Add support for global scale to scaled_matmul_wrapper

[sergey-kozub]

Fuse multiplication by the global scale into the matmul kernel, this results in significant performance improvement.

Note: float16 dtype is disabled in the test as it has insufficient precision.

[gemini-code-assist]

Summary of Changes

Hello @sergey-kozub, 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 introduces support for a global scaling factor within the scaled_matmul_wrapper and its associated internal functions. The primary goal is to enhance performance by fusing the global scale multiplication directly into the matrix multiplication kernel, thereby streamlining the computation. The changes propagate through the internal JAX primitive definitions, including lowering to CUDA, batching, and abstract interpretation, and are exposed to the user through an updated public API. Additionally, specific tests have been adjusted to account for precision limitations with float16.

Highlights

• Global Scale Integration: The scaled_matmul_wrapper and its underlying primitive now natively support a global_scale parameter, allowing for direct integration of a global scaling factor into the matrix multiplication operation.
• Performance Optimization: Multiplication by the global scale is now fused directly into the matmul kernel, which is expected to yield significant performance improvements by reducing separate operations.
• API Enhancement: The public jax.nn.scaled_matmul function has been updated to accept an optional global_scale argument, providing users with access to this new fused functionality.
• Precision Handling: The float16 data type has been explicitly disabled in relevant tests due to its insufficient precision for the nvfp4 mode, ensuring numerical stability.

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 adds support for a global_scale to the scaled_matmul_wrapper, which is a good performance optimization. The changes are mostly consistent, but I've found a critical issue with the custom partitioning logic which has not been updated to support the new arguments. This will likely cause runtime errors. I've also pointed out a couple of minor bugs related to dtype mismatches when creating dummy scale tensors.