icefall_asr_wenetspeech_pruned_transducer_stateless5_offline / log /fast_beam_search /log-decode-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8-2022-07-26-14-02-51
luomingshuang's picture
add pruned rnnt5 offline pretrained models
fe6751e
2022-07-26 14:02:51,547 INFO [decode.py:523] Decoding started
2022-07-26 14:02:51,548 INFO [decode.py:529] Device: cuda:0
2022-07-26 14:02:52,790 INFO [lexicon.py:176] Loading pre-compiled data/lang_char/Linv.pt
2022-07-26 14:02:52,879 INFO [decode.py:540] {'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, 'feature_dim': 80, 'subsampling_factor': 4, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'f8d2dba06c000ffee36aab5b66f24e7c9809f116', 'k2-git-date': 'Thu Apr 21 12:20:34 2022', 'lhotse-version': '1.5.0.dev+git.7cce647.dirty', 'torch-version': '1.11.0', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'pruned-rnnt5-for-wenetspeech', 'icefall-git-sha1': '4bf8392-dirty', 'icefall-git-date': 'Mon Jul 25 16:11:59 2022', 'icefall-path': '/ceph-meixu/luomingshuang/icefall', 'k2-path': '/ceph-ms/luomingshuang/k2_latest/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-meixu/luomingshuang/anaconda3/envs/k2-python/lib/python3.8/site-packages/lhotse-1.5.0.dev0+git.7cce647.dirty-py3.8.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-7-0616225511-78bf4545d8-tv52r', 'IP address': '10.177.77.9'}, 'epoch': 4, 'iter': 0, 'avg': 1, 'use_averaged_model': True, 'exp_dir': PosixPath('pruned_transducer_stateless5/exp_L_offline'), 'lang_dir': 'data/lang_char', 'decoding_method': 'fast_beam_search', 'beam_size': 4, 'beam': 4, 'max_contexts': 4, 'max_states': 8, 'context_size': 2, 'max_sym_per_frame': 1, 'simulate_streaming': False, 'decode_chunk_size': 16, 'left_context': 64, 'num_encoder_layers': 24, 'dim_feedforward': 1536, 'nhead': 8, 'encoder_dim': 384, 'decoder_dim': 512, 'joiner_dim': 512, 'dynamic_chunk_training': False, 'causal_convolution': False, 'short_chunk_size': 25, 'num_left_chunks': 4, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 1500, 'bucketing_sampler': True, 'num_buckets': 300, '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('pruned_transducer_stateless5/exp_L_offline/fast_beam_search'), 'suffix': 'epoch-4-avg-1-beam-4-max-contexts-4-max-states-8', 'blank_id': 0, 'vocab_size': 5537}
2022-07-26 14:02:52,879 INFO [decode.py:542] About to create model
2022-07-26 14:02:53,457 INFO [decode.py:609] Calculating the averaged model over epoch range from 3 (excluded) to 4
2022-07-26 14:03:01,749 INFO [decode.py:632] Number of model parameters: 97487351
2022-07-26 14:03:01,753 INFO [asr_datamodule.py:347] About to create dev dataset
2022-07-26 14:03:06,030 INFO [asr_datamodule.py:368] About to create dev dataloader
2022-07-26 14:03:21,926 INFO [decode.py:443] batch 0/?, cuts processed until now is 197
2022-07-26 14:03:27,326 INFO [decode.py:443] batch 2/?, cuts processed until now is 591
2022-07-26 14:03:32,644 INFO [decode.py:443] batch 4/?, cuts processed until now is 1281
2022-07-26 14:03:37,744 INFO [decode.py:443] batch 6/?, cuts processed until now is 2011
2022-07-26 14:03:43,154 INFO [decode.py:443] batch 8/?, cuts processed until now is 2398
2022-07-26 14:03:48,680 INFO [decode.py:443] batch 10/?, cuts processed until now is 2717
2022-07-26 14:03:54,077 INFO [decode.py:443] batch 12/?, cuts processed until now is 3068
2022-07-26 14:03:59,432 INFO [decode.py:443] batch 14/?, cuts processed until now is 3487
