[Test] Make `test_failure` more stable for batch invariance (#27054)

Wentao Ye committed about 1 year ago

[V1][Spec Decode] Fix greedy temperature detection after sampler refactor (#27077)

Pradyun92 committed about 1 year ago

[Minor] Remove unnecessary error message (#27115)

Zhuohan Li committed about 1 year ago

[Bugfix] Use PIECEWISE cudagraphs on Blackwell if max_model_len > 131072 (#27114)

Michael Goin committed about 1 year ago

[Chore] Remove unused `PolyNorm` layer (#27110)

Isotr0py committed about 1 year ago

[Bugfix] [AITER] [ROCm] Fix Quark MoE Quant Config and AITER Fused MoE quant type logic (#27029)

vllmellm committed about 1 year ago

[ROCm][Bugfix][Model] Fix illegal memory access when running qwen3_moe models with rms_norm (Qwen3-235B-A22B, Qwen3-30B-A3B, etc.) (#26192)

rasmith committed about 1 year ago

[ROCM] MoE fp4 CK kernel (#26545)

Aleksandr Malyshev committed about 1 year ago

[CI] Remove forbidden slash (#27112)

Nicolò Lucchesi committed about 1 year ago

[bugfix] Qwen3-VL fix video incorrect timestamp calculations while do_sample_frames=True (#27104)

燃 committed about 1 year ago

[Data-parallel] Allow DP>1 for world_size > num_gpus on node (8) (#26367)

Patrick von Platen committed about 1 year ago

[CI] Nixl integration tests (#27010)

Nicolò Lucchesi committed about 1 year ago

[torch.compile] Enable attention and allreduce fusion without custom ops enabled (#24604)

Luka Govedič committed about 1 year ago

Fix incorrect docstring for stop_profile() method (#27101)

Yongtao Huang committed about 1 year ago

[Harware][AMD][Model] Triton MoE tuning configs for GLM-4.5 for MI350 and MI355 (#25586)

Reima Karhila (AMD) committed about 1 year ago

[Model]Improve Qwen3VLMoeForConditionalGeneration packed_modules_mapping (#27096)

Jee Jee Li committed about 1 year ago

[Docs] Replace `rst` style double-backtick with `md` single-backtick (#27091)

Harry Mellor committed about 1 year ago

[Docs] Replace all explicit anchors with real links (#27087)

Harry Mellor committed about 1 year ago

[VLM][Refactor] Remove useless func `get_input_positions` in `MRotaryEmbedding` (#27088)

Mengqing Cao committed about 1 year ago

Update troubleshooting.md and remind VLLM_TRACE_FUNCTION usage (#27069)

cong-meta committed about 1 year ago