[misc] simplify megatron resample_data_iterator management

[hjh0119]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @hjh0119, 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 refactors the initialization and management of the resample_data_iterator across gkd_trainer.py and grpo_trainer.py. The changes streamline the data iterator setup by moving its initialization to the train method, eliminating lazy initialization checks in downstream methods, and removing an unnecessary _train_dataset attribute. This improves code clarity and consistency in how resampling data iterators are handled.

Highlights

• Eager Initialization: The resample_data_iterator is now initialized eagerly within the train method for both gkd_trainer.py and grpo_trainer.py, replacing previous lazy initialization patterns.
• Attribute Removal: The _train_dataset attribute has been removed from both trainers, as the train_dataset is now passed directly to the _init_resample_data_iterator method when needed.
• Code Simplification: Redundant lazy initialization checks for resample_data_iterator have been removed from methods like resample_encode_failed_inputs and _dynamic_sampling, simplifying the code flow.
• Iterator Return Value: The _init_resample_data_iterator method now directly returns the first element of the data iterator tuple, streamlining its usage.

🧠 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.

Changelog

• swift/megatron/trainers/gkd_trainer.py
  • Removed the _train_dataset attribute.
  • Initialized resample_data_iterator directly in the train method.
  • Modified _init_resample_data_iterator to accept train_dataset as an argument and return the first element of the iterator tuple.
  • Removed the lazy initialization check for resample_data_iterator from resample_encode_failed_inputs.
  • Updated the condition in _replace_data_iterator to check self.resample_data_iterator instead of self._train_dataset.
• swift/megatron/trainers/grpo_trainer.py
  • Removed the _train_dataset attribute.
  • Initialized resample_data_iterator directly in the train method.
  • Modified _init_resample_data_iterator to accept train_dataset as an argument, return the first element of the iterator tuple, and removed verbose comments.
  • Removed the lazy initialization check for resample_data_iterator from _dynamic_sampling.
  • Removed the lazy initialization check for resample_data_iterator from resample_encode_failed_inputs.

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 refactors the initialization of resample_data_iterator in both gkd_trainer.py and grpo_trainer.py. The previous lazy initialization logic is replaced with an eager initialization at the beginning of the train method. This simplifies the code by removing the _train_dataset attribute and making the data flow more explicit. The changes are consistent across both trainers and improve code clarity and maintainability. The bug where _init_resample_data_iterator returned a tuple, which was then indexed by the caller, has also been fixed by returning the correct iterator element directly. Overall, this is a good refactoring.