Rehy At Video Diffusion Attention

Merge softmax and linear attention for video diffusion models using chunk-wise recurrent reformulation with constant memory usage. Enable efficient distillation from existing softmax models, reducing training cost two orders of magnitude to ~160 GPU hours.

adu2021 Updated

File contents

adu2021/skillxiv/tree/main/skills/skillxiv-v0.0.2-claude-opus-4.6/rehy-at-video-diffusion-attention commit 06214a6d30

Frequently asked questions

npx skillmds@latest add adu2021/rehy-at-video-diffusion-attention