Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 138d137b5b [Refactor][kv_offload]: rename block→chunk (#52615) Ronen Schaffer 2026-09-09 17:45:49 +03:00
• 9ffb8cea96 [CI] Add e2e test for scale-out EC connector flow (#54973) 4 people 2026-09-09 16:38:32 +02:00
• cc4210f671 [Perf] Improve BF16x3 router GEMM accuracy and make it default on sm100 (#55899) 3 people 2026-09-09 22:33:12 +08:00
• 3116c5d06b [Qwen4Exp] Support UVA PLE-offload and Engram tensor parallelism (#54371) Harry Huangandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-09 22:32:34 +08:00
• 62f3bf58a5 [Bugfix][ROCm][DSv4] Skip launch_pdl=True JIT warmup when PDL is unsupported (#56035) jimmy-adams 2026-09-09 21:50:41 +08:00
• e509d32b5a [Transformers backend] Enable QKV-Fuser for Gemma4 (#55690) Thomas OrtnerandHarry Mellor 2026-09-09 15:46:31 +02:00
• 08b3e67b66 [ROCm][Perf] Fix Qwen3-vLLM audio encoder TP when heads are not divisible by TP size (#45900) abrahamzewoudie 2026-09-09 16:28:24 +03:00
• 94848eda60 [Bugfix] Fix unreachable None guard in Molmo2 get_candidate_target_fps (#55893) Adababyandshaolila 2026-09-09 20:37:10 +08:00
• c7e9816c6a [Bugfix][Rust Frontend] Skip undefined token ids in decode and anchor them zero-width (#55240) 3 people 2026-09-09 10:55:04 +01:00
• 474839f846 [CI/Build] Upload CPU nightly image to Docker Hub (#55163) wliao2 2026-09-09 02:13:48 -07:00
• de650b55af [Docs] Update README.md MkDocs to give option to run dev-server on different port. (#54584) 3 people 2026-09-09 17:08:12 +08:00
• 68dcc4fd86 [Rust Frontend] Make --max-model-len optional for the render server (#55328) 3 people 2026-09-09 10:54:54 +02:00
• 5acd95906b [Bugfix] Exclude DP token padding from draft attention metadata (#55458) 4 people 2026-09-09 01:47:45 -07:00
• 114abd1c11 [Bugfix][Frontend] Reject unsupported Responses API input items with 400 instead of 500 (#55974) Nikhil KulkarniandClaude Fable 5.1 2026-09-09 00:51:45 -07:00
• d2906091bf [Bugfix] Fix Mooncake heterogeneous TP with replicated GQA heads (#52516) Yicong WangandOpenAI Codex 2026-09-09 14:18:32 +08:00
• 95cf420ac1 [CI][Test][Spec Decode] Fix CI failure of Qwen3 Omni DSpark loader mock (#56011) Jing Wang 2026-09-09 14:06:15 +08:00
• 385dce36bc [CI/Build][Hardware][NVIDIA] Add public CUDA 13.4 Rubin build path (#54640) 3 people 2026-09-08 21:53:12 -07:00
• bf7eee9d7b [Bugfix][EC Connector] Fix ECExampleConnector load device under TP>1 (#40416) 3 people 2026-09-08 20:44:30 -07:00
• fc6b6e1feb [Rust Frontend] Correctly parse whitespace framing in model output (#55417) Bugen ZhaoandCodex 2026-09-09 13:40:13 +10:00
• 27b63fb8f8 [CI][XPU] skip test_sampling_mask_tensors_match_finite_support on XPU (#55852) Qiming Zhang 2026-09-08 20:16:03 -07:00
• 41541636fd [Refactor][EC Connector] Add backend extension points to ECCPUWorker (#54033) Akine-Ko 2026-09-09 11:13:47 +08:00
• f1aeff6007 [Bugfix][EC Connector] Fail the request, not the engine, when a remote encoding cannot arrive (#55290) Tianyu GuoandOpenAI Codex 2026-09-09 11:13:33 +08:00
• 050f29caf1 [Rust Frontend] Simplify reasoning parser initialization and test organization (#55411) Bugen ZhaoandCodex 2026-09-09 12:57:15 +10:00
