Comment by bigglebear

I would guess a tiny stripped down text diffusion model. It only has 32k context, and for choice mode it can only select from 10 choices.