Commit Graph

  • 36c241e59f update .flake8 yaozengwei 2022-05-02 12:22:24 +08:00
  • aea8a03e00 update decode file yaozengwei 2022-05-02 12:17:43 +08:00
  • 08b37e07a4 minor fix yaozengwei 2022-05-02 00:50:32 +08:00
  • fba9ae0502 First upload of model average codes. yaozengwei 2022-05-01 23:20:00 +08:00
  • ac9655c450 fix a bug of encoder layer index Guo Liyong 2022-04-30 16:30:07 +08:00
  • 6dc2e04462
    Update results. (#340) Fangjun Kuang 2022-04-29 15:49:45 +08:00
  • 7324249d55 Typo fixes. Fangjun Kuang 2022-04-29 15:47:42 +08:00
  • 7d6b801eee Update results. Fangjun Kuang 2022-04-29 15:45:06 +08:00
  • ac84220de9
    Modified conformer with multi datasets (#312) Fangjun Kuang 2022-04-29 15:40:30 +08:00
  • 00fd66459f Fix style issues. Fangjun Kuang 2022-04-29 15:28:42 +08:00
  • 15220797e3 comments about disk usage and training example script Guo Liyong 2022-04-29 15:20:48 +08:00
  • c7000b9c44 Update results. Fangjun Kuang 2022-04-29 15:11:42 +08:00
  • fb61e31904 Fix style issues. Fangjun Kuang 2022-04-29 14:15:19 +08:00
  • 8d2797d7cd Update results. Fangjun Kuang 2022-04-29 14:13:44 +08:00
  • a227bd76b4 Update CI. Fangjun Kuang 2022-04-29 14:06:34 +08:00
  • 5bbce704e2 Merge remote-tracking branch 'dan/master' into modified-conformer-with-multi-datasets Fangjun Kuang 2022-04-29 14:02:53 +08:00
  • 9721a42977 Update results. Fangjun Kuang 2022-04-29 14:01:24 +08:00
  • fc7574f6d2 Add results. Fangjun Kuang 2022-04-29 12:03:22 +08:00
  • c026a97d41 check codebook index range before saving Guo Liyong 2022-04-29 11:57:51 +08:00
  • 76e56fa28f check codebook index range Guo Liyong 2022-04-29 10:54:11 +08:00
  • 9c39d8b009 Merge remote-tracking branch 'k2-fsa/master' yaozengwei 2022-04-29 10:26:06 +08:00
  • 712e1c496f
    Merge 3d0474c98639d06fe638d7dc98769e7fa7c70032 into caab6cfd9216804a33d0530ee7e2d034af818447 Fangjun Kuang 2022-04-28 10:55:04 -04:00
  • 5aaf981d46 fairseq installation Guo Liyong 2022-04-28 20:43:06 +08:00
  • f8bee06acf pipeline to extract codebook index form hubert Guo Liyong 2022-04-28 20:35:16 +08:00
  • 0cb3303a5b codebook index extraction Guo Liyong 2022-04-28 20:34:16 +08:00
  • cc1dcafc70 quantizer training Guo Liyong 2022-04-28 20:31:33 +08:00
  • c26ead5e09 quantizer training data Guo Liyong 2022-04-28 20:26:53 +08:00
  • fb9c0c3971 decoder hubert model Guo Liyong 2022-04-28 00:05:06 +08:00
  • 063c0e21c0 Validate generated manifest files. Fangjun Kuang 2022-04-28 16:22:25 +08:00
  • 1c9936898b Fix training. Fangjun Kuang 2022-04-28 14:25:30 +08:00
  • 026f446a4d Use k2 pruned RNN-T. Fangjun Kuang 2022-04-28 14:13:26 +08:00
  • ea969e9b84 Merge remote-tracking branch 'dan/master' into rnnt-lstm-2022-04-21 Fangjun Kuang 2022-04-28 14:12:08 +08:00
  • caab6cfd92
    Support specifying iteration number of checkpoints for decoding. (#336) Fangjun Kuang 2022-04-28 14:09:22 +08:00
  • 026978e1c0 Merge remote-tracking branch 'dan/master' into rnnt-lstm-2022-04-21 Fangjun Kuang 2022-04-28 11:06:55 +08:00
  • b0e4e5cf31 Minor fixes for decoding. Fangjun Kuang 2022-04-28 10:39:08 +08:00
