_loss=0.03773, over 2370193.62 frames. ], batch size: 42, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:16:27,139 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.688e+02 3.207e+02 3.614e+02 7.710e+02, threshold=6.413e+02, percent-clipped=2.0 2023-05-17 09:16:35,826 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306308.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:16:45,825 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([6.0934, 6.0924, 5.8235, 5.3440, 5.2152, 5.9661, 5.6116, 5.3318], device='cuda:1'), covar=tensor([0.0651, 0.0720, 0.0668, 0.1650, 0.0694, 0.0712, 0.1517, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0664, 0.0591, 0.0540, 0.0667, 0.0443, 0.0762, 0.0820, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-05-17 09:16:49,262 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306327.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 09:17:02,127 INFO [finetune.py:992] (1/2) Epoch 17, batch 10600, loss[loss=0.1907, simple_loss=0.2901, pruned_loss=0.0456, over 12357.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2546, pruned_loss=0.03732, over 2362129.10 frames. ], batch size: 38, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:17:10,801 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306356.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:17:30,660 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306384.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:17:38,229 INFO [finetune.py:992] (1/2) Epoch 17, batch 10650, loss[loss=0.1347, simple_loss=0.2205, pruned_loss=0.0245, over 12280.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2544, pruned_loss=0.03703, over 2372157.59 frames. ], batch size: 28, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:17:38,925 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.440e+02 2.991e+02 3.762e+02 6.396e+02, threshold=5.981e+02, percent-clipped=0.0 2023-05-17 09:18:13,458 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.6556, 2.9435, 3.2963, 4.4510, 2.6540, 4.5042, 4.6713, 4.6716], device='cuda:1'), covar=tensor([0.0121, 0.1070, 0.0534, 0.0159, 0.1254, 0.0276, 0.0128, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0207, 0.0188, 0.0126, 0.0193, 0.0185, 0.0183, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:18:14,685 INFO [finetune.py:992] (1/2) Epoch 17, batch 10700, loss[loss=0.196, simple_loss=0.2873, pruned_loss=0.05233, over 11746.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2557, pruned_loss=0.03739, over 2364155.78 frames. ], batch size: 44, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:18:14,904 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306445.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:18:49,906 INFO [finetune.py:992] (1/2) Epoch 17, batch 10750, loss[loss=0.171, simple_loss=0.2607, pruned_loss=0.04062, over 12079.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2553, pruned_loss=0.03746, over 2362477.21 frames. ], batch size: 42, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:18:50,617 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.600e+02 3.061e+02 3.731e+02 9.933e+02, threshold=6.123e+02, percent-clipped=1.0 2023-05-17 09:19:00,078 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306508.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:19:25,905 INFO [finetune.py:992] (1/2) Epoch 17, batch 10800, loss[loss=0.1634, simple_loss=0.2438, pruned_loss=0.04149, over 12261.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2557, pruned_loss=0.03799, over 2359015.98 frames. ], batch size: 28, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:19:29,549 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306550.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:19:33,867 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306556.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:19:48,221 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.3240, 4.7224, 3.0996, 2.7297, 4.1132, 2.6555, 4.0424, 3.3026], device='cuda:1'), covar=tensor([0.0759, 0.0475, 0.1093, 0.1572, 0.0263, 0.1320, 0.0461, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0264, 0.0178, 0.0204, 0.0145, 0.0186, 0.0205, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:20:02,315 INFO [finetune.py:992] (1/2) Epoch 17, batch 10850, loss[loss=0.1388, simple_loss=0.2251, pruned_loss=0.02629, over 12142.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2547, pruned_loss=0.03745, over 2365229.36 frames. ], batch size: 30, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:20:03,000 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.667e+02 3.157e+02 3.680e+02 8.758e+02, threshold=6.314e+02, percent-clipped=1.0 2023-05-17 09:20:34,802 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2023-05-17 09:20:40,233 INFO [finetune.py:992] (1/2) Epoch 17, batch 10900, loss[loss=0.1754, simple_loss=0.268, pruned_loss=0.04135, over 12193.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2536, pruned_loss=0.03688, over 2368951.73 frames. ], batch size: 35, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:20:41,870 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.3028, 2.0475, 3.6176, 4.2881, 3.8510, 4.2088, 3.7852, 2.8484], device='cuda:1'), covar=tensor([0.0067, 0.0641, 0.0170, 0.0070, 0.0133, 0.0106, 0.0157, 0.0497], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0126, 0.0108, 0.0083, 0.0107, 0.0120, 0.0106, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:21:03,644 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8273, 2.6304, 3.6415, 3.7138, 2.9292, 2.6857, 2.7472, 2.4215], device='cuda:1'), covar=tensor([0.1370, 0.2645, 0.0692, 0.0535, 0.1087, 0.2268, 0.2637, 0.3615], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0402, 0.0289, 0.0313, 0.0287, 0.0331, 0.0413, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:21:14,274 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306692.