2023-11-01 14:47:28,310 INFO [ctc_decode.py:562] Decoding started 2023-11-01 14:47:28,310 INFO [ctc_decode.py:568] Device: cuda:0 2023-11-01 14:47:28,311 INFO [ctc_decode.py:569] {'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': '1.24.3', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '821ebc378e7fb99b8adc81950227963332821e01', 'k2-git-date': 'Wed Jul 19 15:38:25 2023', 'lhotse-version': '1.16.0.dev+git.1db4d97a.clean', 'torch-version': '1.11.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.9', 'icefall-git-branch': 'dev_zipformer_cn', 'icefall-git-sha1': '5b9014f7-dirty', 'icefall-git-date': 'Tue Oct 24 16:08:39 2023', 'icefall-path': '/star-home/jinzengrui/lib/miniconda3/envs/dev39/lib/python3.9/site-packages/icefall-1.0-py3.9.egg', 'k2-path': '/star-home/jinzengrui/lib/miniconda3/envs/dev39/lib/python3.9/site-packages/k2-1.24.3.dev20230721+cuda10.2.torch1.11.0-py3.9-linux-x86_64.egg/k2/__init__.py', 'lhotse-path': '/star-home/jinzengrui/lib/miniconda3/envs/dev39/lib/python3.9/site-packages/lhotse-1.16.0.dev0+git.1db4d97a.clean-py3.9.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-9-0208143539-7dbf569d4f-r7nrb', 'IP address': '10.177.13.150'}, 'frame_shift_ms': 10, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'epoch': 20, 'iter': 0, 'avg': 1, 'use_averaged_model': True, 'exp_dir': PosixPath('zipformer/exp-w-ctc-streaming'), 'bpe_model': 'data/lang_bpe_2000/bpe.model', 'lang_dir': PosixPath('data/lang_bpe_2000'), 'context_size': 2, 'decoding_method': 'ctc-decoding', 'num_paths': 100, 'nbest_scale': 1.0, 'num_encoder_layers': '2,2,3,4,3,2', 'downsampling_factor': '1,2,4,8,4,2', 'feedforward_dim': '512,768,1024,1536,1024,768', 'num_heads': '4,4,4,8,4,4', 'encoder_dim': '192,256,384,512,384,256', 'query_head_dim': '32', 'value_head_dim': '12', 'pos_head_dim': '4', 'pos_dim': 48, 'encoder_unmasked_dim': '192,192,256,256,256,192', 'cnn_module_kernel': '31,31,15,15,15,31', 'decoder_dim': 512, 'joiner_dim': 512, 'causal': True, 'chunk_size': '32', 'left_context_frames': '256', 'use_transducer': True, 'use_ctc': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 300.0, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'res_dir': PosixPath('zipformer/exp-w-ctc-streaming/ctc-decoding'), 'suffix': 'epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model'} 2023-11-01 14:47:31,318 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_2000/Linv.pt 2023-11-01 14:47:38,873 INFO [ctc_decode.py:589] About to create model 2023-11-01 14:47:39,840 INFO [ctc_decode.py:656] Calculating the averaged model over epoch range from 19 (excluded) to 20 2023-11-01 14:47:46,736 INFO [ctc_decode.py:673] Number of model parameters: 70213431 2023-11-01 14:47:46,739 INFO [multi_dataset.py:221] About to get multidataset test cuts 2023-11-01 14:47:46,739 INFO [multi_dataset.py:224] Loading Aidatatang_200zh set in lazy mode 2023-11-01 14:47:46,859 INFO [multi_dataset.py:233] Loading Aishell set in lazy mode 2023-11-01 14:47:46,929 INFO [multi_dataset.py:242] Loading Aishell-2 set in lazy mode 2023-11-01 14:47:46,985 INFO [multi_dataset.py:251] Loading Aishell-4 TEST set in lazy mode 2023-11-01 14:47:47,014 INFO [multi_dataset.py:257] Loading Ali-Meeting set in lazy mode 2023-11-01 14:47:47,068 INFO [multi_dataset.py:266] Loading MagicData set in lazy mode 2023-11-01 14:47:47,147 INFO [multi_dataset.py:275] Loading KeSpeech set in lazy mode 2023-11-01 14:47:47,222 INFO [multi_dataset.py:287] Loading WeNetSpeech set in lazy mode 2023-11-01 14:48:01,939 WARNING [ctc_decode.py:685] Excluding cut with ID: TEST_NET_Y0000000004_0ub4ZzdHzBc_S00023 from decoding, num_frames: 8 2023-11-01 14:48:03,851 INFO [ctc_decode.py:697] Start decoding test set: aidatatang_test 2023-11-01 14:48:06,620 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 80 2023-11-01 14:48:41,924 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 9084 2023-11-01 14:49:02,474 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.1871, 2.0377, 3.0515, 3.3628], device='cuda:0') 2023-11-01 14:49:11,593 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.1252, 1.9225, 2.0306, 2.5157], device='cuda:0') 2023-11-01 14:49:15,233 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 18516 2023-11-01 14:49:42,290 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.0425, 3.2445, 3.2294, 1.5114], device='cuda:0') 2023-11-01 14:49:50,217 INFO [ctc_decode.py:487] batch 