Support Wan2.2-T2V-A14B (dual-expert MoE) GRPO training#8

Open

zhihengy wants to merge 13 commits into

mainfrom

Commits on Jun 16, 2026

diffusion reward: mean-pool PickScore over all generated video frames

zhihengy
and
claude
committed
diffusion(wan2.2): flow-shift override and configurable SDE step strategies

zhihengy
and
claude
committed