Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• f25c580af1 [XPU][Dockerfile] Update UCX install (#53817) liuzhenwei 2026-08-27 12:34:03 +08:00
• 5acc1c4e4b [Model Runner V2][Spec Decode] Skip DP sync before EAGLE/MTP draft prefill (#53694) Giancarlo Delfin 2026-08-26 20:59:35 -07:00
• ca90b9e7d4 [warmup] overlap renderer warmup and engine core initialization (#52764) Ning Xie 2026-08-27 11:42:08 +08:00
• f8d38fbc87 [EC Connector] EC Offloading Connector use events instead of StepTracker (#49994) omerpaz95andClaude Opus 5 2026-08-27 06:41:46 +03:00
• 94d96e2446 [Bugfix] Release CUDA graph profiling memory before KV cache allocation (#53955) Wei Zhao 2026-08-26 21:35:59 -04:00
• 2dc53bae17 [AMD][BugFix] Add gpu_sync_allowed to ROCm AITER FA backend (#53641) rasmith 2026-08-26 20:23:04 -05:00
• 51d1b45182 [Bugfix][GPT-OSS] Fix strict tool-call grammar to accept Harmony renders (#52222) yzong-rhandBen Browning 2026-08-26 21:05:09 -04:00
• d1e5e66ee3 [Bugfix] Restore portable all2all backend default (#53952) vllm-agentandKevin Luu 2026-08-26 16:19:49 -07:00
• 3fd72ffd7b [Bugfix][Rust Frontend] Fix LogprobsTensors wire schema mismatch (#53939) 3 people 2026-08-27 08:16:43 +09:00
• a27e88c862 [XPU][CI] increase timeout of extract_hidden_states tp2 (#53862) liuzhenwei 2026-08-27 06:57:36 +08:00
• f18d0ba90d [Doc] Add Granite 3.1 series to batch invariance tested models (#53650) 3 people 2026-08-26 15:29:10 -07:00
• 76cfe1cd88 [Kimi K3 Perf] Optimize eh_proj linear calculation, 12.9 ~ 25.2% kernel performance improvement (#53942) Wentao Ye 2026-08-26 17:08:02 -04:00
• 161ffd37d2 [Bugfix] Guard tool call argument JSON parsing in chat message postprocessing (#48922) Venkat Balaji SandBen Browning 2026-08-27 01:56:00 +05:30
• 00ca27d0c1 [XPU][TEST]Move LoRA Multimodal to B70 in Intel GPU CI (#53841) xiangdong 2026-08-27 04:04:10 +08:00
• 0a5ad6f0d4 [Model] Remove unused DeepSeek V4 top-k buffer helper (#53697) Woosuk Kwon 2026-08-26 11:59:09 -07:00
• 1a085dadf2 [ROCm][K3] Extend FP8 asm MLA prefill to non-divisor small head counts (#51040) xiaohuguo2023andCursor 2026-08-26 18:47:16 +01:00
• 17da48596c [Bugfix][DP] Synchronize the device on pause completion (#52914) aoshen02 2026-08-27 00:50:33 +08:00
• 8d301f075b [Bugfix][ROCm][Disagg] Fix MoRIIO shared KV memory region registration (#53698) Avinash Paulandavininjamay8 2026-08-26 21:40:57 +05:30
• c71f6f8a81 [Bugfix] Thread kv_transfer_params into engine for /inference/v1/generate (disagg) (#42644) Konstantin DunasandOpenAI Codex 2026-08-26 08:42:52 -07:00
• 2ab187430b [Bugfix] Make Gemma4 MTP suppress_tokens masking CUDA-graph-safe (#53884) Luciano MartinsandLuciano Martins 2026-08-26 12:39:19 -03:00
• 657f9b9ce2 [ROCm][DSV4][Perf] Fuse DeepSeek V4 C4 compressor GEMMs (#53838) Fangzhou Ai 2026-08-26 08:36:14 -07:00
• 080a66a69c [Bugfix][ROCm] Capture CUDA graphs on the current stream (#53818) fanxingran 2026-08-26 23:16:45 +08:00
• 28b484e5d7 [Model] Pixtral: use packed multimodal encoder attention (#52185) Oliver HolworthyandOpenAI Codex 2026-08-26 15:14:40 +01:00
• c47ca4aaeb [Mypy Fix] Mypy fix for "vllm/model_executor/models/[tT]" (#53466) 3 people 2026-08-26 07:12:11 -07:00
