Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• cd64c2dea9 [Docs] Clarify admission control limits apply server-wide, not per DP rank (#55124) Nicolò Lucchesi 2026-09-07 16:08:48 +02:00
• f2d45f26bd [Bugfix][Gemma] Conditionally create KV projections/norms on KV-shared layers (#54917) 3 people 2026-09-07 17:04:00 +03:00
• 42801b3a6b [3/N][warmup][DSv4] Migrate FA4 MLA and shared CuTeDSL kernels (#53565) 3 people 2026-09-07 15:36:35 +02:00
• 1713b9866a [2/N][warmup][DSv4] Migrate sequence and DCP kernels (#53564) 5 people 2026-09-07 15:35:54 +02:00
• ecd600d91e [Pooling] Honor max_embed_len for chunked embeddings (#55551) Taneem Ibrahim 2026-09-07 08:31:40 -05:00
• 167858ea17 [Frontend] Migrate Responses harmony input validation to VLLMValidationError (#55701) Adababyandshaolila 2026-09-07 21:30:11 +08:00
• b339d75a41 [Bugfix][Spec Decode] Resolve n_predict from text_config for Qwen3.5 multimodal MTP (#55369) 3 people 2026-09-07 18:58:26 +05:30
• 6a2a2bb02b [Frontend] Add stateless /v1/responses/render endpoint (#50195) Francisco Javier ArceoandOpenAI Codex 2026-09-07 08:50:42 -04:00
• 5e6f6a8ed4 [Bugfix][Quantization][XPU] Fix moe_wna16 linear weight loading (#52651) Artur Fierka 2026-09-07 14:20:17 +02:00
• 4a806d08ee [Bugfix] OffloadingConnector: stop zeroing offload hits under MTP/EAGLE spec decode (#52771) Kam BasraandClaude 2026-09-07 13:16:05 +01:00
• 94e26dd3dd [Qwen3.8-Flash-Next] Support FP8 indexer cache for QSA (#54890) Thien Tran 2026-09-07 20:15:45 +08:00
• 58ad1f3b89 [Docs] Add OLMo 2 to batch-invariance tested models (#55691) Sean Westfall 2026-09-06 23:56:49 -10:00
• 8648446515 [XPU][LoRA] Support LoRA for DeepSeek V4 on XPU (#53689) Chaojun Zhang 2026-09-07 17:44:12 +08:00
• 9b85112e13 [XPU] Route grouped_topk to the fused _moe_C kernel on XPU (#53580) Marceli FylcekandClaude Opus 5 2026-09-07 11:43:38 +02:00
• 7dbe386833 [Frontend] Migrate Responses API validation errors to VLLMValidationError (#50257) Adababyandshaolila 2026-09-07 17:33:51 +08:00
• e476556189 [ROCm][CI] Add attention-sink support to ROCm AITER sparse MLA (#54404) Andreas KaratzasandOpenAI Codex 2026-09-07 04:26:47 -05:00
• 392db567b2 [CI] [Test] skip test_wna16_cuda_high_bit_skips_humming on non-CUDA platforms (#55660) Chaojun Zhang 2026-09-07 16:57:15 +08:00
• 49eb2accf0 [KVConnector] Guard lmcache_mp_connector state transition with num_external_tokens (#47505) AlexHuang 2026-09-07 16:19:07 +08:00
• 5893426b88 [Bugfix] DSv4 MXFP4 selector: stop narrowing explicit aliases to their BF16 variant (#53586) 4 people 2026-09-07 15:19:13 +08:00
• 34b1e9f7a6 [Bugfix][MooncakeStore] Fix finish-time save crash on hybrid models (#54643) Zhewen Liandzhewenl 2026-09-07 00:18:30 -07:00
• 195bc9c4a1 [CI][ROCm] Temporarily skip unsupported HY-V4 initialization (#55653) Andreas Karatzas 2026-09-07 02:17:36 -05:00
• 6fbb00b188 [EPD] Add ECMooncakeConnector for encoder cache over Mooncake TransferEngine (#41567) 6 people 2026-09-07 14:21:43 +08:00
• f7f060d253 [Bugfix][Audio] Restore soundfile-first automatic decoding (#55642) Andreas Karatzas 2026-09-07 01:21:02 -05:00
• ed29dfae6e [XPU] Use fused_input_norm kernel in FusedInputNorm (#52945) zofiaandKunshang Ji 2026-09-07 13:44:31 +08:00