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:21:16,276 INFO [finetune.py:992] (1/2) Epoch 17, batch 10950, loss[loss=0.1437, simple_loss=0.2368, pruned_loss=0.02527, over 12076.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2543, pruned_loss=0.03728, over 2354093.98 frames. ], batch size: 32, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:21:16,963 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.498e+02 2.941e+02 3.485e+02 6.942e+02, threshold=5.883e+02, percent-clipped=2.0 2023-05-17 09:21:30,441 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.1856, 5.0759, 4.9584, 5.0306, 4.7573, 5.1922, 5.1595, 5.3808], device='cuda:1'), covar=tensor([0.0239, 0.0149, 0.0182, 0.0325, 0.0721, 0.0303, 0.0157, 0.0177], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0209, 0.0201, 0.0260, 0.0254, 0.0232, 0.0184, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0005], device='cuda:1') 2023-05-17 09:21:48,831 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306740.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:21:52,287 INFO [finetune.py:992] (1/2) Epoch 17, batch 11000, loss[loss=0.2359, simple_loss=0.3296, pruned_loss=0.07112, over 10330.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2571, pruned_loss=0.03875, over 2336962.60 frames. ], batch size: 68, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:21:58,212 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306753.0, num_to_drop=1, layers_to_drop={2} 2023-05-17 09:22:28,093 INFO [finetune.py:992] (1/2) Epoch 17, batch 11050, loss[loss=0.2292, simple_loss=0.3026, pruned_loss=0.07786, over 8099.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2615, pruned_loss=0.04137, over 2283129.08 frames. ], batch size: 97, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:22:28,809 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.828e+02 3.526e+02 4.495e+02 1.369e+03, threshold=7.052e+02, percent-clipped=10.0 2023-05-17 09:23:03,933 INFO [finetune.py:992] (1/2) Epoch 17, batch 11100, loss[loss=0.1537, simple_loss=0.2388, pruned_loss=0.03426, over 12018.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2638, pruned_loss=0.04282, over 2245334.33 frames. ], batch size: 31, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:23:07,569 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306850.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:23:38,797 INFO [finetune.py:992] (1/2) Epoch 17, batch 11150, loss[loss=0.162, simple_loss=0.2543, pruned_loss=0.03483, over 12305.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2695, pruned_loss=0.04622, over 2190579.31 frames. ], batch size: 37, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:23:39,464 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 3.233e+02 3.980e+02 4.807e+02 1.094e+03, threshold=7.960e+02, percent-clipped=4.0 2023-05-17 09:23:40,805 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306898.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:24:14,243 INFO [finetune.py:992] (1/2) Epoch 17, batch 11200, loss[loss=0.2169, simple_loss=0.3097, pruned_loss=0.06206, over 11597.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2769, pruned_loss=0.0512, over 2118680.16 frames. ], batch size: 48, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:24:49,628 INFO [finetune.py:992] (1/2) Epoch 17, batch 11250, loss[loss=0.166, simple_loss=0.2641, pruned_loss=0.03397, over 12250.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2824, pruned_loss=0.05477, over 2081207.67 frames. ], batch size: 32, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:24:50,186 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.250e+02 3.461e+02 4.288e+02 5.110e+02 1.210e+03, threshold=8.577e+02, percent-clipped=4.0 2023-05-17 09:25:21,025 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307040.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:25:24,335 INFO [finetune.py:992] (1/2) Epoch 17, batch 11300, loss[loss=0.203, simple_loss=0.2991, pruned_loss=0.05351, over 11308.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2895, pruned_loss=0.05967, over 2001925.00 frames. ], batch size: 55, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:25:26,607 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307048.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 09:25:46,743 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9197, 2.5710, 3.4506, 3.5305, 2.9443, 2.7441, 2.7044, 2.5029], device='cuda:1'), covar=tensor([0.1263, 0.2327, 0.0659, 0.0486, 0.0929, 0.2024, 0.2223, 0.3495], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0393, 0.0282, 0.0305, 0.0280, 0.0322, 0.0404, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:25:55,071 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307088.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:25:59,709 INFO [finetune.py:992] (1/2) Epoch 17, batch 11350, loss[loss=0.1726, simple_loss=0.261, pruned_loss=0.04209, over 12104.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2945, pruned_loss=0.06261, over 1958192.55 frames. ], batch size: 32, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:26:00,344 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 3.468e+02 4.102e+02 4.903e+02 8.634e+02, threshold=8.204e+02, percent-clipped=1.0 2023-05-17 09:26:20,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 09:26:26,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.59 vs. limit=5.0 2023-05-17 09:26:34,225 INFO [finetune.py:992] (1/2) Epoch 17, batch 11400, loss[loss=0.2121, simple_loss=0.3028, pruned_loss=0.06067, over 11072.