upload models and results

data/lang_bpe_500/bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5db9f109de8776c78b1a52bc69a9694acc09bb4c11373e61794e20b24f6e244d
+size 244891

large_bpe_500/decode_results/fast_beam_search/log-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam-2023-01-16-09-53-08

@@ -0,0 +1,107 @@
+2023-01-16 09:53:08,412 INFO [decode.py:565] Decoding started
+2023-01-16 09:53:08,412 INFO [decode.py:571] Device: cuda:0
+2023-01-16 09:53:08,589 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2023-01-16 09:53:08,631 INFO [decode.py:588] { 'activation': 'doubleswish',
+  'avg': 1,
+  'batch_idx_train': 0,
+  'beam': 20.0,
+  'beam_size': 4,
+  'best_train_epoch': -1,
+  'best_train_loss': inf,
+  'best_valid_epoch': -1,
+  'best_valid_loss': inf,
+  'blank_id': 0,
+  'bucketing_sampler': True,
+  'channels': 400,
+  'concatenate_cuts': False,
+  'context_size': 2,
+  'conv_layers': 18,
+  'decoder_dim': 400,
+  'decoding_method': 'fast_beam_search',
+  'drop_last': True,
+  'duration_factor': 1.0,
+  'enable_musan': True,
+  'enable_spec_aug': True,
+  'encoder_dim': 400,
+  'env_info': { 'IP address': '127.0.1.1',
+                'hostname': 'kao-dgxa-f12-u17',
+                'icefall-git-branch': 'tiny',
+                'icefall-git-date': 'Fri Jan 13 07:21:29 2023',
+                'icefall-git-sha1': '5c8e962-dirty',
+                'icefall-path': '/home/jsong/git/icefall',
+                'k2-build-type': 'Release',
+                'k2-git-date': 'Fri Nov 25 08:23:51 2022',
+                'k2-git-sha1': '1feafa064cf3b6c243e6b33b0192601224210937',
+                'k2-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/k2/__init__.py',
+                'k2-version': '1.23.2',
+                'k2-with-cuda': True,
+                'lhotse-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/lhotse/__init__.py',
+                'lhotse-version': '1.7.0',
+                'python-version': '3.9',
+                'torch-cuda-available': True,
+                'torch-cuda-version': '11.3',
+                'torch-version': '1.12.0'},
+  'epoch': 30,
+  'exp_dir': PosixPath('tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug'),
+  'feature_dim': 80,
+  'full_libri': True,
+  'gap': 1.0,
+  'input_strategy': 'PrecomputedFeatures',
+  'iter': 0,
+  'joiner_dim': 400,
+  'lang_dir': PosixPath('data/lang_bpe_500'),
+  'log_interval': 500,
+  'manifest_dir': PosixPath('data/fbank'),
+  'max_contexts': 8,
+  'max_duration': 600,
+  'max_states': 64,
+  'max_sym_per_frame': 1,
+  'nbest_scale': 0.5,
+  'ngram_lm_scale': 0.1,
+  'num_buckets': 30,
+  'num_paths': 100,
+  'num_workers': 2,
+  'on_the_fly_feats': False,
+  'res_dir': PosixPath('tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/fast_beam_search'),
+  'reset_interval': 200,
+  'return_cuts': True,
+  'shuffle': True,
+  'skip_add': True,
+  'spec_aug_time_warp_factor': 80,
+  'subsampling_factor': 4,
+  'suffix': 'epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam',
+  'unk_id': 2,
+  'use_averaged_model': True,
+  'use_double_scores': True,
+  'use_dscnn': True,
+  'valid_interval': 9000,
+  'vocab_size': 500,
+  'warm_step': 5000}
+2023-01-16 09:53:08,631 INFO [decode.py:590] About to create model
+2023-01-16 09:53:11,209 INFO [train.py:426] Encoder MAC ops for 10 seconds of audio is 888.44M
+2023-01-16 09:53:11,213 INFO [decode.py:659] Calculating the averaged model over epoch range from 29 (excluded) to 30
+2023-01-16 09:53:11,613 INFO [decode.py:697] Number of model parameters: 4821330
+2023-01-16 09:53:11,614 INFO [decode.py:698] Parameters for transducer decoding: 4219830
+2023-01-16 09:53:11,614 INFO [asr_datamodule.py:449] About to get test-clean cuts
+2023-01-16 09:53:11,617 INFO [asr_datamodule.py:456] About to get test-other cuts
+2023-01-16 09:53:13,953 INFO [decode.py:459] batch 0/?, cuts processed until now is 43
+2023-01-16 09:53:32,617 INFO [decode.py:459] batch 20/?, cuts processed until now is 1430
+2023-01-16 09:53:53,226 INFO [decode.py:459] batch 40/?, cuts processed until now is 2561
+2023-01-16 09:53:53,817 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/fast_beam_search/recogs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-16 09:53:53,883 INFO [utils.py:536] [test-clean-beam_20.0_max_contexts_8_max_states_64] %WER 7.91% [4160 / 52576, 482 ins, 411 del, 3267 sub ]
+2023-01-16 09:53:54,026 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/fast_beam_search/errs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-16 09:53:54,026 INFO [decode.py:508] 
+For test-clean, WER of different settings are:
+beam_20.0_max_contexts_8_max_states_64	7.91	best for test-clean
+
+2023-01-16 09:53:55,228 INFO [decode.py:459] batch 0/?, cuts processed until now is 52
+2023-01-16 09:54:12,536 INFO [decode.py:459] batch 20/?, cuts processed until now is 1646
+2023-01-16 09:54:31,673 INFO [decode.py:459] batch 40/?, cuts processed until now is 2870
+2023-01-16 09:54:32,255 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/fast_beam_search/recogs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-16 09:54:32,325 INFO [utils.py:536] [test-other-beam_20.0_max_contexts_8_max_states_64] %WER 20.10% [10521 / 52343, 1011 ins, 1417 del, 8093 sub ]
+2023-01-16 09:54:32,481 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/fast_beam_search/errs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-16 09:54:32,481 INFO [decode.py:508] 
+For test-other, WER of different settings are:
+beam_20.0_max_contexts_8_max_states_64	20.1	best for test-other
+
+2023-01-16 09:54:32,481 INFO [decode.py:730] Done!

