[Feature] Support mixed precision

[qimcis]

Addresses #16276

Motivation

Configs using mixed-precision compressed tensors were failing because the code relied on the top-level format. This meant we skipped activation quantization and scheme selection when the global format was "mixed-precision," breaking setups like FP8 attention with INT4 MLPs.

Modifications

Updated parsing to track each group's specific format to determine if activation quantization is needed. The scheme selection and MoE validation now check the per-target format rather than the global one. I also added a regression test for a mixed FP8 + INT4 config to verify that activations are parsed only for the FP8 group and that the correct schemes are chosen.

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

[gemini-code-assist]

Summary of Changes

Hello @qimcis, 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 enhances the handling of mixed-precision compressed tensors by refining how quantization formats are interpreted. Previously, the system struggled with configurations like FP8 attention combined with INT4 MLPs because it used a global format, leading to incorrect activation quantization and scheme selection. The updated logic now processes quantization formats on a per-group basis, ensuring that each component of a mixed-precision setup receives the appropriate quantization treatment. This change improves the flexibility and correctness of mixed-precision model deployments.

Highlights

• Mixed Precision Support: The system now correctly handles mixed-precision compressed tensors by evaluating quantization formats on a per-group basis, rather than relying on a single global format. This resolves issues where activation quantization and scheme selection were incorrectly applied in mixed-precision setups.
• Per-Group Format Tracking: Parsing logic has been updated to track the specific quantization format for each configuration group. This allows for accurate determination of when activation quantization is needed and ensures the correct scheme is selected for each target.
• MoE Validation Update: The validation for Mixture-of-Experts (MoE) configurations now checks the per-target format, ensuring compatibility with mixed-precision settings.
• Regression Test Added: A new regression test has been introduced to verify the proper functioning of mixed FP8 + INT4 configurations, confirming that activations are parsed correctly for the FP8 group and appropriate schemes are chosen.

🧠 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

This pull request effectively adds support for mixed-precision quantization by tracking the format for each configuration group. The changes are logical and consistently applied across configuration parsing, scheme selection, and MoE layers. The addition of regression tests ensures the new functionality is working as expected. The code is well-written, and I have one suggestion to improve type safety.

[gemini-code-assist]

Using Dict[str, Any] reduces type safety and makes the code harder to understand and maintain. A more specific type hint would be better. Since the inner dictionary contains keys 'format' (str), 'weights' (QuantizationArgs), and 'input_activations' (Optional[QuantizationArgs]), consider defining a more precise type. For example, you could import Union and use Dict[str, Dict[str, Union[str, Optional[QuantizationArgs]]]], or define a TypedDict for better clarity.