元記事の説明文
<h2>Highlights</h2>
<p>This release features 459 commits from 230 contributors (63 new)!</p>
<ul>
<li><strong>DeepSeek V4 maturity</strong>: DeepSeek V4 received a major hardening pass this cycle — the model was reorganized into a dedicated <code>vllm/models/deepseek_v4/</code> package (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43004">#43004</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43039">#43039</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43073">#43073</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43077">#43077</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43149">#43149</a>), gained NVFP4 fused MoE support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42209">#42209</a>), full + piecewise CUDA graph (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42604">#42604</a>), and MTP speculative decoding (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43385">#43385</a>). A large set of fused kernels (MegaMoE, <code>mhc</code>, Q-norm, indexer, sparse MLA) and ROCm parity fixes landed alongside accuracy fixes (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42810">#42810</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43710">#43710</a>).</li>
<li><strong>Model Runner V2 advances toward default</strong>: MRv2 is now default for Qwen3 dense models. vLLM will fall back to MRv1 for features that aren't yet supported in MRv2 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39337">#39337</a>). sleep-mode weight reload (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42673">#42673</a>), <code>update_config</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42783">#42783</a>), and shared KV-cache layers (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/35045">#35045</a>), plus many correctness fixes.</li>
<li><strong>Experimental Rust frontend</strong>: A new Rust front-end integration landed (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40848">#40848</a>), with the implementation moved into the tree (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43283">#43283</a>) and a DP Supervisor for data-parallel serving (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40841">#40841</a>).</li>
<li><strong>Batch invariance, faster</strong>: Batch-invariant inference gained Cutlass FP8 support for a <strong>28.9% end-to-end latency improvement</strong> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40408">#40408</a>), compile-mode support on SM80 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42456">#42456</a>), and an NVFP4 Cutlass linear path (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39912">#39912</a>).</li>
<li><strong>Multi-tier KV cache offloading</strong>: A new multi-tier KV cache offloading framework (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40020">#40020</a>) with a Python filesystem secondary tier (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41735">#41735</a>), DSv4 support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43142">#43142</a>), and Mooncake disk offloading (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42689">#42689</a>) extends offloading beyond CPU memory.</li>
</ul>
<h3>Model Support</h3>
<ul>
<li>New architectures: MiniCPM-V 4.6 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41254">#41254</a>), InternS2 Preview (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42705">#42705</a>), OpenVLA (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42654">#42654</a>), MolmoWeb <code>hf_overrides</code> docs (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42163">#42163</a>); EXAONE-4.5 aligned with Transformers update (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42246">#42246</a>).</li>
<li>Speculative decoding: custom callable proposer backend (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39487">#39487</a>), post-norm EAGLE-3 speculators (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42764">#42764</a>), peagle speculators (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41826">#41826</a>), hybrid-attention models in <code>extract_hidden_states</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39949">#39949</a>), non-MTP speculation for NemotronH (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43130">#43130</a>), shared MTP weights in MRv2 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42538">#42538</a>).</li>
<li>DeepSeek V4: NVFP4 MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42209">#42209</a>), CUDA graph full/piecewise (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42604">#42604</a>), MTP (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43385">#43385</a>), model package refactor (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43004">#43004</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43039">#43039</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43073">#43073</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43077">#43077</a>), sparse MLA + compressor refactor (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43149">#43149</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43710">#43710</a>), MegaMoE input-prep kernel move (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43632">#43632</a>).</li>
<li>Qwen3.5/3.6: GDN output-projection flatten (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42311">#42311</a>), GatedDeltaNet Marlin TP≥2 fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/36329">#36329</a>), ViT full CUDA graph (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42151">#42151</a>), runai-streamer weight loading for Qwen3.5/MTP/Qwen3-VL (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42521">#42521</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42716">#42716</a>), KDA chunk-prefill exp2 semantics (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43195">#43195</a>).</li>
<li>Gemma3/Gemma4: mixed-resolution image co-batching crash fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42217">#42217</a>), MoE routing closure fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42250">#42250</a>), tool-parser float-corruption fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42128">#42128</a>), batched vision encoder for image/video (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43169">#43169</a>), multi-GPU fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42630">#42630</a>).</li>
<li>Kimi-K2.5: skip vision-tower dtype conversion under quantization (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42869">#42869</a>), <code>mm_projector</code> dtype fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42081">#42081</a>).</li>