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.297, pruned_loss=0.06381, over 1927067.45 frames. ], batch size: 55, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:26:46,483 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8837, 4.4875, 4.2090, 4.2189, 4.5597, 3.9712, 4.1185, 3.9680], device='cuda:1'), covar=tensor([0.1528, 0.1064, 0.1261, 0.1561, 0.0991, 0.2099, 0.1807, 0.1523], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0504, 0.0409, 0.0453, 0.0467, 0.0441, 0.0408, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:26:47,845 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9179, 4.5510, 4.2779, 4.2755, 4.6197, 4.0847, 4.1900, 4.0412], device='cuda:1'), covar=tensor([0.1491, 0.1058, 0.1422, 0.1594, 0.0974, 0.2049, 0.1870, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0504, 0.0409, 0.0454, 0.0467, 0.0442, 0.0408, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:27:09,346 INFO [finetune.py:992] (1/2) Epoch 17, batch 11450, loss[loss=0.2252, simple_loss=0.3122, pruned_loss=0.06906, over 11581.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.3003, pruned_loss=0.06603, over 1889316.35 frames. ], batch size: 48, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:27:09,990 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.192e+02 3.229e+02 3.849e+02 4.739e+02 1.180e+03, threshold=7.698e+02, percent-clipped=2.0 2023-05-17 09:27:41,781 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2507, 2.9679, 3.6108, 2.3343, 2.6952, 3.0978, 2.9266, 3.1594], device='cuda:1'), covar=tensor([0.0544, 0.1101, 0.0270, 0.1249, 0.1703, 0.1381, 0.1179, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0233, 0.0251, 0.0180, 0.0234, 0.0288, 0.0221, 0.0262], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:27:43,513 INFO [finetune.py:992] (1/2) Epoch 17, batch 11500, loss[loss=0.193, simple_loss=0.2802, pruned_loss=0.05294, over 12073.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.3023, pruned_loss=0.06818, over 1858812.55 frames. ], batch size: 42, lr: 3.28e-03, grad_scale: 16.0 2023-05-17 09:28:19,070 INFO [finetune.py:992] (1/2) Epoch 17, batch 11550, loss[loss=0.2304, simple_loss=0.3071, pruned_loss=0.07688, over 6964.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.3033, pruned_loss=0.06909, over 1849353.85 frames. ], batch size: 98, lr: 3.28e-03, grad_scale: 8.0 2023-05-17 09:28:20,292 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.952e+02 3.645e+02 4.042e+02 4.908e+02 1.002e+03, threshold=8.085e+02, percent-clipped=3.0 2023-05-17 09:28:23,711 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.9540, 4.5110, 4.2275, 4.2467, 4.5649, 4.0120, 4.1773, 4.0897], device='cuda:1'), covar=tensor([0.1695, 0.1065, 0.1205, 0.1824, 0.1082, 0.2153, 0.1672, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0500, 0.0406, 0.0449, 0.0462, 0.0437, 0.0404, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:28:34,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-05-17 09:28:52,095 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307342.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:28:53,866 INFO [finetune.py:992] (1/2) Epoch 17, batch 11600, loss[loss=0.2797, simple_loss=0.341, pruned_loss=0.1092, over 6925.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3048, pruned_loss=0.07089, over 1822092.67 frames. ], batch size: 98, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:28:56,122 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307348.0, num_to_drop=1, layers_to_drop={0} 2023-05-17 09:29:29,975 INFO [finetune.py:992] (1/2) Epoch 17, batch 11650, loss[loss=0.2556, simple_loss=0.3218, pruned_loss=0.09469, over 6416.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3051, pruned_loss=0.07208, over 1791048.33 frames. ], batch size: 98, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:29:30,887 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307396.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:29:31,452 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.341e+02 3.378e+02 3.794e+02 4.382e+02 6.754e+02, threshold=7.588e+02, percent-clipped=0.0 2023-05-17 09:29:34,564 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-05-17 09:29:36,512 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307403.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:29:49,804 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7030, 3.3727, 3.5343, 3.6246, 3.5529, 3.6927, 3.5277, 2.6064], device='cuda:1'), covar=tensor([0.0111, 0.0133, 0.0146, 0.0087, 0.0076, 0.0158, 0.0099, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0080, 0.0086, 0.0075, 0.0061, 0.0095, 0.0084, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:30:05,905 INFO [finetune.py:992] (1/2) Epoch 17, batch 11700, loss[loss=0.1946, simple_loss=0.2936, pruned_loss=0.04781, over 10231.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3041, pruned_loss=0.07189, over 1766807.35 frames. ], batch size: 68, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:30:40,966 INFO [finetune.py:992] (1/2) Epoch 17, batch 11750, loss[loss=0.2266, simple_loss=0.2994, pruned_loss=0.07689, over 6993.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3051, pruned_loss=0.07345, over 1742063.80 frames. ], batch size: 99, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:30:42,311 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.161e+02 3.492e+02 4.034e+02 4.769e+02 1.195e+03, threshold=8.068e+02, percent-clipped=4.0 2023-05-17 09:31:15,032 INFO [finetune.py:992] (1/2) Epoch 17, batch 11800, loss[loss=0.2933, simple_loss=0.3523, pruned_loss=0.1172, over 5879.