• d9105ea800 [Qwen3.8-Flash-Next] Remove torch.compile for NVIDIA implementation (#55272) Thien Tran 2026-09-07 12:57:56 +08:00
• 9cc7793e32 [Bugfix] Gracefully handle unsupported reasoning_effort in chat templates (#54022) frankieandTrae Agent 2026-09-07 12:11:35 +08:00
• f43ef1531d [CI] fix pre-commit (#55630) Jiangyun Zhu 2026-09-07 11:54:22 +08:00
• 3dc7a68ce4 [Perf] Extend Qwen Triton warmup to avoid first-request latency spikes (#54797) vhagorandIsotr0py 2026-09-07 11:46:00 +08:00
• 6748217fb9 [Bugfix] Fix Kimi K3 NVFP4 MoE weight conversion OOM (#55407) Wei Zhao 2026-09-06 23:45:25 -04:00
• c3ec0d29f5 [Bugfix] Fix cuda profiler missing bug (#55237) Wei Zhao 2026-09-06 23:15:57 -04:00
• 1f778486fc [Bugfix][Offloader] Preserve prefetch static-buffer slot ownership (#54975) yulunandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-07 10:58:44 +08:00
• de69e821b7 [ROCm][Perf][DeepSeek V4] Fuse native FP8 shared expert with MXFP4 routed experts (#53161) 3 people 2026-09-06 22:42:21 -04:00
• 294fbb4f59 add 2/3/5/6/7 CUDA support in AutoRound format (#52890) Wenhua Cheng 2026-09-07 09:59:19 +08:00
• 199cb9b964 [Kernel] Remove unused fake implementation (#55535) Jee Jee Li 2026-09-07 08:54:32 +08:00
• 4df80187b5 [Kernel] SM 12.x blockwise FP8: swizzle the CTA raster when the weight exceeds the L2 (#55180) Jürgen Schmied 2026-09-07 01:36:59 +01:00
• 6865e67f0b [Bugfix] Defer adaptive verification until after kernel warmup (#55455) Kevin H. LuuandCodex 2026-09-06 11:31:40 -07:00
• 722d169339 [CI] Align extraction test with canonical auxiliary layer order (#55457) Kevin H. LuuandCodex 2026-09-06 11:22:14 -07:00
• 52358e6e19 [Frontend] Expose multimodal metadata for disaggregated prefill (#54659) zhouyouandCursor Grok 4.6 2026-09-06 22:46:52 +08:00
• 808f8cd3ac [Governance] Add aoshen02 as code owner for RL components (#55529) aoshen02andCodex 2026-09-06 21:26:22 +08:00
• dc02934a07 [Bugfix][KV Offload] Stop offloading the final sampled token's KV slot (#54288) Almog TavorandOr Ozeri 2026-09-06 14:22:19 +02:00
• 569adb5a97 [Bugfix][KV Offload] Fix SWA store reachability during chunked prefill (#54362) 3 people 2026-09-06 03:50:39 -07:00
• dd0760165c [Bugfix] Fall back to T1 when ARC cannot reclaim enough entries from T2 (#55461) Zupeng WangandCodex 2026-09-06 18:28:19 +08:00
• 9afb878a55 [Bugfix][KV Offload] Skip cleaned-up async lookup batches (#55075) 3 people 2026-09-06 18:27:46 +08:00
• a1541f5742 [Perf] Use SDPA for BLIP-2 Q-Former attention (#55285) SunandIsotr0py 2026-09-06 15:04:19 +08:00
• 144e79c810 [Kimi K3] Support internal prefix checkpoints with partial prefix caching and spec-decoding (#53614) Ziming HuangandOpenAI Codex 2026-09-06 13:18:15 +08:00
• 1970f3ed4b Validate scale-out multimodal data before engine handoff (#51898) Clinton ThomasandLucas Bourtoule 2026-09-06 00:58:04 -04:00
• f2e2936f91 [Kernel] Add fused MoE tuned config for E=256,N=512 on NVIDIA A100 80GB PCIe (#55511) Baki Burak Öğün 2026-09-06 06:15:01 +03:00
• 1c344ed41e [CPU] [Feat] Add native AMX-FP8 attention impl for Diamond Rapids (#49410) zhao, zhenhui 2026-09-06 11:06:56 +08:00
• 039ea82660 [Core] Sync DP state on the first step of a wave (#52957) aoshen02 2026-09-06 10:16:58 +08:00