<li>Cohere: enable Cohere MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43143">#43143</a>), pipeline parallelism for Cohere vision (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42819">#42819</a>).</li>
<li>Tool calling: Apertus tool parser (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41154">#41154</a>), Qwen3Coder <code>anyOf</code>/<code>oneOf</code>/<code>$ref</code> resolution re-land (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/37831">#37831</a>), shared <code>coerce_to_schema_type</code> across MiniMax-M2 / DeepSeek-V3.2 / Seed-OSS parsers (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43006">#43006</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43019">#43019</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43140">#43140</a>).</li>
<li>ViT CUDA graph: Qwen2-VL (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41736">#41736</a>), Step3-VL encoder (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42224">#42224</a>), Qwen3.5 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42151">#42151</a>), FlashInfer metadata for Qwen2.5-VL vision attention (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42787">#42787</a>).</li>
</ul>
<h3>Engine Core</h3>
<ul>
<li>Model Runner V2: Qwen3-dense-by-default oracle (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39337">#39337</a>), sleep-mode reload weights (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42673">#42673</a>), <code>update_config</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42783">#42783</a>), shared KV-cache layers (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/35045">#35045</a>), FP32 gumbel sampling (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41775">#41775</a>), auto-fallback to MRv1 with connectors (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42955">#42955</a>), <code>logprob_token_ids</code> correctness (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43125">#43125</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41761">#41761</a>), prompt-logprobs size fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42778">#42778</a>).</li>
<li>KV offloading: multi-tier framework (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40020">#40020</a>), Python filesystem secondary tier (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41735">#41735</a>), DSv4 support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43142">#43142</a>), tier-offload follow-up (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42529">#42529</a>), prefer HND layout (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41928">#41928</a>), <code>reset_cache()</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41956">#41956</a>), per-request tracking (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42507">#42507</a>), store-deferral fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41945">#41945</a>).</li>
<li>MoE refactor: <code>ExpertMapManager</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41046">#41046</a>), experts moved to <code>experts/</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42334">#42334</a>), <code>RoutedExperts</code> alias for FusedMoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40735">#40735</a>), EPLB refactoring for FusedMoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41055">#41055</a>).</li>
<li>Mamba: attention module refactor (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41126">#41126</a>), Mamba2 SSD kernel warmup (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39822">#39822</a>), bf16 SSM cache (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41680">#41680</a>), GPU-side state postprocessing fused kernel (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40172">#40172</a>), run single-token extends as decodes (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42430">#42430</a>).</li>
<li>KV events: emit KV cache metadata (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40984">#40984</a>).</li>
<li>Allocator: manual cumem allocator enable (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/33648">#33648</a>), stream-aware free callback (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43020">#43020</a>).</li>
<li>elastic-EP: stage/commit MoE quant method on reconfigure (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40881">#40881</a>).</li>
</ul>
<h3>Hardware & Performance</h3>
<ul>
<li><strong>NVIDIA Blackwell / SM12x</strong>: FlashInfer b12x MoE + FP4 GEMM for SM120/121 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40082">#40082</a>), per-tensor FP8 CUTLASS on SM12.1 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41215">#41215</a>), <code>head_dim=512</code> for FlashInfer TRTLLM attention (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/38822">#38822</a>), FlashInfer Blackwell GDN prefill (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40717">#40717</a>), GDN prefill kernel for SM100 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43273">#43273</a>).</li>
<li><strong>Performance</strong>: batch-invariant Cutlass FP8 (+28.9% E2E) (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40408">#40408</a>), CutlassFP8 padding pre-processing (+13.5% TTFT) (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42651">#42651</a>), padded NVFP4 quant kernel (+2.4–5.7% E2E) (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42774">#42774</a>), GPU<->CPU sync elimination 1/n (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41429">#41429</a>) and 4/n (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42347">#42347</a>), fused RoPE+KVCache+q_concat for MLA (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40392">#40392</a>), MLA <code>compute_prefill_context</code> / <code>_v_up_proj</code> optimizations (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42460">#42460</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42561">#42561</a>), penalties Triton kernel (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40657">#40657</a>), <code>do_not_specialize</code> in fused FP8 RoPE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42849">#42849</a>), FULL CUDA graph capture for TRITON_MLA decode (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42885">#42885</a>).</li>