• c6ba5e89d6 [Bugfix][Elastic EP] Route only to surviving engines during scale-down (#55772) djramicandAndreas Karatzas 2026-09-09 04:28:28 +02:00
• 21a117ad87 [CI][IR] Speed up vLLM IR test group (#55965) djramic 2026-09-09 04:28:06 +02:00
• 0b066293f3 [Core][KV Connector] Resolve connector block tables for every scheduled request (#54853) 4 people 2026-09-08 19:15:53 -07:00
• a97dacb710 [Docs][EC Connector] CPU EC Connector usage Docs (#54522) omerpaz95 2026-09-09 03:43:08 +03:00
• bc8587f829 [ROCm][Perf][Bugfix] Multi-stream perf improvements; rocprofiler fixes (#55099) Matt 2026-09-08 17:42:34 -07:00
• d98c8c030b [Bugfix] Parse DSML tool calls when the model omits the tool_calls wrapper (#55954) Flora Feng 2026-09-08 20:38:04 -04:00
• 28a73ccfba [Kernel] Add NVFP4 support to the torch linear backend (#53319) Michael LazosandOpenAI Codex 2026-09-08 17:30:05 -07:00
• 4aadb0f14d [Core] Enhance cpu<->gpu sync checking to include paged async copies (#53491) Nick Hill 2026-09-08 17:26:21 -07:00
• 9e137d8d23 [Kernel] Enable optimized FlashInfer add-RMSNorm NVFP4 fusion (#51925) Qidong SuandOpenAI Codex 2026-09-08 20:20:32 -04:00
• c268198715 [Bugfix][Spec Decode] Preserve target parallel config (DCP) for DSpark (#55472) Jing Wang 2026-09-09 07:59:10 +08:00
• 5af4cc33ec [Bugfix] Fix OpenPangu multimodal embedding merge (#55941) Taneem Ibrahim 2026-09-08 18:53:59 -05:00
• 9c2d21046b [Bugfix][Spec Decode] Honour the draft's moe_backend on Model Runner V2 (#54788) 3 people 2026-09-09 01:50:07 +02:00
• c0d8d5413e [Transformers] Generalize merged-column linear fusion (#55301) 3 people 2026-09-08 16:49:37 -07:00
• 60ad959b6f Fix block FP8 MTP in ModelOpt mixed checkpoints (#55513) 3 people 2026-09-08 15:47:21 -07:00
• 28a2ccee78 [ROCm][DI][CI] Enable WideEP Intranode tests (#53195) Chaitanya Sri Krishna Lolla 2026-09-09 03:36:36 +05:30
• 05e52f6336 [CI] Fix ARM64 test dependency builds with GCC 15 (#55877) Kevin H. Luu 2026-09-08 14:27:13 -07:00
• bf339618d1 [Misc] Bump openai to >=2.25.0 to support namespace tools types (#49104) 3 people 2026-09-09 06:11:12 +09:00
• bfb443a6b6 [Kimi Bug] Fix kda ima Triton Error [CUDA]: an illegal memory access was encountered (#55924) Wentao Ye 2026-09-08 17:07:54 -04:00
• 0a742da274 [Perf] Eliminate full-history reasoning scans for structured outputs (#55223) Flora FengandWentao Ye 2026-09-08 17:07:22 -04:00
• 29f46ac475 [Perf] Eliminate full-history reasoning scans for structured outputs (#55223) Flora FengandWentao Ye 2026-09-08 17:07:16 -04:00
• 73f61d12aa [CI][ROCm] Increase timeouts for AMD MI300 jobs (#55919) djramicandKevin H. Luu 2026-09-08 23:06:05 +02:00
• db3814a4f2 [Attention] Require explicit DCP support from attention implementations (#55780) Andreas KaratzasandOpenAI Codex 2026-09-08 16:04:27 -05:00
• cb22234687 [CI] Reuse ColQwen3 models across pooling tests (#55889) Andreas Karatzas 2026-09-08 16:03:54 -05:00
• d29c88f162 [Core] Let SWA layers take the primary block size to avoid inflating the KV block LCM (#53007) 3 people 2026-09-08 16:44:43 -04:00
• 808353f477 [CI] Split long misc test groups by command (#52346) 3 people 2026-09-08 13:44:14 -07:00
• 18615ad1ee [Bugfix] Handle null RoPE parameters for NoPE layers (#55949) Andreas Karatzas 2026-09-08 15:20:05 -05:00
