mirrors/vllm
0
0 Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2026-10-28 09:45:26 +08:00
Code Activity
23,864 Commits 469 Branches 215 Tags
eb703fbb3db3c2ea24656d1f737abf6a5a42f6d7
Commit Graph
23864 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
rasmithandMicah Williamson eb703fbb3d [BugFix] [Build] Fix for where setup.py fails with MAX_JOBS invalid literal error (#45188) 
Signed-off-by: Randall Smith <Randall.Smith@amd.com>
Co-authored-by: Micah Williamson <micah.williamson@amd.com>
 2026-10-26 17:10:53 -04:00
Zhiqi ZhangandFlora Feng 3f45273050 [Bugfix][Tool Parser] Anchor Jamba streamed tool arguments to what was sent (#55669) 
Signed-off-by: Zhiqi Zhang <zhiqizhangg@gmail.com>
Co-authored-by: Flora Feng <4florafeng@gmail.com>
 2026-10-26 21:03:03 +00:00
Yuhan ZhouCodexvllm-agent
 		2841f08b4a [KVConnector][Mooncake] Coalesce descending block ids to reduce descriptor number for Mooncake TE (#63142) 
Signed-off-by: Yuhan Zhou <zhouyuhan_@outlook.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 17:02:18 -04:00
azaidyClaude Opus 5.5vllm-agent
 		533906528c [Bugfix][Spec Decode] DFlash: run the aux hidden state FC in the draft dtype (#63436) 
Signed-off-by: Aliasger Zaidy <aliasger.zaidy@amd.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 20:58:41 +00:00
Lena OnyshchenkoArtur FierkaClaude Opus 5Kunshang Ji
 		12d54c2948 [XPU][MoE] Resize batched Triton MoE tiles and change grid order (#51605) 
Signed-off-by: Lena Onyshchenko <xonyshch@gmail.com>
Signed-off-by: Artur Fierka <artur.fierka@intel.com>
Co-authored-by: Artur Fierka <artur.fierka@intel.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
 2026-10-26 16:57:18 -04:00
Ilya MarkovandRoberto L. Castro f3bdb0e025 [Perf][MXFP4] Optimize activation quant for mxfp4. 3x kernel speedup, 5% throughput improvement (#63101) 
Signed-off-by: ilmarkov <markovilya197@gmail.com>
Co-authored-by: Roberto L. Castro <38211239+LopezCastroRoberto@users.noreply.github.com>
 2026-10-26 16:51:55 -04:00
jxjandOpenAI Codex 01b05cd69d [Bugfix][Frontend] Stop SSL certificate watchers when HTTP serving exits (#55699) 
Signed-off-by: git-jxj <65210887+git-jxj@users.noreply.github.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
 2026-10-26 16:48:13 -04:00
Itay EtelisMatthew BonanniClaude Opus 5.5
 		a2ecc8ac4d [Perf][HiSparse] Reduce residency hash probes (#61263) 
Signed-off-by: Itay Etelis <92247226+Etelis@users.noreply.github.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-26 20:44:15 +00:00
Andreas KaratzasClaudevllm-agent
 		3090266d88 [Bugfix][Kernel] persistent_topk: sync the CTA around radix barrier arrivals (#63475) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 20:40:32 +00:00
Blaze-DSPClaude Opus 5.5vllm-agentmergify[bot]
 		89f0c8c668 [MoE][Model] Gemma 4: skip CUDA-graph padding rows in the routing kernel (#62525) 
Signed-off-by: Blaze-DSP <devsp0703@gmail.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
 2026-10-26 16:37:03 -04:00
liuzhenweiKunshang Jivllm-agentCyrus Leung
 		0a95e4d73c [XPU][FP8] Use per-token dynamic FP8 activation quantization (#61362) 
Signed-off-by: zhenwei-intel <zhenwei.liu@intel.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
 2026-10-26 16:35:40 -04:00
Moein Khazraee c13d4a5a46 [KV Offload] Expose the processed prefix at request finalization (#57646) 
Signed-off-by: Moein Khazraee <moein@nvidia.com>
 2026-10-26 20:31:19 +00:00
ChaunceyLucas WilkinsonClaude Opus 5.5vllm-agent
 		af6665b5dd [Spec Decode] Support PCP with DSpark for DeepSeek-V4.1 (#61994) 
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 15:34:57 -04:00
Nick HillYibai MengClaudeleopckTaoyu ZhuJuntian Liu
 		78e2f79601 [Perf][MRV2] Stage buffer contents by reference and copy straight into UVA (#63633) 
Signed-off-by: Yibai Meng <yibaim@nvidia.com>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Yibai Meng <yibaim@nvidia.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: leopck <4161783+leopck@users.noreply.github.com>
Co-authored-by: Taoyu Zhu <zhutaoyu97@gmail.com>
Co-authored-by: Juntian Liu <Juntianl777@gmail.com>
 2026-10-26 12:00:58 -07:00
liranschourandvllm-agent 9633997299 [KV Offload][P2P] Fail faster: per-mode load timeout, shorter abort-ack and lookup timeouts (#63781) 