<li><strong>AMD ROCm</strong>: DSV4 functionality + accuracy fixes (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42810">#42810</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43679">#43679</a> Tilelang MHC), flash sparse MLA Triton kernels (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41812">#41812</a>), gluon paged MQA logits on gfx950/MI355X (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42062">#42062</a>), RMSNorm+Quant fusion for gfx950 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41825">#41825</a>), AITER FA backend cleanup (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41942">#41942</a>), XGMI backend for MoRI connector (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41753">#41753</a>), QuickReduce min-size override (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41675">#41675</a>), DSV4 MTP (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43385">#43385</a>).</li>
<li><strong>CPU / RISC-V</strong>: RVV-optimized attention kernels for RISC-V Vector Extension (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40119">#40119</a>) with VLEN=256 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42943">#42943</a>), fused GDN for AMX CPU (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42707">#42707</a>), MXFP4 W4A16 MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41922">#41922</a>), experimental Triton + MRv2 on CPU (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43225">#43225</a>), improved CPU thread utilization (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42666">#42666</a>), <code>--cpu-distributed-timeout-seconds</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42968">#42968</a>).</li>
<li><strong>Intel XPU</strong>: GPTQ int4 support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/37844">#37844</a>), mxfp8 MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41918">#41918</a>), FP8 block-scaled quantization (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42952">#42952</a>), custom-op collective behavior (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41354">#41354</a>), multiple sparse-attention kernels (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/37888">#37888</a>), MoE topk routing + MXFP4 fallback (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42951">#42951</a>), CT W4A4 MXFP4 path (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/38896">#38896</a>), reduced XPU MoE host overhead (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42915">#42915</a>).</li>
<li><strong>Kernel ABI</strong>: continued migration to libtorch stable ABI — 5/n (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42339">#42339</a>), 6/n (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42663">#42663</a>), 7/n (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43209">#43209</a>).</li>
<li><strong>Experimental</strong>: breakable CUDA graph (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42304">#42304</a>).</li>
</ul>
<h3>Large Scale Serving</h3>
<ul>
<li>Disaggregated serving (NIXL): lease-renewal TTL for KV blocks on P (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41383">#41383</a>), handshake-failure policy honoring (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40364">#40364</a>), GDN support for PD with NIXL (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41869">#41869</a>), multi-node TP>8 fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39907">#39907</a>), side-channel host-selection fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41806">#41806</a>).</li>
<li>Mooncake: disk offloading in MooncakeStoreConnector (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42689">#42689</a>), HMA support for DSV4 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42828">#42828</a>), operation metrics (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43392">#43392</a>), load-failure propagation (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42788">#42788</a>), block-aligned full hits (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43494">#43494</a>), finish-after-preemption handling (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43281">#43281</a>).</li>
<li>Data parallel: DP Supervisor (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40841">#40841</a>), publish request counts at engine-step start (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41626">#41626</a>), forward <code>X-data-parallel-rank</code> header (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42330">#42330</a>).</li>
<li>EPLB: change default EPLB communicator (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43110">#43110</a>), VLM-wrapper init fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39805">#39805</a>), remove dead <code>torch.accelerator.synchronize()</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40733">#40733</a>).</li>
<li>LoRA: one-shot Triton kernel for MoE LoRA (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42290">#42290</a>), simultaneous 2D & 3D MoE LoRA adapters (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42242">#42242</a>), reduced 2D-weight memory under EP (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42737">#42737</a>), MoE LoRA align-kernel grid fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40131">#40131</a>).</li>
</ul>
<h3>Quantization</h3>
<ul>
<li><strong>MXFP4</strong>: linear layers + compressed-tensors integration (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41664">#41664</a>), CPU W4A16 MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41922">#41922</a>), XPU mxfp8 MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41918">#41918</a>).</li>
<li><strong>NVFP4</strong>: DeepSeek V4 fused MoE (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42209">#42209</a>), ModelOpt W4A16 NVFP4 fused MoE + mixed-precision dispatch (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42566">#42566</a>), batch-invariant NVFP4 Cutlass linear (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39912">#39912</a>), FlashInfer TRTLLM NvFP4 monolithic MoE routing fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43223">#43223</a>), TRTLLM NVFP4 MoE chunking fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43599">#43599</a>).</li>
<li><strong>Quark</strong>: load Quark NVFP4 checkpoints (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/35859">#35859</a>), W8A8 INT8 garbage-output fix on Step-3.5-Flash (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41892">#41892</a>), W4A4 oracle refactor (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41436">#41436</a>).</li>
<li><strong>AutoRound</strong>: W4A16 support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39778">#39778</a>).</li>
<li><strong>ModelOpt</strong>: Qwen3.5/3.6 VLM quantized prefix mapping (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42546">#42546</a>).</li>
