vLLM v0.9.2 — Release Notes & Engineering Impact

## Highlights This release contains 452 commits from 167 contributors (31 new!) **NOTE: This is the last version where V0 engine code and features stay intact. We highly recommend migrating to V1 engine.** ### Engine Core * Priority Scheduling is now implemented in V1 engine (#19057), embedding models in V1 (#16188), Mamba2 in V1 (#19327). * Full CUDA‑Graph execution is now available for all FlashAttention v3 (FA3) and FlashMLA paths, including prefix‑caching. CUDA graph now has a live capture progress bar makes debugging easier (#20301, #18581, #19617, #19501). * FlexAttention update – any head size, FP32 fallback (#20467, #19754). * Shared `CachedRequestData` objects and cached sampler‑ID stores deliver perf enhancements (#20232, #20291). ### Model Support * New families: Ernie 4.5 (+MoE) (#20220), MiniMax‑M1 (#19677, #20297), Slim‑MoE “Phi‑tiny‑MoE‑instruct” (#20286), Tencent HunYuan‑MoE‑V1 (#20114), Keye‑VL‑8B‑Preview (#20126), GLM‑4.1 V (#19331), Gemma‑3 (text‑only, #20134), Tarsier 2 (#19887), Qwen 3 Embedding & Reranker (#19260), dots1 (#18254), GPT‑2 for Sequence Classification (#19663). * Granite hybrid MoE configurations with shared experts are fully supported (#19652). ### Large‑Scale Serving & Engine Improvements * Expert‑Parallel Load Balancer (EPLB) has been added! (#18343, #19790, #19885). * Disaggregated serving enhancements: Avoid stranding blocks in P when aborted in D's waiting queue (#19223), let toy proxy handle /chat/completions (#19730) * Native xPyD P2P NCCL transport as a base case for native PD without external dependency (#18242, #20246). ### Hardware & Performance * NVIDIA Blackwell * SM120: CUTLASS W8A8/FP8 kernels and related tuning, added to Dockerfile (#17280, #19566, #20071, #19794) * SM100: block‑scaled‑group GEMM, INT8/FP8 vectorization, deep‑GEMM kernels, activation‑chunking for MoE, and group‑size 64 for Machete (#19757, #19572, #19168, #19085, #20290, #20331). * Intel GPU (V1) backend with Flash‑Attention support (#19560). * AMD ROCm: full‑graph capture for TritonAttention, quick All‑Reduce, and chunked pre‑fill (#19158, #19744, #18596). * Split‑KV support landed in the unified Triton Attention kernel, boosting long‑context throughput (#19152). * Full‑graph mode enabled in ROCm AITER MLA V1 decode path (#20254). * TPU: dynamic‑grid KV‑cache updates, head‑dim less than 128, tuned paged‑attention kernels, and KV‑padding fixes (#19928, #20235, #19620, #19813, #20048, #20339). * Add models and features supporting matrix. (#20230) ### Quantization * Calibration‑free RTN INT4/INT8 pipeline for effortless, accurate compression (#18768). * Compressed‑Tensor NVFP4 (including MoE) + emulation; FP4 emulation removed on < SM100 devices (#19879, #19990, #19563). * Dynamic MoE‑layer quant (Marlin/GPTQ) and INT8 vectorization primitives (#19395, #20331, #19233). * Bits‑and‑Bytes 0.45 + with improved double‑quant logic and AWQ quality (#20424, #20033, #19431, #20076). ### API · CLI · Frontend * API Server: Eliminate api_key and x_request_id headers middleware overhead (#19946) * New OpenAI‑compatible endpoints: `/v1/audio/translations` & revamped `/v1/audio/transcriptions` (#19615, #20179, #19597). * Token‑level progress bar for `LLM.beam_search` and cached template‑resolution speed‑ups (#19301, #20065). * Image‑object support in `llm.chat`, tool‑choice expansion, and custom‑arg passthroughs enrich multi‑modal agents (#19635, #17177, #16862). * CLI QoL: better parsing for `-O/--compilation-config`, batch‑size‑sweep benchmarking, richer `--help`, faster startup (#20156, #20516, #20430, #19941). * Metrics: Deprecate metrics with gpu_ prefix for non GPU specific metrics (#18354), Export NaNs in logits to scheduler_stats if output is corrupted (#18777) ### Platform & Deployment * No‑privileged CPU / Docker / K8s mode (#19241) and custom default max‑tokens for hosted platforms (#18557). * Security hardening – runtime (cloud)pickle imports forbidden (#18018). * Hermetic builds and wheel slimming (FA2 8.0 + PTX only) shrink supply‑chain surface (#18064, #19336). ## What's Changed * [Docs] Note that alternative structured output backends are supported by @russellb in https://github.com/vllm-project/vllm/pull/19426 * [ROCm][V1] Adding ROCm to the list of plaforms using V1 by default by @gshtras in https://github.com/vllm-project/vllm/pull/19440 * [Model] use AutoWeightsLoader for commandr by @py-andy-c in https://github.com/vllm-project/vllm/pull/19399 * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B-FP8 by @Xu-Wenqing in https://github.com/vllm-project/vllm/pull/19401 * [BugFix] Allow use_cudagraph to work with dynamic VLLM_USE_V1 by @zou3519 in https://github.com/vllm-project/vllm/pull/19390 * [New Model]: Support Qwen3 Embedding & Reranker by @noooop in https://github.com/vllm-project/vllm/pull/19260 * [BugFix] Fix docker build cpu-dev image error by @2niuhe in https://github.com/vllm-project/vllm/pull/19394 * Fix test_max_model_len in tests/entrypoints/llm/test_generate.py by @houseroad in https://github.com/vllm-project/vllm/pull/19451 * [CI] Disable failing GGUF model test by @mgoin in https://github.com/vllm-project/vllm/pull/19454 * [Misc] Remove unused `MultiModalHasher.hash_prompt_mm_data` by @lgeiger in https://github.com/vllm-project/vllm/pull/19422 * Add fused MOE config for Qwen3 30B A3B on B200 by @0xjunhao in https://github.com/vllm-project/vllm/pull/19455 * Fix Typo in Documentation and Function Name by @leopardracer in https://github.com/vllm-project/vllm/pull/19442 * [ROCm] Add rules to automatically label ROCm related PRs by @houseroad in https://github.com/vllm-project/vllm/pull/19405 * [Kernel] Support deep_gemm for linear methods by @artetaout in https://github.com/vllm-project/vllm/pull/19085 * [Doc] Update V1 User Guide for Hardware and Models by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/19474 * [Doc] Fix quantization link titles by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/19478 * [Doc] Support "important" and "announcement" admonitions by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/19479 * [Misc] Reduce warning message introduced in env_override by @houseroad in https://github.com/vllm-project/vllm/pull/19476 * Support non-string values in JSON keys from CLI by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/19471 * Add cache to cuda get_device_capability by @mgoin in https://github.com/vllm-project/vllm/pull/19436 * Fix some typo by @Ximingwang-09 in https://github.com/vllm-project/vllm/pull/19475 * Support no privileged mode on CPU for docker and kubernetes deployments by @louie-tsai in https://github.com/vllm-project/vllm/pull/19241 * [Bugfix] Update the example code, make it work with the latest lmcache by @runzhen in https://github.com/vllm-project/vllm/pull/19453 * [CI] Update FlashInfer to 0.2.6.post1 by @mgoin in https://github.com/vllm-project/vllm/pull/19297 * [doc] fix "Other AI accelerators" getting started page by @davidxia in https://github.com/vllm-project/vllm/pull/19457 * [Misc] Fix misleading ROCm warning by @jeejeelee in https://github.com/vllm-project/vllm/pull/19486 * [Docs] Remove WIP features in V1 guide by @WoosukKwon in https://github.com/vllm-project/vllm/pull/19498 * [Kernels] Add activation chunking logic to FusedMoEModularKernel by @bnellnm in https://github.com/vllm-project/vllm/pull/19168 * [AMD] [Quantization] Add override flag for attention dtype instead of using kv_cache_dtype trigger by @rasmith in https://github.com/vllm-project/vllm/pull/17331 * [UX] Add Feedback During CUDAGraph Capture by @robertgshaw2-redhat in https://github.com/vllm-project/vllm/pull/19501 * [CI/Build] Fix torch nightly CI dependencies by @zou3519 in https://github.com/vllm-project/vllm/pull/19505 * [CI] change spell checker from codespell to typos by @andyxning in https://github.com/vllm-project/vllm/pull/18711 * [BugFix] Force registration of w8a8_block_fp8_matmul_deepgemm via lazy import by @varun-sundar-rabindranath in https://github.com/vllm-project/vllm/pull/19514 * Add Triton