Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 48f17b77bb [Bugfix] Reject mismatched float16/bfloat16 KV cache dtype (#60885) 3 people 2026-10-15 13:36:59 +05:30
• fc8e08a2a6 [Bugfix][KV Offload] Do not let mark_miss() flip a pending or in-flight async lookup probe (#59096) 3 people 2026-10-15 03:57:46 -04:00
• 8fa5ae844a [Frontend] Let an unset --tool-strict-level fall back to the tool parser's default (#61328) Bugen ZhaoandClaude Opus 5.5 2026-10-15 15:48:38 +08:00
• 91ed2859c7 [CPU][Zen] Update zentorch to 2.14.0.1 & enable float16 on Venice and later (#61411) 3 people 2026-10-15 12:53:50 +05:30
• 85c94c1706 [Tests][XPU] Skip modelopt_fp4 online MoE re-quantization test on XPU (#61723) 3 people 2026-10-15 15:23:45 +08:00
• 8f111b22ee [CI] Run 4xB200 steps nightly only (#61541) Kevin H. LuuandClaude Opus 5.5 2026-10-15 00:08:48 -07:00
• 32feecd91f [CI] Pin DeepSeek-V4.1-Flash-NVFP4 eval to its FP8-Engram revision (#61763) Kevin H. LuuandClaude Opus 5.5 2026-10-15 00:05:18 -07:00
• ab593678b5 [CI][XPU] Ignore test_reload.py on Intel CI (#61740) Chaojun ZhangandKevin H. Luu 2026-10-15 15:03:34 +08:00
• a9fe30e9f6 [Bugfix][Multimodal] Recover PyNvVideoCodec decoder slots after failures (#59597) 4 people 2026-10-14 23:17:19 -07:00
• c5f8adc0b1 [KV Offload] Expose OffloadingManager.lock()/unlock() and service the tiering control plane off the scheduler thread (#58168) liranschourandKevin H. Luu 2026-10-15 08:50:56 +03:00
• c54dca96bd [Bugfix][Benchmark] Fix probe latency and duration in bench serve (#60511) HarshilandCyrus Leung 2026-10-14 22:50:30 -07:00
• 5341b5ba7d [KV Offload] Pin only each worker's own offload slots (#59544) 5 people 2026-10-15 08:49:42 +03:00
• 6605802904 [Perf][KV Connector][Mooncake] Send the KV layout once per peer (#59825) 3 people 2026-10-15 13:36:31 +08:00
• a400804ce4 [CI] Relax MTP sharded sampling cross-boot noise bound (#57026) 3 people 2026-10-14 22:31:45 -07:00
• 3f1bcfd82b [Bugfix][KV Offload] Bound unanswered P2P lookup waits (#56836) 7 people 2026-10-15 13:23:34 +08:00
• 81e1c3f183 [XPU][MLA] Integrate XPU MLA support into the existing FLASH_ATTN_MLA backend (#58297) 3 people 2026-10-15 12:47:37 +08:00
• 96b5b0ee4b [Model] Enable torch.compile and CUDA graphs for EmbeddingGemma2 (#61572) Vivaan SharmaandKevin H. Luu 2026-10-15 10:02:39 +05:30
• ff5a91a456 [Feature][Model Loader] Apply model-owned MTP shard selection to RunAI ModelStreamer (#60759) Haiyu ZuoandKevin H. Luu 2026-10-15 12:23:11 +08:00
• 833665cded [ROCm][DI] Enable MoRI FP4 dispatch for mxfp4 w4a4 moe (#59596) Chaitanya Sri Krishna LollaandCursor 2026-10-15 09:50:14 +05:30
• 10d2c6083d [ROCm][DSv4.1] Enable DSpark adaptive verification (#58151) Tuukka SarviandKevin H. Luu 2026-10-15 07:05:43 +03:00
• f46021a7fa [ROCm][Perf] Enable DPA+ETP prefill compute/comm dual-batch-overlap for DeepSeek-V4 (#57773) Shanshan Shen 2026-10-15 11:16:48 +08:00
• 5d5c3d7c43 [Bugfix] Preserve padding token types for cross encoders (#58143) FeanorandTaneem Ibrahim 2026-10-15 11:09:24 +08:00
• 429a3da665 [Bugfix][ROCm][Disagg] Allow MoRIIO HMA per-layer block_len (#58585) 4 people 2026-10-15 07:05:31 +05:30
• de2b869ee9 [Bugfix][Kernel] Register int8 scaled_mm as a custom op (#57331) KaiandKunshang Ji 2026-10-15 12:27:49 +11:00
• 2abd479aee [ROCm][CI] Add AMD job for DeepSeek-V4.1 Mono Decode Numerics (#61720) Aarushi Jain 2026-10-14 20:23:25 -05:00
