[Build] No need to use the more restrictive gencode then necessary

[LucasWilkinson]

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

The a gencode variants are not forward compatible. We should only use them when we need to take advantage of specific a feature (e.g. for 9.0a the a enables the wgmma instructions)

Test Plan / Results

Ran python -m pytest tests/kernels/moe/test_cutlass_moe.py on a hopper device (Failures match main; no new failures)
TODO: test Blackwell

(Optional) Documentation Update

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Summary of Changes

Hello @LucasWilkinson, 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 aims to improve the forward compatibility of the build system by relaxing the gencode requirements for certain CUDA architectures. The change ensures that the build process for CUTLASS MoE kernels uses standard gencode variants unless specific 'a' variant features are explicitly required, which are noted as not being forward compatible.

Highlights

• Build Configuration: I've updated the CMakeLists.txt to use less restrictive gencode variants for CUTLASS MoE kernels. Specifically, the 9.0a and 10.0a gencode variants have been changed to 9.0 and 10.0 respectively for CUTLASS_MOE_DATA_ARCHS.

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 is currently in preview and 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 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

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 correctly changes the CUDA gencode for moe_data.cu from architecture-specific variants (e.g., 9.0a) to base architectures (e.g., 9.0). This is a good improvement for build-time forward compatibility, as this file doesn't appear to use any architecture-specific instructions. The change is logical and well-justified. I've added one comment regarding the maintainability of the surrounding build logic, which could be improved in the future to be less brittle.

[gemini-code-assist]

This change to use base architectures (9.0, 10.0) instead of the more restrictive a-variants is correct and improves forward compatibility for moe_data.cu.

However, the overall logic for compiling this file is a bit brittle. The if condition on the next line uses a hardcoded CUDA version (>= 12.3) which is tied to the requirements of the Hopper MoE kernel. This could cause maintenance issues when new MoE kernels with different CUDA version requirements are added in the future.

A more robust approach would be to compile moe_data.cu if any CUTLASS MoE kernel is enabled, rather than checking for a specific CUDA version. While this is outside the scope of the current change, it's something to consider for future refactoring to improve the build system's maintainability.