Signed-off-by: Liran Schour <lirans@il.ibm.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 20:54:31 +02:00
Qidong SuMax HuCodexBolin SunClaude Fable 5.1Max HuWei Zhao
 		9e3db48873 [MoE] Add locality-domain allocation for FlashInfer CuTeDSL NVFP4 (#61685) 
Signed-off-by: Qidong Su <qidongs@nvidia.com>
Signed-off-by: Max Hu <hyoung2991@gmail.com>
Co-authored-by: Max Hu <maxhu@nvidia.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Bolin Sun <bolins@nvidia.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Max Hu <hyoung2991@gmail.com>
Co-authored-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
 2026-10-26 14:37:01 -04:00
Lucas WilkinsonandClaude Opus 5.5 4135590ffe [Bugfix][MRV2] Skip adaptive verification cost profiling in the CUDA graph memory-profiling pass (#63431) 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
v0.33.0rc0 		2026-10-26 17:59:24 +00:00
Andreas KaratzasandClaude 442f3d989e [CI][ROCm] Take NIXL PD accuracy side-channel ports from below the ephemeral range (#63557) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-10-26 12:41:43 -05:00
Andreas KaratzasClaudevllm-agent
 		ad42b456f9 [CI][ROCm] Take MoRIIO P/D test port blocks from below the ephemeral range (#63453) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 12:41:40 -05:00
Andreas KaratzasandClaude 088e6a4d68 [CI][ROCm] Skip the AITER MoE garbage-rows test for fp8 per-tensor until the AITER fix (#63559) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-10-26 12:41:29 -05:00
Andreas KaratzasandClaude 94b35f3780 [Bugfix][ROCm] Keep the AITER GDN decode fast path off for the DS conv-state layout (#63558) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-10-26 12:40:53 -05:00
Andreas KaratzasandClaude 9744ada582 [Bugfix][ROCm][DSv4.1] Await krdy before K2's combine reads klen in index-source launches (#63576) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
 2026-10-26 12:40:43 -05:00
Kai HoangandFlora Feng eca54bd1d9 [Bugfix][Tool Parser] Keep Step3 parameter values that contain '<' (#60968) 
Signed-off-by: hpdkhoa <hpdkhoa2311@gmail.com>
Co-authored-by: Flora Feng <4florafeng@gmail.com>
 2026-10-26 17:19:57 +00:00
Zhang BowenCodexvllm-agent
 		29be2f2560 [Bugfix][Spec Decode] Preserve DP replica index for standalone draft models (#62301) 
Signed-off-by: MuonChaser <zbw.init@gmail.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 13:15:12 -04:00
Diao ShengjiaBugen ZhaoClaude Opus 5.5vllm-agent
 		20bf65ad33 [Bugfix][Rust Frontend] Read the text config under thinker_config (#61128) 
Signed-off-by: GoDiao <diaoshengjia@gmail.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-27 01:03:29 +08:00
Thuongvu HoBugen ZhaoClaude Opus 5.5vllm-agent
 		c36154742e [Bugfix][Rust Frontend] Use prefiller cache hits for P/D usage (#58885) 
Signed-off-by: Thuongvu Ho <thuongvuho@gmail.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-27 01:02:52 +08:00
Yifan QiaoClaudeBugen Zhaovllm-agent
 		9da5ba7e49 [Rust Frontend] Make HTTP request body limit configurable via VLLM_HTTP_MAX_REQUEST_BODY_SIZE (#47840) 
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-27 01:01:57 +08:00
Nicolò LucchesiAndy LoLucas Wilkinson
 		3e75ccba29 [Bugfix][DCP] Finalize the NIXL DCP interleave size before profiling (#62330) 
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Co-authored-by: Andy Lo <andy@mistral.ai>
Co-authored-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
 2026-10-26 17:41:28 +01:00
Rehan Khanandvllm-agent 5be249d0ad [Kernel] Allow registering OOT block-scaled FP8 linear kernels (#62026) 
Signed-off-by: Rehan Khan <Rehan.Khan7@ibm.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 12:36:48 -04:00
Monishverandvllm-agent c84ff6b3f4 [Kernel][Perf] Add H100 tuned configs for Llama-3.1-8B and Qwen3-8B to the batch-invariant matmul table (#62219) 
Signed-off-by: Monishver Chandrasekaran <monishverchandrasekaran@gmail.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 12:36:04 -04:00
Woosuk KwonCodexvllm-agent
 		4f140a23de [Bugfix] Initialize valid ModelOpt NVFP4 MoE dummy scales (#63698) 
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 09:31:41 -07:00
Roberto L. Castro 33a73daf17 [2/N][Bugfix] Make persistent TopK exact on candidate overflow (#62936) 
Signed-off-by: LopezCastroRoberto <rocastro@redhat.com>
