Zengwei commited on
Commit
0966908
1 Parent(s): 8d435a5
Files changed (28) hide show
  1. decoding_result/fast_beam_search/errs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt +0 -0
  2. decoding_result/fast_beam_search/errs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt +0 -0
  3. decoding_result/fast_beam_search/log-decode-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model-2023-10-08-19-14-56 +30 -0
  4. decoding_result/fast_beam_search/recogs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt +0 -0
  5. decoding_result/fast_beam_search/recogs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt +0 -0
  6. decoding_result/fast_beam_search/wer-summary-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt +2 -0
  7. decoding_result/fast_beam_search/wer-summary-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt +2 -0
  8. decoding_result/greedy_search/errs-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt +0 -0
  9. decoding_result/greedy_search/errs-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt +0 -0
  10. decoding_result/greedy_search/log-decode-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model-2023-10-08-19-09-51 +26 -0
  11. decoding_result/greedy_search/recogs-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt +0 -0
  12. decoding_result/greedy_search/recogs-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt +0 -0
  13. decoding_result/greedy_search/wer-summary-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt +2 -0
  14. decoding_result/greedy_search/wer-summary-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt +2 -0
  15. decoding_result/modified_beam_search/errs-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt +0 -0
  16. decoding_result/modified_beam_search/errs-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt +0 -0
  17. decoding_result/modified_beam_search/log-decode-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model-2023-09-11-11-00-42 +30 -0
  18. decoding_result/modified_beam_search/recogs-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt +0 -0
  19. decoding_result/modified_beam_search/recogs-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt +0 -0
  20. decoding_result/modified_beam_search/wer-summary-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt +2 -0
  21. decoding_result/modified_beam_search/wer-summary-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt +2 -0
  22. exp/decode.sh +2 -2
  23. exp/epoch-50.pt +3 -0
  24. exp/export.sh +2 -2
  25. exp/jit_script.pt +2 -2
  26. exp/pretrained.pt +1 -1
  27. exp/tensorboard/events.out.tfevents.1694324824.de-74279-k2-train-3-0423201227-84b4557756-8lx4n.2538563.0 +3 -0
  28. exp/train.sh +1 -1
decoding_result/fast_beam_search/errs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/fast_beam_search/errs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/fast_beam_search/log-decode-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model-2023-10-08-19-14-56 ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2023-10-08 19:14:56,737 INFO [decode.py:670] Decoding started
2
+ 2023-10-08 19:14:56,738 INFO [decode.py:676] Device: cuda:0
3
+ 2023-10-08 19:14:56,744 INFO [decode.py:686] {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.24.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'c51a0b9684442a88ee37f3ce0af686a04b66855b', 'k2-git-date': 'Mon May 1 21:38:03 2023', 'lhotse-version': '1.12.0.dev+git.891bad1.clean', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'new-zipformer-onnx-decode', 'icefall-git-sha1': 'a53d7102-dirty', 'icefall-git-date': 'Fri Jun 30 11:38:26 2023', 'icefall-path': '/ceph-zw/workspace/zipformer/icefall_zipformer', 'k2-path': '/ceph-zw/workspace/k2/k2/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-zw/workspace/share/lhotse/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-9-0208143539-7dbf569d4f-r7nrb', 