large_bpe_500/decode_results/fast_beam_search/wer-summary-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_20.0_max_contexts_8_max_states_64	7.91

large_bpe_500/decode_results/fast_beam_search/wer-summary-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-1-beam-20.0-max-contexts-8-max-states-64-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_20.0_max_contexts_8_max_states_64	20.1

large_bpe_500/decode_results/greedy_search/log-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam-2023-01-16-09-52-16

@@ -0,0 +1,103 @@
+2023-01-16 09:52:16,827 INFO [decode.py:565] Decoding started
+2023-01-16 09:52:16,827 INFO [decode.py:571] Device: cuda:0
+2023-01-16 09:52:17,018 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2023-01-16 09:52:17,057 INFO [decode.py:588] { 'activation': 'doubleswish',
+  'avg': 1,
+  'batch_idx_train': 0,
+  'beam': 20.0,
+  'beam_size': 4,
+  'best_train_epoch': -1,
+  'best_train_loss': inf,
+  'best_valid_epoch': -1,
+  'best_valid_loss': inf,
+  'blank_id': 0,
+  'bucketing_sampler': True,
+  'channels': 400,
+  'concatenate_cuts': False,
+  'context_size': 2,
+  'conv_layers': 18,
+  'decoder_dim': 400,
+  'decoding_method': 'greedy_search',
+  'drop_last': True,
+  'duration_factor': 1.0,
+  'enable_musan': True,
+  'enable_spec_aug': True,
+  'encoder_dim': 400,
+  'env_info': { 'IP address': '127.0.1.1',
+                'hostname': 'kao-dgxa-f12-u17',
+                'icefall-git-branch': 'tiny',
+                'icefall-git-date': 'Fri Jan 13 07:21:29 2023',
+                'icefall-git-sha1': '5c8e962-dirty',
+                'icefall-path': '/home/jsong/git/icefall',
+                'k2-build-type': 'Release',
+                'k2-git-date': 'Fri Nov 25 08:23:51 2022',
+                'k2-git-sha1': '1feafa064cf3b6c243e6b33b0192601224210937',
+                'k2-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/k2/__init__.py',
+                'k2-version': '1.23.2',
+                'k2-with-cuda': True,
+                'lhotse-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/lhotse/__init__.py',
+                'lhotse-version': '1.7.0',
+                'python-version': '3.9',
+                'torch-cuda-available': True,
+                'torch-cuda-version': '11.3',
+                'torch-version': '1.12.0'},
+  'epoch': 30,
+  'exp_dir': PosixPath('tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug'),
+  'feature_dim': 80,
+  'full_libri': True,
+  'gap': 1.0,
+  'input_strategy': 'PrecomputedFeatures',
+  'iter': 0,
+  'joiner_dim': 400,
+  'lang_dir': PosixPath('data/lang_bpe_500'),
+  'log_interval': 500,
+  'manifest_dir': PosixPath('data/fbank'),
+  'max_contexts': 8,
+  'max_duration': 600,
+  'max_states': 64,
+  'max_sym_per_frame': 1,
+  'nbest_scale': 0.5,
+  'ngram_lm_scale': 0.1,
+  'num_buckets': 30,
+  'num_paths': 100,
+  'num_workers': 2,
+  'on_the_fly_feats': False,
+  'res_dir': PosixPath('tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/greedy_search'),
+  'reset_interval': 200,
+  'return_cuts': True,
+  'shuffle': True,
+  'skip_add': True,
+  'spec_aug_time_warp_factor': 80,
+  'subsampling_factor': 4,
+  'suffix': 'epoch-30-avg-1-context-2-max-sym-per-frame-1-uam',
+  'unk_id': 2,
+  'use_averaged_model': True,
+  'use_double_scores': True,
+  'use_dscnn': True,
+  'valid_interval': 9000,
+  'vocab_size': 500,
+  'warm_step': 5000}
+2023-01-16 09:52:17,058 INFO [decode.py:590] About to create model
+2023-01-16 09:52:19,589 INFO [train.py:426] Encoder MAC ops for 10 seconds of audio is 888.44M
+2023-01-16 09:52:19,594 INFO [decode.py:659] Calculating the averaged model over epoch range from 29 (excluded) to 30
+2023-01-16 09:52:19,977 INFO [decode.py:697] Number of model parameters: 4821330
+2023-01-16 09:52:19,977 INFO [decode.py:698] Parameters for transducer decoding: 4219830
+2023-01-16 09:52:19,977 INFO [asr_datamodule.py:449] About to get test-clean cuts
+2023-01-16 09:52:19,978 INFO [asr_datamodule.py:456] About to get test-other cuts
+2023-01-16 09:52:21,770 INFO [decode.py:459] batch 0/?, cuts processed until now is 43
+2023-01-16 09:52:39,401 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/greedy_search/recogs-test-clean-greedy_search-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam.txt
+2023-01-16 09:52:39,467 INFO [utils.py:536] [test-clean-greedy_search] %WER 8.29% [4359 / 52576, 501 ins, 443 del, 3415 sub ]
+2023-01-16 09:52:39,617 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/greedy_search/errs-test-clean-greedy_search-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam.txt
+2023-01-16 09:52:39,618 INFO [decode.py:508] 
+For test-clean, WER of different settings are:
+greedy_search	8.29	best for test-clean
+
+2023-01-16 09:52:40,281 INFO [decode.py:459] batch 0/?, cuts processed until now is 52
+2023-01-16 09:52:56,315 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/greedy_search/recogs-test-other-greedy_search-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam.txt
+2023-01-16 09:52:56,386 INFO [utils.py:536] [test-other-greedy_search] %WER 21.11% [11052 / 52343, 1006 ins, 1534 del, 8512 sub ]
+2023-01-16 09:52:56,547 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/greedy_search/errs-test-other-greedy_search-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam.txt
+2023-01-16 09:52:56,547 INFO [decode.py:508] 
+For test-other, WER of different settings are:
+greedy_search	21.11	best for test-other
+
+2023-01-16 09:52:56,547 INFO [decode.py:730] Done!