2022-07-26 14:04:04,785 INFO [decode.py:443] batch 16/?, cuts processed until now is 3874
2022-07-26 14:04:10,154 INFO [decode.py:443] batch 18/?, cuts processed until now is 4339
2022-07-26 14:04:15,263 INFO [decode.py:443] batch 20/?, cuts processed until now is 4869
2022-07-26 14:04:20,391 INFO [decode.py:443] batch 22/?, cuts processed until now is 5467
2022-07-26 14:04:25,664 INFO [decode.py:443] batch 24/?, cuts processed until now is 6065
2022-07-26 14:04:30,947 INFO [decode.py:443] batch 26/?, cuts processed until now is 6663
2022-07-26 14:04:35,985 INFO [decode.py:443] batch 28/?, cuts processed until now is 7316
2022-07-26 14:04:41,032 INFO [decode.py:443] batch 30/?, cuts processed until now is 7901
2022-07-26 14:04:46,051 INFO [decode.py:443] batch 32/?, cuts processed until now is 8709
2022-07-26 14:04:51,342 INFO [decode.py:443] batch 34/?, cuts processed until now is 9371
2022-07-26 14:04:56,752 INFO [decode.py:443] batch 36/?, cuts processed until now is 9722
2022-07-26 14:05:01,807 INFO [decode.py:443] batch 38/?, cuts processed until now is 10174
2022-07-26 14:05:06,124 INFO [decode.py:443] batch 40/?, cuts processed until now is 10589
2022-07-26 14:05:09,892 INFO [decode.py:443] batch 42/?, cuts processed until now is 11046
2022-07-26 14:05:13,173 INFO [decode.py:443] batch 44/?, cuts processed until now is 11462
2022-07-26 14:05:16,538 INFO [decode.py:443] batch 46/?, cuts processed until now is 11710
2022-07-26 14:05:20,369 INFO [decode.py:443] batch 48/?, cuts processed until now is 11898
2022-07-26 14:05:23,710 INFO [decode.py:443] batch 50/?, cuts processed until now is 12102
2022-07-26 14:05:27,085 INFO [decode.py:443] batch 52/?, cuts processed until now is 12306
2022-07-26 14:05:30,904 INFO [decode.py:443] batch 54/?, cuts processed until now is 12541
2022-07-26 14:05:34,816 INFO [decode.py:443] batch 56/?, cuts processed until now is 13115
2022-07-26 14:05:38,282 INFO [decode.py:443] batch 58/?, cuts processed until now is 13612
2022-07-26 14:05:40,342 INFO [decode.py:443] batch 60/?, cuts processed until now is 13825
2022-07-26 14:05:40,786 INFO [decode.py:460] The transcripts are stored in pruned_transducer_stateless5/exp_L_offline/fast_beam_search/recogs-DEV-beam_4_max_contexts_4_max_states_8-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8.txt
2022-07-26 14:05:41,258 INFO [utils.py:410] [DEV-beam_4_max_contexts_4_max_states_8] %WER 8.29% [27394 / 330498, 2852 ins, 10852 del, 13690 sub ]
2022-07-26 14:05:42,587 INFO [decode.py:473] Wrote detailed error stats to pruned_transducer_stateless5/exp_L_offline/fast_beam_search/errs-DEV-beam_4_max_contexts_4_max_states_8-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8.txt
2022-07-26 14:05:42,588 INFO [decode.py:490]
For DEV, WER of different settings are:
beam_4_max_contexts_4_max_states_8 8.29 best for DEV
2022-07-26 14:05:51,041 INFO [decode.py:443] batch 0/?, cuts processed until now is 218
2022-07-26 14:05:56,346 INFO [decode.py:443] batch 2/?, cuts processed until now is 635
2022-07-26 14:06:01,980 INFO [decode.py:443] batch 4/?, cuts processed until now is 1814
2022-07-26 14:06:07,485 INFO [decode.py:443] batch 6/?, cuts processed until now is 3064
2022-07-26 14:06:12,718 INFO [decode.py:443] batch 8/?, cuts processed until now is 3520
2022-07-26 14:06:19,112 INFO [decode.py:443] batch 10/?, cuts processed until now is 3789
2022-07-26 14:06:24,449 INFO [decode.py:443] batch 12/?, cuts processed until now is 4173
