deleted
#24
by deleted - opened
deleted
I don't get that? If you first have to load the text encoder, then get vectors for your prompts and conditioning, you can unload the text encoders and the load the diffusion model.
So, the smaller the diffusion model, the more chance it has of fitting into ram?
deleted changed discussion title from Whats the point of these if you still have to use the full-sized text_encoder? to deleted