Commit Graph

  • d83f947366 solve the conflicts luomingshuang 2022-06-01 22:02:57 +08:00
  • 40e3ed2096 solve the conflicts luomingshuang 2022-06-01 22:01:05 +08:00
  • 5d2615d46d change for yml luomingshuang 2022-06-01 21:58:12 +08:00
  • 9629be124d mask the initial cache pkufool 2022-06-01 21:50:02 +08:00
  • 077be8c9c9 change for README.md luomingshuang 2022-06-01 21:30:09 +08:00
  • 5691de6a49 change for yml luomingshuang 2022-06-01 21:01:03 +08:00
  • bd73280e52 change for yml luomingshuang 2022-06-01 20:43:21 +08:00
  • 2c910d1a5c change for workflow yml luomingshuang 2022-06-01 20:35:45 +08:00
  • 56fea10a0e change for test.yml luomingshuang 2022-06-01 20:07:39 +08:00
  • 126e5863f5 change for test.yml luomingshuang 2022-06-01 20:02:34 +08:00
  • 7b98f9dadd change for test.yml luomingshuang 2022-06-01 19:43:34 +08:00
  • 0c73664aef Reduce threshold to 1024 Daniel Povey 2022-06-01 14:42:56 +08:00
  • fbfc98f1d3
    Add streaming Emformer stateless RNN-T. (#390) Fangjun Kuang 2022-06-01 14:31:47 +08:00
  • 7cf7bebf09 change LilcomHdf5Writer to ChunkedLilcomHdf5Writer luomingshuang 2022-06-01 14:16:46 +08:00
  • c901abf332 update READMD.md luomingshuang 2022-06-01 14:03:57 +08:00
  • ca09b9798f Remove decomposition code from checkpoint.py; restore double precision model_avg Daniel Povey 2022-06-01 14:01:58 +08:00
  • 676c257ba8 update code for merging luomingshuang 2022-06-01 13:59:55 +08:00
  • 03e07e80ce More drafts for rebalancing code Daniel Povey 2022-06-01 13:58:42 +08:00
  • 9c9bf4f1e3 Some drafts of rebalancing code in optim.py Daniel Povey 2022-06-01 11:34:19 +08:00
  • 5b340cee85 Minor fixes. Fangjun Kuang 2022-06-01 11:06:44 +08:00
  • bc5c782294 Limit magnitude of linear_pos Daniel Povey 2022-06-01 10:40:54 +08:00
  • a6f4bc77c8 Update results for streaming Emformer. Fangjun Kuang 2022-06-01 08:32:36 +08:00
  • 61619c031e Add activation balancer to stop activations in self_attn from getting too large Daniel Povey 2022-06-01 00:40:45 +08:00
  • 10998bef69 add tail padding yaozengwei 2022-05-31 18:53:20 +08:00
  • 1b909d9178 Merge remote-tracking branch 'dan/master' into streaming-emformer-2022-05-27 Fangjun Kuang 2022-05-31 17:39:39 +08:00
  • 39bed5f5c4 Add streaming Emformer stateless RNN-T. Fangjun Kuang 2022-05-31 17:26:43 +08:00
  • da2ffd4d27 Do average computation in double precision Daniel Povey 2022-05-31 14:39:21 +08:00
  • b2259184b5 Use single precision for model average; increase average-period to 200. Daniel Povey 2022-05-31 14:31:46 +08:00
  • ab9eb0d52c Use decompose=True arg for model averaging Daniel Povey 2022-05-31 14:28:53 +08:00
  • 8d4c987e21 Update checkpoint.py to support decompose argument Daniel Povey 2022-05-31 14:25:45 +08:00
  • bb7ea3141b Merge remote-tracking branch 'k2-fsa/master' yaozengwei 2022-05-31 13:34:23 +08:00
  • 1651fe0d42 Merge changes from pruned_transducer_stateless4->5 Daniel Povey 2022-05-31 13:00:11 +08:00
  • c7cf229f56 Revers pruned_transducer_stateless4 to upstream/master Daniel Povey 2022-05-31 12:45:51 +08:00