'IP address': '10.177.13.150'}, 'epoch': 50, 'iter': 0, 'avg': 30, 'use_averaged_model': True, 'exp_dir': PosixPath('zipformer/exp-large'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'lang_dir': PosixPath('data/lang_bpe_500'), 'decoding_method': 'fast_beam_search', 'beam_size': 4, 'beam': 20.0, 'ngram_lm_scale': 0.01, 'max_contexts': 8, 'max_states': 64, 'context_size': 2, 'max_sym_per_frame': 1, 'num_paths': 200, 'nbest_scale': 0.5, 'num_encoder_layers': '2,2,4,5,4,2', 'downsampling_factor': '1,2,4,8,4,2', 'feedforward_dim': '512,768,1536,2048,1536,768', 'num_heads': '4,4,4,8,4,4', 'encoder_dim': '192,256,512,768,512,256', 'query_head_dim': '32', 'value_head_dim': '12', 'pos_head_dim': '4', 'pos_dim': 48, 'encoder_unmasked_dim': '192,192,256,320,256,192', 'cnn_module_kernel': '31,31,15,15,15,31', 'decoder_dim': 512, 'joiner_dim': 512, 'causal': False, 'chunk_size': '16,32,64,-1', 'left_context_frames': '64,128,256,-1', 'use_transducer': True, 'use_ctc': False, 'full_libri': True, 'mini_libri': False, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 600, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'res_dir': PosixPath('zipformer/exp-large/fast_beam_search'), 'suffix': 'epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
4
+ 2023-10-08 19:14:56,745 INFO [decode.py:688] About to create model
5
+ 2023-10-08 19:14:58,156 INFO [decode.py:755] Calculating the averaged model over epoch range from 20 (excluded) to 50
6
+ 2023-10-08 19:15:12,887 INFO [decode.py:791] Number of model parameters: 148439574
7
+ 2023-10-08 19:15:12,887 INFO [asr_datamodule.py:465] About to get test-clean cuts
8
+ 2023-10-08 19:15:12,894 INFO [asr_datamodule.py:472] About to get test-other cuts
9
+ 2023-10-08 19:15:16,339 INFO [decode.py:562] batch 0/?, cuts processed until now is 43
10
+ 2023-10-08 19:15:55,430 INFO [decode.py:562] batch 20/?, cuts processed until now is 1430
11
+ 2023-10-08 19:16:37,318 INFO [decode.py:562] batch 40/?, cuts processed until now is 2561
12
+ 2023-10-08 19:16:38,663 INFO [decode.py:580] The transcripts are stored in zipformer/exp-large/fast_beam_search/recogs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt
13
+ 2023-10-08 19:16:38,785 INFO [utils.py:562] [test-clean-beam_20.0_max_contexts_8_max_states_64] %WER 2.09% [1098 / 52576, 132 ins, 80 del, 886 sub ]
14
+ 2023-10-08 19:16:39,098 INFO [decode.py:593] Wrote detailed error stats to zipformer/exp-large/fast_beam_search/errs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt
15
+ 2023-10-08 19:16:39,102 INFO [decode.py:610]
16
+ For test-clean, WER of different settings are:
17
+ beam_20.0_max_contexts_8_max_states_64 2.09 best for test-clean
18
+
19
+ 2023-10-08 19:16:42,034 INFO [decode.py:562] batch 0/?, cuts processed until now is 52
20
+ 2023-10-08 19:17:00,934 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([6.5991, 6.6086, 6.0511, 6.2126], device='cuda:0')
21
+ 2023-10-08 19:17:20,966 INFO [decode.py:562] batch 20/?, cuts processed until now is 1647
22
+ 2023-10-08 19:18:01,513 INFO [decode.py:562] batch 40/?, cuts processed until now is 2870
23
+ 2023-10-08 19:18:02,719 INFO [decode.py:580] The transcripts are stored in zipformer/exp-large/fast_beam_search/recogs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt
24
+ 2023-10-08 19:18:02,847 INFO [utils.py:562] [test-other-beam_20.0_max_contexts_8_max_states_64] %WER 4.68% [2448 / 52343, 257 ins, 200 del, 1991 sub ]
25
+ 2023-10-08 19:18:03,160 INFO [decode.py:593] Wrote detailed error stats to zipformer/exp-large/fast_beam_search/errs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt
26
+ 2023-10-08 19:18:03,164 INFO [decode.py:610]
27
+ For test-other, WER of different settings are:
28
+ beam_20.0_max_contexts_8_max_states_64 4.68 best for test-other
29
+
30
+ 2023-10-08 19:18:03,164 INFO [decode.py:822] Done!
decoding_result/fast_beam_search/recogs-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/fast_beam_search/recogs-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/fast_beam_search/wer-summary-test-clean-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ settings WER
2
+ beam_20.0_max_contexts_8_max_states_64 2.09
decoding_result/fast_beam_search/wer-summary-test-other-beam_20.0_max_contexts_8_max_states_64-epoch-50-avg-30-beam-20.0-max-contexts-8-max-states-64-use-averaged-model.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ settings WER
2
+ beam_20.0_max_contexts_8_max_states_64 4.68
decoding_result/greedy_search/errs-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/greedy_search/errs-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/greedy_search/log-decode-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model-2023-10-08-19-09-51 ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2023-10-08 19:09:51,699 INFO [decode.py:670] Decoding started
2
+ 2023-10-08 19:09:51,700 INFO [decode.py:676] Device: cuda:0
3
+ 2023-10-08 19:09:51,704 INFO [decode.py:686] {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.24.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'c51a0b9684442a88ee37f3ce0af686a04b66855b', 'k2-git-date': 'Mon May 1 21:38:03 2023', 'lhotse-version': '1.12.0.dev+git.891bad1.clean', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'new-zipformer-onnx-decode', 'icefall-git-sha1': 'a53d7102-dirty', 'icefall-git-date': 'Fri Jun 30 11:38:26 2023', 'icefall-path': '/ceph-zw/workspace/zipformer/icefall_zipformer', 'k2-path': '/ceph-zw/workspace/k2/k2/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-zw/workspace/share/lhotse/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-9-0208143539-7dbf569d4f-r7nrb', 'IP address': '10.177.13.150'}, 'epoch': 50, 'iter': 0, 'avg': 30, 'use_averaged_model': True, 'exp_dir': PosixPath('zipformer/exp-large'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'lang_dir': PosixPath('data/lang_bpe_500'), 'decoding_method': 'greedy_search', 'beam_size': 4, 'beam': 20.0, 'ngram_lm_scale': 0.01, 'max_contexts': 8, 'max_states': 64, 'context_size': 2, 'max_sym_per_frame': 1, 'num_paths': 200, 'nbest_scale': 0.5, 'num_encoder_layers': '2,2,4,5,4,2', 'downsampling_factor': '1,2,4,8,4,2', 'feedforward_dim': '512,768,1536,2048,1536,768', 'num_heads': '4,4,4,8,4,4', 'encoder_dim': '192,256,512,768,512,256', 'query_head_dim': '32', 'value_head_dim': '12', 'pos_head_dim': '4', 'pos_dim': 48, 'encoder_unmasked_dim': '192,192,256,320,256,192', 'cnn_module_kernel': '31,31,15,15,15,31', 'decoder_dim': 512, 'joiner_dim': 512, 'causal': False, 'chunk_size': '16,32,64,-1', 'left_context_frames': '64,128,256,-1', 'use_transducer': True, 'use_ctc': False, 'full_libri': True, 'mini_libri': False, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 600, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'res_dir': PosixPath('zipformer/exp-large/greedy_search'), 'suffix': 'epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
4
+ 2023-10-08 19:09:51,705 INFO [decode.py:688] About to create model
5
+ 2023-10-08 19:09:53,088 INFO [decode.py:755] Calculating the averaged model over epoch range from 20 (excluded) to 50
6
+ 2023-10-08 19:10:26,088 INFO [decode.py:791] Number of model parameters: 148439574
7
+ 2023-10-08 19:10:26,089 INFO [asr_datamodule.py:465] About to get test-clean cuts
8
+ 2023-10-08 19:10:26,095 INFO [asr_datamodule.py:472] About to get test-other cuts
9
+ 2023-10-08 19:10:28,883 INFO [decode.py:562] batch 0/?, cuts processed until now is 43
10
+ 2023-10-08 19:10:41,405 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([6.2223, 5.6598, 5.7502, 5.9016], device='cuda:0')
11
+ 2023-10-08 19:11:14,220 INFO [decode.py:580] The transcripts are stored in zipformer/exp-large/greedy_search/recogs-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt
12
+ 2023-10-08 19:11:14,339 INFO [utils.py:562] [test-clean-greedy_search] %WER 2.08% [1092 / 52576, 126 ins, 81 del, 885 sub ]
13
+ 2023-10-08 19:11:14,646 INFO [decode.py:593] Wrote detailed error stats to zipformer/exp-large/greedy_search/errs-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt
14
+ 2023-10-08 19:11:14,649 INFO [decode.py:610]
15
+ For test-clean, WER of different settings are:
16
+ greedy_search 2.08 best for test-clean
17
+
18
+ 2023-10-08 19:11:16,701 INFO [decode.py:562] batch 0/?, cuts processed until now is 52
19
+ 2023-10-08 19:11:58,371 INFO [decode.py:580] The transcripts are stored in zipformer/exp-large/greedy_search/recogs-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt
20
+ 2023-10-08 19:11:58,491 INFO [utils.py:562] [test-other-greedy_search] %WER 4.69% [2455 / 52343, 244 ins, 211 del, 2000 sub ]
21
+ 2023-10-08 19:11:58,799 INFO [decode.py:593] Wrote detailed error stats to zipformer/exp-large/greedy_search/errs-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt
22
+ 2023-10-08 19:11:58,803 INFO [decode.py:610]
23
+ For test-other, WER of different settings are:
24
+ greedy_search 4.69 best for test-other
25
+
26
+ 2023-10-08 19:11:58,803 INFO [decode.py:822] Done!
