Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 6f74337c47 [Bugfix][Spec Decode]Preserve user --speculative-config overrides for speculators-format models (#42376) Lee Yongjunandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2026-08-22 01:06:47 +09:00
• 574e6a00ef [Bugfix][Attention] Normalize FlashInfer prefill LSE before merging (#52796) yimdevandusberkeley 2026-08-21 23:31:14 +08:00
• 2740c817ff [Kernel] Add b12x FP4 MoE backend (#52018) Luke Alonsoandmgoin 2026-08-21 08:04:36 -07:00
• 5ee84d3c52 [CI][Bugfix] Use a prompt that survives offload-resume rounding in mamba offload test (#53146) Frederik GossenandClaude 2026-08-21 10:14:52 -04:00
• cda3868f5d [Bugfix] Deterministic MoE combine (reduce_scatterv) under VLLM_BATCH_INVARIANT (#45683) shijuzhaoandshijuzhao 2026-08-21 21:41:31 +08:00
• c8438a3d40 [Doc] Fix dead link in KV transfer README (#53230) Harish Gaikwad 2026-08-21 17:33:41 +05:30
• a60c66e3dc [Bugfix] Fix int32 index overflow in LoRA punica kernels at long context (#53034) Shuai ShaoandJee Jee Li 2026-08-21 03:37:06 -07:00
• 6feafb8b7d [XPU] follow cuda path for mrope on XPU (#53201) Yan Ma 2026-08-21 18:13:27 +08:00
• e85d1b69cf [Bugfix][LoRA] Use an explicit capability flag for tower connector LoRA (#53092) linitra24 2026-08-21 17:45:54 +08:00
• 36bad1b90c [XPU][CI]Add more cases in intel GPU CI and reorganize to align non-xpu part (#51630) xiangdongandKunshang Ji 2026-08-21 17:31:25 +08:00
• 0a21947d71 [XPU][CI]Add parallelism for long-running Intel GPU cases (#52257) xiangdong 2026-08-21 16:02:34 +08:00
• 9fd750f00d [XPU][INC] Add int4 w4a8 (dynamic int8 activation) backend for INC linear layers (#50501) Thanaji Rao ThakkalapelliandKunshang Ji 2026-08-21 01:38:59 -05:00
• f8e0602713 Support kimi k3 nvfp4 checkpoint (#53132) Wei ZhaoandOpenAI Codex 2026-08-21 01:59:44 -04:00
• b389ac2946 [Spec Decode] DFlash2: local convolution + candidate selector (#52816) Zihan Zhang 2026-08-20 22:27:22 -07:00
• df344bf36f [Rust][Benchmark] Load HF datasets from parquet shards via hf-hub, fixing truncated-cache sampling (#51570) Roy Wang 2026-08-21 13:07:43 +08:00
• ba07e4a48f [Bugfix] Fix batch-invariant fp32 matmul OOR on SM89 for N=1 (#52960) vhagor 2026-08-21 11:46:55 +08:00
• 2785c72a14 [Kimi-K3] Extend GEMM-RS to GEMM-AR (#53053) 3 people 2026-08-21 11:44:15 +08:00
• b00f475f09 [MM] Remove text components from ProcessorInputs (#53093) Cyrus Leung 2026-08-21 11:42:53 +08:00
• 2adf4b9e2a [Rust Frontend] Fix Kimi K3 reasoning_effort="none" handling (#53043) ReidandBugen Zhao 2026-08-21 10:47:07 +08:00
• 5df31ea52d [Spec Decode] Enable adaptive verification on DSv4 + sm90 (#52795) Lucas WilkinsonandOpenAI Codex 2026-08-20 22:44:55 -04:00
• df3b3422b4 [Rust Frontend] Add HY3 unified parser and local XGrammar structural-tag builder (#53054) Bugen Zhao 2026-08-21 12:13:54 +10:00
• c6e19b3be2 [Bugfix][Structured Output] Avoid spurious FSM errors after speculative reasoning end (#53046) Chauncey 2026-08-21 10:01:06 +08:00
• 83c5d59209 [Rust Frontend] Replace external protoc with pure Rust lib protox (#52892) 3 people 2026-08-21 11:43:51 +10:00
• 91a893de64 [Bugfix][Spec Decode] Scope DSpark backend inheritance to DeepSeek V4 (#52809) 3 people 2026-08-20 21:42:08 -04:00
