upload the decoding log

Browse files

Files changed (8) hide show

log/log-1best/log-decode-2022-07-20-21-02-14 +27 -0
log/log-attention-decoder/log-decode-2022-07-20-21-02-14 +1199 -0
log/log-ctc-decoding/log-decode-2022-07-21-23-23-55 +27 -0
log/log-ctc-greedy-search/log-decode-2022-07-20-15-54-39 +27 -0
log/log-nbest-oracle/log-decode-2022-07-20-20-48-51 +27 -0
log/log-nbest-rescoring/log-decode-2022-07-20-21-16-48 +192 -0
log/log-nbest/log-decode-2022-07-20-21-02-14 +48 -0
log/log-whole-lattice-rescoring/log-decode-2022-07-20-21-02-14 +207 -0

log/log-1best/log-decode-2022-07-20-21-02-14 ADDED Viewed

	@@ -0,0 +1,27 @@

+2022-07-20 21:02:14,517 INFO [decode.py:653] Decoding started
+2022-07-20 21:02:14,517 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'd812f82-dirty', 'icefall-git-date': 'Wed Jul 20 19:47:36 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/open_source/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': '1best', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 200, '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'}
+2022-07-20 21:02:14,924 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 21:02:14,985 INFO [decode.py:664] device: cuda:0
+2022-07-20 21:02:21,902 INFO [decode.py:821] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 21:02:24,825 INFO [decode.py:837] Number of model parameters: 103071035
+2022-07-20 21:02:24,825 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 21:02:24,840 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 21:02:26,027 INFO [decode.py:588] batch 0/?, cuts processed until now is 15
+2022-07-20 21:02:52,011 INFO [decode.py:588] batch 100/?, cuts processed until now is 2578
+2022-07-20 21:02:52,790 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-no_rescore.txt
+2022-07-20 21:02:52,871 INFO [utils.py:416] [test-clean-no_rescore] %WER 2.93% [1542 / 52576, 148 ins, 217 del, 1177 sub ]
+2022-07-20 21:02:53,105 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-no_rescore.txt
+2022-07-20 21:02:53,106 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+no_rescore	2.93	best for test-clean
+2022-07-20 21:02:54,150 INFO [decode.py:588] batch 0/?, cuts processed until now is 19
+2022-07-20 21:03:19,172 INFO [decode.py:588] batch 100/?, cuts processed until now is 2888
+2022-07-20 21:03:19,539 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-no_rescore.txt
+2022-07-20 21:03:19,620 INFO [utils.py:416] [test-other-no_rescore] %WER 6.37% [3332 / 52343, 258 ins, 592 del, 2482 sub ]
+2022-07-20 21:03:19,850 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-no_rescore.txt
+2022-07-20 21:03:19,851 INFO [decode.py:637]
+For test-other, WER of different settings are:
+no_rescore	6.37	best for test-other
+2022-07-20 21:03:19,851 INFO [decode.py:868] Done!