<li><strong>Framework</strong>: rework <code>quantization_config</code> to use <code>QuantKey</code> with activation override (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41566">#41566</a>), MoE W4A8 CT migrated to oracle (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42680">#42680</a>), AWQ Marlin MoE onto modular WNA16 oracle (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42483">#42483</a>), GPTQ consolidation (<code>gptq_marlin</code> → <code>auto_gptq</code>) (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/38288">#38288</a>).</li>
</ul>
<h3>API & Frontend</h3>
<ul>
<li><strong>Rust frontend</strong>: integration (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40848">#40848</a>), in-tree code move (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43283">#43283</a>), utility call-ID newtype (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43405">#43405</a>), simplified <code>AuthenticationMiddleware</code> path extraction (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43426">#43426</a>).</li>
<li><strong>Responses API</strong>: <code>chat_template_kwargs</code> support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42272">#42272</a>), message-merging fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42189">#42189</a>), empty channel/recipient harmony fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/35540">#35540</a>).</li>
<li><strong>Completions</strong>: <code>thinking_token_budget</code> support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42116">#42116</a>) with inverted-condition fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41674">#41674</a>); map <code>reasoning_effort</code> to <code>enable_thinking</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43401">#43401</a>).</li>
<li><strong>Frontend</strong>: truncation side for OpenAI endpoints (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43260">#43260</a>), normalize <code>reasoning_content</code> → <code>reasoning</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42664">#42664</a>), reworked fastokens integration (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43168">#43168</a>), consolidated Speech-to-Text entrypoints (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42370">#42370</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42274">#42274</a>), beam-search consolidation via <code>BeamSearchMixin</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42946">#42946</a>), score/rerank chat-template instructions (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42412">#42412</a>).</li>
<li><strong>Auth</strong>: API-key authorization for <code>/v2</code> endpoints (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42594">#42594</a>).</li>
<li><strong>Offline API</strong>: pooling offline API split into <code>PoolingOfflineMixin</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42267">#42267</a>), split offline inference APIs/utils (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43553">#43553</a>).</li>
</ul>
<h3>Build & Dependencies</h3>
<ul>
<li>CUDA 12.9 wheel builds switched to PyTorch <code>manylinux_2_28</code> base (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41668">#41668</a>).</li>
<li>FlashInfer bumped to v0.6.11.post2 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41711">#41711</a>); <code>nvidia-cutlass-dsl</code> to 4.5.2 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42991">#42991</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43230">#43230</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43745">#43745</a>); llguidance to 1.7 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42150">#42150</a>); <code>triton_kernels</code> downgraded to v3.5.1 for gpt-oss (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43135">#43135</a>).</li>
<li>Rust frontend build: <code>setuptools-rust</code> dependency (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43287">#43287</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43377">#43377</a>), pinned <code>protoc</code> in rust-build stages (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43292">#43292</a>).</li>
<li>Docker: non-root <code>vllm-openai</code> target (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40275">#40275</a>), build <code>mooncake-transfer-engine</code> from source (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42114">#42114</a>), AINIC & Thor NIC support (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40453">#40453</a>); Python-only installation made optional (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42293">#42293</a>).</li>
<li>vllm-tpu: disable build isolation for CUDA deps (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43038">#43038</a>), tpu-inference docker build fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43360">#43360</a>).</li>
<li><code>humming</code> MoE backend dependency added, reverted, then restored with CuPy runtime fix (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42540">#42540</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43492">#43492</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43530">#43530</a>).</li>
</ul>
<h3>Deprecations & Removals</h3>
<ul>
<li>Removed old locations of <code>get_tokenizer</code> and <code>resolve_hf_chat_template</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/35024">#35024</a>).</li>
<li>Marked env vars now covered by <code>--moe-backend</code> / <code>--linear-backend</code> (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43148">#43148</a>).</li>
<li>Removed deprecated MLA prefill arguments (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42555">#42555</a>).</li>
<li>Removed dead CUDA kernels and dead code (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42767">#42767</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42889">#42889</a>, <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43144">#43144</a>).</li>
</ul>
<h2>Contributors</h2>
