vLLM 0.11.0 release notes

Highlights

This release features 538 commits, 207 contributors (65 new contributors)!

This release completes the removal of V0 engine. V0 engine code including AsyncLLMEngine, LLMEngine, MQLLMEngine, all attention backends, and related components have been removed. V1 is the only engine in the codebase now.
This releases turns on FULL_AND_PIECEWISE as the CUDA graph mode default. This should provide better out of the box performance for most models, particularly fine-grained MoEs, while preserving compatibility with existing models supporting only PIECEWISE mode.

Note: In v0.11.0 (and v0.10.2), --async-scheduling will produce gibberish output in some cases such as preemption and others. This functionality is correct in v0.10.1. We are actively fixing it for the next version.

Model Support

New architectures: DeepSeek-V3.2-Exp (#25896), Qwen3-VL series (#24727), Qwen3-Next (#24526), OLMo3 (#24534), LongCat-Flash (#23991), Dots OCR (#24645), Ling2.0 (#24627), CWM (#25611).
Encoders: RADIO encoder support (#24595), Transformers backend support for encoder-only models (#25174).
Task expansion: BERT token classification/NER (#24872), multimodal models for pooling tasks (#24451).
Data parallel for vision encoders: InternVL (#23909), Qwen2-VL (#25445), Qwen3-VL (#24955).
Speculative decoding: EAGLE3 for MiniCPM3 (#24243) and GPT-OSS (#25246).
Features: Qwen3-VL text-only mode (#26000), EVS video token pruning (#22980), Mamba2 TP+quantization (#24593), MRoPE + YaRN (#25384), Whisper on XPU (#25123), LongCat-Flash-Chat tool calling (#24083).
Performance: GLM-4.1V 916ms TTFT reduction via fused RMSNorm (#24733), GLM-4 MoE SharedFusedMoE optimization (#24849), Qwen2.5-VL CUDA sync removal (#24741), Qwen3-VL Triton MRoPE kernel (#25055), FP8 checkpoints for Qwen3-Next (#25079).
Reasoning: SeedOSS reason parser (#24263).

Engine Core

KV cache offloading: CPU offloading with LRU management (#19848, #20075, #21448, #22595, #24251).
V1 features: Prompt embeddings (#24278), sharded state loading (#25308), FlexAttention sliding window (#24089), LLM.apply_model (#18465).
Hybrid allocator: Pipeline parallel (#23974), varying hidden sizes (#25101).
Async scheduling: Uniprocessor executor support (#24219).
Architecture: Tokenizer group removal (#24078), shared memory multimodal caching (#20452).
Attention: Hybrid SSM/Attention in Triton (#21197), FlashAttention 3 for ViT (#24347).
Performance: FlashInfer RoPE 2x speedup (#21126), fused Q/K RoPE 11% improvement (#24511, #25005), 8x spec decode overhead reduction (#24986), FlashInfer spec decode with 1.14x speedup (#25196), model info caching (#23558), inputs_embeds copy avoidance (#25739).
LoRA: Optimized weight loading (#25403).
Defaults: CUDA graph mode FULL_AND_PIECEWISE (#25444), Inductor standalone compile disabled (#25391).
torch.compile: CUDA graph Inductor partition integration (#24281).

Hardware & Performance

NVIDIA: FP8 FlashInfer MLA decode (#24705), BF16 fused MoE for Hopper/Blackwell expert parallel (#25503).
DeepGEMM: Enabled by default (#24462), 5.5% throughput improvement (#24783).
New architectures: RISC-V 64-bit (#22112), ARM non-x86 CPU (#25166), ARM 4-bit fused MoE (#23809).
AMD: ROCm 7.0 (#25178), GLM-4.5 MI300X tuning (#25703).
Intel XPU: MoE DP accuracy fix (#25465).

Large Scale Serving & Performance

Dual-Batch Overlap (DBO): Overlapping computation mechanism (#23693), DeepEP high throughput + prefill (#24845).
Data Parallelism: torchrun launcher (#24899), Ray placement groups (#25026), Triton DP/EP kernels (#24588).
EPLB: Hunyuan V1 (#23078), Mixtral (#22842), static placement (#23745), reduced overhead (#24573).
Disaggregated serving: KV transfer metrics (#22188), NIXL MLA latent dimension (#25902).
MoE: Shared expert overlap optimization (#24254), SiLU kernel for DeepSeek-R1 (#24054), Enable Allgather/ReduceScatter backend for NaiveAllToAll (#23964).
Distributed: NCCL symmetric memory with 3-4% throughput improvement (#24532), enabled by default for TP (#25070).

