v0.10.0rc2 | FList

What's Changed

[Model] use AutoWeightsLoader for bart by @calvin0327 in https://github.com/vllm-project/vllm/pull/18299
[Model] Support VLMs with transformers backend by @zucchini-nlp in https://github.com/vllm-project/vllm/pull/20543
[bugfix] fix syntax warning caused by backslash by @1195343015 in https://github.com/vllm-project/vllm/pull/21251
[CI] Cleanup modelscope version constraint in Dockerfile by @yankay in https://github.com/vllm-project/vllm/pull/21243
[Docs] Add RFC Meeting to Issue Template by @simon-mo in https://github.com/vllm-project/vllm/pull/21279
Add the instruction to run e2e validation manually before release by @huydhn in https://github.com/vllm-project/vllm/pull/21023
[Bugfix] Fix missing placeholder in logger debug by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/21280
[Model][1/N] Support multiple poolers at model level by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/21227
[Docs] Fix hardcoded links in docs by @hmellor in https://github.com/vllm-project/vllm/pull/21287
[Docs] Make tables more space efficient in supported_models.md by @hmellor in https://github.com/vllm-project/vllm/pull/21291
[Misc] unify variable for LLM instance by @andyxning in https://github.com/vllm-project/vllm/pull/20996
Add Nvidia ModelOpt config adaptation by @Edwardf0t1 in https://github.com/vllm-project/vllm/pull/19815
[Misc] Add sliding window to flashinfer test by @WoosukKwon in https://github.com/vllm-project/vllm/pull/21282
[CPU] Enable shared-memory based pipeline parallel for CPU backend by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/21289
[BugFix] make utils.current_stream thread-safety (#21252) by @simpx in https://github.com/vllm-project/vllm/pull/21253
[Misc] Add dummy maverick test by @minosfuture in https://github.com/vllm-project/vllm/pull/21199
[Attention] Clean up iRoPE in V1 by @LucasWilkinson in https://github.com/vllm-project/vllm/pull/21188
[DP] Fix Prometheus Logging by @robertgshaw2-redhat in https://github.com/vllm-project/vllm/pull/21257
Fix bad lm-eval fork by @mgoin in https://github.com/vllm-project/vllm/pull/21318
[perf] Speed up align sum kernels by @hj-mistral in https://github.com/vllm-project/vllm/pull/21079
[v1][sampler] Inplace logprobs comparison to get the token rank by @houseroad in https://github.com/vllm-project/vllm/pull/21283
[XPU] Enable external_launcher to serve as an executor via torchrun by @chaojun-zhang in https://github.com/vllm-project/vllm/pull/21021
[Doc] Fix CPU doc format by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/21316
[Intel GPU] Ray Compiled Graph avoid NCCL for Intel GPU by @ratnampa in https://github.com/vllm-project/vllm/pull/21338
Revert "[Performance] Performance improvements in non-blockwise fp8 CUTLASS MoE (#20762) by @minosfuture in https://github.com/vllm-project/vllm/pull/21334
[Core] Minimize number of dict lookup in _maybe_evict_cached_block by @Jialin in https://github.com/vllm-project/vllm/pull/21281
[V1] [Hybrid] Add new test to verify that hybrid views into KVCacheTensor are compatible by @tdoublep in https://github.com/vllm-project/vllm/pull/21300
[Refactor] Fix Compile Warning #1444-D by @yewentao256 in https://github.com/vllm-project/vllm/pull/21208
Fix kv_cache_dtype handling for out-of-tree HPU plugin by @kzawora-intel in https://github.com/vllm-project/vllm/pull/21302
[Misc] DeepEPHighThroughtput - Enable Inductor pass by @varun-sundar-rabindranath in https://github.com/vllm-project/vllm/pull/21311
[Bug] DeepGemm: Fix Cuda Init Error by @yewentao256 in https://github.com/vllm-project/vllm/pull/21312
Update fp4 quantize API by @wenscarl in https://github.com/vllm-project/vllm/pull/21327
[Feature][eplb] add verify ep or tp or dp by @lengrongfu in https://github.com/vllm-project/vllm/pull/21102
Add arcee model by @alyosha-swamy in https://github.com/vllm-project/vllm/pull/21296
[Bugfix] Fix eviction cached blocked logic by @simon-mo in https://github.com/vllm-project/vllm/pull/21357
[Misc] Remove deprecated args in v0.10 by @kebe7jun in https://github.com/vllm-project/vllm/pull/21349
[Core] Optimize update checks in LogitsProcessor by @Jialin in https://github.com/vllm-project/vllm/pull/21245
[benchmark] Port benchmark request sent optimization to benchmark_serving by @Jialin in https://github.com/vllm-project/vllm/pull/21209
[Core] Introduce popleft_n and append_n in FreeKVCacheBlockQueue to further optimize block_pool by @Jialin in https://github.com/vllm-project/vllm/pull/21222
[Misc] unify variable for LLM instance v2 by @andyxning in https://github.com/vllm-project/vllm/pull/21356
[perf] Add fused MLA QKV + strided layernorm by @mickaelseznec in https://github.com/vllm-project/vllm/pull/21116
[feat]: add SM100 support for cutlass FP8 groupGEMM by @djmmoss in https://github.com/vllm-project/vllm/pull/20447
[Perf] Cuda Kernel for Per Token Group Quant by @yewentao256 in https://github.com/vllm-project/vllm/pull/21083
Adds parallel model weight loading for runai_streamer by @bbartels in https://github.com/vllm-project/vllm/pull/21330
[feat] Enable mm caching for transformers backend by @zucchini-nlp in https://github.com/vllm-project/vllm/pull/21358
Revert "[Refactor] Fix Compile Warning #1444-D (#21208)" by @yewentao256 in https://github.com/vllm-project/vllm/pull/21384
Add tokenization_kwargs to encode for embedding model truncation by @Receiling in https://github.com/vllm-project/vllm/pull/21033
