Commit Graph

  • f679d6063b inexhaustible train dataloaders Guo Liyong 2021-11-03 17:42:33 +08:00
  • 4ac04b4981
    Merge 25fa6c0690a2649285956fbd2c031209cd3c247b into 810b193dcc3ad3f7a65bc3def63493711c9a084e Fangjun Kuang 2021-11-03 17:00:57 +08:00
  • 810b193dcc
    Clarify the doc about ctc-decoding. (#104) Fangjun Kuang 2021-11-03 07:16:49 +08:00
  • 4949dc6fc0 Clarify the doc about ctc-decoding. Kuang Fangjun 2021-11-03 07:15:35 +08:00
  • 1ac9bb3fd7 WIP: Begin to add decoding scripts. Fangjun Kuang 2021-11-01 22:01:12 +08:00
  • 19828cbf22 Add files form Dan. Fangjun Kuang 2021-11-01 21:58:43 +08:00
  • 3441634f34 Finish preparing the inputs for conformer lm from an nbest object. Fangjun Kuang 2021-11-01 21:34:22 +08:00
  • bf98c0fd27 fix style check Guo Liyong 2021-11-01 19:29:33 +08:00
  • 4ae1701bba giga speech dataloader Guo Liyong 2021-11-01 18:46:08 +08:00
  • 42b437bea6
    Use pre-sorted text to generate token ids for attention decoder. (#98) Fangjun Kuang 2021-10-29 13:46:41 +08:00
  • 312c3716d7 Fix typos. Fangjun Kuang 2021-10-29 13:34:47 +08:00
  • 110ef98ca3 Use pre-sorted text to generate token ids for attention decoder. Fangjun Kuang 2021-10-29 12:29:52 +08:00
  • 12d647d899
    Add a note about the CUDA OOM error. (#94) Fangjun Kuang 2021-10-29 12:17:56 +08:00
  • 6cbba14c3f Fix style issues. Fangjun Kuang 2021-10-29 12:15:09 +08:00
  • a9cdaaeac1 Delete RESULTS.md Mingshuang Luo 2021-10-28 16:00:08 +08:00
  • e2bb9b4350 Update decode.py Mingshuang Luo 2021-10-28 15:36:31 +08:00
  • 5e7c733b11 Update prepare.sh Mingshuang Luo 2021-10-28 15:25:24 +08:00
  • 4beb25c50b Update timit recipe Mingshuang Luo 2021-10-28 14:45:24 +08:00
  • e023a9df98 Add timit recipe for icefall Mingshuang Luo 2021-10-28 13:54:08 +08:00
  • 69c87203ca
    Merge branch 'k2-fsa:master' into master Mingshuang Luo 2021-10-27 21:24:49 +08:00
  • 1b9e4f0fea WIP: Decoding scripts using conformer LM. Fangjun Kuang 2021-10-27 19:54:28 +08:00
  • a7cf41d3ac Add a note about the CUDA OOM error. Fangjun Kuang 2021-10-27 16:59:47 +08:00
  • 8cb7f712e4
    Use GPU for averaging checkpoints if possible. (#84) Fangjun Kuang 2021-10-26 17:10:04 +08:00
  • 25fa6c0690 Support switching modified/standard CTC topo from commandline. Fangjun Kuang 2021-10-25 11:24:26 +08:00
  • 712ead8207
    Fix an error when attention decoder rescoring returns None. (#90) Fangjun Kuang 2021-10-22 19:52:25 +08:00
  • 12d75b43f2 Fix an error when attention decoder rescoring returns None. Fangjun Kuang 2021-10-22 19:50:27 +08:00
  • 902e0b238d
    Merge pull request #82 from pzelasko/feature/find-pessimistic-batches Piotr Żelasko 2021-10-19 11:26:13 -04:00
  • 3cc99d2af2 make flake8 happy Piotr Żelasko 2021-10-19 11:24:54 -04:00
  • d30244e28f
    add a docker file for some users (#87) cdxie 2021-10-19 13:00:59 +08:00
  • e9b8d9eb11
    add a describing file of how to use dockerfile cdxie 2021-10-19 12:36:46 +08:00
  • 6f1c5d32da
    add a docker file for some users cdxie 2021-10-19 11:54:43 +08:00
  • 86f3e0ef37 Make flake8 happy Piotr Żelasko 2021-10-18 09:54:40 -04:00
  • 6fbd7a287c Refactor OOM batch scanning into a local function Piotr Żelasko 2021-10-18 09:53:04 -04:00
  • d509d58f30 Merge branch 'master' into feature/find-pessimistic-batches Piotr Żelasko 2021-10-18 09:47:21 -04:00
