Hi, the image encoder is stored as bfloat16 but the config has it as float32. This causes an unnecessary and somewhat costly upcast in diffusers. Any chance you could have a look? Thanks in advance!
· Sign up or log in to comment
