[Feat] Allow user to set train_batch_size
#177
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
There was a problem hiding this comment.
Summary of Changes
Hello @hiyuchang, 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 separated the batch_size into explore_batch_size and train_batch_size. This clarifies the different roles of batch sizes in the system, with explore_batch_size governing data collection and train_batch_size controlling the actual training process. This change also includes several bug fixes in the configuration management and updates across the codebase to ensure consistency and correctness.
Highlights
- Introduction of
train_batch_size: A new configuration parametertrain_batch_sizehas been added toBufferConfigto explicitly define the batch size used for model training. The existingbatch_sizenow specifically refers to the batch size for data exploration. - Unified Batch Size Logic: The internal
read_batch_sizehas been deprecated and replaced withtrain_batch_sizeacross various components, including buffer readers, sample strategies, and policy loss functions, streamlining how batch sizes are handled internally. - Improved Configuration Management: The configuration manager UI and backend logic have been updated to correctly present and process both
explore_batch_sizeandtrain_batch_size, enhancing user clarity and preventing common configuration errors. - Simplified PPO Mini-Batch Calculation: The calculation of
ppo_mini_batch_sizeno longer requires manual multiplication byrepeat_times, as it now directly leverages thetrain_batch_size, leading to more intuitive and less error-prone batch sizing for PPO algorithms. - Comprehensive Documentation and Example Updates: All relevant documentation, tutorial examples, and configuration files have been updated to reflect the new
train_batch_sizeparameter and the refined batch size semantics, ensuring users have accurate and up-to-date guidance.
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. ↩
Collaborator
Author
|
/unittest-all |
Contributor
There was a problem hiding this comment.
Code Review
This pull request introduces train_batch_size as a new configuration parameter to distinguish it from batch_size, which is now used for the number of tasks. The changes are consistently applied across the codebase, including documentation, examples, tests, and core logic. This is a good refactoring that improves clarity.
I've found a few issues:
- An error message in
trinity/common/verl_config.pywas not updated to use the new parameter name. - The
default_eval_workflow_typeconfiguration seems to have been unintentionally removed from the config manager UI (trinity/manager/config_manager.py), which prevents users from setting it.
Overall, the changes are well-implemented. Addressing the identified issues will further improve the quality of the code.
Collaborator
Author
|
/unittest-all |
pan-x-c
reviewed
Aug 11, 2025
Collaborator
Author
|
/unittest-all |
Summary
Failed Tests
Tests
Github Test Reporter by CTRF 💚 |
Collaborator
Author
|
/unittest-module-common |
Summary
Tests
Github Test Reporter by CTRF 💚 |
Collaborator
Author
|
/unittest-module-trainer |
Summary
Tests
Github Test Reporter by CTRF 💚 |
pan-x-c
approved these changes
Aug 12, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
As the title says.
Also fix some bugs in config manager.
Checklist
Please check the following items before code is ready to be reviewed.