log/log-attention-decoder/log-decode-2022-07-20-21-02-14 ADDED Viewed

	@@ -0,0 +1,1199 @@

+2022-07-20 21:02:14,406 INFO [decode.py:653] Decoding started
+2022-07-20 21:02:14,406 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'd812f82-dirty', 'icefall-git-date': 'Wed Jul 20 19:47:36 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/open_source/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'attention-decoder', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 20, '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'}
+2022-07-20 21:02:14,821 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 21:02:14,886 INFO [decode.py:664] device: cuda:0
+2022-07-20 21:02:21,301 INFO [decode.py:730] Loading pre-compiled G_4_gram.pt
+2022-07-20 21:02:26,829 INFO [decode.py:821] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 21:02:29,635 INFO [decode.py:837] Number of model parameters: 103071035
+2022-07-20 21:02:29,635 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 21:02:29,639 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 21:02:31,600 INFO [decode.py:588] batch 0/?, cuts processed until now is 1
+2022-07-20 21:04:08,163 INFO [decode.py:588] batch 100/?, cuts processed until now is 212
+2022-07-20 21:05:41,871 INFO [decode.py:588] batch 200/?, cuts processed until now is 459
+2022-07-20 21:07:13,153 INFO [decode.py:588] batch 300/?, cuts processed until now is 717
+2022-07-20 21:08:43,103 INFO [decode.py:588] batch 400/?, cuts processed until now is 949
+2022-07-20 21:10:19,640 INFO [decode.py:588] batch 500/?, cuts processed until now is 1163
+2022-07-20 21:11:51,100 INFO [decode.py:588] batch 600/?, cuts processed until now is 1411
+2022-07-20 21:13:20,028 INFO [decode.py:588] batch 700/?, cuts processed until now is 1652
+2022-07-20 21:14:53,913 INFO [decode.py:588] batch 800/?, cuts processed until now is 1867
+2022-07-20 21:16:20,862 INFO [decode.py:588] batch 900/?, cuts processed until now is 2122
+2022-07-20 21:17:51,185 INFO [decode.py:588] batch 1000/?, cuts processed until now is 2321
+2022-07-20 21:19:26,869 INFO [decode.py:588] batch 1100/?, cuts processed until now is 2468
+2022-07-20 21:20:57,079 INFO [decode.py:588] batch 1200/?, cuts processed until now is 2601
+2022-07-20 21:24:52,558 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+ngram_lm_scale_0.6_attention_scale_0.5	2.59	best for test-clean
+ngram_lm_scale_0.7_attention_scale_1.1	2.59
+ngram_lm_scale_0.7_attention_scale_1.2	2.59
+ngram_lm_scale_0.7_attention_scale_1.3	2.59
+ngram_lm_scale_0.9_attention_scale_0.5	2.59
+ngram_lm_scale_0.9_attention_scale_0.6	2.59
+ngram_lm_scale_0.9_attention_scale_1.3	2.59
+ngram_lm_scale_1.0_attention_scale_0.6	2.59
+ngram_lm_scale_1.0_attention_scale_0.7	2.59
+ngram_lm_scale_1.0_attention_scale_1.5	2.59
+ngram_lm_scale_1.1_attention_scale_0.6	2.59
+ngram_lm_scale_1.1_attention_scale_0.9	2.59
+ngram_lm_scale_1.2_attention_scale_1.1	2.59
+ngram_lm_scale_1.5_attention_scale_1.5	2.59
+ngram_lm_scale_0.6_attention_scale_0.9	2.6
+ngram_lm_scale_0.6_attention_scale_1.0	2.6
+ngram_lm_scale_0.6_attention_scale_1.1	2.6
+ngram_lm_scale_0.6_attention_scale_1.2	2.6
+ngram_lm_scale_0.6_attention_scale_1.3	2.6
+ngram_lm_scale_0.6_attention_scale_1.9	2.6
+ngram_lm_scale_0.6_attention_scale_2.0	2.6
+ngram_lm_scale_0.7_attention_scale_0.5	2.6
+ngram_lm_scale_0.7_attention_scale_0.6	2.6
+ngram_lm_scale_0.7_attention_scale_0.7	2.6
+ngram_lm_scale_0.7_attention_scale_0.9	2.6
+ngram_lm_scale_0.7_attention_scale_1.0	2.6
+ngram_lm_scale_0.7_attention_scale_2.3	2.6
+ngram_lm_scale_0.9_attention_scale_0.7	2.6
+ngram_lm_scale_0.9_attention_scale_0.9	2.6
+ngram_lm_scale_0.9_attention_scale_1.0	2.6
+ngram_lm_scale_0.9_attention_scale_1.1	2.6
+ngram_lm_scale_0.9_attention_scale_1.2	2.6
+ngram_lm_scale_0.9_attention_scale_1.5	2.6
+ngram_lm_scale_0.9_attention_scale_1.7	2.6
+ngram_lm_scale_0.9_attention_scale_1.9	2.6
+ngram_lm_scale_1.0_attention_scale_0.9	2.6
+ngram_lm_scale_1.0_attention_scale_1.0	2.6
+ngram_lm_scale_1.0_attention_scale_1.1	2.6
+ngram_lm_scale_1.0_attention_scale_1.7	2.6
+ngram_lm_scale_1.1_attention_scale_0.7	2.6
+ngram_lm_scale_1.1_attention_scale_1.0	2.6
+ngram_lm_scale_1.1_attention_scale_1.7	2.6
+ngram_lm_scale_1.1_attention_scale_1.9	2.6
+ngram_lm_scale_1.1_attention_scale_2.0	2.6
+ngram_lm_scale_1.1_attention_scale_2.1	2.6
+ngram_lm_scale_1.1_attention_scale_2.2	2.6
+ngram_lm_scale_1.2_attention_scale_0.9	2.6
+ngram_lm_scale_1.2_attention_scale_1.0	2.6
+ngram_lm_scale_1.2_attention_scale_2.2	2.6
+ngram_lm_scale_1.2_attention_scale_2.3	2.6
+ngram_lm_scale_1.3_attention_scale_1.0	2.6
+ngram_lm_scale_1.3_attention_scale_1.1	2.6
+ngram_lm_scale_1.3_attention_scale_1.2	2.6
+ngram_lm_scale_1.3_attention_scale_1.3	2.6
+ngram_lm_scale_1.5_attention_scale_1.7	2.6
+ngram_lm_scale_1.5_attention_scale_2.1	2.6
+ngram_lm_scale_1.7_attention_scale_1.9	2.6
+ngram_lm_scale_0.5_attention_scale_1.2	2.61
+ngram_lm_scale_0.5_attention_scale_1.3	2.61
+ngram_lm_scale_0.5_attention_scale_1.5	2.61
+ngram_lm_scale_0.5_attention_scale_1.7	2.61
+ngram_lm_scale_0.5_attention_scale_1.9	2.61
+ngram_lm_scale_0.5_attention_scale_2.0	2.61
+ngram_lm_scale_0.5_attention_scale_2.2	2.61
+ngram_lm_scale_0.6_attention_scale_0.6	2.61
+ngram_lm_scale_0.6_attention_scale_0.7	2.61
+ngram_lm_scale_0.6_attention_scale_1.5	2.61
+ngram_lm_scale_0.6_attention_scale_1.7	2.61
+ngram_lm_scale_0.6_attention_scale_2.1	2.61
+ngram_lm_scale_0.6_attention_scale_2.2	2.61
+ngram_lm_scale_0.6_attention_scale_2.3	2.61
+ngram_lm_scale_0.6_attention_scale_2.5	2.61
+ngram_lm_scale_0.7_attention_scale_0.3	2.61
+ngram_lm_scale_0.7_attention_scale_1.5	2.61
+ngram_lm_scale_0.7_attention_scale_1.7	2.61
+ngram_lm_scale_0.7_attention_scale_1.9	2.61
+ngram_lm_scale_0.7_attention_scale_2.2	2.61
+ngram_lm_scale_0.7_attention_scale_2.5	2.61
+ngram_lm_scale_0.9_attention_scale_2.0	2.61
+ngram_lm_scale_0.9_attention_scale_2.1	2.61
+ngram_lm_scale_0.9_attention_scale_2.2	2.61
+ngram_lm_scale_0.9_attention_scale_2.3	2.61
+ngram_lm_scale_0.9_attention_scale_2.5	2.61
+ngram_lm_scale_0.9_attention_scale_3.0	2.61
+ngram_lm_scale_1.0_attention_scale_0.5	2.61
+ngram_lm_scale_1.0_attention_scale_1.2	2.61
+ngram_lm_scale_1.0_attention_scale_1.3	2.61
+ngram_lm_scale_1.0_attention_scale_1.9	2.61
+ngram_lm_scale_1.0_attention_scale_2.2	2.61
+ngram_lm_scale_1.0_attention_scale_2.3	2.61
+ngram_lm_scale_1.0_attention_scale_2.5	2.61
+ngram_lm_scale_1.0_attention_scale_3.0	2.61
+ngram_lm_scale_1.1_attention_scale_1.1	2.61
+ngram_lm_scale_1.1_attention_scale_1.2	2.61
+ngram_lm_scale_1.1_attention_scale_1.3	2.61
+ngram_lm_scale_1.1_attention_scale_1.5	2.61
+ngram_lm_scale_1.1_attention_scale_2.3	2.61
+ngram_lm_scale_1.2_attention_scale_0.7	2.61
+ngram_lm_scale_1.2_attention_scale_1.2	2.61
+ngram_lm_scale_1.2_attention_scale_1.3	2.61
+ngram_lm_scale_1.2_attention_scale_1.7	2.61
+ngram_lm_scale_1.2_attention_scale_1.9	2.61
+ngram_lm_scale_1.2_attention_scale_2.0	2.61
+ngram_lm_scale_1.2_attention_scale_2.1	2.61
+ngram_lm_scale_1.2_attention_scale_2.5	2.61
+ngram_lm_scale_1.2_attention_scale_5.0	2.61
+ngram_lm_scale_1.3_attention_scale_0.9	2.61
+ngram_lm_scale_1.3_attention_scale_1.5	2.61
+ngram_lm_scale_1.3_attention_scale_1.7	2.61
+ngram_lm_scale_1.3_attention_scale_1.9	2.61
+ngram_lm_scale_1.3_attention_scale_2.2	2.61
+ngram_lm_scale_1.3_attention_scale_2.3	2.61
+ngram_lm_scale_1.3_attention_scale_2.5	2.61
+ngram_lm_scale_1.3_attention_scale_3.0	2.61
+ngram_lm_scale_1.3_attention_scale_5.0	2.61
+ngram_lm_scale_1.5_attention_scale_1.3	2.61
+ngram_lm_scale_1.5_attention_scale_1.9	2.61
+ngram_lm_scale_1.5_attention_scale_2.0	2.61
+ngram_lm_scale_1.5_attention_scale_2.2	2.61
+ngram_lm_scale_1.5_attention_scale_2.3	2.61
+ngram_lm_scale_1.5_attention_scale_2.5	2.61
+ngram_lm_scale_1.5_attention_scale_3.0	2.61
+ngram_lm_scale_1.5_attention_scale_5.0	2.61
+ngram_lm_scale_1.7_attention_scale_2.0	2.61
+ngram_lm_scale_1.7_attention_scale_2.1	2.61
+ngram_lm_scale_1.7_attention_scale_2.2	2.61
+ngram_lm_scale_1.7_attention_scale_2.3	2.61
+ngram_lm_scale_1.7_attention_scale_2.5	2.61
+ngram_lm_scale_1.7_attention_scale_5.0	2.61
+ngram_lm_scale_2.0_attention_scale_2.5	2.61
+ngram_lm_scale_2.3_attention_scale_4.0	2.61
+ngram_lm_scale_2.5_attention_scale_5.0	2.61
+ngram_lm_scale_0.5_attention_scale_0.3	2.62
+ngram_lm_scale_0.5_attention_scale_0.6	2.62
+ngram_lm_scale_0.5_attention_scale_0.7	2.62
+ngram_lm_scale_0.5_attention_scale_0.9	2.62
+ngram_lm_scale_0.5_attention_scale_1.0	2.62
+ngram_lm_scale_0.5_attention_scale_1.1	2.62
+ngram_lm_scale_0.5_attention_scale_2.1	2.62
+ngram_lm_scale_0.5_attention_scale_2.3	2.62
+ngram_lm_scale_0.5_attention_scale_2.5	2.62
+ngram_lm_scale_0.5_attention_scale_3.0	2.62
+ngram_lm_scale_0.6_attention_scale_0.3	2.62
+ngram_lm_scale_0.6_attention_scale_3.0	2.62
+ngram_lm_scale_0.7_attention_scale_2.0	2.62
+ngram_lm_scale_0.7_attention_scale_2.1	2.62
+ngram_lm_scale_0.7_attention_scale_3.0	2.62
+ngram_lm_scale_0.9_attention_scale_4.0	2.62
+ngram_lm_scale_1.0_attention_scale_2.0	2.62
+ngram_lm_scale_1.0_attention_scale_2.1	2.62
+ngram_lm_scale_1.0_attention_scale_4.0	2.62
+ngram_lm_scale_1.0_attention_scale_5.0	2.62
+ngram_lm_scale_1.1_attention_scale_0.5	2.62
+ngram_lm_scale_1.1_attention_scale_2.5	2.62
+ngram_lm_scale_1.1_attention_scale_3.0	2.62
+ngram_lm_scale_1.1_attention_scale_4.0	2.62
+ngram_lm_scale_1.1_attention_scale_5.0	2.62
+ngram_lm_scale_1.2_attention_scale_1.5	2.62
+ngram_lm_scale_1.2_attention_scale_3.0	2.62
+ngram_lm_scale_1.2_attention_scale_4.0	2.62
+ngram_lm_scale_1.3_attention_scale_2.0	2.62
+ngram_lm_scale_1.3_attention_scale_2.1	2.62
+ngram_lm_scale_1.3_attention_scale_4.0	2.62
+ngram_lm_scale_1.5_attention_scale_4.0	2.62
+ngram_lm_scale_1.7_attention_scale_1.7	2.62
+ngram_lm_scale_1.7_attention_scale_3.0	2.62
+ngram_lm_scale_1.7_attention_scale_4.0	2.62
+ngram_lm_scale_1.9_attention_scale_1.9	2.62
+ngram_lm_scale_1.9_attention_scale_2.0	2.62
+ngram_lm_scale_1.9_attention_scale_2.1	2.62
+ngram_lm_scale_1.9_attention_scale_2.2	2.62
+ngram_lm_scale_1.9_attention_scale_2.3	2.62
+ngram_lm_scale_1.9_attention_scale_2.5	2.62
+ngram_lm_scale_1.9_attention_scale_3.0	2.62
+ngram_lm_scale_1.9_attention_scale_4.0	2.62
+ngram_lm_scale_1.9_attention_scale_5.0	2.62
+ngram_lm_scale_2.0_attention_scale_2.0	2.62
+ngram_lm_scale_2.0_attention_scale_2.1	2.62
+ngram_lm_scale_2.0_attention_scale_2.2	2.62
+ngram_lm_scale_2.0_attention_scale_2.3	2.62
+ngram_lm_scale_2.0_attention_scale_3.0	2.62
+ngram_lm_scale_2.0_attention_scale_4.0	2.62
+ngram_lm_scale_2.0_attention_scale_5.0	2.62
+ngram_lm_scale_2.1_attention_scale_4.0	2.62
+ngram_lm_scale_2.1_attention_scale_5.0	2.62
+ngram_lm_scale_2.2_attention_scale_3.0	2.62
+ngram_lm_scale_2.2_attention_scale_4.0	2.62
+ngram_lm_scale_2.2_attention_scale_5.0	2.62
+ngram_lm_scale_2.3_attention_scale_3.0	2.62
+ngram_lm_scale_2.3_attention_scale_5.0	2.62
+ngram_lm_scale_0.3_attention_scale_1.5	2.63
+ngram_lm_scale_0.5_attention_scale_0.5	2.63
+ngram_lm_scale_0.6_attention_scale_4.0	2.63
+ngram_lm_scale_0.7_attention_scale_4.0	2.63
+ngram_lm_scale_0.7_attention_scale_5.0	2.63
+ngram_lm_scale_0.9_attention_scale_5.0	2.63
+ngram_lm_scale_1.2_attention_scale_0.6	2.63
+ngram_lm_scale_1.5_attention_scale_1.2	2.63
+ngram_lm_scale_1.7_attention_scale_1.5	2.63
+ngram_lm_scale_2.0_attention_scale_1.9	2.63
+ngram_lm_scale_2.1_attention_scale_2.1	2.63
+ngram_lm_scale_2.1_attention_scale_2.2	2.63
+ngram_lm_scale_2.1_attention_scale_2.3	2.63
+ngram_lm_scale_2.1_attention_scale_2.5	2.63
+ngram_lm_scale_2.1_attention_scale_3.0	2.63
+ngram_lm_scale_2.2_attention_scale_2.3	2.63
+ngram_lm_scale_2.3_attention_scale_2.5	2.63
+ngram_lm_scale_3.0_attention_scale_4.0	2.63
+ngram_lm_scale_3.0_attention_scale_5.0	2.63
+ngram_lm_scale_0.3_attention_scale_0.5	2.64
+ngram_lm_scale_0.3_attention_scale_0.6	2.64
+ngram_lm_scale_0.3_attention_scale_0.9	2.64
+ngram_lm_scale_0.3_attention_scale_1.2	2.64
+ngram_lm_scale_0.3_attention_scale_1.3	2.64
+ngram_lm_scale_0.3_attention_scale_1.7	2.64
+ngram_lm_scale_0.3_attention_scale_1.9	2.64
+ngram_lm_scale_0.5_attention_scale_0.08	2.64
+ngram_lm_scale_0.5_attention_scale_0.1	2.64
+ngram_lm_scale_0.5_attention_scale_4.0	2.64
+ngram_lm_scale_0.9_attention_scale_0.3	2.64
+ngram_lm_scale_1.2_attention_scale_0.5	2.64
+ngram_lm_scale_1.3_attention_scale_0.7	2.64
+ngram_lm_scale_1.5_attention_scale_1.1	2.64
+ngram_lm_scale_1.9_attention_scale_1.7	2.64
+ngram_lm_scale_2.2_attention_scale_2.2	2.64
+ngram_lm_scale_2.2_attention_scale_2.5	2.64
+ngram_lm_scale_2.3_attention_scale_2.3	2.64
+ngram_lm_scale_2.5_attention_scale_2.5	2.64
+ngram_lm_scale_2.5_attention_scale_3.0	2.64
+ngram_lm_scale_2.5_attention_scale_4.0	2.64
+ngram_lm_scale_0.3_attention_scale_0.3	2.65
+ngram_lm_scale_0.3_attention_scale_0.7	2.65
+ngram_lm_scale_0.3_attention_scale_1.0	2.65
+ngram_lm_scale_0.3_attention_scale_1.1	2.65
+ngram_lm_scale_0.3_attention_scale_2.0	2.65
+ngram_lm_scale_0.3_attention_scale_2.1	2.65
+ngram_lm_scale_0.3_attention_scale_2.2	2.65
+ngram_lm_scale_0.3_attention_scale_2.3	2.65
+ngram_lm_scale_0.3_attention_scale_2.5	2.65
+ngram_lm_scale_0.3_attention_scale_3.0	2.65
+ngram_lm_scale_0.5_attention_scale_0.05	2.65
+ngram_lm_scale_0.6_attention_scale_0.05	2.65
+ngram_lm_scale_0.6_attention_scale_0.08	2.65
+ngram_lm_scale_0.6_attention_scale_0.1	2.65
+ngram_lm_scale_0.6_attention_scale_5.0	2.65
+ngram_lm_scale_1.0_attention_scale_0.3	2.65
+ngram_lm_scale_1.5_attention_scale_1.0	2.65
+ngram_lm_scale_1.9_attention_scale_1.5	2.65
+ngram_lm_scale_2.0_attention_scale_1.7	2.65
+ngram_lm_scale_2.1_attention_scale_2.0	2.65
+ngram_lm_scale_2.2_attention_scale_2.0	2.65
+ngram_lm_scale_2.2_attention_scale_2.1	2.65
+ngram_lm_scale_2.3_attention_scale_2.2	2.65
+ngram_lm_scale_4.0_attention_scale_5.0	2.65
+ngram_lm_scale_0.08_attention_scale_1.0	2.66
+ngram_lm_scale_0.1_attention_scale_1.0	2.66
+ngram_lm_scale_0.1_attention_scale_1.1	2.66
+ngram_lm_scale_0.3_attention_scale_4.0	2.66
+ngram_lm_scale_0.5_attention_scale_0.01	2.66
+ngram_lm_scale_0.5_attention_scale_5.0	2.66
