Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Adding nnx Gemma2-2b (including overall fixes) to examples/gemma #4587

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
Open

Adding nnx Gemma2-2b (including overall fixes) to examples/gemma #4587

wants to merge 4 commits into from

Conversation

mdda
Copy link

@mdda mdda commented Feb 28, 2025

Adds Gemma2-2b (including GQA to Attention and fixes to Block)

Fixes #4567

It includes:

  • New TransformerConfig for gemma2-2b model
  • Renames existing configs to make them uniform
    • NB: This cannot affect existing code, since this module was previously unusable
  • Added additional params reading key adjustment (key in gemma2-2b needs remapping from kaggle download)
  • Adds GQA to Attention module
  • Reorders the operations in Block module so that logits output from overall Transformer are not gibberish
    • logits confirmed to (approximately) match those from GDE gemma (flax linen) model
  • No new documentation provided
    • This change would make the example in the nnx documentation work (did not work before)
  • No additional tests provided

@cgarciae
Copy link
Collaborator

cgarciae commented Mar 4, 2025

Hey Martin! Thanks for doing this.
Some folks are internally also improving the model. Let me merge their changes first and make sure there are no conflicts with those changes.

@mdda
Copy link
Author

mdda commented Mar 5, 2025 •
edited
Loading

Ahhh - Brings back memories of PRs for TensorFlow. Good times! /s

I'll attempt fix the white-space check failures, when you let me know that it isn't a waste of my time.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Some surprises when adding Gemma2-2b to flax/examples/gemma
2 participants
@mdda @cgarciae