Refactor(JAX): Use jax.jit's out_shardings instead of _enforce_jax_state_sharding #21559
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @buildwithsuhana, 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!
I've refactored the JAX trainer's sharding logic to streamline the code. This change moves the responsibility of enforcing JAX state sharding directly into the make_train_function, make_test_function, and make_predict_function methods, where jax.jit is invoked. By doing so, I've eliminated the need for the _enforce_jax_state_sharding method, making the sharding process more integrated and efficient. This refactor simplifies the overall architecture and improves how sharding is handled within the JAX backend.
Highlights
- Code simplification: The internal method
_enforce_jax_state_shardinghas been removed, simplifying the codebase. - Improved sharding integration: The sharding logic is now directly integrated into
make_train_function,make_test_function, andmake_predict_functionby utilizing theout_shardingsargument ofjax.jit. - New test coverage: A new test case,
JAXTrainerShardingTest, has been added to validate the updated sharding behavior within the JAX trainer. - Refactored JAX sharding: The JAX trainer's sharding mechanism has been refactored to leverage
jax.jit'sout_shardingsargument, replacing the previous_enforce_jax_state_shardingmethod.
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 or fill out our survey to provide feedback.
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
-
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. ↩
There was a problem hiding this comment.
Code Review
This pull request refactors the JAX trainer's sharding logic by replacing the _enforce_jax_state_sharding method with the out_shardings argument of jax.jit. This is a good simplification of the codebase. My review includes a suggestion to further refactor the new logic to reduce code duplication and a minor style fix in the new test file. Overall, the changes look good and are well-tested.
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #21559 +/- ##
==========================================
- Coverage 82.73% 82.72% -0.01%
==========================================
Files 567 567
Lines 56456 56454 -2
Branches 8823 8816 -7
==========================================
- Hits 46710 46704 -6
- Misses 7582 7587 +5
+ Partials 2164 2163 -1
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
| train_step = jit(self.train_step, donate_argnums=0) | ||
| out_shardings = None | ||
| if distribution_lib.distribution() is not None: | ||
| state_shardings = self._record_training_state_sharding_spec() |
There was a problem hiding this comment.
self._record_training_state_sharding_spec() does not return anything (and therefore returns None). So you're setting out_shardings to (None, None).
You do want to:
- rename
_record_training_state_sharding_specto something like_get_state_sharding_spec - have it not set
self.attributes - have it return the shardings instead
The worrying part is that no unit test failed. Can we have a unit test that fails when the shardings are not set correctly?
This is a refactor of the JAX trainer's sharding logic to simplify the code.
The logic previously handled by _enforce_jax_state_sharding has been moved into make_train_function, make_test_function, and make_predict_function. This is the ideal location, as it's where we call jax.jit. The now-redundant _enforce_jax_state_sharding method has been removed.