Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 7f4793eaa3 [Nixl][PD] DCP support for MLA models (#50611) Nicolò Lucchesi 2026-08-29 18:03:53 +02:00
• 4fc943b867 [Multimodal] Deprecate PyAV video decoder backend (#54231) Isotr0py 2026-08-29 23:13:28 +08:00
• fd5d3aea94 [Mypy] Fix typing for J models (#54130) Taneem Ibrahim 2026-08-29 07:50:15 -05:00
• cacc429f62 [ROCm][CI] Stage E gating (#50920) Andreas KaratzasandOpenAI Codex 2026-08-28 23:52:23 -07:00
• d3d79ffc1e [Bugfix][Spec Decode] Capture the widest uniform decode batch by default (#50488) 4 people 2026-08-28 22:38:13 -07:00
• a758a9f671 [Perf][KV Connector] Pin token_indices before non_blocking H2D in hf3fs helper (#54293) 3 people 2026-08-28 22:20:17 -07:00
• f4f3bcd4b4 [Test] Assert co-located RayExecutorV2 stores publish distinct ports (#54310) 3 people 2026-08-29 13:13:31 +08:00
• fb68025138 [Bugfix][MRV2] Release model and KV cache on in-process engine shutdown (#54162) Oxana KorzhandClaude Opus 5 2026-08-28 23:09:08 -06:00
• 3958a420f0 [Bugfix][V1] Keep an encoder cache entry until its last occurrence is freed (#54284) Hotragn Pettugani 2026-08-29 01:08:01 -04:00
• 6d4562c59b [Attention][DCP] Enable FlashInfer MLA for DSpark drafting (#54277) Summer Yang 2026-08-28 21:11:28 -07:00
• b2f685834a [Hy4] support Hy4-preview model (#54160) 6 people 2026-08-29 12:07:28 +08:00
• 6c18a54648 [Perf] Avoid h2d copies from non-pinned CPU tensors (#54299) Nick Hill 2026-08-28 21:06:30 -07:00
• 5182f2705b [CI/Build] Use file rendezvous for UniProc loader fixtures (#52367) Stellar鱼 2026-08-29 11:53:04 +08:00
• 738bc8811a [BugFix] Disable TP for Qwen3-Omni audio encoder when heads % TP != 0 (#50858) kaiyuanxieandCalvinXKY 2026-08-29 11:18:04 +08:00
• 46a83642f6 [Perf] Pin CPU tensors before non_blocking H2D in three MM paths (#54292) 3 people 2026-08-28 20:13:40 -07:00
• 93ab92be0c [Bugfix][AMD] Annotate draft KV cache groups on the hybrid grouping path (#52047) 4 people 2026-08-28 20:53:46 -06:00
• 6b110badbb [Bugfix][Mooncake] Save exact Mamba boundary states (#51358) 4 people 2026-08-28 19:40:53 -07:00
• 16d6c376bc [Bugfix][Models] Register sleep-managed runtime buffers (#53507) Ronaldandaoshen02 2026-08-29 10:32:24 +08:00
• 7fd9cc036e [KV Connector] Support MooncakeStore with hybrid DCP prefix caching (#53324) Wei ZhaoandYifan Qiao 2026-08-28 22:10:23 -04:00
• 43196f2458 [Perf][MLA Sparse] Pin req_id_per_token before non_blocking H2D on XPU and ROCm (#54295) 3 people 2026-08-28 19:00:46 -07:00
• 4c6c9d569f [Rust Frontend] Take the raw buffer in mm tensor lowering when possible (#53528) 4 people 2026-08-29 09:05:14 +08:00
• 99013d77d3 [Bugfix][Distributed] Gate cross-node MNNVL custom all-reduce by group capability (#53253) 3 people 2026-08-29 08:49:24 +08:00
• 026d5af7f4 [Bugfix] Fix ncclCommQueryProperties heap overflow with NCCL >= 2.31 (#53008) Xuan-1998andVladimir Aerov 2026-08-28 17:25:00 -07:00
• 68c52b5a93 fix: improve token_ids_cpu swap to copy only valid indices (#36255) zzaebok 2026-08-29 08:22:39 +08:00
• 085e9bb07e [Rust Frontend] Add support for truncate_prompt_tokens and truncation_side (#48584) Pranav ThakurandBugen Zhao 2026-08-29 05:46:37 +05:30