large_bpe_500/decode_results/greedy_search/wer-summary-test-clean-greedy_search-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+greedy_search	8.29

large_bpe_500/decode_results/greedy_search/wer-summary-test-other-greedy_search-epoch-30-avg-1-context-2-max-sym-per-frame-1-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+greedy_search	21.11

large_bpe_500/decode_results/modified_beam_search/log-epoch-30-avg-1-modified_beam_search-beam-size-4-uam-2023-01-16-09-54-54

@@ -0,0 +1,107 @@
+2023-01-16 09:54:54,538 INFO [decode.py:565] Decoding started
+2023-01-16 09:54:54,538 INFO [decode.py:571] Device: cuda:0
+2023-01-16 09:54:54,718 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2023-01-16 09:54:54,758 INFO [decode.py:588] { 'activation': 'doubleswish',
+  'avg': 1,
+  'batch_idx_train': 0,
+  'beam': 20.0,
+  'beam_size': 4,
+  'best_train_epoch': -1,
+  'best_train_loss': inf,
+  'best_valid_epoch': -1,
+  'best_valid_loss': inf,
+  'blank_id': 0,
+  'bucketing_sampler': True,
+  'channels': 400,
+  'concatenate_cuts': False,
+  'context_size': 2,
+  'conv_layers': 18,
+  'decoder_dim': 400,
+  'decoding_method': 'modified_beam_search',
+  'drop_last': True,
+  'duration_factor': 1.0,
+  'enable_musan': True,
+  'enable_spec_aug': True,
+  'encoder_dim': 400,
+  'env_info': { 'IP address': '127.0.1.1',
+                'hostname': 'kao-dgxa-f12-u17',
+                'icefall-git-branch': 'tiny',
+                'icefall-git-date': 'Fri Jan 13 07:21:29 2023',
+                'icefall-git-sha1': '5c8e962-dirty',
+                'icefall-path': '/home/jsong/git/icefall',
+                'k2-build-type': 'Release',
+                'k2-git-date': 'Fri Nov 25 08:23:51 2022',
+                'k2-git-sha1': '1feafa064cf3b6c243e6b33b0192601224210937',
+                'k2-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/k2/__init__.py',
+                'k2-version': '1.23.2',
+                'k2-with-cuda': True,
+                'lhotse-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/lhotse/__init__.py',
+                'lhotse-version': '1.7.0',
+                'python-version': '3.9',
+                'torch-cuda-available': True,
+                'torch-cuda-version': '11.3',
+                'torch-version': '1.12.0'},
+  'epoch': 30,
+  'exp_dir': PosixPath('tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug'),
+  'feature_dim': 80,
+  'full_libri': True,
+  'gap': 1.0,
+  'input_strategy': 'PrecomputedFeatures',
+  'iter': 0,
+  'joiner_dim': 400,
+  'lang_dir': PosixPath('data/lang_bpe_500'),
+  'log_interval': 500,
+  'manifest_dir': PosixPath('data/fbank'),
+  'max_contexts': 8,
+  'max_duration': 600,
+  'max_states': 64,
+  'max_sym_per_frame': 1,
+  'nbest_scale': 0.5,
+  'ngram_lm_scale': 0.1,
+  'num_buckets': 30,
+  'num_paths': 100,
+  'num_workers': 2,
+  'on_the_fly_feats': False,
+  'res_dir': PosixPath('tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/modified_beam_search'),
+  'reset_interval': 200,
+  'return_cuts': True,
+  'shuffle': True,
+  'skip_add': True,
+  'spec_aug_time_warp_factor': 80,
+  'subsampling_factor': 4,
+  'suffix': 'epoch-30-avg-1-modified_beam_search-beam-size-4-uam',
+  'unk_id': 2,
+  'use_averaged_model': True,
+  'use_double_scores': True,
+  'use_dscnn': True,
+  'valid_interval': 9000,
+  'vocab_size': 500,
+  'warm_step': 5000}
+2023-01-16 09:54:54,758 INFO [decode.py:590] About to create model
