train-cnn/train.yaml

taskGroups:
  - taskSpec:
      runnables:
        - container:
            imageUri: us-central1-docker.pkg.dev/<your project>/<your repo>/<your image name>
            volumes: /tmp:/tmp
            commands:
              - /scripts/localize.sh
        - barrier:
            name: all-vms-ready
        - container:
            imageUri: us-central1-docker.pkg.dev/<your project>/<your repo>/<your image name>
            options: --network host
            volumes: /tmp:/tmp
            commands:
              - /scripts/runtorch.sh
        - barrier:
            name: training-complete
      environment:
        variables:
          DATA_ROOT: /mnt/disks/dogs-vs-cats/data
      volumes:
        - nfs:
            server: <your Filestore IP address>
            remotePath: <your Filestore share path, e.g. /share>
          mountPath: /mnt/disks/dogs-vs-cats
    taskCount: 8
    taskCountPerNode: 1
    requireHostsFile: true
allocationPolicy:
  instances:
    - policy:
        bootDisk:
          sizeGb: 250
        machineType: n1-standard-4
        accelerators:
          - type: nvidia-tesla-v100
            count: 1
      installGpuDrivers: true
logsPolicy:
  destination: CLOUD_LOGGING