update Qwen3.5 grpo demo

[hjh0119]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello, 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 focuses on enhancing the Qwen3.5 GRPO demo by optimizing its training parameters and LoRA setup. It also significantly improves the robustness of the GSM8K dataset processing and reward calculation by introducing support for a new answer format. Furthermore, the changes streamline the internal weight management and synchronization processes across different model types, leading to a more efficient and maintainable codebase.

Highlights

• Qwen3.5 GRPO Demo Configuration: The Qwen3.5 GRPO demo has been updated with several configuration changes, including enabling Megatron by default, adjusting batch sizes, and introducing periodic model saving.
• LoRA Configuration Refinement: The LoRA configuration for Qwen3.5 now explicitly targets a comprehensive list of linear projection modules, providing more precise control over adapter application.
• GSM8K Dataset and Reward System Updates: The GSM8K dataset template's maximum length was reduced, and the preprocessor and reward system were enhanced to support a new '\boxed{}' answer format, improving flexibility in answer extraction and format validation.
• Weight Synchronization and Loading Logic Improvements: The weight synchronization mechanisms in both Megatron and Transformers models were refined for better handling of LoRA weights and base model states. Concurrently, the vLLM worker extension's weight loading logic was simplified by expecting pre-normalized weight names from the sender.

🧠 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.

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.

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 updates the GRPO demo for the Qwen3.5 model. The changes include updating default configurations, specifying LoRA target modules, adding checkpoint saving, and modifying the GSM8K prompt and reward logic. A significant and beneficial change is the refactoring of the weight synchronization mechanism, which moves weight name normalization from the sampler to the model side, simplifying the overall logic. I have one minor suggestion to improve code maintainability.