[Example] Clip_B and Clip_V from entropy dynamics

[hiyuchang]

Description

We add support for algorithms in On the Entropy Dynamics in Reinforcement Fine-Tuning of Large Language Models. Contact: @shuminwang-ai.

• Clip_B
• Clip_V

Checklist

Please check the following items before code is ready to be reviewed.

• Code has passed all tests
• Docstrings have been added/updated in Google Style
• Documentation has been updated
• Code is ready for review

[gemini-code-assist]

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!

This pull request significantly enhances the framework's capabilities for fine-tuning Large Language Models using Reinforcement Learning by incorporating advanced entropy dynamics algorithms. It specifically implements the 'Clip_B' method, which refines token-level advantage calculations and introduces a new GRPO variant to manage these computations within the trainer. This integration allows for more nuanced control over the training process, aiming to improve model performance and stability.

Highlights

• New Algorithm Integration: Integrated the 'Clip_B' algorithm for entropy dynamics in Reinforcement Learning (RL) fine-tuning of Large Language Models (LLMs), based on recent research.
• GRPO Variant for Trainer-side Advantage: Introduced a new GRPOverlAlgorithm that delegates advantage computation to the trainer, enabling more flexible advantage function implementations like Clip_B.
• Example Configuration and Documentation: Provided a comprehensive example including a README and a YAML configuration file to demonstrate how to set up and run experiments with the new Clip_B algorithm.

Changelog

• examples/entropy/README.md
  • Added a new README file detailing the setup and execution of Clip_B experiments, including data preparation and a note on future Clip_V implementation.
• examples/entropy/clipb.yaml
  • Added a new YAML configuration file for the Clip_B example, specifying model, algorithm, buffer, explorer, and trainer settings for a math-related dataset.
• examples/entropy/clipb_trainer.patch
  • Added a patch file designed to modify trinity/trainer/verl_trainer.py to prevent the 'entropys' field from being removed from the batch, ensuring it's available for advantage functions like Clip_B.
• trinity/algorithm/init.py
  • Registered the new grpo_verl algorithm type, making it available for use within the system.
• trinity/algorithm/advantage_fn/init.py
  • Registered the new clipb advantage function, allowing it to be selected and used in algorithms.
• trinity/algorithm/advantage_fn/clipb_advantage.py
  • Added the ClipBAdvantageFn class, implementing the Clip_B advantage computation logic, which includes token filtering based on entropy signals and rewards.
• trinity/algorithm/algorithm.py
  • Defined the GRPOverlAlgorithm class, a GRPO variant configured to compute advantages within the trainer, supporting algorithms like Clip_B.
• trinity/common/verl_config.py
  • Added an optional loss_scale_factor field to the Actor configuration class.

Activity

• No specific activity (comments, reviews, or progress updates) has been recorded for this pull request yet.

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 introduces support for the Clip_B algorithm, including a new advantage function and an example configuration. The implementation is largely correct, but there are opportunities for improvement. Specifically, the logic for token filtering in the advantage function can be simplified for better readability. There's also an inconsistency in the standard deviation calculation that should be addressed. A significant maintainability concern is the use of a patch file for modifying core trainer behavior; this change should be integrated directly into the source code. Finally, there are minor issues in the example's README and configuration file that should be fixed.

[pan-x-c]

Suggested change

	from examples.entropy.clipv_dp_actor import DataParallelPPOActor
	from trinity.trainer.verl.dp_actor import DataParallelPPOActor

[pan-x-c]

use patch to implement this