Comment by sorenjan

Thank you, I'm not used to reading this kind of research papers but I think I got the gist of it now.

Can this architecture be used to distill models that need fewer timesteps like LCMs or SDXL turbo?