Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• d154d90d6c [Model Runner V2] batch-sharded sample (#50465) Giancarlo Delfin 2026-08-24 15:49:57 -07:00
• 0d7d5ed0b2 fix(security): enforce VLLM_MAX_AUDIO_CLIP_FILESIZE_MB on all audio paths (#53561) Juan Pérez de Algaba 2026-08-24 23:37:02 +02:00
• 23ab0cfdbc speculative decoding under tensor parallelism (TP>1) , workspace creation select max hidden dim of target and draft model (#52193) Khushali Desai 2026-08-24 14:33:36 -07:00
• f59bb0bd8c [Refactor] Refactor batch invariance folder (#53619) Wentao Ye 2026-08-24 17:31:36 -04:00
• 6a9c69fa85 [Attention][Spec Decode] Support varlen trtllm-gen decode for adaptive verification (#52157) Guan-Ming Chiu 2026-08-25 02:43:57 +08:00
• 342b8ebd8b [Bugfix][Multimodal] Encode text in mixed CLIP/SigLIP pooling batches (#53165) Prudhvi Vudaandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-24 13:45:42 -04:00
• 4f686e182a [MISC] Cleanup deprecated parameters (#53559) Jee Jee LiandCyrus Leung 2026-08-25 01:19:04 +08:00
• 6648eb118d [Core] drop duplicate VLLM_USE_DEEP_GEMM check (#48687) Ning Xie 2026-08-25 00:30:50 +08:00
• 7797b6022c [Kernel][Perf] Per-architecture tuned configs for batch-invariant persistent matmul (~3x decode kernels on RTX 4090D/H20) (#53247) Lio Einaudiandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-24 23:28:51 +08:00
• 29c9af5211 [Kimi K3 Refactor] Add UnfinalizedMoEOutput proto following up for #53152 (#53310) Wentao Ye 2026-08-24 11:21:11 -04:00
• 4c56e62c85 [Bugfix][Kimi K3] Skip absent metadata during CUDA graph profiling (#53581) Kevin H. LuuandCodex 2026-08-24 08:16:28 -07:00
• f620499ee3 [Bugfix][Spec Decode] Reapply group geometry for FlashAttention metadata (#53336) Misha Goin 2026-08-24 11:08:43 -04:00
• 9cef631f30 [Frontend] Move run_batch.py out openai folder (#53500) wang.yuqi 2026-08-24 23:04:10 +08:00
• 22099afc64 [Bugfix][DCP] Handle sparse MLA metadata after DCP Manager refactor (#52377) cjackal 2026-08-25 00:02:48 +09:00
• ecfa7bb373 [MM] Cache common token sequences (#53560) Cyrus Leung 2026-08-24 22:48:10 +08:00
• 4ca856b0b5 [Bugfix] Release worker RPC payload before next dequeue (#51979) 石皮幼鸟andOpenAI Codex 2026-08-24 22:27:36 +08:00
• 9f295fe8ce [Docs][Security] Document multimodal media UUID security implications (#53582) Juan Pérez de Algaba 2026-08-24 16:19:18 +02:00
• 79bb395eea [Pooling] Improve BGE-M3 sync pooling throughput by up to 3.13% (#53464) Taneem Ibrahim 2026-08-24 08:33:32 -04:00
• 41ae917c12 Add Cohere ChatV2 render endpoint (#53219) andrewbcohere 2026-08-24 08:27:57 -04:00
• e239947777 [Bugfix][Frontend] Fix run_batch upload retrying on success and unawaited error body (#50588) Tan ChaoandClaude 2026-08-24 04:33:44 -07:00
• 0ecc284790 [Bugfix][Kernel] Handle kernel block sizes in V2 DCP slot mapping (#51031) Nicolò Lucchesi 2026-08-24 09:28:38 +02:00
• a047e2543d Fix MNNVL Lamport mailbox publication and cleanup (#53000) Ace EldeibandCodex 2026-08-24 02:56:54 -04:00
• 26858770ec [Bugfix] Reuse CUDA streams in packed weight transfer to cap reserved-memory waste (#52951) Quentin Gallouédec 2026-08-24 02:54:41 -04:00
• cc40c3673b [Xeon][doc]add Xeon recipes into table (#53226) Louie Tsai 2026-08-23 23:42:28 -07:00
