Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 4013f3ad6c [ROCm] Fix CI runtime and tests for MI355 DPX (#58244) 6 people 2026-09-24 23:15:11 -05:00
• 44257c2e35 [Feature] Triton kernel dispatcher (#43048) wangxiyuan 2026-09-25 12:07:43 +08:00
• b05f7d5fbc [PP][XPU]Add the flag to control microbatch feature on MRV2+PP (#55145) YiSheng5andKunshang Ji 2026-09-25 12:05:35 +08:00
• 442bd00d9f [Perf][Distributed] Add low-SM multimem reduce-scatter for SM100/SM103 (#55072) Yongye ZhuandSummer Yang 2026-09-24 20:27:21 -07:00
• 8b365ff949 [ROCm][Build] Filter crate tags from vLLM version detection (#57744) Reidandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-25 11:06:05 +08:00
• 4d79d5c638 [ROCm] Cut 69 wasted contiguous copies per decode step from the skinny GEMM path (#58566) 3 people 2026-09-24 20:00:58 -07:00
• f55419c43c [Bugfix][Quantization] Fix MXFP8 startup crash on layers below mm_mxfp8 shape limits (#54223) Samuel Kimandmgoin 2026-09-25 04:54:11 +02:00
• 6936e77ba4 [Perf][Engram] Serialize offloaded lookups and pack host tables into huge pages (#56926) 4 people 2026-09-24 19:52:02 -07:00
• 48d8880d09 [Bugfix][Qwen4Exp] Keep pinned PLE prefetch ids out of the CUDA graph pool (#58489) 3 people 2026-09-24 19:51:50 -07:00
• 3059155b47 [ROCm][Perf] MXFP8 GEMM on native 32x32 block scales for gfx950 (#58510) Fangzhou AiandClaude Opus 5.5 2026-09-24 22:26:30 -04:00
• 055f150df5 [Perf][Rust Frontend] Make histogram observations lock-free (#58574) 3 people 2026-09-25 10:25:10 +08:00
• 34e104f6fe [gRPC] Fix ping tolerance so long non-streaming RPCs are not dropped (#55102) gongwei-130andClaude Fable 5.1 2026-09-24 19:14:32 -07:00
• 65c5e24689 [ROCm][CI][AITER Coverage] Harden MoE sorting-backend/dispatch env-var test matrix (#58393) Divakar Verma 2026-09-24 22:14:02 -04:00
• 32fc82343f [Bugfix][LogitsProcessor] Validate ':' separator in custom logits processor FQCN (#56020) 김재억 2026-09-25 10:49:20 +09:00
• 609731e8f5 [Perf][DSv4.1] Restore the fused query RMSNorm + MXFP8 quantization path (#57679) Juntian Liu 2026-09-24 18:39:03 -07:00
• 3652f35e7c [Bugfix][Quantization] Refresh online NVFP4 scales before reload post-processing (#57954) 4 people 2026-09-24 18:36:08 -07:00
• 16070ed7e9 [Minimax-M3][Perf] Use triton_mrope for vision tower + int64 offset fix for triton_mrope (#58526) 3 people 2026-09-25 09:30:32 +08:00
• ade1056ee5 [Kimi-K3][Perf] Dispatch GEMM for vision patch embedder (#58527) Thien Tran 2026-09-25 09:30:11 +08:00
• 7dbd0a8d22 [Bugfix][MRV2] Treat padded prompt tails as spec-decode rows for hybrid models (#58434) Nick HillandClaude Opus 5.5 2026-09-24 18:29:53 -07:00
• 155b988373 [CI] Run DFlash2 NVFP4 acceptance test on B200; skip it on H200 35GB MIG (#58496) 3 people 2026-09-24 18:29:44 -07:00
• 34ee85d5ca [Perf] Batch Mamba2 prefill SSM state saves, removing GPU<->CPU syncs (#49371) Samuel KimandClaude 2026-09-25 03:11:19 +02:00
• a811738a60 [Bugfix] Accept EOS after grammar finish in outlines backend; reject json_object at validation (#57743) SIDDARTHA REDDY 2026-09-24 20:17:10 -04:00
• 39724758e7 [Bugfix] GLM-5.3-Flash: launch the kpool paged MQA logits in the varlen mode its schedule was built with (#55270) Yifan QiaoandClaude Fable 5.1 2026-09-24 17:13:55 -07:00
• 064bc747ab [Bugfix][Frontend] Count reasoning tokens for Harmony, DeepSeek-V3 and Step3 parsers (#58626) Samyabrata MajiandFlora Feng 2026-09-25 05:42:29 +05:30
