GLM-4.5 and GLM-4.5-Air both support#8804
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @zRzRzRzRzRzRzR, 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!
I've made changes to the determine_num_fused_shared_experts function within glm4_moe.py to enhance support for GLM-4.5 and GLM-4.5-Air models. The primary goal is to refine the logic for shared experts fusion optimization by adjusting the default architecture parameter and removing a specific constraint related to the number of routed experts. This streamlines the conditions under which shared experts fusion can be applied, improving compatibility and flexibility for these models.
Highlights
- Model Architecture Default: I've updated the default "architecture" parameter in the "determine_num_fused_shared_experts" function from "DeepseekV3ForCausalLM" to "Glm4MoeForCausalLM". This change aligns the default behavior with the GLM-4.5 model family.
- Shared Experts Fusion Logic: I've removed the condition "self.config.n_routed_experts != 128" from the checks that determine whether shared experts fusion should be disabled. This makes the fusion optimization more flexible and less restrictive regarding the number of routed experts.
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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| 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 issue 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 or fill out our survey to provide feedback.
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
-
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. ↩
There was a problem hiding this comment.
Code Review
This pull request updates the determine_num_fused_shared_experts method in glm4_moe.py to better support GLM-4.5 and GLM-4.5-Air models. The changes involve updating the default architecture name and removing a check on the number of routed experts to make the shared experts fusion optimization more general.
My review focuses on maintainability. While the changes are functionally correct, the determine_num_fused_shared_experts method is almost a complete duplicate of the one in its parent class, DeepseekV2ForCausalLM. I've suggested a refactoring to reduce this code duplication, which would make the code easier to maintain in the long run.
| def determine_num_fused_shared_experts( | ||
| self, architecture: str = "DeepseekV3ForCausalLM" | ||
| self, architecture: str = "Glm4MoeForCausalLM" | ||
| ): |
There was a problem hiding this comment.
This method is almost a complete duplicate of the parent class DeepseekV2ForCausalLM.determine_num_fused_shared_experts. This code duplication makes maintenance harder, as changes in one place might need to be manually propagated to the other.
To improve maintainability, consider refactoring to reduce the code duplication. One approach is to make the base class method more generic and parameterizable, so that this subclass can call it with its specific values, avoiding the need to override and duplicate the entire method body.
determine_num_fused_shared_experts changed