David Marx

Nice! Looking forward to the change, fingers crossed users will be able to yield output dicts with scores and logits in addition to token_ids (i.e. the motivation for this PR)...

actually better yet, @gante could you maybe point me to the issue or PR I should follow to keep tabs on the "yield from generate" updates?

@gante refactoring messy research code is sorta my jam (just ask poli). Let me know if I could help accelerate some of this roadmap, would be happy to help with...

@gante @zucchini-nlp bump re the GenerationMixin.generate refactoring roadmap and design plans

@gante just checking in.

could you maybe add a separate optional requirements file for streamlit stuff? maybe something like requirements-app.txt, requirements-streamlit.txt, requirements-demo.txt, etc.

@rzzdr you're gonna have to ask someone with perms on the repo, I haven't been in stability's employ for almost a year now.

1. init image just used to initialize the generation process instead of noise. all steering achieved by text prompt. the init image only directly impacts the first frame of the...

so let's say you've got an init image and a text prompt the generation will load up that init image the same as if it were like a "previous frame"...

nb: includes adding MMC to install steps