use-cases/model-fine-tuning-pipeline/fine-tuning/pytorch/manifests/provisioning-request-h100.yaml

apiVersion: v1
kind: PodTemplate
metadata:
  name: h100-job
  namespace: ml-team
template:
  spec:
    nodeSelector:
      cloud.google.com/gke-nodepool: gpu-h100x8-a3h8-dws
    tolerations:
    - key: "nvidia.com/gpu"
      operator: "Exists"
      effect: "NoSchedule"
    - key: "on-demand"
      value: "true"
      operator: "Equal"
      effect: "NoSchedule"    
    containers:
    - name: pi
      image: perl
      command: ["/bin/sh"]
      resources:
        limits:
          cpu: "700m"
          nvidia.com/gpu: 8
        requests:
          cpu: "700m"
          nvidia.com/gpu: 8
    restartPolicy: Never
---
apiVersion: autoscaling.x-k8s.io/v1beta1
kind: ProvisioningRequest
metadata:
  name: h100-job
  namespace: ml-team
spec:
  provisioningClassName: queued-provisioning.gke.io
  parameters:
    maxRunDurationSeconds: "86400"
  podSets:
  - count: 1
    podTemplateRef:
      name: h100-job