Fix min-snr-gamma for v-prediction and ZSNR.#934
Merged
kohya-ss merged 1 commit intokohya-ss:devfrom Nov 25, 2023
Merged
Conversation
This fixes min-snr for vpred+zsnr by dividing directly by SNR+1. The old implementation did it in two steps: (min-snr/snr) * (snr/(snr+1)), which causes division by zero when combined with --zero_terminal_snr
Owner
|
Thank you for this! I didn't fully catch the discussion in the issue, but I believe this works fine. I will merge this sooner. Sorry for waiting a long time. |
nana0304
pushed a commit
to nana0304/sd-scripts
that referenced
this pull request
Jun 4, 2025
Fix min-snr-gamma for v-prediction and ZSNR.
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Based on the discussion in #673
Matches the reference implementation from the paper's author.
Note: With this PR, you should use
--min_snr_gammawithout--scale_v_pred_loss_like_noise_predThe old implementation with
apply_snr_weightandscale_v_prediction_loss_like_noise_predictionsplits the calculation into two steps:The problem here is that with
--zero_terminal_snrit is possible for the first step to divide by zero.This PR simplifies the weight calculation to
min(snr, gamma)/(snr+1)to avoid dividing by zero.This makes
scale_v_prediction_loss_like_noise_predictionunnecessary, but I did not remove it yet to avoid breaking existing workflows and GUIs. Maybe a deprecation notice should be added to it?