Highlights
This release features 584 commits from 270 contributors (76 new)!
- Kimi-K3 performance push: a major optimization effort for Kimi-K3 across the stack — Decode Context Parallel (DCP) support (#50484), fused FlashKDA decode and prefill kernels (#50654, #51311, #52458), SiTU activation support for MegaMoE (#50510), GEMM-RS for sequence parallelism (#52079), combined all-gathers with 1.5~3x kernel-level speedup (#51070), an adaptive speculative token budget delivering ~60% better DSpark TTFT (#51725), and optional shared-expert sharding saving ~17 GiB of memory per GPU (#50912). Kimi-K3 also now runs on ROCm with the V2 model runner (#51653).
- DeepSeek V4: sparse MLA now works end-to-end for plain decode, MTP, and DSpark speculative decoding (#51538), joined by AMD Quark NVFP4 support (#47972), reasoning-effort prompts and mappings (#50580), sparse top-k metadata kernel optimizations (#52084, #51967), narrowed eager CUDA graph regions (#51430, #52401), and ROCm enablement on gfx11 and gfx950 (#47017, #52212).
- Speculative decoding advances: DFlash2 with local convolution and a candidate selector (#52816), DSpark confidence-scheduled verification (#47808), and async scheduling auto-enabled for draft models (#48341).
- Model Runner V2 maturation: E/P/D disaggregation (#38390), weight offloading (#51413), multi-layer MTP KV cache support (#50062), encoder CUDA graphs (#49852), decoder token-wise pooling (#50931) plus Transformers pooling models (#52425), attention-free models (#52374), and
thinking_token_budget support (#46727).
- Tiered KV cache offloading: disk offloading support (#49644), out-of-tree secondary tier managers via
module_path (#51007), partial secondary-tier load results (#50321), tiering metrics (#48798), and a canonical CPU layout for parallelism-agnostic offload (#48414).
- Rust frontend & gRPC: a standalone renderer (#50289), multimodal image inference over gRPC (#50368), explicit data-parallel rank routing (#51178), and RL lifecycle control (#51316), with protobuf schemas now published to Buf (#51276).
- New defaults:
max_num_batched_tokens raised from 8192 to 16384 (#51726), prefix caching enabled by default for Mamba models (#50991), and the Blackwell CUDA graph capture default raised to 1024 (#49390).
- Breaking changes: bitsandbytes support migrated to an out-of-tree plugin (#43529); Transformers bumped to 5.15.0 (#51668); the deprecated
calculate_kv_scales runtime KV scale calculation was removed (#49389); override_attention_dtype was removed (#48684).
Release Artifacts
Python Wheels
| Platform |
Install |
| PyPI (CUDA 13.0) |
pip install vllm |
| PyPI (CUDA 13.0, uv) |
uv pip install vllm --torch-backend=auto |
| ROCm |
pip install vllm --extra-index-url https://wheels.vllm.ai/rocm/0.28.0/rocm722 |
Docker Images
| Platform |
Docker Image |
| CUDA 13.0 (Default) |
docker pull vllm/vllm-openai:v0.28.0 (v0.28.0-cu130 also works) |
| CUDA 12.9 |
docker pull vllm/vllm-openai:v0.28.0-cu129 |
| CUDA 13.0 + Ubuntu 24.04 |
docker pull vllm/vllm-openai:v0.28.0-ubuntu2404 |
| CUDA 12.9 + Ubuntu 24.04 |
docker pull vllm/vllm-openai:v0.28.0-cu129-ubuntu2404 |
| ROCm |
docker pull vllm/vllm-openai-rocm:v0.28.0 |
| CPU |
docker pull vllm/vllm-openai-cpu:v0.28.0 |
| XPU |
docker pull vllm/vllm-openai-xpu:v0.28.0 |
Other Artifacts
Pre-built release artifacts are available in the Assets section at the bottom of this page, including:
- Source distribution tarball
- CUDA 12.9 Python wheels for x86_64 and arm64
- CUDA 13.0 Python wheels for x86_64 and arm64
- CPU Python wheels for x86_64, arm64, and macOS
Model Support
- New models: Muse Glimmer (#51655), Ling 3.0 Flash with BF16, MTP, and parser support (#51045) plus an FP8 variant (#51265) and hybrid MXFP4 routed experts (#52114), Dots3 NOTE native multimodal support (#51255), and Interns2mobius (#51149).
