anodyine/vllm-omni
1
0 Fork 0
Code Issues 4 Pull Requests 2 Actions Packages Projects Releases Wiki Activity
3,096 Commits 235 Branches 31 Tags
2ff0a783d403c60ea585dcec5179b9995c773ece
Commit Graph
3096 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
NATURE 5b002727b0 [Core] Split Omni connector model runner mixin (#6903) 
Signed-off-by: natureofnature <wzliu@connect.hku.hk>
 2026-09-09 13:49:43 +08:00
andyluo7 e0d0879794 [Bugfix][Magi2] Allow import without an active Triton driver (#7239) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
 2026-09-08 22:48:08 -07:00
wangyu 09b720b0d2 [CI] Skip Qwen3-Omni Server VAD multi-turn realtime test (#7279) (#7314) 
Signed-off-by: wangyu <410167048@qq.com>
 2026-09-09 12:23:29 +08:00
xutianle 968f3e379c [Bugfix][CosyVoice3] Resolve hash snapshot pipeline (#6896) 
Signed-off-by: xutianle <xutianle@fudan.edu.cn>
 2026-09-08 20:33:38 -07:00
Allen WuandTRAE CLI f2e30f46b6 [Bugfix][TTS] Isolate shared Higgs v3 reference encode from request cancellation (#7076) 
Signed-off-by: Allen Wu <allenwu2795@gmail.com>
Co-authored-by: TRAE CLI <traecli@bytedance.com>
 2026-09-08 20:32:05 -07:00
Jim Ban 358b7d79c1 [Bugfix][Frontend] Enforce image pixel limits for video input references (#6963) 
Signed-off-by: BANANASJIM <bananasjim1@gmail.com>
 2026-09-08 20:31:54 -07:00
Ziming Wang 43b697541a [Bugfix][Frontend] Clear the P0 multimodal cache through the renderer (#7003) 
Signed-off-by: ZenAlexa <zimingwang945@gmail.com>
 2026-09-08 20:31:42 -07:00
tzhouamandClaude Fable 5.1 285db2fa3c edge results: WP1 edge-profile stream latency (0.6B/1.7B) + init timelines 
1.7B: TTFA p50 43.9 ms, playback_start 43.9 ms, stall 0 (baseline 32.0 / 116.2 / 84.7).
0.6B: TTFA p50 46.1 ms, playback_start 46.1 ms, stall 0 (baseline 27.8 / 95.4 / 67.3).
RTF unchanged (~0.1). Measured on L20X via gpu run --gpus 1 (canhazgpu).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 11:21:48 +08:00
tzhouamandClaude Fable 5.1 8e912db6c8 [Edge][WP6] Slim the AR/TTS import path and add an edge dependency set 
* diffusers is no longer imported by vllm_omni.diffusion.data (annotation only);
  vllm_omni.diffusion.io_support (diffusion registry + distributed state) is
  imported lazily by config_factory and async_omni_engine, so constructing a
  Qwen3-TTS pipeline config / AsyncOmni no longer loads diffusers,
  x_transformers or whisper (guarded by tests/config/test_import_guard.py).
* requirements/common.txt = core.txt (AR/speech runtime) + media.txt
  (diffusion/video/extra audio); VLLM_OMNI_EDGE_BUILD=1 makes setup.py use
  requirements/<device>-edge.txt (core only); `edge` extra adds websockets and
  sounddevice for on-device clients.

Measured on this host (CUDA venv): `import vllm` 7.0-7.9 s, `import vllm_omni`
adds ~20 s of which ~17 s is transformers' own import (transformers.utils.import_utils)
pulled in by vllm_omni.patch / vllm_omni.version; that floor is not
vLLM-Omni's and is unchanged. Guard test: no diffusion module loaded on the TTS path.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 11:14:50 +08:00
tzhouamandClaude Fable 5.1 49a839ecf9 [Edge][WP5] Add codec-token streaming mode with on-device decoding 
* qwen3_tts_talker_only pipeline (single terminal LLM_AR stage,
  final_output_type "latent") + vllm_omni/deploy/qwen3_tts_talker_only.yaml;
  a terminal talker publishes its per-step codes.audio rows to the client
  (omni_client_multimodal_output_keys).
* /v1/audio/speech/stream: session.config.output_mode="codec_tokens" streams
  codec.start (JSON) + binary frames (<IHB seq/n_frames/flags + int16 rows,
  vllm_omni/entrypoints/openai/codec_stream.py) + codec.done; new input.cancel
  message aborts an in-flight generation for both codec and PCM sessions
  (single reader task + session inbox replaces direct socket reads); optional
  session.config.seed; audio.done gains "cancelled" only when it happened.
