anodyine/vllm-omni
1
0 Fork 0
Code Issues 4 Pull Requests 2 Actions Packages Projects Releases Wiki Activity
3,433 Commits 235 Branches 31 Tags
dev/vllm-align
Commit Graph
3433 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
tzhouam 07326d0922 Align source installation documentation with vLLM 0.31.0 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-05 11:52:40 +08:00
tzhouam df14622fb4 Drop CosyVoice regression changes superseded by main 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-05 11:41:24 +08:00
tzhouam 2f17684774 Preserve baseline ASR fixtures for matching-main failure classification 
Keep documentation and narrowband speech evaluator settings identical to Omni main. Matching-main failures remain documented failures rather than reasons to change evaluation or retry for a green build. Retain failed-audio diagnostics and release compatibility fixes.

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 17:54:28 +08:00
tzhouam 0e5e37bde6 Recheck narrowband TTS content with stronger ASR 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 17:46:32 +08:00
tzhouam 223e521320 Use accurate ASR for Qwen3-Omni documentation audio 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 17:12:01 +08:00
tzhouam e81ae8f66f Format failed-speech diagnostic helper and regression coverage 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 16:44:42 +08:00
tzhouam 69b2fcb94e Retain failed speech ASR input without changing accuracy assertions 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 16:40:00 +08:00
tzhouam f7d8d36353 Retain Qwen3-Omni documentation audio for failure diagnosis 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 15:25:06 +08:00
tzhouam 6f5f9b97e3 Merge current Omni main and retain conditioning regression coverage 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 13:50:06 +08:00
tzhouam bd6ab4d50e Supply both completed block hashes in NPU prefill coverage 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 13:45:08 +08:00
tzhouam 4414701161 Use published release images and wire NPU auxiliary outputs 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 13:40:24 +08:00
Sy03 1c1b831f58 [Model] Batch Qwen3-TTS Talker prompt projection and settled-row preprocessing (#8477) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-10-04 05:08:24 +00:00
Sy03 dd765fbf64 [Model] Select CosyVoice3 streaming defaults by device capability (#8433) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-10-03 21:59:36 -07:00
andyluo7 b171731c81 fix: preserve CosyVoice full payload and live conditioning (#8343) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
 2026-10-04 16:03:27 +13:00
andyluo7 09d704907f fix: preserve Torch seeded sampling on ROCm (#8341) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
 2026-10-04 16:03:21 +13:00
andyluo7 c150600896 [Bugfix] Route MOSS reference attention through ROCm FlashAttention (#8463) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
 2026-10-04 16:02:22 +13:00
tzhouam 4ad71d8270 Synchronize CPU ring workers before group teardown 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 04:43:03 +08:00
tzhouam df3807eca4 Keep release alignment focused and clarify baseline fixes 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-04 03:11:35 +08:00
tzhouam aa499fe79d Provide Omni lifecycle state for auxiliary v2 stages 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 23:27:40 +08:00
tzhouam cb8d9f2209 Format realtime reliability fixtures 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 23:10:57 +08:00
tzhouam 37803a8a41 Update realtime invalid-request fixtures and reserve Moss encoder memory 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 23:05:51 +08:00
tzhouam 757dd3b2c4 Leave Code2Wav workspace headroom in two-L4 serving fixture 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:51:54 +08:00
tzhouam 0dd5f3e17d Use blockwise Flux offload within single-L4 memory capacity 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:34:10 +08:00
tzhouam 649b0f763a Fit weekly serving fixtures to assigned GPUs and initialize accuracy groups 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:29:17 +08:00
tzhouam 9361c52de3 Honor Omni stage runner selection in shared vLLM config 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:22:40 +08:00
tzhouam b256e1ec4e Adapt Moss sampling and runner fixtures to vLLM 0.31 interfaces 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:22:39 +08:00
tzhouam 0b6591ae0a test: fit Qwen2.5 online expansion on its declared two GPUs 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:06:27 +08:00
tzhouam 864bc7b22a test: update the Moss runner fixture for release output APIs 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:06:27 +08:00
tzhouam c8d5f99f10 fix: scope CI package version overrides to Omni 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:03:13 +08:00
tzhouam 43e05c6e13 fix: align MRV2 profiling and capture with vLLM 0.31 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 22:03:13 +08:00
tzhouam 16f73d0da2 fix: match vLLM 0.31 generation profiling API 
Accept the upstream randomize_inputs profiling flag while preserving structured codec inputs during generation-stage dummy runs.

