2023-06-15 10:40:32,531 INFO [decode.py:685] Decoding started
2023-06-15 10:40:32,531 INFO [decode.py:691] Device: cuda:0
2023-06-15 10:40:34,119 INFO [lexicon.py:168] Loading pre-compiled data/lang_char/Linv.pt
2023-06-15 10:40:34,422 INFO [decode.py:702] {'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': '1.24.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'c51a0b9684442a88ee37f3ce0af686a04b66855b', 'k2-git-date': 'Mon May 1 21:38:03 2023', 'lhotse-version': '1.14.0.dev+git.0f812851.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'zipformer_wenetspeech', 'icefall-git-sha1': '28d3f6d-clean', 'icefall-git-date': 'Thu Jun 15 10:30:34 2023', 'icefall-path': '/star-kw/kangwei/code/icefall_wenetspeech', 'k2-path': '/ceph-hw/kangwei/code/k2_release/k2/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-hw/kangwei/dev_tools/anaconda3/envs/rnnt2/lib/python3.8/site-packages/lhotse-1.14.0.dev0+git.0f812851.dirty-py3.8.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-10-0221105906-5745685d6b-t8zzx', 'IP address': '10.177.57.19'}, 'epoch': 4, 'iter': 0, 'avg': 1, 'use_averaged_model': True, 'exp_dir': PosixPath('zipformer/exp_L_context_2'), 'lang_dir': PosixPath('data/lang_char'), 'decoding_method': 'modified_beam_search', 'beam_size': 4, 'beam': 20.0, 'ngram_lm_scale': 0.01, 'max_contexts': 8, 'max_states': 64, 'context_size': 2, 'max_sym_per_frame': 1, 'num_paths': 200, 'nbest_scale': 0.5, 'blank_penalty': 2.0, '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': False, 'chunk_size': '16,32,64,-1', 'left_context_frames': '64,128,256,-1', 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 1000, '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, 'training_subset': 'L', 'res_dir': PosixPath('zipformer/exp_L_context_2/modified_beam_search'), 'suffix': 'epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model', 'blank_id': 0, 'vocab_size': 5537}
2023-06-15 10:40:34,423 INFO [decode.py:704] About to create model
2023-06-15 10:40:35,106 INFO [decode.py:771] Calculating the averaged model over epoch range from 3 (excluded) to 4
2023-06-15 10:40:48,495 INFO [decode.py:802] Number of model parameters: 75879898
2023-06-15 10:40:48,496 INFO [asr_datamodule.py:398] About to get dev cuts
2023-06-15 10:40:48,513 INFO [asr_datamodule.py:336] About to create dev dataset
2023-06-15 10:40:49,073 INFO [asr_datamodule.py:354] About to create dev dataloader
2023-06-15 10:40:49,074 INFO [asr_datamodule.py:403] About to get TEST_NET cuts
2023-06-15 10:40:49,085 INFO [asr_datamodule.py:367] About to create test dataset
2023-06-15 10:40:49,136 WARNING [decode.py:811] Exclude cut with ID TEST_NET_Y0000000004_0ub4ZzdHzBc_S00023 from decoding, num_frames : 8.
2023-06-15 10:40:49,664 INFO [asr_datamodule.py:408] About to get TEST_MEETING cuts
2023-06-15 10:40:49,671 INFO [asr_datamodule.py:367] About to create test dataset
2023-06-15 10:41:00,537 INFO [decode.py:581] batch 0/?, cuts processed until now is 130
2023-06-15 10:43:38,395 INFO [decode.py:581] batch 20/?, cuts processed until now is 3192
2023-06-15 10:46:11,954 INFO [decode.py:581] batch 40/?, cuts processed until now is 6421
2023-06-15 10:48:44,660 INFO [decode.py:581] batch 60/?, cuts processed until now is 10176
2023-06-15 10:50:44,867 INFO [decode.py:581] batch 80/?, cuts processed until now is 13727
2023-06-15 10:50:51,948 INFO [decode.py:597] The transcripts are stored in zipformer/exp_L_context_2/modified_beam_search/recogs-DEV-beam_size_4_blank_penalty_2.0-epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model.txt
2023-06-15 10:50:52,333 INFO [utils.py:562] [DEV-beam_size_4_blank_penalty_2.0] %WER 7.75% [25607 / 330498, 2701 ins, 9369 del, 13537 sub ]
2023-06-15 10:50:53,344 INFO [decode.py:610] Wrote detailed error stats to zipformer/exp_L_context_2/modified_beam_search/errs-DEV-beam_size_4_blank_penalty_2.0-epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model.txt
2023-06-15 10:50:53,347 INFO [decode.py:626] 
For DEV, WER of different settings are:
beam_size_4_blank_penalty_2.0	7.75	best for DEV