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.308, pruned_loss=0.07555, over 1703991.31 frames. ], batch size: 99, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:31:27,106 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6999, 2.8860, 2.4257, 2.1941, 2.6744, 2.3033, 2.8503, 2.5304], device='cuda:1'), covar=tensor([0.0566, 0.0542, 0.0861, 0.1309, 0.0288, 0.1092, 0.0476, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0249, 0.0173, 0.0196, 0.0139, 0.0180, 0.0195, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:31:41,871 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7606, 3.7586, 3.7211, 3.8346, 3.6400, 3.6720, 3.5504, 3.7353], device='cuda:1'), covar=tensor([0.1064, 0.0660, 0.1288, 0.0706, 0.1442, 0.1177, 0.0551, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0678, 0.0603, 0.0611, 0.0809, 0.0718, 0.0541, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:31:47,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-05-17 09:31:48,748 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8153, 3.7176, 3.8118, 3.5837, 3.7014, 3.5483, 3.7753, 3.4242], device='cuda:1'), covar=tensor([0.0477, 0.0451, 0.0444, 0.0300, 0.0469, 0.0402, 0.0451, 0.1794], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0269, 0.0291, 0.0266, 0.0263, 0.0262, 0.0239, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:31:50,568 INFO [finetune.py:992] (1/2) Epoch 17, batch 11850, loss[loss=0.254, simple_loss=0.3228, pruned_loss=0.09261, over 6205.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3092, pruned_loss=0.07532, over 1708201.14 frames. ], batch size: 99, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:31:51,996 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.549e+02 3.416e+02 4.242e+02 5.268e+02 1.514e+03, threshold=8.485e+02, percent-clipped=2.0 2023-05-17 09:31:52,847 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307598.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:31:54,517 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.0109, 2.2750, 2.6833, 3.0330, 2.2990, 3.1653, 3.0183, 3.1650], device='cuda:1'), covar=tensor([0.0185, 0.1113, 0.0530, 0.0234, 0.1156, 0.0287, 0.0333, 0.0185], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0179, 0.0119, 0.0184, 0.0176, 0.0174, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:31:57,150 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307604.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:31:59,812 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9319, 2.2739, 2.6646, 2.9771, 2.2753, 3.0781, 2.9677, 3.0966], device='cuda:1'), covar=tensor([0.0176, 0.1077, 0.0491, 0.0207, 0.1139, 0.0303, 0.0303, 0.0183], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0179, 0.0119, 0.0184, 0.0175, 0.0173, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:32:08,483 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307620.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:32:25,778 INFO [finetune.py:992] (1/2) Epoch 17, batch 11900, loss[loss=0.2299, simple_loss=0.3079, pruned_loss=0.0759, over 7311.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3084, pruned_loss=0.07408, over 1697945.19 frames. ], batch size: 99, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:32:35,346 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307659.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:32:37,378 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7891, 2.5073, 3.5243, 3.5733, 2.8206, 2.6081, 2.6650, 2.3826], device='cuda:1'), covar=tensor([0.1443, 0.2823, 0.0665, 0.0522, 0.1126, 0.2447, 0.2845, 0.3869], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0387, 0.0276, 0.0298, 0.0275, 0.0318, 0.0397, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:32:39,397 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307665.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:32:50,831 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307681.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:33:00,035 INFO [finetune.py:992] (1/2) Epoch 17, batch 11950, loss[loss=0.1773, simple_loss=0.2701, pruned_loss=0.04224, over 10505.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3056, pruned_loss=0.07203, over 1685397.55 frames. ], batch size: 69, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:33:01,321 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.326e+02 3.144e+02 3.799e+02 4.450e+02 7.453e+02, threshold=7.599e+02, percent-clipped=0.0 2023-05-17 09:33:02,157 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307698.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:33:35,905 INFO [finetune.py:992] (1/2) Epoch 17, batch 12000, loss[loss=0.2133, simple_loss=0.2962, pruned_loss=0.06518, over 7033.00 frames. ], tot_loss[loss=0.218, simple_loss=0.3004, pruned_loss=0.06774, over 1697500.35 frames. ], batch size: 101, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:33:35,906 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 09:33:40,538 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.2940, 5.8046, 5.6305, 5.4893, 5.8623, 5.2631, 5.1702, 5.5227], device='cuda:1'), covar=tensor([0.1023, 0.0899, 0.1030, 0.1319, 0.0733, 0.1973, 0.2210, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0493, 0.0402, 0.0439, 0.0455, 0.0430, 0.0397, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:33:42,967 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.4738, 5.4112, 5.4002, 5.5092, 5.1796, 5.1520, 5.1571, 5.3266], device='cuda:1'), covar=tensor([0.0683, 0.0493, 0.0938, 0.0477, 0.1666, 0.1387, 0.0498, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0523, 0.0674, 0.0600, 0.0610, 0.0807, 0.0717, 0.0538, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:33:48,315 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6955, 2.8047, 3.9920, 4.2149, 2.8382, 2.6275, 2.6925, 2.0604], device='cuda:1'), covar=tensor([0.1669, 0.3020, 0.0625, 0.0458, 0.1355, 0.2755, 0.3451, 0.4951], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0387, 0.0275, 0.0298, 0.0275, 0.0318, 0.0398, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:33:48,539 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.2707, 3.5800, 3.3253, 3.1332, 2.8081, 2.8211, 3.3013, 2.1521], device='cuda:1'), covar=tensor([0.0543, 0.0135, 0.0186, 0.0243, 0.0394, 0.0376, 0.0228, 0.0701], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0161, 0.0166, 0.0190, 0.0200, 0.0197, 0.0174, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:33:53,997 INFO [finetune.py:1026] (1/2) Epoch 17, validation: loss=0.2896, simple_loss=0.3623, pruned_loss=0.1084, over 1020973.00 frames. 