- Qwen: Qwen3.8 enabled on AMD ROCm (#50068), fused CUDA post-conv MTP decode kernel for Qwen3.5 GDN (#51674), GDN gates aligned with speculative tokens (#51812), and Qwen3.5 fixes for text-only checkpoints (#50734, #50355).
- Transformers modeling backend: MLA support (#48250), hardware-agnostic model definition (#49458), fully generalized input embedding handling (#51247), logit softcapping (#52173), and a hardened multimodal path (#51408, #51657).
- LoRA: vision tower LoRA for Gemma4 (#42662), tower/connector LoRA for Keye (#51780) and Ultravox (#48215).
- Vision encoders: ViT full CUDA graph for Kimi-K2.5 (#50929) and Ernie-4.5-VL (#45254, #51461), torch.compile for the Qwen3-VL encoder (#40116), and long-blocking H2D copies avoided in ViT (#51841).
- MoE: extended EPLB support for Mistral Large 3 and additional MoE backends (#48355), CuTe DSL skinny GEMM extended to GLM-5.2 (#49791).
- Speculative decoding coverage: EAGLE3 support declared on KimiLinear (#52171), Qwen3.6 dSpark acceptance coverage (#51310).
- Correctness: MiniMax-M3 NVFP4 inference (#48929) and compressed-tensors FP8 MoE SwiGLU params (#46845), Gemma3n/Gemma4 variable-length audio batch padding (#50958), Gemma 4 compatibility with the upcoming Transformers version (#49797), and a Qwen3-Omni crash on video without an audio track (#48420).
- Multimodal performance: fused on-device multimodal preprocess normalization (#50411), faster placeholder and token-match scanning (#50716), and repeated prompt-update scans avoided (#51774).
Engine Core
- Speculative decoding: DSpark confidence-scheduled verification (#47808), top-k DSpark Markov projection (#49969), DFlash2 with local convolution and a candidate selector (#52816), async scheduling auto-enabled for draft models (#48341), fused MTP trailing all-reduce with local-argmax draft tokens (#49793), and an adaptive budget for speculative scheduled input tokens (#51725).
- KV cache & scheduling: per-request scheduling for MLA chunked context (#50613), partial-tail prefix reuse with fine-grained prefix matching (#50507), backend-published KV packing in the KV-cache layout refactor (#51612, #51704), LIFO free-block reuse order restored when prefix caching is off (#51482), and silent request skipping in priority scheduling fixed (#49206).
- Performance: continued elimination of GPU<->CPU syncs on the execution path (#51458, #51738, #52369) now guarded by a CI sync check (#43107), new JIT warmup infrastructure with predicate filtering (#49315), the top-k/top-p Triton sampler launched with 8 warps (#51507), detokenization skipped in offline beam search (#50333), Mask Replay (#49577), optimized long-context MLA cache gathers (#51739), and HF revisions resolved to a commit hash once per model load (#49990).
- Hybrid/Mamba: prefix caching on by default (#50991), the final part of the Mamba attention module refactor (#44857), 3D-grid tiling of the state-copy Triton kernels (#49436), and Mamba alignment applied before encoder caps (#51603).
- RL workflows: stateful trainer send over NCCL and sparse NCCL (#50902),
CuMemAllocator.discard() for tag-selective GPU memory release (#52514), level-2 sleep/wake/reload fixed with LoRA enabled (#39935), and rewritten weight-transfer docs with standardized examples (#51729).
- Startup robustness: file:// rendezvous for single-node executors eliminates startup port races (#50999, #51652), frontend processes are watched during engine startup (#43417), a
get_open_port() livelock on DP-reserved ports was fixed (#50965), and NVML is no longer re-initialized on every device-capability check (#50393).
Hardware & Performance
- NVIDIA: FlashInfer XQA decode support on SM12x (#49718), a CuTeDSL fused query kernel on SM100 (#49792), programmatic dependent launch for the DSA decode kernels (#50230), the native DSA decode path for MTP=3 on SM90 (#52164), GB10 fused-MoE FP8 tuning configs (#52502), and B12X dense linear backends (#52016).
- AMD ROCm: torch 2.12 / triton 3.7 stack bump (#50607), AITER and FP8 inference enabled on GFX120x (#43615), DeepSeek-V4 on gfx11 (#47017), optimized Triton sparse-MLA decode on gfx950 (#52212), FlyDSL decode-attention kernel for 4-bit TurboQuant KV cache (#47896) and an fp8 MQA logits kernel on gfx942 (#49544), a fused Kimi-K3 KDA decode kernel (#50654), fused bf16→fp32 router GEMM (#50268), pinned memory on supported WSL2 kernels (#50126), and preshuffled sparse indexing for 16-token blocks (#51216).
- Intel XPU: a torch linear backend including blockwise GEMM (#49664, #50826), MXFP8 linear weights for the INC DeepSeek V4 model (#48476), async-scheduling PP sampled-token broadcast overlapped with compute (#51650), an XPU wheel added to the release pipeline (#52108), tuned Mamba SSU configs for Arc Pro B70 (#50534), and UVA weight offloading fixes (#51770).
- CPU: an MLA backend so DeepSeek-V2/V3 can run on CPU (#49453), a triton-cpu wheel (#52092), GPTQ and AWQ enabled on s390x (#51148) along with tcmalloc (#50841), BF16 MoE routed through zentorch on AMD (#44201), an unquantized MoE backend for Power (VSX) (#51624), unquantized MoE migrated to the modular-kernel experts structure (#50133), and the MXFP4 block scale folded in 2 instructions instead of 4 (#51583).
Large Scale Serving
- E/P/D disaggregation: Model Runner V2 E/P/D support (#38390), duplicate image preprocessing removed with GPU-side preprocessing (#50390), KV consumers may omit multimodal embeddings (#52697), encoder-instance requests kept alive until their images are encoded (#50275), and EC connector scheduler/worker metadata plumbing (#49579, #49585).
- KV offloading: disk offloading for SimpleCPUOffloadConnector (#49644), out-of-tree secondary tier managers via
module_path (#51007), partial secondary-tier load results (#50321), tiering offloading metrics (#48798), data-parallel topology exposed to offloading backends (#51879), a canonical CPU layout for parallelism-agnostic offload (#48414), and quadratic ARC batch eviction avoided (#50992).
- Mooncake: store group semantics (#44956), tenant ID support (#48069), and official wheels in the Docker image (#51067).
- Connectors: transfer mode (push/pull) included in the NIXL compatibility hash (#50620), a MoRIIO per-layer READ-completion barrier (#48534), 2P2D wide-EP with mori-ep/mori-io at dp=ep=16 (#45043), and stale remote cleanup in the push connector (#50234).
- Parallelism: EPLB balancedness calculation fixed and tested (#51813), dense multinode DP rescope (#49212).
Quantization
- Online quantization: online MXFP4 support (#49347), online weight scales shared across TP (#49764), precision preserved in online NVFP4 expert packing (#50029), and the online NVFP4 MoE kernel reused across reloads (#50074).
- NVFP4: batch-invariant NVFP4 MoE via CUTLASS (#40372), KV 4-over-6 scale search (#45187), CuTeDSL MoE with SwiGLU-OAI and ReLU2 activations (#47106), and out_dtype matched to the model dtype (#48861).
- New kernels: block-wise scaled_mm (#49932), DeepSeek-V4 AMD Quark NVFP4 with an emulation kernel (#47972).
- Fixes: dynamic INT8 W8A8 MoE config no longer built as W8A16 (#50833) and a TritonExperts crash (#51411), MXFP4 conversion for FlashInfer CUTLASS (#51038), fp32 weight scales and per-expert checkpoint mapping for MXFP4 (#51419), and fused block-scale orientation (#50727).
