[Misc] Remove unused attention prefix prefill ops functions (#26971)

Lukas Geiger committed about 1 year ago

[BugFix] Fix Siglip2Attention on XPU (#28448)

Fanli Lin committed about 1 year ago

[Kernel] Optimize rms_norm kernel (#27931)

Xin Yang committed about 1 year ago

Prefer FlashAttention MLA as default over FlashMLA (#27363)

Matthew Bonanni committed about 1 year ago

[CI/Build] Refactor Attention backend for test_prefix_prefill from xformers to SDPA (#28424)

Zhewen Li committed about 1 year ago

[Quantization] fix attention quantization of gpt_oss model (#27334)

xuebwang-amd committed about 1 year ago

[ROCm][Quantization] extend AMD Quark to support mixed-precision quantized model (#24239)

xuebwang-amd committed about 1 year ago

[Kernel][Perf] fuse QK Norm and RoPE into one cuda kernel for Qwen Model (#27165)

zhrrr committed about 1 year ago

[NIXL] Generalize block-first backend layouts (FlashInfer-like) (#28282)

Nicolò Lucchesi committed about 1 year ago

Remove weight_scale.T special case for SM90 Block FP8 CUTLASS kernel (#28431)

Michael Goin committed about 1 year ago

[Doc] Fix typo in serving docs (#28474)

the-codeboy committed about 1 year ago

[BugFix] Add test_outputs.py to CI pipeline (#28466)

usberkeley committed about 1 year ago

[BugFix] Fix RuntimeError in PixtralHFAttention on CPU/XPU (#28444)

Fanli Lin committed about 1 year ago

Add @markmc to CODEOWNERS for Observability (#28457)

Mark McLoughlin committed about 1 year ago

[BugFix] Fix Failing Ruff Check (#28469)

jvlunteren committed about 1 year ago

[Kernels] Split up fused_moe/layer.py, isolate more modular kernel code (#28064)

bnellnm committed about 1 year ago

[Docs] Fix grammar in CPU installation guide (#28461)

Maryam Tahhan committed about 1 year ago

Add request timeout override for multi-turn benchmarks (#28386)

Ido Segev committed about 1 year ago

[Model] Pass `mm_features` directly into `get_mrope_input_positions` (#28399)

Cyrus Leung committed about 1 year ago

Fix Fused MoE LoRA Triton kernel bug (#28450)

Chaojun Zhang committed about 1 year ago