2023-05-17 09:33:53,998 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 09:34:10,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-05-17 09:34:28,285 INFO [finetune.py:992] (1/2) Epoch 17, batch 12050, loss[loss=0.1734, simple_loss=0.2677, pruned_loss=0.03953, over 10318.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2966, pruned_loss=0.06494, over 1699168.54 frames. ], batch size: 68, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:34:29,604 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.957e+02 3.433e+02 3.945e+02 6.254e+02, threshold=6.866e+02, percent-clipped=0.0 2023-05-17 09:34:42,591 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.7204, 2.3269, 2.9360, 3.6847, 2.2660, 3.8572, 3.7067, 3.8028], device='cuda:1'), covar=tensor([0.0185, 0.1354, 0.0510, 0.0177, 0.1415, 0.0234, 0.0264, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0200, 0.0178, 0.0119, 0.0185, 0.0175, 0.0173, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:34:50,112 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([1.9461, 2.2006, 2.1961, 2.2110, 1.9113, 1.9239, 2.1784, 1.7209], device='cuda:1'), covar=tensor([0.0368, 0.0213, 0.0246, 0.0225, 0.0397, 0.0279, 0.0188, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0159, 0.0164, 0.0188, 0.0199, 0.0195, 0.0173, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:35:01,331 INFO [finetune.py:992] (1/2) Epoch 17, batch 12100, loss[loss=0.2081, simple_loss=0.2923, pruned_loss=0.06192, over 11710.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2959, pruned_loss=0.06371, over 1713935.57 frames. ], batch size: 44, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:35:20,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.54 vs. limit=5.0 2023-05-17 09:35:33,743 INFO [finetune.py:992] (1/2) Epoch 17, batch 12150, loss[loss=0.1895, simple_loss=0.2831, pruned_loss=0.0479, over 11642.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2962, pruned_loss=0.06379, over 1712053.24 frames. ], batch size: 48, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:35:35,025 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 3.185e+02 3.632e+02 4.072e+02 1.178e+03, threshold=7.264e+02, percent-clipped=3.0 2023-05-17 09:35:42,264 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.4847, 4.3597, 4.3820, 4.4882, 4.1117, 4.5625, 4.5351, 4.5734], device='cuda:1'), covar=tensor([0.0260, 0.0178, 0.0214, 0.0292, 0.0647, 0.0366, 0.0228, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0184, 0.0177, 0.0228, 0.0222, 0.0204, 0.0163, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-05-17 09:36:05,405 INFO [finetune.py:992] (1/2) Epoch 17, batch 12200, loss[loss=0.241, simple_loss=0.3116, pruned_loss=0.08515, over 6833.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2974, pruned_loss=0.06539, over 1682968.86 frames. ], batch size: 98, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:36:10,957 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307954.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:36:14,663 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307960.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:36:21,067 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.7926, 3.0050, 2.4226, 2.2354, 2.7794, 2.3903, 2.9640, 2.6348], device='cuda:1'), covar=tensor([0.0580, 0.0564, 0.0925, 0.1342, 0.0266, 0.1125, 0.0484, 0.0762], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0242, 0.0171, 0.0193, 0.0135, 0.0177, 0.0191, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:36:23,430 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.8952, 4.4470, 4.1432, 4.1683, 4.5134, 3.9121, 4.1129, 3.9751], device='cuda:1'), covar=tensor([0.1449, 0.1315, 0.1536, 0.1989, 0.1065, 0.2421, 0.1931, 0.1448], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0493, 0.0402, 0.0439, 0.0455, 0.0432, 0.0395, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:36:24,606 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307976.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:36:50,394 INFO [finetune.py:992] (1/2) Epoch 18, batch 0, loss[loss=0.1707, simple_loss=0.269, pruned_loss=0.03617, over 12183.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.269, pruned_loss=0.03617, over 12183.00 frames. ], batch size: 35, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:36:50,394 INFO [finetune.py:1017] (1/2) Computing validation loss 2023-05-17 09:37:07,741 INFO [finetune.py:1026] (1/2) Epoch 18, validation: loss=0.289, simple_loss=0.361, pruned_loss=0.1085, over 1020973.00 frames. 2023-05-17 09:37:07,742 INFO [finetune.py:1027] (1/2) Maximum memory allocated so far is 12663MB 2023-05-17 09:37:15,756 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2328, 4.6172, 3.9820, 4.8607, 4.4565, 2.6694, 4.1542, 2.9884], device='cuda:1'), covar=tensor([0.0903, 0.0882, 0.1609, 0.0538, 0.1301, 0.2239, 0.1176, 0.4027], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0362, 0.0344, 0.0311, 0.0355, 0.0266, 0.0333, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:37:19,682 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.214e+02 3.181e+02 3.643e+02 4.435e+02 7.153e+02, threshold=7.285e+02, percent-clipped=0.0 2023-05-17 09:37:20,578 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307998.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:37:46,446 INFO [finetune.py:992] (1/2) Epoch 18, batch 50, loss[loss=0.1774, simple_loss=0.2671, pruned_loss=0.04381, over 12088.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2615, pruned_loss=0.03947, over 541566.90 frames. ], batch size: 32, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:37:57,595 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308046.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:38:22,608 INFO [finetune.py:992] (1/2) Epoch 18, batch 100, loss[loss=0.1658, simple_loss=0.2484, pruned_loss=0.04158, over 12314.