Fix TF trainer bug when first input is None #21630
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 @neo-alex, 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 enhances the robustness of the TensorFlow trainer by addressing an edge case where optional None inputs could cause issues during the calculation of batch-related metrics. It ensures that the training and testing steps can correctly determine the batch size even when some inputs are not provided, leading to more stable model operations.
Highlights
- Tensorflow Trainer Robustness: Fixed a bug in the Tensorflow trainer where
model.fitandmodel.evaluatewould fail to correctly calculatesample_weightif the first flattened input wasNone. The fix ensures that the batch size is now extracted from the first available non-Noneinput.
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
-
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 PR fixes a bug where the Tensorflow trainer would crash if the first input was None. The change correctly finds the first non-None input to determine the batch size. My review focuses on improving the robustness of this fix by handling the edge case where all inputs might be None, which would currently cause an unhandled exception. I've also pointed out that the new logic is duplicated and could be refactored for better maintainability.
| sample_weight=tf.shape( | ||
| next(i for i in tree.flatten(x) if i is not None) | ||
| )[0], |
There was a problem hiding this comment.
This change correctly handles cases where the first input is None. However, it introduces a risk of a StopIteration error if all inputs in x are None. This can be difficult to debug, especially inside a tf.function.
A more robust approach would be to handle this edge case explicitly, for example by raising a ValueError with a clear message.
Also, this logic is duplicated in test_step. Consider extracting it into a private helper method to improve maintainability and ensure consistency.
| sample_weight=tf.shape( | ||
| next(i for i in tree.flatten(x) if i is not None) | ||
| )[0], |
There was a problem hiding this comment.
Similar to the train_step, this change is vulnerable to a StopIteration error if all inputs are None. Explicitly handling this edge case would make the code more robust and prevent potential runtime crashes.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #21630 +/- ##
=======================================
Coverage 82.49% 82.49%
=======================================
Files 572 572
Lines 57451 57451
Branches 8982 8982
=======================================
Hits 47395 47395
Misses 7760 7760
Partials 2296 2296
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:
|
|
Thanks for the fix! Could you please add a test for the same? It helps verify that the issue is actually fixed and we don't see regressions in the future. |
This PR fixes a Tensorflow trainer bug that arises when the first (flatten) input can be None in
model.fit/model.evaluate(which is possible for optional inputs since PR #21548). Note: this PR makes the original code more robust but still assumes that at least one input is not None (to properly extract the batch size).This fix was originally part of the bigger PR #21609 but is now pushed in a small dedicated PR as agreed with @hertschuh here.