• 903a02192f [Bugfix] Handle empty FlatLogprobs slices and delta output (#53704) Tony 2026-08-26 21:00:54 +08:00
• 6a5e8f5979 [DeepEPv2] Support MXFp8 Activation Scale Dispatch (#51398) Robert ShawandRobert Shaw 2026-08-26 08:26:11 -04:00
• 15baeaee98 [Doc] Fix local input path in run-batch examples across docs (#53220) QWERQWERQWE86 2026-08-26 20:17:56 +08:00
• 2cd6c664c1 [MoE] enable all2all fi_one_sided by default (#53311) Artem Perevedentsev 2026-08-26 15:03:45 +03:00
• 31739ceb0b [CI/Build] Improve pre-commit fail message (#53866) Cyrus Leung 2026-08-26 19:22:48 +08:00
• 61d4f56635 [Offloader] Offload submodules that make_layers never reaches (#53120) 4 people 2026-08-26 18:37:00 +08:00
• 7a9993878c [CI] forward fix CRCR report step in the torch-nightly lane (#53732) Andrey Talman 2026-08-26 05:51:26 -04:00
• b821d7b2a6 [CI] Build mamba-ssm with C++20 for torch 2.14 nightly compatibility (#49600) Andrey TalmanandShengqi Chen 2026-08-26 05:31:08 -04:00
• 73bd7c83c4 [Bugfix][Processor] Replace bare asserts with ValueError in DeepseekVLV2/OCR processors (#53854) adisivaprasadandadisivaprasad 2026-08-26 02:10:54 -07:00
• 2267d3b112 [AttentionBackend][HPC-ops] update hpc rope norm to support stride kv cache (#53705) 3 people 2026-08-26 16:20:28 +08:00
• 75aa189ba0 [LoRA] Cleanup VocabParallelEmbedding (#53843) Jee Jee Li 2026-08-26 16:19:35 +08:00
• e376d45e82 [Config] Delegate PCP compatibility checks to PCP manager (#53853) Qiu Chunshuo 2026-08-26 16:16:01 +08:00
• f14369c7f9 [Bugfix][Mooncake] Fix Mamba prefill truncation ordering (#53663) Ziming HuangandOpenAI Codex 2026-08-26 16:08:37 +08:00
• cde7ba92da [CI/Build][The Rock] Use model_class_overrides so spawned worker can use test PredictableLlamaForCausalLM class when worker spawned using Python 3.14 (#49218) rasmithandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-26 02:55:07 -05:00
• 796822d141 [Hardware][AMD][Perf][Bugfix] Update ROCr and clr in base image (#53712) Matt 2026-08-26 00:38:20 -07:00
• ee5ae2ae70 [RL] Add rank-local IPC weight updates (#52497) 4 people 2026-08-26 15:36:57 +08:00
• a5f82882a6 [Bugfix][Model] Honor Molmo2 dummy video num_frames >= 2 override (#53830) Hung Hoang (黄兴） 2026-08-26 13:47:02 +07:00
• b1fbbc2ade BugFix(PCP): use persistent input buffers for PIECEWISE CUDA graphs (#53515) 3 people 2026-08-26 13:58:13 +08:00
• 46638857fd [Bugfix][Structured Output] Preserve reasoning adapters for shared parser engines (#52830) Flora Feng 2026-08-26 00:37:36 -04:00
• 044b05220e [Kernel][Perf] Tune fused_moe FP8 config for Qwen3.5 on L40S (+7%) (#53819) 3 people 2026-08-26 12:02:26 +08:00
• f25c2692fe [Frontend] Add /v1/messages/render endpoint for the Anthropic Messages API (#45803) Hyunkyun Moon 2026-08-26 12:33:59 +09:00
• a5e0045433 [Refactor] Remove dead code quantization 2 (#52821) Wentao Yeandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-25 23:28:44 -04:00
• 7156c63bef [SM100] Hdim 256 optimized (#52980) simon-veitner-redhat 2026-08-26 05:28:40 +02:00
• f6130145c8 [Bugfix][LoRA] Enable tower/connector LoRA for Qwen3-Omni (#53557) linitra24 2026-08-26 10:44:41 +08:00
• 88e1b11313 [Perf] Tune FlashInfer all-reduce thresholds for single-node TP8 on SM103 (#53606) Jee Jee Li 2026-08-26 09:46:56 +08:00
