Optimize RAM usage in generic exporter

[Bond099]

Description of the change

Reference

Colab Notebook

Checklist

• I have added all the necessary unit tests for my change.
• I have verified that my change does not break existing code and works with all backends (TensorFlow, JAX, and PyTorch).
• My PR is based on the latest changes of the main branch (if unsure, rebase the code).
• I have followed the Keras Hub Model contribution guidelines in making these changes.
• I have followed the Keras Hub API design guidelines in making these changes.
• I have signed the Contributor License Agreement.

[gemini-code-assist]

Summary of Changes

Hello @Bond099, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces significant optimizations to the generic model exporter, primarily focusing on reducing RAM consumption when converting Keras models to the Hugging Face safetensors format. By transitioning to a generator-based approach for weight mapping and implementing dynamic sharding, the exporter can now handle much larger models more efficiently, preventing out-of-memory errors during the export process. Additionally, it modularizes tensor transformations, making the export pipeline more robust and extensible for future model types.

Highlights

• Memory Optimization: Refactored the weight export process to use generators and dynamic sharding, significantly reducing RAM usage during the export of large models to the Hugging Face safetensors format.
• Modular Transformations: Introduced a dedicated transform_fn for model-specific weight transformations, improving code organization and deferring expensive operations until tensors are converted to NumPy.
• Dynamic Safetensors Sharding: Implemented logic to automatically shard safetensors files based on a max_shard_size parameter, enabling the export of very large models without encountering out-of-memory issues and generating an index file for sharded models.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces a significant optimization for exporting models by processing weights one by one using a generator, which greatly reduces RAM usage. The addition of model sharding is also a valuable feature for handling large models. The core implementation in hf_exporter.py is robust and well-designed. However, I've identified several critical issues in the refactored weight transformation logic for Gemma models within gemma.py. The current transformations for query, key, and value projection weights are incorrect and will lead to either errors or silently corrupted weights, particularly for models that use Grouped-Query Attention. My review includes specific code suggestions to correct these transformations.