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2566, pruned_loss=0.03851, over 942256.54 frames. ], batch size: 28, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:38:35,361 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.661e+02 3.037e+02 3.590e+02 6.074e+02, threshold=6.073e+02, percent-clipped=0.0 2023-05-17 09:38:58,948 INFO [finetune.py:992] (1/2) Epoch 18, batch 150, loss[loss=0.1415, simple_loss=0.2217, pruned_loss=0.03072, over 12129.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.258, pruned_loss=0.03876, over 1269084.16 frames. ], batch size: 30, lr: 3.27e-03, grad_scale: 4.0 2023-05-17 09:39:34,681 INFO [finetune.py:992] (1/2) Epoch 18, batch 200, loss[loss=0.1543, simple_loss=0.2471, pruned_loss=0.03075, over 12293.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2577, pruned_loss=0.0383, over 1516186.57 frames. ], batch size: 28, lr: 3.27e-03, grad_scale: 4.0 2023-05-17 09:39:47,610 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.638e+02 3.218e+02 3.807e+02 6.334e+02, threshold=6.435e+02, percent-clipped=2.0 2023-05-17 09:39:48,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-05-17 09:39:51,007 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.8918, 4.7889, 4.8461, 4.8945, 4.5533, 4.6034, 4.4140, 4.8039], device='cuda:1'), covar=tensor([0.0717, 0.0654, 0.0955, 0.0655, 0.2020, 0.1440, 0.0579, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0682, 0.0609, 0.0615, 0.0815, 0.0729, 0.0545, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:39:51,074 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308202.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:40:11,470 INFO [finetune.py:992] (1/2) Epoch 18, batch 250, loss[loss=0.1462, simple_loss=0.2344, pruned_loss=0.02898, over 12299.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2573, pruned_loss=0.03835, over 1709010.91 frames. ], batch size: 33, lr: 3.27e-03, grad_scale: 4.0 2023-05-17 09:40:28,785 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308254.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:40:33,073 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308260.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:40:35,306 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308263.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:40:44,434 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308276.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:40:47,194 INFO [finetune.py:992] (1/2) Epoch 18, batch 300, loss[loss=0.1693, simple_loss=0.2551, pruned_loss=0.0418, over 12292.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.257, pruned_loss=0.03795, over 1866545.48 frames. ], batch size: 33, lr: 3.27e-03, grad_scale: 4.0 2023-05-17 09:41:00,538 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.864e+02 2.598e+02 3.095e+02 3.618e+02 6.538e+02, threshold=6.191e+02, percent-clipped=1.0 2023-05-17 09:41:03,439 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308302.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:41:04,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-05-17 09:41:07,615 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308308.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:41:08,442 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308309.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:41:19,204 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308324.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:41:23,332 INFO [finetune.py:992] (1/2) Epoch 18, batch 350, loss[loss=0.1377, simple_loss=0.2269, pruned_loss=0.02425, over 12341.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2563, pruned_loss=0.03764, over 1984235.31 frames. ], batch size: 30, lr: 3.27e-03, grad_scale: 4.0 2023-05-17 09:41:42,851 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.9358, 2.3097, 3.4771, 2.9929, 3.2766, 3.0929, 2.5278, 3.3674], device='cuda:1'), covar=tensor([0.0170, 0.0466, 0.0177, 0.0312, 0.0212, 0.0219, 0.0462, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0208, 0.0193, 0.0189, 0.0219, 0.0169, 0.0200, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:41:52,853 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308370.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:41:59,825 INFO [finetune.py:992] (1/2) Epoch 18, batch 400, loss[loss=0.1483, simple_loss=0.236, pruned_loss=0.03032, over 12355.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2561, pruned_loss=0.03765, over 2070574.22 frames. ], batch size: 31, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:42:12,391 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.569e+02 3.167e+02 3.621e+02 6.825e+02, threshold=6.334e+02, percent-clipped=2.0 2023-05-17 09:42:20,150 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.8116, 2.3208, 3.3194, 2.8377, 3.0949, 3.0120, 2.4852, 3.2442], device='cuda:1'), covar=tensor([0.0184, 0.0457, 0.0215, 0.0303, 0.0232, 0.0232, 0.0434, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0208, 0.0192, 0.0189, 0.0219, 0.0169, 0.0200, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:42:20,947 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6412, 2.6624, 4.4150, 4.4769, 2.7355, 2.5120, 2.8655, 2.0806], device='cuda:1'), covar=tensor([0.1872, 0.3415, 0.0503, 0.0557, 0.1500, 0.2976, 0.3093, 0.4477], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0392, 0.0278, 0.0302, 0.0278, 0.0322, 0.0403, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:42:35,241 INFO [finetune.py:992] (1/2) Epoch 18, batch 450, loss[loss=0.1496, simple_loss=0.2301, pruned_loss=0.03452, over 12364.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2562, pruned_loss=0.03773, over 2136290.12 frames. ], batch size: 30, lr: 3.27e-03, grad_scale: 8.0 2023-05-17 09:43:11,322 INFO [finetune.py:992] (1/2) Epoch 18, batch 500, loss[loss=0.1919, simple_loss=0.2805, pruned_loss=0.05159, over 10463.