<p><a class="user-mention notranslate" href="https://github.com/yewentao256">@yewentao256</a>, <a class="user-mention notranslate" href="https://github.com/haosdent">@haosdent</a>, <a class="user-mention notranslate" href="https://github.com/njhill">@njhill</a>, <a class="user-mention notranslate" href="https://github.com/mgoin">@mgoin</a>, <a class="user-mention notranslate" href="https://github.com/jeejeelee">@jeejeelee</a>, <a class="user-mention notranslate" href="https://github.com/AndreasKaratzas">@AndreasKaratzas</a>, <a class="user-mention notranslate" href="https://github.com/NickLucche">@NickLucche</a>, <a class="user-mention notranslate" href="https://github.com/sfeng33">@sfeng33</a>, <a class="user-mention notranslate" href="https://github.com/noooop">@noooop</a>, <a class="user-mention notranslate" href="https://github.com/WoosukKwon">@WoosukKwon</a>, <a class="user-mention notranslate" href="https://github.com/khluu">@khluu</a>, <a class="user-mention notranslate" href="https://github.com/taneem-ibrahim">@taneem-ibrahim</a>, <a class="user-mention notranslate" href="https://github.com/Dao007forever">@Dao007forever</a>, <a class="user-mention notranslate" href="https://github.com/vadiklyutiy">@vadiklyutiy</a>, <a class="user-mention notranslate" href="https://github.com/bnellnm">@bnellnm</a>, <a class="user-mention notranslate" href="https://github.com/ivanium">@ivanium</a>, <a class="user-mention notranslate" href="https://github.com/tjtanaa">@tjtanaa</a>, <a class="user-mention notranslate" href="https://github.com/mmangkad">@mmangkad</a>, <a class="user-mention notranslate" href="https://github.com/hmellor">@hmellor</a>, <a class="user-mention notranslate" href="https://github.com/DarkLight1337">@DarkLight1337</a>, <a class="user-mention notranslate" href="https://github.com/hickeyma">@hickeyma</a>, <a class="user-mention notranslate" href="https://github.com/zhenwei-intel">@zhenwei-intel</a>, <a class="user-mention notranslate" href="https://github.com/jikunshang">@jikunshang</a>, <a class="user-mention notranslate" href="https://github.com/ronensc">@ronensc</a>, <a class="user-mention notranslate" href="https://github.com/benchislett">@benchislett</a>, <a class="user-mention notranslate" href="https://github.com/hao-aaron">@hao-aaron</a>, <a class="user-mention notranslate" href="https://github.com/arpera">@arpera</a>, <a class="user-mention notranslate" href="https://github.com/zyongye">@zyongye</a>, <a class="user-mention notranslate" href="https://github.com/gau-nernst">@gau-nernst</a>, <a class="user-mention notranslate" href="https://github.com/frida-andersson">@frida-andersson</a>, <a class="user-mention notranslate" href="https://github.com/ZhanqiuHu">@ZhanqiuHu</a>, <a class="user-mention notranslate" href="https://github.com/cleonard530">@cleonard530</a>, <a class="user-mention notranslate" href="https://github.com/akii96">@akii96</a>, <a class="user-mention notranslate" href="https://github.com/bedeks">@bedeks</a>, <a class="user-mention notranslate" href="https://github.com/Isotr0py">@Isotr0py</a>, <a class="user-mention notranslate" href="https://github.com/JasonKeyiL">@JasonKeyiL</a>, <a class="user-mention notranslate" href="https://github.com/bigPYJ1151">@bigPYJ1151</a>, <a class="user-mention notranslate" href="https://github.com/zhewenl">@zhewenl</a>, <a class="user-mention notranslate" href="https://github.com/weizhoublue">@weizhoublue</a>, <a class="user-mention notranslate" href="https://github.com/zxd1997066">@zxd1997066</a>, <a class="user-mention notranslate" href="https://github.com/gnovack">@gnovack</a>, <a class="user-mention notranslate" href="https://github.com/chaojun-zhang">@chaojun-zhang</a>, <a class="user-mention notranslate" href="https://github.com/majian4work">@majian4work</a>, <a class="user-mention notranslate" href="https://github.com/chaunceyjiang">@chaunceyjiang</a>, <a class="user-mention notranslate" href="https://github.com/pschlan-amd">@pschlan-amd</a>, <a class="user-mention notranslate" href="https://github.com/amitz-nv">@amitz-nv</a>, <a class="user-mention notranslate" href="https://github.com/yma11">@yma11</a>, <a class="user-mention notranslate" href="https://github.com/dsikka">@dsikka</a>, <a class="user-mention notranslate" href="https://github.com/tc-mb">@tc-mb</a>, <a class="user-mention notranslate" href="https://github.com/shanjiaz">@shanjiaz</a>, <a class="user-mention notranslate" href="https://github.com/jperezdealgaba">@jperezdealgaba</a>, <a class="user-mention notranslate" href="https://github.com/yzong-rh">@yzong-rh</a>, <a class="user-mention notranslate" href="https://github.com/viktorpusTT">@viktorpusTT</a>, <a class="user-mention notranslate" href="https://github.com/TheEpicDolphin">@TheEpicDolphin</a>, <a class="user-mention notranslate" href="https://github.com/MatthewBonanni">@MatthewBonanni</a>, <a class="user-mention notranslate" href="https://github.com/shen-shanshan">@shen-shanshan</a>, <a class="user-mention notranslate" href="https://github.com/hallerite">@hallerite</a>, <a class="user-mention notranslate" href="https://github.com/zufangzhu">@zufangzhu</a>, <a class="user-mention notranslate" href="https://github.com/bbrowning">@bbrowning</a>, <a class="user-mention notranslate" href="https://github.com/divakar-amd">@divakar-amd</a>, <a class="user-mention notranslate" href="https://github.com/ianliuy">@ianliuy</a>, <a class="user-mention notranslate" href="https://github.com/esmeetu">@esmeetu</a>, <a class="user-mention notranslate" href="https://github.com/rasmith">@rasmith</a>, <a class="user-mention notranslate" href="https://github.com/louie-tsai">@louie-tsai</a>, <a class="user-mention notranslate" href="https://github.com/pmaybank">@pmaybank</a>, <a class="user-mention notranslate" href="https://github.com/liulanze">@liulanze</a>, <a class="user-mention notranslate" href="https://github.com/ZJY0516">@ZJY0516</a>, <a class="user-mention notranslate" href="https://github.com/TheDuyIT">@TheDuyIT</a>, <a class="user-mention notranslate" href="https://github.com/wzhao18">@wzhao18</a>, <a class="user-mention notranslate" href="https://github.com/jinzhen-lin">@jinzhen-lin</a>, <a class="user-mention notranslate" href="https://github.com/BugenZhao">@BugenZhao</a>, <a class="user-mention notranslate" href="https://github.com/ashwing">@ashwing</a>, <a class="user-mention notranslate" href="https://github.com/fuergaosi233">@fuergaosi233</a>, <a class="user-mention notranslate" href="https://github.com/hqhq1025">@hqhq1025</a>, <a class="user-mention notranslate" href="https://github.com/shaharmor98">@shaharmor98</a>, <a class="user-mention notranslate" href="https://github.com/pisceskkk">@pisceskkk</a>, <a class="user-mention notranslate" href="https://github.com/lkm2835">@lkm2835</a>, <a class="user-mention notranslate" href="https://github.com/noa-neria">@noa-neria</a>, <a class="user-mention notranslate" href="https://github.com/Rohan138">@Rohan138</a>, <a class="user-mention notranslate" href="https://github.com/whx-sjtu">@whx-sjtu</a>, <a class="user-mention notranslate" href="https://github.com/vrdn-23">@vrdn-23</a>, <a class="user-mention notranslate" href="https://github.com/alexagriffith">@alexagriffith</a>, <a class="user-mention notranslate" href="https://github.com/Flink-ddd">@Flink-ddd</a>, <a class="user-mention notranslate" href="https://github.com/jeffreywang-anyscale">@jeffreywang-anyscale</a>, <a class="user-mention notranslate" href="https://github.com/skyloevil">@skyloevil</a>, <a class="user-mention notranslate" href="https://github.com/ymoslem">@ymoslem</a>, <a class="user-mention notranslate" href="https://github.com/Lucaskabela">@Lucaskabela</a>, <a