300/?, cuts processed until now is 28179 2023-11-01 14:50:03,993 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.1128, 3.6159, 2.0588, 3.7947], device='cuda:0') 2023-11-01 14:50:18,380 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.6853, 2.1348, 1.8477, 2.7538], device='cuda:0') 2023-11-01 14:50:25,991 INFO [ctc_decode.py:487] batch 400/?, cuts processed until now is 37667 2023-11-01 14:50:59,583 INFO [ctc_decode.py:487] batch 500/?, cuts processed until now is 46172 2023-11-01 14:51:12,661 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aidatatang_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:51:14,382 INFO [utils.py:565] [aidatatang_test-ctc-decoding] %WER 6.75% [31652 / 468933, 4164 ins, 9320 del, 18168 sub ] 2023-11-01 14:51:18,348 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aidatatang_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:51:18,354 INFO [ctc_decode.py:524] For aidatatang_test, WER of different settings are: ctc-decoding 6.75 best for aidatatang_test 2023-11-01 14:51:18,355 INFO [ctc_decode.py:697] Start decoding test set: aidatatang_dev 2023-11-01 14:51:21,787 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 81 2023-11-01 14:51:38,906 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.7173, 2.0871, 1.5568, 2.2526], device='cuda:0') 2023-11-01 14:51:45,093 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.3668, 1.5783, 1.5058, 1.2721, 1.7073, 1.4607, 1.7798, 1.6686], device='cuda:0') 2023-11-01 14:51:57,029 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 9077 2023-11-01 14:52:10,409 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.1061, 1.0307, 1.6031, 1.4386, 1.6898, 1.3798, 1.5740, 1.3794], device='cuda:0') 2023-11-01 14:52:32,461 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 18432 2023-11-01 14:52:55,144 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aidatatang_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:52:55,904 INFO [utils.py:565] [aidatatang_dev-ctc-decoding] %WER 6.17% [14472 / 234524, 1839 ins, 4827 del, 7806 sub ] 2023-11-01 14:52:57,557 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aidatatang_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:52:57,561 INFO [ctc_decode.py:524] For aidatatang_dev, WER of different settings are: ctc-decoding 6.17 best for aidatatang_dev 2023-11-01 14:52:57,564 INFO [ctc_decode.py:697] Start decoding test set: alimeeting_test 2023-11-01 14:53:01,508 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 44 2023-11-01 14:53:17,820 WARNING [ctc_decode.py:685] Excluding cut with ID: R8008_M8016-8062-123 from decoding, num_frames: 6 2023-11-01 14:53:43,012 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 7625 2023-11-01 14:53:48,593 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.5120, 1.2116, 1.5460, 2.4960], device='cuda:0') 2023-11-01 14:53:56,382 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([0.7824, 1.2077, 1.1181, 1.2473, 0.6089, 1.2157, 1.0652, 1.0424], device='cuda:0') 2023-11-01 14:54:17,387 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-alimeeting_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:54:18,088 INFO [utils.py:565] [alimeeting_test-ctc-decoding] %WER 31.44% [65973 / 209845, 4708 ins, 33281 del, 27984 sub ] 2023-11-01 14:54:20,196 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-alimeeting_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:54:20,201 INFO [ctc_decode.py:524] For alimeeting_test, WER of different settings are: ctc-decoding 31.44 best for alimeeting_test 2023-11-01 14:54:20,201 INFO [ctc_decode.py:697] Start decoding test set: alimeeting_eval 2023-11-01 14:54:22,472 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 35 2023-11-01 14:54:44,130 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.8781, 1.6845, 1.9433, 1.6850, 1.6151, 1.8336, 1.5154, 1.8333], device='cuda:0') 2023-11-01 14:54:51,905 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-alimeeting_eval-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:54:52,179 INFO [utils.py:565] [alimeeting_eval-ctc-decoding] %WER 30.00% [24332 / 81111, 1757 ins, 11903 del, 10672 sub ] 2023-11-01 14:54:53,066 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-alimeeting_eval-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:54:53,070 INFO [ctc_decode.py:524] For alimeeting_eval, WER of different settings are: ctc-decoding 30.0 best for alimeeting_eval 2023-11-01 14:54:53,071 INFO [ctc_decode.py:697] Start decoding test set: aishell_test 2023-11-01 14:54:55,116 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 47 2023-11-01 14:54:55,210 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.4958, 1.9935, 2.6885, 3.1444], device='cuda:0') 2023-11-01 14:55:06,238 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.1795, 0.8899, 1.6425, 1.6988, 1.6322, 1.6657, 1.9442, 1.4633], device='cuda:0') 2023-11-01 14:55:13,489 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.4651, 2.1835, 1.9732, 1.4512, 2.1847, 1.6537, 2.2801, 2.1737], device='cuda:0') 2023-11-01 14:55:24,961 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.2220, 3.3997, 2.1962, 2.3623], device='cuda:0') 2023-11-01 14:55:25,776 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 5468 2023-11-01 14:55:37,012 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aishell_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:55:37,738 INFO [utils.py:565] [aishell_test-ctc-decoding] %WER 4.48% [4692 / 104765, 876 ins, 464 del, 3352 sub ] 2023-11-01 14:55:38,965 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aishell_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:55:38,969 INFO [ctc_decode.py:524] For aishell_test, WER of different settings are: ctc-decoding 4.48 best for aishell_test 2023-11-01 14:55:38,970 INFO [ctc_decode.py:697] Start decoding test set: aishell_dev 2023-11-01 14:55:41,751 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 53 2023-11-01 14:55:44,795 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.9071, 2.1266, 2.2245, 1.9673, 2.1795, 2.0215, 1.7538, 2.0710], device='cuda:0') 2023-11-01 14:55:49,530 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.3511, 2.1230, 3.3090, 3.6794], device='cuda:0') 2023-11-01 14:56:13,400 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.3865, 1.9884, 1.8194, 1.3484, 2.0037, 1.5806, 2.0238, 1.9923], device='cuda:0') 2023-11-01 14:56:16,022 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 6034 2023-11-01 14:56:49,224 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 12198 2023-11-01 14:57:02,734 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aishell_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:57:03,428 INFO [utils.py:565] [aishell_dev-ctc-decoding] %WER 3.80% [7798 / 205341, 1443 ins, 675 del, 5680 sub ] 2023-11-01 14:57:05,310 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aishell_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:57:05,314 INFO [ctc_decode.py:524] For aishell_dev, WER of different settings are: ctc-decoding 3.8 best for aishell_dev 2023-11-01 14:57:05,315 INFO [ctc_decode.py:697] Start decoding test set: aishell-2_test 2023-11-01 14:57:07,480 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 83 2023-11-01 14:57:19,559 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.6574, 1.7279, 1.8256, 2.2562], device='cuda:0') 2023-11-01 14:57:27,882 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aishell-2_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:57:28,053 INFO [utils.py:565] [aishell-2_test-ctc-decoding] %WER 5.18% [2564 / 49532, 299 ins, 164 del, 2101 sub ] 2023-11-01 14:57:28,391 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aishell-2_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:57:28,395 INFO [ctc_decode.py:524] For aishell-2_test, WER of different settings are: ctc-decoding 5.18 best for aishell-2_test 2023-11-01 14:57:28,395 INFO [ctc_decode.py:697] Start decoding test set: aishell-2_dev 2023-11-01 14:57:30,517 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 81 2023-11-01 14:57:40,882 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aishell-2_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:57:40,996 INFO [utils.py:565] [aishell-2_dev-ctc-decoding] %WER 4.69% [1162 / 24802, 117 ins, 64 del, 981 sub ] 2023-11-01 14:57:41,180 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aishell-2_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:57:41,183 INFO [ctc_decode.py:524] For aishell-2_dev, WER of different settings are: ctc-decoding 4.69 best for aishell-2_dev 2023-11-01 14:57:41,184 INFO [ctc_decode.py:697] Start decoding test set: aishell-4 2023-11-01 14:57:45,192 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 33 2023-11-01 14:58:16,573 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.4371, 