  • 187534df2e Merge branch 'modified-conformer-with-multi-datasets' of github.com:csukuangfj/icefall into modified-conformer-with-multi-datasets Fangjun Kuang 2022-04-28 06:54:56 +08:00
  • af20922320 Minor fixes. Fangjun Kuang 2022-04-28 06:51:11 +08:00
  • e83e48f2f2 Support specifying iteration number of checkpoints for decoding. Fangjun Kuang 2022-04-28 06:32:50 +08:00
  • 9d48f1ce7d train/decode with codebook loss Guo Liyong 2022-04-27 14:18:12 +08:00
  • 979f574259 add codebook loss Guo Liyong 2022-04-27 12:36:20 +08:00
  • 4b567e480f adapt for S, M and L training subset luomingshuang 2022-04-27 13:43:54 +08:00
  • cbc9c50bfc middle layer outputs from encoder Guo Liyong 2022-04-27 12:33:54 +08:00
  • 9ee57959ec output from middle layer Guo Liyong 2022-04-27 11:20:19 +08:00
  • 8d73423a29 a copy from pruned_transducer_stateless2 Guo Liyong 2022-04-27 10:55:54 +08:00
  • 8cdb893cb9 Merge branch 'master' of https://github.com/k2-fsa/icefall into spgi Desh Raj 2022-04-26 14:06:06 -04:00
  • e61e69fb43 add spgispeech transducer Desh Raj 2022-04-26 14:06:00 -04:00
  • 9aeea3e1af
    Support averaging models with weight tying. (#333) Fangjun Kuang 2022-04-26 13:32:03 +08:00
  • 551786b9bd Merge branch 'model-averaging-shared-params' of https://github.com/csukuangfj/icefall into knowledge_base_1b_merge Daniel Povey 2022-04-26 13:18:09 +08:00
  • 0da522cc4c Support averaging models with weight tying. Fangjun Kuang 2022-04-26 13:14:08 +08:00
  • eba025a6b4 Mess with thresholds for printing Daniel Povey 2022-04-26 10:39:35 +08:00
  • 3ba081e6d9 Add more custom_fwd,custom_bwd' Daniel Povey 2022-04-25 23:58:34 +08:00
  • 2c4478b6d1 Fix for half precision Daniel Povey 2022-04-25 23:03:34 +08:00
  • e718c7ac88 Remove unnecessary copy Daniel Povey 2022-04-25 20:41:00 +08:00
  • f6619a0b20 Remove unnecessary check Daniel Povey 2022-04-25 20:37:06 +08:00
  • 7d457a7781 Add some diagnostics Daniel Povey 2022-04-25 19:34:19 +08:00
  • edaaec09cd Update backprop of sampling.py to be slightly more efficient. Daniel Povey 2022-04-25 19:32:11 +08:00
  • 9a98e6ced6
    fix fp16 option in example usage (#332) pehonnet 2022-04-25 12:51:53 +02:00
  • 355be50e1d
    fix fp16 option in example usage pehonnet 2022-04-25 12:29:46 +02:00
  • d79f5fecf7 Pass model parameters from the command line. Fangjun Kuang 2022-04-25 17:26:43 +08:00
  • bbfa484196 Decrease model size, baseline is one Fangjun is running.. Daniel Povey 2022-04-25 17:07:20 +08:00
  • aea116ea25 Change printing-prob, initial scales Daniel Povey 2022-04-25 14:02:43 +08:00
  • bb7cb82b04 Some fixes/refactoring, make parameters shared Daniel Povey 2022-04-25 13:55:27 +08:00
  • 0d40b4617a Add knowledge-base lookup to model Daniel Povey 2022-04-25 13:40:47 +08:00
  • a359bfe504 Test with CUDA, bug fixes Daniel Povey 2022-04-25 13:19:09 +08:00
  • f8c7e6ffb3 Add some training code. Seems to be training successfully... Daniel Povey 2022-04-24 23:19:46 +08:00
  • df39fc6783 Fix devices Daniel Povey 2022-04-24 22:48:52 +08:00
  • a266922678 First version of sampling.py, tests run. Daniel Povey 2022-04-24 22:29:11 +08:00
  • fe5586e847 Change dirname Daniel Povey 2022-04-24 19:51:27 +08:00
  • 65cd1059f3 Init pruned2_knowledge dir Daniel Povey 2022-04-24 19:50:22 +08:00