+ngram_lm_scale_0.7_attention_scale_0.05	2.66
+ngram_lm_scale_0.7_attention_scale_0.08	2.66
+ngram_lm_scale_0.7_attention_scale_0.1	2.66
+ngram_lm_scale_1.3_attention_scale_0.6	2.66
+ngram_lm_scale_1.5_attention_scale_0.9	2.66
+ngram_lm_scale_1.7_attention_scale_1.1	2.66
+ngram_lm_scale_1.7_attention_scale_1.2	2.66
+ngram_lm_scale_1.7_attention_scale_1.3	2.66
+ngram_lm_scale_2.1_attention_scale_1.7	2.66
+ngram_lm_scale_2.1_attention_scale_1.9	2.66
+ngram_lm_scale_2.3_attention_scale_2.1	2.66
+ngram_lm_scale_0.05_attention_scale_1.0	2.67
+ngram_lm_scale_0.05_attention_scale_1.1	2.67
+ngram_lm_scale_0.05_attention_scale_1.2	2.67
+ngram_lm_scale_0.08_attention_scale_1.1	2.67
+ngram_lm_scale_0.08_attention_scale_1.2	2.67
+ngram_lm_scale_0.08_attention_scale_1.3	2.67
+ngram_lm_scale_0.08_attention_scale_1.5	2.67
+ngram_lm_scale_0.08_attention_scale_2.5	2.67
+ngram_lm_scale_0.1_attention_scale_1.2	2.67
+ngram_lm_scale_0.1_attention_scale_1.3	2.67
+ngram_lm_scale_0.1_attention_scale_1.5	2.67
+ngram_lm_scale_0.1_attention_scale_2.5	2.67
+ngram_lm_scale_0.3_attention_scale_0.05	2.67
+ngram_lm_scale_0.3_attention_scale_0.1	2.67
+ngram_lm_scale_0.6_attention_scale_0.01	2.67
+ngram_lm_scale_2.0_attention_scale_1.5	2.67
+ngram_lm_scale_2.2_attention_scale_1.9	2.67
+ngram_lm_scale_2.3_attention_scale_1.9	2.67
+ngram_lm_scale_2.3_attention_scale_2.0	2.67
+ngram_lm_scale_2.5_attention_scale_2.3	2.67
+ngram_lm_scale_3.0_attention_scale_3.0	2.67
+ngram_lm_scale_0.01_attention_scale_1.3	2.68
+ngram_lm_scale_0.05_attention_scale_1.3	2.68
+ngram_lm_scale_0.05_attention_scale_1.5	2.68
+ngram_lm_scale_0.05_attention_scale_4.0	2.68
+ngram_lm_scale_0.08_attention_scale_1.7	2.68
+ngram_lm_scale_0.08_attention_scale_2.3	2.68
+ngram_lm_scale_0.08_attention_scale_3.0	2.68
+ngram_lm_scale_0.08_attention_scale_4.0	2.68
+ngram_lm_scale_0.1_attention_scale_0.9	2.68
+ngram_lm_scale_0.1_attention_scale_1.7	2.68
+ngram_lm_scale_0.1_attention_scale_1.9	2.68
+ngram_lm_scale_0.1_attention_scale_2.0	2.68
+ngram_lm_scale_0.1_attention_scale_2.1	2.68
+ngram_lm_scale_0.1_attention_scale_2.2	2.68
+ngram_lm_scale_0.1_attention_scale_2.3	2.68
+ngram_lm_scale_0.1_attention_scale_3.0	2.68
+ngram_lm_scale_0.1_attention_scale_4.0	2.68
+ngram_lm_scale_0.3_attention_scale_0.01	2.68
+ngram_lm_scale_0.3_attention_scale_0.08	2.68
+ngram_lm_scale_0.3_attention_scale_5.0	2.68
+ngram_lm_scale_0.7_attention_scale_0.01	2.68
+ngram_lm_scale_1.1_attention_scale_0.3	2.68
+ngram_lm_scale_1.9_attention_scale_1.3	2.68
+ngram_lm_scale_2.5_attention_scale_2.2	2.68
+ngram_lm_scale_0.01_attention_scale_0.9	2.69
+ngram_lm_scale_0.01_attention_scale_1.0	2.69
+ngram_lm_scale_0.01_attention_scale_1.1	2.69
+ngram_lm_scale_0.01_attention_scale_1.2	2.69
+ngram_lm_scale_0.01_attention_scale_1.5	2.69
+ngram_lm_scale_0.01_attention_scale_2.0	2.69
+ngram_lm_scale_0.01_attention_scale_4.0	2.69
+ngram_lm_scale_0.01_attention_scale_5.0	2.69
+ngram_lm_scale_0.05_attention_scale_0.9	2.69
+ngram_lm_scale_0.05_attention_scale_1.7	2.69
+ngram_lm_scale_0.05_attention_scale_1.9	2.69
+ngram_lm_scale_0.05_attention_scale_2.0	2.69
+ngram_lm_scale_0.05_attention_scale_2.1	2.69
+ngram_lm_scale_0.05_attention_scale_2.2	2.69
+ngram_lm_scale_0.05_attention_scale_2.3	2.69
+ngram_lm_scale_0.05_attention_scale_2.5	2.69
+ngram_lm_scale_0.05_attention_scale_3.0	2.69
+ngram_lm_scale_0.05_attention_scale_5.0	2.69
+ngram_lm_scale_0.08_attention_scale_0.7	2.69
+ngram_lm_scale_0.08_attention_scale_0.9	2.69
+ngram_lm_scale_0.08_attention_scale_1.9	2.69
+ngram_lm_scale_0.08_attention_scale_2.0	2.69
+ngram_lm_scale_0.08_attention_scale_2.1	2.69
+ngram_lm_scale_0.08_attention_scale_2.2	2.69
+ngram_lm_scale_0.08_attention_scale_5.0	2.69
+ngram_lm_scale_0.1_attention_scale_0.3	2.69
+ngram_lm_scale_0.1_attention_scale_0.7	2.69
+ngram_lm_scale_0.1_attention_scale_5.0	2.69
+ngram_lm_scale_0.9_attention_scale_0.1	2.69
+ngram_lm_scale_1.3_attention_scale_0.5	2.69
+ngram_lm_scale_2.2_attention_scale_1.7	2.69
+ngram_lm_scale_0.01_attention_scale_1.7	2.7
+ngram_lm_scale_0.01_attention_scale_1.9	2.7
+ngram_lm_scale_0.01_attention_scale_2.1	2.7
+ngram_lm_scale_0.01_attention_scale_2.2	2.7
+ngram_lm_scale_0.01_attention_scale_2.3	2.7
+ngram_lm_scale_0.01_attention_scale_2.5	2.7
+ngram_lm_scale_0.01_attention_scale_3.0	2.7
+ngram_lm_scale_0.05_attention_scale_0.6	2.7
+ngram_lm_scale_0.05_attention_scale_0.7	2.7
+ngram_lm_scale_0.08_attention_scale_0.1	2.7
+ngram_lm_scale_0.08_attention_scale_0.3	2.7
+ngram_lm_scale_0.08_attention_scale_0.5	2.7
+ngram_lm_scale_0.08_attention_scale_0.6	2.7
+ngram_lm_scale_0.1_attention_scale_0.1	2.7
+ngram_lm_scale_0.1_attention_scale_0.5	2.7
+ngram_lm_scale_0.1_attention_scale_0.6	2.7
+ngram_lm_scale_0.9_attention_scale_0.08	2.7
+ngram_lm_scale_1.7_attention_scale_1.0	2.7
+ngram_lm_scale_1.9_attention_scale_1.2	2.7
+ngram_lm_scale_2.1_attention_scale_1.5	2.7
+ngram_lm_scale_2.5_attention_scale_2.0	2.7
+ngram_lm_scale_2.5_attention_scale_2.1	2.7
+ngram_lm_scale_4.0_attention_scale_4.0	2.7
+ngram_lm_scale_0.01_attention_scale_0.3	2.71
+ngram_lm_scale_0.01_attention_scale_0.7	2.71
+ngram_lm_scale_0.05_attention_scale_0.3	2.71
+ngram_lm_scale_0.05_attention_scale_0.5	2.71
+ngram_lm_scale_0.08_attention_scale_0.08	2.71
+ngram_lm_scale_0.1_attention_scale_0.05	2.71
+ngram_lm_scale_0.1_attention_scale_0.08	2.71
+ngram_lm_scale_1.5_attention_scale_0.7	2.71
+ngram_lm_scale_2.0_attention_scale_1.3	2.71
+ngram_lm_scale_2.3_attention_scale_1.7	2.71
+ngram_lm_scale_0.01_attention_scale_0.5	2.72
+ngram_lm_scale_0.01_attention_scale_0.6	2.72
+ngram_lm_scale_0.05_attention_scale_0.08	2.72
+ngram_lm_scale_0.05_attention_scale_0.1	2.72
+ngram_lm_scale_0.08_attention_scale_0.05	2.72
+ngram_lm_scale_0.1_attention_scale_0.01	2.72
+ngram_lm_scale_0.9_attention_scale_0.05	2.72
+ngram_lm_scale_3.0_attention_scale_2.5	2.72
+ngram_lm_scale_0.05_attention_scale_0.05	2.73
+ngram_lm_scale_0.08_attention_scale_0.01	2.73
+ngram_lm_scale_0.9_attention_scale_0.01	2.73
+ngram_lm_scale_1.7_attention_scale_0.9	2.73
+ngram_lm_scale_2.5_attention_scale_1.9	2.73
+ngram_lm_scale_0.01_attention_scale_0.08	2.74
+ngram_lm_scale_0.01_attention_scale_0.1	2.74
+ngram_lm_scale_0.05_attention_scale_0.01	2.74
+ngram_lm_scale_1.9_attention_scale_1.1	2.74
+ngram_lm_scale_2.2_attention_scale_1.5	2.74
+ngram_lm_scale_0.01_attention_scale_0.01	2.75
+ngram_lm_scale_0.01_attention_scale_0.05	2.75
+ngram_lm_scale_1.0_attention_scale_0.1	2.75
+ngram_lm_scale_1.2_attention_scale_0.3	2.75
+ngram_lm_scale_5.0_attention_scale_5.0	2.75
+ngram_lm_scale_1.5_attention_scale_0.6	2.76
+ngram_lm_scale_2.0_attention_scale_1.2	2.76
+ngram_lm_scale_1.0_attention_scale_0.08	2.77
+ngram_lm_scale_2.1_attention_scale_1.3	2.77
+ngram_lm_scale_2.3_attention_scale_1.5	2.78
+ngram_lm_scale_1.9_attention_scale_1.0	2.79
+ngram_lm_scale_2.5_attention_scale_1.7	2.79
+ngram_lm_scale_1.0_attention_scale_0.05	2.8
+ngram_lm_scale_2.0_attention_scale_1.1	2.8
+ngram_lm_scale_2.1_attention_scale_1.2	2.81
+ngram_lm_scale_3.0_attention_scale_2.3	2.81
+ngram_lm_scale_1.0_attention_scale_0.01	2.82
+ngram_lm_scale_1.1_attention_scale_0.1	2.82
+ngram_lm_scale_1.3_attention_scale_0.3	2.82
+ngram_lm_scale_2.2_attention_scale_1.3	2.82
+ngram_lm_scale_1.7_attention_scale_0.7	2.83
+ngram_lm_scale_3.0_attention_scale_2.2	2.83
+ngram_lm_scale_1.1_attention_scale_0.08	2.84
+ngram_lm_scale_1.5_attention_scale_0.5	2.84
+ngram_lm_scale_3.0_attention_scale_2.1	2.85
+ngram_lm_scale_1.1_attention_scale_0.05	2.86
+ngram_lm_scale_1.9_attention_scale_0.9	2.86
+ngram_lm_scale_2.0_attention_scale_1.0	2.86
+ngram_lm_scale_2.1_attention_scale_1.1	2.86
+ngram_lm_scale_2.2_attention_scale_1.2	2.87
+ngram_lm_scale_2.3_attention_scale_1.3	2.87
+ngram_lm_scale_1.1_attention_scale_0.01	2.89
+ngram_lm_scale_2.5_attention_scale_1.5	2.89
+ngram_lm_scale_3.0_attention_scale_2.0	2.9
+ngram_lm_scale_1.2_attention_scale_0.1	2.92
+ngram_lm_scale_2.0_attention_scale_0.9	2.92
+ngram_lm_scale_1.2_attention_scale_0.08	2.93
+ngram_lm_scale_1.7_attention_scale_0.6	2.93
+ngram_lm_scale_2.1_attention_scale_1.0	2.93
+ngram_lm_scale_2.2_attention_scale_1.1	2.95
+ngram_lm_scale_2.3_attention_scale_1.2	2.95
+ngram_lm_scale_1.2_attention_scale_0.05	2.97
+ngram_lm_scale_4.0_attention_scale_3.0	2.97
+ngram_lm_scale_3.0_attention_scale_1.9	2.99
+ngram_lm_scale_5.0_attention_scale_4.0	2.99
+ngram_lm_scale_1.3_attention_scale_0.1	3.0
+ngram_lm_scale_1.5_attention_scale_0.3	3.0
+ngram_lm_scale_1.7_attention_scale_0.5	3.0
+ngram_lm_scale_1.2_attention_scale_0.01	3.01
+ngram_lm_scale_1.9_attention_scale_0.7	3.01
+ngram_lm_scale_2.1_attention_scale_0.9	3.01
+ngram_lm_scale_2.2_attention_scale_1.0	3.01
+ngram_lm_scale_1.3_attention_scale_0.08	3.02
+ngram_lm_scale_2.3_attention_scale_1.1	3.02
+ngram_lm_scale_2.5_attention_scale_1.3	3.02
+ngram_lm_scale_1.3_attention_scale_0.05	3.09
+ngram_lm_scale_1.9_attention_scale_0.6	3.1
+ngram_lm_scale_2.0_attention_scale_0.7	3.1
+ngram_lm_scale_2.2_attention_scale_0.9	3.1
+ngram_lm_scale_2.3_attention_scale_1.0	3.1
+ngram_lm_scale_2.5_attention_scale_1.2	3.1
+ngram_lm_scale_3.0_attention_scale_1.7	3.1
+ngram_lm_scale_1.3_attention_scale_0.01	3.15
+ngram_lm_scale_2.5_attention_scale_1.1	3.18
+ngram_lm_scale_2.3_attention_scale_0.9	3.19
+ngram_lm_scale_4.0_attention_scale_2.5	3.19
+ngram_lm_scale_1.9_attention_scale_0.5	3.21
+ngram_lm_scale_2.1_attention_scale_0.7	3.21
+ngram_lm_scale_2.0_attention_scale_0.6	3.22
+ngram_lm_scale_1.7_attention_scale_0.3	3.23
+ngram_lm_scale_3.0_attention_scale_1.5	3.24
+ngram_lm_scale_2.5_attention_scale_1.0	3.27
+ngram_lm_scale_4.0_attention_scale_2.3	3.27
+ngram_lm_scale_1.5_attention_scale_0.1	3.29
+ngram_lm_scale_2.2_attention_scale_0.7	3.29
+ngram_lm_scale_2.1_attention_scale_0.6	3.3
+ngram_lm_scale_2.0_attention_scale_0.5	3.31
+ngram_lm_scale_1.5_attention_scale_0.08	3.32
+ngram_lm_scale_4.0_attention_scale_2.2	3.32
+ngram_lm_scale_2.5_attention_scale_0.9	3.34
+ngram_lm_scale_3.0_attention_scale_1.3	3.34
+ngram_lm_scale_2.3_attention_scale_0.7	3.35
+ngram_lm_scale_2.2_attention_scale_0.6	3.36
+ngram_lm_scale_5.0_attention_scale_3.0	3.36
+ngram_lm_scale_2.1_attention_scale_0.5	3.37
+ngram_lm_scale_1.5_attention_scale_0.05	3.38
+ngram_lm_scale_4.0_attention_scale_2.1	3.38
+ngram_lm_scale_3.0_attention_scale_1.2	3.42
+ngram_lm_scale_4.0_attention_scale_2.0	3.42
+ngram_lm_scale_1.5_attention_scale_0.01	3.45
+ngram_lm_scale_1.9_attention_scale_0.3	3.45
+ngram_lm_scale_2.3_attention_scale_0.6	3.46
+ngram_lm_scale_4.0_attention_scale_1.9	3.48
+ngram_lm_scale_2.2_attention_scale_0.5	3.51
+ngram_lm_scale_3.0_attention_scale_1.1	3.53
+ngram_lm_scale_1.7_attention_scale_0.1	3.54
+ngram_lm_scale_2.5_attention_scale_0.7	3.54
+ngram_lm_scale_1.7_attention_scale_0.08	3.58
+ngram_lm_scale_2.0_attention_scale_0.3	3.58
+ngram_lm_scale_5.0_attention_scale_2.5	3.59
+ngram_lm_scale_3.0_attention_scale_1.0	3.61
+ngram_lm_scale_2.3_attention_scale_0.5	3.62
+ngram_lm_scale_4.0_attention_scale_1.7	3.64
+ngram_lm_scale_1.7_attention_scale_0.05	3.65
+ngram_lm_scale_2.1_attention_scale_0.3	3.67
+ngram_lm_scale_2.5_attention_scale_0.6	3.67
+ngram_lm_scale_5.0_attention_scale_2.3	3.71
+ngram_lm_scale_1.7_attention_scale_0.01	3.72
+ngram_lm_scale_3.0_attention_scale_0.9	3.72
+ngram_lm_scale_4.0_attention_scale_1.5	3.76
+ngram_lm_scale_5.0_attention_scale_2.2	3.76
+ngram_lm_scale_2.2_attention_scale_0.3	3.78
+ngram_lm_scale_1.9_attention_scale_0.1	3.79
+ngram_lm_scale_2.5_attention_scale_0.5	3.81
+ngram_lm_scale_5.0_attention_scale_2.1	3.81
+ngram_lm_scale_1.9_attention_scale_0.08	3.83
+ngram_lm_scale_5.0_attention_scale_2.0	3.87
+ngram_lm_scale_1.9_attention_scale_0.05	3.88
+ngram_lm_scale_2.3_attention_scale_0.3	3.88
+ngram_lm_scale_2.0_attention_scale_0.1	3.9
+ngram_lm_scale_3.0_attention_scale_0.7	3.92
+ngram_lm_scale_5.0_attention_scale_1.9	3.93
+ngram_lm_scale_4.0_attention_scale_1.3	3.94
+ngram_lm_scale_2.0_attention_scale_0.08	3.95
+ngram_lm_scale_1.9_attention_scale_0.01	3.96
+ngram_lm_scale_4.0_attention_scale_1.2	3.99
+ngram_lm_scale_2.0_attention_scale_0.05	4.0
+ngram_lm_scale_2.1_attention_scale_0.1	4.01
+ngram_lm_scale_3.0_attention_scale_0.6	4.02
+ngram_lm_scale_2.1_attention_scale_0.08	4.05
+ngram_lm_scale_2.5_attention_scale_0.3	4.05
+ngram_lm_scale_5.0_attention_scale_1.7	4.06
+ngram_lm_scale_2.0_attention_scale_0.01	4.1
+ngram_lm_scale_4.0_attention_scale_1.1	4.1
+ngram_lm_scale_2.1_attention_scale_0.05	4.11
+ngram_lm_scale_2.2_attention_scale_0.1	4.12
+ngram_lm_scale_2.2_attention_scale_0.08	4.15
+ngram_lm_scale_3.0_attention_scale_0.5	4.16
+ngram_lm_scale_2.1_attention_scale_0.01	4.18
+ngram_lm_scale_2.2_attention_scale_0.05	4.19
+ngram_lm_scale_4.0_attention_scale_1.0	4.19
+ngram_lm_scale_5.0_attention_scale_1.5	4.19
+ngram_lm_scale_2.3_attention_scale_0.1	4.2
+ngram_lm_scale_2.3_attention_scale_0.08	4.23
+ngram_lm_scale_2.2_attention_scale_0.01	4.27
+ngram_lm_scale_2.3_attention_scale_0.05	4.28
+ngram_lm_scale_4.0_attention_scale_0.9	4.28
+ngram_lm_scale_2.3_attention_scale_0.01	4.34
+ngram_lm_scale_2.5_attention_scale_0.1	4.35
+ngram_lm_scale_5.0_attention_scale_1.3	4.35
+ngram_lm_scale_2.5_attention_scale_0.08	4.39
+ngram_lm_scale_3.0_attention_scale_0.3	4.4
+ngram_lm_scale_5.0_attention_scale_1.2	4.41
+ngram_lm_scale_2.5_attention_scale_0.05	4.43
+ngram_lm_scale_4.0_attention_scale_0.7	4.45
+ngram_lm_scale_5.0_attention_scale_1.1	4.46
+ngram_lm_scale_2.5_attention_scale_0.01	4.48
+ngram_lm_scale_4.0_attention_scale_0.6	4.54
+ngram_lm_scale_5.0_attention_scale_1.0	4.54
+ngram_lm_scale_5.0_attention_scale_0.9	4.62
+ngram_lm_scale_3.0_attention_scale_0.1	4.64
+ngram_lm_scale_4.0_attention_scale_0.5	4.65
+ngram_lm_scale_3.0_attention_scale_0.08	4.67
+ngram_lm_scale_3.0_attention_scale_0.05	4.7
+ngram_lm_scale_3.0_attention_scale_0.01	4.73
+ngram_lm_scale_5.0_attention_scale_0.7	4.76
+ngram_lm_scale_4.0_attention_scale_0.3	4.8
+ngram_lm_scale_5.0_attention_scale_0.6	4.85
+ngram_lm_scale_5.0_attention_scale_0.5	4.91
+ngram_lm_scale_4.0_attention_scale_0.1	4.99
+ngram_lm_scale_4.0_attention_scale_0.08	5.01
+ngram_lm_scale_4.0_attention_scale_0.05	5.02
