b1584aec7c
fix: stabilize CPU SDPA mask broadcasting
...
Use an explicit broadcastable attention mask shape during MiniCPM incremental decoding so CPU runtimes avoid a PyTorch SDPA dimension error without changing attention semantics.
Made-with: Cursor
2026-04-13 15:38:53 +08:00
5510503182
Merge pull request #246 from sharziki/fix/unclosed-file-handles
...
fix: close file handles in from_local() config loading
2026-04-11 13:10:04 +08:00
fb46aad9a5
fix: close file handles in from_local() config loading
...
Use context managers when reading config.json in VoxCPMModel.from_local()
and VoxCPM2Model.from_local() to prevent file descriptor leaks. Also add
explicit encoding="utf-8" to avoid locale-dependent decode errors.
Closes #235
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com >
2026-04-11 00:01:14 -04:00
e4e049624c
update finetuning pipeline and runtime device handling
...
Support optional ref_audio samples in finetuning and make runtime device selection explicit while keeping auto fallback behavior consistent. Also ignore the local app override file to avoid accidental commits.
Made-with: Cursor
2026-04-11 11:08:50 +08:00
abf01b9bf3
Merge pull request #229 from kuishou68/fix/issue-228-validate-text-type-order
...
fix: correct isinstance/strip order in _generate() to prevent AttributeError on non-string input
2026-04-10 10:30:15 +08:00
4f4a5b9f6c
fix: correct type-check order in _generate() to prevent AttributeError on non-string input
...
The previous guard `not text.strip() or not isinstance(text, str)` called
.strip() before verifying that text is actually a string, causing an
AttributeError (e.g. for int input) instead of the intended ValueError.
Swap operand order so isinstance check short-circuits first.
Closes #228
2026-04-09 16:13:40 +00:00
79c0cf68dd
chore: remove accidentally committed app_local.py
...
Made-with: Cursor
2026-04-09 16:05:18 +08:00
75cfa3e9b8
fix: use uncompiled feat_encoder for prefill to prevent CUDA Graph dynamic shape accumulation ( #209 )
2026-04-09 16:00:17 +08:00
5611bd08a0
optim app.py
2026-04-09 00:30:19 +08:00
364eff6840
update readme: python version
2026-04-08 23:07:38 +08:00
6d10932b09
update readme
2026-04-08 18:48:58 +08:00
68af4fe502
fix: ft log and setting
2026-04-08 18:15:17 +08:00
ee3649c1b3
fix: streaming decode
2026-04-08 17:25:54 +08:00
82d77d445c
fix: decode chunksize for audiovae_v2
2026-04-08 16:31:36 +08:00
8f95d13073
update readme: 30-language asr result on internal benchmark
2026-04-08 15:36:56 +08:00
df38f0a167
update readme for modelscope download
2026-04-08 11:29:19 +08:00
9adfaf6996
update demo for zh
2026-04-08 00:15:16 +08:00
46cfce0c97
fix VoxCPM2 training sample_rate: 48000 -> 16000 (match AudioVAE encoder)
...
Made-with: Cursor
2026-04-07 22:59:18 +08:00
da700f264e
update ZH readme
2026-04-07 18:04:56 +08:00
9da570d409
remove wechat link
2026-04-07 15:29:12 +08:00
9374524c47
update readme
2026-04-06 23:01:16 +08:00
ec6d30e996
update readme
2026-04-06 22:56:06 +08:00
a010d621ff
update readme
...
Made-with: Cursor
2026-04-06 22:09:24 +08:00
3f005b0dbd
Enhance README formatting and community section for better visibility
2026-04-06 19:50:29 +08:00
039c6e9f92
update
2026-04-06 17:15:10 +08:00
5734ab36b6
Update README
2026-04-06 16:24:12 +08:00
746631c38d
update
2026-04-06 16:10:50 +08:00
07b8b5c01f
update readme
2026-04-06 15:53:58 +08:00
f738cc9946
update
2026-04-03 18:46:29 +08:00
0c2cf23617
Update app.py UI, adjust streaming_prefix_len, remove legacy docs
...
- Refine app.py: Ultimate Cloning naming, NFE slider, i18n polish
- Change streaming_prefix_len default from 3 to 4 for smoother decoding
- Remove legacy docs/ directory (migrated to ReadTheDocs)
Made-with: Cursor
2026-04-03 18:42:41 +08:00
b823d8107c
Merge branch 'dev_2.0' of https://github.com/OpenBMB/VoxCPM into dev_2.0
2026-04-03 17:44:46 +08:00
a87739426f
add voxcpm2 finetune conf
2026-04-03 14:23:15 +08:00
12c2b8ff98
update readme
2026-04-02 21:01:23 +08:00
30c300cfe8
adjust default cfg range
2026-04-02 18:14:35 +08:00
addee2c550
surport voxcpm2 cli
2026-04-01 21:15:55 +08:00
42c428164c
feat: add no_rope support for residual LM and fix streaming continuation decoding
...
- Add `residual_lm_no_rope` config option in VoxCPMConfig and propagate to MiniCPMModel
- Add `no_rope` field to MiniCPM4Config; make RoPE embedding optional in MiniCPMModel and MiniCPMAttention
- Add `streaming_prefix_len` parameter to generation interface
- Fix non-streaming audio decode in continuation mode to trim leading prefix patches consistently
- Refactor streaming prefix context preparation: distinguish continuation vs. zero-shot via feat_mask trailing bit instead of audio_mask sum
Made-with: Cursor
2026-03-31 17:07:33 +08:00
d9cf376e16
update voxcpm2
2026-03-31 11:50:37 +08:00
23ed7ffeee
fix: fix some bugs in resuming multi-GPU training
2026-03-13 18:43:07 +08:00
7823e14b82
Merge pull request #188 from haosenwang1018/fix/bare-excepts
...
fix: use specific exceptions instead of bare except
2026-03-03 11:49:00 +08:00
8df79de636
fix: use specific exceptions instead of bare except
...
- lora_ft_webui.py: except (JSONDecodeError, OSError) for config file
- voxcpm.py: except ImportError for triton availability check
2026-02-24 22:19:45 +00:00
acaadb19e9
Merge pull request #186 from symhsym/patch-1
...
Update train_voxcpm_finetune.py
2026-02-11 18:05:39 +08:00
07e526a231
Update train_voxcpm_finetune.py
...
修改了issue#185中提到的问题，在训练时进行validate会对原模型执行to(torch.bfloat16)然后to(torch.float32)的操作，这样可能导致模型数值浮动，因此这个修改让validate步骤保留原模型数值
2026-02-11 11:17:47 +08:00
7aadc6c94e
Merge pull request #161 from s3ldc/cli-arg-validation
...
Improve CLI argument validation and help text
2026-01-24 13:06:30 +08:00
8f3a91cac8
cli: improve argument validation and help text for VoxCPM CLI
2026-01-20 14:33:58 +05:30
e72fb42c38
Merge pull request #147 from zanellig/main
...
Fix README's feature checkboxes
2026-01-19 12:41:22 +08:00
6dd63a534f
fix: feature checkboxes
2026-01-18 18:03:24 -03:00
79e75f259e
Fix: optimize save ckpt function
2026-01-16 16:22:34 +08:00
e8dd956fc2
Print all log messages to stderr instead of stdout
2026-01-12 15:30:45 +08:00
db75a7269b
Merge pull request #141 from vytskalt/main
...
Print debug messages to stderr instead of stdout
2026-01-12 15:06:51 +08:00
f2e203d5e2
print debug messages to stderr instead of stdout
2026-01-09 20:05:52 +02:00
刘鑫