feat(moe): add remaining CUTLASS unified MoE runners - #4610

Merged

feih-nv merged 11 commits into

flashinfer-ai:mainfrom

feih-nv:feih/unified-moe-cutlass-runners

Aug 26, 2026

Commits on Aug 25, 2026

feat(moe): add remaining CUTLASS unified MoE runners
feih-nv
committed
dd6b977
refactor(moe): remove CutlassConfig placeholder
feih-nv
committed
7c5eb79
fix(moe): harden CUTLASS unified runners against review findings
feih-nv
committed
2cb364c
test(moe): require concrete CUTLASS autotune tactics
feih-nv
committed
8196308
fix(moe): reject unaligned MXFP4 and static per-tensor FP8 scales
feih-nv
committed
490bbae
fix(moe): enable CUTLASS MXFP8xMXFP8 on SM103/SM107
feih-nv
committed
7389ddd
docs(moe): clarify MXFP8 arch skips and CUDA 12.8 gate
feih-nv
committed
04bb287
fix(moe): align flat CUTLASS FP8 block CUDA 12.8 error text
feih-nv
committed
c92762b
fix(moe): require 128-aligned CUTLASS MXFP8xMXFP8 geometry
feih-nv
committed
dde1fba
docs(moe): show CutlassNvfp4Config in the unified API examples
feih-nv
committed
030146b
docs(moe): drop CUTLASS appendix from the MVP as-built walkthrough
feih-nv
committed
c85999c