[JAX] Fix breakage due to default dtype handling in ctc_loss.

[divyashreepathihalli]

from cl/794615827
"_Keras is the only user of a deprecated JAX api (jax_default_dtype_bits) for electing float32 default dtypes. This API will be removed in an upcoming JAX release.

A recent JAX change to default dtype handling if that flag is set broke a Keras test (cl/794553125). Since the behavior that Keras is relying on is deprecated it is probably best to update Keras not to depend on that default dtype handling in the first place in the test that broke (ctc_loss)._"

[gemini-code-assist]

Summary of Changes

Hello @divyashreepathihalli, 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 critical breakage in Keras's ctc_loss functionality when used with JAX. The issue arose because Keras was implicitly relying on a deprecated JAX API for default dtype handling, which was affected by a recent JAX update. The core change involves explicitly setting the data type for one-hot encoding operations within the ctc_loss implementation in Keras's JAX backend. By providing the dtype argument to jax.nn.one_hot, the code no longer depends on the problematic deprecated behavior, ensuring compatibility and stability with newer JAX versions.

Highlights

• Explicit Dtype Specification: The jax.nn.one_hot function calls within the ctc_loss implementation in keras/src/backend/jax/nn.py now explicitly pass the dtype argument.
• Addressing Deprecated JAX API Usage: This change ensures that the one-hot encoded tensors have the correct data type, preventing implicit reliance on JAX's default dtype handling, which was causing issues due to a deprecated API.
• Fixing Keras Test Breakage: The fix directly resolves a test breakage in Keras's ctc_loss that occurred after a recent JAX update related to default dtypes.

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 addresses a breakage in ctc_loss for the JAX backend, caused by a change in JAX's default dtype handling. The fix involves explicitly specifying the dtype in two jax.nn.one_hot calls within the ctc_loss function. This change is correct and makes the implementation more robust by removing the dependency on the deprecated default dtype behavior. The chosen dtypes match the other tensors used in the subsequent operations, ensuring correctness. The changes are well-targeted and I approve them.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.73%. Comparing base (45c98ec) to head (555269c).

Additional details and impacted files

@@           Coverage Diff           @@
##           master   #21588   +/-   ##
=======================================
  Coverage   82.73%   82.73%           
=======================================
  Files         567      567           
  Lines       56678    56678           
  Branches     8839     8839           
=======================================
  Hits        46895    46895           
  Misses       7609     7609           
  Partials     2174     2174           

Flag	Coverage Δ
keras	82.54% <100.00%> (ø)
keras-jax	63.64% <100.00%> (ø)
keras-numpy	58.22% <0.00%> (+<0.01%)	⬆️
keras-openvino	34.61% <0.00%> (ø)
keras-tensorflow	64.19% <0.00%> (ø)
keras-torch	63.80% <0.00%> (ø)

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.