darts-gpu.yaml

---
apiVersion: kubeflow.org/v1beta1
kind: Experiment
metadata:
  namespace: kubeflow
  name: darts-gpu
spec:
  parallelTrialCount: 1
  maxTrialCount: 1
  maxFailedTrialCount: 1
  objective:
    type: maximize
    objectiveMetricName: Best-Genotype
  metricsCollectorSpec:
    collector:
      kind: StdOut
    source:
      filter:
        metricsFormat:
          - "([\\w-]+)=(Genotype.*)"
  algorithm:
    algorithmName: darts
    algorithmSettings:
      - name: num_epochs
        value: "3"
  nasConfig:
    graphConfig:
      numLayers: 3
    operations:
      - operationType: separable_convolution
        parameters:
          - name: filter_size
            parameterType: categorical
            feasibleSpace:
              list:
                - "3"
      - operationType: dilated_convolution
        parameters:
          - name: filter_size
            parameterType: categorical
            feasibleSpace:
              list:
                - "3"
                - "5"
      - operationType: avg_pooling
        parameters:
          - name: filter_size
            parameterType: categorical
            feasibleSpace:
              list:
                - "3"
      - operationType: max_pooling
        parameters:
          - name: filter_size
            parameterType: categorical
            feasibleSpace:
              list:
                - "3"
      - operationType: skip_connection
  trialTemplate:
    primaryContainerName: training-container
    trialParameters:
      - name: algorithmSettings
        description: Algorithm settings of DARTS Experiment
        reference: algorithm-settings
      - name: searchSpace
        description: Search Space of DARTS Experiment
        reference: search-space
      - name: numberLayers
        description: Number of Neural Network layers
        reference: num-layers
    trialSpec:
      apiVersion: batch/v1
      kind: Job
      spec:
        template:
          spec:
            containers:
              - name: training-container
                image: docker.io/kubeflowkatib/darts-cnn-cifar10-gpu:latest
                command:
                  - python3
                  - run_trial.py
                  - --algorithm-settings="${trialParameters.algorithmSettings}"
                  - --search-space="${trialParameters.searchSpace}"
                  - --num-layers="${trialParameters.numberLayers}"
                resources:
                  limits:
                    nvidia.com/gpu: 1
            restartPolicy: Never