[bugfix] fix mcore_bridge deepseek-v3

[Jintao-Huang]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @Jintao-Huang, 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 implements a crucial bugfix to enhance compatibility and correct behavior of the mcore_bridge with deepseek-v3 models. The change involves a precise adjustment to the tensor parallelism configuration, ensuring that a newly recognized projection layer (linear_q_up_proj) within these models is correctly handled for distributed processing. This modification is essential for the stable and efficient execution of deepseek-v3 models within the existing framework.

Highlights

• Bugfix for DeepSeek-V3: This pull request addresses a bug related to the mcore_bridge functionality when working with deepseek-v3 models, ensuring proper operation.
• Tensor Parallelism Configuration: The _get_tp_split_dim function in gpt_bridge.py has been updated to include linear_q_up_proj in the set of keys that determine tensor parallel split dimensions, specifically for MLA (Multi-Layer Attention) layers.

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 addresses a bug in the Megatron Core bridge for deepseek-v3 models. The change correctly adds linear_q_up_proj to the set of keys for layers that are split along dimension 0 for tensor parallelism. This ensures that the q_b_proj layer within the multi-latent attention mechanism is properly sharded as a ColumnLinear layer, which is crucial for correct tensor parallel execution. The fix is accurate and well-targeted.

While reviewing this function, I noticed a potential related issue. The linear_q_down_proj and linear_kv_down_proj layers, which are also part of the multi-latent attention mechanism, are not configured for tensor parallelism splitting in _get_tp_split_dim. If these are intended to be RowParallelLinear layers, their weights should be split along dimension 1. This would require adding their keys to a set of keys that are split on dimension 1. This is outside the scope of this PR, but it might be worth investigating in a follow-up to ensure full and efficient tensor parallelism support for this architecture.

Overall, the change in this PR is correct and I approve it.