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

Some workloads (like our notebook docs compiler) require us to run multiple servers on a single GPU, but it's not yet very convenient to do so. We need an easier way to allocate GPU memory.

Related resources

No response