Quantization

FP8: Per-token-group quantization (#24342), hardware-accelerated instructions (#24757), torch.compile KV cache (#22758), paged attention update (#22222).
FP4: NVFP4 for dense models (#25609), Gemma3 (#22771), Llama 3.1 405B (#25135).
W4A8: Faster preprocessing (#23972).
Compressed tensors: Blocked FP8 for MoE (#25219).

API & Frontend

OpenAI: Prompt logprobs for all tokens (#24956), logprobs=-1 for full vocab (#25031), reasoning streaming events (#24938), Responses API MCP tools (#24628, #24985), health 503 on dead engine (#24897).
Multimodal: Media UUID caching (#23950), image path format (#25081).
Tool calling: XML parser for Qwen3-Coder (#25028), Hermes-style tokens (#25281).
CLI: --enable-logging (#25610), improved --help (#24903).
Config: Speculative model engine args (#25250), env validation (#24761), NVTX profiling (#25501), guided decoding backward compatibility (#25615, #25422).
Metrics: V1 TPOT histogram (#24015), hidden deprecated gpu_ metrics (#24245), KV cache GiB units (#25204, #25479).
UX: Removed misleading quantization warning (#25012).

Security

https://github.com/vllm-project/vllm/security/advisories/GHSA-wr9h-g72x-mwhm

Dependencies

PyTorch 2.8 for CPU (#25652), FlashInfer 0.3.1 (#24470), CUDA 13 (#24599), ROCm 7.0 (#25178).
Build requirements: C++17 now enforced globally (#24823).
TPU: Deprecated xm.mark_step in favor of torch_xla.sync (#25254).

V0 Deprecation

Engines: AsyncLLMEngine (#25025), LLMEngine (#25033), MQLLMEngine (#25019), core (#25321), model runner (#25328), MP executor (#25329).
Components: Attention backends (#25351), encoder-decoder (#24907), output processor (#25320), sampling metadata (#25345), Sequence/Sampler (#25332).
Interfaces: LoRA (#25686), async output processor (#25334), MultiModalPlaceholderMap (#25366), seq group methods (#25330), placeholder attention (#25510), input embeddings (#25242), multimodal registry (#25362), max_seq_len_to_capture (#25543), attention classes (#25541), hybrid models (#25400), backend suffixes (#25489), compilation fallbacks (#25675), default args (#25409).

