Add Linux Aarch64 G3 runners to vLLM bms

[ghost]

PyTorch already have G3 runners (see https://hud.pytorch.org/runners/pytorch?search=m7g).
This patch lets us leverage those in the vLLM benchmark dashboard,
on the same model that's used to measure x86 performance.

This PR aims to fix vllm-project/vllm#26019

[huydhn]

I'm not sure about this part, setting the DEVICE_NAME to arm64-cpu is fine. But I think we need to keep ON_CPU=1 here for both x86 and arm. That's a flag from vLLM itself to run benchmark on CPU https://github.com/vllm-project/vllm/blob/94666612a938380cb643c1555ef9aa68b7ab1e53/docs/contributing/benchmarks.md?plain=1#L1177

[ioghiban]

Alright, I skipped setting ON_CPU for arm since this env var is used in .github/scripts/run-sglang-performance-benchmarks.sh expecting either NUMA nodes or GPUs, of which neither should be available on the existing Arm Neoverse runners. So, if we set ON_CPU to 1 in this case, I'd expect some warnings there.

[ioghiban]

Also, in vllm/.buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh we check these env vars and again expect NUMA nodes for CPUs, so a distinction needs to be made here for the available Aarch64 runners, see this PR

[meta-cla]

Hi @ghost!

Thank you for your pull request.

We require contributors to sign our Contributor License Agreement, and yours needs attention.

You currently have a record in our system, but the CLA is no longer valid, and will need to be resubmitted.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!