• e57bb4ace7 [Core] Remove the legacy Ray compiled-graph executor (#61712) Woosuk KwonandClaude Opus 5.5 2026-10-14 18:13:27 -07:00
• bfdbfa1632 [Bugfix][PCP] Align MoE routing padding with gathered tokens (#60946) 4 people 2026-10-15 09:04:00 +08:00
• 8530aed2b7 [Spec Decode] Remove Medusa and MLP speculator support (#61701) Woosuk KwonandClaude Opus 5.5 2026-10-14 17:29:23 -07:00
• 273a04964e [ROCm] Bump AITER to v0.1.25 and relax the DSv4.1 mono MoE tolerance (#61697) josh-schell-amdandClaude Opus 5.5 2026-10-14 18:50:24 -05:00
• 7840bede98 [ROCm][CI] Add AMD jobs for test groups that only exist in tes… (#59678) Aarushi JainandAndreas Karatzas 2026-10-14 18:00:44 -05:00
• 4b61f300c0 [CI] Fix /ci run reporting a failed merge on fork PRs (#61684) Kevin H. Luu 2026-10-14 15:26:15 -07:00
• a4d553a46a Revert "[Build] Update deep_gemm git tag for locality_domain update" (#61703) Micah Williamson 2026-10-14 16:57:44 -05:00
• 9a648b4649 [ROCm][Kimi-K3] Don't read W_UK_T before weight loading in the MLA wrapper (#61654) Rohan PotdarandClaude Opus 5.5 2026-10-14 16:44:55 -05:00
• 4c5ea86f53 [Bugfix][Parser] Stream Gemma4 string tool arguments incrementally (#61494) Ben BrowningandFlora Feng 2026-10-14 17:25:41 -04:00
• 984c242ede [Quantization] Support compressed-tensors MXFP4 MoE with FP8 activations via Mxfp4MoEMethod (#59815) 3 people 2026-10-14 16:33:35 -04:00
• fe98d90a9d [Dependency] Update Humming and remove NVFP4 scale workaround (#61556) 4 people 2026-10-15 04:32:36 +08:00
• 2e8a5930f6 [Benchmark] Remove the temporary bench forward dummy loader (#61676) Woosuk KwonandClaude Opus 5.5 2026-10-14 13:30:18 -07:00
• 7c3f7e5500 [Perf][DSv4.1] Length-aware candidate block selection in the decode indexer (#61039) Juntian Liu 2026-10-14 13:15:58 -07:00
• 2622ee21aa [ROCm][BugFix] Don't use AITER rms_norm as an IR OP on gfx90a when AITER always on. (#59290) rasmithandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-14 14:48:37 -05:00
• e73fa00981 [Bugfix][Frontend] Apply the JSON schema in Cohere Chat v2 response_format (#60586) alpandmottopanikeiku 2026-10-14 12:42:20 -07:00
• ded7ae35e5 [Build] Update deep_gemm git tag for locality_domain update (#61491) Chris LeonardandShengqi Chen 2026-10-14 15:41:16 -04:00
• 388a610dfb [Perf][DFlash] Capture context combine and anchor in the draft CUDA graph (#59511) Juntian LiuandClaude Opus 5.5 2026-10-14 12:40:52 -07:00
• 4975fda882 [Dependency] Upgrade FlashInfer to 0.7.1 (#60948) Wei ZhaoandJee Jee Li 2026-10-14 15:31:08 -04:00
• 9b5c6707a3 [Bugfix][MRV2] Capture DCP FULL cudagraphs with KV context (#61496) Ayushman Singh 2026-10-14 15:20:48 -04:00
• 222e00fc44 [CI/Build] Fix test_deepseek_v41_mxfp8_scale_loading after #57797 (#61678) Stefano CastagnettaandMisha Goin 2026-10-14 12:17:59 -07:00
• e96c822660 [Bugfix][ROCm][MiniMax-M3] Reject unbuilt index-head counts in the AITER indexer (#60990) Andreas KaratzasandClaude 2026-10-14 14:13:48 -05:00
• 21e207efa1 [Bugfix][MRV2][Spec Decode] Honor EAGLE3 single-state draft configuration (#61563) Sandeep AgrawalandKevin H. Luu 2026-10-14 12:09:06 -07:00
• 06265beb25 [ROCm] Enable fp8 query for Triton attention (#60296) amd-xavierwangandJartX 2026-10-14 15:03:20 -04:00
• 857ddac4e5 [Perf][MRV2] Skip empty draft-token resets on request admission (#61640) 3 people 2026-10-14 21:01:01 +02:00
• c1407a7e86 [Spec Decode][MRV2] Warm up top-k/top-p and greedy rejection kernels for speculative decoding (#59260) 3 people 2026-10-14 11:59:21 -07:00
