Hi team, #C02116BBNTU I'm experiencing an issue with GPU detection in our Metaflow jobs. Here’s a quick summary: • Manual Pods: When I launch pods manually with

runtimeClassName: nvidia

, the GPU is detected and

nvidia-smi

works correctly. • Metaflow Pods: However, jobs run via Metaflow don’t see the GPU, as it appears we can’t set `runtimeClassName`in the Metaflow Kubernetes integration. Could we explore a workaround—such as a mutating webhook or a patch—to allow setting

runtimeClassName

for Metaflow pods? Thanks!

Hi! Are you using the nvidia device plugin for k8s? it will automatically do the right thing for you