• a4d70bef37 [Model Runner V2] Reserve CUDA graph memory (#53306) 3 people 2026-08-23 23:11:19 -07:00
• 8c2bbe00d5 [Bugfix][LoRA] Add multimodal module mapping for Muse-Glimmer (#53513) linitra24 2026-08-24 13:51:43 +08:00
• 460c08bc8a [LoRA] Add Qwen3-Omni multimodal LoRA support (#52786) linitra24 2026-08-24 13:25:23 +08:00
• e6e1af4ca1 [Perf] Tune FlashInfer all-reduce selection on SM103 (#53318) Summer Yang 2026-08-23 22:19:59 -07:00
• f94666b60d [Bugfix][XPU] Skip oneCCL warm-up all_reduce when world_size == 1 (#52389) DaeHyun Sung 2026-08-24 13:31:33 +09:00
• 702e1d7186 [LoRA] feat: Support LoRA for DeepSeek V4 (#53361) ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟andJee Jee Li 2026-08-24 06:57:57 +03:00
• e8888b2d68 [Model] Add FP8 quantization support for ModernBERT (#53101) ThunGuoandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-24 11:04:19 +08:00
• 2ec6f0d71e Reject oversized media before fully downloading it (#51896) Clinton ThomasandLucas Bourtoule 2026-08-23 21:15:09 -04:00
• 585bb07c70 feat: add SSE keep-alive comments for idle streaming responses (#51034) Dieter Schwarzmann 2026-08-24 03:13:02 +02:00
• a7195188a4 Add MTP support for Nemotron VL models (#53121) danisereb 2026-08-24 03:59:56 +03:00
• cd329413e2 [Misc] Use VLLMValidationError in Cohere request validation (#52467) FrankandCodex 2026-08-23 17:52:27 -07:00
• 185cada36b [Model] Fix KV cache layout and optimize Dots3 NOTE Omni encoders (#53460) 范裕达 2026-08-24 01:02:35 +08:00
• 6cd9713463 [RL] P2P RDT weight sync (#43375) Aaron Hao 2026-08-23 09:42:22 -07:00
• 39eba6ac36 [Misc] Use VLLMValidationError in offline inference input validation (#53432) Jensen ChenandClaude Opus 5 2026-08-23 08:43:51 -07:00
• b26039b09f Add routed expert loading for gpt-oss (#52209) Wyett 2026-08-23 02:59:08 -07:00
• e25c586b90 [CI/Build] Pin Cython below 3.3 for arm64 tilelang sdist (#53358) Kevin H. LuuandOpenAI Codex 2026-08-23 01:09:27 -07:00
• ebcd606467 [MM] Simplify prompt updates: replace PromptSeq with list[int] (#53372) Cyrus LeungandQwen Code 2026-08-23 14:50:32 +08:00
• a3561ef8e4 [Bugfix][Model] Mistral3: fix image placeholder grid for processor size overrides (#52874) anmolgupt 2026-08-22 21:34:38 -07:00
• d569e71bd0 [CI][AMD] Honor single-node Docker workload timeout (#52547) Andreas Karatzas 2026-08-22 22:37:55 -05:00
• 9dba2c9b37 [Rust Frontend][RL]: report engine world size over gRPC (#53204) Biswa Panda 2026-08-22 19:31:23 -07:00
• 30b34171b1 [ROCm][CI] Restore attention coverage after KV-cache layout refactor (#53351) Andreas Karatzas 2026-08-22 20:00:25 -05:00
• bbe8b23e1a [Deprecation] Remove dead use_prefill_decode_attention flag (#52557) 3 people 2026-08-22 15:38:10 -07:00
• 2f55ef254c [Model] Add Qwen3-Omni DSpark support (#52560) Zhou248 2026-08-23 06:35:22 +08:00
• a014e35f38 [Pooling] Fix Rust pooling endpoint for benchmark (#53352) Taneem Ibrahim 2026-08-22 14:45:09 -05:00
• 811d12ee26 [Mypy Fix] Mypy fix for "vllm/model_executor/models/[eE][fF]" (#53381) Wentao Ye 2026-08-22 15:41:27 -04:00
• 236f78cc5c [Core][RL] Support sparse checkpoint updates through native weight loaders (#50723) Shuolei Wangandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-23 01:39:58 +08:00
