Upgrade vLLM to v0.11.2#4368

Closed

wangxiyuan wants to merge 40 commits intovllm-project:mainfrom

wangxiyuan:4142

Commits on Nov 24, 2025

fix break by vllm commit: Support LoRA with speculative decoding #21068

leo-pony
authored and
wangxiyuan
committed
[Hybrid] Pass kernel block size to builders #27753

leo-pony
authored and
wangxiyuan
committed
fix the main-to-main break by:[Bug] Fix env string 0 same to True #28159

leo-pony
authored and
wangxiyuan
committed
[Core] Async scheduling + structured outputs compatibility#26866

leo-pony
authored and
wangxiyuan
committed
fix structure output break bduring adapt to llm: Async scheduling + structured outputs compatibility#26866

leo-pony
authored and
wangxiyuan
committed
fix structured outputs compatibility

22dimensions
authored and
wangxiyuan
committed
fix mtp breaks in modelrunner and format fix

leo-pony
authored and
wangxiyuan
committed
fix mypy issues

leo-pony
authored and
wangxiyuan
committed
model runner execute model support v0.11.0 branch

leo-pony
authored and
wangxiyuan
committed
fix format issue

leo-pony
authored and
wangxiyuan
committed
update to releases/v0.11.1

22dimensions
authored and
wangxiyuan
committed
fix break by vllm:[BugFix][VL] Fix FA selection on Qwen2.5-VL #27790

leo-pony
authored and
wangxiyuan
committed
fix scheduler

22dimensions
authored and
wangxiyuan
committed
skip ut, nightly, v0.11.0

leo-pony
authored and
wangxiyuan
committed
Skip 1th e2e full

leo-pony
authored and
wangxiyuan
committed
skip has tested cases

leo-pony
authored and
wangxiyuan
committed
Fix vllm break:Support LoRA with speculative decoding:#21068

leo-pony
authored and
wangxiyuan
committed
remove skip of nightly a2

leo-pony
authored and
wangxiyuan
committed
fix the deepseek mtp break

leo-pony
authored and
wangxiyuan
committed
Add comments for deepseek torchair mtp break, vllm PR:27922

leo-pony
authored and
wangxiyuan
committed
Add comments fRestore full test cases

leo-pony
authored and
wangxiyuan
committed
Enable torchair test in single card full test

leo-pony
authored and
wangxiyuan
committed
just to trigger ci test

leo-pony
authored and
wangxiyuan
committed
fix vllm break by: Enable sequence parallelism matching w/o custom ops enabled #27126

leo-pony
authored and
wangxiyuan
committed
vllm break of PR:https://github.com/vllm-project/vllm/pull/24794

leo-pony
authored and
wangxiyuan
committed
adapt qwen3 next

22dimensions
authored and
wangxiyuan
committed
fix the undefine splitting_ops in test_sp_for_qwen3_moe

leo-pony
authored and
wangxiyuan
committed
make light test take effect

leo-pony
authored and
wangxiyuan
committed
fix vllm break: Fix backend selection for encoder-only models (#28534)

leo-pony
authored and
wangxiyuan
committed
fix vllm break:Refactor CUDA attention backend selection logic#24794

leo-pony
authored and
wangxiyuan
committed
fix break of vllm:Rename clashing method names for vLLM model protocol #27583

leo-pony
authored and
wangxiyuan
committed
fix lint error

leo-pony
authored and
wangxiyuan
committed
fix vllm break: Avoid bytecode hook and simplify TorchCompileWrapperWithCustomDipatch #25110

leo-pony
authored and
wangxiyuan
committed
skip A3-nightly test

leo-pony
authored and
wangxiyuan
committed
replace VisionPatchEmbed for better performance

shen-shanshan
authored and
wangxiyuan
committed
fix

shen-shanshan
authored and
wangxiyuan
committed
fix vllm break: [Model] Pass mm_features directly into get_mrope_input_positions#28399

leo-pony
authored and
wangxiyuan
committed
fix vllm break:[Misc] Make SchedulerConfig.max_model_len init-only #28733

leo-pony
authored and
wangxiyuan
committed
fix vllm break:[V1] Support MP Executor for multi node distributed inference#23691

leo-pony
authored and
wangxiyuan
committed
update version
wangxiyuan
committed