* serving_speech._generate_codec_chunks aligns the cumulative code tensor to
  output token ids (drops prefill placeholder/EOS rows) and emits deltas.
* TTS adapters: supports_codec_stream / codec_stream_spec (Qwen3-TTS: 16
  codebooks, 2048 ids, 12.5 Hz, 24 kHz, decoder qwen3_tts_tokenizer_v2_12hz).
* examples/online_serving/text_to_speech/qwen3_tts/edge_decoder_client.py
  decodes locally with Qwen3TTSTokenizerV2Decoder (ramp schedule, 72-frame
  context, cancel/reset), codec_parity_check.py and
  benchmarks/tts/codec_stream_e2e.py compare against the two-stage server.
* docs/serving/codec_token_streaming.md; CPU unit tests for the wire format,
  alignment, session config and handler (frames, unsupported adapter, cancel);
  L3 e2e tests/e2e/online_serving/test_qwen3_tts_codec_stream.py.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 11:10:13 +08:00
tzhouamandClaude Fable 5.1 35e43b9983 [Edge][WP3] Add CPU OmniPlatform (AR + generation stages on vLLM's CPU backend) 
* vllm_omni/platforms/cpu/platform.py: CPUOmniPlatform(OmniPlatform, CpuPlatform)
  selected when VLLM_TARGET_DEVICE=cpu or a +cpu vLLM wheel is installed
  (builtin plugin "cpu"; explicit CPU target short-circuits accelerator probes
  like vLLM does). NUMA-based memory queries, one logical device, no graph
  wrapper (sentinel type for isinstance checks), inductor opt-in via
  VLLM_OMNI_CPU_INDUCTOR=1, diffusion stages unsupported.
* vllm_omni/platforms/cpu/worker: CPUARWorker / CPUGenerationWorker compose
  vLLM's CPUWorker with the Omni mixin (XPU pattern); CPUARModelRunner /
  CPUGenerationModelRunner compose the Omni runners with CPUModelRunner
  (no graphs, no async omni output; sleep/wake return error ACKs).
* patch.py skips the CuMem patch on CPU-only torch; Qwen3-TTS speaker encoder
  and prompt-embeds dtype follow the model dtype (float32/float16 on CPU).
* vllm_omni/deploy/qwen3_tts.yaml: `platforms: cpu:` overlay (eager, no async
  scheduling, absolute kv_cache_memory_bytes, disjoint OMP thread ranges).
* requirements/cpu.txt (+psutil, intel-openmp), docs/getting_started/installation/cpu.md.

Measured (this x86 Xeon 8480C host, vllm 0.28.0+cpu, bf16, 12+4 threads):
Qwen3-TTS 0.6B CustomVoice end-to-end on CPU, 12 prompts: TTFA p50 292 ms,
RTF ~1.7 (25-frame chunk ~4.6 s), init 185 s. Result JSON + init timeline
committed under benchmarks/tts/edge_results/.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 11:02:14 +08:00
tzhouamandClaude Fable 5.1 55192de984 [Edge][WP4] Add hardware adaptation: probe, derived overrides, auto profile, calibration 
* vllm_omni/edge/hardware_probe.py: HardwareProfile from /proc, /sys and torch
  (ISA flags incl. AVX-512/AMX and NEON/dotprod/i8mm/SVE, physical cores,
  big.LITTLE clusters, NUMA, RAM, CUDA discrete vs Tegra unified memory, NPU
  via override), hardware classes jetson / cuda_discrete / arm64_cpu / x86_cpu /
  npu_phone; VLLM_OMNI_HW_PROFILE=<json> overrides the probe (emulation, tests).
* vllm_omni/edge/adapt.py: derive_overrides -> dtype, absolute KV budgets from
  available memory minus weights, max_num_seqs, per-stage disjoint CPU thread
  ranges on the big cluster, eager/graph switches for small or unified memory,
  adaptive-chunk parameters, timeouts from a load cost model.
* vllm_omni/edge/calibrate.py: chunk-schedule calibration from measured chunk
  arrivals with a per-hardware-fingerprint cache under VLLM_CACHE_ROOT.
* vllm_omni/edge/auto_profile.py + deploy_profile="auto": edge/<model>.yaml ->
  edge/hardware/<class>.yaml -> derived overrides -> cached calibration,
  materialized as a temp deploy YAML (platform overlays are superseded).