• d29f7f5c92 [Bugfix] Load untied Gemma LM head weights (#53170) Kevin H. Luu 2026-08-20 17:47:08 -07:00
• f32b17b6d6 [Rust Frontend] Support --generation-config vllm (#53044) Bugen Zhao 2026-08-21 10:31:25 +10:00
• 0a5a55136f [CI][Docker] Pin remaining manylinux builder images (#53172) Kevin H. LuuandOpenAI Codex 2026-08-20 16:17:02 -07:00
• 7cfb97e337 [Frontend][Core][Spec Decode] Per-request acceptance stats in OpenAI API responses (#48915) 3 people 2026-08-20 15:42:49 -07:00
• 2f41c894e3 Fix seed loss when batch contains unseeded requests (#51866) MKQuantum 2026-08-20 18:38:53 -04:00
• 54ba80d961 [CI][Docker] Pin manylinux2_28-builder:cuda13.0 to the release/2.13 image (#52994) Andrey Talman 2026-08-20 18:19:40 -04:00
• bfb6c13499 [Bugfix][MoE] Tune FlashInfer experts to scheduler token limit (#52989) Misha GoinandOpenAI Codex 2026-08-20 16:26:24 -04:00
• 4f6885fffc [DSV4][Kernel] Fuse shared experts into MegaMoE (#53040) Canlin Guo 2026-08-21 03:07:02 +08:00
• 3b829cf176 [CI/Build][ROCm] Keep the CUDA-only kernel tests out of the ROCm run (#53113) stefankoncarevic 2026-08-20 20:40:33 +02:00
• d56bbf3995 [Bugfix] Support MistralCommonBackend tokenizers in structured output (#52720) Thanh Phan 2026-08-21 01:08:50 +07:00
• 7c8b68b9ce [Bugfix][MiniMax-M3] Keep FP8 query allocation stable across CUDA graph replay (#51203) kyleliang-nv 2026-08-20 10:45:28 -07:00
• 00f7f25828 [Misc] Don't allow language-model-only used with encoder CG together (#53127) Isotr0py 2026-08-21 01:26:47 +08:00
• ae25628995 upgrade tpu-inference to v0.27.0 (#53088) Dennis Yeh 2026-08-21 00:21:39 +08:00
• 1fe3a1571a Reduce AutoWeightsLoader kwargs (#53106) Harry Mellor 2026-08-20 16:40:11 +01:00
• 2dd17225c6 Fix Transformers modelling backend RMSNormFuser.fuse performance (#52766) Harry Mellor 2026-08-20 16:21:23 +01:00
• 01af92e175 [Feature][Model Runner V2] Support extract_hidden_states speculation (#49811) 4 people 2026-08-20 23:13:43 +08:00
• cb09dd7488 [Core][Multimodal] Skip redundant placeholder scan when token match succeeds (#52925) Yiqinandshenyiqin 2026-08-20 23:04:29 +08:00
• bd8865a299 [Kernel] Add FlashInfer TRTLLM MXFP8 linear backend (#52204) SeonjinandMisha Goin 2026-08-20 07:07:22 -07:00
• de216b6e64 [Bugfix] Skip MM processor cache inserts larger than capacity (#53016) Prudhvi Vuda 2026-08-20 09:42:50 -04:00
• 4b7cb949a9 [Docker] Update to nixl-1.3.2 (#51777) Sandeep MaddipatlaandKunshang Ji 2026-08-20 06:03:32 -07:00
• df1376907b [Model] Add tower and connector LoRA support for LFM2-VL (#51498) Zupeng Wang 2026-08-20 20:33:43 +08:00
• 6259572b28 [Docs] Use incremental builds for C++ changes in AGENTS.md (#53098) Thien TranandOpenAI Codex 2026-08-20 19:08:50 +08:00
• 6df7adc17f [Bugfix][GDN] Reset speculative decode count for an empty draft schedule (#53077) Kevin H. Luu 2026-08-20 03:33:41 -07:00
• f0c14b4f77 Fix weight tying (#51665) Harry Mellor 2026-08-20 11:31:55 +01:00
• 6e85feb1b7 [3/N] Harden Transformers modelling backend multi-modal path (#51827) Harry Mellor 2026-08-20 11:24:41 +01:00
• 1eab6fef01 [CI] replace shellcheck script with shellcheck-py hook (#52572) Turner JabbourandClaude Sonnet 5 2026-08-20 04:09:59 -06:00
