[WIP] feat: support moe hybrid parallel

[Prayer3th]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @Prayer3th, 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 Mixture-of-Experts (MoE) implementation by introducing support for hybrid parallelism, which combines tensor and expert parallelism. The changes involve a refactoring of the EPMoE layer to explicitly manage both parallelism dimensions, consolidating the MoE architecture by deprecating FusedMoE. Additionally, the weight sharding strategies have been updated, and internal logic has been simplified to accommodate this more advanced parallelism scheme, aiming to boost the efficiency and scalability of MoE models.

Highlights

• Refactoring of Expert Parallelism Parameter: The expert_parallel_size parameter has been consistently renamed to ep_size across the codebase, including in the EPMoE layer and its instantiations, for improved clarity and consistency.
• Introduction of Hybrid Parallelism Logic: The EPMoE layer now includes explicit calculations for tp_size (tensor parallelism size) derived from world_size and ep_size, indicating the integration of a hybrid parallelism strategy that combines tensor and expert parallelism.
• Unified MoE Implementation: The FusedMoE class has been removed from qwen3_moe.py, and the EPMoE class is now exclusively used for MoE layers, simplifying the architecture and consolidating all MoE parallelism logic into a single implementation.
• Updated Weight Sharding for Hybrid Parallelism: The partitioning specifications for expert weights (wi_0, wi_1, wo) have been revised in moe.py and qwen3_moe.py to align with the new hybrid parallelism strategy, utilizing `
• Simplified Expert Shard ID Calculation: The expert_shard_id calculation within the _forward method of EPMoE has been simplified, removing the dependency on data_index and tensor_size and now relying solely on tensor_index, which streamlines the expert dispatch process.

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. ↩