songweige / rich-text-to-image

Rich-Text-to-Image Generation

Home Page:https://rich-text-to-image.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

UnboundLocalError: local variable 'attn_probs' referenced before assignment

howardgriffin opened this issue · comments

How to solve this problem?

Btw, I have installed xformers. When I setting this line https://github.com/SongweiGe/rich-text-to-image/blob/7780c3784e27aa803d6e5138807409e6d2973891/models/attention.py#L449 to False, everything goes OK

Thanks for reporting this. The xformer option was inherited from the diffusers codebase. The current implementation does not support xformer attention as we need to manipulate attention_probs.