Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Samples attention #14

Merged
merged 6 commits into from
Feb 19, 2025
Merged

Samples attention #14

merged 6 commits into from
Feb 19, 2025

Conversation

rockerBOO
Copy link

@rockerBOO rockerBOO commented Feb 18, 2025

  • Flash attention is a flag instead of using the dtype. Disabled for the moment
  • Sample images are now rendering but they produce only noise. Might help with debugging the model if they were actually working.
  • Fixed load from disk cache was not in the new order
  • Documented a lot of the functions

Attention processor with SDPA should be working on torch 2.6 but produced NaN on a lower torch when I was testing things. Maybe some torch version will get it working.

Sample images come in through CFG as 2 of the input latent and the positive and negative prompts.
[noise] * 2 paired with [postive_prompt, negative_prompt]

I rebased with SD3 i believe for the extra commits here.

Going to poke a little bit at fixing the sample images as that loop should solve some other issues. Then I would want to move to fixing batch size but should be fairly close with batch size.

@sdbds sdbds merged commit 6597631 into sdbds:lumina Feb 19, 2025
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants