Commit Graph

  • 61f3c87d48 Merge branch 'master' into streaming-conformer pkufool 2022-06-25 09:50:37 +08:00
  • 934f25f987 move model parameters to train.py pkufool 2022-06-25 09:39:39 +08:00
  • 7ce69a2b03 fix typo Jun Wang 2022-06-25 09:39:15 +08:00
  • 146d7c5a93 Bug fix Daniel Povey 2022-06-24 19:49:13 +08:00
  • 29eab49966 fix a error for conformer.py luomingshuang 2022-06-24 19:32:24 +08:00
  • c0ea334738
    fix bug of concatenating list to tuple (#444) Tiance Wang 2022-06-24 19:31:09 +08:00
  • 117d348f70 This time get it right, last time was starting from wrong base. Daniel Povey 2022-06-24 19:00:58 +08:00
  • 6002e19836 Change max_fullcov_size back to 1023.. doing it this way because I forgot to commit the changes on this branch Daniel Povey 2022-06-24 18:59:34 +08:00
  • 662b9c2e2e
    fix bug of concatenating list to tuple Tiance Wang 2022-06-24 18:45:24 +08:00
  • c2ec901e73 Change param_pow from 0.75 to 0.5 Daniel Povey 2022-06-24 17:14:41 +08:00
  • 1b93d6f575 Merge branch 'master' into pruned-rnnt5-recipe-for-tal-csasr luomingshuang 2022-06-24 16:35:03 +08:00
  • ba78791e9d add wer and cer for Chinese and English respectively luomingshuang 2022-06-24 16:33:44 +08:00
  • f327407308 Avoid error if svd fails Daniel Povey 2022-06-24 13:20:20 +08:00
  • 2233c852fd Increase max_fullcov_size from 1023 to 2048 Daniel Povey 2022-06-24 13:02:17 +08:00
  • c391bfd100
    fix errors for soft connection (#443) Mingshuang Luo 2022-06-24 10:40:46 +08:00
  • 9352287030 fix errors for soft connection luomingshuang 2022-06-24 10:26:18 +08:00
  • c1334d4da6 change for utils.py luomingshuang 2022-06-23 20:16:01 +08:00
  • 0475d75d15
    [Ready to be merged] Add RNN-LM to Conformer-CTC decoding (#439) ezerhouni 2022-06-23 13:37:03 +02:00
  • 8039787583 Update Results.md Erwan 2022-06-23 13:20:51 +02:00
  • dc25ab909a Fix help Erwan 2022-06-23 13:18:23 +02:00
  • 73a281725c Add export for rnn_lm Erwan 2022-06-23 12:41:06 +02:00
  • 1ab9553fd4 change for conformer.py luomingshuang 2022-06-23 16:49:43 +08:00
  • 5b5371f801 do some changes for merging luomingshuang 2022-06-23 16:27:41 +08:00
  • 9cc3f61056 do some changes for merging luomingshuang 2022-06-23 16:24:56 +08:00
  • 683b8e1504 Some code reworking and fixes, rationalizing how speedup is done and fix an issue affecting learning rate. Daniel Povey 2022-06-23 15:40:51 +08:00
  • c34344e98f Change param_pow,grad_pow to 0.75,0.95 Daniel Povey 2022-06-23 14:04:11 +08:00
  • 3767f3d616 Reducd param_pow from 0.8 to 0.7 Daniel Povey 2022-06-23 11:12:12 +08:00
  • ca17f23da3 Reduce powers to 0.8,0.85 Daniel Povey 2022-06-23 11:09:27 +08:00
  • 09282ca28c Reduce some print statements, convert some to info Daniel Povey 2022-06-23 10:41:08 +08:00
  • 7f756b2910 Introduce grad_pow option, set it to 0.9 and param_pow to 0.85 Daniel Povey 2022-06-23 10:33:14 +08:00
  • 0e54471d2d remove absolute path Erwan 2022-06-22 15:13:46 +02:00
  • 66afaf402d Fix code according to review Erwan 2022-06-22 15:04:25 +02:00