• b3a87b2d04 [Bugfix][KVConnector] Finalize saves on steps without a forward (#57775) 3 people 2026-09-24 17:06:36 -07:00
• 7833ff25dd Release prompt_embeds tensor when its InputBatch slot is freed (#57988) Khushali Desai 2026-09-24 17:00:44 -07:00
• 7a9406690e [KV Connector] Fix DecodeBench fp8 fill values and add a startup fill mode (#58472) 4 people 2026-09-24 16:55:47 -07:00
• 71891bdb94 [Perf] Parallelize registered CUDA Triton kernel warmup at startup (#58582) Misha GoinandCodex 2026-09-24 19:52:00 -04:00
• d5051abaf1 [Bugfix][Mamba] Restore prompt-tail prefix-cache hits with MTP (#58368) 3 people 2026-09-25 02:46:42 +03:00
• 4f72bd0152 [Bugfix][Quantization] Give LM heads standard linear metadata (#58444) Misha Goin 2026-09-24 19:39:49 -04:00
• 9ac3b7c69f [ROCm][CI] Mirror the DSv4-Flash disaggregated DP EP group on MI355 (#58558) stefankoncarevicandAndreas Karatzas 2026-09-25 01:38:25 +02:00
• f6aa2919bc [Bugfix][Core] Keep every multimodal feature in the partial-block KV event (#58288) 3 people 2026-09-25 07:25:01 +08:00
• bd4cb3fe4e [BUGFIX] fix ovis2_5 multimodal tokens (#52623) Miłosz Grunwald 2026-09-25 01:18:54 +02:00
• 6a86bf626e [PD][PushConnector] Record last activity of remotes on the D side (#52245) snadampalandNicolò Lucchesi 2026-09-24 18:15:32 -05:00
• f5b40fcdfe [CI] Report to CRCR after all jobs finish, gated on the build's long pole (#58628) Andrey Talman 2026-09-24 19:02:30 -04:00
• 7edb27f68e [XPU][CI] enable prompt embeds tests on XPU (#58283) Fanli LinandKunshang Ji 2026-09-25 06:56:59 +08:00
• e6c07ea576 [Bugfix][KV Cache] Fix incremental multimodal block hashing (#51694) Jellow 2026-09-25 06:38:03 +08:00
• 36f94d5fe8 [Bugfix][Outlines] Fix EOS termination and unconstrained masks after rejected drafts (#58612) Nick Hill 2026-09-24 14:55:56 -07:00
• 04730e8270 [DFlash] Capture the context K/V precompute in the draft CUDA graph (#57632) Juntian LiuandClaude Opus 5.5 2026-09-24 14:38:04 -07:00
• e33de821c0 [ROCm][CI] skip the ROCm MRV1 default where MRV1 cannot serve the config (#58535) stefankoncarevicandAndreas Karatzas 2026-09-24 22:36:36 +02:00
• 0908116dd9 [Bugfix] Pass quant_config to DiffusionGemma's ParallelLMHead (#48521) Aaron KangandMisha Goin 2026-09-24 16:17:02 -04:00
• ab3de6edf2 [Perf] DiffusionGemma: constrained reads over the request's logprob_token_ids (#58216) 5 people 2026-09-24 14:15:08 -06:00
• 92e9c6bf13 [ROCm] Fix misrouting race-condition in multi-decode P/D disagg with mori-io (#51681) 4 people 2026-09-24 21:52:43 +02:00
• 90a9515006 [Cleanup] Remove online quantization support in fp8.py in favor of online shorthands (#53585) 3 people 2026-09-24 21:48:14 +02:00
• 58aa2f1ef1 [Refactor] Move auxiliary files out of the repository root (#58572) Misha Goin 2026-09-24 15:38:56 -04:00
• cd06e81c41 [Bugfix][Quantization] Add Humming to the W4A8 (INT4xFP8) MoE oracle (#58427) Misha Goin 2026-09-24 15:37:33 -04:00
• e773247f8a [Fast Start] Wait for weight cache daemon readiness (#58370) Xun Sun 2026-09-25 03:18:04 +08:00
• e30559b58f [Core] Skip JIT monitor when JIT warmup is disabled (#58590) Nick HillandKimi Code 2026-09-24 12:08:16 -07:00
• 074d57fce1 [Bugfix] Keep JIT warmup under enforce-eager when fault tolerance is on (#58593) Nick HillandKimi 2026-09-24 12:07:32 -07:00
• 5b3f280a6d [Bugfix] Count unsplit Idefics3 image patches (#48760) 3 people 2026-09-25 02:36:26 +08:00
• 8b84e15066 [transformer] RMSNorm matching for alternative rsqrt (#54461) Thomas OrtnerandHarry Mellor 2026-09-24 18:33:06 +02:00
• 09fe178dba Revert "[DSpark] Support pipeline-parallel targets in aggregated serving (#56956)" (#58484) Nick Hill 2026-09-24 08:48:36 -07:00
• d636456b1f [Frontend] Remove the slow tokenizer mode (#58545) Harry MellorandClaude Opus 5.5 2026-09-24 17:48:28 +02:00
• 344fcc252a [Bugfix] Resolve the Hub revision once per repo (#56092) 4 people 2026-09-24 17:48:24 +02:00
• a7d203efe5 fix(config): apply presence_penalty/frequency_penalty from override-generation-config (#50769) 3 people 2026-09-24 23:48:18 +08:00
• 26f49e336a [Perf][Attention] Bound FlashInfer prefill dequantization scratch (#57918) Yueh-Ting (eop) Chen 2026-09-24 22:42:08 +08:00
• 3c42385d7d [Refactor] Remove dead or duplicate tests (#58446) Wentao Ye 2026-09-24 10:29:56 -04:00
• d881812500 [Chore] Use Transformers v5 names and drop redundant processor use_fast (#58550) Harry MellorandClaude Opus 5.5 2026-09-24 16:14:10 +02:00
• 484c211fe8 [Bugfix][Pooling] Fix JinaVL label configuration and restore multimodal tests (#57347) Linze Shi 2026-09-24 06:51:25 -07:00
• cccf7e1376 Remove .gemini/ and CLAUDE.md (#58541) Harry MellorandClaude Opus 5.5 2026-09-24 14:57:23 +02:00
• 721d0e5c11 [ROCm][DSv4.1][Perf] Emit MXFP8 from the sparse decode reduce and run wo_a as a grouped FP8 GEMM (#58456) Fangzhou AiandClaude Opus 5.5 2026-09-24 07:58:14 -04:00
• dcfc17e0b1 [XPU] enable XPU GRAPH by default (#51600) liuzhenwei 2026-09-24 19:17:25 +08:00
• f5a78f2ad7 [Multimodal] Reuse the supplied tokenizer in the MiniMax-M3 VL processor (#58460) Zijing Liu 2026-09-24 03:47:50 -07:00
• 5747d4500a [Perf][Frontend] Offload streaming derender detokenization (#57528) SHREY GAJJAR 2026-09-24 03:39:21 -07:00
• f34a0e0797 [CPU] Gate the AVX10.2 paths on compiler support (#58133) 3 people 2026-09-24 15:37:53 +05:30
• cfd5c20286 [XPU][UT] Align HF and vLLM inputs for Qwen2 embedding test by preventing Sentence Transformers from applying chat template (#58117) Ziyang Ma 2026-09-24 18:07:37 +08:00
• e430f7421f [Bugfix][Rust Frontend] Prevent MM timing from enabling debug tracing (#58378) 3 people 2026-09-24 17:22:23 +08:00
• bbd7c24d6e [MRV2] Validate MRV2 entrypoint logits processors (#57728) Taneem Ibrahim 2026-09-24 05:07:26 -04:00
• 5963795ec7 [Attention][CPU] Run Zen CPU encoder attention on zentorch SDPA (#54508) 3 people 2026-09-24 14:35:44 +05:30
• 14f98cbe5d [Bugfix][XPU] store the pointer raw bit pattern instead of its numeric value (#54514) Yejing LaiandKunshang Ji 2026-09-24 17:05:26 +08:00
• c0c35ddd8e [Bugfix][CI] Fix the flaky sharded-sampling tests, and the engine teardown need (#58342) 3 people 2026-09-24 11:05:11 +02:00
• 23110a0c9e [Bugfix] Capture prefill kernels for mixed FULL graphs (#58275) Andreas KaratzasandOpenAI Codex 2026-09-24 04:05:07 -05:00
• bbc4ddeca9 [Bugfix][V1] Reject encoder-cache hits with mismatched embedding counts (#57696) jackLeiandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-24 01:40:47 -07:00
• b44895cf93 [CPU] Use pre-built triton (#58140) Li, Jiang 2026-09-24 16:12:34 +08:00
• 28282ffe34 [Feature][Frontend] Request JSON body debug logging on --enable-log-requests flag (#58163) Talor Abramovich 2026-09-24 11:08:40 +03:00
