Skip to content

Conversation

@sayakpaul
Copy link
Member

What does this PR do?

Follow-up of #8528. Thanks to @Slickytail for contributing the functional weighting schemes generously.

@sayakpaul sayakpaul requested a review from kashif June 16, 2024 19:52
@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@sayakpaul
Copy link
Member Author

@kashif a gentle ping.

@kashif
Copy link
Contributor

kashif commented Jun 18, 2024

thanks! great refactor @sayakpaul

@sayakpaul sayakpaul merged commit 4edde13 into main Jun 18, 2024
@Nikolai10
Copy link

@sayakpaul, @kashif: issue #8592 sets the default value for weighting_scheme to logit_normal, see https://github.com/huggingface/diffusers/pull/8592/files.

This issue (#8591), however, overrides it back to sigma_sqrt -- by accident?

At least in my experiments sigma_sqrt often yields NaN values...

@sayakpaul sayakpaul deleted the refactor-sd3-training-tidbits branch June 18, 2024 17:28
@sayakpaul
Copy link
Member Author

My bad. Would you mind opening a PR to fix these?

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.

5 participants