2023-06-15 10:50:53,740 WARNING [decode.py:811] Exclude cut with ID TEST_NET_Y0000000004_0ub4ZzdHzBc_S00023 from decoding, num_frames : 8.
2023-06-15 10:51:02,926 INFO [decode.py:581] batch 0/?, cuts processed until now is 146
2023-06-15 10:53:24,884 INFO [decode.py:581] batch 20/?, cuts processed until now is 4116
2023-06-15 10:54:16,808 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([1.6708, 3.2472, 2.3909, 1.6948], device='cuda:0')
2023-06-15 10:55:52,912 INFO [decode.py:581] batch 40/?, cuts processed until now is 8601
2023-06-15 10:57:59,652 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([3.4302, 2.1882, 1.9218, 2.0420], device='cuda:0')
2023-06-15 10:58:14,384 INFO [decode.py:581] batch 60/?, cuts processed until now is 14082
2023-06-15 11:00:40,619 INFO [decode.py:581] batch 80/?, cuts processed until now is 18750
2023-06-15 11:02:27,662 INFO [decode.py:581] batch 100/?, cuts processed until now is 24487
2023-06-15 11:02:33,797 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([1.6972, 2.0084, 1.8694, 1.7287, 2.0439, 1.8600, 2.0036, 2.1735],
       device='cuda:0')
2023-06-15 11:02:42,857 INFO [decode.py:597] The transcripts are stored in zipformer/exp_L_context_2/modified_beam_search/recogs-TEST_NET-beam_size_4_blank_penalty_2.0-epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model.txt
2023-06-15 11:02:43,368 INFO [utils.py:562] [TEST_NET-beam_size_4_blank_penalty_2.0] %WER 8.81% [36627 / 415746, 3892 ins, 7732 del, 25003 sub ]
2023-06-15 11:02:44,713 INFO [decode.py:610] Wrote detailed error stats to zipformer/exp_L_context_2/modified_beam_search/errs-TEST_NET-beam_size_4_blank_penalty_2.0-epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model.txt
2023-06-15 11:02:44,717 INFO [decode.py:626] 
For TEST_NET, WER of different settings are:
beam_size_4_blank_penalty_2.0	8.81	best for TEST_NET

2023-06-15 11:02:54,221 INFO [decode.py:581] batch 0/?, cuts processed until now is 93
2023-06-15 11:03:07,981 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([2.0809, 3.3829, 2.4716, 4.2258], device='cuda:0')
2023-06-15 11:05:22,695 INFO [decode.py:581] batch 20/?, cuts processed until now is 2345
2023-06-15 11:07:55,098 INFO [decode.py:581] batch 40/?, cuts processed until now is 4929
2023-06-15 11:10:00,446 INFO [decode.py:581] batch 60/?, cuts processed until now is 7955
2023-06-15 11:10:30,188 INFO [decode.py:597] The transcripts are stored in zipformer/exp_L_context_2/modified_beam_search/recogs-TEST_MEETING-beam_size_4_blank_penalty_2.0-epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model.txt
2023-06-15 11:10:30,445 INFO [utils.py:562] [TEST_MEETING-beam_size_4_blank_penalty_2.0] %WER 13.67% [30121 / 220385, 3284 ins, 12079 del, 14758 sub ]
2023-06-15 11:10:31,135 INFO [decode.py:610] Wrote detailed error stats to zipformer/exp_L_context_2/modified_beam_search/errs-TEST_MEETING-beam_size_4_blank_penalty_2.0-epoch-4-avg-1-modified_beam_search-beam-size-4-blank-penalty-2.0-use-averaged-model.txt
2023-06-15 11:10:31,138 INFO [decode.py:626] 
For TEST_MEETING, WER of different settings are:
beam_size_4_blank_penalty_2.0	13.67	best for TEST_MEETING

2023-06-15 11:10:31,138 INFO [decode.py:849] Done!