Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 92b40f5a14 [CPU][Perf] refactor paged attention for Arm CPUs (#56045) 3 people 2026-09-20 05:25:13 +01:00
• 70fc7bd695 [Feature][EPD] Support dynamic EPD (EC-connector) proxy (#54176) 3 people 2026-09-20 12:24:41 +08:00
• 1596fa5f88 [ROCm][DSv4.1][Perf] Reuse the decode topk ragged metadata across layers (#57434) Fangzhou AiandShanshan Shen 2026-09-19 23:46:33 -04:00
• 03f8301387 Add one-step recipe serving (#57306) Louie Tsai 2026-09-19 20:24:09 -07:00
• 1c3ef2ad4c [XPU][UT]Fix test_mnnvl_alltoall device and distributed backend (#57591) YiSheng5andKunshang Ji 2026-09-20 10:53:00 +08:00
• f4fc5aae7f [XPU][UT]Using the device_control_env_var to restrict visible device on different platform (#57581) YiSheng5andKunshang Ji 2026-09-20 10:52:50 +08:00
• 708c2495c0 [Test][Core] Compute the expected hybrid prefix-cache hit instead of hard-coding it (#57467) Fanli Lin 2026-09-20 10:16:12 +08:00
• 5780c0125a Merge remote-tracking branch 'origin/main' into sync/lesj/qwen4exp-bugfixes lesj0610 2026-09-20 09:27:57 +09:00
• a7fda4c88b [Bugfix] Fix SM100 fp8_ds_mla cache scales (#49435) 6 people 2026-09-20 03:15:07 +03:00
• 674b6d95d6 [CI][Bugfix] Fix MoE reprocess test mock after #57405's kernel refactor (#57641) 3 people 2026-09-19 14:42:21 -07:00
• 36fa72d2d0 [GLM5.3 Perf] Size the GLM-5 sparse indexer decode workspace, 3072 MiB GPU memory saved (#57701) Wentao Ye 2026-09-19 16:57:01 -04:00
• 133b71e0be [Bugfix] Take padded path for ragged decode batches in sparse_attn_indexer (#52500) 3 people 2026-09-19 13:14:29 -07:00
• 5302d1fe40 [Model Runner V2] Support custom logits processors (#56497) 3 people 2026-09-19 10:46:49 -07:00
• 751f6807d9 [Refactor] Use dynamic MM cache in processor (#57674) Cyrus Leung 2026-09-19 23:38:00 +08:00
• 41c4a3ed4e [Kernel][Perf] Add sm_120 (RTX PRO 6000 / RTX 50) tuned configs for batch-invariant persistent matmul (#57456) Lio Einaudi 2026-09-19 21:30:55 +08:00
• fbe8a157fb [ROCm][Perf] Enable the AITER GDN decode fast path for flat qkvz layouts (#53623) 3 people 2026-09-19 16:05:04 +03:00
• 211e252d0b [Bugfix][Model] Fix MiMo-V2.5 fused fp8 qkv_proj sharding (pre-shard count is num_key_value_heads; MTP path too) (#57508) vllmellm 2026-09-19 21:05:00 +08:00
• 4cc15f2121 [Quantization] Let ModelOpt MXFP8 layers load pre-processed weights Purpose (#57316) siyuandIsotr0py 2026-09-19 17:16:39 +08:00
• a5a30471ff [Tests] Cover get_unhashed_block_ids_all_groups (#55928) Aditya Ganesh Kumar 2026-09-19 17:13:54 +08:00
• c3b4844634 [Docs] Explain pointer-alignment JIT specialization in Triton skill (#57669) Woosuk Kwon 2026-09-19 00:53:42 -07:00
• 38537331d3 [Bugfix][DSA] Avoid runtime JIT for offset candidate end buffers (#57667) Woosuk Kwon 2026-09-19 00:27:11 -07:00
• 59e6682c21 [CI][AMD] Bump torchao to v18 for Python 3.14 (#53009) Ryan RockandAndreas Karatzas 2026-09-19 02:13:16 -05:00
• 6af44e2887 [Bugfix][CPU] Fix DeepSeek V4.1 import without Triton (#57654) Jiang WuandJiang Wu 2026-09-18 22:46:13 -07:00
• a8d1aa9c99 [Bugfix][Multimodal] Frame the multi-modal hash digest input (#54283) 3 people 2026-09-19 00:06:48 -04:00
• 01c7bf8813 [Bugfix][Model] Fix Aria expert weight names and layout (#57487) Jiang WuandJiang Wu 2026-09-18 21:06:30 -07:00