  • 6370d519d2 Minor fixes. Fangjun Kuang 2021-10-18 17:23:08 +08:00
  • b1c08e3f9d Use pre-computed alignment in conformer CTC training. Fangjun Kuang 2021-10-18 16:45:25 +08:00
  • 3effcb4225
    Fix typos. (#85) Fangjun Kuang 2021-10-18 16:17:14 +08:00
  • 270a9f7dff Fix typos. Fangjun Kuang 2021-10-18 16:16:26 +08:00
  • 53b79fafa7
    Add MMI training with word pieces as modelling unit. (#6) Fangjun Kuang 2021-10-18 15:20:32 +08:00
  • f76ef6e58a Disable macOS tests for now. Fangjun Kuang 2021-10-18 15:19:57 +08:00
  • 3ac9b4595d Fix style issues. Fangjun Kuang 2021-10-18 15:04:27 +08:00
  • 0663b97599 Fix typos. Fangjun Kuang 2021-10-18 15:00:37 +08:00
  • 00dac43130 Fix style issues. Fangjun Kuang 2021-10-18 14:58:07 +08:00
  • f383666c40 Add doc about how to check and use extracted alignments. Fangjun Kuang 2021-10-18 14:53:32 +08:00
  • d7023c3c4b Merge remote-tracking branch 'dan/master' into mmi Fangjun Kuang 2021-10-18 14:38:50 +08:00
  • b8dbad5156 Update decoding script. Fangjun Kuang 2021-10-18 14:38:07 +08:00
  • 4890e27b45
    Extract framewise alignment information using CTC decoding (#39) Fangjun Kuang 2021-10-18 14:24:33 +08:00
  • 07140e5d5c Add doc about how to extract framewise alignments. Fangjun Kuang 2021-10-18 14:21:21 +08:00
  • 1c603c3bce Merge remote-tracking branch 'dan/master' into ctc-ali Fangjun Kuang 2021-10-18 14:07:20 +08:00
  • 47519b36a4 Use GPU for averaging checkpoints if possible. Fangjun Kuang 2021-10-18 11:32:10 +08:00
  • 7d65f6480a
    Merge 8e650a584134e9cd42216427ac1f2f2a0ae45b74 into bd7c2f7645c0aea4cb482cc1f60836907a61d36b Daniel Povey 2021-10-17 00:34:19 +03:30
  • bd7c2f7645
    fix conformer typo in docs (#83) Jan "yenda" Trmal 2021-10-15 19:46:17 -04:00
  • 6b44a71539 fix conformer typo in docs Jan Trmal 2021-10-15 17:15:35 -04:00
  • 403d1744ff Introduce backprop in finding OOM batches Piotr Żelasko 2021-10-15 10:05:13 -04:00
  • 8ae39529c6
    Merge d4440b421c3700982c7f9e20bea37e48fb144b7e into fee1f84b20a5a704428c5eac80de2ac4033e1b27 Fangjun Kuang 2021-10-15 20:39:33 +08:00
  • 77c5eca310
    Merge b7d4a4f983730d4b6f9a6b2e774c784efae03e34 into fee1f84b20a5a704428c5eac80de2ac4033e1b27 Fangjun Kuang 2021-10-15 17:35:49 +08:00
  • 060117a9ff Reformatting Piotr Żelasko 2021-10-14 21:40:14 -04:00
  • 1c7c79f2fc Find CUDA OOM batches before starting training Piotr Żelasko 2021-10-14 21:28:11 -04:00
  • 7edf94264e Script showing example installation with conda Piotr Żelasko 2021-10-14 16:41:27 -04:00
  • fee1f84b20
    Test pre-trained model in CI (#80) Fangjun Kuang 2021-10-15 00:41:33 +08:00
  • 45c893a78a Minor fixes. Fangjun Kuang 2021-10-15 00:17:42 +08:00
  • 4c485251e9 Merge remote-tracking branch 'dan/master' into test-pretrained Fangjun Kuang 2021-10-15 00:09:42 +08:00
  • 5401ce199d
    Update ctc-decoding on pretrained.py and conformer_ctc.rst (#78) Mingshuang Luo 2021-10-14 23:29:06 +08:00
  • f489f9bd82 Minor fixes. Fangjun Kuang 2021-10-14 22:06:20 +08:00
  • f7ae6e9ebc Minor fixes. Fangjun Kuang 2021-10-14 21:55:42 +08:00
  • 4ab52eeea4 Clone pre-trained model from GitHub. Fangjun Kuang 2021-10-14 21:20:58 +08:00
  • b975bdef9e Disable decoder layers in pretrained.py if it is not used. Fangjun Kuang 2021-10-14 21:16:59 +08:00