* benchmarks/tts/hw_emulation.py: emulated classes (ATEN_CPU_CAPABILITY,
  taskset, profile overrides) driving stream_latency_bench with auto profiles.
* CPU-only tests for probe snapshots, derivation rules, calibration and auto.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 11:01:40 +08:00
tzhouamandClaude Fable 5.1 073d31b9cd [Edge][WP2] Add initialization timeline and cold-start profiler 
* vllm_omni/engine/init_timeline.py: InitTimeline phase collector; engine
  phases (stages_init, g3_device_lock, g2_spawn/ready) and worker phases
  (proc_entry, engine_core_init, init_device, load_model/load_weights,
  talker_mtp_init, attn_prewarm, cudagraph_capture, talker_mtp_capture,
  code2wav_cudagraph, predictor_setup_compile/warmup/cudagraph) recorded via
  worker_phase/timed_phase; workers append JSONL under
  VLLM_OMNI_INIT_TIMELINE_DIR (inherited at spawn) and the engine merges them.
  Enabled by VLLM_OMNI_INIT_TIMELINE=<json> or log_stats; table logged at INFO,
  partial table on orchestrator startup timeout, suggested timeouts logged.
* omni_base.py: named deploy profiles default to stage_init_timeout=900 /
  init_timeout=1500 unless given explicitly.
* gpu_model_runner._init_talker_mtp tolerates max_cudagraph_capture_size=None;
  gpu_ar_model_runner.shutdown skips CUDA graph teardown on CPU.
* benchmarks/tts/init_profile.py: runs stream_latency_bench --init-only per
  configuration (default, cold_cache, eager_stage1, eager_both,
  parallel_stage_init, edge, ...) and tabulates phase durations.
* stream_latency_bench: --parallel-stage-init / --deploy-profile flags; the
  resource poller no longer shadows Thread._stop and skips GPU sampling on CPU.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 11:01:39 +08:00
tzhouamandClaude Fable 5.1 f804bbd679 [Edge][WP0] Record Qwen3-TTS streaming baselines on 1x L20X 
Default deploy config (1 -> 25 frame schedule), 12 prompts x 3 repeats,
warm-up excluded, one GPU via `gpu run`:
  1.7B CustomVoice: TTFA p50 32.0 ms, playback-start p50 116 ms (stall at
    TTFA 85 ms), RTF_total 0.086, chunk gap p50 163 ms, init 271 s (cold caches)
  0.6B CustomVoice: TTFA p50 27.8 ms, playback-start p50 95 ms (stall 67 ms),
    RTF_total 0.077, chunk gap p50 146 ms, init 152 s
Per-request chunk arrivals, memory and the full summaries are in the JSON
(force-added: *.json is gitignored repo-wide).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 10:49:37 +08:00
tzhouamandClaude Fable 5.1 b22dcc59f7 [Edge][WP1] Add edge deploy profile and deploy_profile selector 
* vllm_omni/deploy/edge/qwen3_tts.yaml: single-device Qwen3-TTS profile with a
  ramped codec chunk schedule (2->4->8->16->25 frames) that removes the ~80 ms
  playback stall of the default 1->25 schedule, decode batch 1, max_num_seqs 4.
* deploy_profile="edge" (Omni/AsyncOmni kwarg, --deploy-profile CLI, headless
  serve) resolves vllm_omni/deploy/<profile>/<model_type>.yaml; mutually
  exclusive with deploy_config; explicit profiles never fall back silently.
* tests/helpers/stage_config.py: modify_stage_config anchors a relative
  base_config before writing the temp YAML so overlay-derived configs resolve.
