vllm
[Bugfix] Fix fused MoE IMA (sans chunking) by using int64 for strides
#34279

Merged

Commits

[Bugfix] Annotate fused MOE triton kernel strides as tl.int64

tlrmchlsmth committed 177 days ago

Loading