job.yml

apiVersion: batch/v1
kind: Job
metadata:
  name: test-experiment
  namespace: aiml-valmadre-research
  labels:
    gcp_secret_file: data-manager
spec:
  template:
    spec:
      restartPolicy: Never
      volumes:
        - name: dataset-volume
          persistentVolumeClaim:
            claimName: hier-cls-data
        - name: shared-memory
          emptyDir:
            medium: Memory
      containers:
        - name: test-experiment-container
          image: docker.aiml.team/jack.valmadre/hier-class:latest
          stdin: true
          tty: true
          command: ["/bin/sh", "-c"]
          args:
            - git clone https://deep-purple:$SECRET_SECRET_KEY@gitlab.aiml.team/jack.valmadre/hier-class.git &&
              cd v2p &&
              ./run.sh 

          resources:
            limits:
              nvidia.com/gpu: 2
              memory: 128Gi
              cpu: 8000m
            requests:
              memory: 128Gi
              cpu: 8000m
          volumeMounts:
            - mountPath: /data
              name: dataset-volume
            - mountPath: /dev/shm
              name: shared-memory
          env:
            - name: TORCH_HOME
              value: /data
            - name: SECRET_SECRET_KEY
              valueFrom:
                secretKeyRef:
                  name: deep-purple
                  key: token