Allow passing variables to a function with @custom_gradient.

[hertschuh]

With the JAX backend, passing a variable to a function with a custom gradient would fail.

This adds a wrapper to unwrap variables.

Fixes #21105

[gemini-code-assist]

Summary of Changes

Hello @hertschuh, 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 addresses a compatibility issue within the JAX backend where KerasVariable objects were not correctly handled by functions utilizing @custom_gradient. By introducing a wrapper that unwraps these variables to their raw values, the system now ensures seamless integration and proper gradient computation in such scenarios. The accompanying test updates validate this fix, enhancing the robustness of custom gradient definitions in Keras with JAX.

Highlights

• JAX Custom Gradient Variable Handling: Modified the custom_gradient function in the JAX backend to automatically unwrap KerasVariable instances to their underlying .value before passing them to JAX's custom_gradient mechanism.
• Bug Fix: This change resolves an issue where passing a KerasVariable directly to a function decorated with @custom_gradient would cause a failure in the JAX backend.
• Test Coverage: Expanded the test_custom_gradient in keras/src/ops/core_test.py to include a parameterized test case that explicitly verifies the correct behavior when a KerasVariable is passed to a custom-gradient-decorated function.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

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 by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

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 pull request 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 aims to fix an issue with @custom_gradient on the JAX backend when keras.Variable instances are passed as arguments. The approach of wrapping the function to unwrap variables is sound, and the added tests effectively validate the fix for both tensor and variable inputs. However, there's a small but critical issue in the implementation of the wrapper function that needs to be addressed.

[gemini-code-assist]

Using tree.map_shape_structure here is incorrect. This function is designed to treat shape-like tuples as leaves and will attempt to traverse into KerasVariable objects since they are not shape-like tuples and are registered as PyTrees. This will apply the lambda function to the children of the KerasVariable instead of the variable itself, defeating the purpose of this wrapper.

You should use tree.map_structure instead. It will correctly treat KerasVariable instances as leaves and apply the unwrapping function to them.

Suggested change

	args, kwargs = tree.map_shape_structure(
	lambda x: x.value if isinstance(x, KerasVariable) else x,
	(args, kwargs),
	)
	args, kwargs = tree.map_structure(
	lambda x: x.value if isinstance(x, KerasVariable) else x,
	(args, kwargs),
	)

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.80%. Comparing base (8a37f9d) to head (8768da4).

Additional details and impacted files

@@           Coverage Diff           @@
##           master   #22069   +/-   ##
=======================================
  Coverage   82.80%   82.80%           
=======================================
  Files         592      592           
  Lines       62501    62505    +4     
  Branches     9789     9789           
=======================================
+ Hits        51751    51757    +6     
+ Misses       8215     8213    -2     
  Partials     2535     2535           

Flag	Coverage Δ
keras	82.63% <100.00%> (+<0.01%)	⬆️
keras-jax	62.38% <100.00%> (+<0.01%)	⬆️
keras-numpy	56.47% <0.00%> (-0.01%)	⬇️
keras-openvino	37.64% <0.00%> (+<0.01%)	⬆️
keras-tensorflow	63.63% <0.00%> (-0.01%)	⬇️
keras-torch	62.41% <0.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JyotinderSingh]

Thanks for the fix!