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2557, pruned_loss=0.03751, over 2195315.56 frames. ], batch size: 68, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:43:24,083 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.639e+02 3.231e+02 3.895e+02 6.138e+02, threshold=6.461e+02, percent-clipped=0.0 2023-05-17 09:43:24,349 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.6489, 3.7565, 3.4193, 3.2841, 2.9415, 2.8824, 3.6794, 2.6112], device='cuda:1'), covar=tensor([0.0449, 0.0155, 0.0191, 0.0247, 0.0518, 0.0441, 0.0184, 0.0490], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0163, 0.0167, 0.0193, 0.0203, 0.0199, 0.0176, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:43:47,326 INFO [finetune.py:992] (1/2) Epoch 18, batch 550, loss[loss=0.1806, simple_loss=0.2739, pruned_loss=0.04361, over 12301.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2546, pruned_loss=0.03708, over 2239291.52 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:44:02,102 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([3.2383, 4.6864, 3.0187, 2.6935, 4.1065, 2.7040, 3.9869, 3.2313], device='cuda:1'), covar=tensor([0.0733, 0.0641, 0.1090, 0.1515, 0.0256, 0.1268, 0.0474, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0253, 0.0177, 0.0198, 0.0140, 0.0183, 0.0198, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:44:03,552 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308552.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:44:07,537 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308558.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:44:23,002 INFO [finetune.py:992] (1/2) Epoch 18, batch 600, loss[loss=0.1576, simple_loss=0.2506, pruned_loss=0.03227, over 12350.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2529, pruned_loss=0.03684, over 2270255.34 frames. ], batch size: 35, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:44:36,546 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.569e+02 2.931e+02 3.751e+02 8.067e+02, threshold=5.861e+02, percent-clipped=1.0 2023-05-17 09:44:47,768 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308613.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:44:52,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=2.80 vs. limit=5.0 2023-05-17 09:44:56,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-05-17 09:44:59,649 INFO [finetune.py:992] (1/2) Epoch 18, batch 650, loss[loss=0.1696, simple_loss=0.2709, pruned_loss=0.03415, over 12202.00 frames. ], tot_loss[loss=0.163, simple_loss=0.253, pruned_loss=0.0365, over 2297544.77 frames. ], batch size: 35, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:45:24,965 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308665.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:45:27,795 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308669.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:45:35,304 INFO [finetune.py:992] (1/2) Epoch 18, batch 700, loss[loss=0.1671, simple_loss=0.2611, pruned_loss=0.0366, over 11813.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2544, pruned_loss=0.03672, over 2310666.32 frames. ], batch size: 44, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:45:48,061 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.567e+02 3.046e+02 3.784e+02 7.394e+02, threshold=6.092e+02, percent-clipped=4.0 2023-05-17 09:45:51,998 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-05-17 09:46:10,683 INFO [finetune.py:992] (1/2) Epoch 18, batch 750, loss[loss=0.1701, simple_loss=0.2546, pruned_loss=0.04283, over 12277.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.254, pruned_loss=0.03676, over 2321091.52 frames. ], batch size: 33, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:46:10,877 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308730.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:46:12,303 INFO [zipformer.py:625] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308732.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:46:35,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-05-17 09:46:35,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-05-17 09:46:46,597 INFO [finetune.py:992] (1/2) Epoch 18, batch 800, loss[loss=0.1565, simple_loss=0.2517, pruned_loss=0.03061, over 12156.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2545, pruned_loss=0.03692, over 2329917.28 frames. ], batch size: 36, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:46:50,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-05-17 09:46:53,643 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-05-17 09:46:56,251 INFO [zipformer.py:625] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308793.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:46:59,534 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.100e+02 2.666e+02 2.982e+02 3.536e+02 6.354e+02, threshold=5.964e+02, percent-clipped=1.0 2023-05-17 09:47:23,174 INFO [finetune.py:992] (1/2) Epoch 18, batch 850, loss[loss=0.1723, simple_loss=0.2676, pruned_loss=0.03851, over 12018.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2541, pruned_loss=0.03674, over 2346868.06 frames. ], batch size: 40, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:47:43,485 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308858.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:47:58,891 INFO [finetune.py:992] (1/2) Epoch 18, batch 900, loss[loss=0.1833, simple_loss=0.2754, pruned_loss=0.04562, over 11661.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2539, pruned_loss=0.03682, over 2349813.45 frames. ], batch size: 48, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:48:11,876 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.597e+02 3.126e+02 3.640e+02 8.251e+02, threshold=6.251e+02, percent-clipped=4.0 2023-05-17 09:48:18,276 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308906.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:48:19,678 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308908.