+ngram_lm_scale_5.0_attention_scale_0.3	5.06
+ngram_lm_scale_4.0_attention_scale_0.01	5.07
+ngram_lm_scale_5.0_attention_scale_0.1	5.22
+ngram_lm_scale_5.0_attention_scale_0.08	5.24
+ngram_lm_scale_5.0_attention_scale_0.05	5.25
+ngram_lm_scale_5.0_attention_scale_0.01	5.27
+2022-07-20 21:24:54,113 INFO [decode.py:588] batch 0/?, cuts processed until now is 2
+2022-07-20 21:26:25,190 INFO [decode.py:588] batch 100/?, cuts processed until now is 249
+2022-07-20 21:27:58,029 INFO [decode.py:588] batch 200/?, cuts processed until now is 540
+2022-07-20 21:29:23,092 INFO [decode.py:588] batch 300/?, cuts processed until now is 834
+2022-07-20 21:30:51,288 INFO [decode.py:588] batch 400/?, cuts processed until now is 1104
+2022-07-20 21:32:21,736 INFO [decode.py:588] batch 500/?, cuts processed until now is 1350
+2022-07-20 21:33:54,991 INFO [decode.py:588] batch 600/?, cuts processed until now is 1640
+2022-07-20 21:35:18,198 INFO [decode.py:588] batch 700/?, cuts processed until now is 1933
+2022-07-20 21:36:49,206 INFO [decode.py:588] batch 800/?, cuts processed until now is 2175
+2022-07-20 21:38:16,845 INFO [decode.py:588] batch 900/?, cuts processed until now is 2440
+2022-07-20 21:38:31,538 INFO [decode.py:783] Caught exception:
+CUDA out of memory. Tried to allocate 740.00 MiB (GPU 0; 31.75 GiB total capacity; 28.28 GiB already allocated; 447.50 MiB free; 29.92 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
+2022-07-20 21:38:31,538 INFO [decode.py:789] num_arcs before pruning: 349313
+2022-07-20 21:38:31,538 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
+2022-07-20 21:38:31,557 INFO [decode.py:803] num_arcs after pruning: 8796
+2022-07-20 21:39:44,765 INFO [decode.py:588] batch 1000/?, cuts processed until now is 2656
+2022-07-20 21:41:20,435 INFO [decode.py:588] batch 1100/?, cuts processed until now is 2839
+2022-07-20 21:46:22,023 INFO [decode.py:637]
+For test-other, WER of different settings are:
+ngram_lm_scale_1.1_attention_scale_0.9	5.54	best for test-other
+ngram_lm_scale_1.5_attention_scale_1.5	5.54
+ngram_lm_scale_1.5_attention_scale_1.7	5.54
+ngram_lm_scale_1.7_attention_scale_2.0	5.54
+ngram_lm_scale_1.7_attention_scale_2.1	5.54
+ngram_lm_scale_1.2_attention_scale_1.1	5.55
+ngram_lm_scale_1.3_attention_scale_0.9	5.55
+ngram_lm_scale_1.3_attention_scale_1.1	5.55
+ngram_lm_scale_1.3_attention_scale_1.2	5.55
+ngram_lm_scale_1.7_attention_scale_1.9	5.55
+ngram_lm_scale_1.0_attention_scale_0.7	5.56
+ngram_lm_scale_1.2_attention_scale_0.6	5.56
+ngram_lm_scale_1.2_attention_scale_1.0	5.56
+ngram_lm_scale_1.2_attention_scale_1.2	5.56
+ngram_lm_scale_1.3_attention_scale_1.0	5.56
+ngram_lm_scale_1.5_attention_scale_1.3	5.56
+ngram_lm_scale_1.5_attention_scale_1.9	5.56
+ngram_lm_scale_1.5_attention_scale_2.0	5.56
+ngram_lm_scale_1.7_attention_scale_2.2	5.56
+ngram_lm_scale_1.9_attention_scale_2.2	5.56
+ngram_lm_scale_1.9_attention_scale_2.3	5.56
+ngram_lm_scale_1.9_attention_scale_2.5	5.56
+ngram_lm_scale_2.0_attention_scale_2.5	5.56
+ngram_lm_scale_2.0_attention_scale_3.0	5.56
+ngram_lm_scale_2.2_attention_scale_3.0	5.56
+ngram_lm_scale_2.5_attention_scale_4.0	5.56
+ngram_lm_scale_3.0_attention_scale_5.0	5.56
+ngram_lm_scale_0.9_attention_scale_0.6	5.57
+ngram_lm_scale_1.0_attention_scale_0.9	5.57
+ngram_lm_scale_1.1_attention_scale_1.0	5.57
+ngram_lm_scale_1.1_attention_scale_1.1	5.57
+ngram_lm_scale_1.2_attention_scale_0.7	5.57
+ngram_lm_scale_1.2_attention_scale_0.9	5.57
+ngram_lm_scale_1.2_attention_scale_1.3	5.57
+ngram_lm_scale_1.2_attention_scale_1.5	5.57
+ngram_lm_scale_1.3_attention_scale_0.7	5.57
+ngram_lm_scale_1.3_attention_scale_1.3	5.57
+ngram_lm_scale_1.5_attention_scale_2.1	5.57
+ngram_lm_scale_1.5_attention_scale_2.2	5.57
+ngram_lm_scale_1.5_attention_scale_2.3	5.57
+ngram_lm_scale_1.5_attention_scale_2.5	5.57
+ngram_lm_scale_1.7_attention_scale_1.7	5.57
+ngram_lm_scale_1.7_attention_scale_2.3	5.57
+ngram_lm_scale_1.7_attention_scale_2.5	5.57
+ngram_lm_scale_1.7_attention_scale_3.0	5.57
+ngram_lm_scale_1.9_attention_scale_2.0	5.57
+ngram_lm_scale_1.9_attention_scale_2.1	5.57
+ngram_lm_scale_1.9_attention_scale_3.0	5.57
+ngram_lm_scale_2.0_attention_scale_2.3	5.57
+ngram_lm_scale_2.0_attention_scale_4.0	5.57
+ngram_lm_scale_2.1_attention_scale_3.0	5.57
+ngram_lm_scale_2.1_attention_scale_4.0	5.57
+ngram_lm_scale_2.3_attention_scale_3.0	5.57
+ngram_lm_scale_0.9_attention_scale_0.7	5.58
+ngram_lm_scale_1.0_attention_scale_1.0	5.58
+ngram_lm_scale_1.0_attention_scale_1.1	5.58
+ngram_lm_scale_1.1_attention_scale_0.5	5.58
+ngram_lm_scale_1.1_attention_scale_0.6	5.58
+ngram_lm_scale_1.1_attention_scale_1.2	5.58
+ngram_lm_scale_1.1_attention_scale_1.5	5.58
+ngram_lm_scale_1.3_attention_scale_1.5	5.58
+ngram_lm_scale_1.3_attention_scale_1.7	5.58
+ngram_lm_scale_1.3_attention_scale_1.9	5.58
+ngram_lm_scale_1.3_attention_scale_2.0	5.58
+ngram_lm_scale_1.3_attention_scale_2.1	5.58
+ngram_lm_scale_1.3_attention_scale_2.2	5.58
+ngram_lm_scale_2.0_attention_scale_2.2	5.58
+ngram_lm_scale_2.2_attention_scale_4.0	5.58
+ngram_lm_scale_2.3_attention_scale_4.0	5.58
+ngram_lm_scale_0.9_attention_scale_0.9	5.59
+ngram_lm_scale_1.0_attention_scale_1.2	5.59
+ngram_lm_scale_1.1_attention_scale_0.7	5.59
+ngram_lm_scale_1.1_attention_scale_1.3	5.59
+ngram_lm_scale_1.2_attention_scale_1.7	5.59
+ngram_lm_scale_1.2_attention_scale_1.9	5.59
+ngram_lm_scale_1.2_attention_scale_2.1	5.59
+ngram_lm_scale_1.3_attention_scale_2.3	5.59
+ngram_lm_scale_1.5_attention_scale_1.0	5.59
+ngram_lm_scale_1.5_attention_scale_1.1	5.59
+ngram_lm_scale_1.5_attention_scale_1.2	5.59
+ngram_lm_scale_1.7_attention_scale_1.5	5.59
+ngram_lm_scale_1.9_attention_scale_1.9	5.59
+ngram_lm_scale_1.9_attention_scale_4.0	5.59
+ngram_lm_scale_2.0_attention_scale_2.1	5.59
+ngram_lm_scale_2.1_attention_scale_2.3	5.59
+ngram_lm_scale_2.1_attention_scale_2.5	5.59
+ngram_lm_scale_2.5_attention_scale_5.0	5.59
+ngram_lm_scale_0.9_attention_scale_1.0	5.6
+ngram_lm_scale_1.0_attention_scale_0.6	5.6
+ngram_lm_scale_1.0_attention_scale_1.3	5.6
+ngram_lm_scale_1.1_attention_scale_1.7	5.6
+ngram_lm_scale_1.1_attention_scale_1.9	5.6
+ngram_lm_scale_1.2_attention_scale_2.0	5.6
+ngram_lm_scale_1.7_attention_scale_1.3	5.6
+ngram_lm_scale_2.0_attention_scale_2.0	5.6
+ngram_lm_scale_2.1_attention_scale_2.2	5.6
+ngram_lm_scale_2.2_attention_scale_5.0	5.6
+ngram_lm_scale_2.3_attention_scale_2.5	5.6
+ngram_lm_scale_2.3_attention_scale_5.0	5.6
+ngram_lm_scale_3.0_attention_scale_4.0	5.6
+ngram_lm_scale_0.9_attention_scale_0.5	5.61
+ngram_lm_scale_1.0_attention_scale_0.5	5.61
+ngram_lm_scale_1.0_attention_scale_1.5	5.61
+ngram_lm_scale_1.1_attention_scale_2.0	5.61
+ngram_lm_scale_1.2_attention_scale_0.5	5.61
+ngram_lm_scale_1.2_attention_scale_2.2	5.61
+ngram_lm_scale_1.3_attention_scale_2.5	5.61
+ngram_lm_scale_1.5_attention_scale_0.9	5.61
+ngram_lm_scale_1.5_attention_scale_3.0	5.61
+ngram_lm_scale_1.7_attention_scale_4.0	5.61
+ngram_lm_scale_1.9_attention_scale_1.7	5.61
+ngram_lm_scale_2.0_attention_scale_1.7	5.61
+ngram_lm_scale_2.0_attention_scale_1.9	5.61
+ngram_lm_scale_2.0_attention_scale_5.0	5.61
+ngram_lm_scale_2.1_attention_scale_5.0	5.61
+ngram_lm_scale_2.2_attention_scale_2.5	5.61
+ngram_lm_scale_2.5_attention_scale_3.0	5.61
+ngram_lm_scale_0.9_attention_scale_1.1	5.62
+ngram_lm_scale_1.0_attention_scale_1.7	5.62
+ngram_lm_scale_1.1_attention_scale_2.1	5.62
+ngram_lm_scale_1.2_attention_scale_2.3	5.62
+ngram_lm_scale_1.7_attention_scale_1.2	5.62
+ngram_lm_scale_1.9_attention_scale_1.5	5.62
+ngram_lm_scale_2.1_attention_scale_1.9	5.62
+ngram_lm_scale_2.1_attention_scale_2.0	5.62
+ngram_lm_scale_2.1_attention_scale_2.1	5.62
+ngram_lm_scale_2.2_attention_scale_2.2	5.62
+ngram_lm_scale_2.2_attention_scale_2.3	5.62
+ngram_lm_scale_0.7_attention_scale_0.7	5.63
+ngram_lm_scale_0.9_attention_scale_0.3	5.63
+ngram_lm_scale_0.9_attention_scale_1.2	5.63
+ngram_lm_scale_0.9_attention_scale_1.3	5.63
+ngram_lm_scale_1.0_attention_scale_0.3	5.63
+ngram_lm_scale_1.0_attention_scale_1.9	5.63
+ngram_lm_scale_1.1_attention_scale_2.2	5.63
+ngram_lm_scale_1.3_attention_scale_0.6	5.63
+ngram_lm_scale_1.9_attention_scale_5.0	5.63
+ngram_lm_scale_2.2_attention_scale_2.0	5.63
+ngram_lm_scale_2.2_attention_scale_2.1	5.63
+ngram_lm_scale_2.3_attention_scale_2.1	5.63
+ngram_lm_scale_2.3_attention_scale_2.3	5.63
+ngram_lm_scale_0.7_attention_scale_0.3	5.64
+ngram_lm_scale_0.7_attention_scale_0.5	5.64
+ngram_lm_scale_0.7_attention_scale_0.6	5.64
+ngram_lm_scale_0.7_attention_scale_0.9	5.64
+ngram_lm_scale_0.7_attention_scale_1.0	5.64
+ngram_lm_scale_0.9_attention_scale_1.5	5.64
+ngram_lm_scale_1.0_attention_scale_2.0	5.64
+ngram_lm_scale_1.1_attention_scale_2.3	5.64
+ngram_lm_scale_1.2_attention_scale_2.5	5.64
+ngram_lm_scale_1.5_attention_scale_5.0	5.64
+ngram_lm_scale_2.2_attention_scale_1.9	5.64
+ngram_lm_scale_2.3_attention_scale_2.2	5.64
+ngram_lm_scale_0.7_attention_scale_1.1	5.65
+ngram_lm_scale_0.9_attention_scale_1.7	5.65
+ngram_lm_scale_0.9_attention_scale_1.9	5.65
+ngram_lm_scale_1.0_attention_scale_2.1	5.65
+ngram_lm_scale_1.0_attention_scale_2.2	5.65
+ngram_lm_scale_1.0_attention_scale_2.3	5.65
+ngram_lm_scale_1.1_attention_scale_2.5	5.65
+ngram_lm_scale_1.3_attention_scale_3.0	5.65
+ngram_lm_scale_1.7_attention_scale_1.1	5.65
+ngram_lm_scale_1.7_attention_scale_5.0	5.65
+ngram_lm_scale_2.5_attention_scale_2.5	5.65
+ngram_lm_scale_0.6_attention_scale_0.5	5.66
+ngram_lm_scale_0.6_attention_scale_0.6	5.66
+ngram_lm_scale_0.6_attention_scale_0.9	5.66
+ngram_lm_scale_0.7_attention_scale_1.2	5.66
+ngram_lm_scale_0.9_attention_scale_2.3	5.66
+ngram_lm_scale_1.0_attention_scale_2.5	5.66
+ngram_lm_scale_1.2_attention_scale_3.0	5.66
+ngram_lm_scale_1.5_attention_scale_4.0	5.66
+ngram_lm_scale_2.3_attention_scale_2.0	5.66
+ngram_lm_scale_0.5_attention_scale_0.6	5.67
+ngram_lm_scale_0.5_attention_scale_0.7	5.67
+ngram_lm_scale_0.5_attention_scale_0.9	5.67
+ngram_lm_scale_0.6_attention_scale_0.7	5.67
+ngram_lm_scale_0.6_attention_scale_1.0	5.67
+ngram_lm_scale_0.6_attention_scale_1.1	5.67
+ngram_lm_scale_0.7_attention_scale_1.3	5.67
+ngram_lm_scale_0.9_attention_scale_2.0	5.67
+ngram_lm_scale_0.9_attention_scale_2.1	5.67
+ngram_lm_scale_0.9_attention_scale_2.2	5.67
+ngram_lm_scale_0.9_attention_scale_2.5	5.67
+ngram_lm_scale_1.1_attention_scale_3.0	5.67
+ngram_lm_scale_1.3_attention_scale_4.0	5.67
+ngram_lm_scale_2.0_attention_scale_1.5	5.67
+ngram_lm_scale_2.1_attention_scale_1.7	5.67
+ngram_lm_scale_2.5_attention_scale_2.3	5.67
+ngram_lm_scale_4.0_attention_scale_5.0	5.67
+ngram_lm_scale_0.5_attention_scale_1.0	5.68
+ngram_lm_scale_0.6_attention_scale_0.3	5.68
+ngram_lm_scale_1.0_attention_scale_3.0	5.68
+ngram_lm_scale_1.2_attention_scale_4.0	5.68
+ngram_lm_scale_2.3_attention_scale_1.9	5.68
+ngram_lm_scale_2.5_attention_scale_2.2	5.68
+ngram_lm_scale_3.0_attention_scale_3.0	5.68
+ngram_lm_scale_0.5_attention_scale_0.5	5.69
+ngram_lm_scale_0.5_attention_scale_1.1	5.69
+ngram_lm_scale_0.6_attention_scale_1.2	5.69
+ngram_lm_scale_1.1_attention_scale_0.3	5.69
+ngram_lm_scale_1.3_attention_scale_0.5	5.69
+ngram_lm_scale_1.7_attention_scale_1.0	5.69
+ngram_lm_scale_1.9_attention_scale_1.3	5.69
+ngram_lm_scale_0.5_attention_scale_1.2	5.7
+ngram_lm_scale_0.6_attention_scale_1.3	5.7
+ngram_lm_scale_0.6_attention_scale_1.5	5.7
+ngram_lm_scale_0.7_attention_scale_1.5	5.7
+ngram_lm_scale_0.7_attention_scale_1.7	5.7
+ngram_lm_scale_0.7_attention_scale_1.9	5.7
+ngram_lm_scale_0.7_attention_scale_2.0	5.7
+ngram_lm_scale_0.9_attention_scale_3.0	5.7
+ngram_lm_scale_2.1_attention_scale_1.5	5.7
+ngram_lm_scale_2.2_attention_scale_1.7	5.7
+ngram_lm_scale_0.5_attention_scale_1.3	5.71
+ngram_lm_scale_0.6_attention_scale_1.7	5.71
+ngram_lm_scale_0.7_attention_scale_2.1	5.71
+ngram_lm_scale_0.7_attention_scale_2.2	5.71
+ngram_lm_scale_1.1_attention_scale_4.0	5.71
+ngram_lm_scale_1.3_attention_scale_5.0	5.71
+ngram_lm_scale_2.5_attention_scale_2.1	5.71
+ngram_lm_scale_0.5_attention_scale_0.3	5.72
+ngram_lm_scale_0.6_attention_scale_1.9	5.72
+ngram_lm_scale_1.0_attention_scale_4.0	5.72
+ngram_lm_scale_1.5_attention_scale_0.7	5.72
+ngram_lm_scale_1.9_attention_scale_1.2	5.72
+ngram_lm_scale_2.3_attention_scale_1.7	5.72
+ngram_lm_scale_2.5_attention_scale_2.0	5.72
+ngram_lm_scale_0.5_attention_scale_1.5	5.73
+ngram_lm_scale_0.6_attention_scale_0.1	5.73
+ngram_lm_scale_0.7_attention_scale_2.3	5.73
+ngram_lm_scale_0.9_attention_scale_4.0	5.73
+ngram_lm_scale_1.2_attention_scale_5.0	5.73
+ngram_lm_scale_2.0_attention_scale_1.3	5.73
+ngram_lm_scale_0.5_attention_scale_1.7	5.74
+ngram_lm_scale_0.5_attention_scale_1.9	5.74
+ngram_lm_scale_0.5_attention_scale_2.0	5.74
+ngram_lm_scale_0.5_attention_scale_2.2	5.74
+ngram_lm_scale_0.5_attention_scale_2.3	5.74
+ngram_lm_scale_0.6_attention_scale_0.08	5.74
+ngram_lm_scale_0.6_attention_scale_2.0	5.74
+ngram_lm_scale_0.6_attention_scale_2.1	5.74
+ngram_lm_scale_0.6_attention_scale_2.2	5.74
+ngram_lm_scale_0.6_attention_scale_2.5	5.74
+ngram_lm_scale_0.7_attention_scale_0.1	5.74
+ngram_lm_scale_0.7_attention_scale_2.5	5.74
+ngram_lm_scale_0.7_attention_scale_3.0	5.74
+ngram_lm_scale_1.1_attention_scale_5.0	5.74
+ngram_lm_scale_1.7_attention_scale_0.9	5.74
+ngram_lm_scale_4.0_attention_scale_4.0	5.74
+ngram_lm_scale_0.3_attention_scale_1.1	5.75
+ngram_lm_scale_0.3_attention_scale_1.2	5.75
+ngram_lm_scale_0.5_attention_scale_2.1	5.75
+ngram_lm_scale_0.5_attention_scale_2.5	5.75
+ngram_lm_scale_0.6_attention_scale_0.05	5.75
+ngram_lm_scale_0.6_attention_scale_2.3	5.75
+ngram_lm_scale_0.6_attention_scale_3.0	5.75
+ngram_lm_scale_0.7_attention_scale_0.05	5.75
+ngram_lm_scale_0.7_attention_scale_0.08	5.75
+ngram_lm_scale_0.9_attention_scale_5.0	5.75
+ngram_lm_scale_1.0_attention_scale_5.0	5.75
+ngram_lm_scale_2.2_attention_scale_1.5	5.75
+ngram_lm_scale_2.5_attention_scale_1.9	5.75
+ngram_lm_scale_0.3_attention_scale_0.9	5.76
+ngram_lm_scale_0.3_attention_scale_1.0	5.76
+ngram_lm_scale_0.3_attention_scale_1.3	5.76
+ngram_lm_scale_0.5_attention_scale_0.1	5.76
+ngram_lm_scale_0.7_attention_scale_4.0	5.76