• 9d75dd4544 [CI] Deflake pooling shards with GPU teardown fixtures between tests (#57371) Kevin H. LuuandKimi 2026-09-18 20:30:02 -07:00
• 2fd6268e6d [CI] Deflake MTEB score tests with per-model mteb_tol and reruns (#57364) Kevin H. Luu 2026-09-18 20:29:00 -07:00
• 038cb14bd7 [CI][Bugfix] Correct the Laguna DFlash acceptance-length reference (#57647) 3 people 2026-09-18 19:44:48 -06:00
• 468663a5e5 [ROCm] Bump AITER to v0.1.22.post1 (#56885) Micah Williamson 2026-09-18 20:36:45 -05:00
• 90531daf6d [Core][Frontend] Bind KV-event publishers at port 0 and expose the bound endpoints (#55844) 5 people 2026-09-19 09:11:05 +08:00
• 729ebac498 [CI] Reclaim GPU memory between model initialization tests (#57362) Kevin H. Luu 2026-09-18 17:04:14 -07:00
• a1bf8ac12d [Bugfix][MRV2] Match fast-prefill padding to active LoRA batches (#56456) waizuichougou 2026-09-19 06:13:23 +08:00
• 1dc2d854c1 [Feat][Model] Support encoder-side SWA-bounded replay for DeepSeek-V4.1-Flash (#56227) 3 people 2026-09-18 15:01:24 -07:00
• 6aab78f477 [Bugfix][KVConnector] Make ExampleHiddenStatesConnector abort-safe (#56841) ys2025-AI 2026-09-19 05:53:02 +08:00
• 62af3df734 [Bugfix][MLA] Reserve sparse prefill buffers before KV cache sizing (#57575) 3 people 2026-09-18 17:22:18 -04:00
• 0390299309 [ROCm][CI] Shard MI300 Entrypoints Integration (Pooling) (#57583) Aarushi Jain 2026-09-18 16:16:29 -05:00
• 2c3fb4c545 [CI][ROCm] Deprecate DinD for MI250 test groups (#56162) Andreas KaratzasandOpenAI Codex 2026-09-18 16:13:35 -05:00
• 49e0427ca5 [Bugfix][NIXL] Avoid receive reports for notification-only requests (#57570) Nicolò LucchesiandOpenAI Codex 2026-09-18 22:33:02 +02:00
• 63d9ad0a3a [Model] DiffusionGemma: honor logprob_token_ids on the converging step (#57417) Matt MastracciandClaude Fable 5.1 2026-09-18 14:11:27 -06:00
• 50d812b66a [Perf][DSV4.1] Optimize MegaMoE staging and NVFP4 cache gathers (#57604) Woosuk KwonandCodex 2026-09-18 12:57:19 -07:00
• 23e26e0588 [KV Offloading] Back-pressure detection and remediation (#50045) bnellnmandItay Etelis 2026-09-18 14:47:48 -04:00
• 71fc70d3ae [ROCm][DSv4] Fix sparse-indexer logits collapse on gfx950/gfx942 (#50455) frida-andersson 2026-09-18 18:57:44 +02:00
• 8a5cf54387 [Multimodal] Type dummy options per modality (#57576) Harry MellorandClaude Opus 5 2026-09-18 17:29:11 +01:00
• 56cac80b14 [Deprecation] Remove assistant_token_mask support (#57520) Cyrus Leung 2026-09-19 00:17:51 +08:00
• 164dca8d1e [Nixl] Separate transport-failure metrics from KV expiry (#55854) Nicolò Lucchesi 2026-09-18 18:08:59 +02:00
• a7156060c1 [Core] Make parallel sampling (n>1) reqs admission atomic (#53936) Nicolò Lucchesi 2026-09-18 18:08:44 +02:00
• 4991f97669 [XPU] sleep mode: fix KV cache release test (#57485) Yan MaandKunshang Ji 2026-09-18 23:49:17 +08:00
• 017dced6a6 [DeepSeek V4] Fix fused MoE expert distribution (#57465) Itay AlroyandTyler Michael Smith 2026-09-18 18:17:07 +03:00
• dbf4b89cf2 [Misc] Remove unnecessary Transformers version guards (#57556) Cyrus Leung 2026-09-18 23:01:35 +08:00
• 1cdf1689e5 [Quantization][XPU] Enable int8_w8a8 MoE on the Triton backend for XPU (#53162) 3 people 2026-09-18 16:38:15 +02:00
• 70164bdadc [Frontend] Add stream reasoning and tool calls from the derender endpoint (#50550) Martin Hickey 2026-09-18 15:36:27 +01:00