+2023-01-16 09:54:57,321 INFO [train.py:426] Encoder MAC ops for 10 seconds of audio is 888.44M
+2023-01-16 09:54:57,325 INFO [decode.py:659] Calculating the averaged model over epoch range from 29 (excluded) to 30
+2023-01-16 09:54:57,706 INFO [decode.py:697] Number of model parameters: 4821330
+2023-01-16 09:54:57,706 INFO [decode.py:698] Parameters for transducer decoding: 4219830
+2023-01-16 09:54:57,706 INFO [asr_datamodule.py:449] About to get test-clean cuts
+2023-01-16 09:54:57,707 INFO [asr_datamodule.py:456] About to get test-other cuts
+2023-01-16 09:55:01,888 INFO [decode.py:459] batch 0/?, cuts processed until now is 43
+2023-01-16 09:55:55,346 INFO [decode.py:459] batch 20/?, cuts processed until now is 1430
+2023-01-16 09:56:40,761 INFO [decode.py:459] batch 40/?, cuts processed until now is 2561
+2023-01-16 09:56:42,416 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/modified_beam_search/recogs-test-clean-beam_size_4-epoch-30-avg-1-modified_beam_search-beam-size-4-uam.txt
+2023-01-16 09:56:42,520 INFO [utils.py:536] [test-clean-beam_size_4] %WER 7.74% [4072 / 52576, 492 ins, 384 del, 3196 sub ]
+2023-01-16 09:56:42,664 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/modified_beam_search/errs-test-clean-beam_size_4-epoch-30-avg-1-modified_beam_search-beam-size-4-uam.txt
+2023-01-16 09:56:42,664 INFO [decode.py:508] 
+For test-clean, WER of different settings are:
+beam_size_4	7.74	best for test-clean
+
+2023-01-16 09:56:45,681 INFO [decode.py:459] batch 0/?, cuts processed until now is 52
+2023-01-16 09:57:37,910 INFO [decode.py:459] batch 20/?, cuts processed until now is 1646
+2023-01-16 09:58:20,895 INFO [decode.py:459] batch 40/?, cuts processed until now is 2870
+2023-01-16 09:58:22,601 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/modified_beam_search/recogs-test-other-beam_size_4-epoch-30-avg-1-modified_beam_search-beam-size-4-uam.txt
+2023-01-16 09:58:22,672 INFO [utils.py:536] [test-other-beam_size_4] %WER 19.89% [10412 / 52343, 1054 ins, 1301 del, 8057 sub ]
+2023-01-16 09:58:22,832 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_4m_bpe500_halfdelay_specaug/modified_beam_search/errs-test-other-beam_size_4-epoch-30-avg-1-modified_beam_search-beam-size-4-uam.txt
+2023-01-16 09:58:22,832 INFO [decode.py:508] 
+For test-other, WER of different settings are:
+beam_size_4	19.89	best for test-other
+
+2023-01-16 09:58:22,832 INFO [decode.py:730] Done!

large_bpe_500/decode_results/modified_beam_search/wer-summary-test-clean-beam_size_4-epoch-30-avg-1-modified_beam_search-beam-size-4-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_size_4	7.74

large_bpe_500/decode_results/modified_beam_search/wer-summary-test-other-beam_size_4-epoch-30-avg-1-modified_beam_search-beam-size-4-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_size_4	19.89

large_bpe_500/exp/cpu_jit.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20897e2653f7130fc2c1a52621cf01bb990010659c886124c961620c58d99d5f
+size 19576790

large_bpe_500/exp/pretrained.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be438679835e07dc63718fe490fd838cc532d04f0e51248314f116676bd10851
+size 19539685

large_bpe_500/exp/tensorboard/events.out.tfevents.1673245407.kao-dgxa-f12-u17.4075883.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7694fbf23c8fd399fd58b484a3837fa8f38ac8db1a098dd40e2512677a338e04
+size 618506

middle_bpe_500/decode_results/fast_beam_search/log-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam-2023-01-09-01-15-28

@@ -0,0 +1,107 @@
+2023-01-09 01:15:28,211 INFO [decode.py:565] Decoding started
+2023-01-09 01:15:28,211 INFO [decode.py:571] Device: cuda:0
+2023-01-09 01:15:28,390 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2023-01-09 01:15:28,425 INFO [decode.py:588] { 'activation': 'doubleswish',
+  'avg': 2,
+  'batch_idx_train': 0,
+  'beam': 20.0,
+  'beam_size': 4,
+  'best_train_epoch': -1,
+  'best_train_loss': inf,
+  'best_valid_epoch': -1,
+  'best_valid_loss': inf,
+  'blank_id': 0,
+  'bucketing_sampler': True,
+  'channels': 300,
+  'concatenate_cuts': False,
+  'context_size': 2,
+  'conv_layers': 18,
+  'decoder_dim': 256,
+  'decoding_method': 'fast_beam_search',
+  'drop_last': True,
+  'duration_factor': 1.0,
+  'enable_musan': True,
+  'enable_spec_aug': True,
+  'encoder_dim': 256,
+  'env_info': { 'IP address': '127.0.1.1',
+                'hostname': 'kao-dgxa-f12-u17',
+                'icefall-git-branch': 'tiny',
+                'icefall-git-date': 'Mon Jan 2 00:08:32 2023',
+                'icefall-git-sha1': '2fd970b-dirty',
+                'icefall-path': '/home/jsong/git/icefall',
+                'k2-build-type': 'Release',
+                'k2-git-date': 'Fri Nov 25 08:23:51 2022',
+                'k2-git-sha1': '1feafa064cf3b6c243e6b33b0192601224210937',
+                'k2-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/k2/__init__.py',
+                'k2-version': '1.23.2',
+                'k2-with-cuda': True,
+                'lhotse-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/lhotse/__init__.py',
+                'lhotse-version': '1.7.0',
+                'python-version': '3.9',
+                'torch-cuda-available': True,
+                'torch-cuda-version': '11.3',
+                'torch-version': '1.12.0'},
+  'epoch': 30,
+  'exp_dir': PosixPath('tiny_transducer_ctc/exp_2m_bpe500_halfdelay'),
+  'feature_dim': 80,
+  'full_libri': True,
+  'gap': 1.0,
+  'input_strategy': 'PrecomputedFeatures',
+  'iter': 0,
+  'joiner_dim': 256,
+  'lang_dir': PosixPath('data/lang_bpe_500'),
+  'log_interval': 500,
+  'manifest_dir': PosixPath('data/fbank'),
+  'max_contexts': 8,
+  'max_duration': 600,
+  'max_states': 64,
+  'max_sym_per_frame': 1,
+  'nbest_scale': 0.5,
+  'ngram_lm_scale': 0.1,
+  'num_buckets': 30,
+  'num_paths': 100,
+  'num_workers': 2,
+  'on_the_fly_feats': False,
+  'res_dir': PosixPath('tiny_transducer_ctc/exp_2m_bpe500_halfdelay/fast_beam_search'),
+  'reset_interval': 200,
+  'return_cuts': True,
+  'shuffle': True,
+  'skip_add': True,
+  'spec_aug_time_warp_factor': 80,
+  'subsampling_factor': 4,
+  'suffix': 'epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam',
+  'unk_id': 2,
+  'use_averaged_model': True,
+  'use_double_scores': True,
+  'use_dscnn': True,
+  'valid_interval': 9000,
+  'vocab_size': 500,
+  'warm_step': 5000}
+2023-01-09 01:15:28,425 INFO [decode.py:590] About to create model
+2023-01-09 01:15:30,875 INFO [train.py:425] Encoder MAC ops for 10 seconds of audio is 501.07M
+2023-01-09 01:15:30,878 INFO [decode.py:659] Calculating the averaged model over epoch range from 28 (excluded) to 30
+2023-01-09 01:15:31,196 INFO [decode.py:697] Number of model parameters: 2735794
+2023-01-09 01:15:31,196 INFO [decode.py:698] Parameters for transducer decoding: 2350294
+2023-01-09 01:15:31,196 INFO [asr_datamodule.py:443] About to get test-clean cuts
+2023-01-09 01:15:31,197 INFO [asr_datamodule.py:450] About to get test-other cuts
+2023-01-09 01:15:33,792 INFO [decode.py:459] batch 0/?, cuts processed until now is 43
+2023-01-09 01:15:52,777 INFO [decode.py:459] batch 20/?, cuts processed until now is 1430
+2023-01-09 01:16:13,995 INFO [decode.py:459] batch 40/?, cuts processed until now is 2561
+2023-01-09 01:16:14,590 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_2m_bpe500_halfdelay/fast_beam_search/recogs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-09 01:16:14,663 INFO [utils.py:536] [test-clean-beam_20.0_max_contexts_8_max_states_64] %WER 9.69% [5096 / 52576, 599 ins, 487 del, 4010 sub ]
+2023-01-09 01:16:14,812 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_2m_bpe500_halfdelay/fast_beam_search/errs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-09 01:16:14,812 INFO [decode.py:508] 
+For test-clean, WER of different settings are:
+beam_20.0_max_contexts_8_max_states_64	9.69	best for test-clean
+
+2023-01-09 01:16:16,043 INFO [decode.py:459] batch 0/?, cuts processed until now is 52
+2023-01-09 01:16:33,449 INFO [decode.py:459] batch 20/?, cuts processed until now is 1646
+2023-01-09 01:16:53,170 INFO [decode.py:459] batch 40/?, cuts processed until now is 2870
+2023-01-09 01:16:53,753 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_2m_bpe500_halfdelay/fast_beam_search/recogs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-09 01:16:53,825 INFO [utils.py:536] [test-other-beam_20.0_max_contexts_8_max_states_64] %WER 23.58% [12345 / 52343, 1208 ins, 1744 del, 9393 sub ]
+2023-01-09 01:16:53,988 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_2m_bpe500_halfdelay/fast_beam_search/errs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam.txt
+2023-01-09 01:16:53,988 INFO [decode.py:508] 
+For test-other, WER of different settings are:
+beam_20.0_max_contexts_8_max_states_64	23.58	best for test-other
+
+2023-01-09 01:16:53,988 INFO [decode.py:730] Done!

middle_bpe_500/decode_results/fast_beam_search/wer-summary-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_20.0_max_contexts_8_max_states_64	9.69

middle_bpe_500/decode_results/fast_beam_search/wer-summary-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-30-avg-2-beam-20.0-max-contexts-8-max-states-64-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_20.0_max_contexts_8_max_states_64	23.58

middle_bpe_500/decode_results/greedy_search/log-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam-2023-01-09-01-12-47

@@ -0,0 +1,103 @@
+2023-01-09 01:12:47,525 INFO [decode.py:565] Decoding started
+2023-01-09 01:12:47,525 INFO [decode.py:571] Device: cuda:0
+2023-01-09 01:12:47,708 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2023-01-09 01:12:47,749 INFO [decode.py:588] { 'activation': 'doubleswish',
+  'avg': 2,
+  'batch_idx_train': 0,
+  'beam': 20.0,
+  'beam_size': 4,
+  'best_train_epoch': -1,
+  'best_train_loss': inf,
+  'best_valid_epoch': -1,
+  'best_valid_loss': inf,
+  'blank_id': 0,
+  'bucketing_sampler': True,
+  'channels': 300,
+  'concatenate_cuts': False,
+  'context_size': 2,
+  'conv_layers': 18,
+  'decoder_dim': 256,
+  'decoding_method': 'greedy_search',
+  'drop_last': True,
+  'duration_factor': 1.0,
+  'enable_musan': True,
+  'enable_spec_aug': True,
+  'encoder_dim': 256,
+  'env_info': { 'IP address': '127.0.1.1',
+                'hostname': 'kao-dgxa-f12-u17',
+                'icefall-git-branch': 'tiny',
+                'icefall-git-date': 'Mon Jan 2 00:08:32 2023',
+                'icefall-git-sha1': '2fd970b-dirty',
+                'icefall-path': '/home/jsong/git/icefall',
+                'k2-build-type': 'Release',
+                'k2-git-date': 'Fri Nov 25 08:23:51 2022',
+                'k2-git-sha1': '1feafa064cf3b6c243e6b33b0192601224210937',
+                'k2-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/k2/__init__.py',
+                'k2-version': '1.23.2',
+                'k2-with-cuda': True,
+                'lhotse-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/lhotse/__init__.py',
+                'lhotse-version': '1.7.0',
+                'python-version': '3.9',
+                'torch-cuda-available': True,
+                'torch-cuda-version': '11.3',
+                'torch-version': '1.12.0'},
+  'epoch': 30,
+  'exp_dir': PosixPath('tiny_transducer_ctc/exp_2m_bpe500_halfdelay'),
+  'feature_dim': 80,
+  'full_libri': True,
+  'gap': 1.0,
+  'input_strategy': 'PrecomputedFeatures',
+  'iter': 0,
+  'joiner_dim': 256,
+  'lang_dir': PosixPath('data/lang_bpe_500'),
+  'log_interval': 500,
+  'manifest_dir': PosixPath('data/fbank'),
+  'max_contexts': 8,
+  'max_duration': 600,
+  'max_states': 64,
+  'max_sym_per_frame': 1,
+  'nbest_scale': 0.5,
+  'ngram_lm_scale': 0.1,
+  'num_buckets': 30,
+  'num_paths': 100,
+  'num_workers': 2,
+  'on_the_fly_feats': False,
+  'res_dir': PosixPath('tiny_transducer_ctc/exp_2m_bpe500_halfdelay/greedy_search'),
+  'reset_interval': 200,
+  'return_cuts': True,
+  'shuffle': True,
+  'skip_add': True,
+  'spec_aug_time_warp_factor': 80,
+  'subsampling_factor': 4,
+  'suffix': 'epoch-30-avg-2-context-2-max-sym-per-frame-1-uam',
+  'unk_id': 2,
+  'use_averaged_model': True,
+  'use_double_scores': True,
+  'use_dscnn': True,
+  'valid_interval': 9000,
+  'vocab_size': 500,
+  'warm_step': 5000}
+2023-01-09 01:12:47,750 INFO [decode.py:590] About to create model
+2023-01-09 01:12:50,232 INFO [train.py:425] Encoder MAC ops for 10 seconds of audio is 501.07M
+2023-01-09 01:12:50,235 INFO [decode.py:659] Calculating the averaged model over epoch range from 28 (excluded) to 30
+2023-01-09 01:12:50,572 INFO [decode.py:697] Number of model parameters: 2735794
+2023-01-09 01:12:50,572 INFO [decode.py:698] Parameters for transducer decoding: 2350294
+2023-01-09 01:12:50,572 INFO [asr_datamodule.py:443] About to get test-clean cuts
+2023-01-09 01:12:50,574 INFO [asr_datamodule.py:450] About to get test-other cuts
+2023-01-09 01:12:52,369 INFO [decode.py:459] batch 0/?, cuts processed until now is 43
+2023-01-09 01:13:09,470 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_2m_bpe500_halfdelay/greedy_search/recogs-test-clean-greedy_search-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam.txt
+2023-01-09 01:13:09,537 INFO [utils.py:536] [test-clean-greedy_search] %WER 10.26% [5394 / 52576, 598 ins, 569 del, 4227 sub ]
+2023-01-09 01:13:09,683 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_2m_bpe500_halfdelay/greedy_search/errs-test-clean-greedy_search-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam.txt
+2023-01-09 01:13:09,683 INFO [decode.py:508] 
+For test-clean, WER of different settings are:
+greedy_search	10.26	best for test-clean
+
+2023-01-09 01:13:10,353 INFO [decode.py:459] batch 0/?, cuts processed until now is 52
+2023-01-09 01:13:25,888 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_2m_bpe500_halfdelay/greedy_search/recogs-test-other-greedy_search-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam.txt
+2023-01-09 01:13:25,960 INFO [utils.py:536] [test-other-greedy_search] %WER 25.13% [13156 / 52343, 1217 ins, 1939 del, 10000 sub ]
+2023-01-09 01:13:26,121 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_2m_bpe500_halfdelay/greedy_search/errs-test-other-greedy_search-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam.txt
+2023-01-09 01:13:26,121 INFO [decode.py:508] 
+For test-other, WER of different settings are:
+greedy_search	25.13	best for test-other
+
+2023-01-09 01:13:26,121 INFO [decode.py:730] Done!

