Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

Currently, fa3 is also in comm_ops. For the NVCC flags, we contain sm75, sm80, sm89, sm100, and sm100a. However, we only need sm90 and sm90a. Therefore, we can consider using a new so and switching to the new one. This change will significantly reduce the wheel size.

Related resources

No response