  • 741dcd1d6d Move pruned_transducer_stateless4 to pruned_transducer_stateless7 Daniel Povey 2022-05-31 12:45:28 +08:00
  • 8f877efec5 Remove pruned_transducer_stateless4b Daniel Povey 2022-05-31 12:29:45 +08:00
  • 7011956c6c Merge remote-tracking branch 'upstream/master' into cain3d_clean_merge Daniel Povey 2022-05-31 12:17:45 +08:00
  • 7307f1c6bc fix typos yaozengwei 2022-05-30 22:37:47 +08:00
  • c3df609805 Revert lrate changes Daniel Povey 2022-05-30 16:21:20 +08:00
  • 90024c308f predicted masked codebook indexes only Guo Liyong 2022-05-30 14:33:48 +08:00
  • 0c33543ce7 copy asr_datamodule.py to stateless6 Guo Liyong 2022-05-30 14:31:20 +08:00
  • b01c09a693 Remove the natural gradient stuff while keeping cosmetic changes. Daniel Povey 2022-05-30 11:56:11 +08:00
  • 496fb963c8 icefall format Guo Liyong 2022-05-30 10:49:08 +08:00
  • 03fa9957a6 direct copy from lhotse Guo Liyong 2022-05-30 10:47:19 +08:00
  • 0325e3a04e Add torch.jit.export pkufool 2022-05-29 16:25:01 +08:00
  • 8a96f29a11 Further increase learning rate Daniel Povey 2022-05-28 17:48:12 +08:00
  • 605838da55 Fix style pkufool 2022-05-29 07:48:52 +08:00
  • aecaecfb17 Fix style pkufool 2022-05-29 07:35:52 +08:00
  • fb3f3d2526 Fix style pkufool 2022-05-29 07:33:54 +08:00
  • b23db42486 Fix conv cache error, support async streaming decoding pkufool 2022-05-29 07:06:44 +08:00
  • 2b8ea98fc2 Improve documentation; remove unused code. Daniel Povey 2022-05-28 19:22:09 +08:00
  • 603b707fa8 monir style fix yaozengwei 2022-05-28 19:17:19 +08:00
  • 26b443eac3 add rule in .flake8 yaozengwei 2022-05-28 19:11:04 +08:00
  • 4c32afae20
    Update Dockerfile Teo Wen Shen 2022-05-28 19:54:44 +09:00
  • 66b8636878 copy files from existing branch yaozengwei 2022-05-28 18:16:22 +08:00
  • 79b5ac21ca Changed Dockerfile Teo 2022-05-28 10:08:28 +00:00
  • 295595d334 Revert the exclusion of dim=500 Daniel Povey 2022-05-28 17:49:16 +08:00
  • cebadffc9d minor fix yaozengwei 2022-05-28 14:06:24 +08:00
  • a6ca4d283a update RESULTS.md and add decode command yaozengwei 2022-05-28 14:02:26 +08:00
  • c4ee2bc0af
    [Ready to merge]stateless6: states4 + hubert distillation. (#387) LIyong.Guo 2022-05-28 12:37:50 +08:00
  • 0b645662f9 Increase learning rates. Daniel Povey 2022-05-28 12:12:34 +08:00
  • e771472a30 Speed up learning rate schedule. Daniel Povey 2022-05-28 11:30:45 +08:00
  • 111dc643ac fix typo Guo Liyong 2022-05-28 01:39:41 +08:00
  • bd81457222 adjust results format Guo Liyong 2022-05-28 01:37:51 +08:00
  • 093fbd1234 update results by stateless6 Guo Liyong 2022-05-28 01:31:52 +08:00
  • b3329f333a make export.py call stateless6/train.py instead of stateless2/train.py Guo Liyong 2022-05-28 01:21:11 +08:00
  • 828defb019 Some temp code for loading old state dict Daniel Povey 2022-05-28 00:29:47 +08:00
  • d89cb53a3b Revert max_eff_lr to initial_lr Daniel Povey 2022-05-27 21:42:20 +08:00
  • 7aa47408af Bug fixes to avoid inf alpha Daniel Povey 2022-05-27 21:41:05 +08:00