1.9602, 2.5218, 3.4740], device='cuda:0') 2023-11-01 14:58:23,848 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 5000 2023-11-01 14:58:27,338 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.0054, 1.8766, 2.1849, 2.5415], device='cuda:0') 2023-11-01 14:58:48,415 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.7196, 1.4419, 1.6940, 2.3319], device='cuda:0') 2023-11-01 14:58:53,724 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-aishell-4-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:58:54,660 INFO [utils.py:565] [aishell-4-ctc-decoding] %WER 19.59% [35385 / 180665, 5239 ins, 10885 del, 19261 sub ] 2023-11-01 14:58:56,297 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-aishell-4-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 14:58:56,300 INFO [ctc_decode.py:524] For aishell-4, WER of different settings are: ctc-decoding 19.59 best for aishell-4 2023-11-01 14:58:56,301 INFO [ctc_decode.py:697] Start decoding test set: magicdata_test 2023-11-01 14:58:59,144 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 57 2023-11-01 14:59:03,781 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.4844, 2.5461, 2.2005, 1.8129], device='cuda:0') 2023-11-01 14:59:34,926 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 6425 2023-11-01 14:59:41,444 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.3595, 3.8809, 3.9809, 4.5196], device='cuda:0') 2023-11-01 15:00:10,369 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 13211 2023-11-01 15:00:18,571 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.9834, 1.8399, 1.9684, 2.3691], device='cuda:0') 2023-11-01 15:00:34,434 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.7582, 2.9792, 2.5871, 2.1582], device='cuda:0') 2023-11-01 15:00:41,569 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.7768, 4.4343, 3.7572, 3.8698], device='cuda:0') 2023-11-01 15:00:43,333 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.9072, 1.5017, 1.8388, 2.6391], device='cuda:0') 2023-11-01 15:00:44,628 INFO [ctc_decode.py:487] batch 300/?, cuts processed until now is 20136 2023-11-01 15:01:18,367 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-magicdata_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:01:19,198 INFO [utils.py:565] [magicdata_test-ctc-decoding] %WER 9.83% [23510 / 239091, 1804 ins, 13912 del, 7794 sub ] 2023-11-01 15:01:21,185 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-magicdata_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:01:21,190 INFO [ctc_decode.py:524] For magicdata_test, WER of different settings are: ctc-decoding 9.83 best for magicdata_test 2023-11-01 15:01:21,190 INFO [ctc_decode.py:697] Start decoding test set: magicdata_dev 2023-11-01 15:01:24,288 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 52 2023-11-01 15:01:37,866 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.7325, 4.3410, 2.7236, 4.5414], device='cuda:0') 2023-11-01 15:01:48,856 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.2312, 1.7215, 2.1283, 3.1109], device='cuda:0') 2023-11-01 15:01:56,976 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 5919 2023-11-01 15:02:25,871 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.2261, 1.7625, 2.5821, 3.0766], device='cuda:0') 2023-11-01 15:02:30,480 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 11646 2023-11-01 15:02:33,221 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-magicdata_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:02:33,600 INFO [utils.py:565] [magicdata_dev-ctc-decoding] %WER 10.85% [12676 / 116800, 903 ins, 7031 del, 4742 sub ] 2023-11-01 15:02:34,423 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-magicdata_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:02:34,427 INFO [ctc_decode.py:524] For magicdata_dev, WER of different settings are: ctc-decoding 10.85 best for magicdata_dev 2023-11-01 15:02:34,428 INFO [ctc_decode.py:697] Start decoding test set: kespeech-asr_test 2023-11-01 15:02:37,432 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 45 2023-11-01 15:02:46,001 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.2753, 2.0451, 3.1021, 3.3815], device='cuda:0') 2023-11-01 15:02:53,138 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.1011, 3.3499, 2.1994, 2.2218], device='cuda:0') 2023-11-01 15:03:14,972 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 4867 2023-11-01 