Fused MoE kernel config for E=16 on B200 by @b8zhong in https://github.com/vllm-project/vllm/pull/19518 * [Frontend] Improve error message in tool_choice validation by @22quinn in https://github.com/vllm-project/vllm/pull/19239 * [BugFix] Work-around incremental detokenization edge case error by @njhill in https://github.com/vllm-project/vllm/pull/19449 * [BugFix] Handle missing sep_token for Qwen3-Reranker in Score API by @strutive07 in https://github.com/vllm-project/vllm/pull/19522 * [AMD][Kernel][BugFix] fix test_rocm_compressed_tensors_w8a8 for rocm by @rasmith in https://github.com/vllm-project/vllm/pull/19509 * Fix typo by @2niuhe in https://github.com/vllm-project/vllm/pull/19525 * [Security] Prevent new imports of (cloud)pickle by @russellb in https://github.com/vllm-project/vllm/pull/18018 * [Bugfix][V1] Allow manual FlashAttention for Blackwell by @mgoin in https://github.com/vllm-project/vllm/pull/19492 * [Bugfix] Respect num-gpu-blocks-override in v1 by @jmswen in https://github.com/vllm-project/vllm/pull/19503 * [Quantization] Improve AWQ logic by @jeejeelee in https://github.com/vllm-project/vllm/pull/19431 * [Doc] Add V1 column to supported models list by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/19523 * [NixlConnector] Drop `num_blocks` check by @NickLucche in https://github.com/vllm-project/vllm/pull/19532 * [Perf] Vectorize static / dynamic INT8 quant kernels by @yewentao256 in https://github.com/vllm-project/vllm/pull/19233 * Fix TorchAOConfig skip layers by @mobicham in https://github.com/vllm-project/vllm/pull/19265 * [torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass by @ProExpertProg in https://github.com/vllm-project/vllm/pull/16756 * [doc] Make top navigation sticky by @reidliu41 in https://github.com/vllm-project/vllm/pull/19540 * [Spec Decode][Benchmark] Generalize spec decode offline benchmark to more methods and datasets by @ekagra-ranjan in https://github.com/vllm-project/vllm/pull/18847 * [Misc] Turn MOE_DP_CHUNK_SIZE into an env var by @varun-sundar-rabindranath in https://github.com/vllm-project/vllm/pull/19506 * [Bugfix] Enforce contiguous input for dynamic_per_token FP8/INT8 quant by @mgoin in https://github.com/vllm-project/vllm/pull/19452 * [Doc] Unify structured outputs examples by @aarnphm in https://github.com/vllm-project/vllm/pull/18196 * [V1] Resolve failed concurrent structred output requests by @russellb in https://github.com/vllm-project/vllm/pull/19565 * Revert "[Build/CI] Add tracing deps to vllm container image (#15224)" by @kouroshHakha in https://github.com/vllm-project/vllm/pull/19378 * [BugFix] : Fix Batched DeepGemm Experts by @varun-sundar-rabindranath in https://github.com/vllm-project/vllm/pull/19515 * [Bugfix] Fix EAGLE vocab embedding for multimodal target model by @zixi-qi in https://github.com/vllm-project/vllm/pull/19570 * [Doc] uses absolute links for structured outputs by @aarnphm in https://github.com/vllm-project/vllm/pull/19582 * [doc] fix incorrect link by @reidliu41 in https://github.com/vllm-project/vllm/pull/19586 * [Misc] Correct broken docs link by @Zerohertz in https://github.com/vllm-project/vllm/pull/19553 * [CPU] Refine default config for the CPU backend by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/19539 * [Fix] bump mistral common to support magistral by @princepride in https://github.com/vllm-project/vllm/pull/19533 * [Fix] The zip function in Python 3.9 does not have the strict argument by @princepride in https://github.com/vllm-project/vllm/pull/19549 * use base version for version comparison by @BoyuanFeng in https://github.com/vllm-project/vllm/pull/19587 * [torch.compile] reorganize the cache directory to support compiling multiple models by @youkaichao in https://github.com/vllm-project/vllm/pull/19064 * [BugFix] Honor `enable_caching` in connector-delayed kvcache load case by @njhill in https://github.com/vllm-project/vllm/pull/19435 * [Model] Fix minimax model cache & lm_head precision by @qscqesze in https://github.com/vllm-project/vllm/pull/19592 * [Refactor] Remove unused variables in `moe_permute_unpermute_kernel.inl` by @yewentao256 in https://github.com/vllm-project/vllm/pull/19573 * [doc][mkdocs] fix the duplicate Supported features sections in GPU docs by @reidliu41 in https://github.com/vllm-project/vllm/pull/19606 * [CUDA] Enable full cudagraph for FlashMLA by @ProExpertProg in https://github.com/vllm-project/vllm/pull/18581 * [Doc] Add troubleshooting section to k8s deployment by @annapendleton in https://github.com/vllm-project/vllm/pull/19377 * [torch.compile] Use custom ops when use_inductor=False by @WoosukKwon in https://github.com/vllm-project/vllm/pull/19618 * Adding "AMD: Multi-step Tests" to amdproduction. by @Concurrensee in https://github.com/vllm-project/vllm/pull/19508 * [BugFix] Fix DP Coordinator incorrect debug log message by @njhill in https://github.com/vllm-project/vllm/pull/19624 * [V1][Metrics] Deprecate metrics with gpu_ prefix for non GPU specific metrics. by @sahelib25 in https://github.com/vllm-project/vllm/pull/18354 * [Bugfix][1/n] Fix the speculative decoding test by setting the target dtype by @houseroad in https://github.com/vllm-project/vllm/pull/19633 * [Misc] Modularize CLI Argument Parsing in Benchmark Scripts by @reidliu41 in https://github.com/vllm-project/vllm/pull/19593 * [Bugfix] Fix auto dtype casting for BatchFeature by @Isotr0py in https://github.com/vllm-project/vllm/pull/19316 * [Hardware][NVIDIA][kernel] Fp4 MOE quant kernel optimization by @jiahanc in https://github.com/vllm-project/vllm/pull/19500 * Only build CUTLASS MoE kernels on Hopper by @huydhn in https://github.com/vllm-project/vllm/pull/19648 * [Bugfix] Don't attempt to use triton if no driver is active by @kzawora-intel in https://github.com/vllm-project/vllm/pull/19561 * [Fix] Convert kv_transfer_config from dict to KVTransferConfig by @maobaolong in https://github.com/vllm-project/vllm/pull/19262 * [Perf] Further tunings for SM100 FP8 CUTLASS kernel by @ilmarkov in https://github.com/vllm-project/vllm/pull/19566 * [Bugfix][2/n] Fix speculative decoding CI - Fix test_ngram_e2e_greedy_correctness by @houseroad in https://github.com/vllm-project/vllm/pull/19644 * [Kernel] Raise verbose error and consolidate `num_heads/num_kv_heads` divisibility check by @22quinn in https://github.com/vllm-project/vllm/pull/19339 * [Benchmark] Refactor benchmark script for fp8 & int8 by @yewentao256 in https://github.com/vllm-project/vllm/pull/19627 * Enable prefix caching with full cuda graphs by @WoosukKwon in https://github.com/vllm-project/vllm/pull/19617 * [CI/Build] Fix torch nightly CI dependencies part 2 by @zou3519 in https://github.com/vllm-project/vllm/pull/19589 * [Misc] Remove duplicate multiproc method setting for CPU platform by @Isotr0py in https://github.com/vllm-project/vllm/pull/19649 * [MISC] Remove unused variableds in C++ by @houseroad in https://github.com/vllm-project/vllm/pull/19609 * [Bugfix][Core] Prefix caching causes incorrect outputs due to outdated ComputedBlocksTracker by @quanliu1991 in https://github.com/vllm-project/vllm/pull/18957 * [Misc][Frontend] passthrough `bad_words` by @f14-bertolotti in https://github.com/vllm-project/vllm/pull/19564 * [Misc] Fix skipped max-model-len validation when deriving max model length from tokenizer config by @yeqcharlotte in https://github.com/vllm-project/vllm/pull/19660 * [TPU] support attention head dim smaller than 128 by @yaochengji in https://github.com/vllm-project/vllm/pull/19620 * [MISC] typo fix by @andyxning in https://github.com/vllm-project/vllm/pull/19672 * [CI] Add mteb testing for rerank models by @noooop in https://github.com/vllm-project/vllm/pull/19344 * [Docs] Move multiproc doc to v1 dir by @russellb in https://github.com/vllm-project/vllm/pull/19651 * [Kernel] GGUF MMVQ kernel for multiple input vectors by @SzymonOzog in https://github.com/vllm-project/vllm/pull/18754 * [BugFix] Don't catch BaseException when dumping execute_model errors by @njhill in https://github.com/vllm-project/vllm/pull/19626 * [DOC] Add