• 911548247a [Distributed][MoonEP] BF16 integration of MoonEP balanced EP backend (#52101) Kevin Liandkevinli 2026-09-18 07:34:01 -07:00
• 2bdae2a5a8 [fix] Mistral-Large-3 accuracy regression on main (#57563) Julien DebacheandHarry Mellor 2026-09-18 16:19:16 +02:00
• 76d517fd12 [Quantization] Support kimi-k3 routed expert quant (#57430) Kyle SayersandMisha Goin 2026-09-18 10:03:59 -04:00
• 44dd18fe0b [Model][Gemma4] Load Weights with AutoWeightsLoader (#55911) 3 people 2026-09-18 15:07:30 +03:00
• 2b8a5495a8 [Mypy] Fix mypy typing for N/O models (#54142) Taneem Ibrahim 2026-09-18 06:50:01 -05:00
• e562003adb [Tokenizer] Drop dead Mistral tokenizer shims for transformers#41962 (#56325) 3 people 2026-09-18 19:46:41 +08:00
• c58532c86b [Frontend] Upgrade XGrammar to 0.2.7 and Rust structural tags to 0.3.0 (#57272) 3 people 2026-09-18 19:44:05 +08:00
• cc09352e5c [Bugfix][Multimodal] Preserve DeepSeek V4 image block spacing (#56882) 4 people 2026-09-18 18:44:39 +08:00
• 04a3a00eb7 [CI] Raise GSM8K startup max wait to 2400s for flaky 30B+ MoE eval configs (#57361) Kevin H. Luu 2026-09-18 03:40:54 -07:00
• f2b6152ec2 [CI] Add flaky rerun markers / timeout skips to sibling tests lacking them (#57359) Kevin H. LuuandKimi 2026-09-18 03:40:32 -07:00
• 4c6c1a40b3 [ROCm][Bugfix] Fix intermittent ROCR host segfault (#57328) Matt 2026-09-18 03:13:45 -07:00
• 5397f967b9 [CI] Add residual timeout headroom after JIT rollback (#56649) Kevin H. LuuandOpenAI Codex 2026-09-18 03:11:07 -07:00
• 5bb596201a [CI] Raise Elastic EP Scaling step timeout 30m -> 40m (#57287) vllm-agentandkimi-no-na-wa 2026-09-18 03:10:38 -07:00
• b346479065 [CPU] Add CPU FP8 W8A8 linear/MoE support (#49942) Yuwen ZhouandLi, Jiang <jiang1.li@intel.com> 2026-09-18 17:59:42 +08:00
• 6df2b1a8c6 [Quantization] Support native Quark W4A16 INT4/UINT4 exports in vLLM (#48606) meng-amd 2026-09-18 17:52:46 +08:00
• 67a8a3f926 [Bugfix][CPU] Fix macOS multimodal SHM cache initialization (#57142) shaohuaxiandCodex 2026-09-18 02:05:13 -07:00
• 32636580a6 [Bugfix][MLA] TritonMLA: fix illegal memory access on causal multi-token decode (#51065) olka-amdandClaude Opus 5 2026-09-18 11:49:53 +03:00
• 8da75d6190 C3x SM100 FP8 blockwise - Pad activation scales to multiple of 4 (#57377) Thien TranandCodex 2026-09-18 16:39:27 +08:00
• 95d723700b [Rust Frontend] Return sampling masks over gRPC (#56777) Biswa PandaandBugen Zhao 2026-09-18 01:26:17 -07:00
• d5f0a6e829 [Bugfix][DBO] Fix DeepEP low-latency profiling crash with DP+EP+DBO (#57502) Micah WilliamsonandAndreas Karatzas 2026-09-18 01:33:41 -05:00
• 42a33039eb [Perf] Fuse CohereASR relative attention score accumulation (#55190) SunandCodex 2026-09-18 14:18:41 +08:00
• 39e33db7f3 [Frontend] Fix the parsing of missing string= in DeepSeek V4 (#56271) 3 people 2026-09-18 13:11:24 +08:00
• dee37d8911 [Frontend] Add --tool-strict-level for server-side control for structural tag activation (#56268) 3 people 2026-09-18 12:32:39 +08:00
• d12c276853 [Bugfix][ROCm] Alias SparseAttnIndexerKpool.forward_cuda to forward_native (GLM-5.3-Flash boot crash) (#57425) 3 people 2026-09-17 20:36:55 -07:00