  • b54d9a256d Minor fixes. Fangjun Kuang 2022-04-24 15:25:34 +08:00
  • b1c3705fbe Compute the Nbest oracle WER for RNN-T decoding. Fangjun Kuang 2022-04-24 15:10:30 +08:00
  • 85ac3a8000 Minor fixes. Fangjun Kuang 2022-04-23 16:53:01 +08:00
  • 51cc6486cd Add random combine from #229. Fangjun Kuang 2022-04-23 16:35:02 +08:00
  • 1614a68017 Copy files for editing. Fangjun Kuang 2022-04-23 16:23:25 +08:00
  • d30023c8de
    Merge a36b86cb239fe5bed12a318091f970483fc45920 into b3e6bf66dfbfe2f53a6f3471bc1302d0e038f927 Zengwei Yao 2022-04-22 12:11:16 +02:00
  • b3e6bf66df
    Add modified beam search decoding for streaming inference with emformer model (#327) Zengwei Yao 2022-04-22 18:06:07 +08:00
  • a36b86cb23 Merge branch 'streaming_new' into streaming yaozengwei 2022-04-22 17:08:42 +08:00
  • b612b3dc50 Merge branch 'streaming_decoding' into streaming_new yaozengwei 2022-04-22 17:04:17 +08:00
  • d766dc5aee
    Fix some typos. (#329) whsqkaak 2022-04-22 16:54:59 +09:00
  • 4f4c6f3b1e Fix some typos. LeeSH 2022-04-22 16:49:46 +09:00
  • 0930748b61 test restore dynamic sampler luomingshuang 2022-04-22 14:42:09 +08:00
  • ece99a862b Minor fix for transducer_emformer/streaming_feature_extractor.py yaozengwei 2022-04-22 11:23:23 +08:00
  • e97c9fbdbf Sorted imports for transducer_emformer/streaming_feature_extractor.py yaozengwei 2022-04-22 11:04:50 +08:00
  • 8fde2acd97 Merge branch 'streaming_decoding_new' into streaming_decoding yaozengwei 2022-04-21 21:04:18 +08:00
  • 83a5052cf0 Merge remote-tracking branch 'k2-fsa/master' into streaming_decoding_new yaozengwei 2022-04-21 20:40:27 +08:00
  • d20a852f61 Fixed docs. yaozengwei 2022-04-21 19:55:30 +08:00
  • cf0ce8db32 Fixed streaming decoding codes for emformer model. yaozengwei 2022-04-21 19:48:35 +08:00
  • 01be91217b WIP: Implement BMUF for distributed training. Fangjun Kuang 2022-04-21 18:07:27 +08:00
  • d6390fd107 Update params Guanbo Wang 2022-04-21 02:30:54 -04:00
  • 52b3ed2920 Use a stateless decoder for transducer_lstm. Fangjun Kuang 2022-04-21 13:58:43 +08:00
  • 3d0474c986 Fix style issues. Fangjun Kuang 2022-04-21 11:49:52 +08:00
  • e4d45adf5a Change model.py and joiner.py to use torchaudio's RNN-T loss. Fangjun Kuang 2022-04-21 11:01:08 +08:00
  • e83dcdc3b4 Copy files for editing. Fangjun Kuang 2022-04-21 10:37:24 +08:00
  • e9f0975868 Merge remote-tracking branch 'origin/modified-conformer-with-multi-datasets' into modified-conformer-with-multi-datasets Fangjun Kuang 2022-04-20 17:22:29 +08:00
  • 65fd981747 Disable speed perturbe for XL subset. Fangjun Kuang 2022-04-20 17:21:31 +08:00
  • 24db3a1934 update emformer_pruned_transducer_stateless/emformer.py yaozengwei 2022-04-20 14:21:45 +08:00
  • 18a1e959f7 process and compute fbank features for S and M subset luomingshuang 2022-04-20 14:02:35 +08:00
  • 3607c516d6
    Update results for torchaudio RNN-T. (#322) Fangjun Kuang 2022-04-20 11:15:10 +08:00
  • aadfa68b6b Update results for torchaudio RNN-T. Fangjun Kuang 2022-04-20 11:12:28 +08:00
  • 42f8afd264 Merge branch 'streaming_decoding' into streaming yaozengwei 2022-04-20 11:10:53 +08:00