[hotfix] fix mixtral with tensor-level compressed-tensor quantization #8721
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @ch-wan, 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 implemented a hotfix to address an issue with Mixtral's tensor-level compressed-tensor quantization. This change is crucial for ensuring the stability of our nightly tests, which were failing due to this specific problem. The modifications involve refining the handling of MoE layers within the quantization process to correctly interact with the FusedMoE implementation and ensure proper expert iteration.
Highlights
- Dependency Update: Imported
FusedMoEfromsglang.srt.layers.moe.fused_moe_triton, indicating a dependency on a specific fused MoE implementation. - Type Hint Refinement: Updated the type annotation for the
layerparameter inprocess_weights_after_loadingfrom a generictorch.nn.Moduleto the more specificFusedMoEclass, improving type safety and clarity. - Expert Count Property Correction: Corrected the property used to iterate over experts from
layer.local_num_expertstolayer.num_local_experts, resolving a potential issue with accessing the correct number of local 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 introduces a hotfix for Mixtral models using tensor-level compressed-tensor quantization. The changes correctly update the code to use the FusedMoE type hint and align with an attribute rename from local_num_experts to num_local_experts. The changes are accurate and address the issue. I have one suggestion to improve consistency in type hinting within the same file.
| @@ -189,7 +190,7 @@ def create_weights( | |||
| layer.w13_input_scale = None | |||
| layer.w2_input_scale = None | |||
|
|
|||
| def process_weights_after_loading(self, layer: torch.nn.Module) -> None: | |||
| def process_weights_after_loading(self, layer: FusedMoE) -> None: | |||
There was a problem hiding this comment.
Using the more specific FusedMoE type hint for the layer parameter is a good improvement for type safety and readability. For consistency, you might consider applying the same change to the process_weights_after_loading method in the CompressedTensorsWNA16MoEMethod class (line 470), which currently uses the generic torch.nn.Module.
Motivation
Fix our nightly test.
Modifications
Accuracy Test
Benchmark & Profiling
Checklist