[gkd] top-k-logits & teacher server

[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 introduces significant enhancements to the Generalized Knowledge Distillation (GKD) framework by optimizing resource utilization and increasing deployment flexibility. It enables GKD training to be more efficient by allowing the use of a subset of teacher model logits and by supporting external teacher model services, which is crucial for scaling up to very large teacher models and distributed training environments.

Highlights

• Top-K Logits for GKD: Introduced a 'Top-K' mode for Generalized Knowledge Distillation (GKD) to compute KL divergence using only the top-K most probable tokens from the teacher model. This significantly reduces memory usage and computational overhead, especially for models with large vocabularies.
• External Teacher Model API Support: Added functionality to use an external API service (e.g., vLLM) as the teacher model. This allows for decoupling the teacher model from the training process, enabling the use of larger teacher models or shared teacher services, and providing more flexible deployment options.
• Comprehensive Documentation and Examples: Updated both Chinese and English documentation with detailed explanations of the Top-K mode and external teacher API, including mathematical principles, usage examples, and notes on limitations. New example shell scripts are provided for both standard and Megatron-SWIFT GKD training with an external teacher server.
• Robust Argument Validation: Implemented new command-line arguments (--teacher_model_server, --gkd_logits_topk) and added validation logic to ensure correct configuration, such as requiring gkd_logits_topk when using an external server and preventing conflicts with other features like liger kernel.
• Dedicated Teacher API Client: Developed a new TeacherAPIClient to handle asynchronous and synchronous fetching of log probabilities from OpenAI-compatible API endpoints, complete with response parsing and server health checks.

🧠 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.

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

此拉取请求通过引入对外部教师模型 API 和 Top-K KL 散度计算模式的支持，显著增强了 GKD 训练算法。这些改进提供了更大的灵活性和内存效率，尤其适用于大型教师模型。所有相关文档（中文和英文）都已更新，并提供了新的示例脚本。实现包括对参数配置的严格验证以及一个专用的 API 客户端，该客户端具有全面的解析和分布式通信逻辑。此外，还包含了 API 客户端的单元测试和集成测试，以确保新功能的可靠性。