+ngram_lm_scale_2.0_attention_scale_1.2	5.76
+ngram_lm_scale_0.5_attention_scale_0.08	5.77
+ngram_lm_scale_0.5_attention_scale_3.0	5.77
+ngram_lm_scale_0.6_attention_scale_4.0	5.77
+ngram_lm_scale_0.7_attention_scale_5.0	5.77
+ngram_lm_scale_1.9_attention_scale_1.1	5.77
+ngram_lm_scale_3.0_attention_scale_2.5	5.77
+ngram_lm_scale_0.3_attention_scale_0.6	5.78
+ngram_lm_scale_0.3_attention_scale_0.7	5.78
+ngram_lm_scale_0.3_attention_scale_1.5	5.78
+ngram_lm_scale_0.5_attention_scale_0.01	5.78
+ngram_lm_scale_0.5_attention_scale_0.05	5.78
+ngram_lm_scale_0.6_attention_scale_0.01	5.78
+ngram_lm_scale_0.9_attention_scale_0.1	5.78
+ngram_lm_scale_2.1_attention_scale_1.3	5.78
+ngram_lm_scale_0.3_attention_scale_0.5	5.79
+ngram_lm_scale_0.3_attention_scale_1.7	5.79
+ngram_lm_scale_0.7_attention_scale_0.01	5.79
+ngram_lm_scale_0.3_attention_scale_2.0	5.8
+ngram_lm_scale_0.3_attention_scale_2.1	5.8
+ngram_lm_scale_0.3_attention_scale_2.2	5.8
+ngram_lm_scale_0.5_attention_scale_4.0	5.8
+ngram_lm_scale_0.6_attention_scale_5.0	5.8
+ngram_lm_scale_0.9_attention_scale_0.08	5.8
+ngram_lm_scale_1.2_attention_scale_0.3	5.8
+ngram_lm_scale_0.3_attention_scale_1.9	5.81
+ngram_lm_scale_0.3_attention_scale_2.3	5.81
+ngram_lm_scale_0.5_attention_scale_5.0	5.81
+ngram_lm_scale_1.5_attention_scale_0.6	5.81
+ngram_lm_scale_0.3_attention_scale_2.5	5.82
+ngram_lm_scale_2.3_attention_scale_1.5	5.82
+ngram_lm_scale_0.3_attention_scale_3.0	5.83
+ngram_lm_scale_0.3_attention_scale_4.0	5.83
+ngram_lm_scale_0.3_attention_scale_5.0	5.83
+ngram_lm_scale_1.0_attention_scale_0.1	5.83
+ngram_lm_scale_2.0_attention_scale_1.1	5.83
+ngram_lm_scale_2.1_attention_scale_1.2	5.83
+ngram_lm_scale_5.0_attention_scale_5.0	5.83
+ngram_lm_scale_0.08_attention_scale_1.3	5.84
+ngram_lm_scale_0.08_attention_scale_1.7	5.84
+ngram_lm_scale_0.1_attention_scale_1.3	5.84
+ngram_lm_scale_0.9_attention_scale_0.05	5.84
+ngram_lm_scale_1.9_attention_scale_1.0	5.84
+ngram_lm_scale_3.0_attention_scale_2.3	5.84
+ngram_lm_scale_0.08_attention_scale_1.2	5.85
+ngram_lm_scale_0.08_attention_scale_1.5	5.85
+ngram_lm_scale_0.08_attention_scale_1.9	5.85
+ngram_lm_scale_0.1_attention_scale_1.0	5.85
+ngram_lm_scale_0.1_attention_scale_1.1	5.85
+ngram_lm_scale_0.1_attention_scale_1.2	5.85
+ngram_lm_scale_0.1_attention_scale_1.5	5.85
+ngram_lm_scale_0.1_attention_scale_1.7	5.85
+ngram_lm_scale_0.1_attention_scale_1.9	5.85
+ngram_lm_scale_0.1_attention_scale_2.0	5.85
+ngram_lm_scale_0.1_attention_scale_2.1	5.85
+ngram_lm_scale_0.1_attention_scale_2.2	5.85
+ngram_lm_scale_0.3_attention_scale_0.3	5.85
+ngram_lm_scale_2.5_attention_scale_1.7	5.85
+ngram_lm_scale_0.05_attention_scale_1.3	5.86
+ngram_lm_scale_0.08_attention_scale_1.0	5.86
+ngram_lm_scale_0.08_attention_scale_2.0	5.86
+ngram_lm_scale_0.08_attention_scale_2.1	5.86
+ngram_lm_scale_0.08_attention_scale_2.2	5.86
+ngram_lm_scale_0.1_attention_scale_0.9	5.86
+ngram_lm_scale_0.1_attention_scale_2.3	5.86
+ngram_lm_scale_0.1_attention_scale_2.5	5.86
+ngram_lm_scale_1.0_attention_scale_0.08	5.86
+ngram_lm_scale_2.2_attention_scale_1.3	5.86
+ngram_lm_scale_0.05_attention_scale_1.5	5.87
+ngram_lm_scale_0.05_attention_scale_1.7	5.87
+ngram_lm_scale_0.05_attention_scale_1.9	5.87
+ngram_lm_scale_0.05_attention_scale_2.0	5.87
+ngram_lm_scale_0.05_attention_scale_2.5	5.87
+ngram_lm_scale_0.08_attention_scale_0.9	5.87
+ngram_lm_scale_0.08_attention_scale_1.1	5.87
+ngram_lm_scale_0.08_attention_scale_2.3	5.87
+ngram_lm_scale_0.08_attention_scale_2.5	5.87
+ngram_lm_scale_0.1_attention_scale_3.0	5.87
+ngram_lm_scale_0.9_attention_scale_0.01	5.87
+ngram_lm_scale_0.01_attention_scale_1.5	5.88
+ngram_lm_scale_0.01_attention_scale_1.7	5.88
+ngram_lm_scale_0.01_attention_scale_2.5	5.88
+ngram_lm_scale_0.01_attention_scale_3.0	5.88
+ngram_lm_scale_0.05_attention_scale_1.0	5.88
+ngram_lm_scale_0.05_attention_scale_1.1	5.88
+ngram_lm_scale_0.05_attention_scale_1.2	5.88
+ngram_lm_scale_0.05_attention_scale_2.1	5.88
+ngram_lm_scale_0.05_attention_scale_2.2	5.88
+ngram_lm_scale_0.05_attention_scale_2.3	5.88
+ngram_lm_scale_0.08_attention_scale_3.0	5.88
+ngram_lm_scale_0.3_attention_scale_0.1	5.88
+ngram_lm_scale_3.0_attention_scale_2.2	5.88
+ngram_lm_scale_0.01_attention_scale_1.9	5.89
+ngram_lm_scale_0.01_attention_scale_2.0	5.89
+ngram_lm_scale_0.01_attention_scale_2.1	5.89
+ngram_lm_scale_0.05_attention_scale_0.9	5.89
+ngram_lm_scale_0.05_attention_scale_3.0	5.89
+ngram_lm_scale_0.08_attention_scale_4.0	5.89
+ngram_lm_scale_0.1_attention_scale_0.7	5.89
+ngram_lm_scale_0.1_attention_scale_4.0	5.89
+ngram_lm_scale_0.1_attention_scale_5.0	5.89
+ngram_lm_scale_0.3_attention_scale_0.08	5.89
+ngram_lm_scale_0.01_attention_scale_1.3	5.9
+ngram_lm_scale_0.01_attention_scale_2.2	5.9
+ngram_lm_scale_0.01_attention_scale_2.3	5.9
+ngram_lm_scale_0.05_attention_scale_4.0	5.9
+ngram_lm_scale_0.05_attention_scale_5.0	5.9
+ngram_lm_scale_0.08_attention_scale_0.7	5.9
+ngram_lm_scale_0.08_attention_scale_5.0	5.9
+ngram_lm_scale_0.01_attention_scale_1.1	5.91
+ngram_lm_scale_0.01_attention_scale_1.2	5.91
+ngram_lm_scale_0.01_attention_scale_4.0	5.91
+ngram_lm_scale_0.01_attention_scale_5.0	5.91
+ngram_lm_scale_0.1_attention_scale_0.6	5.91
+ngram_lm_scale_0.3_attention_scale_0.05	5.91
+ngram_lm_scale_1.3_attention_scale_0.3	5.91
+ngram_lm_scale_1.5_attention_scale_0.5	5.91
+ngram_lm_scale_1.7_attention_scale_0.7	5.91
+ngram_lm_scale_1.9_attention_scale_0.9	5.91
+ngram_lm_scale_2.0_attention_scale_1.0	5.91
+ngram_lm_scale_2.1_attention_scale_1.1	5.91
+ngram_lm_scale_0.01_attention_scale_0.9	5.92
+ngram_lm_scale_0.01_attention_scale_1.0	5.92
+ngram_lm_scale_0.05_attention_scale_0.6	5.92
+ngram_lm_scale_0.05_attention_scale_0.7	5.92
+ngram_lm_scale_0.08_attention_scale_0.6	5.92
+ngram_lm_scale_0.1_attention_scale_0.5	5.92
+ngram_lm_scale_1.0_attention_scale_0.05	5.92
+ngram_lm_scale_0.01_attention_scale_0.7	5.94
+ngram_lm_scale_0.05_attention_scale_0.5	5.94
+ngram_lm_scale_0.08_attention_scale_0.5	5.94
+ngram_lm_scale_2.2_attention_scale_1.2	5.94
+ngram_lm_scale_2.3_attention_scale_1.3	5.95
+ngram_lm_scale_3.0_attention_scale_2.1	5.95
+ngram_lm_scale_0.01_attention_scale_0.6	5.96
+ngram_lm_scale_0.3_attention_scale_0.01	5.96
+ngram_lm_scale_1.1_attention_scale_0.1	5.97
+ngram_lm_scale_1.0_attention_scale_0.01	5.98
+ngram_lm_scale_2.5_attention_scale_1.5	5.98
+ngram_lm_scale_0.01_attention_scale_0.5	5.99
+ngram_lm_scale_0.08_attention_scale_0.3	5.99
+ngram_lm_scale_0.1_attention_scale_0.3	5.99
+ngram_lm_scale_3.0_attention_scale_2.0	6.0
+ngram_lm_scale_1.1_attention_scale_0.08	6.01
+ngram_lm_scale_1.7_attention_scale_0.6	6.01
+ngram_lm_scale_0.05_attention_scale_0.3	6.02
+ngram_lm_scale_2.1_attention_scale_1.0	6.02
+ngram_lm_scale_2.0_attention_scale_0.9	6.03
+ngram_lm_scale_2.2_attention_scale_1.1	6.03
+ngram_lm_scale_2.3_attention_scale_1.2	6.03
+ngram_lm_scale_0.01_attention_scale_0.3	6.04
+ngram_lm_scale_4.0_attention_scale_3.0	6.05
+ngram_lm_scale_3.0_attention_scale_1.9	6.07
+ngram_lm_scale_0.08_attention_scale_0.08	6.08
+ngram_lm_scale_0.08_attention_scale_0.1	6.08
+ngram_lm_scale_0.1_attention_scale_0.08	6.08
+ngram_lm_scale_0.1_attention_scale_0.1	6.08
+ngram_lm_scale_1.1_attention_scale_0.05	6.08
+ngram_lm_scale_5.0_attention_scale_4.0	6.08
+ngram_lm_scale_0.1_attention_scale_0.05	6.09
+ngram_lm_scale_0.05_attention_scale_0.1	6.11
+ngram_lm_scale_0.08_attention_scale_0.05	6.11
+ngram_lm_scale_0.05_attention_scale_0.08	6.12
+ngram_lm_scale_0.1_attention_scale_0.01	6.12
+ngram_lm_scale_2.1_attention_scale_0.9	6.12
+ngram_lm_scale_2.2_attention_scale_1.0	6.12
+ngram_lm_scale_2.5_attention_scale_1.3	6.12
+ngram_lm_scale_0.05_attention_scale_0.05	6.13
+ngram_lm_scale_0.08_attention_scale_0.01	6.13
+ngram_lm_scale_1.2_attention_scale_0.1	6.13
+ngram_lm_scale_2.3_attention_scale_1.1	6.13
+ngram_lm_scale_0.01_attention_scale_0.1	6.14
+ngram_lm_scale_1.2_attention_scale_0.08	6.15
+ngram_lm_scale_1.9_attention_scale_0.7	6.15
+ngram_lm_scale_1.1_attention_scale_0.01	6.16
+ngram_lm_scale_0.01_attention_scale_0.08	6.17
+ngram_lm_scale_1.7_attention_scale_0.5	6.17
+ngram_lm_scale_0.05_attention_scale_0.01	6.18
+ngram_lm_scale_1.5_attention_scale_0.3	6.2
+ngram_lm_scale_0.01_attention_scale_0.05	6.21
+ngram_lm_scale_1.2_attention_scale_0.05	6.21
+ngram_lm_scale_2.2_attention_scale_0.9	6.24
+ngram_lm_scale_0.01_attention_scale_0.01	6.25
+ngram_lm_scale_1.3_attention_scale_0.1	6.25
+ngram_lm_scale_2.3_attention_scale_1.0	6.25
+ngram_lm_scale_3.0_attention_scale_1.7	6.25
+ngram_lm_scale_2.0_attention_scale_0.7	6.26
+ngram_lm_scale_1.9_attention_scale_0.6	6.27
+ngram_lm_scale_2.5_attention_scale_1.2	6.27
+ngram_lm_scale_1.2_attention_scale_0.01	6.28
+ngram_lm_scale_1.3_attention_scale_0.08	6.29
+ngram_lm_scale_2.5_attention_scale_1.1	6.35
+ngram_lm_scale_2.3_attention_scale_0.9	6.36
+ngram_lm_scale_4.0_attention_scale_2.5	6.36
+ngram_lm_scale_1.3_attention_scale_0.05	6.38
+ngram_lm_scale_2.0_attention_scale_0.6	6.39
+ngram_lm_scale_2.1_attention_scale_0.7	6.39
+ngram_lm_scale_3.0_attention_scale_1.5	6.4
+ngram_lm_scale_1.9_attention_scale_0.5	6.42
+ngram_lm_scale_2.5_attention_scale_1.0	6.44
+ngram_lm_scale_1.3_attention_scale_0.01	6.45
+ngram_lm_scale_4.0_attention_scale_2.3	6.49
+ngram_lm_scale_1.7_attention_scale_0.3	6.5
+ngram_lm_scale_2.2_attention_scale_0.7	6.5
+ngram_lm_scale_2.1_attention_scale_0.6	6.51
+ngram_lm_scale_2.0_attention_scale_0.5	6.54
+ngram_lm_scale_4.0_attention_scale_2.2	6.54
+ngram_lm_scale_2.5_attention_scale_0.9	6.56
+ngram_lm_scale_3.0_attention_scale_1.3	6.59
+ngram_lm_scale_5.0_attention_scale_3.0	6.59
+ngram_lm_scale_1.5_attention_scale_0.1	6.62
+ngram_lm_scale_4.0_attention_scale_2.1	6.63
+ngram_lm_scale_2.3_attention_scale_0.7	6.64
+ngram_lm_scale_1.5_attention_scale_0.08	6.67
+ngram_lm_scale_2.2_attention_scale_0.6	6.68
+ngram_lm_scale_2.1_attention_scale_0.5	6.7
+ngram_lm_scale_3.0_attention_scale_1.2	6.72
+ngram_lm_scale_4.0_attention_scale_2.0	6.73
+ngram_lm_scale_1.5_attention_scale_0.05	6.75
+ngram_lm_scale_2.3_attention_scale_0.6	6.79
+ngram_lm_scale_4.0_attention_scale_1.9	6.82
+ngram_lm_scale_1.9_attention_scale_0.3	6.83
+ngram_lm_scale_2.2_attention_scale_0.5	6.83
+ngram_lm_scale_1.5_attention_scale_0.01	6.84
+ngram_lm_scale_3.0_attention_scale_1.1	6.84
+ngram_lm_scale_2.5_attention_scale_0.7	6.86
+ngram_lm_scale_5.0_attention_scale_2.5	6.9
+ngram_lm_scale_3.0_attention_scale_1.0	6.92
+ngram_lm_scale_1.7_attention_scale_0.1	6.93
+ngram_lm_scale_2.0_attention_scale_0.3	6.95
+ngram_lm_scale_4.0_attention_scale_1.7	6.96
+ngram_lm_scale_1.7_attention_scale_0.08	6.98
+ngram_lm_scale_2.3_attention_scale_0.5	6.98
+ngram_lm_scale_5.0_attention_scale_2.3	7.02
+ngram_lm_scale_2.5_attention_scale_0.6	7.04
+ngram_lm_scale_1.7_attention_scale_0.05	7.06
+ngram_lm_scale_3.0_attention_scale_0.9	7.07
+ngram_lm_scale_5.0_attention_scale_2.2	7.09
+ngram_lm_scale_2.1_attention_scale_0.3	7.1
+ngram_lm_scale_4.0_attention_scale_1.5	7.14
+ngram_lm_scale_1.7_attention_scale_0.01	7.15
+ngram_lm_scale_5.0_attention_scale_2.1	7.18
+ngram_lm_scale_2.5_attention_scale_0.5	7.23
+ngram_lm_scale_2.2_attention_scale_0.3	7.24
+ngram_lm_scale_5.0_attention_scale_2.0	7.25
+ngram_lm_scale_1.9_attention_scale_0.1	7.26
+ngram_lm_scale_1.9_attention_scale_0.08	7.31
+ngram_lm_scale_4.0_attention_scale_1.3	7.34
+ngram_lm_scale_2.3_attention_scale_0.3	7.35
+ngram_lm_scale_5.0_attention_scale_1.9	7.35
+ngram_lm_scale_3.0_attention_scale_0.7	7.36
+ngram_lm_scale_1.9_attention_scale_0.05	7.38
+ngram_lm_scale_2.0_attention_scale_0.1	7.39
+ngram_lm_scale_2.0_attention_scale_0.08	7.42
+ngram_lm_scale_4.0_attention_scale_1.2	7.46
+ngram_lm_scale_1.9_attention_scale_0.01	7.49
+ngram_lm_scale_2.0_attention_scale_0.05	7.49
+ngram_lm_scale_2.1_attention_scale_0.1	7.51
+ngram_lm_scale_3.0_attention_scale_0.6	7.52
+ngram_lm_scale_5.0_attention_scale_1.7	7.55
+ngram_lm_scale_2.1_attention_scale_0.08	7.56
+ngram_lm_scale_2.5_attention_scale_0.3	7.56
+ngram_lm_scale_4.0_attention_scale_1.1	7.57
+ngram_lm_scale_2.0_attention_scale_0.01	7.61
+ngram_lm_scale_2.1_attention_scale_0.05	7.62
+ngram_lm_scale_2.2_attention_scale_0.1	7.62
+ngram_lm_scale_3.0_attention_scale_0.5	7.65
+ngram_lm_scale_2.2_attention_scale_0.08	7.66
+ngram_lm_scale_4.0_attention_scale_1.0	7.69
+ngram_lm_scale_5.0_attention_scale_1.5	7.7
+ngram_lm_scale_2.3_attention_scale_0.1	7.71
+ngram_lm_scale_2.1_attention_scale_0.01	7.72
+ngram_lm_scale_2.2_attention_scale_0.05	7.72
+ngram_lm_scale_2.3_attention_scale_0.08	7.75
+ngram_lm_scale_2.2_attention_scale_0.01	7.79
+ngram_lm_scale_4.0_attention_scale_0.9	7.79
+ngram_lm_scale_2.3_attention_scale_0.05	7.81
+ngram_lm_scale_5.0_attention_scale_1.3	7.89
+ngram_lm_scale_2.3_attention_scale_0.01	7.92
+ngram_lm_scale_2.5_attention_scale_0.1	7.92
+ngram_lm_scale_2.5_attention_scale_0.08	7.96
+ngram_lm_scale_3.0_attention_scale_0.3	7.97
+ngram_lm_scale_5.0_attention_scale_1.2	7.98
+ngram_lm_scale_2.5_attention_scale_0.05	8.01
+ngram_lm_scale_4.0_attention_scale_0.7	8.04
+ngram_lm_scale_5.0_attention_scale_1.1	8.06
+ngram_lm_scale_2.5_attention_scale_0.01	8.08
+ngram_lm_scale_4.0_attention_scale_0.6	8.14
+ngram_lm_scale_5.0_attention_scale_1.0	8.17
+ngram_lm_scale_5.0_attention_scale_0.9	8.27
+ngram_lm_scale_4.0_attention_scale_0.5	8.28
+ngram_lm_scale_3.0_attention_scale_0.1	8.3
+ngram_lm_scale_3.0_attention_scale_0.08	8.34
+ngram_lm_scale_3.0_attention_scale_0.05	8.38
+ngram_lm_scale_5.0_attention_scale_0.7	8.43
+ngram_lm_scale_3.0_attention_scale_0.01	8.46
+ngram_lm_scale_4.0_attention_scale_0.3	8.48
+ngram_lm_scale_5.0_attention_scale_0.6	8.51
+ngram_lm_scale_5.0_attention_scale_0.5	8.58
+ngram_lm_scale_4.0_attention_scale_0.1	8.68
+ngram_lm_scale_4.0_attention_scale_0.08	8.71
+ngram_lm_scale_4.0_attention_scale_0.05	8.73
+ngram_lm_scale_5.0_attention_scale_0.3	8.75
+ngram_lm_scale_4.0_attention_scale_0.01	8.8
+ngram_lm_scale_5.0_attention_scale_0.1	8.97
+ngram_lm_scale_5.0_attention_scale_0.08	8.98
+ngram_lm_scale_5.0_attention_scale_0.05	9.0
+ngram_lm_scale_5.0_attention_scale_0.01	9.03
+2022-07-20 21:46:22,023 INFO [decode.py:868] Done!

