[Bugfix] Drop stale layer kwarg from online MXFP4 kernel creation (fix precommit) (#51442)

Nick Hill committed 3 months ago

[CI] Fix Batch Invariance (B200) (#51417)

Jiangyun Zhu committed 3 months ago

[DSv32/GLM Perf] Skip short prefill topk for dense mha layer, 97.9% kernel level latency reduction (#51298)

Wentao Ye committed 3 months ago

[Profiler] Stamp vLLM version/commit into torch profiler trace metadata (#51389)

Elvir Crnčević committed 3 months ago

Add NVFP4 KV 4-over-6 scale search (#45187)

Wei-Ming Chen committed 3 months ago

[Bugfix] when loading weights skip empty expert bias if model does not support them (#50937)

Walter Beller-Morales committed 3 months ago

[Frontend] Disable uvicorn signal handlers instead of racing them (#50916)

Nick Hill committed 3 months ago

[Online quantization] Add online MXFP4 quantization support (#49347)

fxmarty-amd committed 3 months ago

[MRv2 Feature] MR v2 weight offloading support (#51413)

Wentao Ye committed 3 months ago

[Perf] Raise Blackwell CUDA graph capture default to 1024 (#49390)

Lucas Wilkinson committed 3 months ago

[K3 Perf] Optimize k3 dspark fused kv, 4.5~4.6x kernel performance improvement (#50585)

Wentao Ye committed 3 months ago

[Test] Add packed DeepSeek-V4 KV zeroer geometry regression (#51288)

coltonottley committed 3 months ago

[Test] Add ROCm AITER MLA op registration and env gating tests (#50930)

Aarushi Jain committed 3 months ago

[ROCm][CI][Bugfix] Do not microbatch a step that splits a prefix from its writer (#51402)

stefankoncarevic committed 3 months ago

[Attention] Mamba attention module refactor - Final part (#44857)

wangxiyuan committed 3 months ago

[1/N] Harden Transformers modelling backend multi-modal path (#51408)

Harry Mellor committed 3 months ago

[Refactor] refactor humming linear and moe backends to use explicit layer configs (#49610)

Jinzhen Lin committed 3 months ago

[ROCm] Add tuned selective_state_update float32 config for AMD Instinct MI325X (#50007)

vanshbhatia-amd committed 3 months ago

[Bugfix][ROCm] Fix ROCM_AITER_FA & ROCM_AITER_UNIFIED_ATTN QK-Norm+RoPE+KVCache fusion for the packed KV-cache [BLOCKS, HEADS, BLOCK_SIZE, 2*HEAD_DIM] layout (#49373)

Jack Hu committed 3 months ago

[ROCm] Enable pinned memory on supported WSL2 kernels (#50126)

Flora Cui committed 3 months ago