What's Changed

[Qwen3-Next] MoE configs for H20 TP=1,2,4,8 by @jeejeelee in https://github.com/vllm-project/vllm/pull/24707
[DOCs] Update ROCm installation docs section by @gshtras in https://github.com/vllm-project/vllm/pull/24691
Enable conversion of multimodal models to pooling tasks by @maxdebayser in https://github.com/vllm-project/vllm/pull/24451
Fix implementation divergence for BLOOM models between vLLM and HuggingFace when using prompt embeds by @qthequartermasterman in https://github.com/vllm-project/vllm/pull/24686
[Bugfix] Fix MRoPE dispatch on CPU by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/24712
[BugFix] Fix Qwen3-Next PP by @njhill in https://github.com/vllm-project/vllm/pull/24709
[CI] Fix flaky test v1/worker/test_gpu_model_runner.py::test_kv_cache_stride_order by @heheda12345 in https://github.com/vllm-project/vllm/pull/24640
[CI] Add ci_envs for convenient local testing by @noooop in https://github.com/vllm-project/vllm/pull/24630
[CI/Build] Skip prompt embeddings tests on V1-only CPU backend by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/24721
[Misc][gpt-oss] Add gpt-oss label to PRs that mention harmony or related to builtin tool call by @heheda12345 in https://github.com/vllm-project/vllm/pull/24717
[Bugfix] Fix BNB name match by @jeejeelee in https://github.com/vllm-project/vllm/pull/24735
[Kernel] [CPU] refactor cpu_attn.py:_run_sdpa_forward for better memory access by @ignaciosica in https://github.com/vllm-project/vllm/pull/24701
[sleep mode] save memory for on-the-fly quantization by @youkaichao in https://github.com/vllm-project/vllm/pull/24731
[Multi Modal] Add FA3 in VIT by @wwl2755 in https://github.com/vllm-project/vllm/pull/24347
[Multimodal] Remove legacy multimodal fields in favor of MultiModalFeatureSpec by @sfeng33 in https://github.com/vllm-project/vllm/pull/24548
[Doc]: fix typos in various files by @didier-durand in https://github.com/vllm-project/vllm/pull/24726
[Docs] Fix warnings in mkdocs build (continued) by @Zerohertz in https://github.com/vllm-project/vllm/pull/24740
[Bugfix] Fix MRoPE dispatch on XPU by @yma11 in https://github.com/vllm-project/vllm/pull/24724
[Qwen3-Next] MoE configs for H100 TP=1,2 and TP2/EP by @elvircrn in https://github.com/vllm-project/vllm/pull/24739
[Core] Shared memory based object store for Multimodal data caching and IPC by @dongluw in https://github.com/vllm-project/vllm/pull/20452
[Bugfix][Frontend] Fix --enable-log-outputs does not match the documentation by @kebe7jun in https://github.com/vllm-project/vllm/pull/24626
[Models] Optimise and simplify _validate_and_reshape_mm_tensor by @lgeiger in https://github.com/vllm-project/vllm/pull/24742
[Models] Prevent CUDA sync in Qwen2.5-VL by @lgeiger in https://github.com/vllm-project/vllm/pull/24741
[Model] Switch to Fused RMSNorm in GLM-4.1V model by @SamitHuang in https://github.com/vllm-project/vllm/pull/24733
[UX] Remove AsyncLLM torch profiler disabled log by @mgoin in https://github.com/vllm-project/vllm/pull/24609
[CI] Speed up model unit tests in CI by @afeldman-nm in https://github.com/vllm-project/vllm/pull/24253
[Bugfix] Fix incompatibility between #20452 and #24548 by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/24754
[CI] Trigger BC Linter when labels are added/removed by @zhewenl in https://github.com/vllm-project/vllm/pull/24767
[Benchmark] Allow arbitrary headers to be passed to benchmarked endpoints by @smarterclayton in https://github.com/vllm-project/vllm/pull/23937
[Compilation Bug] Fix Inductor Graph Output with Shape Issue by @yewentao256 in https://github.com/vllm-project/vllm/pull/24772
Invert pattern order to make sure that out_proj layers are identified by @anmarques in https://github.com/vllm-project/vllm/pull/24781
[Attention][FlashInfer] Enable FP8 FlashInfer (TRTLLM) MLA decode by @MatthewBonanni in https://github.com/vllm-project/vllm/pull/24705
Add FLASHINFER_MLA to backend selector test by @MatthewBonanni in https://github.com/vllm-project/vllm/pull/24753
[Qwen3Next] Fixes the cuda graph capture conditions under large batch sizes (#24660) by @sighingnow in https://github.com/vllm-project/vllm/pull/24667
