Support Frequency Aware Dropout #2385
Open
gesen2egee wants to merge 1 commit into
Open
Conversation
gesen2egee
changed the title
gesen2egee
marked this pull request as ready for review
June 15, 2026 10:27
gesen2egee
marked this pull request as draft
June 15, 2026 14:39
gesen2egee
marked this pull request as ready for review
June 16, 2026 16:02
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
https://arxiv.org/abs/2603.27199
The previous branch added FAD options to the dataset config dataclasses, but the concrete DreamBooth, fine-tuning, and ControlNet dataset constructors did not accept those dataset-level parameters. As a result, normal dataset creation could fail before training started. The sFAD flag also defaulted to enabled through argparse, which made it difficult to opt out, and the step schedule used 0.1 to 0.8 instead of the paper's 0 to 1 scaling.
Wildcard frequency statistics were also estimated by random sampling, which made the per-subset token frequencies non-deterministic. Finally, FAD captions were still considered text-encoder-output cacheable, which could freeze a dynamic caption path.