decoding_result/greedy_search/recogs-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/greedy_search/recogs-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/greedy_search/wer-summary-test-clean-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ settings WER
2
+ greedy_search 2.08
decoding_result/greedy_search/wer-summary-test-other-greedy_search-epoch-50-avg-30-context-2-max-sym-per-frame-1-use-averaged-model.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ settings WER
2
+ greedy_search 4.69
decoding_result/modified_beam_search/errs-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/modified_beam_search/errs-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/modified_beam_search/log-decode-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model-2023-09-11-11-00-42 ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2023-09-11 11:00:42,917 INFO [decode.py:670] Decoding started
2
+ 2023-09-11 11:00:42,918 INFO [decode.py:676] Device: cuda:0
3
+ 2023-09-11 11:00:42,920 INFO [decode.py:686] {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.24.1', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'c51a0b9684442a88ee37f3ce0af686a04b66855b', 'k2-git-date': 'Mon May 1 21:38:03 2023', 'lhotse-version': '1.12.0.dev+git.891bad1.clean', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'new-zipformer-onnx-decode', 'icefall-git-sha1': 'a53d7102-clean', 'icefall-git-date': 'Fri Jun 30 11:38:26 2023', 'icefall-path': '/ceph-zw/workspace/zipformer/icefall_zipformer', 'k2-path': '/ceph-zw/workspace/k2/k2/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-zw/workspace/share/lhotse/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-7-1218101249-5d97868c7c-98fdf', 'IP address': '10.177.22.20'}, 'epoch': 50, 'iter': 0, 'avg': 30, 'use_averaged_model': True, 'exp_dir': PosixPath('zipformer/exp-large'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'lang_dir': PosixPath('data/lang_bpe_500'), 'decoding_method': 'modified_beam_search', 'beam_size': 4, 'beam': 20.0, 'ngram_lm_scale': 0.01, 'max_contexts': 8, 'max_states': 64, 'context_size': 2, 'max_sym_per_frame': 1, 'num_paths': 200, 'nbest_scale': 0.5, 'num_encoder_layers': '2,2,4,5,4,2', 'downsampling_factor': '1,2,4,8,4,2', 'feedforward_dim': '512,768,1536,2048,1536,768', 'num_heads': '4,4,4,8,4,4', 'encoder_dim': '192,256,512,768,512,256', 'query_head_dim': '32', 'value_head_dim': '12', 'pos_head_dim': '4', 'pos_dim': 48, 'encoder_unmasked_dim': '192,192,256,320,256,192', 'cnn_module_kernel': '31,31,15,15,15,31', 'decoder_dim': 512, 'joiner_dim': 512, 'causal': False, 'chunk_size': '16,32,64,-1', 'left_context_frames': '64,128,256,-1', 'use_transducer': True, 'use_ctc': False, 'full_libri': True, 'mini_libri': False, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 600, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'res_dir': PosixPath('zipformer/exp-large/modified_beam_search'), 'suffix': 'epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
4
+ 2023-09-11 11:00:42,921 INFO [decode.py:688] About to create model
5
+ 2023-09-11 11:00:43,870 INFO [decode.py:755] Calculating the averaged model over epoch range from 20 (excluded) to 50
6
+ 2023-09-11 11:01:04,161 INFO [decode.py:791] Number of model parameters: 148439574
7
+ 2023-09-11 11:01:04,161 INFO [asr_datamodule.py:465] About to get test-clean cuts
8
+ 2023-09-11 11:01:04,164 INFO [asr_datamodule.py:472] About to get test-other cuts
9
+ 2023-09-11 11:01:09,312 INFO [decode.py:562] batch 0/?, cuts processed until now is 43
10
+ 2023-09-11 11:02:34,610 INFO [decode.py:562] batch 20/?, cuts processed until now is 1430
11
+ 2023-09-11 11:03:44,556 INFO [decode.py:562] batch 40/?, cuts processed until now is 2574
12
+ 2023-09-11 11:03:47,304 INFO [decode.py:580] The transcripts are stored in zipformer/exp-large/modified_beam_search/recogs-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt
13
+ 2023-09-11 11:03:47,453 INFO [utils.py:562] [test-clean-beam_size_4] %WER 2.06% [1083 / 52576, 129 ins, 79 del, 875 sub ]
14
+ 2023-09-11 11:03:47,626 INFO [decode.py:593] Wrote detailed error stats to zipformer/exp-large/modified_beam_search/errs-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt
15
+ 2023-09-11 11:03:47,626 INFO [decode.py:610]
16
+ For test-clean, WER of different settings are:
17
+ beam_size_4 2.06 best for test-clean
18
+
19
+ 2023-09-11 11:03:52,458 INFO [decode.py:562] batch 0/?, cuts processed until now is 52
20
+ 2023-09-11 11:04:14,440 INFO [zipformer.py:1728] name=None, attn_weights_entropy = tensor([4.7445, 4.2710, 3.7179, 4.0052], device='cuda:0')
21
+ 2023-09-11 11:05:15,705 INFO [decode.py:562] batch 20/?, cuts processed until now is 1646
22
+ 2023-09-11 11:06:22,455 INFO [decode.py:562] batch 40/?, cuts processed until now is 2870
23
+ 2023-09-11 11:06:25,111 INFO [decode.py:580] The transcripts are stored in zipformer/exp-large/modified_beam_search/recogs-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt
24
+ 2023-09-11 11:06:25,187 INFO [utils.py:562] [test-other-beam_size_4] %WER 4.63% [2426 / 52343, 261 ins, 195 del, 1970 sub ]
25
+ 2023-09-11 11:06:25,430 INFO [decode.py:593] Wrote detailed error stats to zipformer/exp-large/modified_beam_search/errs-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt
26
+ 2023-09-11 11:06:25,431 INFO [decode.py:610]
27
+ For test-other, WER of different settings are:
28
+ beam_size_4 4.63 best for test-other
29
+
30
+ 2023-09-11 11:06:25,431 INFO [decode.py:822] Done!