  • 0787583580 Increase max_eff_lr Daniel Povey 2022-05-27 21:03:08 +08:00
  • fd0e9d4bad Fix bug for scalars. Daniel Povey 2022-05-27 20:48:41 +08:00
  • 503b79252f Add new update and max_eff_lr Daniel Povey 2022-05-27 20:44:10 +08:00
  • 7e12c46f27 Merge branch 'pruned4_result' of github.com:yaozengwei/icefall into pruned4_result yaozengwei 2022-05-27 18:18:19 +08:00
  • bc659d5dcf set default value of --use-averaged-model to True yaozengwei 2022-05-27 18:17:26 +08:00
  • 9bc7c31350
    Update RESULT.md Zengwei Yao 2022-05-27 18:13:37 +08:00
  • 525c1ea277 update RESULT.md about pruned_transducer_stateless4 yaozengwei 2022-05-27 18:07:07 +08:00
  • 4efe920401 More consistent use of eps. Daniel Povey 2022-05-27 17:41:19 +08:00
  • eed864a3db Change power from 0.66 to 1.0, like natural gradient. Daniel Povey 2022-05-27 16:45:42 +08:00
  • 89fad8cc5a Change power to 0.66 Daniel Povey 2022-05-27 16:39:54 +08:00
  • 9d36a13aa4
    Update egs/librispeech/ASR/pruned_transducer_stateless6/hubert_xlarge.py LIyong.Guo 2022-05-27 15:06:23 +08:00
  • 85f35698ef fix type error of dst_manifest_dir Guo Liyong 2022-05-27 14:48:11 +08:00
  • 50641cdeab Merge branch 'stateless6' of github.com:glynpu/icefall into stateless6 Guo Liyong 2022-05-27 13:49:39 +08:00
  • 973fd83cf1 add intro of statless6 in README Guo Liyong 2022-05-27 13:33:22 +08:00
  • 8a4b3497a6
    Update egs/librispeech/ASR/pruned_transducer_stateless6/hubert_xlarge.py LIyong.Guo 2022-05-27 13:08:17 +08:00
  • 80dfdd1cfa a short intro to distillation framework Guo Liyong 2022-05-27 12:38:45 +08:00
  • f921a7be0e Merge branch 'stateless6' of github.com:glynpu/icefall into stateless6 Guo Liyong 2022-05-27 12:26:59 +08:00
  • 6bc387cf46 check fairseq and quantization Guo Liyong 2022-05-27 12:26:43 +08:00
  • 6e1a92ac7c
    Update egs/librispeech/ASR/pruned_transducer_stateless6/hubert_xlarge.py LIyong.Guo 2022-05-27 11:24:30 +08:00
  • 6fff82b917
    Update egs/librispeech/ASR/pruned_transducer_stateless6/hubert_xlarge.py LIyong.Guo 2022-05-27 11:23:43 +08:00
  • 91432397cf Merge branch 'stateless6' of github.com:glynpu/icefall into stateless6 Guo Liyong 2022-05-27 11:04:50 +08:00
  • 3f4b7fac84 add results of 100hours Guo Liyong 2022-05-27 11:04:28 +08:00
  • c10481a726 fix comment Guo Liyong 2022-05-27 10:58:15 +08:00
  • 95e921ce8b
    Update egs/librispeech/ASR/pruned_transducer_stateless6/hubert_xlarge.py LIyong.Guo 2022-05-27 10:54:39 +08:00
  • aa9e287b3a usage Guo Liyong 2022-05-27 10:37:29 +08:00
  • d9a6aff863 example usage Guo Liyong 2022-05-27 10:28:34 +08:00
  • 14a4d1d6f2 fix typo Guo Liyong 2022-05-27 09:54:48 +08:00
  • d8f68abff8 distillation with hubert Guo Liyong 2022-05-26 22:00:40 +08:00
  • 1f37eb5d0c a copy of stateless4 as base Guo Liyong 2022-05-26 21:46:08 +08:00
  • 545316636b Merge remote-tracking branch 'origin/master' yaozengwei 2022-05-26 21:55:56 +08:00
  • fbbc24f941 Merge remote-tracking branch 'k2-fsa/master' yaozengwei 2022-05-26 21:54:40 +08:00
  • 61e7929c60 Remove unused arg Daniel Povey 2022-05-26 15:18:03 +08:00