  • 73a31db1b2 Remove rnn_lm directory from egs Erwan 2022-06-22 10:47:50 +02:00
  • 4929faeed4 copy files from conv_emformer_transducer_stateless2 yaozengwei 2022-06-22 14:41:04 +08:00
  • 9c37c16326 modify .flake8 yaozengwei 2022-06-22 12:28:26 +08:00
  • dc89b61b80
    Add fast_beam_search_nbest. (#420) Fangjun Kuang 2022-06-22 00:09:25 +08:00
  • e032b03657 Minor fixes. Fangjun Kuang 2022-06-22 00:08:33 +08:00
  • 284cbf7ed1 Support LG for fast beam search. Fangjun Kuang 2022-06-21 22:55:18 +08:00
  • 42e3e883fd refactor init states for stream yaozengwei 2022-06-21 22:07:36 +08:00
  • f5af662b7b Support LG for pruned_transducer_stateless2. Fangjun Kuang 2022-06-21 21:55:08 +08:00
  • 136ee53447 Support LG decoding in pruned_transducer_stateless Fangjun Kuang 2022-06-21 21:31:33 +08:00
  • 7100c33820
    Add pruned RNN-T for aishell. (#436) Fangjun Kuang 2022-06-21 21:17:22 +08:00
  • 15d5748da3 Add links to sherpa. Fangjun Kuang 2022-06-21 21:09:11 +08:00
  • 5b19011edb fix doc in conv_emformer_transducer_stateless/emformer.py yaozengwei 2022-06-21 20:52:42 +08:00
  • d3daeaf5cd
    Upload extracted codebook indexes (#429) Zengwei Yao 2022-06-21 19:16:59 +08:00
  • 54d3ac9641 fix typo yaozengwei 2022-06-21 16:03:54 +08:00
  • ec9e4cffe2 Support using log_add in LG decoding with fast_beam_search. Fangjun Kuang 2022-06-21 15:08:07 +08:00
  • 5fe60dec43 Minor fixes on decode stream pkufool 2022-06-20 20:01:47 +08:00
  • 8d2f4ba098 Fix date format Erwan 2022-06-20 13:34:42 +02:00
  • 2eb21668e0 Update results and move rnn directory Erwan 2022-06-20 11:42:46 +02:00
  • 6898b012a9 Minor fixes. Fangjun Kuang 2022-06-20 17:03:50 +08:00
  • 91b2765cfd
    Fixs spelling mistake (#438) 2xwwx2 2022-06-20 16:41:04 +08:00
  • c02be4bece
    Fixs spelling mistake 2xwwx2 2022-06-20 16:40:11 +08:00
  • 998091ef52
    do some changes for export.py (#437) Mingshuang Luo 2022-06-20 14:57:08 +08:00
  • 38e58a46a2 do some changes for export.py luomingshuang 2022-06-20 14:27:19 +08:00
  • 4124cd7241 Reduce debug frequencies Daniel Povey 2022-06-20 13:48:42 +08:00
  • c10a9889fa Reduce grad_eps to 1e-14 Daniel Povey 2022-06-20 13:45:53 +08:00
  • a42d96dfe0
    Fix warmup (#435) Zengwei Yao 2022-06-20 13:40:01 +08:00
  • 2d811da27e Update CI. Fangjun Kuang 2022-06-20 12:49:55 +08:00
  • 2d684f11bf support torch script. Fangjun Kuang 2022-06-20 12:30:49 +08:00
  • b560221eea Change param_pow to 0.75 Daniel Povey 2022-06-20 00:29:07 +08:00
  • cf816a9138 Change param_pow to 0.0 Daniel Povey 2022-06-19 22:44:17 +08:00
  • ad5337fe95 Set param_pow to 0.5 Daniel Povey 2022-06-19 22:14:34 +08:00
  • 493b96e3c4 Add pruned RNN-T for aishell. Fangjun Kuang 2022-06-19 20:02:23 +08:00
  • 8acc398428 Reduce debug freq Daniel Povey 2022-06-19 11:21:48 +08:00
  • 29dfe37f22 Implement min,max param rms Daniel Povey 2022-06-18 22:22:23 +08:00
  • b2a4c33ddd delete comments yaozengwei 2022-06-18 20:07:59 +08:00
  • d7f4920206 fix warmup when scan_pessimistic_batches_for_oom yaozengwei 2022-06-18 19:58:05 +08:00
  • 74c14f5f5d Merge remote-tracking branch 'k2-fsa/master' yaozengwei 2022-06-18 17:48:51 +08:00
  • 3fb3cc4e23 Make random cov have correct diagonal; reset grad_min_rand to 0. Daniel Povey 2022-06-18 17:04:44 +08:00
  • 3f0b57403e Min-rand-scale 0.2 for grad cov Daniel Povey 2022-06-18 16:26:37 +08:00
  • e996f7f371 First version I am running, of the speedup code Daniel Povey 2022-06-18 15:17:51 +08:00
  • a6bf32c3e0 Fast code seems to pass tests. Daniel Povey 2022-06-18 12:56:16 +08:00
  • 5f64396112 Merge branch 'master' into streaming-conformer pkufool 2022-06-18 09:08:34 +08:00
  • ab788980c9
    Fix an error introduced by supporting torchscript for torch 1.6.0 (#434) Fangjun Kuang 2022-06-18 08:57:20 +08:00
  • 251bff3f41 Fix an error introduced by supporting torchscript for torch 1.6.0 Fangjun Kuang 2022-06-18 08:54:03 +08:00
  • 6a937ffa40 Minor fixes for streaming decoding pkufool 2022-06-18 08:41:52 +08:00
  • 88ed814197 Fix jit export for torch 1.6 pkufool 2022-06-18 08:35:25 +08:00
  • 208bbb6325 fix bug in conv_emformer_transducer_stateless2/emformer.py yaozengwei 2022-06-17 22:52:05 +08:00
  • d53f69108f
    Support torch 1.6.0 (#433) Fangjun Kuang 2022-06-17 22:24:47 +08:00
  • 03d54a303f Support torch 1.6.0 Fangjun Kuang 2022-06-17 22:22:35 +08:00
  • ea5cd69e3b Possibly fix bug RE learning rate Daniel Povey 2022-06-17 20:36:36 +08:00
  • cbccc1dd91 Fix bias_correction2, power of 0.5 Daniel Povey 2022-06-17 20:20:16 +08:00
  • c27bb1c554 correct the files, ln -> cp yaozengwei 2022-06-17 19:16:22 +08:00
  • 5d877ef25e change tail padding length from right_context_length to chunk_length yaozengwei 2022-06-17 18:54:55 +08:00
  • 2d13b682fe Revert param_rel_eps to old value Daniel Povey 2022-06-17 18:22:27 +08:00
  • 92fb1c52c8 Reduce debug statement frequency Daniel Povey 2022-06-17 16:03:50 +08:00
  • d9a3eba370 Fix bug that caused it to use too much memory Daniel Povey 2022-06-17 16:02:47 +08:00
  • 6873aa1a13 Bug fix to print statement Daniel Povey 2022-06-17 15:35:07 +08:00
  • 0c455db55d comment some lines, random combine from 1/3 layers, on linear layers in combiner yaozengwei 2022-06-17 15:20:05 +08:00
  • 90622f88ad Decrease param_rel_eps default from 1e-4 to 1e-2 Daniel Povey 2022-06-17 15:10:46 +08:00
  • 7f0756e156 Implement structured version of conformer Daniel Povey 2022-06-17 15:10:21 +08:00
  • 5aa142df10 fix style yaozengwei 2022-06-17 14:51:56 +08:00
  • 2fe4af8c99 Prevent crash due to error in C_diff Daniel Povey 2022-06-17 12:35:55 +08:00
  • 6c499dcd66 Augment debug output Daniel Povey 2022-06-17 12:32:09 +08:00
  • 5c3e4f506f Reduce debug frequency Daniel Povey 2022-06-17 12:17:48 +08:00
  • 47df144253 Bug fix for scalar case Daniel Povey 2022-06-17 12:16:12 +08:00
  • 9e92d13a33 Some bug fixes.. seems to be working. Daniel Povey 2022-06-17 12:04:21 +08:00
  • 827a37c7fc Closer to finished Daniel Povey 2022-06-17 11:10:43 +08:00
  • 7833da7d2b add enable-distillation argument option, fix monir typos yaozengwei 2022-06-17 11:05:50 +08:00