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:48:35,346 INFO [finetune.py:992] (1/2) Epoch 18, batch 950, loss[loss=0.1427, simple_loss=0.2257, pruned_loss=0.02985, over 12277.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2537, pruned_loss=0.03687, over 2350756.60 frames. ], batch size: 28, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:49:01,241 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308965.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:49:11,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-05-17 09:49:11,610 INFO [finetune.py:992] (1/2) Epoch 18, batch 1000, loss[loss=0.1488, simple_loss=0.2371, pruned_loss=0.03022, over 12337.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2534, pruned_loss=0.03691, over 2356103.98 frames. ], batch size: 30, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:49:24,362 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.709e+02 3.115e+02 3.712e+02 7.348e+02, threshold=6.229e+02, percent-clipped=2.0 2023-05-17 09:49:35,675 INFO [zipformer.py:625] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309013.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:49:44,119 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309025.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:49:47,179 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.2879, 4.7525, 4.0587, 4.9184, 4.4287, 2.8642, 4.1865, 3.1644], device='cuda:1'), covar=tensor([0.0876, 0.0737, 0.1506, 0.0594, 0.1252, 0.1840, 0.1133, 0.3326], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0383, 0.0365, 0.0336, 0.0377, 0.0280, 0.0352, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:49:47,611 INFO [finetune.py:992] (1/2) Epoch 18, batch 1050, loss[loss=0.1482, simple_loss=0.2478, pruned_loss=0.02427, over 12148.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2532, pruned_loss=0.03663, over 2369330.02 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:49:54,803 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([4.9583, 4.6291, 4.7636, 4.8275, 4.6714, 4.9668, 4.7794, 2.6031], device='cuda:1'), covar=tensor([0.0109, 0.0063, 0.0083, 0.0061, 0.0053, 0.0085, 0.0075, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0080, 0.0085, 0.0075, 0.0061, 0.0095, 0.0083, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-05-17 09:50:09,291 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-05-17 09:50:18,143 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-05-17 09:50:23,753 INFO [finetune.py:992] (1/2) Epoch 18, batch 1100, loss[loss=0.1884, simple_loss=0.283, pruned_loss=0.04692, over 12055.00 frames. ], tot_loss[loss=0.163, simple_loss=0.253, pruned_loss=0.03653, over 2375581.95 frames. ], batch size: 37, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:50:29,561 INFO [zipformer.py:625] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309088.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:50:36,383 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 2.166e+02 2.763e+02 3.161e+02 3.707e+02 8.705e+02, threshold=6.323e+02, percent-clipped=2.0 2023-05-17 09:50:41,587 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.3271, 3.1051, 3.0115, 2.9935, 2.6696, 2.5390, 3.0700, 2.2285], device='cuda:1'), covar=tensor([0.0415, 0.0214, 0.0212, 0.0207, 0.0412, 0.0372, 0.0198, 0.0479], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0165, 0.0168, 0.0193, 0.0202, 0.0201, 0.0177, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:50:59,908 INFO [finetune.py:992] (1/2) Epoch 18, batch 1150, loss[loss=0.1428, simple_loss=0.2328, pruned_loss=0.02645, over 11806.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2529, pruned_loss=0.03619, over 2377305.66 frames. ], batch size: 26, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:51:35,937 INFO [finetune.py:992] (1/2) Epoch 18, batch 1200, loss[loss=0.1674, simple_loss=0.2591, pruned_loss=0.03789, over 11282.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2521, pruned_loss=0.03594, over 2371783.12 frames. ], batch size: 55, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:51:37,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-05-17 09:51:39,638 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([5.3380, 6.0803, 5.7632, 5.5996, 6.1968, 5.4228, 5.6111, 5.5954], device='cuda:1'), covar=tensor([0.1346, 0.1007, 0.1252, 0.1984, 0.1027, 0.2340, 0.2054, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0514, 0.0415, 0.0457, 0.0473, 0.0450, 0.0412, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-05-17 09:51:49,451 INFO [optim.py:368] (1/2) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.620e+02 3.079e+02 3.784e+02 9.052e+02, threshold=6.158e+02, percent-clipped=1.0 2023-05-17 09:51:57,027 INFO [zipformer.py:625] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309208.0, num_to_drop=0, layers_to_drop=set() 2023-05-17 09:52:12,689 INFO [finetune.py:992] (1/2) Epoch 18, batch 1250, loss[loss=0.167, simple_loss=0.264, pruned_loss=0.03499, over 12201.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2523, pruned_loss=0.03585, over 2373156.40 frames. ], batch size: 35, lr: 3.26e-03, grad_scale: 8.0 2023-05-17 09:52:13,674 INFO [zipformer.py:1454] (1/2) attn_weights_entropy = tensor([2.5954, 3.6667, 3.3852, 3.2908, 2.9070, 2.8005, 3.6830, 2.3933], device='cuda:1'), covar=tensor([0.0449, 0.0214, 0.0223, 0.0228, 0.0456, 0.0431, 0.0149, 0.0529], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0166, 0.0170, 0.0194, 0.0204, 0.0202, 0.0178, 0.0207], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-05-17 09:52:15,031 INFO [finetune.py:1294] (1/2) Saving batch to pruned_transducer_stateless7/exp_giga_finetune/batch-a689ee27-eec1-83b6-15a8-f48f39643825.pt 2023-05-17 09:52:15,087 INFO [finetune.py:1300] (1/2) features shape: torch.Size([55, 904, 80]) 2023-05-17 09:52:15,089 INFO [finetune.py:1304] (1/2) num tokens: 2397