log/log-ctc-decoding/log-decode-2022-07-21-23-23-55 ADDED Viewed

	@@ -0,0 +1,27 @@

+2022-07-21 23:23:55,621 INFO [decode.py:743] Decoding started
+2022-07-21 23:23:55,622 INFO [decode.py:744] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.17', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '9e88318adfd7c80290a96f8a888d279d45dc1564', 'k2-git-date': 'Mon Jul 18 16:26:06 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'b9fda2c-dirty', 'icefall-git-date': 'Thu Jul 21 22:06:09 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/open_source/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'ctc-decoding', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'rnn_lm_exp_dir': 'rnn_lm/exp', 'rnn_lm_epoch': 7, 'rnn_lm_avg': 2, 'rnn_lm_embedding_dim': 2048, 'rnn_lm_hidden_dim': 2048, 'rnn_lm_num_layers': 4, 'rnn_lm_tie_weights': False, 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 200, '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'}
+2022-07-21 23:23:56,049 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-21 23:23:56,127 INFO [decode.py:754] device: cuda:0
+2022-07-21 23:24:03,033 INFO [decode.py:916] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-21 23:24:05,616 INFO [decode.py:932] Number of model parameters: 103071035
+2022-07-21 23:24:05,616 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-21 23:24:05,632 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-21 23:24:06,872 INFO [decode.py:678] batch 0/?, cuts processed until now is 15
+2022-07-21 23:24:41,077 INFO [decode.py:678] batch 100/?, cuts processed until now is 2578
+2022-07-21 23:24:41,935 INFO [decode.py:699] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-ctc-decoding.txt
+2022-07-21 23:24:42,014 INFO [utils.py:416] [test-clean-ctc-decoding] %WER 2.98% [1568 / 52576, 162 ins, 113 del, 1293 sub ]
+2022-07-21 23:24:42,241 INFO [decode.py:711] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-ctc-decoding.txt
+2022-07-21 23:24:42,244 INFO [decode.py:727]
+For test-clean, WER of different settings are:
+ctc-decoding	2.98	best for test-clean
+2022-07-21 23:24:43,208 INFO [decode.py:678] batch 0/?, cuts processed until now is 19
+2022-07-21 23:25:17,339 INFO [decode.py:678] batch 100/?, cuts processed until now is 2888
+2022-07-21 23:25:17,768 INFO [decode.py:699] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-ctc-decoding.txt
+2022-07-21 23:25:17,849 INFO [utils.py:416] [test-other-ctc-decoding] %WER 7.14% [3735 / 52343, 375 ins, 280 del, 3080 sub ]
+2022-07-21 23:25:18,086 INFO [decode.py:711] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-ctc-decoding.txt
+2022-07-21 23:25:18,087 INFO [decode.py:727]
+For test-other, WER of different settings are:
+ctc-decoding	7.14	best for test-other
+2022-07-21 23:25:18,087 INFO [decode.py:989] Done!

log/log-ctc-greedy-search/log-decode-2022-07-20-15-54-39 ADDED Viewed

	@@ -0,0 +1,27 @@

+2022-07-20 15:54:39,939 INFO [decode.py:653] Decoding started
+2022-07-20 15:54:39,939 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 4, 'output_beam': 4, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': '0738a44-dirty', 'icefall-git-date': 'Mon Jul 18 14:44:26 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'ctc-greedy-search', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 200, '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'}
+2022-07-20 15:54:40,359 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 15:54:40,416 INFO [decode.py:664] device: cuda:0
+2022-07-20 15:54:46,648 INFO [decode.py:817] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 15:54:49,543 INFO [decode.py:833] Number of model parameters: 103071035
+2022-07-20 15:54:49,544 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 15:54:49,602 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 15:54:50,853 INFO [decode.py:588] batch 0/?, cuts processed until now is 15
+2022-07-20 15:55:16,670 INFO [decode.py:588] batch 100/?, cuts processed until now is 2578
+2022-07-20 15:55:17,385 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-ctc-greedy-search.txt
+2022-07-20 15:55:17,464 INFO [utils.py:410] [test-clean-ctc-greedy-search] %WER 2.98% [1568 / 52576, 162 ins, 113 del, 1293 sub ]
+2022-07-20 15:55:17,698 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-ctc-greedy-search.txt
+2022-07-20 15:55:17,699 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+ctc-greedy-search	2.98	best for test-clean
+2022-07-20 15:55:18,620 INFO [decode.py:588] batch 0/?, cuts processed until now is 19
+2022-07-20 15:55:43,222 INFO [decode.py:588] batch 100/?, cuts processed until now is 2888
+2022-07-20 15:55:43,554 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-ctc-greedy-search.txt
+2022-07-20 15:55:43,643 INFO [utils.py:410] [test-other-ctc-greedy-search] %WER 7.14% [3735 / 52343, 375 ins, 280 del, 3080 sub ]
+2022-07-20 15:55:43,890 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-ctc-greedy-search.txt
+2022-07-20 15:55:43,891 INFO [decode.py:637]
+For test-other, WER of different settings are:
+ctc-greedy-search	7.14	best for test-other
+2022-07-20 15:55:43,891 INFO [decode.py:864] Done!

log/log-nbest-oracle/log-decode-2022-07-20-20-48-51 ADDED Viewed

	@@ -0,0 +1,27 @@

+2022-07-20 20:48:51,656 INFO [decode.py:653] Decoding started
+2022-07-20 20:48:51,657 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'd812f82-dirty', 'icefall-git-date': 'Wed Jul 20 19:47:36 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/open_source/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'nbest-oracle', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 200, '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'}
+2022-07-20 20:48:52,068 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 20:48:52,133 INFO [decode.py:664] device: cuda:0
+2022-07-20 20:48:58,588 INFO [decode.py:821] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 20:49:01,210 INFO [decode.py:837] Number of model parameters: 103071035
+2022-07-20 20:49:01,211 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 20:49:01,216 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 20:49:02,456 INFO [decode.py:588] batch 0/?, cuts processed until now is 15
+2022-07-20 20:49:33,689 INFO [decode.py:588] batch 100/?, cuts processed until now is 2578
+2022-07-20 20:49:34,651 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-oracle_100_nbest_scale_0.5.txt
+2022-07-20 20:49:34,728 INFO [utils.py:416] [test-clean-oracle_100_nbest_scale_0.5] %WER 1.53% [805 / 52576, 56 ins, 111 del, 638 sub ]
+2022-07-20 20:49:34,958 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-oracle_100_nbest_scale_0.5.txt
+2022-07-20 20:49:34,958 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+oracle_100_nbest_scale_0.5	1.53	best for test-clean
+2022-07-20 20:49:35,889 INFO [decode.py:588] batch 0/?, cuts processed until now is 19
+2022-07-20 20:50:05,816 INFO [decode.py:588] batch 100/?, cuts processed until now is 2888
+2022-07-20 20:50:06,184 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-oracle_100_nbest_scale_0.5.txt
+2022-07-20 20:50:06,262 INFO [utils.py:416] [test-other-oracle_100_nbest_scale_0.5] %WER 3.47% [1818 / 52343, 111 ins, 321 del, 1386 sub ]
+2022-07-20 20:50:06,486 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-oracle_100_nbest_scale_0.5.txt
+2022-07-20 20:50:06,487 INFO [decode.py:637]
+For test-other, WER of different settings are:
+oracle_100_nbest_scale_0.5	3.47	best for test-other
+2022-07-20 20:50:06,487 INFO [decode.py:868] Done!