• 1a16c2ad2c [CI][Test] Deflake the rms_norm scaling-property assertions (#54271) Oxana KorzhandClaude Opus 5 2026-08-28 18:08:20 -06:00
• 9662ab0835 [Bugfix] Set breakable graph env before Ray actor import (#53293) Ace EldeibandCodex 2026-08-28 19:11:50 -04:00
• ae5b8e4a8d [ROCm][PERF] Enable AITER PA gluon decode for MiniMax-M3 MTP and dense layers (#52849) 3 people 2026-08-28 15:54:39 -07:00
• b131311fb4 [CI/Build] Add advisory PR title format check (#54263) Bugen Zhao 2026-08-29 08:50:47 +10:00
• df14152ac6 [Model] Support speculative decoding method for PLaMo3 (#54239) Shinichi Hemmi 2026-08-29 07:31:07 +09:00
• fd98d32f73 [ROCm][CI] Fix test_ray_v2_executor (#54249) Charlie Fu 2026-08-28 17:11:42 -05:00
• c274d36103 [Bugfix] Keep the Moondream3 MoE all-reduce out of the fused-path try (#54152) 3 people 2026-08-28 16:06:42 -06:00
• 2aac565cae [Core][KV Connector] Start async KV loads after the forward launch when no sync loads are scheduled (#53333) Summer YangandNick Hill 2026-08-28 14:58:00 -07:00
• 96242aa50d [Bugfix][MRV2] Release layer-bound KV cache memory in shutdown() (#54246) Nick HillandKimi Code 2026-08-28 14:27:09 -07:00
• 70732942c6 [Bugfix][ROCm] Pre-allocate wvSplitKrc static workspaces before KV init (#54247) Nick HillandKimi Code 2026-08-28 14:22:53 -07:00
• 21fa2c5a2a [Mypy] Fix mypy typing for model interfaces and H/I models (#54079) Taneem IbrahimandWentao Ye 2026-08-28 15:49:19 -05:00
• 74850f9f66 [Rust Frontend] Optimize SSE streaming hot path (#51321) 3 people 2026-08-29 04:35:38 +08:00
• 06cccf8730 [Bugfix] Fix int32 token offset overflow in fused SiLU block quant (#53409) 4 people 2026-08-28 13:23:36 -07:00
• 67e86d1e6f [BugFix] Bind RayExecutorV2 TCPStore before publishing its port (#50969) 5 people 2026-08-29 03:27:22 +08:00
• 2f1cba799e [ROCm] remove VLLM_ROCM_USE_AITER_FP4_ASM_GEMM environment variable; make w4a4 use the preshuffle triton+asm by default (#53141) afriedriandAndreas Karatzas 2026-08-28 13:59:53 -05:00
• 31c579503d update quark docs to include online quantization (#52736) hangy-amd 2026-08-29 02:23:06 +08:00
• d9dabfa351 [Kimi-K3][Kernel] Optimize the low-M fused latent MoE tail (#54168) Yongye ZhuandOpenAI Codex 2026-08-28 10:45:36 -07:00
• 6f91e3d953 [Perf] Split xdrope_positions H2D copy into per-row transfers (#53412) Keyi LiandKeyi Li 2026-08-28 10:28:43 -07:00
• ffe690eca9 [MM][CG] Support ViT full CUDA graph for Idefics3 and SmolVLM (#47625) Chipmunk 2026-08-29 02:16:31 +09:00
• 06569a8696 [ROCm][Quantization][MOE] Enable fused shared experts for block-quantized FP8 (#53097) xuebwang-amd 2026-08-29 00:23:07 +08:00
• 11012d2a35 [Rust Frontend] Reduce copy in auxiliary frame resolution (#54148) Bugen Zhao 2026-08-29 02:01:17 +10:00
• 94a54f581e [XPU]bump up vllm_xpu_kernels to 0.1.14.1 (#54203) Xinyu Chen 2026-08-28 22:12:17 +08:00
• d1922cb5a7 [Bugfix] Restore multimodal support on the plain "vllm" throughput backend (#52168) Michał Ganczarenko 2026-08-28 12:28:25 +02:00
• 5f213ed159 [ROCm][CI] Warm up the RLHF dev server before the pause/resume timing checks (#53594) stefankoncarevic 2026-08-28 11:07:04 +02:00
• 3bb19cd8a3 [Bugfix][CI/Build] Fail closed when selected precompiled CUDA variant is unavailable (#52545) Jung Jiyu 2026-08-28 18:00:48 +09:00