• bf13ecc2f7 [Perf][MRV2] Share token-to-request mappings across KV cache groups (#57102) Juntian LiuandWoosuk Kwon 2026-09-17 20:14:36 -07:00
• 2909ad8fa4 [Bugfix] Attach request-level tools to existing system message in DeepSeek V4 Python renderer (#51856) JasonandChauncey 2026-09-17 23:07:25 -04:00
• 70df48dc3d [Bugfix][KV Cache][GLM-5.3-Flash] Disable slot mapping kernel for the kpool tail buffer (#57317) Simon Danielsson 2026-09-18 04:42:54 +02:00
• 7b94293627 [Bugfix][ROCm][KV Offload] Use private pinned tensors for CPU KV offload (#57160) Ren Yuzhou 2026-09-18 10:35:23 +08:00
• 2bbdfcfcfe [Perf][GLM5.3-Flash] Use cooperative top-k for small GLM decode batches (#57327) Chaunceyandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-18 10:31:35 +08:00
• 64563d0ec4 [Bugfix][DSv4.1] Preserve NaN-scored candidate block indices (#57454) Woosuk KwonandCodex 2026-09-17 19:00:10 -07:00
• 2c88fb131c [Bugfix] DiffusionGemma: hand out stashed logprobs only on the committing step (#57414) Matt Mastracci 2026-09-17 18:24:58 -06:00
• 213c379fca [ROCm][Bugfix] Gate AITER MXFP8 MoE on the aiter enable flag (#57426) Rohan PotdarandClaude 2026-09-17 19:06:05 -05:00
• 48f663c37f [XPU] Fix incorrect context-key normalization for Qwen DFlash-based models (#56431) Yan Ma 2026-09-18 07:52:01 +08:00
• 092bdd6d57 [Frontend][Core] Add release_kv_cache_memory() API (#44890) 4 people 2026-09-18 07:12:12 +08:00
• a524f80b1f [AMD][Bugfix] Make the nested-RoPE patch reach automatic validation (#57289) 3 people 2026-09-17 17:09:06 -06:00
• ccde9c61e9 [Bugfix] Restore KV cache metadata GET method for external event consumer (#56925) Wei Zhao 2026-09-17 19:01:35 -04:00
• e0050f287a [Bugfix][ROCm] Add record_logical_topk_ready to ROCMAiterMLASparseImpl (GLM-5.3-Flash boot crash) (#57252) 3 people 2026-09-17 15:34:56 -07:00
• 80447d2765 [Bugfix][DSv4.1] Fix FlashInfer DSpark non-causal attention (#57432) Woosuk KwonandCodex 2026-09-17 15:14:37 -07:00
• d3074acdda [AMD][CI][The Rock] Fix language models standard for The Rock on mi355 (#53837) rasmithandAndreas Karatzas 2026-09-17 17:13:13 -05:00
• e8a2a0a8bb [Bugfix][Frontend] Reject stop strings on --tokens-only servers instead of silently ignoring them (#57058) 4 people 2026-09-17 15:02:00 -07:00
• ac78445bc9 [MoE] Encapsulate TRT-LLM BF16 weight layout handling (#57405) Misha Goin 2026-09-17 17:47:30 -04:00
• db7a24c230 [Perf] Add fused DFlash2 grouped convolution (#55960) Misha Goin 2026-09-17 17:38:59 -04:00
• 668d6c3a77 [CI] Shard multimodal Processor 1->4 (#56316) 4 people 2026-09-17 14:11:40 -07:00
• 9612f77077 [Bugfix] Release stale FlashMLA workspace views after growth (#56902) Maroon AyoubandRobert Shaw 2026-09-17 23:23:15 +03:00
• acc2ed2a5f [ROCm][CI][The Rock 10] Fix (MI355) Quantized Models failure on The Rock 10 with Triton 3.8.x (#54849) rasmithandShanshan Shen 2026-09-17 15:20:35 -05:00
• 2b02c6c29b [Model] Add Cohere2MoE Eagle3 auxiliary hidden states (#49819) Doug Brown 2026-09-17 16:09:36 -04:00
• 9a5bd373cf [CI] Retire Weight Loading smoke tests (#57398) Misha Goin 2026-09-17 16:04:03 -04:00
• ac2f0ea82c [MoE][Bugfix] Skip SP padded rows in grouped MoE routing (#56079) 4 people 2026-09-17 21:36:19 +02:00
• 67e5b0acc9 [Bugfix][HiSparse][NIXL] Import full blocks without tail prefill on D (#57049) 3 people 2026-09-17 21:26:10 +02:00