log/log-nbest-rescoring/log-decode-2022-07-20-21-16-48 ADDED Viewed

	@@ -0,0 +1,192 @@

+2022-07-20 21:16:48,757 INFO [decode.py:653] Decoding started
+2022-07-20 21:16:48,757 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'd812f82-dirty', 'icefall-git-date': 'Wed Jul 20 19:47:36 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'nbest-rescoring', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, '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'}
+2022-07-20 21:16:49,153 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 21:16:49,217 INFO [decode.py:664] device: cuda:0
+2022-07-20 21:16:54,771 INFO [decode.py:730] Loading pre-compiled G_4_gram.pt
+2022-07-20 21:16:59,308 INFO [decode.py:821] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 21:17:01,993 INFO [decode.py:837] Number of model parameters: 103071035
+2022-07-20 21:17:01,993 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 21:17:01,997 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 21:17:03,382 INFO [decode.py:588] batch 0/?, cuts processed until now is 2
+2022-07-20 21:17:47,578 INFO [decode.py:588] batch 100/?, cuts processed until now is 326
+2022-07-20 21:18:29,336 INFO [decode.py:588] batch 200/?, cuts processed until now is 709
+2022-07-20 21:19:07,578 INFO [decode.py:588] batch 300/?, cuts processed until now is 1126
+2022-07-20 21:19:46,337 INFO [decode.py:588] batch 400/?, cuts processed until now is 1489
+2022-07-20 21:20:27,408 INFO [decode.py:588] batch 500/?, cuts processed until now is 1823
+2022-07-20 21:21:06,308 INFO [decode.py:588] batch 600/?, cuts processed until now is 2211
+2022-07-20 21:21:50,370 INFO [decode.py:588] batch 700/?, cuts processed until now is 2474
+2022-07-20 21:22:35,645 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.1.txt
+2022-07-20 21:22:35,734 INFO [utils.py:410] [test-clean-lm_scale_0.1] %WER 2.81% [1476 / 52576, 202 ins, 102 del, 1172 sub ]
+2022-07-20 21:22:36,154 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.1.txt
+2022-07-20 21:22:36,184 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.2.txt
+2022-07-20 21:22:36,264 INFO [utils.py:410] [test-clean-lm_scale_0.2] %WER 2.75% [1447 / 52576, 199 ins, 105 del, 1143 sub ]
+2022-07-20 21:22:36,496 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.2.txt
+2022-07-20 21:22:36,525 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.3.txt
+2022-07-20 21:22:36,607 INFO [utils.py:410] [test-clean-lm_scale_0.3] %WER 2.72% [1428 / 52576, 195 ins, 107 del, 1126 sub ]
+2022-07-20 21:22:36,840 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.3.txt
+2022-07-20 21:22:36,870 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.4.txt
+2022-07-20 21:22:36,949 INFO [utils.py:410] [test-clean-lm_scale_0.4] %WER 2.69% [1412 / 52576, 192 ins, 112 del, 1108 sub ]
+2022-07-20 21:22:37,184 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.4.txt
+2022-07-20 21:22:37,213 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.5.txt
+2022-07-20 21:22:37,288 INFO [utils.py:410] [test-clean-lm_scale_0.5] %WER 2.68% [1409 / 52576, 188 ins, 115 del, 1106 sub ]
+2022-07-20 21:22:37,516 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.5.txt
+2022-07-20 21:22:37,544 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.6.txt
+2022-07-20 21:22:37,621 INFO [utils.py:410] [test-clean-lm_scale_0.6] %WER 2.68% [1411 / 52576, 181 ins, 121 del, 1109 sub ]
+2022-07-20 21:22:37,845 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.6.txt
+2022-07-20 21:22:37,872 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.7.txt
+2022-07-20 21:22:37,948 INFO [utils.py:410] [test-clean-lm_scale_0.7] %WER 2.68% [1409 / 52576, 173 ins, 131 del, 1105 sub ]
+2022-07-20 21:22:38,179 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.7.txt
+2022-07-20 21:22:38,206 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.8.txt
+2022-07-20 21:22:38,282 INFO [utils.py:410] [test-clean-lm_scale_0.8] %WER 2.71% [1425 / 52576, 169 ins, 144 del, 1112 sub ]
+2022-07-20 21:22:38,656 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.8.txt
+2022-07-20 21:22:38,684 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.9.txt
+2022-07-20 21:22:38,762 INFO [utils.py:410] [test-clean-lm_scale_0.9] %WER 2.74% [1441 / 52576, 156 ins, 164 del, 1121 sub ]
+2022-07-20 21:22:38,987 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.9.txt
+2022-07-20 21:22:39,016 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.0.txt
+2022-07-20 21:22:39,102 INFO [utils.py:410] [test-clean-lm_scale_1.0] %WER 2.82% [1485 / 52576, 156 ins, 197 del, 1132 sub ]
+2022-07-20 21:22:39,328 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.0.txt
+2022-07-20 21:22:39,356 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.1.txt
+2022-07-20 21:22:39,432 INFO [utils.py:410] [test-clean-lm_scale_1.1] %WER 2.91% [1530 / 52576, 153 ins, 234 del, 1143 sub ]
+2022-07-20 21:22:39,667 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.1.txt
+2022-07-20 21:22:39,703 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.2.txt
+2022-07-20 21:22:39,779 INFO [utils.py:410] [test-clean-lm_scale_1.2] %WER 3.04% [1600 / 52576, 150 ins, 283 del, 1167 sub ]
+2022-07-20 21:22:40,004 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.2.txt
+2022-07-20 21:22:40,036 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.3.txt
+2022-07-20 21:22:40,112 INFO [utils.py:410] [test-clean-lm_scale_1.3] %WER 3.23% [1697 / 52576, 150 ins, 347 del, 1200 sub ]
+2022-07-20 21:22:40,341 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.3.txt
+2022-07-20 21:22:40,372 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.4.txt
+2022-07-20 21:22:40,448 INFO [utils.py:410] [test-clean-lm_scale_1.4] %WER 3.41% [1792 / 52576, 149 ins, 420 del, 1223 sub ]
+2022-07-20 21:22:40,674 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.4.txt
+2022-07-20 21:22:40,705 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.5.txt
+2022-07-20 21:22:40,950 INFO [utils.py:410] [test-clean-lm_scale_1.5] %WER 3.62% [1902 / 52576, 146 ins, 493 del, 1263 sub ]
+2022-07-20 21:22:41,181 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.5.txt
+2022-07-20 21:22:41,209 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.6.txt
+2022-07-20 21:22:41,286 INFO [utils.py:410] [test-clean-lm_scale_1.6] %WER 3.81% [2001 / 52576, 140 ins, 573 del, 1288 sub ]
+2022-07-20 21:22:41,511 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.6.txt
+2022-07-20 21:22:41,538 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.7.txt
+2022-07-20 21:22:41,614 INFO [utils.py:410] [test-clean-lm_scale_1.7] %WER 3.99% [2100 / 52576, 142 ins, 651 del, 1307 sub ]
+2022-07-20 21:22:41,841 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.7.txt
+2022-07-20 21:22:41,868 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.8.txt
+2022-07-20 21:22:41,949 INFO [utils.py:410] [test-clean-lm_scale_1.8] %WER 4.16% [2187 / 52576, 143 ins, 722 del, 1322 sub ]
+2022-07-20 21:22:42,176 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.8.txt
+2022-07-20 21:22:42,204 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.9.txt
+2022-07-20 21:22:42,281 INFO [utils.py:410] [test-clean-lm_scale_1.9] %WER 4.32% [2272 / 52576, 141 ins, 776 del, 1355 sub ]
+2022-07-20 21:22:42,508 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.9.txt
+2022-07-20 21:22:42,536 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_2.0.txt
+2022-07-20 21:22:42,612 INFO [utils.py:410] [test-clean-lm_scale_2.0] %WER 4.54% [2386 / 52576, 143 ins, 847 del, 1396 sub ]
+2022-07-20 21:22:42,843 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_2.0.txt
+2022-07-20 21:22:42,844 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+lm_scale_0.5	2.68	best for test-clean
+lm_scale_0.6	2.68
+lm_scale_0.7	2.68
+lm_scale_0.4	2.69
+lm_scale_0.8	2.71
+lm_scale_0.3	2.72
+lm_scale_0.9	2.74
+lm_scale_0.2	2.75
+lm_scale_0.1	2.81
+lm_scale_1.0	2.82
+lm_scale_1.1	2.91
+lm_scale_1.2	3.04
+lm_scale_1.3	3.23
+lm_scale_1.4	3.41
+lm_scale_1.5	3.62
+lm_scale_1.6	3.81
+lm_scale_1.7	3.99
+lm_scale_1.8	4.16
+lm_scale_1.9	4.32
+lm_scale_2.0	4.54
+2022-07-20 21:22:43,967 INFO [decode.py:588] batch 0/?, cuts processed until now is 2
+2022-07-20 21:23:26,720 INFO [decode.py:588] batch 100/?, cuts processed until now is 377
+2022-07-20 21:24:06,356 INFO [decode.py:588] batch 200/?, cuts processed until now is 811
+2022-07-20 21:24:45,879 INFO [decode.py:588] batch 300/?, cuts processed until now is 1265
+2022-07-20 21:25:24,826 INFO [decode.py:588] batch 400/?, cuts processed until now is 1684
+2022-07-20 21:26:06,471 INFO [decode.py:588] batch 500/?, cuts processed until now is 2061
+2022-07-20 21:26:50,778 INFO [decode.py:588] batch 600/?, cuts processed until now is 2495
+2022-07-20 21:27:36,625 INFO [decode.py:588] batch 700/?, cuts processed until now is 2781
+2022-07-20 21:28:29,094 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.1.txt
+2022-07-20 21:28:29,184 INFO [utils.py:410] [test-other-lm_scale_0.1] %WER 6.16% [3224 / 52343, 397 ins, 263 del, 2564 sub ]
+2022-07-20 21:28:29,431 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.1.txt
+2022-07-20 21:28:29,464 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.2.txt
+2022-07-20 21:28:29,553 INFO [utils.py:410] [test-other-lm_scale_0.2] %WER 6.01% [3148 / 52343, 379 ins, 262 del, 2507 sub ]
+2022-07-20 21:28:29,800 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.2.txt
+2022-07-20 21:28:29,830 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.3.txt
+2022-07-20 21:28:29,912 INFO [utils.py:410] [test-other-lm_scale_0.3] %WER 5.93% [3105 / 52343, 365 ins, 269 del, 2471 sub ]
+2022-07-20 21:28:30,172 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.3.txt
+2022-07-20 21:28:30,206 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.4.txt
+2022-07-20 21:28:30,286 INFO [utils.py:410] [test-other-lm_scale_0.4] %WER 5.84% [3057 / 52343, 347 ins, 291 del, 2419 sub ]
+2022-07-20 21:28:30,523 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.4.txt
+2022-07-20 21:28:30,556 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.5.txt
+2022-07-20 21:28:30,640 INFO [utils.py:410] [test-other-lm_scale_0.5] %WER 5.77% [3021 / 52343, 333 ins, 301 del, 2387 sub ]
+2022-07-20 21:28:30,885 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.5.txt
+2022-07-20 21:28:30,920 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.6.txt
+2022-07-20 21:28:31,004 INFO [utils.py:410] [test-other-lm_scale_0.6] %WER 5.78% [3027 / 52343, 326 ins, 333 del, 2368 sub ]
+2022-07-20 21:28:31,246 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.6.txt
+2022-07-20 21:28:31,285 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.7.txt
+2022-07-20 21:28:31,367 INFO [utils.py:410] [test-other-lm_scale_0.7] %WER 5.82% [3046 / 52343, 314 ins, 369 del, 2363 sub ]
+2022-07-20 21:28:31,607 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.7.txt
+2022-07-20 21:28:31,641 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.8.txt
+2022-07-20 21:28:31,732 INFO [utils.py:410] [test-other-lm_scale_0.8] %WER 5.82% [3045 / 52343, 297 ins, 412 del, 2336 sub ]
+2022-07-20 21:28:31,974 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.8.txt
+2022-07-20 21:28:32,006 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.9.txt
+2022-07-20 21:28:32,086 INFO [utils.py:410] [test-other-lm_scale_0.9] %WER 5.89% [3084 / 52343, 285 ins, 463 del, 2336 sub ]
+2022-07-20 21:28:32,543 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.9.txt
+2022-07-20 21:28:32,576 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.0.txt
+2022-07-20 21:28:32,660 INFO [utils.py:410] [test-other-lm_scale_1.0] %WER 6.04% [3159 / 52343, 273 ins, 530 del, 2356 sub ]
+2022-07-20 21:28:32,903 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.0.txt
+2022-07-20 21:28:32,936 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.1.txt
+2022-07-20 21:28:33,019 INFO [utils.py:410] [test-other-lm_scale_1.1] %WER 6.20% [3244 / 52343, 268 ins, 598 del, 2378 sub ]
+2022-07-20 21:28:33,282 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.1.txt
+2022-07-20 21:28:33,312 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.2.txt
+2022-07-20 21:28:33,397 INFO [utils.py:410] [test-other-lm_scale_1.2] %WER 6.31% [3302 / 52343, 257 ins, 667 del, 2378 sub ]
+2022-07-20 21:28:33,635 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.2.txt
+2022-07-20 21:28:33,667 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.3.txt
+2022-07-20 21:28:33,750 INFO [utils.py:410] [test-other-lm_scale_1.3] %WER 6.58% [3444 / 52343, 250 ins, 787 del, 2407 sub ]
+2022-07-20 21:28:33,989 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.3.txt
+2022-07-20 21:28:34,021 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.4.txt
+2022-07-20 21:28:34,105 INFO [utils.py:410] [test-other-lm_scale_1.4] %WER 6.88% [3603 / 52343, 246 ins, 921 del, 2436 sub ]
+2022-07-20 21:28:34,341 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.4.txt
+2022-07-20 21:28:34,373 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.5.txt
+2022-07-20 21:28:34,468 INFO [utils.py:410] [test-other-lm_scale_1.5] %WER 7.09% [3713 / 52343, 235 ins, 1031 del, 2447 sub ]
+2022-07-20 21:28:34,706 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.5.txt
+2022-07-20 21:28:34,738 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.6.txt
+2022-07-20 21:28:34,820 INFO [utils.py:410] [test-other-lm_scale_1.6] %WER 7.32% [3834 / 52343, 227 ins, 1131 del, 2476 sub ]
+2022-07-20 21:28:35,258 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.6.txt
+2022-07-20 21:28:35,287 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.7.txt
+2022-07-20 21:28:35,380 INFO [utils.py:410] [test-other-lm_scale_1.7] %WER 7.54% [3945 / 52343, 223 ins, 1227 del, 2495 sub ]
+2022-07-20 21:28:35,616 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.7.txt
+2022-07-20 21:28:35,649 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.8.txt
+2022-07-20 21:28:35,732 INFO [utils.py:410] [test-other-lm_scale_1.8] %WER 7.75% [4056 / 52343, 221 ins, 1320 del, 2515 sub ]
+2022-07-20 21:28:35,969 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.8.txt
+2022-07-20 21:28:35,998 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.9.txt
+2022-07-20 21:28:36,106 INFO [utils.py:410] [test-other-lm_scale_1.9] %WER 7.97% [4170 / 52343, 222 ins, 1401 del, 2547 sub ]
+2022-07-20 21:28:36,339 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.9.txt
+2022-07-20 21:28:36,367 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_2.0.txt
+2022-07-20 21:28:36,447 INFO [utils.py:410] [test-other-lm_scale_2.0] %WER 8.18% [4281 / 52343, 220 ins, 1479 del, 2582 sub ]
+2022-07-20 21:28:36,681 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_2.0.txt
+2022-07-20 21:28:36,681 INFO [decode.py:637]
+For test-other, WER of different settings are:
+lm_scale_0.5	5.77	best for test-other
+lm_scale_0.6	5.78
+lm_scale_0.7	5.82
+lm_scale_0.8	5.82
+lm_scale_0.4	5.84
+lm_scale_0.9	5.89
+lm_scale_0.3	5.93
+lm_scale_0.2	6.01
+lm_scale_1.0	6.04
+lm_scale_0.1	6.16
+lm_scale_1.1	6.2
+lm_scale_1.2	6.31
+lm_scale_1.3	6.58
+lm_scale_1.4	6.88
+lm_scale_1.5	7.09
+lm_scale_1.6	7.32
+lm_scale_1.7	7.54
+lm_scale_1.8	7.75
+lm_scale_1.9	7.97
+lm_scale_2.0	8.18
+2022-07-20 21:28:36,682 INFO [decode.py:868] Done!

log/log-nbest/log-decode-2022-07-20-21-02-14 ADDED Viewed

	@@ -0,0 +1,48 @@

+2022-07-20 21:02:14,473 INFO [decode.py:653] Decoding started
+2022-07-20 21:02:14,473 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'd812f82-dirty', 'icefall-git-date': 'Wed Jul 20 19:47:36 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/open_source/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'nbest', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 20, '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'}
+2022-07-20 21:02:14,869 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 21:02:14,927 INFO [decode.py:664] device: cuda:0
+2022-07-20 21:02:22,144 INFO [decode.py:821] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 21:02:24,985 INFO [decode.py:837] Number of model parameters: 103071035
+2022-07-20 21:02:24,986 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 21:02:24,988 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 21:02:26,237 INFO [decode.py:588] batch 0/?, cuts processed until now is 1
+2022-07-20 21:03:00,467 INFO [decode.py:588] batch 100/?, cuts processed until now is 212
+2022-07-20 21:03:32,099 INFO [decode.py:588] batch 200/?, cuts processed until now is 459
+2022-07-20 21:04:01,741 INFO [decode.py:588] batch 300/?, cuts processed until now is 717
+2022-07-20 21:04:31,150 INFO [decode.py:588] batch 400/?, cuts processed until now is 949
+2022-07-20 21:05:03,456 INFO [decode.py:588] batch 500/?, cuts processed until now is 1163
+2022-07-20 21:05:33,288 INFO [decode.py:588] batch 600/?, cuts processed until now is 1411
+2022-07-20 21:06:01,818 INFO [decode.py:588] batch 700/?, cuts processed until now is 1652
+2022-07-20 21:06:33,143 INFO [decode.py:588] batch 800/?, cuts processed until now is 1867
+2022-07-20 21:06:59,955 INFO [decode.py:588] batch 900/?, cuts processed until now is 2122
+2022-07-20 21:07:30,299 INFO [decode.py:588] batch 1000/?, cuts processed until now is 2321
+2022-07-20 21:08:02,342 INFO [decode.py:588] batch 1100/?, cuts processed until now is 2468
+2022-07-20 21:08:32,616 INFO [decode.py:588] batch 1200/?, cuts processed until now is 2601
+2022-07-20 21:08:37,175 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-no_rescore-nbest-scale-0.5-100.txt
+2022-07-20 21:08:37,260 INFO [utils.py:416] [test-clean-no_rescore-nbest-scale-0.5-100] %WER 2.94% [1547 / 52576, 149 ins, 218 del, 1180 sub ]
+2022-07-20 21:08:37,488 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-no_rescore-nbest-scale-0.5-100.txt
+2022-07-20 21:08:37,489 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+no_rescore-nbest-scale-0.5-100	2.94	best for test-clean
+2022-07-20 21:08:38,299 INFO [decode.py:588] batch 0/?, cuts processed until now is 2
+2022-07-20 21:09:09,190 INFO [decode.py:588] batch 100/?, cuts processed until now is 249
+2022-07-20 21:09:38,759 INFO [decode.py:588] batch 200/?, cuts processed until now is 540
+2022-07-20 21:10:06,577 INFO [decode.py:588] batch 300/?, cuts processed until now is 834
+2022-07-20 21:10:34,377 INFO [decode.py:588] batch 400/?, cuts processed until now is 1104
+2022-07-20 21:11:05,680 INFO [decode.py:588] batch 500/?, cuts processed until now is 1350
+2022-07-20 21:11:36,410 INFO [decode.py:588] batch 600/?, cuts processed until now is 1640
+2022-07-20 21:12:03,632 INFO [decode.py:588] batch 700/?, cuts processed until now is 1933
+2022-07-20 21:12:32,513 INFO [decode.py:588] batch 800/?, cuts processed until now is 2175
+2022-07-20 21:13:02,611 INFO [decode.py:588] batch 900/?, cuts processed until now is 2440
+2022-07-20 21:13:33,536 INFO [decode.py:588] batch 1000/?, cuts processed until now is 2656
+2022-07-20 21:14:04,886 INFO [decode.py:588] batch 1100/?, cuts processed until now is 2839
+2022-07-20 21:14:34,022 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-no_rescore-nbest-scale-0.5-100.txt
+2022-07-20 21:14:34,110 INFO [utils.py:416] [test-other-no_rescore-nbest-scale-0.5-100] %WER 6.39% [3346 / 52343, 259 ins, 590 del, 2497 sub ]
+2022-07-20 21:14:34,427 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-no_rescore-nbest-scale-0.5-100.txt
+2022-07-20 21:14:34,428 INFO [decode.py:637]
+For test-other, WER of different settings are:
+no_rescore-nbest-scale-0.5-100	6.39	best for test-other
+2022-07-20 21:14:34,428 INFO [decode.py:868] Done!