class="user-mention notranslate" href="https://github.com/kg6-sleipnir">@kg6-sleipnir</a>, <a class="user-mention notranslate" href="https://github.com/woernfl">@woernfl</a>, <a class="user-mention notranslate" href="https://github.com/tdoublep">@tdoublep</a>, <a class="user-mention notranslate" href="https://github.com/GOavi101">@GOavi101</a>, <a class="user-mention notranslate" href="https://github.com/jmamou">@jmamou</a>, <a class="user-mention notranslate" href="https://github.com/PeaBrane">@PeaBrane</a>, <a class="user-mention notranslate" href="https://github.com/KaivalyaMDabhadkar">@KaivalyaMDabhadkar</a>, <a class="user-mention notranslate" href="https://github.com/BWAAEEEK">@BWAAEEEK</a>, <a class="user-mention notranslate" href="https://github.com/MrZ20">@MrZ20</a>, <a class="user-mention notranslate" href="https://github.com/afierka-intel">@afierka-intel</a>, <a class="user-mention notranslate" href="https://github.com/JoursBleu">@JoursBleu</a>, <a class="user-mention notranslate" href="https://github.com/hissu-hyvarinen">@hissu-hyvarinen</a>, <a class="user-mention notranslate" href="https://github.com/mwawrzos">@mwawrzos</a>, <a class="user-mention notranslate" href="https://github.com/CynicDora">@CynicDora</a>, <a class="user-mention notranslate" href="https://github.com/NoeliaBentancor">@NoeliaBentancor</a>, <a class="user-mention notranslate" href="https://github.com/johncalesp">@johncalesp</a>, <a class="user-mention notranslate" href="https://github.com/fynnsu">@fynnsu</a>, <a class="user-mention notranslate" href="https://github.com/fxmarty-amd">@fxmarty-amd</a>, <a class="user-mention notranslate" href="https://github.com/walterbm">@walterbm</a>, <a class="user-mention notranslate" href="https://github.com/liangel-02">@liangel-02</a>, <a class="user-mention notranslate" href="https://github.com/lgeiger">@lgeiger</a>, <a class="user-mention notranslate" href="https://github.com/he-yufeng">@he-yufeng</a>, <a class="user-mention notranslate" href="https://github.com/abinggo">@abinggo</a>, <a class="user-mention notranslate" href="https://github.com/KrxGu">@KrxGu</a>, <a class="user-mention notranslate" href="https://github.com/hks-9697-v2">@hks-9697-v2</a>, <a class="user-mention notranslate" href="https://github.com/Sarah-Salah">@Sarah-Salah</a>, <a class="user-mention notranslate" href="https://github.com/rebklee">@rebklee</a>, <a class="user-mention notranslate" href="https://github.com/aoshen02">@aoshen02</a>, <a class="user-mention notranslate" href="https://github.com/haic0">@haic0</a>, <a class="user-mention notranslate" href="https://github.com/libinta">@libinta</a>, <a class="user-mention notranslate" href="https://github.com/Zhenzhong1">@Zhenzhong1</a>, <a class="user-mention notranslate" href="https://github.com/xhx1022">@xhx1022</a>, <a class="user-mention notranslate" href="https://github.com/b-mu">@b-mu</a>, <a class="user-mention notranslate" href="https://github.com/WindChimeRan">@WindChimeRan</a>, <a class="user-mention notranslate" href="https://github.com/tpopp">@tpopp</a>, <a class="user-mention notranslate" href="https://github.com/charlifu">@charlifu</a>, <a class="user-mention notranslate" href="https://github.com/chengyinie">@chengyinie</a>, <a class="user-mention notranslate" href="https://github.com/ricky-chaoju">@ricky-chaoju</a>, <a class="user-mention notranslate" href="https://github.com/lyd1992">@lyd1992</a>, <a class="user-mention notranslate" href="https://github.com/daniel-devlab">@daniel-devlab</a>, <a class="user-mention notranslate" href="https://github.com/paulyu12">@paulyu12</a>, <a class="user-mention notranslate" href="https://github.com/bobofang11235">@bobofang11235</a>, <a class="user-mention notranslate" href="https://github.com/laudney">@laudney</a>, <a class="user-mention notranslate" href="https://github.com/BadrBasowid">@BadrBasowid</a>, <a class="user-mention notranslate" href="https://github.com/maeehart">@maeehart</a>, <a class="user-mention notranslate" href="https://github.com/PatchouliTIS">@PatchouliTIS</a>, <a class="user-mention notranslate" href="https://github.com/chunxiaozheng">@chunxiaozheng</a>, <a class="user-mention notranslate" href="https://github.com/blake-snc">@blake-snc</a>, <a class="user-mention notranslate" href="https://github.com/southfreebird">@southfreebird</a>, <a class="user-mention notranslate" href="https://github.com/rbrugaro-amd">@rbrugaro-amd</a>, <a class="user-mention notranslate" href="https://github.com/rasdani">@rasdani</a>, <a class="user-mention notranslate" href="https://github.com/dusthunter">@dusthunter</a>, <a class="user-mention notranslate" href="https://github.com/qizzzh">@qizzzh</a>, <a class="user-mention notranslate" href="https://github.com/ProExpertProg">@ProExpertProg</a>, <a class="user-mention notranslate" href="https://github.com/qianlihuang">@qianlihuang</a>, <a class="user-mention notranslate" href="https://github.com/alec-flowers">@alec-flowers</a>, <a class="user-mention notranslate" href="https://github.com/JisoLya">@JisoLya</a>, <a class="user-mention notranslate" href="https://github.com/gaozihao-shy">@gaozihao-shy</a>, <a class="user-mention notranslate" href="https://github.com/rishaps">@rishaps</a>, <a class="user-mention notranslate" href="https://github.com/xyang16">@xyang16</a>, <a class="user-mention notranslate" href="https://github.com/wendyliu235">@wendyliu235</a>, <a class="user-mention notranslate" href="https://github.com/hlin99">@hlin99</a>, <a class="user-mention notranslate" href="https://github.com/tianmu-li">@tianmu-li</a>, <a class="user-mention notranslate" href="https://github.com/yuwenzho">@yuwenzho</a>, <a class="user-mention notranslate" href="https://github.com/inisis">@inisis</a>, <a class="user-mention notranslate" href="https://github.com/kfirtoledo">@kfirtoledo</a>, <a class="user-mention notranslate" href="https://github.com/roikoren755">@roikoren755</a>, <a class="user-mention notranslate" href="https://github.com/liranschour">@liranschour</a>, <a class="user-mention notranslate" href="https://github.com/vllm-agent">@vllm-agent</a>, <a class="user-mention notranslate" href="https://github.com/blancsw">@blancsw</a>, <a class="user-mention notranslate" href="https://github.com/netanel-haber">@netanel-haber</a>, <a class="user-mention notranslate" href="https://github.com/BowenBao">@BowenBao</a>, <a class="user-mention notranslate" href="https://github.com/czhu-cohere">@czhu-cohere</a>, <a class="user-mention notranslate" href="https://github.com/amitport">@amitport</a>, <a class="user-mention notranslate" href="https://github.com/tuukkjs">@tuukkjs</a>, <a class="user-mention