[Perf] Move gc.freeze logic from EngineCoreProc to EngineCore for better coverage (#27896)

Jialin Ouyang committed about 1 year ago

[PERF] Allreduce fusion. Support torch native matching. Tuning of the thresholds (#24248)

Ilya Markov committed about 1 year ago

[ROCm] Add missing gemm_a8w8_blockscale import (#28378)

Yong Hoon Shin committed about 1 year ago

[CI] Fix Plugin Tests Tests (#28413)

Robert Shaw committed about 1 year ago

[Frontend][2/n] remove empty content from _parse_tool_calls_from_content (#28331)

Andrew Xia committed about 1 year ago

[BugFix] Fix DeepGEMM over-allocating workspace (#28254)

Lucas Wilkinson committed about 1 year ago

[Misc] Add more scoping for improved trace (#28329)

Wei Wei committed about 1 year ago

[Bugfix][EPLB] Disabled shared expert overlap when EPLB is enabled (#28377)

Sage Moore committed about 1 year ago

[FA/Chore] Bump FA version for FP8 two-level accumulation (#27889)

Jonas M. Kübler committed about 1 year ago

[Bugfix] Fix llguidance backend, rollback when EOS was encountered (#25905)

Rémi Delacourt committed about 1 year ago

[Performance] Support FP8 flashinfer TRTLLM MOE on Qwen3 and Qwen-3next (#27492)

jiahanc committed about 1 year ago

[Bugfix] Fix persistent_masked_m_silu_mul_quant tests (#28366)

Varun Sundar Rabindranath committed about 1 year ago

[V0 Deprecation] Remove unused `context_len` and `seq_len` from M-RoPE (#28395)

Cyrus Leung committed about 1 year ago

[RFC][ROCm][AITER] Keep all AITER kernels in `_aiter_ops` class like `_custom_ops` and `_ipex_ops` (#24490)

vllmellm committed about 1 year ago

[Kernel] Optimization of the mm_k operator. (#28280)

caozuoba committed about 1 year ago

[Rocm][fused_moe][fp4] view weight to torch.float4_e2m1fn_x2 when running aiter fused moe for fp4 model (#27474)

zejunchen-zejun committed about 1 year ago

Add @tjtanaa to codeowner for ROCm and multi-modal (#28360)

TJian committed about 1 year ago

[Fix] optimize visual token mask with caching and multi-token support (#28374)

Ferrebo committed about 1 year ago

[bugfix] fix siglip batch text output error (#28365)

Yu Jiaqi committed about 1 year ago

[Metrics] Refactor LoRA state tracking (#26801)

Mark McLoughlin committed about 1 year ago