 2026-10-26 12:25:16 -04:00
Rohan PotdarandClaude Opus 5.5 536854bf10 [ROCm] Enable expert parallelism for AITER Triton W4A16 MoE (#59255) 
Signed-off-by: Rohan Potdar <rohan.potdar@amd.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-26 11:18:35 -05:00
Luu Bao LongClaude Opus 5mergify[bot]vllm-agent
 		7f56f453ed [Bugfix][Core][MRV2] Profile the sampler with the params the warm-up actually uses (#57926) 
Signed-off-by: Luu Bao Long <longbaoluu68@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 12:05:29 -04:00
djramic df8b5d3781 [ROCm][CI] Raise MI300 Core Operation Kernels timeout to 90 min (#63760) 
Signed-off-by: Djordje Ramic <djoramic@amd.com>
 2026-10-26 10:59:23 -05:00
Aaron HaoandClaude Opus 5.5 50a184f435 [RL] Declare weight ownership and layout on ParamMeta (#63256) 
Signed-off-by: Aaron Hao <aaronhao@inferact.ai>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-26 23:42:30 +08:00
Chen Cheng 61b2527bdb [Perf][MoE] Add H200 FP8 fused MoE config for Beam-1.0 TP8 (#62763) 
Signed-off-by: Chen Cheng <85005591+ischencheng@users.noreply.github.com>
 2026-10-26 15:39:00 +00:00
Li Wangandwangli 0abec7309e [Model] Add native support for Lfm2BidirectionalModel (#62201) 
Signed-off-by: wangli <wangli858794774@gmil.com>
Co-authored-by: wangli <wangli858794774@gmil.com>
 2026-10-26 15:31:39 +00:00
ztysthJack BrownCodexvllm-agent
 		e113018f50 [Bugfix][KV Offload] Preserve running request prefixes across LRU store batches (#61050) 
Signed-off-by: Jack Brown <rnk.liulw@gmail.com>
Co-authored-by: Jack Brown <rnk.liulw@gmail.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 17:30:22 +02:00
Avaya AggarwalClaudeJee Jee Livllm-agent
 		14d593115d [Bugfix][LoRA] Fix double pinning of 3D MoE LoRA and EP dummy LoRA sizing (#61636) 
Signed-off-by: onepunchmonk <aggarwal.avaya@yahoo.com>
Signed-off-by: OnePunchMonk <aggarwal.avaya@yahoo.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 23:20:59 +08:00
liranschourandvllm-agent 51d2a85594 [Bugfix][KV Offload][P2P] Never block in ZmqConnection.send() (#63708) 
Signed-off-by: Liran Schour <lirans@il.ibm.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 16:55:56 +02:00
javieratomicClaude Opus 5.5vllm-agent
 		14aefcd4f9 [Bugfix][KV Offload] Hold promotions after a cache reset until jobs report (#62929) 
Signed-off-by: Javier López <59578422+javieratomic@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 16:55:43 +02:00
Sunandvllm-agent 8a4b018f1c [Bugfix][KV Offload] Roll back failed P2P NIXL registration (#63628) 
Signed-off-by: levius <2114377220@qq.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 16:54:54 +02:00
Tony LinandNicolò Lucchesi 74340c7300 [docs] correct NIXL hybrid SSM heterogeneous TP support (#63036) 
Signed-off-by: Tony Lin <tony.lin@intel.com>
Co-authored-by: Nicolò Lucchesi <nicolo.lucchesi@mistral.ai>
 2026-10-26 14:16:11 +00:00
Vedant AgarwalClaudevllm-agent
 		8c6e770a1c [Bugfix][KV Offload][OBJ] Namespace object keys by blocks_per_chunk (#63655) 
Signed-off-by: Vedant Agarwal <vedantagwl10@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 16:15:53 +02:00
sudhanshu shuklasudhanshu shuklavllm-agent
 		2f1cb151fd [Bugfix] Preserve fp16 in legacy fused MoE LoRA (#59931) 
Signed-off-by: sudhanshu shukla <sudhanshu112233shukla@users.noreply.github.com>
Co-authored-by: sudhanshu shukla <sudhanshu112233shukla@users.noreply.github.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 21:45:35 +08:00
Misha Goin fcb787055e [Perf][DeepSeek V4] Declare MXFP8 MoE activations and run them as block FP8 on Hopper (#61875) 2026-10-26 09:37:42 -04:00
Shanshan Shenandvllm-agent d0d681e40d [ROCm][Perf] Make MXFP4 the default indexer K cache dtype for DeepSeek-V4 (#63669) 
Signed-off-by: shen-shanshan <467638484@qq.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 21:14:55 +08:00
Sy03andvllm-agent 373614f66d [Bugfix][Multimodal] Release consumed MRV2 CPU inputs (#57368) 
Signed-off-by: Sy03 <1370724210@qq.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
 2026-10-26 20:56:11 +08:00
VegetogandCodex a64fc94081 [Bugfix][KV Offload] Preserve new loads during same-step resume (#60574) 
Signed-off-by: Vegetog <110553275+Vegetog@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
 2026-10-26 14:51:12 +02:00