• e0aaef85f3 [2/N][KV Connector][NIXL] Support per-region transfer geometry (#53780) 3 people 2026-09-08 16:01:12 -04:00
• 2c9d68f80b [Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809) Roderick Wuandmgoin 2026-09-08 15:45:12 -04:00
• 1a522b6949 [ROCm] [Docker] Upgrade default AINIC repo to ship libionic 54.0-187-1 (#54112) shikamd123 2026-09-08 12:21:18 -07:00
• 1b2c591cd0 [Bugfix] speedup nvfp4 kv for FMHA (#55031) sychen52andCyrus Leung 2026-09-08 10:25:42 -07:00
• bcca76e7db [Test] Dequantize NVFP4 KV cache scales in the layout the kernel writes (#55908) Raya Elena Solanoandmgoin 2026-09-08 19:13:39 +02:00
• 8ebc5b0a18 [Bugfix] Fix NVFP4 fused SiLU+mul scale allocation and global scale direction (#55643) 4 people 2026-09-08 23:26:07 +08:00
• 6b15bea080 [Refactor] Remove utils dead code (#53941) Wentao Ye 2026-09-08 11:22:09 -04:00
• 7c2f1ff495 [Core] Scope PCP-DP validation to GPU manager (#54523) Qiu Chunshuo 2026-09-08 23:18:01 +08:00
• 8e1f97e709 [Qwen3.8-Flash-Next] Tune FP8 TP2/TP4 Triton MoE on B200 (#55890) Canlin GuoandOpenAI Codex 2026-09-08 22:45:18 +08:00
• b105d8284f [LoRA] Clarify target module matching logic (#55865) linitra24 2026-09-08 22:30:15 +08:00
• 07950d4734 [Kimi Bug] Fix kimi k3 startup cuda graph issue with recoverSSM (#55774) Wentao Ye 2026-09-08 10:29:53 -04:00
• 6ddbab03de [4/N][warmup][DSv4] Migrate common attention kernels (#50176) Roberto L. CastroandCodex 2026-09-08 16:25:48 +02:00
• f998862d46 [Bugfix] Fix Kimi K3 loading with interleaved weight streams (#53379) Sherif WalyandNicolò Lucchesi 2026-09-08 15:08:51 +01:00
• 613ab20f7f [Model] Enable torch.compile for Sarvam MLA (#55817) Mohit SinglaandCodex 2026-09-08 19:31:30 +05:30
• 263c4ff95f [Bugfix][Spec Decode] Cache the Mamba state at the block-grid position of EAGLE resume (#53945) 4 people 2026-09-08 08:59:58 -05:00
• 414057a3d3 [Bugfix] Apply attention sinks in the Transformers backend (#52156) 3 people 2026-09-08 15:54:07 +02:00
• 472d8c9125 [Perf] Fuse DeepEncoder relative bias in Triton attention (#55629) 3 people 2026-09-08 21:22:04 +08:00
• 5ebce2391e [Bugfix][KV Offload] Respect prefix-cache bypass in SimpleCPUOffload (#54998) Andy LoandOpenAI Codex 2026-09-08 14:14:19 +01:00
• f6326f53bd [Perf][GDN] Enable the FlashInfer GDN prefill kernel on SM12x (#55715) Stefano Castagnetta 2026-09-08 14:26:03 +02:00
• 25047604fe [ROCm][Perf] Remove AITER paged-MQA outputs guard for DeepSeek-V4 (#55808) Shanshan Shen 2026-09-08 19:58:26 +08:00
• 34b9899c8f [CI] Increase ColQwen3 pooling test memory budget on H200 MIG (#55878) Kevin H. Luu 2026-09-08 04:39:01 -07:00
• 13cf9e05c1 [Perf][Quant][NVFP4] Prefer W4A4 linear kernels over weight-only ones on SM120/121 (#55170) Stefano Castagnetta 2026-09-08 13:06:45 +02:00
• e41a17e606 [ROCm][Quantization] Support AMD Quark per-block FP8 for fused MoE layers (#52263) jimmy-adams 2026-09-08 18:30:26 +08:00
• 5db652225f [Bugfix] Validate extension integers before engine serialization (#55606) Kevin H. LuuandCodex 2026-09-08 01:30:06 -07:00
• 96eccb8f49 [CI] Restore clean GPU state before OAI Triton MoE tests (#55453) Kevin H. LuuandCodex 2026-09-08 01:15:21 -07:00