15:03:50,988 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 9965 2023-11-01 15:03:56,531 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.3316, 2.5555, 2.7021, 3.2418], device='cuda:0') 2023-11-01 15:04:11,631 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.5980, 2.3235, 2.1412, 1.9799, 2.0662, 2.0400, 2.2173, 1.5323], device='cuda:0') 2023-11-01 15:04:25,088 INFO [ctc_decode.py:487] batch 300/?, cuts processed until now is 15124 2023-11-01 15:05:01,096 INFO [ctc_decode.py:487] batch 400/?, cuts processed until now is 19643 2023-11-01 15:05:02,644 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-kespeech-asr_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:05:03,573 INFO [utils.py:565] [kespeech-asr_test-ctc-decoding] %WER 16.48% [46778 / 283772, 3325 ins, 13461 del, 29992 sub ] 2023-11-01 15:05:05,728 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-kespeech-asr_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:05:05,733 INFO [ctc_decode.py:524] For kespeech-asr_test, WER of different settings are: ctc-decoding 16.48 best for kespeech-asr_test 2023-11-01 15:05:05,733 INFO [ctc_decode.py:697] Start decoding test set: kespeech-asr_dev_phase1 2023-11-01 15:05:07,532 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 44 2023-11-01 15:05:24,275 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-kespeech-asr_dev_phase1-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:05:24,392 INFO [utils.py:565] [kespeech-asr_dev_phase1-ctc-decoding] %WER 14.45% [4572 / 31634, 359 ins, 1475 del, 2738 sub ] 2023-11-01 15:05:24,638 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-kespeech-asr_dev_phase1-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:05:24,643 INFO [ctc_decode.py:524] For kespeech-asr_dev_phase1, WER of different settings are: ctc-decoding 14.45 best for kespeech-asr_dev_phase1 2023-11-01 15:05:24,645 INFO [ctc_decode.py:697] Start decoding test set: kespeech-asr_dev_phase2 2023-11-01 15:05:26,528 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 47 2023-11-01 15:05:26,936 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.2724, 2.7972, 2.2002, 1.7790], device='cuda:0') 2023-11-01 15:05:39,597 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.4564, 3.7457, 3.9398, 4.5298], device='cuda:0') 2023-11-01 15:05:43,397 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-kespeech-asr_dev_phase2-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:05:43,594 INFO [utils.py:565] [kespeech-asr_dev_phase2-ctc-decoding] %WER 8.53% [2723 / 31928, 230 ins, 1390 del, 1103 sub ] 2023-11-01 15:05:43,957 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-kespeech-asr_dev_phase2-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:05:43,961 INFO [ctc_decode.py:524] For kespeech-asr_dev_phase2, WER of different settings are: ctc-decoding 8.53 best for kespeech-asr_dev_phase2 2023-11-01 15:05:43,961 INFO [ctc_decode.py:697] Start decoding test set: wenetspeech-meeting_test 2023-11-01 15:05:46,509 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 28 2023-11-01 15:06:08,526 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.4926, 2.2071, 2.6642, 2.4292, 2.3721, 2.5718, 2.1505, 2.4094], device='cuda:0') 2023-11-01 15:06:27,984 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 3776 2023-11-01 15:06:41,581 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.1046, 1.8950, 2.1569, 1.9787, 1.9776, 2.0899, 1.8083, 1.9956], device='cuda:0') 2023-11-01 15:06:59,966 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.9375, 1.5815, 1.8459, 2.8367], device='cuda:0') 2023-11-01 15:07:06,228 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 8092 2023-11-01 15:07:10,351 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-wenetspeech-meeting_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:07:11,026 INFO [utils.py:565] [wenetspeech-meeting_test-ctc-decoding] %WER 9.73% [21453 / 220385, 2225 ins, 5243 del, 13985 sub ] 2023-11-01 15:07:12,598 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-wenetspeech-meeting_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:07:12,602 INFO [ctc_decode.py:524] For wenetspeech-meeting_test, WER of different settings are: ctc-decoding 9.73 best for wenetspeech-meeting_test 2023-11-01 15:07:12,602 INFO [ctc_decode.py:697] Start decoding test set: wenetspeech-net_test 2023-11-01 15:07:13,069 WARNING [ctc_decode.py:685] Excluding