update grpo notebooks to have per_device_train_batch_size=num_generat… #130
Conversation
Summary of ChangesHello @mmathew23, 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 updates several GRPO notebooks to improve clarity and control over the training batch size. By explicitly setting the Highlights
Using Gemini Code AssistThe 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
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 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
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request makes a valuable improvement by explicitly setting per_device_train_batch_size to match num_generations across all GRPO notebooks and their corresponding Python scripts. This change enhances clarity for users, making the relationship between these two important hyperparameters obvious. The update has been applied consistently across a large number of files, which is great to see. I have one minor suggestion to further improve the clarity.
| optim = "adamw_8bit", | ||
| logging_steps = 1, | ||
| per_device_train_batch_size = 1, | ||
| per_device_train_batch_size = 4, |
There was a problem hiding this comment.
To further improve clarity and prevent potential user errors, consider adding a comment to explicitly state the relationship between per_device_train_batch_size and num_generations. This will help users who might modify num_generations to also update per_device_train_batch_size accordingly.
| per_device_train_batch_size = 4, | |
| per_device_train_batch_size = 4, # Should match num_generations |
| optim = "adamw_torch_fused", | ||
| logging_steps = 1, | ||
| per_device_train_batch_size = 1, | ||
| per_device_train_batch_size = 4, |
There was a problem hiding this comment.
To improve clarity and prevent potential user errors, it would be beneficial to add a comment indicating that per_device_train_batch_size should match num_generations. This makes the relationship between these parameters explicit for anyone modifying the script.
| per_device_train_batch_size = 4, | |
| per_device_train_batch_size = 4, # Should match num_generations |
| optim = "paged_adamw_8bit", | ||
| logging_steps = 1, | ||
| per_device_train_batch_size = 1, | ||
| per_device_train_batch_size = 6, |
There was a problem hiding this comment.
To improve clarity and prevent potential user errors, it would be beneficial to add a comment indicating that per_device_train_batch_size should match num_generations. This makes the relationship between these parameters explicit for anyone modifying the script.
| per_device_train_batch_size = 6, | |
| per_device_train_batch_size = 6, # Should match num_generations |
1 participant
instead of implicitly letting train batch size get set to num_generations behind the scenes, make it explicit and clear for users.