Hacker News Reader

I would be very greatfull to see how one can leverage it not for LLMs but for Stable Diffusion models

Why couldn't it be applied to SD?

That's Flash Attention 1, but yeah there's nothing stopping someone from submitting a PR.

This enables V1. V2 is still yet to be integrated into xformers. The team replied saying it should happen this week.

SD1 uses a head dimension that is not supported by the old version of Flash Attention but is supported by this version.