API & Frontend
- New capabilities: request priority parsed from an HTTP header (#51089), session ID plumbing into requests (#48048),
count_reasoning_tokens in the streaming parser engine (#45802), content_parts on /inference/v1/generate (#51478), model optional on all /derender request classes (#51463), output token IDs logged at DEBUG level (#52098), and vLLM Recipes connected to native config-based deployment and benchmarking (#51308, #51878).
- Rust frontend: a standalone renderer (#50289), gRPC multimodal image inference (#50368), explicit data-parallel rank routing (#51178), RL lifecycle control (#51316), dynamic tools from developer messages (#51144), protobuf schemas published to Buf (#51276), and MiniJinja upgraded to 2.22 (#51235).
- Anthropic API: 4xx returned for client-caused errors on
/v1/messages (#52246), disable_parallel_tool_use preserved (#52021), and stop sequences bounded (#51997).
- Cohere: upstreamed parser fixes (#51998), stop sequences reported correctly (#51556), and vectorized binary embedding bit-packing (#52277).
- Structured output: request stop tokens masked in xgrammar until the grammar terminates (#49227, #50595), NUL bytes rejected in
structured_outputs.regex (#51796), negative token IDs rejected as out-of-vocabulary (#51795), and VLLMValidationError raised from validators (#52394).
- Robustness: uvicorn signal handlers disabled instead of racing them (#50916), a consolidated entrypoint exception handler (#52261), 400 instead of 500 on non-object JSON bodies (#51654, #52528), generation inputs bounded before expensive work (#51447), and
cache_salt now required to be non-empty (#50816).
Security
- Fixed a DoS via sample-rate forgery that bypassed the audio decode duration guard (#49948); the audio decode duration limit is now also enforced in NanoNemotronVL (#50221).
- DeepStream classified as a GPU backend with pixel limits enforced (#50755).
_load_ov2_processor guarded with resolve_trust_remote_code (#52952).- Documentation now warns that
--api-key does not gate all endpoints (#51999).
Dependencies
- Transformers 5.15.0 (#51668), huggingface-hub 1.27.0 (#51422), fastsafetensors upgrade (#50827).
- ROCm: torch 2.12, triton 3.7, torchaudio, torchvision (#50607).
- Runtime image upgraded to Ubuntu 24.04, picking up rdma-core > 44 (#51058).
- DeepGEMM pinned to the deepseek-ai nv_dev tip (#52035), DeepEP pinned by full commit hash (#52028), FlashAttention 3 built with the torch stable API (#49599).
Breaking Changes & Deprecations
- bitsandbytes support is now an out-of-tree plugin (#43529).
- The deprecated
calculate_kv_scales runtime KV scale calculation was removed (#49389).
override_attention_dtype was removed (#48684).reasoning_content output removal is documented as a breaking client change (#50624).- KV offload tiering metrics renamed from
kv_offload_tiering_block_{queries,hits} to ..._chunk_... (#52812).
- MoE legacy code removed (#51078).
New Contributors
Contributors