• f4eccdadef [Bugfix][Multimodal] Bound renderer warmup to the prefill token budget (#55448) lucamotzandOpenAI Codex 2026-09-05 18:43:28 +02:00
• 7985444339 [Kernel][HY V4] Add Triton iHC pre/post fallback (#55059) linitra24andJee Jee Li 2026-09-05 23:12:50 +08:00
• 52bc900d93 [Bugfix][Kernel] Build fused GDN MTP decode for SM110 (#53835) 3 people 2026-09-05 22:59:50 +08:00
• 28e605fb33 [Bugfix][Qwen4Exp] fix state index strides in fused PLE conv (#55375) Harry Huang 2026-09-05 22:02:20 +08:00
• bc96d76aa9 [Kernel] Fall back from persistent top-k on low-shared-memory GPUs (#54110) lucamotzandLuca Motz 2026-09-05 14:58:37 +02:00
• 7fbd44cbe0 [Bugfix][DSv4] Seed the -1 sentinel in the prefill sparse index workspace (#55299) aoshen02andaoshen02 2026-09-05 18:31:04 +08:00
• 2902ca17e3 [Bugfix] complete VLLMValidationError migration in chat_utils.py (#50254) Adababyandshaolila 2026-09-05 17:56:12 +08:00
• e473e9036f [Docs][Models] Use the official FunASR Nano vLLM checkpoint (#54944) 3 people 2026-09-05 15:29:49 +08:00
• 32601ef7a1 [Perf][Multimodal] Avoid duplicate text embedding in Qwen2.5-Omni (#55415) waizuichougou 2026-09-05 14:39:32 +08:00
• 8369affa54 [Feat] Add EPLB support for GLM-5.3-Flash (#55119) 3 people 2026-09-05 13:28:33 +08:00
• e962733e08 [Security] Validate cache salts before they reach LMCache (#51444) 3 people 2026-09-04 23:56:06 -04:00
• 4ee2595512 [Attention] Sync FA with upstream (#54819) GroupieSteven 2026-09-04 23:48:32 -04:00
• d87a440f88 [Core][MRV2] Support eagle3 spec decode with pipeline parallel (#50514) 3 people 2026-09-04 20:45:34 -07:00
• d4ef5e53fc [CI][ROCm] Disable Transformers nightly groups (#55409) Andreas KaratzasandOpenAI Codex 2026-09-04 21:51:01 -05:00
• ea5f41edd7 [CI][ROCm] Restore Wikitext coverage for Qwen OCP-MX (#55410) Andreas KaratzasandOpenAI Codex 2026-09-04 21:50:54 -05:00
• a7f9be9612 [CI][ROCm] Raise AMD job and server readiness timeouts (#55308) Andreas KaratzasandOpenAI Codex 2026-09-04 21:50:51 -05:00
• 16328c7a77 [Perf] Kimi K3 nvfp4 Align in_proj weights by 128 to avoid elementwise copy (#55242) Wei ZhaoandOpenAI Codex 2026-09-04 22:46:47 -04:00
• 385ba6b5b0 [Bugfix][Quantization] Register Quark per-block FP8 scales as weight_scale (#54770) 3 people 2026-09-05 10:44:24 +08:00
• 6cbb3c154e [Perf][GDN] Build cudagraph-capture metadata without a device sync (#55404) 3 people 2026-09-04 17:40:17 -07:00
• 8277c42e4c [Perf] Ensure async h2d copies are pinned in more places (#55202) Nick Hill 2026-09-04 17:17:01 -07:00
• 874df9373d [Bugfix] Preserve Mamba state for padded prompt tails (#55178) Nathan Sala 2026-09-05 02:44:28 +03:00
• eb74fbb3e7 [Bugfix][NIXL] Don't assert when a failed transfer is cleaned up twice (#54518) 3 people 2026-09-04 16:20:32 -07:00
• 685074cd61 [Bugfix][V2] Warm up kernels before capturing CUDA graphs (#55341) 3 people 2026-09-05 06:44:18 +08:00
• 8f269a93bb Revert "[CI] Remove deleted nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 and its arch aliases" (#55392) Misha Goin 2026-09-04 17:45:24 -04:00
• 3f41d102c5 [1/2][Model Runner V2] DBO support, eager mode (#50945) 3 people 2026-09-05 05:17:39 +08:00
• 784cac7c42 [Mypy] Fix mypy typing for P models (#54169) Taneem IbrahimandHarry Mellor 2026-09-04 15:47:11 -05:00
