[Performance] Reduce DeepGEMM N dim restriction from 128 to 64 multiplier (#28687)

Alexander Matveev committed 12 months ago

[Feature] EPLB on Qwen3VLMoe and CompressedTensorsWNA16MoEMethod (#28849)

JartX committed 12 months ago

[CI] Fix precommit `rope_theta` issue (#29040)

Wentao Ye committed 12 months ago

[Bugfix] Move flashinfer kernel check into ```__init__``` function of ```FusedMoE``` (#29018)

Max Hu committed 12 months ago

[BugFix] Fix false assertion with spec-decode=[2,4,..] and TP>2 (#29036)

Lucas Wilkinson committed 12 months ago

[ROCm][CI] Fix Weight Loading With Multiple GPU Tests on ROCm (#28984)

Micah Williamson committed 12 months ago

[DeepSeek] Fix DeepSeek V3.2 Rope Embedding (#28968)

Yongye Zhu committed 12 months ago

[MoE] Nvfp4 Masked Gemm: Add flashinfer grouped_gemm_nt_masked (#25990)

Shu Wang committed 12 months ago

[BugFix] Ray with multiple nodes (#28873)

Julien Denize committed 12 months ago

[Bug] Fix Batch Invariant MLA test (#28967)

Wentao Ye committed 12 months ago

[CI/Build] Fix test_prefix_prefill for AMD (#28905)

Ryan Rock committed 12 months ago

[Feature] Prefill Context Parallel (PCP) basic support (#28718)

Qiu committed 12 months ago

Eagle: MM Cuda Graphs with MRope (#28896)

Izzy Putterman committed 12 months ago

Upstreaming aiter triton attention backend as a new backend (#28701)

Aleksandr Malyshev committed 12 months ago

[Bugfix] Fix precision corruption when shared_experts_stream=None (#28942)

杰兮 committed 12 months ago

[Kernels] Improve H200 Fused MoE Config (#28992)

Robert Shaw committed 12 months ago

[Core] Reuse created spec tokens lists to mitigate GC cost (#28917)

Jialin Ouyang committed 12 months ago

Re-enable FlashInfer for Llama4 on Blackwell in e2e fusion tests (#28966)

Copilot committed 12 months ago

GLM-V video segmentation solution adjustment (#28941)

Yuxuan Zhang committed 12 months ago

Update `rope_scaling` to `rope_parameters` in preparation for Transformers v5 (#28542)

Harry Mellor committed 12 months ago