Commits vllm-project/vllm

Address review: use registered envs.VLLM_DISABLE_PUSH_ALLREDUCE

Alexander Matveev committed 17 days ago

6d32045b

Address review: register VLLM_DISABLE_PUSH_ALLREDUCE in envs.py

Alexander Matveev committed 17 days ago

41f6a766

Address review: add architecture-specific threshold selection

Alexander Matveev committed 17 days ago

d35dfbb2

Address review: add type annotation and cleanup for push_ar_comm

Alexander Matveev committed 17 days ago

66eb0286

Address review: bind test sockets to localhost instead of all interfaces

Alexander Matveev committed 17 days ago

7b879d0a

Address review: add CUDA error checking and runtime buffer overflow guard

Alexander Matveev committed 17 days ago

4640e0d2

perf: add push-based allreduce for small tensor reductions

alexm-redhat committed 17 days ago

e3b4fdaf

[Frontend] Skip structural tags for auto tool_choice without strict mode (#45600)

sfeng33 committed 17 days ago

Verified cd9078fe

[Perf] Optimize DSv4 prefill chunk planning, 4.0% E2E Throughput Improvement (#45061)

yewentao256 committed 17 days ago

Verified e18fe932

[Bugfix] Chat Completions Harmony Refactor Clean up (#45464)

yzong-rh committed 17 days ago

Verified 51ec5cf0

[KV Offloading] Implement `reset_cache` for `TieringOffloadingManager` (#44541)

ronensc committed 17 days ago

Verified 7e612a0f

[Model] Add MiniMax M3 support (#45381)

youkaichao committed 17 days ago

Verified 0a1c5034

[AMD][Bugfix][Quantization] Honor fused-name match in is_layer_skipped (#43981)

ZiguanWang committed 17 days ago

Verified a3195fab

[Chore] Consolidate reasoning/tool parser attributes into unified Parser in chat serving (#45548)

sfeng33 committed 17 days ago

Verified 0d809796

[Bugfix] Two-phase KV allocation for cross-group prefix cache hits (supersedes #33775) (#44409)

Saddss committed 17 days ago

Verified 588db183

Remove redundant Triton KV cache dtype asserts and enforce architectural support (fp8 >= sm89) (#43914)

mikekg committed 17 days ago

Verified fa63bb9d

Fix the E8M0 scale computation in the MXFP4 (W4A4) MOE CUTLASS kernel (#43557)

xin3he committed 17 days ago

Verified 5ed15f42

(security) Enforce audio upload size limit before full file materialization (#45510)

jperezdealgaba committed 17 days ago

Verified b997071e

[Bugfix] Unset HF's default max_new_tokens for DiffusionGemma (#45417)

martin-kukla committed 17 days ago

Verified 6c5872ef

[Docs] Update the online serving docs. (#45676)

noooop committed 17 days ago

Verified 1d88c4da

[ROCm][Doc] Add installation notes about python version requirement (#45671)

vllmellm committed 17 days ago

Verified 25c53d12

[XPU] skip UT test_with_ngram_gpu_spec_decoding (#44423)

Yejing-Lai committed 17 days ago

Verified 9872921c

[Bugfix][Rust Frontend] Make metrics respect --served-model-name (#45465)

reidliu41 committed 17 days ago

Verified c17e2f7c

[Rust Frontend] Support `parallel_tool_calls = false` (#44760)

FAUST-BENCHOU committed 17 days ago

Verified 40eac9a9

[Models] Fix MiMo v2.x QKV TP sharding + FP4 support (#45200)

TheEpicDolphin committed 17 days ago

Verified b5adb027

[Rust Frontend] Add external→internal request-id map for abort() (#45137)

sahilsGit committed 17 days ago

Verified 64833f81

[Bugfix][Rust] Sync EngineCoreReadyResponse with the Python dataclass (#45557)

wseaton committed 17 days ago

Verified ddad5dbd

[Bugfix] Reject out-of-range temperature values in SamplingParams (#44965)

panpan0000 committed 17 days ago

Verified ebb0a71a

[Feature][Frontend] Report multimodal token counts in usage.prompt_tokens_details (#45458)

Sunt-ing committed 17 days ago

Verified 48df95c4

[Model] Remove XverseForCausalLM (#45638)

xianbaoqian committed 17 days ago

Verified 7df4fe1b