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 21:23:18 +08:00
tzhouam 2974f8be59 test: use the vLLM 0.31 model-length property in graph fixtures 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 21:20:08 +08:00
tzhouam ed876de879 rebase: target the official vLLM 0.31.0 release 
Use upstream release images at db9527a46873, preserve custom processor conditioning, and include weekly jobs in full Buildkite release validation.

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 21:11:54 +08:00
tzhouam 008fa25263 Merge main into dev/vllm-align for vLLM 0.31 release 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 21:05:17 +08:00
tzhouam 84cbf94b1b fix: preserve Omni processor conditioning on vLLM 0.31 
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
 2026-10-03 21:03:26 +08:00
andyluo7 ee8fdab1dc [CI/Build] Fix MOSS CPU tests on ROCm workers (#8447) 
Signed-off-by: andyluo7 <andy.luo@amd.com>
 2026-10-03 09:13:51 +08:00
Sy03andYueqian Lin 3bc3f1a7d0 [Model] Optimize MOSS Local 1.5 MRV2 serving and progressive chunks (#8213) 
Signed-off-by: Sy03 <1370724210@qq.com>
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Yueqian Lin <linyueqian@outlook.com>
 2026-10-02 18:03:14 +00:00
Zheng Wengangandamy-why-3459 527982d886 [Bugfix] Skip async-chunk SHM put for stage-0-final requests (#7245) 
Signed-off-by: ZhengWG <zwg0606@gmail.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
 2026-10-02 09:14:39 -07:00
Sy03 2ba5074ca0 [Model] Optimize CosyVoice3 packed streaming with bounded HiFT graphs (#8420) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-10-02 08:49:50 -07:00
Yueqian Lin 4025bb30f6 [Perf][AuK] Graph the encoder prefill, cache reference latents and fuse the codec activation (#8305) 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-10-02 15:43:55 +00:00
Yueqian Lin 13c66da752 [Perf][AuK] Precompute adaLN modulations per request and run the conv position embedding as one GEMM (#8328) 
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
 2026-10-02 08:12:29 -07:00
Nick CaoandCodex 0e6791e9c4 [Bugfix] Initialize eager state in Higgs test fixture (#8426) 
Signed-off-by: Nick Cao <ncao@redhat.com>
Co-authored-by: Codex <noreply@openai.com>
 2026-10-02 10:22:52 -04:00
Zheng Wengangandamy-why-3459 eb1699bfca [FEAT][Fronted] Support incremental prefill for realtime video streaming (#5894) 
Signed-off-by: ZhengWG <zwg0606@gmail.com>
Signed-off-by: Zheng Wengang <zwg0606@gmail.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
 2026-10-02 22:18:51 +08:00
Sy03 e372aa3a41 [Bugfix] Restore Fish Speech and CosyVoice3 audio quality at high concurrency (#8422) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-10-02 07:11:04 -07:00
Sy03princeprideClaude Opus 5.5
 		ae58c2d8b9 [Model] YuE2: async single-stage synthesis and CUDA graph acceleration (#8407) 
Signed-off-by: Sy03 <1370724210@qq.com>
Signed-off-by: princepride <wangzhipeng628@gmail.com>
Signed-off-by: Wang Zhipeng <wangzhipeng628@gmail.com>
Co-authored-by: princepride <wangzhipeng628@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-02 05:17:06 -07:00
linzhenpl07Claude Opus 5Zhou Taichang
 		b10270a01c [AR-Diffusion] Page in kernel-legal blocks so the default resolution can run (#6481) 
Signed-off-by: linzhenpl07 <linzhenpl07@gmail.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Zhou Taichang <tzhouam@connect.ust.hk>
 2026-10-02 20:05:04 +08:00
Sy03 e84014dffb [Model] Add single-stage Qwen3-TTS with in-Talker streaming codec decode (#8259) 
Signed-off-by: Sy03 <1370724210@qq.com>
 2026-10-02 21:34:30 +10:00
Rahul Steiger 596487fe62 [Core] Prototype diffusion attention execution contracts with FA4 (#7379) 2026-10-02 17:58:47 +08:00
MaciejBalaNVand汪志鹏 97d6c693a6 [feature] Add HSDP pre-sharded option (#7948) 
Signed-off-by: Maciej Bala <mbala@nvidia.com>
Co-authored-by: 汪志鹏 <wangzhipeng628@gmail.com>
 2026-10-02 09:35:09 +00:00
汪志鹏 b742f86136 [CI/Build] Restore streaming executor fixture contract (#8418) 
Signed-off-by: princepride <wangzhipeng628@gmail.com>
 2026-10-02 01:45:56 -07:00