middle_bpe_500/decode_results/greedy_search/wer-summary-test-clean-greedy_search-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+greedy_search	10.26

middle_bpe_500/decode_results/greedy_search/wer-summary-test-other-greedy_search-epoch-30-avg-2-context-2-max-sym-per-frame-1-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+greedy_search	25.13

middle_bpe_500/decode_results/modified_beam_search/log-epoch-30-avg-2-modified_beam_search-beam-size-4-uam-2023-01-09-01-17-06

@@ -0,0 +1,107 @@
+2023-01-09 01:17:06,332 INFO [decode.py:565] Decoding started
+2023-01-09 01:17:06,332 INFO [decode.py:571] Device: cuda:0
+2023-01-09 01:17:06,514 INFO [lexicon.py:168] Loading pre-compiled data/lang_bpe_500/Linv.pt
+2023-01-09 01:17:06,554 INFO [decode.py:588] { 'activation': 'doubleswish',
+  'avg': 2,
+  'batch_idx_train': 0,
+  'beam': 20.0,
+  'beam_size': 4,
+  'best_train_epoch': -1,
+  'best_train_loss': inf,
+  'best_valid_epoch': -1,
+  'best_valid_loss': inf,
+  'blank_id': 0,
+  'bucketing_sampler': True,
+  'channels': 300,
+  'concatenate_cuts': False,
+  'context_size': 2,
+  'conv_layers': 18,
+  'decoder_dim': 256,
+  'decoding_method': 'modified_beam_search',
+  'drop_last': True,
+  'duration_factor': 1.0,
+  'enable_musan': True,
+  'enable_spec_aug': True,
+  'encoder_dim': 256,
+  'env_info': { 'IP address': '127.0.1.1',
+                'hostname': 'kao-dgxa-f12-u17',
+                'icefall-git-branch': 'tiny',
+                'icefall-git-date': 'Mon Jan 2 00:08:32 2023',
+                'icefall-git-sha1': '2fd970b-dirty',
+                'icefall-path': '/home/jsong/git/icefall',
+                'k2-build-type': 'Release',
+                'k2-git-date': 'Fri Nov 25 08:23:51 2022',
+                'k2-git-sha1': '1feafa064cf3b6c243e6b33b0192601224210937',
+                'k2-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/k2/__init__.py',
+                'k2-version': '1.23.2',
+                'k2-with-cuda': True,
+                'lhotse-path': '/home/jsong/miniconda3/envs/k2/lib/python3.9/site-packages/lhotse/__init__.py',
+                'lhotse-version': '1.7.0',
+                'python-version': '3.9',
+                'torch-cuda-available': True,
+                'torch-cuda-version': '11.3',
+                'torch-version': '1.12.0'},
+  'epoch': 30,
+  'exp_dir': PosixPath('tiny_transducer_ctc/exp_2m_bpe500_halfdelay'),
+  'feature_dim': 80,
+  'full_libri': True,
+  'gap': 1.0,
+  'input_strategy': 'PrecomputedFeatures',
+  'iter': 0,
+  'joiner_dim': 256,
+  'lang_dir': PosixPath('data/lang_bpe_500'),
+  'log_interval': 500,
+  'manifest_dir': PosixPath('data/fbank'),
+  'max_contexts': 8,
+  'max_duration': 600,
+  'max_states': 64,
+  'max_sym_per_frame': 1,
+  'nbest_scale': 0.5,
+  'ngram_lm_scale': 0.1,
+  'num_buckets': 30,
+  'num_paths': 100,
+  'num_workers': 2,
+  'on_the_fly_feats': False,
+  'res_dir': PosixPath('tiny_transducer_ctc/exp_2m_bpe500_halfdelay/modified_beam_search'),
+  'reset_interval': 200,
+  'return_cuts': True,
+  'shuffle': True,
+  'skip_add': True,
+  'spec_aug_time_warp_factor': 80,
+  'subsampling_factor': 4,
+  'suffix': 'epoch-30-avg-2-modified_beam_search-beam-size-4-uam',
+  'unk_id': 2,
+  'use_averaged_model': True,