[Bugfix] Decode Tokenized IDs to Strings for hf_processor in llm.chat() with model_impl=transformers by @ariG23498 in https://github.com/vllm-project/vllm/pull/21353
[CI/Build] Fix test failure due to updated model repo by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/21375
Fix Flashinfer Allreduce+Norm enable disable calculation based on fi_allreduce_fusion_max_token_num by @xinli-git in https://github.com/vllm-project/vllm/pull/21325
[Model] Add Qwen3CoderToolParser by @ranpox in https://github.com/vllm-project/vllm/pull/21396
[Misc] Copy HF_TOKEN env var to Ray workers by @ruisearch42 in https://github.com/vllm-project/vllm/pull/21406
[BugFix] Fix ray import error mem cleanup bug by @joerunde in https://github.com/vllm-project/vllm/pull/21381
[CI/Build] Fix model executor tests by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/21387
[Bugfix][ROCm][Build] Fix build regression on ROCm by @gshtras in https://github.com/vllm-project/vllm/pull/21393
Simplify weight loading in Transformers backend by @hmellor in https://github.com/vllm-project/vllm/pull/21382
[BugFix] Update python to python3 calls for image; fix prefix & input calculations. by @ericehanley in https://github.com/vllm-project/vllm/pull/21391
[BUGFIX] deepseek-v2-lite failed due to fused_qkv_a_proj name update by @xuechendi in https://github.com/vllm-project/vllm/pull/21414
[Bugfix][CUDA] fixes CUDA FP8 kv cache dtype supported by @elvischenv in https://github.com/vllm-project/vllm/pull/21420
Changing "amdproduction" allocation. by @Alexei-V-Ivanov-AMD in https://github.com/vllm-project/vllm/pull/21409
[Bugfix] Fix nightly transformers CI failure by @Isotr0py in https://github.com/vllm-project/vllm/pull/21427
[Core] Add basic unit test for maybe_evict_cached_block by @Jialin in https://github.com/vllm-project/vllm/pull/21400
[Cleanup] Only log MoE DP setup warning if DP is enabled by @mgoin in https://github.com/vllm-project/vllm/pull/21315
add clear messages for deprecated models by @youkaichao in https://github.com/vllm-project/vllm/pull/21424
[Bugfix] ensure tool_choice is popped when tool_choice:null is passed in json payload by @gcalmettes in https://github.com/vllm-project/vllm/pull/19679
Fixed typo in profiling logs by @sergiopaniego in https://github.com/vllm-project/vllm/pull/21441
[Docs] Fix bullets and grammars in tool_calling.md by @windsonsea in https://github.com/vllm-project/vllm/pull/21440
[Sampler] Introduce logprobs mode for logging by @houseroad in https://github.com/vllm-project/vllm/pull/21398
Mamba V2 Test not Asserting Failures. by @fabianlim in https://github.com/vllm-project/vllm/pull/21379
[Misc] fixed nvfp4_moe test failures due to invalid kwargs by @chenyang78 in https://github.com/vllm-project/vllm/pull/21246
[Docs] Clean up v1/metrics.md by @windsonsea in https://github.com/vllm-project/vllm/pull/21449
[Model] add Hunyuan V1 Dense Model support. by @kzjeef in https://github.com/vllm-project/vllm/pull/21368
[V1] Check all pooling tasks during profiling by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/21299
[Bugfix][Qwen][DCA] fixes bug in dual-chunk-flash-attn backend for qwen 1m models. by @sighingnow in https://github.com/vllm-project/vllm/pull/21364
[Tests] Add tests for headless internal DP LB by @njhill in https://github.com/vllm-project/vllm/pull/21450
[Core][Model] PrithviMAE Enablement on vLLM v1 engine by @christian-pinto in https://github.com/vllm-project/vllm/pull/20577
Add test case for compiling multiple graphs by @sarckk in https://github.com/vllm-project/vllm/pull/21044
[TPU][TEST] Fix the downloading issue in TPU v1 test 11. by @QiliangCui in https://github.com/vllm-project/vllm/pull/21418
[Core] Add reload_weights RPC method by @22quinn in https://github.com/vllm-project/vllm/pull/20096
[V1] Fix local chunked attention always disabled by @sarckk in https://github.com/vllm-project/vllm/pull/21419
[V0 Deprecation] Remove Prompt Adapters by @mgoin in https://github.com/vllm-project/vllm/pull/20588
[Core] Freeze gc during cuda graph capture to speed up init by @mgoin in https://github.com/vllm-project/vllm/pull/21146
feat(gguf_loader): accept HF repo paths & URLs for GGUF by @hardikkgupta in https://github.com/vllm-project/vllm/pull/20793
[Frontend] Set MAX_AUDIO_CLIP_FILESIZE_MB via env var instead of hardcoding by @deven-labovitch in https://github.com/vllm-project/vllm/pull/21374
[Misc] Add dummy maverick test to CI by @minosfuture in https://github.com/vllm-project/vllm/pull/21324
[XPU][UT] increase intel xpu CI test scope by @Liangliang-Ma in https://github.com/vllm-project/vllm/pull/21492
[Bugfix] Fix casing warning by @MatthewBonanni in https://github.com/vllm-project/vllm/pull/21468
[Bugfix] Fix example disagg_example_p2p_nccl_xpyd.sh zombie process by @david6666666 in https://github.com/vllm-project/vllm/pull/21437
[BugFix]: Batch generation from prompt_embeds fails for long prompts by @KazusatoOoko in https://github.com/vllm-project/vllm/pull/21390
[BugFix] Fix KVConnector TP worker aggregation by @njhill in https://github.com/vllm-project/vllm/pull/21473
[DP] Internal Load Balancing Per Node [one-pod-per-node] by @robertgshaw2-redhat in https://github.com/vllm-project/vllm/pull/21238
Dump input metadata on crash for async scheduling by @WoosukKwon in https://github.com/vllm-project/vllm/pull/21258
[BugFix] Set CUDA_VISIBLE_DEVICES before spawning the subprocesses by @yinghai in https://github.com/vllm-project/vllm/pull/21211
Add think chunk by @juliendenize in https://github.com/vllm-project/vllm/pull/21333

