anodyine/vllm-omni
1
0 Fork 0
Code Issues 4 Pull Requests 2 Actions Packages Projects Releases Wiki Activity
1,369 Commits 235 Branches 31 Tags
3bd77d3f8f2a14e1a70cddaebf5f980154b52f0e
Commit Graph
218 Commits
Author SHA1 Message Date
Yueqian Lin 3bd77d3f8f Merge branch 'main' into feat/moss-tts-nano 
Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
 2026-04-21 17:50:21 -04:00
Nick Cao 339a6da28c [Bugfix] treewide: drop references to librosa (#2996) 
Signed-off-by: Nick Cao <ncao@redhat.com>
 2026-04-21 16:04:15 +00:00
Yueqian LinandSYLAR 3f504b45d2 [Config Refactor] Migrate 5 TTS models (VoxCPM2 / CosyVoice3 / MiMo Audio / Voxtral TTS / Fish Speech S2 Pro) to Pipeline + Deploy schema (#2958) 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
 2026-04-21 21:17:53 +08:00
Haco ad7c966b1d [BugFix] add missing subtalker sampling config to Qwen3-TTS deploy YAML (#2940) 
Signed-off-by: xiaohajiayou <923390377@qq.com>
 2026-04-21 14:47:22 +08:00
Jared WenandHongsheng Liu 7e28eda929 [Bugfix] Fix GLM-Image output dimensions and image edit pipeline (#2320) 
Signed-off-by: JaredforReal <w13431838023@gmail.com>
Signed-off-by: Jared Wen <w13431838023@gmail.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
 2026-04-20 19:24:06 +08:00
SYLAR cd384d95f3 [Config Refactor 2.5/N] Centralize pipeline registry (#2915) 
Signed-off-by: lishunyang <lishunyang12@163.com>
 2026-04-20 01:18:03 +08:00
Sy03 26edc7f2fb [Bugfix][VoxCPM2]: Fix vectorized_gather OOB under concurrent prefill+decode batches (#2903) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-19 20:39:14 +12:00
SYLARreidliu41xiaohajiayouAlex BrooksHongsheng Liu
 		a3903810fe [Config Refactor][2/N] Pipeline + Deploy Config Schema (#2383) 
Signed-off-by: lishunyang <lishunyang12@163.com>
Signed-off-by: reidliu41 <reid201711@gmail.com>
Signed-off-by: Alex Brooks <albrooks@redhat.com>
Co-authored-by: reidliu41 <reid201711@gmail.com>
Co-authored-by: xiaohajiayou <75477391+xiaohajiayou@users.noreply.github.com>
Co-authored-by: Alex Brooks <albrooks@redhat.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
 2026-04-19 14:36:00 +12:00
Sy03 a683b1dd75 [Bugfix][VoxCPM2] Fix voice-clone decode loop by padding prefill prompt (#2894) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-18 14:04:44 +00:00
ChenWenjing 4124a1f858 [Feature]Support audio streaming input and output-phase2 (#2581) 
Signed-off-by: CHEN <116010019@link.cuhk.edu.cn>
 2026-04-18 15:42:41 +08:00
Nick Cao d2c23d7bd0 [Refactor] Remove resampy dependency (#2891) 
Signed-off-by: Nick Cao <ncao@redhat.com>
 2026-04-18 03:20:36 +00:00
Nick CaoandClaude f2edb81a59 [Cleanup] Remove dead runtime.defaults config parameters (#2343) 
Signed-off-by: Nick Cao <ncao@redhat.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-04-17 23:36:41 +08:00
Yuanheng Zhao c0ccbb872f [Model] Add Ming-flash-omni-2.0 Thinker Stage (#1822) 
Signed-off-by: yuanheng <jonathan.zhaoyh@gmail.com>
Signed-off-by: Yuanheng Zhao <jonathan.zhaoyh@gmail.com>
 2026-04-17 19:10:02 +08:00
Bingyu (Spencer) Liu a5a4998ece [Feature] Support Prefill-Decode disaggregation via vLLM KV transfer (#2220) 
Signed-off-by: LiuBingyu <liubingyu62@gmail.com>
 2026-04-17 18:43:17 +08:00
Nick CaoandClaude 1e3bb36521 [Refactor] Remove sox from dependencies (#2745) 
Signed-off-by: Nick Cao <ncao@redhat.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-04-17 01:24:47 +00:00
TaffyOfficialTaffyOfficialClaude Opus 4.6
 		c3ca5daafb Feat/Add HunyuanImage-3.0-Instruct ar part support: (#2713) 
Signed-off-by: TaffyOfficial <2324465096@qq.com>
Co-authored-by: TaffyOfficial <2324465096@qq.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
 2026-04-16 23:15:25 +08:00
Sy03 7d64a7c996 [BugFix][VoxCPM2]: split multichar Chinese tokens to match training tokenization (#2832) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-16 10:07:48 -05:00
Sy03 322620fd57 [Fix][Fish Speech] Remove redundant get_vocab() in control token encoding (#2842) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-16 12:47:39 +00:00
Juan Pablo ZuluagaandHongsheng Liu 82f8c93343 [Feat][Qwen3-Omni] Shared code predictor module for Qwen3-TTS and Qwen3-Omni (#2375) 
Signed-off-by: JuanPZuluaga <juanz9312@gmail.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
 2026-04-15 15:14:57 +08:00
IsleOfDawnlightCeleste-jqlyj-jjjYueqian Lin
 		4bf4c63147 Add voxcpm model support. (#2467) 
Signed-off-by: Celeste-jq <591998922@qq.com>
Signed-off-by: lyj-jjj <liuyingjun5@huawei.com>
Signed-off-by: IsleOfDawnlight <stellamou@qq.com>
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Celeste-jq <591998922@qq.com>
Co-authored-by: lyj-jjj <liuyingjun5@huawei.com>
Co-authored-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-15 15:04:58 +08:00
Sy03 38d5f2d530 [Perf] VoxCPM2: Speedup by manual CUDA Graph capture for scaffold/residual forward (#2803) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-15 14:09:22 +08:00
Yueqian Lin 02e5dc747d [Perf] VoxCPM2: streaming VAE + compile optimization (45% RTF reduction) (#2758) 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-15 09:58:25 +08:00
Yueqian Lin 033e4041be fix(moss-tts-nano): remove unused voice variable (ruff F841) 
The upstream inference_stream() API has no voice/spk parameter;
voice preset selection is not yet wired into the call. Remove the
dead assignment to silence ruff F841.

Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-14 13:11:43 -04:00
Yueqian Lin 38aaaebeda feat(moss-tts-nano): enable progressive streaming via AR runner 
Switch from GPUGenerationWorker to AR worker with OmniARScheduler,
using the VoxCPM-style generator pattern for streaming:

- inference_stream() stored per-request in _stream_gens dict
- Each forward() call yields one audio chunk via next(generator)
- compute_logits() emits EOS only when last chunk is yielded
- AR scheduler loops until EOS, enabling progressive audio output

TTFP reduced from ~3.1s to ~0.11s (30x improvement).

Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-14 13:06:44 -04:00
Yueqian Lin fc81b8504a feat(moss-tts-nano): add online serving, Gradio demo, and fix /v1/audio/speech integration 
- Register moss_tts_nano as TTS model stage in serving_speech.py
  (model detection, validation, prompt building)
- Fix registry mod_relname: moss_tts_nano -> modeling_moss_tts_nano
- Fix stage config: is_comprehension=true (required for generate task)
- Fix default mode: voice_clone -> continuation (built-in presets)
- Add compute_logits stub for VllmModelForTextGeneration protocol
- Remove unused _sentinel nn.Parameter
- Add Gradio demo with AudioWorklet streaming player (48kHz stereo)
- Add online/offline serving docs and launch scripts

TODO: Single-stage generation models don't support true streaming
(progressive audio chunks). Current TTFP = full generation time.
Needs multi-step scheduling support in GPUGenerationModelRunner.

Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-14 13:06:44 -04:00
Yueqian Lin 571c8d4235 fix(moss-tts-nano): also patch torchaudio.save for soundfile fallback 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-14 13:06:12 -04:00
Yueqian Lin 0b8190d07f fix(moss-tts-nano): patch torchaudio.load to use soundfile when torchcodec unavailable 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-14 13:06:12 -04:00
Yueqian Lin da6b1b5741 feat(moss-tts-nano): add single-stage MOSS-TTS-Nano TTS support 
Integrates OpenMOSS-Team/MOSS-TTS-Nano (0.1B) as a single
GPUGenerationWorker stage.  Both the AR LM and MOSS-Audio-Tokenizer-Nano
codec run inside MossTTSNanoForGeneration.forward(), removing the need for
an inter-stage connector.

Key design choices:
- Weights loaded in load_weights() not __init__ (avoids pre-CUDA alloc)
- trust_remote_code delegates to upstream HF model classes
- codec path read from config.audio_tokenizer_pretrained_name_or_path
- inference_stream() collects progressive audio chunks for low latency
- 48 kHz stereo output; voice clone + continuation modes supported


Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-14 13:06:12 -04:00
Yiyang "Ian" LiuandSYLAR bc4a659f03 [Bugfix] Make mrope kwargs optional in HunyuanImage3 get_mrope_input_positions (#2654) 
Signed-off-by: Yiyang Liu <yiyangliu@microsoft.com>
Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
 2026-04-14 14:41:28 +00:00
Yiyang "Ian" LiuandCopilot 53a9cf49a6 fix: do not apply FP8 quant config to vision/audio encoders for pre-quantized checkpoints (#2702) 
Signed-off-by: Yiyang Liu <37043548+ianliuy@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
 2026-04-14 13:52:32 +00:00
Chen-Yo Sun dd1389173b [Voxtral TTS] Fix Voxtral TTS input with text and ref_audio (#2750) 
Signed-off-by: Chen-Yo Sun <chenyo.sun@mistral.ai>
 2026-04-13 22:37:45 +00:00
Sy03Yueqian LinYueqian Lin
 		14f7910900 [Perf]: Speedup VoxCPM2 TTS performance and Support PagedAttention (#2690) 
Signed-off-by: Sy03 <1370724210@qq.com>
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
 2026-04-14 09:16:47 +12:00
Chen-Yo Sun c9e2e3e8d7 [Voxtral TTS] Correct decode steps param in Voxtral TTS (#2524) 
Signed-off-by: Chen-Yo Sun <chenyo.sun@mistral.ai>
 2026-04-13 17:53:35 +00:00
Yueqian Lin d9e745ce2c [Fix] VoxCPM2: support raw audio for voice cloning via OpenAI API (#2720) 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-04-13 06:30:56 +00:00
Sy03 cb4d13a658 [Perf][Fish Speech] Enable CUDA Graph capture for Fast AR code predictor (#2520) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-13 04:53:35 +00:00
NATURE e12250119b [Bugfix] Fix Bagel online mode for 1. Hang after several requests 2. Non-deterministic image quality regression. (#2458) 
Signed-off-by: natureofnature <wzliu@connect.hku.hk>
 2026-04-13 03:36:20 +00:00
ReidandHongsheng Liu 38dfe56fdf [Bugfix] Validate speaker in chat endpoint and fix case-insensitive lookup (#2407) 
Signed-off-by: reidliu41 <reid201711@gmail.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
 2026-04-12 00:14:16 -04:00
Juan Pablo Zuluaga f7e8df9dbd [TTS][OmniVoice] Add voice cloning support for OmniVoice TTS (#2676) 
Signed-off-by: JuanPZuluaga <juanz9312@gmail.com>
 2026-04-11 12:14:03 -04:00
Nick Cao eccee21c6c [Perf] Use global CUDA graph pool for MiMo Audio (#2657) 
Signed-off-by: Nick Cao <ncao@redhat.com>
 2026-04-11 13:45:26 +00:00
teithandYueqian Lin 001f2e32e2 [BUG FIX]: prevent EngineCore crash when Qwen TTS Base task is missing ref_text (#2203) 
Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
 2026-04-11 02:46:38 +00:00
Yueqian LinandSYLAR a41174e083 [Model] VoxCPM2 native AR TTS support (#2658) 
Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
Co-authored-by: SYLAR <lishunyang12@users.noreply.github.com>
 2026-04-11 10:21:26 +08:00
Nick CaoandClaude 2bc183f6f0 [Refactor] Remove dependency on librosa (#2273) 
Signed-off-by: Nick Cao <ncao@redhat.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-04-11 02:11:31 +08:00
Yueqian Lin 94232436bb [Feat][FishSpeech] Cache DAC-encoded ref audio for voice cloning (#2609) 
Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
 2026-04-10 10:34:03 +08:00
Sy03 694be6f2e1 [Bugfix] Fix Fish Speech voice clone FileNotFoundError on multi-GPU (#2606) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-09 20:34:34 +00:00
Dogeun KimYejoon Leeaidas
 		e2b0ee4b87 [Model] Add Dynin-omni model in vllm-omni (#1759) 
Signed-off-by: kdg6245@snu.ac.kr <kdg6245@snu.ac.kr>
Signed-off-by: Yejoon Lee (IPAI) <leeyejoon@snu.ac.kr>
Signed-off-by: aidas (arpa-kt) <aidaslab@gmail.com>
Signed-off-by: Dogeun Kim <82812668+DOGEUNNKIM@users.noreply.github.com>
Co-authored-by: Yejoon Lee (IPAI) <leeyejoon@snu.ac.kr>
Co-authored-by: aidas (arpa-kt) <aidaslab@gmail.com>
 2026-04-09 16:37:34 +08:00
Nick CaoandClaude 149b9f179a [Perf][Qwen3-TTS][Voxtral-TTS] Share CUDA graph memory pool across decoder capture sizes (#2386) 
Signed-off-by: Nick Cao <ncao@redhat.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-04-08 20:40:17 -04:00
Zhengyuan Su (苏政渊)andClaude 486d77d797 [Feature] Enable LoRA adapter injection for BAGEL (#2490) 
Signed-off-by: Zhengyuan Su <su.zhengyuan@u.nus.edu>
Signed-off-by: Claude <noreply@anthropic.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-04-06 19:30:52 +08:00
Will.houwillamhouClaudeHappy
 		e23b2634d1 fix(qwen3_tts): align code predictor buffer dtype with model parameters (#2470) 
Signed-off-by: willamhou <willamhou@ceresman.com>
Co-authored-by: willamhou <willamhou@ceresman.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Happy <yesreply@happy.engineering>
 2026-04-05 20:54:15 +00:00
Sy03 8b57c6205e [Perf][Fish Speech] Free unused DAC codec components to save VRAM (#2430) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-04-05 20:49:18 +00:00
Sy03andYueqian Lin f6cfacdd16 [Perf][Qwen3-TTS] Free unused decoder in Talker SpeechTokenizer to VRAM (#2429) 
Signed-off-by: Sy03 <1370724210@qq.com>
Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
 2026-04-05 19:04:37 +00:00