• 00b7847c80 [Perf] Use Conv3dLayer for MiniMax M3 patch embedding (#58512) Thien TranandOpenAI Codex 2026-09-24 15:41:36 +08:00
• 1291bdb85e [EPD][Model Loader] Skip language-model checkpoint shards for --mm-encoder-only (#58086) 3 people 2026-09-24 15:41:15 +08:00
• c3f5270270 [ROCm][CI] Add the MI355 TurboQuant t3nc mirror (#58432) Andreas KaratzasandOpenAI Codex 2026-09-24 02:16:09 -05:00
• 67ddac3ea8 [CPU][GDN] Support NIXL DS convolution-state layout (#53300) 3 people 2026-09-23 23:55:20 -07:00
• 153f2ba1b9 [Compile][CI] Honor Triton cache overrides and add AMD timeout headroom (#58474) Andreas KaratzasandCodex 2026-09-24 01:27:07 -05:00
• e028554a1f [Rust Frontend] Support --sse-keep-alive-interval (#58306) 3 people 2026-09-24 14:19:14 +08:00
• cf78797cb7 [Rust Frontend] Pass vision preprocessing context for Nemotron-H (#57634) Biswa PandaandCodex 2026-09-23 23:08:26 -07:00
• ee2f3cdf12 [Bugfix][Spec Decode] Separate DSpark width from MTP stage validation (#54631) lucamotzandClaude Opus 5.5 2026-09-24 08:08:22 +02:00
• 44bf011f44 [Dependency] Upgrade FlashInfer version to 0.7.0 (#58069) 4 people 2026-09-24 01:45:23 -04:00
• a1b6763df6 [Rust Frontend] Accept custom chat roles for HF templates (#58311) Bugen ZhaoandClaude Opus 5.5 2026-09-24 13:43:34 +08:00
• a5603f21ae [Rust Frontend] Recognize new frontend-owned serve args as unsupported or no-op (#58330) Bugen ZhaoandClaude Opus 5.5 2026-09-24 13:43:24 +08:00
• 8b06a343ec [Bugfix][CPU][MoE] Fix out-of-bounds write and segfault when router weights are fp32 (#56168) 3 people 2026-09-24 01:15:17 -04:00
• 7f1a5398e9 Fix full logprobs in token-in/token-out responses (#58488) aoshen02 2026-09-24 12:46:16 +08:00
• 70fc359d25 [Bugfix][KV Offload] Retain offload event metadata through batch translation (#57453) Kapil AryaandOr Ozeri 2026-09-23 21:09:12 -07:00
• 4fb767ff0a [Kernel] Resubmit PR 48666 - Gemma4 FP8 KV FA4 head dim 512 backend selection (#53175) Jhao-Ting Chen 2026-09-23 20:29:06 -07:00
• e78e367c6e [ROCm] Give turboquant boundary layers a layout-compatible backend (#54988) 3 people 2026-09-24 05:00:07 +02:00
• e6dc16cebd [PCP] Support prefill context parallelism with data parallelism (#57075) 7 people 2026-09-23 22:32:36 -04:00
• 6697a7dd30 [Scheduler] Tune --long-prefill-token-threshold adaptiveness (#58459) Robert Shaw 2026-09-23 22:23:18 -04:00
• e9f167129c [CI][ROCM] Add the Fusion E2E TP2 Quick group on MI355, and the AITER MLA fix it needs (#58369) 4 people 2026-09-24 04:01:58 +02:00
• bcdacfc1ff [Test][Determinism] Cover chunked prefill in the batch-invariance suite (#55612) Bob OkandWentao Ye 2026-09-24 04:21:56 +03:00
• 9ef37771be [Bugfix][DSA] Bound DeepSelect sentinel columns in the sparse top-k remap (#58215) Juntian LiuandClaude Opus 5.5 2026-09-23 17:54:27 -07:00
• e0f03eb4f4 [CI] Share BF16 baselines across quantization comparison tests (#58469) Aarushi JainandAndreas Karatzas 2026-09-23 19:26:09 -05:00
• 7ef0d69ea0 [5/12][ci-selector][CI] Skip the Proton GPU test when another CUPTI tool is injected (#58455) Kevin H. LuuandClaude Opus 5.5 2026-09-23 17:25:41 -07:00
• 310f15d354 [Bugfix][MRV2] Align dummy idx_mapping dtype to avoid runtime jit (#58462) Nick Hill 2026-09-23 16:54:48 -07:00
• 89d9fd83e2 [CI] Disable JIT warmup by default in VllmRunner (#58452) Misha Goin 2026-09-23 19:53:58 -04:00