notranslate" href="https://github.com/revit13">@revit13</a>, <a class="user-mention notranslate" href="https://github.com/ofirzaf">@ofirzaf</a>, <a class="user-mention notranslate" href="https://github.com/qyYue1389">@qyYue1389</a>, <a class="user-mention notranslate" href="https://github.com/junyanxu">@junyanxu</a>, <a class="user-mention notranslate" href="https://github.com/gracie-guo">@gracie-guo</a>, <a class="user-mention notranslate" href="https://github.com/sagearc">@sagearc</a>, <a class="user-mention notranslate" href="https://github.com/xinyu-intel">@xinyu-intel</a>, <a class="user-mention notranslate" href="https://github.com/yiwen101">@yiwen101</a>, <a class="user-mention notranslate" href="https://github.com/DomBrown">@DomBrown</a>, <a class="user-mention notranslate" href="https://github.com/tomeras91">@tomeras91</a>, <a class="user-mention notranslate" href="https://github.com/Dogacel">@Dogacel</a>, <a class="user-mention notranslate" href="https://github.com/maxdebayser">@maxdebayser</a>, <a class="user-mention notranslate" href="https://github.com/fadara01">@fadara01</a>, <a class="user-mention notranslate" href="https://github.com/Terrencezzj">@Terrencezzj</a>, <a class="user-mention notranslate" href="https://github.com/izikgo">@izikgo</a>, <a class="user-mention notranslate" href="https://github.com/wangrui6">@wangrui6</a>, <a class="user-mention notranslate" href="https://github.com/kebe7jun">@kebe7jun</a>, <a class="user-mention notranslate" href="https://github.com/rishitdholakia13">@rishitdholakia13</a>, <a class="user-mention notranslate" href="https://github.com/j9smith">@j9smith</a>, <a class="user-mention notranslate" href="https://github.com/meena-at-work">@meena-at-work</a>, <a class="user-mention notranslate" href="https://github.com/dllehr-amd">@dllehr-amd</a>, <a class="user-mention notranslate" href="https://github.com/alexeldeib">@alexeldeib</a>, <a class="user-mention notranslate" href="https://github.com/sonusflow">@sonusflow</a>, <a class="user-mention notranslate" href="https://github.com/lucianommartins">@lucianommartins</a>, <a class="user-mention notranslate" href="https://github.com/AAISSJ">@AAISSJ</a>, <a class="user-mention notranslate" href="https://github.com/DaoyuanLi2816">@DaoyuanLi2816</a>, <a class="user-mention notranslate" href="https://github.com/zexplorerhj">@zexplorerhj</a>, <a class="user-mention notranslate" href="https://github.com/zhangxin81">@zhangxin81</a>, <a class="user-mention notranslate" href="https://github.com/velonica0">@velonica0</a>, <a class="user-mention notranslate" href="https://github.com/fuscof-ibm">@fuscof-ibm</a>, <a class="user-mention notranslate" href="https://github.com/anishesg">@anishesg</a>, <a class="user-mention notranslate" href="https://github.com/zhengluo-nv">@zhengluo-nv</a>, <a class="user-mention notranslate" href="https://github.com/ylangtsou">@ylangtsou</a>, <a class="user-mention notranslate" href="https://github.com/fangyuchu">@fangyuchu</a>, <a class="user-mention notranslate" href="https://github.com/zx3xyy">@zx3xyy</a>, <a class="user-mention notranslate" href="https://github.com/simondanielsson">@simondanielsson</a>, <a class="user-mention notranslate" href="https://github.com/ruizhang99">@ruizhang99</a>, <a class="user-mention notranslate" href="https://github.com/zixi-qi">@zixi-qi</a>, @xwu-intel, @yufufi, @wdhongtw, @mrjunwan-lang, @wangxiyuan, @wasnertobias, @ilmarkov, @sychen52, @zhandaz, @russellb, @SandishKumarHN, @juhi10071998, @itayalroy, @djmmoss, @SumanthRH, @mayuyuace, @zhougit86, @meenchen, @lucifer1004, @popkart-EZ, @jzakrzew, @ffggs, @huanghua1994, @orozery, @danisereb, @rshavitt, @Yihuki, @QingZhou-YangHY, @Jie-Fang, @bbartels</p>
<h2>New Contributors</h2>
<ul>
<li><a class="user-mention notranslate" href="https://github.com/abinggo">@abinggo</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42128">#42128</a></li>
<li><a class="user-mention notranslate" href="https://github.com/afierka-intel">@afierka-intel</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40327">#40327</a></li>
<li><a class="user-mention notranslate" href="https://github.com/alexagriffith">@alexagriffith</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41987">#41987</a></li>
<li><a class="user-mention notranslate" href="https://github.com/alexeldeib">@alexeldeib</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43255">#43255</a></li>
<li><a class="user-mention notranslate" href="https://github.com/amitport">@amitport</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41666">#41666</a></li>
<li><a class="user-mention notranslate" href="https://github.com/anishesg">@anishesg</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43079">#43079</a></li>
<li><a class="user-mention notranslate" href="https://github.com/bedeks">@bedeks</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40269">#40269</a></li>
<li><a class="user-mention notranslate" href="https://github.com/blake-snc">@blake-snc</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/35568">#35568</a></li>
<li><a class="user-mention notranslate" href="https://github.com/blancsw">@blancsw</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41154">#41154</a></li>
<li><a class="user-mention notranslate" href="https://github.com/bobofang11235">@bobofang11235</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42604">#42604</a></li>
<li><a class="user-mention notranslate" href="https://github.com/BWAAEEEK">@BWAAEEEK</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42233">#42233</a></li>
<li><a class="user-mention notranslate" href="https://github.com/CynicDora">@CynicDora</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/39487">#39487</a></li>
<li><a class="user-mention notranslate" href="https://github.com/daniel-devlab">@daniel-devlab</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42479">#42479</a></li>
<li><a class="user-mention notranslate" href="https://github.com/DaoyuanLi2816">@DaoyuanLi2816</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42905">#42905</a></li>
<li><a class="user-mention notranslate" href="https://github.com/Dogacel">@Dogacel</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42764">#42764</a></li>
<li><a class="user-mention notranslate" href="https://github.com/DomBrown">@DomBrown</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42080">#42080</a></li>
<li><a class="user-mention notranslate" href="https://github.com/dusthunter">@dusthunter</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42594">#42594</a></li>
<li>@ffggs made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43414">#43414</a></li>
<li><a class="user-mention notranslate" href="https://github.com/frida-andersson">@frida-andersson</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41825">#41825</a></li>