@yewentao256, @AndreasKaratzas, @njhill, @mgoin, @aoshen02, @khluu, @hmellor, @stefankoncarevic, @taneem-ibrahim, @LucasWilkinson, @chaunceyjiang, @jikunshang, @zufangzhu, @bigPYJ1151, @NickLucche, @askliar, @fxmarty-amd, @Rohan138, @gty111, @zhenwei-intel, @Isotr0py, @jeejeelee, @ZJY0516, @wangxiyuan, @BugenZhao, @yma11, @zyongye, @DarkLight1337, @jperezdealgaba, @zhou9402, @connorcarpenter15, @kliuae, @lucifer1004, @Alex-ai-future, @aarushjain29, @TheEpicDolphin, @chaojun-zhang, @pmanczak, @WoosukKwon, @BabyDrangoner, @noooop, @almogtavor, @hongxiayang, @fuscof-ibm, @gau-nernst, @zxd1997066, @tlrmchlsmth, @music-dino, @zexplorerhj, @S1ro1, @jdebache, @sfeng33, @mayuyuace, @ganeshr10, @benchislett, @tzulingk, @gcanlin, @ivanium, @divakar-amd, @R3hankhan123, @sagearc, @vhagor, @ronensc, @micah-wil, @qyYue1389, @vanshbhatia-amd, @hao-aaron, @chengy-sysu, @elvircrn, @taking-lying-flat, @omerpaz95, @Etelis, @vllmellm, @frank-suwen, @KernelClint, @Fangzhou-Ai, @louie-tsai, @simondanielsson, @maxyanghu, @dmai-afk, @KurodaKanbei, @ziqifan617, @bastefaniak, @ECMGit, @haregali, @cmiyai, @fede-kamel, @drakosha, @vineethsaivs, @zcxGGmu, @TQCB, @skysnow2001, @shenoyvvarun, @karen-sy, @fattchris, @RyanJHamby, @shikamd123, @namgyu-youn, @zzt93, @abmfy, @reidliu41, @Rapisurazurite, @tandixit95, @mganczarenko, @yimdev, @anujbolewar, @LiuYinfeng01, @lk-chen, @NVShreyas, @huangzhilin-hzl, @varoudis, @Yejing-Lai, @mkhazraee, @jzakrzew, @TrainToGPB, @waynehacking8, @zixi-qi, @Sundaresan-G, @mindungil, @bitborne, @Wauplin, @jacobzhang22, @zhewenl, @bnellnm, @pisceskkk, @Lin-z-w, @gabriel-peracio, @SilenNaihin, @baodii, @YunzhuLu, @xwu-intel, @BWAAEEEK, @thisjiang, @maobaolong, @anhtra3889, @JaredforReal, @lvhan028, @xiaolong-intel, @andyxning, @cleonard530, @gnovack, @MatthewBonanni, @wangxian001, @lengrongfu, @Tejas-Raj01, @simon-mo, @vitamin-chaos, @arpera, @jairitAge, @jimmy-adams, @ILikeIneine, @woosebastian, @haic0, @edwinlim0919, @fcui-amd, @jhu960213, @jinzhen-lin, @coltonottley, @walterbm, @meenchen, @matteso1, @djramic, @gchinora, @davidjpyu, @tianmu-li, @xiaopusun, @majunze2001, @Vegetog, @puririshi98, @janeyx99, @RobbieJ, @oonyshch, @thegoldenflow, @Srinivasoo7, @fatday, @acheamponge, @efschu, @rajfirke, @fanxingran, @xudonlyu, @lcskrishna, @xijiaat, @GirasoleY, @d4l3k, @samuelkim7, @tarukumar, @acmore, @theminghuang, @khushali9, @wzhao18, @Priyjain-amd, @yiz-liu, @lkm2835, @dmholtz, @Dao007forever, @liushujia122, @LopezCastroRoberto, @UgaTheDev, @tuukkjs, @aditi-amd, @guan404ming, @yiliu30, @zou3519, @Luosuu, @JoursBleu, @varun-sundar-rabindranath, @mpashkovskii, @yu-xin-c, @WillZZZy, @vrdn-23, @xyang16, @ccrhx4, @tanpinsiang, @russellb, @fxfxfxfxfxfxfxfx, @afriedri, @yifjiang, @Akashcodes732, @HF-001, @ovidiusm, @arthurgao2003, @TomerBN-Nvidia, @hotTea123, @vx120, @bohnstingl, @qwerqwerqwe8688-jpg, @jasonozuzu-cohere, @vineetatiwari27, @ruirui6946, @linitra24, @syedalijaseem, @nickus, @yzong-rh, @s3woz, @jhaotingc, @lukealonso, @Jie-Fang, @kzwrime, @xianbaoqian, @velonica0, @ccaadaro, @yisustc, @fangchenli, @iwannagotobed, @zobinHuang, @rchalamala, @shanjiaz, @jamesETsmith, @stacyroberts, @guanxingithub, @biswapanda, @shanewidanagama, @UranusSeven, @hsusul, @tobymao, @mispa-ms, @jeffreywang88, @SayHelloToWorld, @jyan-R, @oops-oom, @shantipriya-amd, @andakai, @akii96, @shen-shanshan, @Kaif10, @yitingdc, @positive666, @pavelzak, @SubSir, @ywang96