+  'use_double_scores': True,
+  'use_dscnn': True,
+  'valid_interval': 9000,
+  'vocab_size': 500,
+  'warm_step': 5000}
+2023-01-09 01:17:06,555 INFO [decode.py:590] About to create model
+2023-01-09 01:17:08,981 INFO [train.py:425] Encoder MAC ops for 10 seconds of audio is 501.07M
+2023-01-09 01:17:08,984 INFO [decode.py:659] Calculating the averaged model over epoch range from 28 (excluded) to 30
+2023-01-09 01:17:09,308 INFO [decode.py:697] Number of model parameters: 2735794
+2023-01-09 01:17:09,308 INFO [decode.py:698] Parameters for transducer decoding: 2350294
+2023-01-09 01:17:09,308 INFO [asr_datamodule.py:443] About to get test-clean cuts
+2023-01-09 01:17:09,309 INFO [asr_datamodule.py:450] About to get test-other cuts
+2023-01-09 01:17:13,410 INFO [decode.py:459] batch 0/?, cuts processed until now is 43
+2023-01-09 01:18:06,305 INFO [decode.py:459] batch 20/?, cuts processed until now is 1430
+2023-01-09 01:18:50,859 INFO [decode.py:459] batch 40/?, cuts processed until now is 2561
+2023-01-09 01:18:52,480 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_2m_bpe500_halfdelay/modified_beam_search/recogs-test-clean-beam_size_4-epoch-30-avg-2-modified_beam_search-beam-size-4-uam.txt
+2023-01-09 01:18:52,547 INFO [utils.py:536] [test-clean-beam_size_4] %WER 9.43% [4959 / 52576, 592 ins, 455 del, 3912 sub ]
+2023-01-09 01:18:52,730 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_2m_bpe500_halfdelay/modified_beam_search/errs-test-clean-beam_size_4-epoch-30-avg-2-modified_beam_search-beam-size-4-uam.txt
+2023-01-09 01:18:52,731 INFO [decode.py:508] 
+For test-clean, WER of different settings are:
+beam_size_4	9.43	best for test-clean
+
+2023-01-09 01:18:55,707 INFO [decode.py:459] batch 0/?, cuts processed until now is 52
+2023-01-09 01:19:46,663 INFO [decode.py:459] batch 20/?, cuts processed until now is 1646
+2023-01-09 01:20:28,554 INFO [decode.py:459] batch 40/?, cuts processed until now is 2870
+2023-01-09 01:20:30,170 INFO [decode.py:478] The transcripts are stored in tiny_transducer_ctc/exp_2m_bpe500_halfdelay/modified_beam_search/recogs-test-other-beam_size_4-epoch-30-avg-2-modified_beam_search-beam-size-4-uam.txt
+2023-01-09 01:20:30,285 INFO [utils.py:536] [test-other-beam_size_4] %WER 23.53% [12315 / 52343, 1237 ins, 1620 del, 9458 sub ]
+2023-01-09 01:20:30,446 INFO [decode.py:491] Wrote detailed error stats to tiny_transducer_ctc/exp_2m_bpe500_halfdelay/modified_beam_search/errs-test-other-beam_size_4-epoch-30-avg-2-modified_beam_search-beam-size-4-uam.txt
+2023-01-09 01:20:30,446 INFO [decode.py:508] 
+For test-other, WER of different settings are:
+beam_size_4	23.53	best for test-other
+
+2023-01-09 01:20:30,446 INFO [decode.py:730] Done!

middle_bpe_500/decode_results/modified_beam_search/wer-summary-test-clean-beam_size_4-epoch-30-avg-2-modified_beam_search-beam-size-4-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_size_4	9.43

middle_bpe_500/decode_results/modified_beam_search/wer-summary-test-other-beam_size_4-epoch-30-avg-2-modified_beam_search-beam-size-4-uam.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_size_4	23.53

middle_bpe_500/exp/cpu_jit.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc21aa9bb4d0233243f57c76f52847e4c55f11b97cd71ae2e18128e30214b16d
+size 11207430