cut with ID: TEST_NET_Y0000000004_0ub4ZzdHzBc_S00023 from decoding, num_frames: 8 2023-11-01 15:07:15,284 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 43 2023-11-01 15:07:19,716 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.6678, 1.3518, 1.8001, 2.7190], device='cuda:0') 2023-11-01 15:07:24,947 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.8001, 4.3394, 2.1421, 4.5083], device='cuda:0') 2023-11-01 15:07:25,339 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.8812, 2.1852, 2.2367, 1.8576, 2.0715, 2.4661, 2.0930, 1.6511], device='cuda:0') 2023-11-01 15:07:25,392 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.9651, 2.0863, 1.8320, 2.9962], device='cuda:0') 2023-11-01 15:07:26,290 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.7713, 2.6142, 1.8248, 2.6063], device='cuda:0') 2023-11-01 15:07:30,476 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.1002, 2.0256, 2.4191, 1.9466, 2.1978, 2.2908, 1.7404, 2.3669], device='cuda:0') 2023-11-01 15:07:54,856 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 7009 2023-11-01 15:07:58,817 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([0.9261, 3.7728, 3.7106, 1.4652], device='cuda:0') 2023-11-01 15:08:08,226 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.4854, 1.3401, 1.6919, 2.5682], device='cuda:0') 2023-11-01 15:08:32,168 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 14995 2023-11-01 15:08:37,643 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.3924, 1.8270, 1.6473, 1.1533, 2.0223, 1.3789, 1.8598, 1.9455], device='cuda:0') 2023-11-01 15:08:41,772 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([2.0806, 2.0476, 2.4033, 2.0766, 2.1347, 2.1956, 1.8576, 2.2775], device='cuda:0') 2023-11-01 15:08:48,253 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.8197, 2.5915, 1.8798, 2.7089], device='cuda:0') 2023-11-01 15:09:07,856 INFO [ctc_decode.py:487] batch 300/?, cuts processed until now is 22693 2023-11-01 15:09:18,835 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-wenetspeech-net_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:09:21,819 INFO [utils.py:565] [wenetspeech-net_test-ctc-decoding] %WER 12.03% [49994 / 415746, 3354 ins, 20928 del, 25712 sub ] 2023-11-01 15:09:25,204 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-wenetspeech-net_test-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:09:25,213 INFO [ctc_decode.py:524] For wenetspeech-net_test, WER of different settings are: ctc-decoding 12.03 best for wenetspeech-net_test 2023-11-01 15:09:25,215 INFO [ctc_decode.py:697] Start decoding test set: wenetspeech_dev 2023-11-01 15:09:28,056 INFO [ctc_decode.py:487] batch 0/?, cuts processed until now is 39 2023-11-01 15:10:02,475 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.8177, 2.9916, 1.8380, 1.7392], device='cuda:0') 2023-11-01 15:10:05,218 INFO [ctc_decode.py:487] batch 100/?, cuts processed until now is 4983 2023-11-01 15:10:06,864 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([0.8310, 3.8471, 3.8342, 1.5184], device='cuda:0') 2023-11-01 15:10:14,365 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([1.9438, 2.0733, 2.3426, 2.7923], device='cuda:0') 2023-11-01 15:10:16,598 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.6166, 2.8729, 1.9620, 2.0345], device='cuda:0') 2023-11-01 15:10:41,636 INFO [ctc_decode.py:487] batch 200/?, cuts processed until now is 10268 2023-11-01 15:10:47,303 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([3.5680, 1.8801, 2.7909, 3.2492], device='cuda:0') 2023-11-01 15:10:52,045 INFO [zipformer.py:1858] name=None, attn_weights_entropy = tensor([4.2542, 3.5412, 1.9220, 1.7750], device='cuda:0') 2023-11-01 15:11:04,812 INFO [ctc_decode.py:501] The transcripts are stored in zipformer/exp-w-ctc-streaming/ctc-decoding/recogs-wenetspeech_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:11:05,947 INFO [utils.py:565] [wenetspeech_dev-ctc-decoding] %WER 10.24% [33842 / 330498, 1990 ins, 18068 del, 13784 sub ] 2023-11-01 15:11:08,184 INFO [ctc_decode.py:510] Wrote detailed error stats to zipformer/exp-w-ctc-streaming/ctc-decoding/errs-wenetspeech_dev-epoch-20-avg-1-chunk-32-left-context-256-use-averaged-model.txt 2023-11-01 15:11:08,189 INFO [ctc_decode.py:524] For wenetspeech_dev, WER of different settings are: ctc-decoding 10.24 best for wenetspeech_dev 2023-11-01 15:11:08,190 INFO [ctc_decode.py:716] Done!