2022-07-26 14:06:29,638 INFO [decode.py:443] batch 14/?, cuts processed until now is 4661
2022-07-26 14:06:36,002 INFO [decode.py:443] batch 16/?, cuts processed until now is 5016
2022-07-26 14:06:41,194 INFO [decode.py:443] batch 18/?, cuts processed until now is 5459
2022-07-26 14:06:46,219 INFO [decode.py:443] batch 20/?, cuts processed until now is 6077
2022-07-26 14:06:52,526 INFO [decode.py:443] batch 22/?, cuts processed until now is 6640
2022-07-26 14:06:57,527 INFO [decode.py:443] batch 24/?, cuts processed until now is 7302
2022-07-26 14:07:03,130 INFO [decode.py:443] batch 26/?, cuts processed until now is 8133
2022-07-26 14:07:09,686 INFO [decode.py:443] batch 28/?, cuts processed until now is 8686
2022-07-26 14:07:15,245 INFO [decode.py:443] batch 30/?, cuts processed until now is 9936
2022-07-26 14:07:20,301 INFO [decode.py:443] batch 32/?, cuts processed until now is 10675
2022-07-26 14:07:25,631 INFO [decode.py:443] batch 34/?, cuts processed until now is 11836
2022-07-26 14:07:30,886 INFO [decode.py:443] batch 36/?, cuts processed until now is 12245
2022-07-26 14:07:37,137 INFO [decode.py:443] batch 38/?, cuts processed until now is 12541
2022-07-26 14:07:43,724 INFO [decode.py:443] batch 40/?, cuts processed until now is 13275
2022-07-26 14:07:49,147 INFO [decode.py:443] batch 42/?, cuts processed until now is 13969
2022-07-26 14:07:55,213 INFO [decode.py:443] batch 44/?, cuts processed until now is 15596
2022-07-26 14:08:00,674 INFO [decode.py:443] batch 46/?, cuts processed until now is 16757
2022-07-26 14:08:06,166 INFO [decode.py:443] batch 48/?, cuts processed until now is 18091
2022-07-26 14:08:11,665 INFO [decode.py:443] batch 50/?, cuts processed until now is 19425
2022-07-26 14:08:16,808 INFO [decode.py:443] batch 52/?, cuts processed until now is 19962
2022-07-26 14:08:20,733 INFO [decode.py:443] batch 54/?, cuts processed until now is 20601
2022-07-26 14:08:25,786 INFO [decode.py:443] batch 56/?, cuts processed until now is 20963
2022-07-26 14:08:29,464 INFO [decode.py:443] batch 58/?, cuts processed until now is 21633
2022-07-26 14:08:33,238 INFO [decode.py:443] batch 60/?, cuts processed until now is 22016
2022-07-26 14:08:36,729 INFO [decode.py:443] batch 62/?, cuts processed until now is 22223
2022-07-26 14:08:38,783 INFO [decode.py:443] batch 64/?, cuts processed until now is 22442
2022-07-26 14:08:44,202 INFO [decode.py:443] batch 66/?, cuts processed until now is 22604
2022-07-26 14:08:50,474 INFO [decode.py:443] batch 68/?, cuts processed until now is 22702
2022-07-26 14:08:53,576 INFO [decode.py:443] batch 70/?, cuts processed until now is 23022
2022-07-26 14:08:58,560 INFO [decode.py:443] batch 72/?, cuts processed until now is 23139
2022-07-26 14:09:01,791 INFO [decode.py:443] batch 74/?, cuts processed until now is 23946
2022-07-26 14:09:03,513 INFO [decode.py:443] batch 76/?, cuts processed until now is 24133
2022-07-26 14:09:06,205 INFO [decode.py:443] batch 78/?, cuts processed until now is 24774
2022-07-26 14:09:06,703 INFO [decode.py:460] The transcripts are stored in pruned_transducer_stateless5/exp_L_offline/fast_beam_search/recogs-TEST_NET-beam_4_max_contexts_4_max_states_8-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8.txt
2022-07-26 14:09:07,402 INFO [utils.py:410] [TEST_NET-beam_4_max_contexts_4_max_states_8] %WER 9.00% [37399 / 415747, 4138 ins, 8380 del, 24881 sub ]
