Add Gemma3 bf16 related changes - #1441
Merged
Nenad Banfic (nenad1002) merged 4 commits intoMay 2, 2025
Merged
Conversation
Contributor
Author
|
Automatically merging— the final branch will be reviewed once all multimodal changes have been incorporated. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The Gemma3 model must compute layer norms and residual connections in FP32 to ensure accurate logits. Additionally, all logits are now explicitly cast to FP32. This PR removes the incorrect code paths and refactors the builder script so that it cleanly supports both Gemma and non-Gemma models.