reasoning capability to vLLM streamlit code by @Navanit-git in https://github.com/vllm-project/vllm/pull/19557 * [Feature]:Allow for Granite MoE Hybrid models with _only_ shared experts. by @shawntan in https://github.com/vllm-project/vllm/pull/19652 * [Bugfix] Fix TP inference for Flex attention backend by @Isotr0py in https://github.com/vllm-project/vllm/pull/19657 * [MISC] bump huggingface_hub pkg to 0.33.0 by @andyxning in https://github.com/vllm-project/vllm/pull/19547 * [Bugfix] fix missing 'finish_reason': null in streaming chat by @chaunceyjiang in https://github.com/vllm-project/vllm/pull/19662 * [Kernels] Use empty for modular MoE workspaces by @bnellnm in https://github.com/vllm-project/vllm/pull/19667 * [Model] Add support for MiniMaxM1ForCausalLM (shares architecture with MiniMaxText01ForCausalLM) by @qscqesze in https://github.com/vllm-project/vllm/pull/19677 * [V1] Change return type on get_multimodal_embeddings() by @russellb in https://github.com/vllm-project/vllm/pull/19446 * [Quantization] Remove FP4 emulation; Fall-back to marlin for device < 100 by @dsikka in https://github.com/vllm-project/vllm/pull/19563 * [Fix] Fall back to Gloo when NCCL backend is unavailable by @conroy-cheers in https://github.com/vllm-project/vllm/pull/19641 * [doc] add project flag to gcloud TPU command by @davidxia in https://github.com/vllm-project/vllm/pull/19664 * [Wheel Size] Only build FA2 8.0+PTX by @LucasWilkinson in https://github.com/vllm-project/vllm/pull/19336 * [Frontend] add chunking audio for > 30s audio by @nguyenhoangthuan99 in https://github.com/vllm-project/vllm/pull/19597 * [DOC] fix doc typos by @diliu0349 in https://github.com/vllm-project/vllm/pull/19600 * Fixes IMA for TP w/ flex-attention by @drisspg in https://github.com/vllm-project/vllm/pull/19712 * [Core] add remove_seq_from_computed_blocks_tracker to BlockSpaceManager by @quanliu1991 in https://github.com/vllm-project/vllm/pull/19686 * [Doc] Add missing llava family multi-image examples by @Isotr0py in https://github.com/vllm-project/vllm/pull/19698 * Add a doc on how to update PyTorch version by @huydhn in https://github.com/vllm-project/vllm/pull/19705 * [Kernel] Add Split-KV Support to Unified Triton Attention Kernel by @jvlunteren in https://github.com/vllm-project/vllm/pull/19152 * [doc][mkdocs] Add edit button to documentation by @reidliu41 in https://github.com/vllm-project/vllm/pull/19637 * [doc] split "Other AI Accelerators" tabs by @davidxia in https://github.com/vllm-project/vllm/pull/19708 * [V1][Kernel] Flashinfer HND KV cache layout by @NickLucche in https://github.com/vllm-project/vllm/pull/19280 * [Mis] remove duplicate engine status checks by @googs1025 in https://github.com/vllm-project/vllm/pull/19647 * [Bugfix] Update multimodel models mapping to fit new checkpoint after Transformers v4.52 by @Isotr0py in https://github.com/vllm-project/vllm/pull/19151 * [Perf] Optimize `moe_align_block_size` CUDA kernel by @yewentao256 in https://github.com/vllm-project/vllm/pull/19572 * Remove sm120 arch from sm100 cutlass kernel arch list by @mgoin in https://github.com/vllm-project/vllm/pull/19716 * [Misc] Update lmcache connector with the latest connector apis by @YaoJiayi in https://github.com/vllm-project/vllm/pull/19441 * [Bugfix] Fix faulty triton importing logic when using Ray for DP by @mgoin in https://github.com/vllm-project/vllm/pull/19734 * [Feature][ROCm] Add full graph capture support for TritonAttentionBackend by @charlifu in https://github.com/vllm-project/vllm/pull/19158 * [TPU] Update torch version to include paged attention kernel change by @Chenyaaang in https://github.com/vllm-project/vllm/pull/19706 * [MISC] correct copy_blocks src_to_dists param type by @andyxning in https://github.com/vllm-project/vllm/pull/19696 * [MISC] correct DeviceConfig device field static type analysis by @andyxning in https://github.com/vllm-project/vllm/pull/19699 * [Misc] Add __str__ for RequestStatus by @lk-chen in https://github.com/vllm-project/vllm/pull/19780 * [V1] Add API docs for EncoderCacheManager by @russellb in https://github.com/vllm-project/vllm/pull/19294 * [V1][P/D] An native implementation of xPyD based on P2P NCCL by @Abatom in https://github.com/vllm-project/vllm/pull/18242 * [V1] Decouple GPU and TPU `InputBatch` by @afeldman-nm in https://github.com/vllm-project/vllm/pull/19778 * [Minor] Zero-initialize attn output buffer by @WoosukKwon in https://github.com/vllm-project/vllm/pull/19784 * [doc] fix the incorrect label by @reidliu41 in https://github.com/vllm-project/vllm/pull/19787 * [Platform] Allow platform use V1 Engine by default by @wangxiyuan in https://github.com/vllm-project/vllm/pull/19792 * [Qwen] Add tagging rule for Qwen related PRs by @houseroad in https://github.com/vllm-project/vllm/pull/19799 * [Hardware][AMD] integrate aiter chunked prefill into vllm by @Zzz9990 in https://github.com/vllm-project/vllm/pull/18596 * [Bugfix] fix RAY_CGRAPH_get_timeout is not set successfully by @chaunceyjiang in https://github.com/vllm-project/vllm/pull/19725 * [Docs] Add Huzaifa Sidhpurwala to vuln mgmt team doc by @russellb in https://github.com/vllm-project/vllm/pull/19808 * [v1] Support mamba2 by @heheda12345 in https://github.com/vllm-project/vllm/pull/19327 * docs: fix Slack bulletpoint in README by @nathan-weinberg in https://github.com/vllm-project/vllm/pull/19811 * Disable "Forbid direct 'import triton'" check for `vllm/triton_utils/importing.py` in an extensible way by @afeldman-nm in https://github.com/vllm-project/vllm/pull/19783 * [Core] Do not copy array during hashing by @lgeiger in https://github.com/vllm-project/vllm/pull/19484 * [TPU] Update torch-xla version to include paged attention tuned block change by @QiliangCui in https://github.com/vllm-project/vllm/pull/19813 * [Core] More fixes to MultiModalEmbeddings type handling by @russellb in https://github.com/vllm-project/vllm/pull/19715 * [Multimodal] Use fast processor for Qwen2/2.5-VL by @WoosukKwon in https://github.com/vllm-project/vllm/pull/19789 * [BugFix] Fix use_cudagraph=False by @zou3519 in https://github.com/vllm-project/vllm/pull/19612 * [Frontend] Expose custom args in OpenAI APIs by @afeldman-nm in https://github.com/vllm-project/vllm/pull/16862 * Fix FA2 fallback for Blackwell V1 by @mgoin in https://github.com/vllm-project/vllm/pull/19781 * [Misc][ROCm] Enforce no unused variable in ROCm C++ files by @houseroad in https://github.com/vllm-project/vllm/pull/19796 * [Quantization] Modify the logic of BNB double quantization by @jeejeelee in https://github.com/vllm-project/vllm/pull/19742 * Support embedding models in V1 by @maxdebayser in https://github.com/vllm-project/vllm/pull/16188 * [Bugfix] Fix the linter by @houseroad in https://github.com/vllm-project/vllm/pull/19826 * [Bugfix] Add check_health to v1 async client. by @kouroshHakha in https://github.com/vllm-project/vllm/pull/19821 * Mark invariant normalizer in Gemma as non-persistent by @yhtang in https://github.com/vllm-project/vllm/pull/19788 * [ROCm] [AITER] [Bugfix] Patch for AITER commit `648764942e552a8bb5fe16026703716a81f05374` by @tjtanaa in https://github.com/vllm-project/vllm/pull/18990 * [Misc] [ROCm] Prevent surplus tensor reshape by @zsolt-borbely-htec in https://github.com/vllm-project/vllm/pull/19803 * raise exception for pin_lora by @andyxning in https://github.com/vllm-project/vllm/pull/19809 * [Minor] Allow redirecting model path for HfRunner in test by @Isotr0py in https://github.com/vllm-project/vllm/pull/19795 * Add xLAM tool parser support by @zuxin666 in https://github.com/vllm-project/vllm/pull/17148 * [Frontend] Add optional token-level progress bar to `LLM.beam_search` by @NekoMimiUnagi in https://github.com/vllm-project/vllm/pull/19301 * Fixing Chunked Prefill Test. by @Alexei-V-Ivanov-AMD in https://github.com/vllm-project/vllm/pull/19762 * [Doc] Update V1 user guide for embedding models by @22quinn in https://github.com/vllm-project/vllm/pull/19842 * [CI][CPU] Improve dummy Triton interfaces and fix the CPU CI by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/19838 * [Core][Bugfix] Fix Online MM Beam Search by @alex-jw-brooks in https://github.com/vllm-project/vllm/pull/19688 * [Frontend] early return chat format resolution when specified by @xzbdmw in https://github.com/vllm-project/vllm/pull/19735 * [Benchmark][Bugfix] Fix Dataset Length Calculation by @robertgshaw2-redhat in https://github.com/vllm-project/vllm/pull/19868 * [CI/Build][Bugfix] Fix deadlock on v1 engine test CI by @Isotr0py in https://github.com/vllm-project/vllm/pull/19872 * [CI][Neuron] Fail and exit on first error by @elaineyz in https://github.com/vllm-project/vllm/pull/19622 * [Benchmark] Fix `Value of type "SampleRequest" is not indexable` by @b8zhong in https://github.com/vllm-project/vllm/pull/18032 * [Chore]: qwen3-moe-type-hints-mistake by @Xerxes-cn in https://github.com/vllm-project/vllm/pull/19860 * [Bugfix] Enable PP with AITER+V1 by @qli88 in https://github.com/vllm-project/vllm/pull/19822 * [Bugfix][Ray] Set the cuda context eagerly in the ray worker by @kouroshHakha in https://github.com/vllm-project/vllm/pull/19583 * [Misc] update cuda version by @reidliu41 in https://github.com/vllm-project/vllm/pull/19526 * [Misc] refactor example - openai_transcription_client by @reidliu41 in https://github.com/vllm-project/vllm/pull/19851 * [Kernel] correct cpu worker function parameter type by @andyxning in https://github.com/vllm-project/vllm/pull/19745 * [Fix] import regex instead of re by @tdoublep in https://github.com/vllm-project/vllm/pull/19875 * [Model] GPT2ForSequenceClassification model by @nie3e in https://github.com/vllm-project/vllm/pull/19663 * [custom_op][vllm-plugin] update custom_op class to use op_registry by @xuechendi in https://github.com/vllm-project/vllm/pull/19164 * Export NaNs in logits to scheduler_stats if output is corrupted by @vladmihailescu in https://github.com/vllm-project/vllm/pull/18777 * [CPU][CI] Fallback sliding window to v0 and fix CPU pooling model tests by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/19901 * [Kernel] mark TorchSDPABackend swap_blocks NotImplementedError by @andyxning in https://github.com/vllm-project/vllm/pull/19749 * [Misc] Clean up useless code by @wangxiyuan in https://github.com/vllm-project/vllm/pull/19889 * Fix: Check the type of params to be a Sequence not list. by @rabinadk1 in https://github.com/vllm-project/vllm/pull/19910 * [Bugfix] Fix bnb 8bit model weights loading by @Isotr0py in https://github.com/vllm-project/vllm/pull/19917 * [New model support]Support Tarsier2 by @princepride in https://github.com/vllm-project/vllm/pull/19887 * [doc] add contact us in community by @reidliu41 in https://github.com/vllm-project/vllm/pull/19922 * [Multimodal] Optimize Qwen2/2.5-VL startup time by @WoosukKwon in https://github.com/vllm-project/vllm/pull/19756 * [Docs] Add GPT2ForSequenceClassification to supported models in docs by @nie3e in https://github.com/vllm-project/vllm/pull/19932 * [Misc] add vllm_config in __init__ by @andyxning in https://github.com/vllm-project/vllm/pull/19866 * [MISC] add cpu_kvcache_space_bytes to CacheConfig by @andyxning in https://github.com/vllm-project/vllm/pull/19812 * [Benchmark] fix request loss if "ping" is returned by @sywangyi in https://github.com/vllm-project/vllm/pull/19535 * [CI/Build] Auto tag perf benchmarks related PRs by @22quinn in https://github.com/vllm-project/vllm/pull/19943 * [doc] use snippets for contact us by @reidliu41 in https://github.com/vllm-project/vllm/pull/19944 * [Misc] Update model-specific PR tagging by @ywang96 in https://github.com/vllm-project/vllm/pull/19949 * [Misc] Simplify vllm bench cli subcommand implementation by @yeqcharlotte in https://github.com/vllm-project/vllm/pull/19948 * [Chore] dedup logs by @aarnphm in https://github.com/vllm-project/vllm/pull/19955 * [BugFix] Add an env to disable moe chunking to work around compile incompatibility by @yeqcharlotte in https://github.com/vllm-project/vllm/pull/19642 * [Perf][CLI] Improve overall startup time by @aarnphm in https://github.com/vllm-project/vllm/pull/19941 * [Core] feat: Implement Priority Scheduling in V1 Engine by @amitm02 in https://github.com/vllm-project/vllm/pull/19057 * [Misc] Configurable timeout for execute_model RPC calls via env var by @jinqinn in https://github.com/vllm-project/vllm/pull/19544 * Fix(models/siglip): Add compatibility for Gemma models quantized by llm-compressor by @Flink-ddd in https://github.com/vllm-project/vllm/pull/19643 * [doc] Fold long code blocks to improve readability by @reidliu41 in https://github.com/vllm-project/vllm/pull/19926 * [P/D][NixlConnector] Support `tp_size > num_kv_heads` deployments by @NickLucche in https://github.com/vllm-project/vllm/pull/19691 * [BugFix][P/D] Fix for cases where _recving_transfers can be cleaned up when *all* transfer done by @lk-chen in https://github.com/vllm-project/vllm/pull/19874 * [Doc] Update V1 status for decoder-only embedding models by @Isotr0py in https://github.com/vllm-project/vllm/pull/19952 * [doc] use MkDocs collapsible blocks - supplement by @reidliu41 in https://github.com/vllm-project/vllm/pull/19973 * [Bugfix] Fix CI bitsandbytes failure by @jeejeelee in https://github.com/vllm-project/vllm/pull/19969 * [doc] improve readability for long commands by @reidliu41 in https://github.com/vllm-project/vllm/pull/19920 * [Docs] Fix syntax highlighting of shell commands by @lgeiger in https://github.com/vllm-project/vllm/pull/19870 * [EP+DP] Optimize the little operations in the DeepGEMM + DeepEP low latency case by @tlrmchlsmth in https://github.com/vllm-project/vllm/pull/19885 * [Bugfix][v1] Fix step pooler implementation and step pooling usage in v1 by @Isotr0py in https://github.com/vllm-project/vllm/pull/19956 * [Misc] Add type alias `ReqId` and `EngineId` for better readability by @lk-chen in https://github.com/vllm-project/vllm/pull/19880 * [Feature] Support sequence parallelism for static fp8 quantization by @cascade812 in https://github.com/vllm-project/vllm/pull/19181 * [CI/Build] Push latest tag for cpu and neuron docker image by @22quinn in https://github.com/vllm-project/vllm/pull/19897 * Feat Dynamic Quantization for MoE Layers in GPTQ Marlin Backend by @Jun-Howie in https://github.com/vllm-project/vllm/pull/19395 * [Bugfix][Benchmark] Fix Marlin benchmark by @22quinn in https://github.com/vllm-project/vllm/pull/19929 * [TPU] Fix tpu model runner test by @Chenyaaang in https://github.com/vllm-project/vllm/pull/19995 * Update test case parameter to have the throughput above 8.0 by @QiliangCui in https://github.com/vllm-project/vllm/pull/19994 * [Misc][Tools][Benchmark] Add profile to autotune script by @Chenyaaang in https://github.com/vllm-project/vllm/pull/19711 * [doc] Fix broken link in the installation for CPU by @yankay in https://github.com/vllm-project/vllm/pull/19980 * add some examples for other benchmark scripts by @reidliu41 in https://github.com/vllm-project/vllm/pull/19893 * [PERF] Speedup of MRoPE prepare inputs by @vadiklyutiy in https://github.com/vllm-project/vllm/pull/19939 * [Bugfix][CPU] Fix InputBatch for pooling models in the CPU v1 by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/20014 * refactor example - qwen3_reranker by @reidliu41 in https://github.com/vllm-project/vllm/pull/19847 * [Fix][V1] Remove --scheduling-policy oracle by @amitm02 in https://github.com/vllm-project/vllm/pull/20010 * [Perf] Improve/Fix-regression for FA3 in High QPS regimes by @LucasWilkinson in https://github.com/vllm-project/vllm/pull/19463 * [Misc][Benchmarking] Add variable request-rate ("ramp-up") to the benchmarking client. by @dtransposed in https://github.com/vllm-project/vllm/pull/19423 * [BugFix] Fix multi-node offline data parallel by @njhill in https://github.com/vllm-project/vllm/pull/19937 * [P/D] Asynchronously do _nixl_handshake by @lk-chen in https://github.com/vllm-project/vllm/pull/19836 * [Feature] Integrate new deepgemm by @yewentao256 in https://github.com/vllm-project/vllm/pull/19820 * [Easy] Remove submodule added in #19463 by @b8zhong in https://github.com/vllm-project/vllm/pull/20039 * use .dev for version comparison with pytorch nightly release by @BoyuanFeng in https://github.com/vllm-project/vllm/pull/20031 * cmake: Update vllm_flash_attn for vllm_kernels by @seemethere in https://github.com/vllm-project/vllm/pull/20032 * [Llama4] Update `attn_temperature_tuning` by @b8zhong in https://github.com/vllm-project/vllm/pull/19997 * Revert "[Feature] Integrate new deepgemm (#19820)" by @yewentao256 in https://github.com/vllm-project/vllm/pull/20049 * Revert "Fix(models/siglip): Add compatibility for Gemma models quantized by llm-compressor" by @Isotr0py in https://github.com/vllm-project/vllm/pull/20030 * Move to a faster base64 implementation by @h-avsha in https://github.com/vllm-project/vllm/pull/19984 * [Frontend] speed up import time of vllm.config by @davidxia in https://github.com/vllm-project/vllm/pull/18036 * [Refactor] Remove duplicate `ceil_div` by @yewentao256 in https://github.com/vllm-project/vllm/pull/20023 * [Feat][CLI] enforce-include-usage by @max-wittig in https://github.com/vllm-project/vllm/pull/19695 * [Kernels][Bugfix] Use torch op for all kernels in FusedMoE forward. Add additional testing for cudagraphs. by @bnellnm in https://github.com/vllm-project/vllm/pull/19717 * [Chore] debloat some initial logs by @aarnphm in https://github.com/vllm-project/vllm/pull/19438 * [BugFix] Fix full-cuda-graph illegal memory access in FA3 by @LucasWilkinson in https://github.com/vllm-project/vllm/pull/20057 * [doc] add reference link for Intel XPU by @reidliu41 in https://github.com/vllm-project/vllm/pull/20064 * [Doc] Guide for Incremental Compilation Workflow by @mgoin in https://github.com/vllm-project/vllm/pull/19109 * [V1][Speculative Decoding] Fix DeepSeek MTP by @cjackal in https://github.com/vllm-project/vllm/pull/20022 * [Frontend] Add `/v1/audio/translations` OpenAI API endpoint by @NickLucche in https://github.com/vllm-project/vllm/pull/19615 * [Quantization] Add compressed-tensors emulations support for NVFP4 by @dsikka in https://github.com/vllm-project/vllm/pull/19879 * [Fix] Support cls pooling in ModernBertPooler by @lsz05 in https://github.com/vllm-project/vllm/pull/20067 * static_scaled_fp8_quant should not run when scale.numel is not 1 by @eldarkurtic in https://github.com/vllm-project/vllm/pull/20076 * [PD] let toy proxy handle /chat/completions by @lk-chen in https://github.com/vllm-project/vllm/pull/19730 * [Misc] Add parallel state `node_count` function by @njhill in https://github.com/vllm-project/vllm/pull/20045 * Fix the path to the testing script. by @QiliangCui in https://github.com/vllm-project/vllm/pull/20082 * [Bugfix] default set cuda_graph_sizes to max_num_seqs for v1 engine by @izhuhaoran in https://github.com/vllm-project/vllm/pull/20062 * [TPU][Bugfix] fix kv cache padding by @yaochengji in https://github.com/vllm-project/vllm/pull/20048 * [P/D] Avoid stranding blocks in P when aborted in D's waiting queue by @njhill in https://github.com/vllm-project/vllm/pull/19223 * [TPU] Add TPU specific var VLLM_TPU_MOST_MODEL_LEN by @Chenyaaang in https://github.com/vllm-project/vllm/pull/19919 * [CI] Add SM120 to the Dockerfile by @mgoin in https://github.com/vllm-project/vllm/pull/19794 * [Bugfix] Fix Mistral tool-parser regex for nested JSON by @mgoin in https://github.com/vllm-project/vllm/pull/20093 * [PD] Skip `tp_size` exchange with rank0 by @NickLucche in https://github.com/vllm-project/vllm/pull/19413 * [Benchmark][Bug] Fix multiple bugs in bench and add args to spec_decode offline by @ekagra-ranjan in https://github.com/vllm-project/vllm/pull/20083 * [Bugfix] Allow `CUDA_VISIBLE_DEVICES=''` in `Platform.device_id_to_physical_device_id` by @eicherseiji in https://github.com/vllm-project/vllm/pull/18979 * [Doc] Update docs for New Model Implementation by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20115 * [Refactor] Remove unused library by @yewentao256 in https://github.com/vllm-project/vllm/pull/20099 * [CPU] Fix torch version in x86 CPU backend by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/19258 * [Misc] Use collapsible blocks for benchmark examples. by @reidliu41 in https://github.com/vllm-project/vllm/pull/20017 * [Docs] Improve frameworks/helm.md by @windsonsea in https://github.com/vllm-project/vllm/pull/20113 * [Bugfix][V1][ROCm] Fix AITER Flash Attention Backend (Fix API Break and Local Attention Logic: affecting Llama4) by @tjtanaa in https://github.com/vllm-project/vllm/pull/19904 * Revert "[Bugfix] default set cuda_graph_sizes to max_num_seqs for v1 engine" by @mgoin in https://github.com/vllm-project/vllm/pull/20128 * [Bug Fix] Fix address/port already in use error for pplx test by @yewentao256 in https://github.com/vllm-project/vllm/pull/20094 * [Doc] Automatically signed-off by PyCharm by @noooop in https://github.com/vllm-project/vllm/pull/20120 * [Doc] Auto sign-off for VSCode by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20132 * [Doc] Rename page titles by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20130 * Spam folks if config.py changes by @tlrmchlsmth in https://github.com/vllm-project/vllm/pull/20131 * [Hardware][Intel GPU] Add v1 Intel GPU support with Flash attention backend. by @jikunshang in https://github.com/vllm-project/vllm/pull/19560 * [TPU] add kv cache update kernel by @yaochengji in https://github.com/vllm-project/vllm/pull/19928 * [Refactor] Rename commnication utils by @yewentao256 in https://github.com/vllm-project/vllm/pull/20091 * [Doc] correct LoRA capitalization by @kyolebu in https://github.com/vllm-project/vllm/pull/20135 * [Feature] Expert Parallelism Load Balancer (EPLB) by @abmfy in https://github.com/vllm-project/vllm/pull/18343 * [CI Failure] Fix OOM with test_oot_registration_embedding by @mgoin in https://github.com/vllm-project/vllm/pull/20144 * [Quantization] Bump to use latest `compressed-tensors` by @dsikka in https://github.com/vllm-project/vllm/pull/20033 * [Perf] SM100 FP8 GEMM Optimizations after cutlass_profiler by @ilmarkov in https://github.com/vllm-project/vllm/pull/20071 * [Bugfix] Build moe_data for both sm100 and sm90 by @mgoin in https://github.com/vllm-project/vllm/pull/20086 * [Feature][Rocm] add quick all reduce for rocm by @lihaoyang-amd in https://github.com/vllm-project/vllm/pull/19744 * [CI] Sync test dependency with test.in for torch nightly by @yangw-dev in https://github.com/vllm-project/vllm/pull/19632 * [Fix] Fix gemma CI test failing on main by @tdoublep in https://github.com/vllm-project/vllm/pull/20124 * [Model][1/N] Automatic conversion of CrossEncoding model by @noooop in https://github.com/vllm-project/vllm/pull/20012 * [Perf][Frontend]: eliminate api_key and x_request_id headers middleware overhead by @Yazan-Sharaya in https://github.com/vllm-project/vllm/pull/19946 * Quick Fix by adding conditional import for flash_attn_varlen_func in flash_attn by @xuechendi in https://github.com/vllm-project/vllm/pull/20143 * Gemma3n (Text-only) by @robertgshaw2-redhat in https://github.com/vllm-project/vllm/pull/20134 * [Bugfix] Fix flaky failure when getting DP ports by @mgoin in https://github.com/vllm-project/vllm/pull/20151 * [Perf][Frontend] Cached resolution for resolving chat templates by @ilyal-cerebras in https://github.com/vllm-project/vllm/pull/20065 * [Fix][ROCm] Remove unused variables to fix build error on GFX11/12 by @hyoon1 in https://github.com/vllm-project/vllm/pull/19891 * [Fix][torch.compile] Enable custom ops by default when Inductor off by @ProExpertProg in https://github.com/vllm-project/vllm/pull/20102 * [Bugfix] Mark 'hidden_states' as mutable in moe_forward registration. by @bnellnm in https://github.com/vllm-project/vllm/pull/20152 * [Bugfix] Fix some narrowing conversion warnings by @tlrmchlsmth in https://github.com/vllm-project/vllm/pull/20141 * [CI/Build] Allow hermetic builds by @fabiendupont in https://github.com/vllm-project/vllm/pull/18064 * [CI Fix] Pin tests/models/registry.py MiniMaxText01ForCausalLM to revision due to model changes by @mgoin in https://github.com/vllm-project/vllm/pull/20199 * [Misc] Add type assertion of request_id for LLMEngine.add_request by @SHA-4096 in https://github.com/vllm-project/vllm/pull/19700 * Fix num_token_padding support for static per-tensor scaled_fp8_quant by @mgoin in https://github.com/vllm-project/vllm/pull/20188 * fix ci issue distributed 4 gpu test by @yewentao256 in https://github.com/vllm-project/vllm/pull/20204 * [Bugfix] Properly reject requests with empty list guided_choice by @mgoin in https://github.com/vllm-project/vllm/pull/20195 * [BugFix] Fix the incorrect func name in the comments. (config.py) by @1195343015 in https://github.com/vllm-project/vllm/pull/20185 * [CI/Build] Add new CI job to validate Hybrid Models for every PR by @tdoublep in https://github.com/vllm-project/vllm/pull/20147 * [Frontend] Generalize `v1/audio/transcriptions` endpoint by @NickLucche in https://github.com/vllm-project/vllm/pull/20179 * [Bugfix] Correct behavior of GraniteMoeHybrid for TensorParallel execution by @s3woz in https://github.com/vllm-project/vllm/pull/20137 * [Refactor] Create a function util and cache the results for `has_deepgemm`, `has_deepep`, `has_pplx` by @yewentao256 in https://github.com/vllm-project/vllm/pull/20187 * [CI Fix] Try fixing eagle e2e test OOM by reducing block allocation by @mgoin in https://github.com/vllm-project/vllm/pull/20213 * [Quantization] Add compressed-tensors NVFP4 MoE Support by @dsikka in https://github.com/vllm-project/vllm/pull/19990 * Fix cuda_archs_loose_intersection when handling sm_*a by @huydhn in https://github.com/vllm-project/vllm/pull/20207 * [Model] support dots1 by @redmoe-moutain in https://github.com/vllm-project/vllm/pull/18254 * [BUGFIX][DEEPSEEK][MODEL_LOAD] fix w13, w2 weight not initialized assert by @xuechendi in https://github.com/vllm-project/vllm/pull/20202 * [Misc] Fix import by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20233 * [doc] Add Slack and Forum to the top navigation by @reidliu41 in https://github.com/vllm-project/vllm/pull/20208 * [Bugfix] Skip loading extra parameters for modelopt Qwen3 MoE model by @noiji in https://github.com/vllm-project/vllm/pull/19598 * [Bugfix] Fix processor initialization in transformers 4.53.0 by @Isotr0py in https://github.com/vllm-project/vllm/pull/20244 * [Quantization] Improve BitsAndBytesModelLoader by @jeejeelee in https://github.com/vllm-project/vllm/pull/20242 * [Docs] Fix 1-2-3 list in v1/prefix_caching.md by @windsonsea in https://github.com/vllm-project/vllm/pull/20243 * [Bugfix] fix quark ptpc by @lihaoyang-amd in https://github.com/vllm-project/vllm/pull/20251 * [Spec Decode] Refactor spec decoding into a separate function by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20238 * [Spec Decode] Clean up spec decode example by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20240 * [Optimization] Use Shared `CachedRequestData` Instance Across All Requests by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20232 * [Unit Test] Add unit test for deep gemm by @yewentao256 in https://github.com/vllm-project/vllm/pull/20090 * [Core] [Bugfix] [Multimodal] Fix multimodal profiling and generation for SFT/PTQed models by @kylesayrs in https://github.com/vllm-project/vllm/pull/20058 * [Refactor] Remove useless pdb comment by @yewentao256 in https://github.com/vllm-project/vllm/pull/20266 * [Bugfix][V1][P/D]Fix the issue of occasional garbled output for P2pNcclConnector by @Abatom in https://github.com/vllm-project/vllm/pull/20263 * [CLI] Improve CLI arg parsing for `-O`/`--compilation-config` by @ProExpertProg in https://github.com/vllm-project/vllm/pull/20156 * [Bugfix] Fix include prompt in stream response when echo=true by @fyuan1316 in https://github.com/vllm-project/vllm/pull/15233 * [Misc] Fix spec decode example by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20296 * [Example] add one-click runnable example for P2P NCCL XpYd by @KuntaiDu in https://github.com/vllm-project/vllm/pull/20246 * [CI][Intel Gaudi][vllm-Plugin]Add CI for hpu-plugin-v1-test by @xuechendi in https://github.com/vllm-project/vllm/pull/20196 * [Doc] add config and troubleshooting guide for NCCL & GPUDirect RDMA by @chewong in https://github.com/vllm-project/vllm/pull/15897 * [Feature] A calibration-free RTN-based quantization for accurate and accelerated INT4/INT8 inference by @sakogan in https://github.com/vllm-project/vllm/pull/18768 * [V1] Only print cudagraph tqdm on rank 0 with `is_global_first_rank` by @mgoin in https://github.com/vllm-project/vllm/pull/19516 * Fix `numel()` downcast in vllm/csrc/moe/moe_align_sum_kernels.cu +2 by @r-barnes in https://github.com/vllm-project/vllm/pull/17082 * [Misc] add xgrammar for arm64 by @prashantgupta24 in https://github.com/vllm-project/vllm/pull/18359 * Enable ZP Support for Machete by @czhu-cohere in https://github.com/vllm-project/vllm/pull/20268 * [CPU] Update custom ops for the CPU backend by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/20255 * [Bugfix] Fix deepep tests by @varun-sundar-rabindranath in https://github.com/vllm-project/vllm/pull/20288 * [Misc] remove redundant char by @kebe7jun in https://github.com/vllm-project/vllm/pull/20287 * [BugFix][V1][ROCm] Triton MLA uses V0 backend on V1 engine by @tywuAMD in https://github.com/vllm-project/vllm/pull/19067 * [doc] fix the incorrect logo in dark mode by @reidliu41 in https://github.com/vllm-project/vllm/pull/20289 * [Perf] Validate @config in pre-commit instead of dynamically by @lionelvillard in https://github.com/vllm-project/vllm/pull/20200 * [Quant] [Bugfix] Fix quantization config matching with `hf_to_vllm_mapper` by @kylesayrs in https://github.com/vllm-project/vllm/pull/20046 * [Misc] Minor refactor of NIXL background handshake by @NickLucche in https://github.com/vllm-project/vllm/pull/20068 * Add GLM4.1V model (Draft) by @zRzRzRzRzRzRzR in https://github.com/vllm-project/vllm/pull/19331 * [Model]Add Tencent HunYuanMoEV1 Model Support by @aiyiwang2025 in https://github.com/vllm-project/vllm/pull/20114 * [Misc] Minor refactoring for scheduler by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20299 * [Docs] Update transcriptions API to use openai client with `stream=True` by @NickLucche in https://github.com/vllm-project/vllm/pull/20271 * [CUDA graphs] Enable full cuda graphs with FA3 AoT scheduling by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20301 * [Frontend] Expand tools even if tool_choice="none" by @okdshin in https://github.com/vllm-project/vllm/pull/17177 * [V1] [ROCm] Enable EP with AITER Fused MoE by @tjtanaa in https://github.com/vllm-project/vllm/pull/20270 * [Optimization] Cache sampled token ids in model runner by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20291 * remove unused variables in marlin_template.h by @zhoutianzi666 in https://github.com/vllm-project/vllm/pull/20236 * [Refactor] Refactor import utils by @yewentao256 in https://github.com/vllm-project/vllm/pull/20269 * Enable group size 64 for Machete by @czhu-cohere in https://github.com/vllm-project/vllm/pull/20290 * [Kernel][Bugfix] Fixup some warnings in nvfp4_blockwise_moe when CUDA < 12.8 by @tlrmchlsmth in https://github.com/vllm-project/vllm/pull/20324 * [UT][intel GPU] use current_platform instead of device hardcode in v1 tests by @Liangliang-Ma in https://github.com/vllm-project/vllm/pull/20169 * [Refactor] Remove duplicate `find_free_port` by @yewentao256 in https://github.com/vllm-project/vllm/pull/20333 * [Refactor] Remove Unused Env `VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON` by @yewentao256 in https://github.com/vllm-project/vllm/pull/20334 * [Misc][Doc] Add missing comment for LLM by @draftbk in https://github.com/vllm-project/vllm/pull/20285 * [FIX][Intel GPU]fix ipex flash_attn_varlen_func api missing parameter by @jikunshang in https://github.com/vllm-project/vllm/pull/20348 * [Bugfix] Fix dynamic rotary embedding by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20343 * fix[Docs]: link anchor is incorrect #20309 by @yyzxw in https://github.com/vllm-project/vllm/pull/20315 * [Doc][TPU] Add models and features supporting