log/log-whole-lattice-rescoring/log-decode-2022-07-20-21-02-14 ADDED Viewed

	@@ -0,0 +1,207 @@

+2022-07-20 21:02:14,439 INFO [decode.py:653] Decoding started
+2022-07-20 21:02:14,440 INFO [decode.py:654] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 8, 'dim_feedforward': 2048, 'encoder_dim': 512, 'num_encoder_layers': 12, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.15.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '350e449fe5d7b4231f77eb0add764782eed9f5d2', 'k2-git-date': 'Thu May 26 22:56:24 2022', 'lhotse-version': '1.3.0.dev+git.232f3eb.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'conformer_ctc2', 'icefall-git-sha1': 'd812f82-dirty', 'icefall-git-date': 'Wed Jul 20 19:47:36 2022', 'icefall-path': '/home/storage23/wangquandong/experiment/open_source/icefall', 'k2-path': '/home/storage23/wangquandong/tools/k2/k2/python/k2/__init__.py', 'lhotse-path': '/home/storage23/wangquandong/tools/lhotse/lhotse/__init__.py', 'hostname': 'tj1-asr-train-v100-30.kscn', 'IP address': '10.38.22.183'}, 'epoch': 30, 'iter': 0, 'avg': 8, 'method': 'whole-lattice-rescoring', 'use_averaged_model': True, 'num_decoder_layers': 6, 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'lm_dir': PosixPath('data/lm'), 'full_libri': True, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 20, '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'}
+2022-07-20 21:02:14,840 INFO [lexicon.py:176] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2022-07-20 21:02:14,895 INFO [decode.py:664] device: cuda:0
+2022-07-20 21:02:21,292 INFO [decode.py:730] Loading pre-compiled G_4_gram.pt
+2022-07-20 21:02:26,887 INFO [decode.py:821] Calculating the averaged model over epoch range from 22 (excluded) to 30
+2022-07-20 21:02:30,017 INFO [decode.py:837] Number of model parameters: 103071035
+2022-07-20 21:02:30,018 INFO [asr_datamodule.py:444] About to get test-clean cuts
+2022-07-20 21:02:30,022 INFO [asr_datamodule.py:451] About to get test-other cuts
+2022-07-20 21:02:31,504 INFO [decode.py:588] batch 0/?, cuts processed until now is 1
+2022-07-20 21:03:17,856 INFO [decode.py:588] batch 100/?, cuts processed until now is 212
+2022-07-20 21:03:58,002 INFO [decode.py:588] batch 200/?, cuts processed until now is 459
+2022-07-20 21:04:39,673 INFO [decode.py:588] batch 300/?, cuts processed until now is 717
+2022-07-20 21:05:25,026 INFO [decode.py:588] batch 400/?, cuts processed until now is 949
+2022-07-20 21:06:10,220 INFO [decode.py:588] batch 500/?, cuts processed until now is 1163
+2022-07-20 21:06:51,165 INFO [decode.py:588] batch 600/?, cuts processed until now is 1411
+2022-07-20 21:07:30,465 INFO [decode.py:588] batch 700/?, cuts processed until now is 1652
+2022-07-20 21:08:14,019 INFO [decode.py:588] batch 800/?, cuts processed until now is 1867
+2022-07-20 21:08:52,156 INFO [decode.py:588] batch 900/?, cuts processed until now is 2122
+2022-07-20 21:09:33,815 INFO [decode.py:588] batch 1000/?, cuts processed until now is 2321
+2022-07-20 21:10:18,006 INFO [decode.py:588] batch 1100/?, cuts processed until now is 2468
+2022-07-20 21:11:00,039 INFO [decode.py:588] batch 1200/?, cuts processed until now is 2601
+2022-07-20 21:11:06,549 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.1.txt
+2022-07-20 21:11:06,631 INFO [utils.py:416] [test-clean-lm_scale_0.1] %WER 2.82% [1482 / 52576, 214 ins, 100 del, 1168 sub ]
+2022-07-20 21:11:06,864 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.1.txt
+2022-07-20 21:11:06,893 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.2.txt
+2022-07-20 21:11:06,973 INFO [utils.py:416] [test-clean-lm_scale_0.2] %WER 2.74% [1440 / 52576, 206 ins, 102 del, 1132 sub ]
+2022-07-20 21:11:07,204 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.2.txt
+2022-07-20 21:11:07,234 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.3.txt
+2022-07-20 21:11:07,313 INFO [utils.py:416] [test-clean-lm_scale_0.3] %WER 2.70% [1421 / 52576, 200 ins, 105 del, 1116 sub ]
+2022-07-20 21:11:07,543 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.3.txt
+2022-07-20 21:11:07,572 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.4.txt
+2022-07-20 21:11:07,649 INFO [utils.py:416] [test-clean-lm_scale_0.4] %WER 2.67% [1403 / 52576, 195 ins, 109 del, 1099 sub ]
+2022-07-20 21:11:07,879 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.4.txt
+2022-07-20 21:11:07,907 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.5.txt
+2022-07-20 21:11:07,984 INFO [utils.py:416] [test-clean-lm_scale_0.5] %WER 2.66% [1400 / 52576, 188 ins, 112 del, 1100 sub ]
+2022-07-20 21:11:08,213 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.5.txt
+2022-07-20 21:11:08,241 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.6.txt
+2022-07-20 21:11:08,318 INFO [utils.py:416] [test-clean-lm_scale_0.6] %WER 2.68% [1407 / 52576, 181 ins, 119 del, 1107 sub ]
+2022-07-20 21:11:08,549 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.6.txt
+2022-07-20 21:11:08,577 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.7.txt
+2022-07-20 21:11:08,794 INFO [utils.py:416] [test-clean-lm_scale_0.7] %WER 2.68% [1409 / 52576, 173 ins, 130 del, 1106 sub ]
+2022-07-20 21:11:09,022 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.7.txt
+2022-07-20 21:11:09,050 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.8.txt
+2022-07-20 21:11:09,127 INFO [utils.py:416] [test-clean-lm_scale_0.8] %WER 2.70% [1422 / 52576, 169 ins, 142 del, 1111 sub ]
+2022-07-20 21:11:09,369 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.8.txt
+2022-07-20 21:11:09,400 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_0.9.txt
+2022-07-20 21:11:09,476 INFO [utils.py:416] [test-clean-lm_scale_0.9] %WER 2.74% [1443 / 52576, 156 ins, 167 del, 1120 sub ]
+2022-07-20 21:11:09,705 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_0.9.txt
+2022-07-20 21:11:09,735 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.0.txt
+2022-07-20 21:11:09,811 INFO [utils.py:416] [test-clean-lm_scale_1.0] %WER 2.84% [1491 / 52576, 156 ins, 203 del, 1132 sub ]
+2022-07-20 21:11:10,041 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.0.txt
+2022-07-20 21:11:10,070 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.1.txt
+2022-07-20 21:11:10,147 INFO [utils.py:416] [test-clean-lm_scale_1.1] %WER 2.94% [1545 / 52576, 151 ins, 246 del, 1148 sub ]
+2022-07-20 21:11:10,377 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.1.txt
+2022-07-20 21:11:10,407 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.2.txt
+2022-07-20 21:11:10,485 INFO [utils.py:416] [test-clean-lm_scale_1.2] %WER 3.09% [1623 / 52576, 146 ins, 300 del, 1177 sub ]
+2022-07-20 21:11:10,714 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.2.txt
+2022-07-20 21:11:10,743 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.3.txt
+2022-07-20 21:11:10,820 INFO [utils.py:416] [test-clean-lm_scale_1.3] %WER 3.30% [1734 / 52576, 149 ins, 376 del, 1209 sub ]
+2022-07-20 21:11:11,203 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.3.txt
+2022-07-20 21:11:11,232 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.4.txt
+2022-07-20 21:11:11,310 INFO [utils.py:416] [test-clean-lm_scale_1.4] %WER 3.51% [1846 / 52576, 148 ins, 462 del, 1236 sub ]
+2022-07-20 21:11:11,539 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.4.txt
+2022-07-20 21:11:11,567 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.5.txt
+2022-07-20 21:11:11,644 INFO [utils.py:416] [test-clean-lm_scale_1.5] %WER 3.79% [1992 / 52576, 144 ins, 567 del, 1281 sub ]
+2022-07-20 21:11:11,873 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.5.txt
+2022-07-20 21:11:11,901 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.6.txt
+2022-07-20 21:11:11,978 INFO [utils.py:416] [test-clean-lm_scale_1.6] %WER 4.07% [2140 / 52576, 131 ins, 698 del, 1311 sub ]
+2022-07-20 21:11:12,209 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.6.txt
+2022-07-20 21:11:12,237 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.7.txt
+2022-07-20 21:11:12,315 INFO [utils.py:416] [test-clean-lm_scale_1.7] %WER 4.38% [2303 / 52576, 131 ins, 826 del, 1346 sub ]
+2022-07-20 21:11:12,544 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.7.txt
+2022-07-20 21:11:12,572 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.8.txt
+2022-07-20 21:11:12,649 INFO [utils.py:416] [test-clean-lm_scale_1.8] %WER 4.69% [2465 / 52576, 131 ins, 955 del, 1379 sub ]
+2022-07-20 21:11:12,881 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.8.txt
+2022-07-20 21:11:12,910 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_1.9.txt
+2022-07-20 21:11:12,989 INFO [utils.py:416] [test-clean-lm_scale_1.9] %WER 4.99% [2625 / 52576, 133 ins, 1068 del, 1424 sub ]
+2022-07-20 21:11:13,366 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_1.9.txt
+2022-07-20 21:11:13,405 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-clean-lm_scale_2.0.txt
+2022-07-20 21:11:13,494 INFO [utils.py:416] [test-clean-lm_scale_2.0] %WER 5.31% [2794 / 52576, 130 ins, 1197 del, 1467 sub ]
+2022-07-20 21:11:13,727 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-clean-lm_scale_2.0.txt
+2022-07-20 21:11:13,728 INFO [decode.py:637]
+For test-clean, WER of different settings are:
+lm_scale_0.5	2.66	best for test-clean
+lm_scale_0.4	2.67
+lm_scale_0.6	2.68
+lm_scale_0.7	2.68
+lm_scale_0.3	2.7
+lm_scale_0.8	2.7
+lm_scale_0.2	2.74
+lm_scale_0.9	2.74
+lm_scale_0.1	2.82
+lm_scale_1.0	2.84
+lm_scale_1.1	2.94
+lm_scale_1.2	3.09
+lm_scale_1.3	3.3
+lm_scale_1.4	3.51
+lm_scale_1.5	3.79
+lm_scale_1.6	4.07
+lm_scale_1.7	4.38
+lm_scale_1.8	4.69
+lm_scale_1.9	4.99
+lm_scale_2.0	5.31
+2022-07-20 21:11:14,658 INFO [decode.py:588] batch 0/?, cuts processed until now is 2
+2022-07-20 21:11:54,814 INFO [decode.py:588] batch 100/?, cuts processed until now is 249
+2022-07-20 21:12:32,635 INFO [decode.py:588] batch 200/?, cuts processed until now is 540
+2022-07-20 21:13:09,226 INFO [decode.py:588] batch 300/?, cuts processed until now is 834
+2022-07-20 21:13:44,899 INFO [decode.py:588] batch 400/?, cuts processed until now is 1104
+2022-07-20 21:14:26,260 INFO [decode.py:588] batch 500/?, cuts processed until now is 1350
+2022-07-20 21:15:05,158 INFO [decode.py:588] batch 600/?, cuts processed until now is 1640
+2022-07-20 21:15:40,242 INFO [decode.py:588] batch 700/?, cuts processed until now is 1933
+2022-07-20 21:16:18,132 INFO [decode.py:588] batch 800/?, cuts processed until now is 2175
+2022-07-20 21:16:56,866 INFO [decode.py:588] batch 900/?, cuts processed until now is 2440
+2022-07-20 21:17:04,052 INFO [decode.py:783] Caught exception:
+CUDA out of memory. Tried to allocate 740.00 MiB (GPU 0; 31.75 GiB total capacity; 28.28 GiB already allocated; 445.50 MiB free; 29.93 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
+2022-07-20 21:17:04,052 INFO [decode.py:789] num_arcs before pruning: 349313
+2022-07-20 21:17:04,052 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
+2022-07-20 21:17:04,072 INFO [decode.py:803] num_arcs after pruning: 8796
+2022-07-20 21:17:41,206 INFO [decode.py:588] batch 1000/?, cuts processed until now is 2656
+2022-07-20 21:18:21,669 INFO [decode.py:588] batch 1100/?, cuts processed until now is 2839
+2022-07-20 21:18:59,437 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.1.txt
+2022-07-20 21:18:59,529 INFO [utils.py:416] [test-other-lm_scale_0.1] %WER 6.24% [3266 / 52343, 440 ins, 240 del, 2586 sub ]
+2022-07-20 21:18:59,774 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.1.txt
+2022-07-20 21:18:59,807 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.2.txt
+2022-07-20 21:18:59,890 INFO [utils.py:416] [test-other-lm_scale_0.2] %WER 6.04% [3161 / 52343, 414 ins, 245 del, 2502 sub ]
+2022-07-20 21:19:00,128 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.2.txt
+2022-07-20 21:19:00,161 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.3.txt
+2022-07-20 21:19:00,243 INFO [utils.py:416] [test-other-lm_scale_0.3] %WER 5.94% [3107 / 52343, 396 ins, 249 del, 2462 sub ]
+2022-07-20 21:19:00,480 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.3.txt
+2022-07-20 21:19:00,512 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.4.txt
+2022-07-20 21:19:00,594 INFO [utils.py:416] [test-other-lm_scale_0.4] %WER 5.85% [3060 / 52343, 379 ins, 268 del, 2413 sub ]
+2022-07-20 21:19:01,006 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.4.txt
+2022-07-20 21:19:01,035 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.5.txt
+2022-07-20 21:19:01,118 INFO [utils.py:416] [test-other-lm_scale_0.5] %WER 5.77% [3022 / 52343, 351 ins, 286 del, 2385 sub ]
+2022-07-20 21:19:01,356 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.5.txt
+2022-07-20 21:19:01,385 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.6.txt
+2022-07-20 21:19:01,467 INFO [utils.py:416] [test-other-lm_scale_0.6] %WER 5.76% [3017 / 52343, 332 ins, 320 del, 2365 sub ]
+2022-07-20 21:19:01,708 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.6.txt
+2022-07-20 21:19:01,736 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.7.txt
+2022-07-20 21:19:01,816 INFO [utils.py:416] [test-other-lm_scale_0.7] %WER 5.77% [3020 / 52343, 318 ins, 348 del, 2354 sub ]
+2022-07-20 21:19:02,052 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.7.txt
+2022-07-20 21:19:02,081 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.8.txt
+2022-07-20 21:19:02,162 INFO [utils.py:416] [test-other-lm_scale_0.8] %WER 5.78% [3028 / 52343, 294 ins, 398 del, 2336 sub ]
+2022-07-20 21:19:02,398 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.8.txt
+2022-07-20 21:19:02,427 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_0.9.txt
+2022-07-20 21:19:02,507 INFO [utils.py:416] [test-other-lm_scale_0.9] %WER 5.86% [3066 / 52343, 278 ins, 464 del, 2324 sub ]
+2022-07-20 21:19:02,743 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_0.9.txt
+2022-07-20 21:19:02,771 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.0.txt
+2022-07-20 21:19:02,852 INFO [utils.py:416] [test-other-lm_scale_1.0] %WER 6.02% [3149 / 52343, 268 ins, 540 del, 2341 sub ]
+2022-07-20 21:19:03,087 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.0.txt
+2022-07-20 21:19:03,115 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.1.txt
+2022-07-20 21:19:03,196 INFO [utils.py:416] [test-other-lm_scale_1.1] %WER 6.20% [3247 / 52343, 258 ins, 636 del, 2353 sub ]
+2022-07-20 21:19:03,431 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.1.txt
+2022-07-20 21:19:03,460 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.2.txt
+2022-07-20 21:19:03,686 INFO [utils.py:416] [test-other-lm_scale_1.2] %WER 6.40% [3351 / 52343, 248 ins, 749 del, 2354 sub ]
+2022-07-20 21:19:03,923 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.2.txt
+2022-07-20 21:19:03,951 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.3.txt
+2022-07-20 21:19:04,036 INFO [utils.py:416] [test-other-lm_scale_1.3] %WER 6.67% [3489 / 52343, 241 ins, 891 del, 2357 sub ]
+2022-07-20 21:19:04,280 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.3.txt
+2022-07-20 21:19:04,309 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.4.txt
+2022-07-20 21:19:04,392 INFO [utils.py:416] [test-other-lm_scale_1.4] %WER 7.08% [3705 / 52343, 234 ins, 1071 del, 2400 sub ]
+2022-07-20 21:19:04,628 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.4.txt
+2022-07-20 21:19:04,656 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.5.txt
+2022-07-20 21:19:04,737 INFO [utils.py:416] [test-other-lm_scale_1.5] %WER 7.42% [3886 / 52343, 224 ins, 1253 del, 2409 sub ]
+2022-07-20 21:19:04,973 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.5.txt
+2022-07-20 21:19:05,001 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.6.txt
+2022-07-20 21:19:05,081 INFO [utils.py:416] [test-other-lm_scale_1.6] %WER 7.84% [4102 / 52343, 214 ins, 1437 del, 2451 sub ]
+2022-07-20 21:19:05,317 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.6.txt
+2022-07-20 21:19:05,346 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.7.txt
+2022-07-20 21:19:05,426 INFO [utils.py:416] [test-other-lm_scale_1.7] %WER 8.29% [4338 / 52343, 213 ins, 1641 del, 2484 sub ]
+2022-07-20 21:19:05,667 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.7.txt
+2022-07-20 21:19:05,696 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.8.txt
+2022-07-20 21:19:05,777 INFO [utils.py:416] [test-other-lm_scale_1.8] %WER 8.78% [4596 / 52343, 209 ins, 1859 del, 2528 sub ]
+2022-07-20 21:19:06,031 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.8.txt
+2022-07-20 21:19:06,082 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_1.9.txt
+2022-07-20 21:19:06,310 INFO [utils.py:416] [test-other-lm_scale_1.9] %WER 9.27% [4854 / 52343, 205 ins, 2085 del, 2564 sub ]
+2022-07-20 21:19:06,549 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_1.9.txt
+2022-07-20 21:19:06,578 INFO [decode.py:609] The transcripts are stored in pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/recogs-test-other-lm_scale_2.0.txt
+2022-07-20 21:19:06,659 INFO [utils.py:416] [test-other-lm_scale_2.0] %WER 9.72% [5090 / 52343, 204 ins, 2272 del, 2614 sub ]
+2022-07-20 21:19:06,896 INFO [decode.py:621] Wrote detailed error stats to pruned_transducer_stateless_smooth/quandongtest_8gpu_ctc_attnetion2-exp/errs-test-other-lm_scale_2.0.txt
+2022-07-20 21:19:06,897 INFO [decode.py:637]
+For test-other, WER of different settings are:
+lm_scale_0.6	5.76	best for test-other
+lm_scale_0.5	5.77
+lm_scale_0.7	5.77
+lm_scale_0.8	5.78
+lm_scale_0.4	5.85
+lm_scale_0.9	5.86
+lm_scale_0.3	5.94
+lm_scale_1.0	6.02
+lm_scale_0.2	6.04
+lm_scale_1.1	6.2
+lm_scale_0.1	6.24
+lm_scale_1.2	6.4
+lm_scale_1.3	6.67
+lm_scale_1.4	7.08
+lm_scale_1.5	7.42
+lm_scale_1.6	7.84
+lm_scale_1.7	8.29
+lm_scale_1.8	8.78
+lm_scale_1.9	9.27
+lm_scale_2.0	9.72
+2022-07-20 21:19:06,897 INFO [decode.py:868] Done!