• 9eb9d9d395 [Perf] Support internal prefill checkpoints for Mamba prefix caching, 9%~25% TTFT improvement (#52789) 3 people 2026-08-22 13:32:40 -04:00
• 020829f66b [MM] Remove renderer_applies_updates flag (#53385) Cyrus Leung 2026-08-23 01:25:55 +08:00
• 610bfc58e9 [ROCm] Ship rocprofiler-sdk 1.3.2 in Dockerfile.rocm_base to fix torch.profiler traces (#53182) Rohan Potdar 2026-08-22 12:16:25 -05:00
• da329cc303 [Bugfix] Fix speculative decoding for short_conv (LFM2) models (#50272) 3 people 2026-08-22 15:34:22 +02:00
• 704f12aa1e [Pooling] Report input throughput for batched requests (#53213) Taneem Ibrahimandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-22 08:01:40 -05:00
• 10704541aa [CI][ARM64] Restore known-good CUDA 13 builder image (#53374) Kevin H. LuuandOpenAI Codex 2026-08-22 03:50:33 -07:00
• 040700aaa6 [MM] Address comments on #53275 (#53364) Cyrus Leung 2026-08-22 18:27:59 +08:00
• 0b19ebcacd [MM] Simplify _apply_hf_processor_main (#53275) Cyrus Leung 2026-08-22 16:49:37 +08:00
• 7ca49fbe4b [Refactor][Model Runner V2][Multimodal] Move the encoder-only path out of the shared runner (#53176) Tianyu Guo 2026-08-22 13:11:44 +08:00
• e9d1398d9e [Bugfix][Kimi K3] Enable deferred MoE finalization before weight loading (#53327) Yongye ZhuandOpenAI Codex 2026-08-21 20:23:32 -07:00
• b2db227a7c [Bugfix][R3] Unwrap UniformTypeKVCacheSpecs when selecting the routed-experts KV group (#53240) ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟 2026-08-22 05:13:53 +03:00
• 7f4a1b7e24 [Bugfix][ROCM] Fix the MXFP8 block scale exponent (#53110) stefankoncarevic 2026-08-22 03:26:19 +02:00
• 8bdc70ec7b [6/N][KV-Cache Layout Refactor] Standardize KV cache layout (#51718) Lucas Wilkinson 2026-08-21 20:47:49 -04:00
• 9ff7041b55 [Bugfix][Spec Decode] Use group geometry for FlashAttention metadata (#53002) Misha Goin 2026-08-21 20:07:19 -04:00
• a34f2ab70a [Test] Add focused hybrid MTP prefix-cache regressions (#53189) Misha Goin 2026-08-21 20:05:39 -04:00
• a556f3fccb Forward Anthropic vllm_xargs to sampling params (#53308) Maroon Ayoub 2026-08-22 01:23:21 +03:00
• f94dcde5c5 Fix Cohere ChatV2 citation and tool handling issues (#52175) andrewbcohere 2026-08-21 18:23:17 -04:00
• f37d5868b9 [CI/Build][ROCm] Run the TileLang HIP symbol checks in their own interpreter (#53117) stefankoncarevic 2026-08-21 23:17:42 +02:00
• e3f6026503 Revert "Remove native Hunyuan V1 and VL implementations" (#53296) Kevin H. LuuandAndreas Karatzas 2026-08-21 13:39:58 -07:00
• 592e06f2ae Revert "[ROCm][Perf] Kimi-K3 Fused kernels for KDA prefill" (#53294) Kevin H. LuuandAndreas Karatzas 2026-08-21 13:34:13 -07:00
• 0a3a4f7f35 [Bugfix][KV Connector][NIXL] Support PCP producers (#52779) Lucas WilkinsonandOpenAI Codex 2026-08-21 16:22:01 -04:00
• 3e47a9a824 [CI] Fix MultiConnector accuracy test lifecycle (#53023) 3 people 2026-08-21 15:19:19 -05:00
• 29b7c2f7d4 [ROCm][CI] aiter kernel ops - enable rope test (#52854) Divakar Verma 2026-08-21 16:17:34 -04:00
• a0af854f5d [ROCm][CI] Stabilize MI355 FlyDSL MoE accuracy test (#53024) Andreas Karatzas 2026-08-21 15:17:18 -05:00
• b5f7fcc79d [ROCm][CI] Add float16 dtype and unsupported head size tests for paged attention (#53177) Divakar Verma 2026-08-21 16:16:45 -04:00