  • 2de12b195e Fix CI errors. Fangjun Kuang 2021-10-14 21:04:52 +08:00
  • 493c8812fd Install a CPU version of PyTorch. Fangjun Kuang 2021-10-14 20:56:15 +08:00
  • 3e03e860a6 Install kaldifeat Fangjun Kuang 2021-10-14 20:41:00 +08:00
  • 3b725c724a Minor fixes. Fangjun Kuang 2021-10-14 20:27:32 +08:00
  • da5dc4c504
    Merge a133571dc7d073290b927877b54ca4cad06b0ad9 into f2387fe523c6f89987f3723bfa967095e8de5127 Fangjun Kuang 2021-10-14 20:20:43 +08:00
  • bfa77040ca Add CI to run pre-trained models. Fangjun Kuang 2021-10-14 20:20:35 +08:00
  • f2387fe523
    Fix a bug introduced while supporting torch script. (#79) Fangjun Kuang 2021-10-14 20:09:38 +08:00
  • e6b004bfc5 Fix a bug introduced while supporting torch script. Fangjun Kuang 2021-10-14 20:01:57 +08:00
  • 19891e26d5 Update pretrained.py and conformer_ctc.rst Mingshuang Luo 2021-10-14 19:51:36 +08:00
  • 9176da7574 Update pretrained.py Mingshuang Luo 2021-10-14 19:21:02 +08:00
  • 870d4d09ba Update pretrained.py Mingshuang Luo 2021-10-14 19:15:47 +08:00
  • a22b638820 Update pretrained.py Mingshuang Luo 2021-10-14 19:00:25 +08:00
  • 40db9985ec Update ctc-decoding on pretrained.py and conformer_ctc.rst Mingshuang Luo 2021-10-14 18:27:15 +08:00
  • f3fd2792ae
    Merge branch 'k2-fsa:master' into master Mingshuang Luo 2021-10-14 16:40:19 +08:00
  • 5016ee3c95
    Give an informative message when users provide an unsupported decoding method (#77) Fangjun Kuang 2021-10-14 16:20:35 +08:00
  • 02721a79ec Give an informative message when users provide an unsupported decoding method. Fangjun Kuang 2021-10-14 15:47:29 +08:00
  • 39bc8cae94
    Add ctc decoding to pretrained.py on conformer_ctc (#75) Mingshuang Luo 2021-10-13 12:20:16 +08:00
  • 4790668da4 Update the running logs Mingshuang Luo 2021-10-13 11:37:37 +08:00
  • e035d2749c Update conformer_ctc.rst Mingshuang Luo 2021-10-13 11:05:01 +08:00
  • c8afe05e73 fix the style issue Mingshuang Luo 2021-10-13 10:27:16 +08:00
  • 33b88eee2b Update pretrained.py Mingshuang Luo 2021-10-13 10:23:27 +08:00
  • 524afc02ba update ctc-decoding for pretrained.py on conformer_ctc Mingshuang Luo 2021-10-13 00:52:40 +08:00
  • 391432b356
    Update train.py ("10"--->"params.log_interval") (#76) Mingshuang Luo 2021-10-12 21:30:31 +08:00
  • e5feabb232
    Update train.py Mingshuang Luo 2021-10-12 20:18:15 +08:00
  • 906d0ad6f8
    Update train.py Mingshuang Luo 2021-10-12 20:16:51 +08:00
  • 18d7dd2696
    Update train.py Mingshuang Luo 2021-10-12 20:15:49 +08:00
  • 597c5efdb1
    Use LossRecord to record and print the loss for the training process (#62) Mingshuang Luo 2021-10-12 15:58:03 +08:00
  • beb54ddb61
    Support torch script. (#65) Fangjun Kuang 2021-10-12 14:55:05 +08:00
  • 9d3775f721 Add documentation about how to deploy a trained model. Fangjun Kuang 2021-10-12 14:42:28 +08:00
  • 7fd9d291f3 update pretrained.py and conformer_ctc.rst Mingshuang Luo 2021-10-12 11:36:04 +08:00
  • 26d0df55de Merge remote-tracking branch 'dan/master' into jit Fangjun Kuang 2021-10-12 11:05:39 +08:00
  • 76a7bf13b2 Fix style issues. Fangjun Kuang 2021-10-12 11:05:03 +08:00
  • 6bc027ee8a Add ctc-decoding to pretrained.py Mingshuang Luo 2021-10-12 11:00:08 +08:00