New Contributors

@chaojun-zhang made their first contribution in https://github.com/vllm-project/vllm/pull/21021
@alyosha-swamy made their first contribution in https://github.com/vllm-project/vllm/pull/21296
@bbartels made their first contribution in https://github.com/vllm-project/vllm/pull/21330
@Receiling made their first contribution in https://github.com/vllm-project/vllm/pull/21033
@ariG23498 made their first contribution in https://github.com/vllm-project/vllm/pull/21353
@xinli-git made their first contribution in https://github.com/vllm-project/vllm/pull/21325
@ranpox made their first contribution in https://github.com/vllm-project/vllm/pull/21396
@ericehanley made their first contribution in https://github.com/vllm-project/vllm/pull/21391
@sergiopaniego made their first contribution in https://github.com/vllm-project/vllm/pull/21441
@hardikkgupta made their first contribution in https://github.com/vllm-project/vllm/pull/20793
@deven-labovitch made their first contribution in https://github.com/vllm-project/vllm/pull/21374
@MatthewBonanni made their first contribution in https://github.com/vllm-project/vllm/pull/21468
@david6666666 made their first contribution in https://github.com/vllm-project/vllm/pull/21437
@KazusatoOoko made their first contribution in https://github.com/vllm-project/vllm/pull/21390

Full Changelog: https://github.com/vllm-project/vllm/compare/v0.10.0rc1...v0.10.0rc2