* async_omni_engine.py also carries the init-timeline hooks used by WP2.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 10:27:33 +08:00
tzhouamandClaude Fable 5.1 b70207f57f [Edge][WP0] Add offline streaming-latency benchmark for chunked TTS 
Adds benchmarks/tts/stream_latency_bench.py (AsyncOmni driver reusing the
Qwen3-TTS example's prompt builder; --model override, warm-up/repeat,
--init-only, per-chunk arrivals, peak RSS and per-process GPU memory) and
benchmarks/tts/stream_latency_metrics.py (TTFA, underrun-free playback start,
stall at TTFA, RTF streamed/total, chunk gaps, percentiles) with CPU unit
tests. Results go under benchmarks/tts/edge_results/stream_latency/.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-09 10:24:53 +08:00
Rakesh KariyaCursorClaude Opus 5
 		2bae550c04 [Frontend] Opt-in WebSocket TTS split_granularity and session seed (#7046) 
Signed-off-by: rk9595 <rakesh.kariya@somaiya.edu>
Signed-off-by: Rakesh Kariya <rakesh.kariya@somaiya.edu>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
 2026-09-08 19:15:59 -07:00
zhengjiaandCursor 81c1c460b3 [Bugfix][Frontend] Accept the msgpack-numpy package's numpy markers on the OpenPI endpoint (#6051) 
Signed-off-by: zjli2013 <leezhengjiang@126.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
 2026-09-08 19:15:48 -07:00
NancyFyongandNancyFyong e74fe3c6e8 [Model] Add Audio8 TTS Preview 0.6B (DualAR, 44.1 kHz codec) (#6157) 
Signed-off-by: NancyFyong <NancyFyong@users.noreply.github.com>
Co-authored-by: NancyFyong <NancyFyong@users.noreply.github.com>
 2026-09-08 19:00:25 -07:00
boatman cdf027874c [Feat][OmniVoice]Support Varlen Attn, Request-Batch and Step-Execution (#6408) 
Signed-off-by: boatman <109857087+sphinxkkkbc@users.noreply.github.com>
 2026-09-08 18:58:38 -07:00
eval d7678e2bb3 [Bugfix][MiniCPM-o] Fix the audio_embeds input path (#5730) 
Signed-off-by: eval-dev <0xe5bca0@gmail.com>
Signed-off-by: eval <74645252+eval-dev@users.noreply.github.com>
 2026-09-08 18:41:42 -07:00
Zhichao Zhang 5078c7ad66 [Bugfix][MiMo-Audio] Align code2wav decode with tokenizer device (#6539) 
Signed-off-by: chaosansui <zzc15560846421@163.com>
Signed-off-by: Zhichao Zhang <60429419+smartDream-chao@users.noreply.github.com>
 2026-09-08 18:38:11 -07:00
Hongsheng Liu f67cc52f74 [Doc] Add AI usage policy for contributions (#7305) 
Signed-off-by: hsliuustc0106 <liuhongsheng4@huawei.com>
 2026-09-09 09:17:24 +08:00
Anjie HouandClaude Opus 5 b6f4df320d [Refactor][Diffusion] Resolve offload topology through one plan resolver (#7209) 
Signed-off-by: specture724 <specture724@gmail.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
 2026-09-09 08:17:38 +08:00
Tianyao Wu b3dd45874a [Bugfix] Build the forced-aligner prompt without a chat template (word timestamps one bin late) (#7240) 
Signed-off-by: Tianyao Wu <rayroy31@gmail.com>
 2026-09-08 19:27:38 +00:00
Lei KeandKrystalRay 5e4baea64d [Bugfix][NPU] Limit MiniMax H3 modulation grid size (#6794) 
Signed-off-by: KrystalRay <keeleiray@gmail.com>
Co-authored-by: KrystalRay <keeleiray@gmail.com>
 2026-09-08 09:08:46 -07:00
Harry MellorHongsheng LiuClaude Opus 5
 		fdaf4e31bd Add vLLM-Omni library info to Hugging Face Hub requests (#5381) 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
 2026-09-08 10:08:38 -06:00
liangmenghuangandClaude Fable 5.1 7a80b79698 [Bugfix][Model] GR00T-N1.7: honor the per-request seed for flow-matching noise (#7253) 
Signed-off-by: liangmengh <liangmengh@nvidia.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-08 15:53:20 +00:00
andyluo7 de1b0708bd [CI][ROCm] Route LTX2 Ulysses parity to two-GPU lane (#7234) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
 2026-09-08 22:59:37 +08:00
Hongsheng Liu 7422a85cc7 [Bugfix] Skip HWR store-size scans when no limit is configured (#7131) 
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
 2026-09-08 22:54:49 +08:00
YancyandAsthenia a31690041d [Bugfix][Examples] Use --profiler-config flag in offline TTS examples (#6763) 