• f956e1c343 [Kimi-K3][Bugfix] Fix low-latency GEMM fallback initialization (#54167) Thien Tran 2026-08-28 16:32:20 +08:00
• 9236159bb6 [ROCm][CI] Keep startup profiling from aborting when free memory grows (#53591) stefankoncarevic 2026-08-28 09:02:40 +02:00
• c01b50e390 [Test] Fix mock_current_vllm_config missing kernel_config in test_dflash2 (#54132) Qiming Zhang 2026-08-27 22:28:43 -07:00
• 6f7df92a8e [CPU][MLA] Fix prefill backend selection so MLA runs end-to-end on CPU (#51471) maobaolongandCodex 2026-08-28 12:36:55 +08:00
• e6bfe03ad7 Count store offers, not lookups, for CPU offload store_threshold (#52227) 3 people 2026-08-27 21:21:06 -06:00
• 53bf990502 [Bugfix][KV Cache] Prevent negative external block allocation (#52707) haic0andCursor 2026-08-28 11:10:45 +08:00
• b74ac0bc61 [Bugfix] Allocate packed outputs in fused_q_kv_rmsnorm so q_b_proj keeps its low-latency GEMM path at decode (#53109) Roy WangandClaude Fable 5 2026-08-28 10:08:27 +08:00
• f3b91f1af2 Remove wrongly added e2e test (#54099) Elvir Crnčević 2026-08-28 04:04:17 +02:00
• 8cfba696ee [Misc] Separate adaptive verification config validation (#54108) Nick Hill 2026-08-27 19:03:01 -07:00
• 5bbc58c0ad [Bugfix] Remove race in fused groupwise RMSNorm quantization (#54111) Misha GoinandOpenAI Codex 2026-08-27 22:01:22 -04:00
• 6ec92bcbc8 [Kimi-K3] Merge MLA gate into QKV-A projection (#54015) Thien Tran 2026-08-28 08:47:18 +08:00
• b2a6e9dca0 fix(build): correct preprocessor guard for GDN decode to fix Ampere c… (#52743) prakharPant 2026-08-28 05:51:44 +05:30
• d04d3e0a07 [Bugfix] Revert renderer warmup overlap to avoid fork deadlock (#54023) Kevin H. Luu 2026-08-27 16:36:33 -07:00
• 3cc015acd2 [Rust Frontend][gRPC] Enforce LoRA path validation across transports (#53756) Connor CarpenterandBugen Zhao 2026-08-27 19:00:12 -04:00
• e5377e6727 [Bugfix][Model] Fix K3 DSpark config for 96-head drafts (#54005) qizixiandOpenAI Codex 2026-08-27 15:58:28 -07:00
• 7c877062ac [kernel] Integrate FlashInfer BF16 CuTeDSL Low Latency GEMM (#50572) 3 people 2026-08-28 06:29:12 +08:00
• f79a2f5582 Add support for loading dflash2 model in speculators format (#53797) Fynn Schmitt-Ulms 2026-08-27 18:12:01 -04:00
• 479eeb32d2 [Bugfix][MRV2] Isolate sleep-mode KV allocations (#53508) Ronaldandaoshen02 2026-08-28 06:08:28 +08:00
• d262964e8c Upgrade tpu-inference to v0.28.0 (#54020) Dennis Yeh 2026-08-28 05:25:00 +08:00
• de9250ac9e [Attention] Enable masked MHA for GLM-5 head dimensions (#53785) Matthew BonanniandOpenAI Codex 2026-08-27 17:17:12 -04:00
• 3a9bfc209f [Bugfix][Parser] Scope reasoning-end detection to the current turn via turn-boundary tokens (#54089) Mikhail Podvitskii 2026-08-27 22:57:32 +02:00
• 7d5769b2d9 [Perf][DSv4] Use native CUDA SwiGLU clamp kernel for Humming MoE (throughput +1.40%) (#53685) Chauncey 2026-08-28 04:52:45 +08:00
• 9818bb3db8 [Kimi Perf] Tune hopper low latency gemm kernel, 4%~97% performance improvement (#54088) Wentao Ye 2026-08-27 16:48:53 -04:00
• 79378fe6e8 [Bugfix] Avoid TCPStore port collision for co-located non-DP Ray engines (#53666) Jeffrey Wang 2026-08-27 12:58:14 -07:00
• 9db222c6be [Rust Frontend] Align OpenAI request and response edge cases (#53218) Bugen ZhaoandNick Hill 2026-08-28 05:23:38 +10:00