matrix. by @QiliangCui in https://github.com/vllm-project/vllm/pull/20230 * [TPU] kv cache update kernel supports dynamic grid by @yaochengji in https://github.com/vllm-project/vllm/pull/20235 * [Frontend] Support configurable mm placeholder strings & flexible video sampling policies via CLI flags. by @huachenheli in https://github.com/vllm-project/vllm/pull/20105 * [Model][VLM] Support Keye-VL-8B-Preview by @Kwai-Keye in https://github.com/vllm-project/vllm/pull/20126 * [Bugfix] Keye-VL compatibility with `tok_kwargs` (#20058) by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20353 * [Docs] Fix indentations for 2-level items in deprecation_policy.md by @windsonsea in https://github.com/vllm-project/vllm/pull/20352 * [Docs] Make TPU ref prettier in google_tpu.md by @windsonsea in https://github.com/vllm-project/vllm/pull/20356 * [Model] Add Ernie4.5 and Ernie4.5MoE Model Support by @CSWYF3634076 in https://github.com/vllm-project/vllm/pull/20220 * [Build/CI] Automatically tag DeepSeek related PRs by @houseroad in https://github.com/vllm-project/vllm/pull/20370 * [NVIDIA] Support Cutlass w8a8 FP8 for Blackwell Geforce GPUs (sm120) by @kaln27 in https://github.com/vllm-project/vllm/pull/17280 * [Bugfix] Fix the max_seq_len limit of 16384 for DeepSeek models by @huaqiangwang in https://github.com/vllm-project/vllm/pull/20322 * [Model] Adds support for SlimMoE models Phi-tiny-MoE-instruct by @zichongli5 in https://github.com/vllm-project/vllm/pull/20286 * Documentation update tool_calling: mapping back to function from response by @cronoik-inceptionai in https://github.com/vllm-project/vllm/pull/20373 * [Kernels] MoE refactor by @bnellnm in https://github.com/vllm-project/vllm/pull/19636 * [V1] LogitsProcessor programming model by @afeldman-nm in https://github.com/vllm-project/vllm/pull/16728 * [Minor] Clean up incorrect comment in test by @njhill in https://github.com/vllm-project/vllm/pull/20382 * [Misc] add handler HF_TOKEN is emptry string by @lengrongfu in https://github.com/vllm-project/vllm/pull/20369 * [ROCm][FEAT] Enable Full Graph Mode in AITER MLA V1 Attn Backend (Decode Phase only) by @vllmellm in https://github.com/vllm-project/vllm/pull/20254 * [DP] Support external DP Load Balancer mode by @njhill in https://github.com/vllm-project/vllm/pull/19790 * [Docs] Update EAGLE example by @NickLucche in https://github.com/vllm-project/vllm/pull/20375 * [Bugfix] Fixes for FlashInfer's TORCH_CUDA_ARCH_LIST by @tlrmchlsmth in https://github.com/vllm-project/vllm/pull/20136 * [BugFix] Fix DP headless mode arg validation by @njhill in https://github.com/vllm-project/vllm/pull/20398 * Enable CPU nightly performance benchmark and its Markdown report by @louie-tsai in https://github.com/vllm-project/vllm/pull/18444 * [Bugfix] Fix import of CutlassExpertsFp8 in compressed_tensors_moe.py by @bnellnm in https://github.com/vllm-project/vllm/pull/20381 * [Misc] Small: Fix video loader return type annotations. by @huachenheli in https://github.com/vllm-project/vllm/pull/20389 * [Bugfix][CI/CD][CPU] Fix CPU CI tests by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/20383 * [TPU] Add a case to cover RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8 by @QiliangCui in https://github.com/vllm-project/vllm/pull/20385 * [Feature] Support MiniMax-M1 function calls features by @qscqesze in https://github.com/vllm-project/vllm/pull/20297 * [Tests] Update online DP tests to verify that requests are balanced by @njhill in https://github.com/vllm-project/vllm/pull/20157 * [Misc] Add rules to label Speculative Decoding Related PRs by @draftbk in https://github.com/vllm-project/vllm/pull/20406 * [doc] fix link by @reidliu41 in https://github.com/vllm-project/vllm/pull/20417 * [Docs] Replace two list with tables in intel_gaudi.md by @windsonsea in https://github.com/vllm-project/vllm/pull/20414 * [Core] Move multimodal placeholder from chat utils to model definition by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20355 * [Kernel] refactor cpu worker v0 cache dtype by @andyxning in https://github.com/vllm-project/vllm/pull/20080 * [CI/Build][CPU] Enable cross compilation in CPU release pipeline by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/20423 * [Quantization] Bump to use latest bitsandbytes by @jeejeelee in https://github.com/vllm-project/vllm/pull/20424 * [Model][2/N] Automatic conversion of CrossEncoding model by @noooop in https://github.com/vllm-project/vllm/pull/19978 * [Misc] Automatically tag PRs to add new models by @Isotr0py in https://github.com/vllm-project/vllm/pull/20222 * [Frontend] improve vllm bench <bench_type> --help display by @reidliu41 in https://github.com/vllm-project/vllm/pull/20430 * [Bugfix] Fix flaky `test_streaming_response` test by @NickLucche in https://github.com/vllm-project/vllm/pull/20363 * [Frontend] fix duplicate output for bench subcmd by @reidliu41 in https://github.com/vllm-project/vllm/pull/20446 * [CI] Trimming some failing test groups from AMDPRODUCTION. by @Alexei-V-Ivanov-AMD in https://github.com/vllm-project/vllm/pull/20390 * [Misc] Clean up InternVL family config registration by @Isotr0py in https://github.com/vllm-project/vllm/pull/19992 * [Misc] adjust for ipv6 for mookcacke url parse by @andyxning in https://github.com/vllm-project/vllm/pull/20107 * [Misc] Remove _maybe_ignore_quant_config from GLM4.1v by @zRzRzRzRzRzRzR in https://github.com/vllm-project/vllm/pull/20432 * [Kernel] Enable fp8 support for pplx and BatchedTritonExperts. by @bnellnm in https://github.com/vllm-project/vllm/pull/18864 * [Misc] Fix `Unable to detect current VLLM config. Defaulting to NHD kv cache layout` warning by @NickLucche in https://github.com/vllm-project/vllm/pull/20400 * [Bugfix] Register reducer even if transformers_modules not available by @eicherseiji in https://github.com/vllm-project/vllm/pull/19510 * Change warn_for_unimplemented_methods to debug by @mgoin in https://github.com/vllm-project/vllm/pull/20455 * [Platform] Add custom default max tokens by @gmarinho2 in https://github.com/vllm-project/vllm/pull/18557 * Add ignore consolidated file in mistral example code by @princepride in https://github.com/vllm-project/vllm/pull/20420 * [Misc] small update by @reidliu41 in https://github.com/vllm-project/vllm/pull/20462 * [Structured Outputs][V1] Skipping with models doesn't contain tokenizers by @aarnphm in https://github.com/vllm-project/vllm/pull/20365 * [Perf] Optimize Vectorization Utils for Int 8 Quantization Kernels by @yewentao256 in https://github.com/vllm-project/vllm/pull/20331 * [Misc] Add SPDX-FileCopyrightText by @jeejeelee in https://github.com/vllm-project/vllm/pull/20428 * Support Llama 4 for fused_marlin_moe by @mgoin in https://github.com/vllm-project/vllm/pull/20457 * [Bug][Frontend] Fix structure of transcription's decoder_prompt by @sangbumlikeagod in https://github.com/vllm-project/vllm/pull/18809 * [Model][3/N] Automatic conversion of CrossEncoding model by @noooop in https://github.com/vllm-project/vllm/pull/20168 * [Doc] Fix classification table in list of supported models by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20489 * [CI] add kvcache-connector dependency definition and add into CI build by @panpan0000 in https://github.com/vllm-project/vllm/pull/18193 * [Misc] Small: Remove global media connector. Each test should have its own test connector object. by @huachenheli in https://github.com/vllm-project/vllm/pull/20395 * Enable V1 for Hybrid SSM/Attention Models by @tdoublep in https://github.com/vllm-project/vllm/pull/20016 * [feat]: CUTLASS block scaled group gemm for SM100 by @djmmoss in https://github.com/vllm-project/vllm/pull/19757 * [CI Bugfix] Fix pre-commit failures on main by @mgoin in https://github.com/vllm-project/vllm/pull/20502 * [Doc] fix mutltimodal_inputs.md gh examples link by @GuyStone in https://github.com/vllm-project/vllm/pull/20497 * [Misc] Add security warning for development mode endpoints by @reidliu41 in https://github.com/vllm-project/vllm/pull/20508 * [doc] small fix by @reidliu41 in