[PyTorch]Add Casting-Free FP8-Flow-MoE Blockwise Optimizations#1

Closed

xiaoxi-wangfj wants to merge 3 commits into

FP8-FLow-Moe-MDfrom

Commits on Dec 22, 2025

[FP8-Flow-MoE] Fuse permute+pad and unpermute+unpad ops for FP8 optimization
xiaoxi-wangfj
committed

Commits on Dec 25, 2025

[FP8-Flow-MoE] fix dirty pad scale
xiaoxi-wangfj
committed

Commits on Dec 26, 2025

[PyTorch]: add moe fp8 flow under blockwise recipe
xiaoxi-wangfj
committed