gemma : more consistent attention scaling for v2 and v3 #13951
Merged
Conversation
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
ggerganov
changed the title
furyhawk
pushed a commit
to furyhawk/llama.cpp
that referenced
this pull request
Jun 6, 2025
* gemma : fix attn scale for 27B * cont : apply scale before attn * cont : consistent attention scaling
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
outdated
fix #12433 (comment)
I suspect the reference configs for Gemma 27B v2 and v3 are borked:
https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/config.py#L173
https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/config.py#L289
It does not make sense to normalize the Q tensor with
hidden_size / num_heads. It should be normalized withhead_size, like all other models.This change improves PPL and fixes the catastrophic generation at large contexts (see #12433 (comment))