• d125b540b7 [Agents] Add CUDA IMA debugging skill (#53702) Thien TranandCodex 2026-08-26 09:09:24 +08:00
• a447955aca [Kimi Bug] Fix k3 torch.AcceleratorError: CUDA error: an illegal memory access was encountered (#53773) Wentao Ye 2026-08-25 20:38:48 -04:00
• bc2d63e650 [Kimi K3][Kernel] Enable low-latency decode GEMM dispatch on SM100 (#53534) Canlin Guo 2026-08-26 07:32:28 +08:00
• bc39ded3c4 [Bugfix][KV Offload] Defer request-level cascade of in-flight primary keys (#53329) Almog Tavor 2026-08-26 01:20:03 +02:00
• 80771bbbdd [Core] Disable fuse_allreduce_rms under VLLM_BATCH_INVARIANT (non-deterministic under TP) (#51292) 3 people 2026-08-25 12:16:09 -07:00
• 0e30bd62fe [CI Bug] Fix kimi test AssertionError: Aligned Mamba state indices must be precomputed (#53766) Wentao Ye 2026-08-25 15:06:14 -04:00
• 384d079697 [Perf] Autotune batch invariance triton kernel in blackwell, 33.6% E2E latency reduction (#53649) Wentao Ye 2026-08-25 15:06:01 -04:00
• e6cc0899b6 [Bugfix][MRV2] Run cudagraph memory profiling in a throwaway graph pool (#53682) Nick Hill 2026-08-25 11:52:15 -07:00
• c9331d8064 [CI] Preserve Rust Docker cache across commits (#53290) Misha GoinandOpenAI Codex 2026-08-25 14:13:36 -04:00
• 299ebd094a [Bugfix][Multimodal] Reject malformed base64 audio with 400 instead of 500 (#53744) Moe Huzaifa 2026-08-25 12:31:42 -05:00
• 7de96050c5 [Spec Decode] Enable adaptive DSpark on SM100 sparse MLA (#52783) Misha Goin 2026-08-25 13:31:15 -04:00
• 70021a0081 [Mypy Fix] Mypy fix for "vllm/model_executor/models/[gG]" (#53616) Wentao Ye 2026-08-25 12:47:16 -04:00
• 217a0b5cef [Bugfix][Frontend] Apply the stop string limit to Cohere requests (#53750) Moe Huzaifa 2026-08-25 11:41:08 -05:00
• 3e833933dc [Bugfix][Frontend] Keep credentials out of the Rust frontend launch log (#53738) Moe Huzaifa 2026-08-25 11:32:12 -05:00
• 19406fae28 [Bugfix][Tokenizer] Replace bare asserts in the DeepSeek V4 encoder (#53747) Moe Huzaifa 2026-08-25 11:30:23 -05:00
• 3454335f7b [Bugfix] Resolve B12X modules before Dynamo tracing (#53326) Luke Alonso 2026-08-25 09:20:15 -07:00
• bc11ecaf4e [CI] Add GSM8K accuracy test for amd/DeepSeek-V4-Flash-MXFP4 (#50632) Colin ZandAndreas Karatzas 2026-08-25 09:06:58 -07:00
• 10ade93979 [XPU] update key supported models (#53494) Yan Ma 2026-08-25 23:10:48 +08:00
• 41729fc53b [K3 Perf] Optimize k3 mamba metadata preparation, 6.6~7.6x kernel performance improvement (#52388) Wentao Ye 2026-08-25 11:00:28 -04:00
• d5cadcee86 [Feature][DSpark]: Logprobs adaptive verification (#52242) Naveenraj KamalakannanandBenjamin Chislett 2026-08-25 10:22:33 -04:00
• ca29cd48ba [Quantization][Humming] Support MXFP4 weight + block-FP8 activation for MoE (#51332) 4 people 2026-08-25 16:21:00 +02:00
• 9c8e90eb26 [Bugfix] Handle parenthesized Gemma4 tool calls (#53657) Taneem Ibrahim 2026-08-25 08:21:42 -04:00
• 2014e47485 [Pooling UX] Improve serve --task error guidance (#53467) Taneem Ibrahim 2026-08-25 08:20:10 -04:00
• afc91fa0d2 [Profiler] Fix start_profile permanently no-op after max_iterations auto-stop (#51839) Elvir Crnčević 2026-08-25 14:13:29 +02:00
• d9fbe526c0 [ROCm][CI] Skip ModernBERT FP8 MTEB test when no FP8 ScaledMM kernel exists (#53589) djramic 2026-08-25 10:07:09 +02:00