• ae94e9c9e8 [Bugfix] Don't crash DeepGEMM MegaMoE warmup when get_bf16_mega_gate_config is missing (#61657) simon-veitner-redhatandClaude 2026-10-14 20:58:35 +02:00
• 3f729dafcc [Perf][ROCm][GLM-5.3-Flash] AITER prefill indexer topk (#60753) Simon Danielsson 2026-10-14 13:48:56 -05:00
• e534024e49 [ROCm][Bugfix][MLA] Fix FP8 MLA prefill unpad for interleave-padded heads (#60907) lucas-santos-amdandDouglas Lehr 2026-10-14 15:34:28 -03:00
• d1d8e6c096 [Perf][Nemotron-Parse] Do image rescale/normalize on the device (#60500) 3 people 2026-10-14 11:33:22 -07:00
• 0b6d07cbe2 [CI/Build] Use FileStore for DeepEP test rendezvous (#56595) Kevin H. LuuandOpenAI Codex 2026-10-14 11:29:12 -07:00
• d3f6b2a7b3 [Frontend] Clean Client Facing 500 Report (#58043) Robert ShawandClaude 2026-10-14 14:27:58 -04:00
• 47232f3c7f [Bugfix][CI] Fix renderers tests broken by tool name removal from warn logs (#61652) Ganesh RandCyrus Leung 2026-10-14 23:44:58 +05:30
• 52a8cc5316 [ROCm][CI] Add dtype to the KDA metadata test's fake model config (#61650) Rohan PotdarandClaude Opus 5.5 2026-10-14 13:04:14 -05:00
• 9d3dfa2886 [Benchmark] Add vllm bench forward: exact-shape forward timing (#61339) Woosuk KwonandClaude Opus 5.5 2026-10-14 10:52:26 -07:00
• 2c74fd1873 [Model Runner V2] Enable batch-sharded sampling for GPT-OSS (#61257) TaimysandOpenAI Codex 2026-10-15 01:11:11 +08:00
• 89cab55086 [Model] Support Decision 3.0 decision models (vllm-sr/d3) (#61565) Xunzhuo 2026-10-15 01:10:55 +08:00
• 3895a8babe [Bugfix] Fix TP loading of small block-FP8 MLA gates in BailingMoeV3 (#61642) zexplorerhjandCodex 2026-10-15 01:09:23 +08:00
• e2e3e2b834 [CI] Use an ungated model in the DGX Spark XQA test (#61416) SeongJun LeeandWei Zhao 2026-10-15 02:04:07 +09:00
• 3e851a7803 [Bugfix][Attention] Make the query contiguous for TRTLLM ragged MLA prefill (#61634) Harry MellorandClaude Opus 5.5 2026-10-14 17:38:52 +01:00
• 355553ac22 [ROCm][Perf] Drop two dead per-step ops from the AITER sparse MLA metadata build (#58354) Zheng GongandCursor 2026-10-15 00:15:51 +08:00
• 13dbf96093 [Model Loader] Apply EP weight filtering in InstantTensor (#61167) 3 people 2026-10-14 19:10:10 +03:00
• eaf25f675a [NIXL] Prepare KV transfer descriptors as NIXL 1.5 strided lists (#61032) Nick HillandIlia Yastrebov 2026-10-14 09:07:24 -07:00
• 9e2b579aab [Model] Support EAGLE3/DFlash aux hidden states for GraniteMoe and GraniteMoeShared (#61110) Nan ZhangandRobert Shaw 2026-10-14 12:01:14 -04:00
• e2aabc8b15 [Bugfix][KV Offload][P2P] Keep store pins and AbortAck until NIXL transfers stop (#59790) 4 people 2026-10-14 23:31:31 +08:00
• 5d88c034f6 [Security] Reject local verifier references in Speculators metadata (#61587) Juan Pérez de AlgabaandOpenAI Codex 2026-10-14 17:30:20 +02:00
• bf6c80c7e8 [Misc] Remove tool name from warn logs (#59294) 3 people 2026-10-14 17:03:11 +02:00
• efad400092 [HiSparse] Prefault the shared host pool in parallel (#60869) Itay EtelisandCodex 2026-10-14 17:28:26 +03:00
• 02653fa22c [Bugfix][KV Offload] Fix inaccurate proportional store dropping (#59140) 4 people 2026-10-14 07:05:21 -07:00
• b4b7eca0ef [Bugfix][Quantization] Clamp FP8 scales before UE8M0 rounding (#58032) Ashish RaoandVadim Gimpelson 2026-10-14 06:56:42 -07:00
• 7115b70c01 [Bugfix][KV Offload][P2P] Drop a timed-out store job's blocks from the round's parked supply (#59329) 3 people 2026-10-14 09:22:28 -04:00