• 5b1e7a812b [BugFix][Mooncake] Fix Mooncake saves from sparse Mamba block tables (#51362) Ziming Huang 2026-08-20 18:05:43 +08:00
• 727274a75b [Rust Frontend] Fix Qwen parser auto-detection (#51169) 3 people 2026-08-20 12:39:05 +03:00
• 30e2394c83 [Bugfix] Record non-ImportError attention backend probe failures instead of crashing engine init (#51703) Eoin-HoustounandClaude Fable 5 2026-08-20 10:36:00 +01:00
• 6b68db441e [ROCm] Fix DeepSeek V4 indexer numerics and coverage (#50803) Andreas Karatzas 2026-08-20 04:34:00 -05:00
• 44cf3f0466 [XPU][Tests] Make tests device-agnostic (#51968) pmanczak 2026-08-20 11:25:41 +02:00
• 38e9cefdef [Bugfix] Return HTTP 400 instead of 501 for unknown chat roles in DeepSeek encoders (#53071) JC-ut0 2026-08-20 09:23:41 +00:00
• c8de519917 [Kernel][Kimi] fused vision q/k roper kernel (#50400) rongfu.lengandIsotr0py 2026-08-20 17:15:47 +08:00
• 5d4d470470 [CI] Fix nonexistent dependency for data-parallel example test selection (#53026) Taneem Ibrahim 2026-08-20 04:11:00 -05:00
• c0a25c089a [Bugfix][Security] Guard _load_ov2_processor with resolve_trust_remote_code (#52952) Juan Pérez de Algaba 2026-08-20 11:09:49 +02:00
• 963fcfa48c [Rust][Benchmark] Align speed-bench CLI flags with Python and add flag parity test (#51592) 3 people 2026-08-20 16:50:51 +08:00
• 4666a8ba9e [Refactor] Remove InputPreprocessor (#53064) Cyrus Leung 2026-08-20 16:39:48 +08:00
• 14617c2b6c [Bugfix] DeepEP-V2: expert_tokens_meta must be None on the decode/cudagraph path (empty recv_expert_num_tokens) (#52632) Anton AandRoger Wang 2026-08-20 04:11:44 -04:00
• a34fd69106 [CI][Bugfix] Update distributed DP API server test path (#52939) Kevin H. Luu 2026-08-20 01:10:06 -07:00
• 16cfe728d8 [Bugfix][Rust Frontend] Reject n > 1 in the /inference/v1/generate route (#52844) Quentin GallouédecandBugen Zhao 2026-08-20 03:41:06 -04:00
• d66300a1ba [Bugfix][EPD] Fix encoder round-robin fan-out (#52491) Ankit Nakhawa 2026-08-20 03:30:40 -04:00
• a1c5b1fd9f [Core][V1] Support trace_decode_token_ids for deterministic decode replay (#46701) 6 people 2026-08-20 15:12:43 +08:00
• 754e1c3de4 [CI][XPU] Skip test_fused_shared_expert.py on XPU (#53035) Qiming Zhang 2026-08-19 23:56:09 -07:00
• d626108b18 [ROCm][Perf] Fuse DeepSeek-V4 mHC post/pre and RMSNorm with AITER (#52737) Shanshan Shen 2026-08-20 12:51:04 +08:00
• cd5035379c [ROCm] [Bugfix] Preserve CPU query offsets during capture (#51585) akii96 2026-08-20 07:33:37 +03:00
• 9b2aef5124 [Bugfix] Video loading: sample over presentable frames, not header sample count (MP4 edit-list trims) (#48608) 3 people 2026-08-20 06:23:09 +02:00
• 11baa0ebd8 [Attention] Avoid redundant mask compute in GDN metadata build (#52078) Xin Yang 2026-08-19 20:52:42 -07:00
• 4f66bc3e0d [CI][ROCm] Standardize AMD test job labels by device (#52976) Andreas Karatzas 2026-08-19 22:33:25 -05:00
• e85dd21497 fix: report stop_sequence stop_reason in Anthropic Messages API (#45807) Yufeng HeandChauncey 2026-08-20 11:23:43 +08:00
• f85e060ec2 [Doc] Update Gaudi HPU committers (#52726) pmanczakandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-20 04:59:22 +02:00
• 76fb6d210a [Bugfix][Core] Reserve the KV null block when validating max_model_len (#47272) 3 people 2026-08-20 11:49:23 +09:00