• 12c84b98dd [Config][EC] Normalize producer-only encoder config (#53656) Tianyu Guo 2026-08-25 15:53:04 +08:00
• 59d7fc92ea [Bugfix][DeepSeek V4] Handle trailing system messages in prompt rendering (#51262) Jiahao LiangandChauncey 2026-08-25 15:49:44 +08:00
• 23310bc3de [sleep functionality] code refactor about sleep/wake_up (#50431) Ning Xie 2026-08-25 15:45:20 +08:00
• 5e379a361e [Agents] Add kernel microbenchmark skill (#53688) Thien TranandCodex 2026-08-25 15:25:27 +08:00
• 06ecec7a84 [Bugfix][MM] Fix JinaVL processing cache order (#53553) Jiatai WangandOpenAI Codex 2026-08-25 00:03:05 -07:00
• 735daf8ef6 [Frontend] Move cli_args.py and dp_supervisor.py out openai folder (#53659) wang.yuqi 2026-08-25 15:02:32 +08:00
• c74dd2d208 [Bugfix][MRV2] Dispatch uniform decode to a padded FULL cudagraph (#53407) 3 people 2026-08-25 07:59:38 +01:00
• 4af586e185 Vllm recipes tool improve (#53325) Louie Tsai 2026-08-24 22:53:47 -07:00
• 48d7132962 [Model] Migrate FlexOlmo, Olmo3 and Hunyuan V1/VL to the Transformers modeling backend (#53615) 4 people 2026-08-24 22:39:05 -07:00
• 4f3e255dca [Bugfix][Model] deepseek-vl2: restore original DeepseekV2Config defaults for omitted language_config fields (#51302) Seunghyuk Park (shepark) 2026-08-24 22:03:40 -07:00
• a9a17e7095 Dflash2 load fix (#53435) Karan Shelar 2026-08-25 10:02:43 +05:30
• 1f9444a34f [Rust Frontend][gRPC] Add LoRA lifecycle control (#52840) 3 people 2026-08-25 00:13:19 -04:00
• af119619c4 Exclude the cpu backend from vLLM's active-Triton-driver count (#53530) (#53530) 3 people 2026-08-24 21:13:13 -07:00
• 8fe9317f2e [Model][MoE] DeepSeek-V4: add opt-in FlashInfer moe_ep expert backend (#49636) 3 people 2026-08-24 20:35:01 -07:00
• d3e2888c75 [Bugfix][Frontend] Redact hf_token in the non-default args log (#53625) Moe Huzaifa 2026-08-24 22:06:46 -05:00
• b2dd9ce73d [Docs] Fix docstring continuation indentation in routed_experts.py (#53604) Harry MellorandClaude Opus 5 2026-08-24 19:27:16 -07:00
• d4c4ceb31e [Bugfix][NIXL] Fix Mamba prefill truncation ordering (#53523) Ziming HuangandRoger Wang 2026-08-25 10:13:27 +08:00
• cbe3966f9c [XPU] Fix sparse-MLA metadata sync (#52066) Libin TangandKunshang Ji 2026-08-24 19:05:35 -07:00
• 7b7e5cbfcd [Bugfix][LoRA] Restore tower and connector LoRA support for LFM2-VL (#53519) Zupeng Wang 2026-08-25 09:53:04 +08:00
• 34c3824a11 [CI] Increase Entrypoints Unit timeout after launcher suite growth (#53618) Kevin H. LuuandCodex 2026-08-24 18:48:57 -07:00
• 07ef21bc69 [Kernel][Perf] Enable fused QK-norm + partial MRoPE + gate for Qwen3.6 (#52676) kiroxuandOpenAI Codex 2026-08-25 09:28:49 +08:00
• 9bf0d4717e [Bugfix] Keep grid dims for XD-RoPE models on a prefix-cache hit (#53456) drakoshaandClaude 2026-08-25 04:14:32 +03:00
• 09fddeb4ce [Bugfix] BailingMoeV3 KDA: skip absent metadata during CUDA graph profiling (#53593) Nick Hill 2026-08-24 18:02:50 -07:00
• e3dde1ee9b [CI] Report torch-nightly results to PyTorch CRCR (#51830) Andrey Talman 2026-08-24 21:00:13 -04:00
• 7ca336929c [Model] Remove ten deprecated model architectures (#53608) 3 people 2026-08-24 17:06:00 -07:00
• a0f1b9ad05 [CI][The Rock] Increase flex attention abs tol (#53646) rasmith 2026-08-24 18:57:03 -05:00