• 2d09fdcdc2 [Quantization] Auto-disable DeepGEMM for Kolibri 1 on SM12x (#61274) Luca MotzandClaude Opus 5.5 2026-10-14 14:59:50 +02:00
• 3723a1ea96 [Bugfix][Transformers backend] Fix CUDA graph capture for dynamic-rope models (#58064) 4 people 2026-10-14 20:56:54 +08:00
• 2631afd3fe [Bugfix][Platform] Fall back to torch when device-level NVML memory is unsupported (GB10) (#57378) hclandHarry Mellor 2026-10-14 20:45:29 +08:00
• a68dbbad05 [Bugfix][KV Offload] Retain offload mappings after unsafe KVCR startup failure (#57500) Kapil AryaandOr Ozeri 2026-10-14 05:43:04 -07:00
• 775fefa804 [Metrics][KV Offload] Add vllm:kv_offload_config_info - generic info metrics for offloading the offloading connector (#56867) 3 people 2026-10-14 15:34:37 +03:00
• 3a5a6b11a3 [Perf][Model] Batch PaddleOCR-VL vision encoder and projector (#61458) ssingh231219 2026-10-14 13:59:44 +02:00
• d39c2fb05a [Bugfix][Gemma4] Quantize audio towers independently (#59935) 3 people 2026-10-14 16:57:44 +05:30
• 23fbc2ec2f [Model] Support precomputed image embeddings for LightOnOCR (#61585) Lazurite 2026-10-14 18:54:09 +08:00
• 81503550f3 [ROCm] Disable AITER FP8/FP4 BMM for MLA by default (#53123) xuebwang 2026-10-14 18:46:02 +08:00
• c67c69e657 [Bugfix][KV Offload] Track allocated block IDs when num_external_tokens is 0 (#47046) linshi wangandOr Ozeri 2026-10-14 03:34:10 -07:00
• 6ecb39eaa7 [Docs] Update structured decisions docs. (#61588) wang.yuqi 2026-10-14 18:28:22 +08:00
• 19aba3e4d5 [Bugfix][RL] Load attention sinks through weight_loader for layerwise reload (#57797) Ronaldandaoshen02 2026-10-14 18:17:58 +08:00
• f9b65c30c7 [ROCm][Perf] Make AITER FlyDSL GDN prefill default on gfx942/gfx950 (#60645) 3 people 2026-10-14 13:11:57 +03:00
• 3f466c852c [Bugfix][Parser] Recover genuine tool calls after unclosed Qwen3 <function= opener in prose (#59928) 3 people 2026-10-14 10:01:25 +00:00
• 8a8e2baa02 [Bugfix][Rust Frontend] Inject VLLM_USE_RUST_FRONTEND=1 into vllm-rs managed engine envs (#61586) 3 people 2026-10-14 17:09:08 +08:00
• be4ce2e505 [Bugfix] Share PCP-aware local GPU indexing for NUMA binding (#61237) Sherif Waly 2026-10-14 09:54:03 +01:00
• 937d6c19df [Perf][GDN] Skip FLA chunk metadata when FlashInfer GDN prefill is active (#61569) Thien TranandKimi Code 2026-10-14 16:46:54 +08:00
• d36158f919 [CI] Merge the target branch into stale PRs on /ci run (#61592) Kevin H. LuuandMisha Goin 2026-10-14 01:43:03 -07:00
• bb1438a93f [CI] Fix Unlimited-OCR loading after upstream model_type rename (#61590) Artem Perevedentsev 2026-10-14 11:24:24 +03:00
• 467b024f21 [Rust Frontend] Report vllm-proto schema version in ServerInfo (#61533) 3 people 2026-10-14 01:21:48 -07:00
• aa0c433a3e [Frontend] Accept --reasoning-parser none as the unset default (#61525) 3 people 2026-10-14 00:50:35 -07:00
• c9bfa574ce [Online quantization] Extend re-quantization API to MOE (showcase NVFP4 modelopt/CT/quark to any online MOE method) (#60276) 3 people 2026-10-14 09:49:37 +02:00
• e1726d9142 [Frontend] Add cache salt and tracing to openai decisions (#61550) ChaunceyandCyrus Leung 2026-10-14 15:48:44 +08:00
• 4dc5add6ea [XPU] [Linear] [MXFP8] enable torch as mxfp8 linear backend (#51933) zofia 2026-10-14 15:48:38 +08:00
• ec8a372566 [ROCm][Bugfix] Pin AITER back to v0.1.24.post1 for DSv4.1 mono decode numerics (#61559) Ronald 2026-10-14 15:19:38 +08:00