• 4aab2b0ebe [Model Runner V2] Use MRV2 for all models by default (#53183) Nick Hill 2026-08-27 11:59:34 -07:00
• a18dbe49ad [Doc] Add EXAONE-4.0-1.2B to batch invariance tested models (#53839) Paarth SharmaandWentao Ye 2026-08-27 14:38:03 -04:00
• 674c284dbd Fix Humming MoE activation_output aliasing (#54056) Elvir Crnčević 2026-08-27 20:28:38 +02:00
• d7b6a35967 [Bugfix] Handle malformed namespace tools (#53763) Taneem Ibrahim 2026-08-27 14:18:16 -04:00
• 07242faa46 [Attention][DCP] Use FlashInfer native CP for MLA decode (#54012) Summer Yang 2026-08-27 10:32:45 -07:00
• 32ad1400d7 [ROCm][Perf] Fuse SWA q/kv RMSNorm and q FP8 group quant for DeepSeek-V4 (#53540) Shanshan Shen 2026-08-28 00:37:54 +08:00
• 9650dc73e2 [CI/Build][Hardware][NVIDIA] Add opt-in Rubin Docker builds (#53443) Shang Wangandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-27 12:06:44 -04:00
• d21c5b50a1 [Perf][GLM5.2] Fuse sparse MLA Q concatenation with head padding (#53878) Chauncey 2026-08-27 23:54:32 +08:00
• 76c0c6530e [Bugfix][Scheduler] Don't pad spec decode up to max_model_len (#53962) Nick Hill 2026-08-27 08:30:22 -07:00
• f18e29834b [Bugfix] Preserve parallel HY-V3 calls delivered in one streaming delta (#53965) Taneem Ibrahim 2026-08-27 11:26:10 -04:00
• d85708f7a4 [1/N][KV Connector] Identify externally transferable KV cache groups (#53779) Matthew Bonanni 2026-08-27 11:21:24 -04:00
• c7b0467c25 buffer size insuffient Dspark sd for FlashInfer MNNVL allreduce (#50932) Khushali Desai 2026-08-27 08:01:08 -07:00
• 6d11122607 [Elastic EP] Preserve AOT cache reuse during scaling (#53378) Itay Alroy 2026-08-27 17:55:44 +03:00
• 75dea9b4ae [Bugfix] Update FlashMLA for sparse decode workspace fix (#53755) Matthew BonanniandOpenAI Codex 2026-08-27 09:37:29 -04:00
• 4a6a3272e8 [Bugfix][Frontend] Let pooling requests set padding (#51157) Hert4 2026-08-27 18:06:20 +07:00
• 42e7942ac9 fix(config): guard LlamaBidirectionalConfig against missing hf_config.pooling (#50536) hcl 2026-08-27 18:56:07 +08:00
• 478ec3ea41 [Bugfix][KV Offload] Handle padded GPU cache storage (#54021) Jiangyun ZhuandOpenAI Codex 2026-08-27 18:52:59 +08:00
• 6c4aa8bec3 [RL] Support checkpoint-coordinate sparse NCCL weight updates (#53751) 3 people 2026-08-27 18:19:51 +08:00
• 50708f95ae [Bugfix][OpenAI] Fix streamed completion logprob offsets with echo (#47815) Ting SUN 2026-08-27 18:19:07 +08:00
• acd0af90e8 [Bugfix] Raise clear error on interleaved multimodal placeholder overcount (#53999) Hung Hoang (黄兴） 2026-08-27 16:49:40 +07:00
• fd57c4b7af [Rocm][CI] add dockerfile.xpu to rocm ci artifact (#53949) Charlie Fu 2026-08-27 03:06:48 -05:00
• 79651d6085 [Tools][Recipes] Improve sweep recommendations and short-alias parsing (#53946) Louie Tsai 2026-08-26 23:58:23 -07:00
• aa640684cc [Kimi K3][Kernel] Support DS conv-state layout in fused KDA decode kernel (#53396) Canlin Guo 2026-08-27 14:54:50 +08:00
• 833fa6233c [Benchmark] Warn on warm prefix cache for random serve runs (#53920) Zupeng WangandOpenAI 2026-08-27 13:51:17 +08:00
• b3af042abd Bugfix: use PCP slot mappings for PIECEWISE capture (#53869) Qiu Chunshuo 2026-08-27 13:39:36 +08:00