decoding_result/modified_beam_search/recogs-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/modified_beam_search/recogs-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt ADDED
The diff for this file is too large to render. See raw diff
 
decoding_result/modified_beam_search/wer-summary-test-clean-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ settings WER
2
+ beam_size_4 2.06
decoding_result/modified_beam_search/wer-summary-test-other-beam_size_4-epoch-50-avg-30-modified_beam_search-beam-size-4-use-averaged-model.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ settings WER
2
+ beam_size_4 4.63
exp/decode.sh CHANGED
@@ -1,8 +1,8 @@
1
  export CUDA_VISIBLE_DEVICES="0"
2
  for m in greedy_search modified_beam_search fast_beam_search; do
3
  ./zipformer/decode.py \
4
- --epoch 40 \
5
- --avg 16 \
6
  --exp-dir zipformer/exp-large \
7
  --max-duration 600 \
8
  --causal 0 \
 
1
  export CUDA_VISIBLE_DEVICES="0"
2
  for m in greedy_search modified_beam_search fast_beam_search; do
3
  ./zipformer/decode.py \
4
+ --epoch 50 \
5
+ --avg 30 \
6
  --exp-dir zipformer/exp-large \
7
  --max-duration 600 \
8
  --causal 0 \
exp/epoch-50.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f330c6b92b5ddd05ab32c163396a62069fa339e881e2a85cad5a8bcaacdc27f7
3
+ size 2376172406
exp/export.sh CHANGED
@@ -5,8 +5,8 @@
5
  --feedforward-dim 512,768,1536,2048,1536,768 \
6
  --encoder-dim 192,256,512,768,512,256 \
7
  --encoder-unmasked-dim 192,192,256,320,256,192 \
8
- --epoch 40 \
9
- --avg 16 \
10
  --jit 1
11
 
12
 
 
5
  --feedforward-dim 512,768,1536,2048,1536,768 \
6
  --encoder-dim 192,256,512,768,512,256 \
7
  --encoder-unmasked-dim 192,192,256,320,256,192 \
8
+ --epoch 50 \
9
+ --avg 30 \
10
  --jit 1
11
 
12
 
exp/jit_script.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7e02e5a8fc72bb83fd644d46456787c5b0971c310605df892bb6e7a372433903
3
- size 596529834
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0492252e90f0e7384311c2eb0f8266c0aa3860cc760b22ac0d8f3573eb2dc1b0
3
+ size 596533034
exp/pretrained.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d932bcff9148dd6d57dd5ef803279de9de6a4d80662360c76fb9bd7cee203e39
3
  size 594239019
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d52b77b75d7bd62da42075a495d50fb351cec31068ca9bbc383d38917562476c
3
  size 594239019
exp/tensorboard/events.out.tfevents.1694324824.de-74279-k2-train-3-0423201227-84b4557756-8lx4n.2538563.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c368e9a26552d64e0fb63f649e4a49e775f3c82cd47eb612c9eaa5b52fc54d3c
3
+ size 265820
exp/train.sh CHANGED
@@ -1,7 +1,7 @@
1
  export CUDA_VISIBLE_DEVICES="0,1,2,3"
2
  ./zipformer/train.py \
3
  --world-size 4 \
4
- --num-epochs 40 \
5
  --start-epoch 1 \
6
  --use-fp16 1 \
7
  --exp-dir zipformer/exp-large \
 
1
  export CUDA_VISIBLE_DEVICES="0,1,2,3"
2
  ./zipformer/train.py \
3
  --world-size 4 \
4
+ --num-epochs 50 \
5
  --start-epoch 1 \
6
  --use-fp16 1 \
7
  --exp-dir zipformer/exp-large \