Fangjun Kuang
fba5e67d5e
Fix CI tests. ( #1974 )
...
- Introduce unified AMP helpers (create_grad_scaler, torch_autocast) to handle
deprecations in PyTorch ≥2.3.0
- Replace direct uses of torch.cuda.amp.GradScaler and torch.cuda.amp.autocast
with the new utilities across all training and inference scripts
- Update all torch.load calls to include weights_only=False for compatibility with
newer PyTorch versions
2025-07-01 13:47:55 +08:00
Fangjun Kuang
d4d4f281ec
Revert "Replace deprecated pytorch methods ( #1814 )" ( #1841 )
...
This reverts commit 3e4da5f78160d3dba3bdf97968bd7ceb8c11631f.
2024-12-18 16:49:57 +08:00
Li Peng
3e4da5f781
Replace deprecated pytorch methods ( #1814 )
...
* Replace deprecated pytorch methods
- torch.cuda.amp.GradScaler(...) => torch.amp.GradScaler("cuda", ...)
- torch.cuda.amp.autocast(...) => torch.amp.autocast("cuda", ...)
* Replace `with autocast(...)` with `with autocast("cuda", ...)`
Co-authored-by: Li Peng <lipeng@unisound.ai>
2024-12-16 10:24:16 +08:00
zr_jin
242002e0bd
Strengthened style constraints ( #1527 )
2024-03-04 23:28:04 +08:00
zr_jin
5445ea6df6
Use shuffled LibriSpeech cuts instead ( #1450 )
...
* use shuffled LibriSpeech cuts instead
* leave the old code in comments for reference
2024-01-08 15:09:21 +08:00
Fangjun Kuang
8136ad775b
Use high_freq -400 in computing fbank features. ( #1447 )
...
See also https://github.com/k2-fsa/sherpa-onnx/issues/514
2024-01-04 13:59:32 +08:00
zr_jin
a81396b482
Use tokens.txt to replace bpe.model ( #1162 )
2023-08-12 16:53:59 +08:00
Fangjun Kuang
f5de2e90c6
Fix style issues. ( #937 )
2023-03-08 22:56:04 +08:00
pehonnet
07243d136a
remove key from result filename ( #936 )
...
Co-authored-by: pe-honnet <pe.honnet@telepathy.ai>
2023-03-08 21:06:07 +08:00
Zengwei Yao
4e832fa6b0
fix reduction conformer_ctc3/train.py ( #908 )
2023-02-14 20:45:38 +08:00
Zengwei Yao
25ee50e27c
add ctc-greedy-search with timestamps ( #905 )
2023-02-13 19:45:09 +08:00
Zengwei Yao
af735eb75b
Get alignments using lhotse workflows align-with-torchaudio ( #888 )
...
* add lhotse workflow align-with-torchaudio
* modify related decode.py files
2023-02-08 21:54:35 +08:00
Zengwei Yao
d12e6f098c
Get (start, end) timestamps for CTC models ( #876 )
...
* parse timestamps and texts for BPE-based models
* parse timestamps (frame indexes) and texts for other cases
* add test functions
* add parse_fsa_timestamps_and_texts function, test in conformer_ctc3/decode.py
* calculate symbol delay for (start, end) timestamps
2023-02-07 21:43:16 +08:00
Zengwei Yao
5a05b95730
add params.hlg_scale ( #880 )
2023-02-06 23:21:46 +08:00
Zengwei Yao
b25c234c51
Add Zipformer-MMI ( #746 )
...
* Minor fix to conformer-mmi
* Minor fixes
* Fix decode.py
* add training files
* train with ctc warmup
* add pruned_transducer_stateless7_mmi
* add zipformer_mmi/mmi_decode.py, using HP as decoding graph
* add mmi_decode.py
* remove pruned_transducer_stateless7_mmi
* rename zipformer_mmi/train_with_ctc.py as zipformer_mmi/train.py
* remove unused method
* rename mmi_decode.py
* add export.py pretrained.py jit_pretrained.py ...
* add RESULTS.md
* add CI test
* add docs
* add README.md
Co-authored-by: pkufool <wkang.pku@gmail.com>
2022-12-11 21:30:39 +08:00
Wei Kang
c25c8c6ad1
Add need_repeat_flag in phone based ctc graph compiler ( #727 )
...
* Fix is_repeat_token in icefall
* Fix phone based recipe
* Update egs/librispeech/ASR/conformer_ctc3/train.py
Co-authored-by: Fangjun Kuang <csukuangfj@gmail.com>
* Fix black
Co-authored-by: Fangjun Kuang <csukuangfj@gmail.com>
2022-12-04 17:20:17 +08:00
Zengwei Yao
8eb4b9d96d
Combining rnnt loss and k2-ctc loss for Dan's Zipformer ( #683 )
...
* init files
* add ctc as auxiliary loss and ctc_decode.py
* tuning the scalar of HLG score for 1best, nbest and nbest-oracle
* rename to pruned_transducer_stateless7_ctc
* fix doc
* fix bug, recover the hlg scores
* modify ctc_decode.py, move out the hlg scale
* fix hlg_scale
* add export.py and pretrained.py, and so on
* upload files, update README.md and RESULTS.md
* add CI test
2022-12-03 19:01:10 +08:00
Zengwei Yao
ece728d895
Apply delay penalty on k2 ctc loss ( #669 )
...
* add init files
* fix bug, apply delay penalty
* fix decoding code and getting timestamps
* add option applying delay penalty on ctc log-prob
* fix bug of streaming decoding
* minor change for bpe-based case
* add test_model.py
* add README.md
* add CI
2022-11-28 22:34:02 +08:00