https://github.com/vllm-project/vllm/pull/20506 * [Misc] Remove the unused LoRA test code by @jeejeelee in https://github.com/vllm-project/vllm/pull/20494 * Fix unknown attribute of topk_indices_dtype in CompressedTensorsW8A8Fp8MoECutlassMethod by @luccafong in https://github.com/vllm-project/vllm/pull/20507 * [v1] Re-add fp32 support to v1 engine through FlexAttention by @Isotr0py in https://github.com/vllm-project/vllm/pull/19754 * [Misc] Add logger.exception for TPU information collection failures by @reidliu41 in https://github.com/vllm-project/vllm/pull/20510 * [Misc] remove unused import by @reidliu41 in https://github.com/vllm-project/vllm/pull/20517 * test_attention compat with coming xformers change by @bottler in https://github.com/vllm-project/vllm/pull/20487 * [BUG] Fix #20484. Support empty sequence in cuda penalty kernel by @vadiklyutiy in https://github.com/vllm-project/vllm/pull/20491 * [Bugfix] Fix missing per_act_token parameter in compressed_tensors_moe by @luccafong in https://github.com/vllm-project/vllm/pull/20509 * [BugFix] Fix: ImportError when building on hopper systems by @LucasWilkinson in https://github.com/vllm-project/vllm/pull/20513 * [TPU][Bugfix] fix the MoE OOM issue by @yaochengji in https://github.com/vllm-project/vllm/pull/20339 * [Frontend] Support image object in llm.chat by @sfeng33 in https://github.com/vllm-project/vllm/pull/19635 * [Benchmark] Add support for multiple batch size benchmark through CLI in `benchmark_moe.py` + Add Triton Fused MoE kernel config for FP8 E=16 on B200 by @b8zhong in https://github.com/vllm-project/vllm/pull/20516 * [Misc] call the pre-defined func by @reidliu41 in https://github.com/vllm-project/vllm/pull/20518 * [V0 deprecation] Remove V0 CPU/XPU/TPU backends by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20412 * [V1] Support any head size for FlexAttention backend by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20467 * [BugFix][Spec Decode] Fix spec token ids in model runner by @WoosukKwon in https://github.com/vllm-project/vllm/pull/20530 * [Bugfix] Add `use_cross_encoder` flag to use correct activation in `ClassifierPooler` by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/20527 ## New Contributors * @py-andy-c made their first contribution in https://github.com/vllm-project/vllm/pull/19399 * @2niuhe made their first contribution in https://github.com/vllm-project/vllm/pull/19394 * @leopardracer made their first contribution in https://github.com/vllm-project/vllm/pull/19442 * @artetaout made their first contribution in https://github.com/vllm-project/vllm/pull/19085 * @runzhen made their first contribution in https://github.com/vllm-project/vllm/pull/19453 * @strutive07 made their first contribution in https://github.com/vllm-project/vllm/pull/19522 * @yewentao256 made their first contribution in https://github.com/vllm-project/vllm/pull/19233 * @mobicham made their first contribution in https://github.com/vllm-project/vllm/pull/19265 * @kouroshHakha made their first contribution in https://github.com/vllm-project/vllm/pull/19378 * @BoyuanFeng made their first contribution in https://github.com/vllm-project/vllm/pull/19587 * @sahelib25 made their first contribution in https://github.com/vllm-project/vllm/pull/18354 * @jiahanc made their first contribution in https://github.com/vllm-project/vllm/pull/19500 * @quanliu1991 made their first contribution in https://github.com/vllm-project/vllm/pull/18957 * @f14-bertolotti made their first contribution in https://github.com/vllm-project/vllm/pull/19564 * @Navanit-git made their first contribution in https://github.com/vllm-project/vllm/pull/19557 * @nguyenhoangthuan99 made their first contribution in https://github.com/vllm-project/vllm/pull/19597 * @diliu0349 made their first contribution in https://github.com/vllm-project/vllm/pull/19600 * @Zzz9990 made their first contribution in https://github.com/vllm-project/vllm/pull/18596 * @yhtang made their first contribution in https://github.com/vllm-project/vllm/pull/19788 * @zsolt-borbely-htec made their first contribution in https://github.com/vllm-project/vllm/pull/19803 * @zuxin666 made their first contribution in https://github.com/vllm-project/vllm/pull/17148 * @NekoMimiUnagi made their first contribution in https://github.com/vllm-project/vllm/pull/19301 * @xzbdmw made their first contribution in https://github.com/vllm-project/vllm/pull/19735 * @Xerxes-cn made their first contribution in https://github.com/vllm-project/vllm/pull/19860 * @nie3e made their first contribution in https://github.com/vllm-project/vllm/pull/19663 * @vladmihailescu made their first contribution in https://github.com/vllm-project/vllm/pull/18777 * @rabinadk1 made their first contribution in https://github.com/vllm-project/vllm/pull/19910 * @amitm02 made their first contribution in https://github.com/vllm-project/vllm/pull/19057 * @jinqinn made their first contribution in https://github.com/vllm-project/vllm/pull/19544 * @Flink-ddd made their first contribution in https://github.com/vllm-project/vllm/pull/19643 * @Jun-Howie made their first contribution in https://github.com/vllm-project/vllm/pull/19395 * @seemethere made their first contribution in https://github.com/vllm-project/vllm/pull/20032 * @h-avsha made their first contribution in https://github.com/vllm-project/vllm/pull/19984 * @max-wittig made their first contribution in https://github.com/vllm-project/vllm/pull/19695 * @lsz05 made their first contribution in https://github.com/vllm-project/vllm/pull/20067 * @kyolebu made their first contribution in https://github.com/vllm-project/vllm/pull/20135 * @lihaoyang-amd made their first contribution in https://github.com/vllm-project/vllm/pull/19744 * @Yazan-Sharaya made their first contribution in https://github.com/vllm-project/vllm/pull/19946 * @ilyal-cerebras made their first contribution in https://github.com/vllm-project/vllm/pull/20065 * @fabiendupont made their first contribution in https://github.com/vllm-project/vllm/pull/18064 * @SHA-4096 made their first contribution in https://github.com/vllm-project/vllm/pull/19700 * @1195343015 made their first contribution in https://github.com/vllm-project/vllm/pull/20185 * @redmoe-moutain made their first contribution in https://github.com/vllm-project/vllm/pull/18254 * @noiji made their first contribution in https://github.com/vllm-project/vllm/pull/19598 * @chewong made their first contribution in https://github.com/vllm-project/vllm/pull/15897 * @sakogan made their first contribution in https://github.com/vllm-project/vllm/pull/18768 * @czhu-cohere made their first contribution in https://github.com/vllm-project/vllm/pull/20268 * @aiyiwang2025 made their first contribution in https://github.com/vllm-project/vllm/pull/20114 * @okdshin made their first contribution in https://github.com/vllm-project/vllm/pull/17177 * @zhoutianzi666 made their first contribution in https://github.com/vllm-project/vllm/pull/20236 * @yyzxw made their first contribution in https://github.com/vllm-project/vllm/pull/20315 * @Kwai-Keye made their first contribution in https://github.com/vllm-project/vllm/pull/20126 * @CSWYF3634076 made their first contribution in https://github.com/vllm-project/vllm/pull/20220 * @kaln27 made their first contribution in https://github.com/vllm-project/vllm/pull/17280 * @huaqiangwang made their first contribution in https://github.com/vllm-project/vllm/pull/20322 * @zichongli5 made their first contribution in https://github.com/vllm-project/vllm/pull/20286 * @cronoik-inceptionai made their first contribution in https://github.com/vllm-project/vllm/pull/20373 * @sangbumlikeagod made their first contribution in https://github.com/vllm-project/vllm/pull/18809 * @djmmoss made their first contribution in https://github.com/vllm-project/vllm/pull/19757 * @GuyStone made their first contribution in https://github.com/vllm-project/vllm/pull/20497 * @bottler made their first contribution in https://github.com/vllm-project/vllm/pull/20487 **Full Changelog**: https://github.com/vllm-project/vllm/compare/v0.9.1...v0.9.2