[Core] Support async scheduling with uniproc executor by @njhill in https://github.com/vllm-project/vllm/pull/24219
[Frontend][Multimodal] Allow skipping media data when UUIDs are provided. by @huachenheli in https://github.com/vllm-project/vllm/pull/23950
[Model] Add Olmo3 model implementation by @2015aroras in https://github.com/vllm-project/vllm/pull/24534
[Bugfix] Fix GPUModelRunner has no attribute lora_manager by @jeejeelee in https://github.com/vllm-project/vllm/pull/24762
[Chore] Remove unused batched RoPE op & kernel by @WoosukKwon in https://github.com/vllm-project/vllm/pull/24789
[Docs] Fix warnings in mkdocs build (continued) by @Zerohertz in https://github.com/vllm-project/vllm/pull/24791
[Docs] Remove Neuron install doc as backend no longer exists by @hmellor in https://github.com/vllm-project/vllm/pull/24396
[Doc]: Remove 404 hyperlinks by @rozeappletree in https://github.com/vllm-project/vllm/pull/24785
[Perf] Use NVIDIA hardware-accelerated instruction for float to fp8_e4m3 quantization by @elvischenv in https://github.com/vllm-project/vllm/pull/24757
[Kernels][DP/EP] Optimize Silu Kernel for R1 by @elvircrn in https://github.com/vllm-project/vllm/pull/24054
[Core][Multimodal] Cache supports_kw by @lgeiger in https://github.com/vllm-project/vllm/pull/24773
[CI Failure] Fix test_flashinfer_cutlass_mxfp4_mxfp8_fused_moe by @mgoin in https://github.com/vllm-project/vllm/pull/24750
[Misc] Correct an outdated comment. by @russellb in https://github.com/vllm-project/vllm/pull/24765
[Doc]: fix typos in various files by @didier-durand in https://github.com/vllm-project/vllm/pull/24798
[CI][Spec Decode] Adjust threshold for flaky ngram spec decoding test again by @wwl2755 in https://github.com/vllm-project/vllm/pull/24771
Remove redundant assignment in xfer_buffers, This is a little fix by @ChenTaoyu-SJTU in https://github.com/vllm-project/vllm/pull/24732
[Minor] Simplify duplicative device check for cuda by @ziliangpeng in https://github.com/vllm-project/vllm/pull/24793
[Chore] Minor simplification for non-PP path by @WoosukKwon in https://github.com/vllm-project/vllm/pull/24810
[Multi Modal][Performance] Fused Q,K's apply_rope into one by @wwl2755 in https://github.com/vllm-project/vllm/pull/24511
[Misc] Improve s3_utils type hints with BaseClient by @Zerohertz in https://github.com/vllm-project/vllm/pull/24825
[Perf] Fix DeepGEMM Contiguous Layout Issue, 5.5% Throughput Improvement by @yewentao256 in https://github.com/vllm-project/vllm/pull/24783
fix type of sampling rate for encode_base64 by @co63oc in https://github.com/vllm-project/vllm/pull/24826
[Benchmarks] Throw usage error when using dataset-name random and dataset-path together by @yeqcharlotte in https://github.com/vllm-project/vllm/pull/24819
Force use C++17 globally to avoid compilation error by @chenfengjin in https://github.com/vllm-project/vllm/pull/24823
[Chore] Remove ipex_ops warning by @robertgshaw2-redhat in https://github.com/vllm-project/vllm/pull/24835
[Spec Decoding]Support Spec Decoding Metrics in DP Mode by @wuhang2014 in https://github.com/vllm-project/vllm/pull/24049
[Hybrid Allocator] Support Pipeline Parallel by @heheda12345 in https://github.com/vllm-project/vllm/pull/23974
[Docs] Have a try to improve frameworks/streamlit.md by @windsonsea in https://github.com/vllm-project/vllm/pull/24841
[kv cache] update num_free_blocks in the end by @andyxning in https://github.com/vllm-project/vllm/pull/24228
[Frontend] Skip stop in reasoning content by @gaocegege in https://github.com/vllm-project/vllm/pull/14550
[Bugfix] MiDashengLM model contact error under concurrent testing by @bingchen-mi in https://github.com/vllm-project/vllm/pull/24738
[Doc]: fix typos in various files by @didier-durand in https://github.com/vllm-project/vllm/pull/24821
[Misc] rename interval to max_recent_requests by @andyxning in https://github.com/vllm-project/vllm/pull/24229
[Misc] Own KVConnectors installation by @NickLucche in https://github.com/vllm-project/vllm/pull/24867
[P/D]kv_output_aggregator support heterogeneous by @LCAIZJ in https://github.com/vllm-project/vllm/pull/23917
