Vouch request: brianwtaylor

[brianwtaylor]

What do you want to work on?

Fix #181 — GPU setup failures shouldn't take down the whole cluster. I'd add a fallback path so the cluster still comes up without GPU workloads if something in the GPU chain breaks (missing runtime, device-plugin timeout, driver mismatch, etc), with clear log output about what failed.

I have two DGX Sparks (GB10, aarch64, cgroup v2) connected over QFD so I can test both success and failure paths. I've been contributing to NemoClaw — cgroup v2 preflight fix (PR #62, merged).

Why this change?

Too many moving parts in GPU setup for it to be a hard gate. Spark users especially hit edge cases with unified memory and aarch64 that discrete GPU hosts don't see. Cluster should still be usable.

Checklist

• I wrote this request myself (not AI-generated)
• I have read CONTRIBUTING.md