<li><a class="user-mention notranslate" href="https://github.com/fuergaosi233">@fuergaosi233</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43488">#43488</a></li>
<li><a class="user-mention notranslate" href="https://github.com/gaozihao-shy">@gaozihao-shy</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42869">#42869</a></li>
<li><a class="user-mention notranslate" href="https://github.com/gracie-guo">@gracie-guo</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42626">#42626</a></li>
<li><a class="user-mention notranslate" href="https://github.com/haic0">@haic0</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40453">#40453</a></li>
<li><a class="user-mention notranslate" href="https://github.com/hks-9697-v2">@hks-9697-v2</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42521">#42521</a></li>
<li><a class="user-mention notranslate" href="https://github.com/hlin99">@hlin99</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42740">#42740</a></li>
<li><a class="user-mention notranslate" href="https://github.com/inisis">@inisis</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41710">#41710</a></li>
<li><a class="user-mention notranslate" href="https://github.com/izikgo">@izikgo</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42938">#42938</a></li>
<li><a class="user-mention notranslate" href="https://github.com/j9smith">@j9smith</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41215">#41215</a></li>
<li><a class="user-mention notranslate" href="https://github.com/junyanxu">@junyanxu</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42671">#42671</a></li>
<li><a class="user-mention notranslate" href="https://github.com/KaivalyaMDabhadkar">@KaivalyaMDabhadkar</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42333">#42333</a></li>
<li><a class="user-mention notranslate" href="https://github.com/libinta">@libinta</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41689">#41689</a></li>
<li>@lucifer1004 made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43433">#43433</a></li>
<li><a class="user-mention notranslate" href="https://github.com/meena-at-work">@meena-at-work</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40082">#40082</a></li>
<li>@mrjunwan-lang made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43360">#43360</a></li>
<li><a class="user-mention notranslate" href="https://github.com/MrZ20">@MrZ20</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42394">#42394</a></li>
<li><a class="user-mention notranslate" href="https://github.com/mwawrzos">@mwawrzos</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42498">#42498</a></li>
<li><a class="user-mention notranslate" href="https://github.com/NoeliaBentancor">@NoeliaBentancor</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42250">#42250</a></li>
<li>@ovidiusm made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42542">#42542</a></li>
<li><a class="user-mention notranslate" href="https://github.com/paulyu12">@paulyu12</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42306">#42306</a></li>
<li>@QingZhou-YangHY made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43579">#43579</a></li>
<li><a class="user-mention notranslate" href="https://github.com/qizzzh">@qizzzh</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/41680">#41680</a></li>
<li><a class="user-mention notranslate" href="https://github.com/qyYue1389">@qyYue1389</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42289">#42289</a></li>
<li><a class="user-mention notranslate" href="https://github.com/rasdani">@rasdani</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42481">#42481</a></li>
<li><a class="user-mention notranslate" href="https://github.com/rebklee">@rebklee</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42098">#42098</a></li>
<li><a class="user-mention notranslate" href="https://github.com/revit13">@revit13</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42926">#42926</a></li>
<li><a class="user-mention notranslate" href="https://github.com/ruizhang99">@ruizhang99</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43260">#43260</a></li>
<li><a class="user-mention notranslate" href="https://github.com/Sarah-Salah">@Sarah-Salah</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42441">#42441</a></li>
<li><a class="user-mention notranslate" href="https://github.com/sonusflow">@sonusflow</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/36329">#36329</a></li>
<li><a class="user-mention notranslate" href="https://github.com/TheDuyIT">@TheDuyIT</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40131">#40131</a></li>
<li><a class="user-mention notranslate" href="https://github.com/tuukkjs">@tuukkjs</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42880">#42880</a></li>
<li><a class="user-mention notranslate" href="https://github.com/vllm-agent">@vllm-agent</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42913">#42913</a></li>
<li><a class="user-mention notranslate" href="https://github.com/wangrui6">@wangrui6</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/40326">#40326</a></li>
<li>@wasnertobias made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43001">#43001</a></li>
<li><a class="user-mention notranslate" href="https://github.com/weizhoublue">@weizhoublue</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42830">#42830</a></li>
<li><a class="user-mention notranslate" href="https://github.com/woernfl">@woernfl</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42397">#42397</a></li>
<li>@xwu-intel made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/37888">#37888</a></li>
<li>@Yihuki made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42933">#42933</a></li>
<li><a class="user-mention notranslate" href="https://github.com/yiwen101">@yiwen101</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42654">#42654</a></li>
<li><a class="user-mention notranslate" href="https://github.com/ylangtsou">@ylangtsou</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43038">#43038</a></li>
<li>@yufufi made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42972">#42972</a></li>
<li><a class="user-mention notranslate" href="https://github.com/zhengluo-nv">@zhengluo-nv</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/43105">#43105</a></li>
<li>@zhougit86 made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42739">#42739</a></li>
<li><a class="user-mention notranslate" href="https://github.com/zx3xyy">@zx3xyy</a> made their first contribution in <a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/42855">#42855</a></li>
</ul>