[UT] enhance free kv cache block queue popleft_n by @andyxning in https://github.com/vllm-project/vllm/pull/24220
[XPU] Set consistent default KV cache layout by @NickLucche in https://github.com/vllm-project/vllm/pull/24745
[Misc] Fix examples openai_pooling_client.py by @noooop in https://github.com/vllm-project/vllm/pull/24853
[Model]: support Ling2.0 by @ant-yy in https://github.com/vllm-project/vllm/pull/24627
[Bugfix] Fix GLM4.1V multimodal processor with compatability for Transformers v4.56 by @Isotr0py in https://github.com/vllm-project/vllm/pull/24822
Fp8 paged attention update by @xiao-llm in https://github.com/vllm-project/vllm/pull/22222
Reinstate existing torch script by @hmellor in https://github.com/vllm-project/vllm/pull/24729
[USAGE] Improve error handling for weight initialization in Unquantized… by @koiker in https://github.com/vllm-project/vllm/pull/20321
Move MultiModalConfig from config/__init__.py to config/multimodal.py by @hmellor in https://github.com/vllm-project/vllm/pull/24659
[Transform] Deterministic Hadacore Transforms by @kylesayrs in https://github.com/vllm-project/vllm/pull/24106
Update num_tokens_across_dp to use nccl instead of gloo by @SageMoore in https://github.com/vllm-project/vllm/pull/24105
Bump Flashinfer to 0.3.1 by @bbartels in https://github.com/vllm-project/vllm/pull/24868
[gpt-oss] Add IncompleteDetails to ResponsesRepsonse by @qandrew in https://github.com/vllm-project/vllm/pull/24561
[gpt-oss][1a] create_responses stream outputs BaseModel type, api server is SSE still by @qandrew in https://github.com/vllm-project/vllm/pull/24759
[Performance] Remove redundant clone() calls in cutlass_mla by @alexm-redhat in https://github.com/vllm-project/vllm/pull/24891
[Bug] Fix Cutlass Scaled MM Compilation Error by @yewentao256 in https://github.com/vllm-project/vllm/pull/24887
[ci] fix wheel names for arm wheels by @simon-mo in https://github.com/vllm-project/vllm/pull/24898
[Tests] fix initialization of kv hash in tests by @mickaelseznec in https://github.com/vllm-project/vllm/pull/24273
[Compile] Fix noop_elimination pass and add tests for noop_elimination by @ZJY0516 in https://github.com/vllm-project/vllm/pull/24880
HuggingFace -> Hugging Face in Integration with Hugging Face docs by @sergiopaniego in https://github.com/vllm-project/vllm/pull/24889
Updated CODEOWNERS for flashinfer, mla, fused_moe by @mgoin in https://github.com/vllm-project/vllm/pull/24906
[Deprecation] Remove DeepGEMM Old Symbol Wrapper by @yewentao256 in https://github.com/vllm-project/vllm/pull/24902
[ROCm][Bugfix] Fix the case where there's bias by @gshtras in https://github.com/vllm-project/vllm/pull/24895
Add pytest-cov and .coveragerc by @rzabarazesh in https://github.com/vllm-project/vllm/pull/24778
[Bug] Fix is_flashmla_supported Check Error by @yewentao256 in https://github.com/vllm-project/vllm/pull/24774
[CI] Small Accuracy Eval Test for Deepseek Model by @yewentao256 in https://github.com/vllm-project/vllm/pull/24259
[Metrics] Hide deprecated metrics with gpu_ prefix by @markmc in https://github.com/vllm-project/vllm/pull/24245
[Docs] Update instructions for how to using existing torch binary by @zou3519 in https://github.com/vllm-project/vllm/pull/24892
Upgrade flashinfer to 0.3.1 by @houseroad in https://github.com/vllm-project/vllm/pull/24470
[XPU] Fix circular import error. by @jikunshang in https://github.com/vllm-project/vllm/pull/24927
Remove V0 Encoder-Decoder Support by @WoosukKwon in https://github.com/vllm-project/vllm/pull/24907
[Bugfix] Fix sequence parallelism bug when enable pipeline parallelism by @cascade812 in https://github.com/vllm-project/vllm/pull/24021
[Bug] [Spec Dec]: Fix kv_cache dtype mismatch for Eagle3 drafter on FP8 target by @vllmellm in https://github.com/vllm-project/vllm/pull/24505
[QWEN NEXT] Fused MoE kernels Optimization configs by @samanamp in https://github.com/vllm-project/vllm/pull/24924
[benchmark] Add triton version in the moe tuned config by @jeejeelee in https://github.com/vllm-project/vllm/pull/24769