2022-07-26 14:09:08,955 INFO [decode.py:473] Wrote detailed error stats to pruned_transducer_stateless5/exp_L_offline/fast_beam_search/errs-TEST_NET-beam_4_max_contexts_4_max_states_8-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8.txt
2022-07-26 14:09:08,956 INFO [decode.py:490]
For TEST_NET, WER of different settings are:
beam_4_max_contexts_4_max_states_8 9.0 best for TEST_NET
2022-07-26 14:09:15,513 INFO [decode.py:443] batch 0/?, cuts processed until now is 140
2022-07-26 14:09:22,081 INFO [decode.py:443] batch 2/?, cuts processed until now is 404
2022-07-26 14:09:27,913 INFO [decode.py:443] batch 4/?, cuts processed until now is 954
2022-07-26 14:09:33,571 INFO [decode.py:443] batch 6/?, cuts processed until now is 1546
2022-07-26 14:09:39,528 INFO [decode.py:443] batch 8/?, cuts processed until now is 1890
2022-07-26 14:09:46,009 INFO [decode.py:443] batch 10/?, cuts processed until now is 2193
2022-07-26 14:09:52,294 INFO [decode.py:443] batch 12/?, cuts processed until now is 2515
2022-07-26 14:09:58,064 INFO [decode.py:443] batch 14/?, cuts processed until now is 2927
2022-07-26 14:10:04,315 INFO [decode.py:443] batch 16/?, cuts processed until now is 3249
2022-07-26 14:10:10,026 INFO [decode.py:443] batch 18/?, cuts processed until now is 3660
2022-07-26 14:10:18,015 INFO [decode.py:443] batch 20/?, cuts processed until now is 3805
2022-07-26 14:10:23,584 INFO [decode.py:443] batch 22/?, cuts processed until now is 4523
2022-07-26 14:10:29,533 INFO [decode.py:443] batch 24/?, cuts processed until now is 4940
2022-07-26 14:10:34,658 INFO [decode.py:443] batch 26/?, cuts processed until now is 5525
2022-07-26 14:10:40,192 INFO [decode.py:443] batch 28/?, cuts processed until now is 5899
2022-07-26 14:10:45,625 INFO [decode.py:443] batch 30/?, cuts processed until now is 6331
2022-07-26 14:10:49,097 INFO [decode.py:443] batch 32/?, cuts processed until now is 6658
2022-07-26 14:10:52,814 INFO [decode.py:443] batch 34/?, cuts processed until now is 6705
2022-07-26 14:10:56,367 INFO [decode.py:443] batch 36/?, cuts processed until now is 6911
2022-07-26 14:11:00,138 INFO [decode.py:443] batch 38/?, cuts processed until now is 7333
2022-07-26 14:11:03,260 INFO [decode.py:443] batch 40/?, cuts processed until now is 7558
2022-07-26 14:11:07,566 INFO [decode.py:443] batch 42/?, cuts processed until now is 7890
2022-07-26 14:11:10,640 INFO [decode.py:443] batch 44/?, cuts processed until now is 8002
2022-07-26 14:11:14,286 INFO [decode.py:443] batch 46/?, cuts processed until now is 8128
2022-07-26 14:11:18,849 INFO [decode.py:443] batch 48/?, cuts processed until now is 8221
2022-07-26 14:11:22,694 INFO [decode.py:443] batch 50/?, cuts processed until now is 8363
2022-07-26 14:11:24,318 INFO [decode.py:460] The transcripts are stored in pruned_transducer_stateless5/exp_L_offline/fast_beam_search/recogs-TEST_MEETING-beam_4_max_contexts_4_max_states_8-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8.txt
2022-07-26 14:11:24,621 INFO [utils.py:410] [TEST_MEETING-beam_4_max_contexts_4_max_states_8] %WER 14.93% [32908 / 220385, 2417 ins, 17261 del, 13230 sub ]
2022-07-26 14:11:25,431 INFO [decode.py:473] Wrote detailed error stats to pruned_transducer_stateless5/exp_L_offline/fast_beam_search/errs-TEST_MEETING-beam_4_max_contexts_4_max_states_8-epoch-4-avg-1-beam-4-max-contexts-4-max-states-8.txt
2022-07-26 14:11:25,432 INFO [decode.py:490]
For TEST_MEETING, WER of different settings are:
beam_4_max_contexts_4_max_states_8 14.93 best for TEST_MEETING
2022-07-26 14:11:25,432 INFO [decode.py:731] Done!