• c0233fcf01 [Model] Remove unused DeepseekV32Indexer forward (#53021) Woosuk KwonandOpenAI Codex 2026-08-19 19:39:52 -07:00
• fbb4c04db4 [ROCm][CI] Speed up test_rocm_aiter_qk_norm_rope_kvcache_fusion (#53004) 3 people 2026-08-19 21:20:21 -05:00
• 6a962071bd [Distributed] Enable FlashInfer all-reduce by default (#52998) Woosuk KwonandOpenAI Codex 2026-08-19 18:17:43 -07:00
• 58e5ee0158 [refactor] consolidate cp attn ops (#52839) Summer YangandWoosuk Kwon 2026-08-19 18:04:11 -07:00
• bf2866f8bf [KV Connector] Add decode offloading to Mooncake Store consumers (#52466) Guanyi Chenandz-zanez 2026-08-20 08:40:35 +08:00
• d4f4d3f40f [Model Runner V2][Spec Decode] Fix draft logits cache column stride in gumbel_sample (#53017) Giancarlo Delfin 2026-08-19 17:39:30 -07:00
• 0a111cca2c [kv_offload] fix(metrics): rename kv_offload_tiering_block_{queries,hits} → chunk (#52812) Ronen Schaffer 2026-08-20 03:24:07 +03:00
• 5bf0dbd6fd [Bugfix] vLLM crashes at startup when DeepEP v2 is used with --enforce-eager wiht TRTLLM Bf16 (#51824) Sage Moore 2026-08-19 17:18:47 -07:00
• 823ec22b78 [ROCm]: Bump triton 3.7 commit (#52819) Rohan Potdar 2026-08-19 17:39:08 -05:00
• c205726108 [CI] Fix and extend PR/issue auto-labeling (#51459) 3 people 2026-08-19 14:45:44 -07:00
• d591d1d511 [Bugfix] Add Kimi K3 MoE support to benchmark_moe.py (#50082) vanshbhatia-amdandVansh Bhatia 2026-08-19 15:42:30 -04:00
• 541c6d64c1 [Bugfix][Quantization] Support CT block FP8 with Marlin (#52966) Misha GoinandOpenAI Codex 2026-08-19 15:41:26 -04:00
• cb58bb9c1e [CI] Harden RemoteVLLMServer GPU cleanup checks (#52282) Andreas KaratzasandOpenAI Codex 2026-08-19 14:17:17 -05:00
• 480d4f0d15 [ROCm][CI] Enable modular OAI Triton MoE tests (#46434) Andreas Karatzas 2026-08-19 13:46:12 -05:00
• 755492e37d Revert "[Kernel] Gemma-4 FA4 FP8 Kernel" (#52987) Roger Wang 2026-08-19 10:48:46 -07:00
• f76d71d7ce [CI/Build] Fix CPU platform pre-commit formatting (#52981) Misha GoinandOpenAI Codex 2026-08-19 13:32:37 -04:00
• 9b5f3454f2 [LoRA] Avoid false target matches for unsupported module types (#52313) linitra24andJee Jee Li 2026-08-20 00:48:11 +08:00
• 0c8c3f41cf [Bugfix][V1] Sync mamba_block_size via EngineCoreReadyResponse (#50809) lxyand“新颐” 2026-08-20 00:44:36 +08:00
• 3a386cfaf5 [ROCm] Give EngineCore cleanup grace after request abort (#52281) 3 people 2026-08-19 11:39:59 -05:00
• e9e1630e93 [Model] Support bidirectional (encoder-only) attention for DeepSeek e… (#52948) EgorandClaude 2026-08-19 19:36:27 +03:00
• 54dd98be28 [CT] Support Humming for WNA16 MoE (#48918) 5 people 2026-08-20 00:31:32 +08:00
• c676232313 [Bugfix][Quantization] Fix OCP MX MoE emulation silently skipping mxfp6 activation QDQ (#52704) xuebwang-amd 2026-08-20 00:29:29 +08:00
• 583a00257d [ROCm] [Bugfix] Fix Triton fused shared expert alignment (#51632) akii96 2026-08-19 19:14:52 +03:00
• 525b7bbb3a [Bugfix][CPU] Enable C++ causal_conv1d GDN path and float32 SSM cache on non-AMX AVX-512BF16 CPUs (#49688) Dinesh Chitlangia 2026-08-19 11:50:59 -04:00
• 17dbd42930 [ROCm] Add UE8M0 scale packing for Triton silu_mul_quant (#37835) Andreas Karatzas 2026-08-19 10:20:05 -05:00