[Bugfix] remove duplicate tokens streamed in required tool choice streaming by @Jason-CKY in https://github.com/vllm-project/vllm/pull/23312
[Mamba] Support TP>1 with quantization for mamba2 mixer in case n_groups % tp_size == 0 by @tomeras91 in https://github.com/vllm-project/vllm/pull/24593
[Feat][EPLB] A novel static EPLB placement strategy for MoE models. by @cboss6 in https://github.com/vllm-project/vllm/pull/23745
Move SpeculativeConfig from config/__init__.py to config/speculative.py by @hmellor in https://github.com/vllm-project/vllm/pull/24904
[Docs] move benchmarks README to contributing guides by @yeqcharlotte in https://github.com/vllm-project/vllm/pull/24820
feat: Add Grafana and Perces monitoring dashboards for vLLM by @liangwen12year in https://github.com/vllm-project/vllm/pull/23498
(doc): set cmake c++ compatible standard when building on MacOS CPU. by @teekenl in https://github.com/vllm-project/vllm/pull/23483
[CI] Add Decode Context Parallelism (DCP) test to CI by @minosfuture in https://github.com/vllm-project/vllm/pull/24487
[Model] Clean up and simplify Mamba2 Metadata Usage in both V0 and V1 by @cyang49 in https://github.com/vllm-project/vllm/pull/24331
[Core][MultiModalHasher] Don't convert memoryviews to bytes during hashing by @lgeiger in https://github.com/vllm-project/vllm/pull/24925
[Core/DBO][1/N] Add Dual-Batch Overlap mechanism to VLLM by @SageMoore in https://github.com/vllm-project/vllm/pull/23693
[Bugfix] Fix unable to run encoder model when disable_hybrid_kv_cache_manager is true by @lianyiibo in https://github.com/vllm-project/vllm/pull/24571
[Misc] Add removed encoder-decoder models to previously supported models list by @Isotr0py in https://github.com/vllm-project/vllm/pull/24961
Directly get max encoder len from VLLM config in V1 by @Sugar-zsg in https://github.com/vllm-project/vllm/pull/24866
[gpt-oss][1b] streaming add item id, content id by @qandrew in https://github.com/vllm-project/vllm/pull/24788
[MISC] Add code owners of vllm/v1 to vllm/v1/core by @heheda12345 in https://github.com/vllm-project/vllm/pull/24928
[ROCm] Add dependencies for ROCm by @Concurrensee in https://github.com/vllm-project/vllm/pull/24900
[gpt-oss][1][bugfix] fix streaming final output by @qandrew in https://github.com/vllm-project/vllm/pull/24466
Use kwargs for long lists of EngineCoreRequest arguments in tests and fix extra kwargs by @qthequartermasterman in https://github.com/vllm-project/vllm/pull/24987
fp8 kv cache support fix for torch.compile by @maleksan85 in https://github.com/vllm-project/vllm/pull/22758
[Perf] Reuse workspace for FP8+FP4 Marlin MoE by @mgoin in https://github.com/vllm-project/vllm/pull/20500
[CI][Bugfix] Fix failing Blackwell test by @MatthewBonanni in https://github.com/vllm-project/vllm/pull/24993
[CI] GPT-OSS GPQA eval test for Blackwell by @mgoin in https://github.com/vllm-project/vllm/pull/24920
[FP8] Extend per-token-group quantization support to QuantFP8 by @tahsintunan in https://github.com/vllm-project/vllm/pull/24342
Removes source compilation of nixl dependency by @bbartels in https://github.com/vllm-project/vllm/pull/24874
[Doc] Add --force-overwrite option to generate_cmake_presets.py by @elvischenv in https://github.com/vllm-project/vllm/pull/24375
[Core] Use CpuGpuBuffer for block table tensors by @njhill in https://github.com/vllm-project/vllm/pull/24795
[Benchmarks] Add MMVU video dataset support and clean up deprecated datasets by @Isotr0py in https://github.com/vllm-project/vllm/pull/24719
[UX] Enforce valid choices for envs like VLLM_ATTENTION_BACKEND, etc by @mgoin in https://github.com/vllm-project/vllm/pull/24761
[Docs] fix invalid doc link by @yyzxw in https://github.com/vllm-project/vllm/pull/25017
[UX] Remove "quantization is not fully optimized yet" log by @mgoin in https://github.com/vllm-project/vllm/pull/25012
[misc] fix typo in value error by @prashantgupta24 in https://github.com/vllm-project/vllm/pull/24995
[Core] Get num_encoder_tokens from scheduler config by @russellb in https://github.com/vllm-project/vllm/pull/24989
[V0 Deprecation] Remove MQLLMEngine by @WoosukKwon in https://github.com/vllm-project/vllm/pull/25019