• 5093e4844a [Bugfix][Spec Decode] Drop FlashAttention's AOT schedule for a sliding-window DFlash drafter (#54374) Zihan Zhang 2026-09-04 13:42:01 -07:00
• c81ace1859 [Perf] Read VidCom2 frame budgets once (#55331) Sun 2026-09-05 04:11:42 +08:00
• 3284af6bf1 [Bugfix] Reject 0 or non-positive max concurrency (#54887) Taneem Ibrahim 2026-09-04 14:59:29 -05:00
• 2524051385 [Bugfix][Spec Decode] Honour the draft's attention_backend on Model Runner V2 (#54826) Stefano Castagnetta 2026-09-04 21:23:19 +02:00
• 5690b02c03 [Quantization] Support online quantization with partially pre-quantized checkpoints (#51392) 3 people 2026-09-04 20:46:06 +02:00
• a11dfcff89 [CI] Surface why the CRCR nightly report cannot read its Buildkite secret (#54860) Andrey Talman 2026-09-04 14:10:30 -04:00
• 701a744490 [CI] Raise AMD Spec Decode Eagle 1 job timeout to 35min (#55136) Jared WenandKimi 2026-09-05 01:31:54 +08:00
• 8cd95f7de7 [Bugfix][MLA] Restore DSpark cache-group capability under optimized Python (#55234) Summer Yang 2026-09-04 10:24:31 -07:00
• 1ff5edb023 [Bug-fix] Fix MoE fused sum row offsets (#50220) yzyyzyhhhandyaozhiyuan.666 2026-09-05 00:39:02 +08:00
• 761c5861e3 [Rust Frontend][gRPC] Preserve multimodal metadata for remote-prefill decode (#54814) Connor Carpenter 2026-09-04 12:37:53 -04:00
• 8ad2076c4a [Transformers backend] Find attention with a fuser and attach vLLM's layer to it (#54941) Thomas OrtnerandHarry Mellor 2026-09-04 18:26:01 +02:00
• 99a1ab8e4c [ROCm][CI] Bump ROCk release image build timeout to 3h (#55354) Rohan PotdarandClaude Opus 4.8 2026-09-04 11:16:54 -05:00
• 6a039f465e [feat] add torchcodec as audio loader and implement selective audio backend (#51826) 3 people 2026-09-05 00:16:10 +08:00
• 131e0285f6 [Frontend] Warn when removed guided-decoding fields are present in a request (#54285) erdholionandIlir 2026-09-04 18:15:40 +02:00
• 5e729d84eb [CI] Only run GitHub Actions on the main repo (#55349) Harry Mellor 2026-09-04 17:09:49 +01:00
• f19431e5c8 [Bugfix][MoE] Allow TRTLLM FP8 block-scale MoE with SwiGLU clamp (#55069) 4 people 2026-09-04 23:34:21 +08:00
• a85d0738da [Bugfix] Fix double BOS in LLM.chat() for multimodal models (#55288) summer 2026-09-04 22:23:15 +08:00
• a69e75b9b6 Fast Start (#54921) 3 people 2026-09-04 21:32:06 +08:00
• 07ee97e023 [Test] Split test_sampling_mask_preserves_top_k_boundary_ties to remove Triton-kernel-specific assumption Description (#55315) Qiming Zhang 2026-09-04 06:16:50 -07:00
• 31a8a26662 [Qwen3.8-Flash-Next] Improve QSA sparse GQA for prefill and short-ctx decode (#54873) Thien TranandKimi 2026-09-04 21:08:02 +08:00
• 8340fe1bb9 [CI/Build] Fix flaky failures in CPU CI image building (#55317) Li, Jiang 2026-09-04 20:27:02 +08:00
• 8b6de0eb9a [Security] Cap GLMGA video sampling to prevent request-driven resource exhaustion (#54935) Juan Pérez de AlgabaandIsotr0py 2026-09-04 14:02:41 +02:00
• c615b1fd67 [Mypy] Fix mypy typing for L models (#54177) Taneem Ibrahim 2026-09-04 06:50:43 -05:00
• 3ff4f02dfe [AMD][kimik3][ROCm][Perf] Fuse MLA q/kv RMSNorm in AMD Kimi-K3 MLA wrapper (#52494) Rita Brugarolas 2026-09-04 03:14:51 -07:00
• fd4a151262 [Kernel] Reuse Qwen4Exp HC combine-norm for MTP input (#54687) Canlin Guo 2026-09-04 17:57:56 +08:00