• ce6c241ddc [ROCm][Perf][M3] Fused allreduce+GemmaRMSNorm fast path (#54787) Taoyu ZhuandShanshan Shen 2026-09-08 15:21:49 +08:00
• 81dabe2e5d [Bugfix][KV Offload] Validate SWA coverage at unaligned cache-hit boundaries (#55712) ahmed xijiaat 2026-09-08 15:15:48 +08:00
• 5133e1d285 [Frontend] Migrate input-validation errors to VLLMValidationError (harmony/mistral/chat_utils/params) (#55735) Adababyandshaolila 2026-09-08 14:45:40 +08:00
• 755541443f [Bugfix][ROCm] Mask paged attention V cache padding (#53856) aoshen02 2026-09-08 14:39:23 +08:00
• 782f36cd0c [Bugfix][InternVL] Stop the video parser consuming image_embeds (#55779) Subhash Polisetti 2026-09-07 22:19:41 -07:00
• 6c73b08dec [Bugfix] reject string file on translation like transcription (#55618) 3 people 2026-09-08 06:02:53 +01:00
• 9ea8f3ffc3 [Bugfix] Honor STEP token pooling in DispatchPooler.for_seq_cls (#55307) weihan 2026-09-08 12:13:09 +08:00
• 7fa2c63796 [Bugfix] Autotune FlashInfer deferred MoE decode kernels before CUDA graph capture (#55377) 3 people 2026-09-07 21:02:23 -07:00
• eb6b619ab2 [Bugfix][KV Offload] Register mixed page sizes in one cache group (#54756) He LiuandKimi 2026-09-08 11:50:15 +08:00
• 9c297e3b8b [Kernel] PDL enablement for fusedQKNormRopeKernel (#55755) Jee Jee LiandYongye Zhu 2026-09-08 11:49:03 +08:00
• 607a6e48c5 [CI/Build] Fix unused fake implementation testing (#55797) Jee Jee Li 2026-09-08 11:36:18 +08:00
• a69402aaa8 [Perf] Integrate FlashInfer KDA kernels (#55364) Wei Zhao 2026-09-07 23:29:30 -04:00
• 12579c7f01 [Kernel][Perf] Tune H20 block-FP8 MoE low-batch configs (+21%) (#54668) leo 2026-09-08 11:20:21 +08:00
• 869f78732b [Kimi Bug] Fix kimi k3 AssertionError assert 0 <= checkpoint_idx < len(blocks) (#55747) Wentao Ye 2026-09-07 22:34:48 -04:00
• 6b5a12c0f8 [ROCm][CI] Enable HY-V4 model initialization on ROCm (#54405) Andreas Karatzas 2026-09-07 20:55:44 -05:00
• 9dbdf8e9af [Bugfix][Quantization][MoE] Normalise an unset group_size on the compressed-tensors WNA16 MoE path (#53163) Artur FierkaandClaude 2026-09-08 03:42:46 +02:00
• 06334fae2d [Build] Remove obsolete TPU Dockerfile (#55376) Woosuk KwonandOpenAI Codex 2026-09-07 18:07:27 -07:00
• 54da70c1eb [Feature] Support EAGLE3 for Sarvam (#53052) 3 people 2026-09-08 05:38:35 +05:30
• fdfa0a1659 [Pooling] Honor request_id from request bodies (#55665) Taneem Ibrahim 2026-09-07 18:55:20 -05:00
• 252ed87621 [Bugfix] Restore Responses validation error boundary (#55761) Taneem Ibrahim 2026-09-07 17:12:18 -05:00
• 537af2c3a4 [CI] Recover empty multi-node Docker networks and finish partial cleanup (#55454) Kevin H. LuuandCodex 2026-09-07 12:53:13 -07:00
• 58f09211dd [CI] Synchronize shared offload unlink test before observing pathname (#55604) Kevin H. LuuandCodex 2026-09-07 12:52:36 -07:00
• 51da0ca66c [CI/Build] Unskip ColQwen3 multimodal pooling tests on Transformers v5 (#55588) Sahil Patel 2026-09-07 16:15:39 +01:00
• ad2f18e8af [CI] reduce npu CI use time and add timeout (#55731) Zeyu Yang 2026-09-07 22:51:09 +08:00
• 70584f69b1 [Model] Add Cohere Compass model (#54774) 3 people 2026-09-07 10:36:49 -04:00
• 7cc89a7dda [Tests] Update SarvamMLA transformers v5 compatibility reason to hf (#55728) AbhiandAntigravity 2026-09-07 19:57:07 +05:30