[Model] Support Qwen3-VL Model Series by @ywang96 in https://github.com/vllm-project/vllm/pull/24727
[Rocm] [quantization] Fix quark ptpc moe and add test case by @haoyangli-amd in https://github.com/vllm-project/vllm/pull/24649
Add more documentation and improve usability of lognormal dist (benchmark_serving_multi_turn) by @pliops-daniels in https://github.com/vllm-project/vllm/pull/23255
[XPU] Fix xpu model runner call torch.cuda APIs by @jikunshang in https://github.com/vllm-project/vllm/pull/25011
[EPLB] Support EPLB for Mixtral Model by @rouchenzi in https://github.com/vllm-project/vllm/pull/22842
[Core][MultiModalHasher] Hash images without converting image mode by @lgeiger in https://github.com/vllm-project/vllm/pull/24969
[Model] Pass param prefix to LLMHead by @whx-sjtu in https://github.com/vllm-project/vllm/pull/24862
[Model] Apply SharedFusedMoE to glm4_moe. by @whx-sjtu in https://github.com/vllm-project/vllm/pull/24849
[Core] Remove tokenizer group in vLLM by @zhuohan123 in https://github.com/vllm-project/vllm/pull/24078
[Docs] Fix griffe warning in base_static_graph.py by @windsonsea in https://github.com/vllm-project/vllm/pull/25018
[DP] Create placement groups by ray_device_key by @xinyu-intel in https://github.com/vllm-project/vllm/pull/25026
[Frontend] Support returning all prompt logprobs by @chaunceyjiang in https://github.com/vllm-project/vllm/pull/24956
[BugFix] enable DOTALL to match multi-line tool_call parameters in extract_tool_call_required_streaming by @shijun-yin in https://github.com/vllm-project/vllm/pull/24668
[Misc] Avoid use of deprecated AutoModelForVision2Seq by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/25065
Add RADIO Vision Encoder Support to vLLM by @danielafrimi in https://github.com/vllm-project/vllm/pull/24595
[Bugfix] Fix Stream usage in CPU model runner and OneDNN kernel check by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/25046
Apply fixes for CUDA 13 by @Aidyn-A in https://github.com/vllm-project/vllm/pull/24599
[fix] lora benchmarks pass no_lora_flag_cpu by @dolpm in https://github.com/vllm-project/vllm/pull/23774
[Bugfix][Qwen3-Next] fixes the varlen issue in qwen3-next's MTP implementation. by @sighingnow in https://github.com/vllm-project/vllm/pull/24957
[Docs] improve code formatting and comments for eliminate griffe build warning. by @samzong in https://github.com/vllm-project/vllm/pull/25010
Remove old cutlass mla by @MatthewBonanni in https://github.com/vllm-project/vllm/pull/23961
[Docs] vllm/benchmarks/datasets.py fix docstring param format. by @samzong in https://github.com/vllm-project/vllm/pull/24970
[CI Bugfix] Fix failing test_invalid_env by @mgoin in https://github.com/vllm-project/vllm/pull/25078
[V0 Deprecation] Remove V0 Core tests by @WoosukKwon in https://github.com/vllm-project/vllm/pull/25082
cleanup: remove adapter commons by @simon-mo in https://github.com/vllm-project/vllm/pull/25045
Remove unused find_cuda_init helper script by @simon-mo in https://github.com/vllm-project/vllm/pull/25044
[V0 Deprecation] Remove unused output processor util by @WoosukKwon in https://github.com/vllm-project/vllm/pull/25023
Change log level from info to debug for IOProcessor by @mgoin in https://github.com/vllm-project/vllm/pull/24999
[CI] Revert back prepare_prompts and check_answers by @WoosukKwon in https://github.com/vllm-project/vllm/pull/25087
[V0 Deprecation] Remove V0 tests in test_sequence.py by @WoosukKwon in https://github.com/vllm-project/vllm/pull/25088
[CI Bugfix] Fix failing test_model_load_with_params tests due to tokenizer refactor by @mgoin in https://github.com/vllm-project/vllm/pull/25086
[V1] Logits processor docs by @afeldman-nm in https://github.com/vllm-project/vllm/pull/22919
[Misc] Update owners for KV connector and V1 offloading by @ApostaC in https://github.com/vllm-project/vllm/pull/25041
[Bugfix] Update import path for bc_linter_include by @mmangkad in https://github.com/vllm-project/vllm/pull/24766
[BUG] Exclude .pth file

These notes run past the length kept in the archive. The rest is on the publisher’s page.