• 88eb946cb1 [ROCm][CI] Stabilize MI355 FusedMoE test group (#53025) Andreas Karatzas 2026-08-21 15:15:25 -05:00
• c0ff33404b Revert compile-cache device index regression on CPU (#53304) Kevin H. Luu 2026-08-21 13:08:13 -07:00
• f15ea66b30 [ROCm][Test] Use platform FP8 dtype in ModelOpt FP8_PB_WO test (#53268) djramic 2026-08-21 22:06:19 +02:00
• e6f35d3c69 [DSv4 Perf] Adaptive topk width for dsv4, making #50004 back (#52823) Wentao Ye 2026-08-21 15:11:45 -04:00
• d6c2fec9fd [Cleanup][MLA] Remove FlashInfer DSpark DCP support (#53139) Summer Yang 2026-08-21 11:42:52 -07:00
• d9e0ace7a0 [Bugfix][Attention] Fall back to native FlashInfer decode when XQA cannot serve a KV-cache group's head_dim (#53111) 3 people 2026-08-21 20:41:24 +02:00
• ba53da60bb Revert "[Bugfix][MoE] Tune FlashInfer experts to scheduler token limit" (#52989) (#53186) vllm-agentandvllm-ci-failure-analyzer 2026-08-21 11:39:28 -07:00
• 7a2fdbaac4 [K3 Perf] Fuse MXFP4 top-k finalization into latent-tail, ~5% E2E latency reduction (#53152) 3 people 2026-08-21 13:29:59 -04:00
• fe76112ff2 [ROCm][Perf] Optimize DeepSeek V4 C4A top-k with AITER (#52882) Fangzhou Ai 2026-08-21 13:00:39 -04:00
• 1baf372bfc [Frontend] Use VLLMValidationError for batch request URL validation (#50191) Tan Chao 2026-08-21 09:07:24 -07:00
• 47cd1c8885 [Core] Add dynamo_timed tracing for print_readable (#40834) Frederik Gossen 2026-08-21 12:07:21 -04:00
• e00a034ad1 (security) fix: enforce decoder prompt-length validation for skip-che… (#46588) Juan Pérez de Algaba 2026-08-21 18:07:19 +02:00
• b41dc4ec72 [Bugfix] Return HTTP 500 for non-streaming generate errors (#49195) Wei-Cheng (Wayne) Chiu 2026-08-22 00:07:17 +08:00
• 6dcc5d7cae [Bugfix] Include device index in compile cache paths (#38962) Neil SchemenauerandTyler Michael Smith 2026-08-21 09:07:16 -07:00
• 463aa5e30f [ROCm][Perf] Kimi-K3 Fused kernels for KDA prefill (#52606) 3 people 2026-08-22 00:07:13 +08:00
• d53b1c2efc Remove native Hunyuan V1 and VL implementations (#53272) Tiezhen WANGandOpenAI Codex 2026-08-22 00:07:10 +08:00
• 27ec8ac626 [Bugfix] Fix MTP draft model using local cache path instead of S3 URL with runai_streamer (#42079) SoluMilken 2026-08-22 00:07:08 +08:00
• 72aedcc426 [DCP] Default query replication for GLM sparse attention (#50382) 3 people 2026-08-21 12:07:05 -04:00
• cd7b7c265a [ROCm] Cpu offload for ROCm 7.13+ to align the hipMemcpyBatchAsync params and perf in 7.14x (#43018) Hongxia YangandYifan Qiao 2026-08-21 12:07:03 -04:00
• c1d7a3808d [Bugfix] Fix HYV3 shared_mlp prefix for compressed-tensors ignore matching (#48682) DCo 2026-08-22 00:07:01 +08:00
• 1183f04b74 [Bugfix] test_batch_inference_correctness now uses batch invariance (#48040) Morrison TurnanskyandTyler Michael Smith 2026-08-21 12:06:59 -04:00
• fbb17e780b [Bugfix] Fix six quantization exception messages split across positional args (#50479) ErenAta16 2026-08-21 19:06:56 +03:00
• 18aa245a99 using existing uvicorn configuration for dp supervisor (#52473) Greg Pereira 2026-08-21 09:06:54 -07:00
• 08b73d07cc [Frontend] Prevent Kimi K3 reserved markers in response text (#52889) Bugen ZhaoandOpenAI Codex 2026-08-22 02:06:52 +10:00
• 6d8cd88e8d [Docs] document cache salting for prefix cache timing side-channel mitigation (#39082) Russell Bryant 2026-08-21 12:06:49 -04:00