Signed-off-by: Asthenia <asthenia0412@gmail.com>
Co-authored-by: Asthenia <asthenia0412@gmail.com>
 2026-09-08 13:56:27 +00:00
Honghan Zhuandamy-why-3459 d59d166485 [Bugfix] Preserve additional_config for LLM stages (#7272) 
Signed-off-by: zhuhh97 <zhuhonghan@huawei.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
 2026-09-08 20:44:49 +08:00
Shuolei Wang 5378b77be4 [Model][JoyAI-VL-Interaction] Add native multi-stage pipeline (#5352) 
Signed-off-by: Shuolei Wang <shuoleiwang123@gmail.com>
 2026-09-08 05:07:08 -07:00
liangmenghuangandClaude Fable 5.1 d7740336ff [Bugfix][Model] GR00T-N1.7: honor processor_config image pipeline (use_albumentations / crop_fraction / letter_box_transform) to match Isaac-GR00T (#7083) 
Signed-off-by: liangmengh <liangmengh@nvidia.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
 2026-09-08 18:59:01 +08:00
LHXuuu c9c25b33c3 [Frontend] Add server-side VAD for turn-based omni models (#6618) 
Signed-off-by: LHXuuu <xulianhao.xlh@antgroup.com>
 2026-09-08 03:17:00 -07:00
Alicia 031612f250 [Bugfix] Fix video benchmark timeout and failure accounting (#7259) 
Signed-off-by: Alicia <115451386+congw729@users.noreply.github.com>
 2026-09-08 09:40:08 +00:00
andyluo7andwangyu b58ff5cb8b [CI/Build] Wait for duplex reaper recovery in retry test (#7225) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
Co-authored-by: wangyu <53896905+yenuo26@users.noreply.github.com>
 2026-09-08 17:10:51 +08:00
tlysanhuo 7c2e30afd3 [Bugfix] Bound omni benchmark per-request timeout to 15 min by default (#7130) 
Signed-off-by: tly <2200895168@qq.com>
 2026-09-08 02:00:25 -07:00
Chenguang Zheng f3fabe4145 [Bugfix] Keep --disable-log-stats out of stage config resolution (#7237) 
Signed-off-by: Chenguang ZHENG <645327136@qq.com>
 2026-09-08 16:19:07 +08:00
Honghan Zhu c086554d1f [Perf]Boogu-Image Skip dense attention masks (#6871) 
Signed-off-by: zhuhh97 <zhuhonghan@huawei.com>
 2026-09-08 00:53:20 -07:00
wuli666 c704aeccae [Bugfix] Fix Step-Audio2 L4 CI KV cache allocation (#6839) 
Signed-off-by: wuli666 <421774554@qq.com>
 2026-09-08 14:17:33 +08:00
Chenguang Zheng 087ac46318 [Refactor 1/N] Centralize Omni config resolution (#5140) 
Signed-off-by: Chenguang ZHENG <645327136@qq.com>
 2026-09-08 10:30:26 +08:00
Deep Shahandamy-why-3459 1413eaab7c [Bugfix][Qwen3-Omni] Read code-predictor RoPE theta from rope_parameters (#7228) 
Signed-off-by: Deep Shah <deep@socratic.co>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
 2026-09-08 10:25:27 +08:00
ZacheryAU b1ec136bd7 [Frontend] Add omni benchmark support for image and video endpoints (#4728) 2026-09-08 09:28:21 +08:00
Yueqian Lin e1c59edec6 [Refactor] fish_speech: vendor DAC codec modules, drop external fish-speech dependency (#4871) 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-09-08 09:22:36 +08:00
boatman 043f03d576 [Migrate]Move model-specific helpers out of serving_speech (#6697) 
Signed-off-by: boatman <109857087+sphinxkkkbc@users.noreply.github.com>
 2026-09-08 08:59:45 +08:00
Jierui (Jerry) Xu c3498fb869 [Model] Add Boogu-Image Turbo pipeline resolution (#6961) 
Signed-off-by: Jerry Xu <xjr2423@gmail.com>
 2026-09-08 08:59:23 +08:00
Anjie HouClaude Sonnet 5Hongsheng Liu
 		6942a5b600 [1/N] Stream Wan VAE chunks to the media consumer (#7016) 
Signed-off-by: specture724 <specture724@gmail.com>
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
 2026-09-08 07:41:13 +08:00
Rahul SteigerandNils Hoffmann 3b335814cc [Bugfix][Cosmos3] Fix distributed Transfer output envelope (#7205) 
Signed-off-by: Rahul Steiger <rsteiger@nvidia.com>
Co-authored-by: Nils Hoffmann <nhoffmann@nvidia.com>
 2026-09-08 07:34:08 +08:00
Pujitha PaladuguPujitha Paladuguamy-why-3459
 		cf36f5e474 [Bugfix][MiniCPM-o] Allow ragged audio_feature_lens across a batch (#7071) 
Signed-off-by: Pujitha Paladugu <10557236+pujitha24@users.noreply.github.com>
Co-authored-by: Pujitha Paladugu <10557236+pujitha24@users.noreply.github.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
 2026-09-07 23:51:13 +08:00