training_service.yml

all:
  logCollection: http

kubeflow:
  maxExecDuration: 15m
  nniManagerIp:
  # use a small trial number to make IT faster
  maxTrialNum: 2
  trialConcurrency: 2

  kubeflowConfig:
    operator: tf-operator
    apiVersion: v1
    storage: azureStorage
    keyVault:
      vaultName:
      name:
    azureStorage:
      accountName:
      azureShare:
    namespace: kubeflow
  trial:
    worker:
      replicas: 1
      command:
      gpuNum: 1
      cpuNum: 1
      memoryMB: 8192
      image:
  trainingServicePlatform: kubeflow

frameworkcontroller:
  maxExecDuration: 15m
  nniManagerIp:
  # use a small trial number to make IT faster
  maxTrialNum: 2
  trialConcurrency: 2
  frameworkcontrollerConfig:
    serviceAccountName: frameworkcontroller
    storage: azureStorage
    keyVault:
      vaultName:
      name:
    azureStorage:
      accountName:
      azureShare:
    namespace: kubeflow
  trial:
    taskRoles:
      - name: worker
        taskNum: 1
        command:
        gpuNum: 1
        cpuNum: 1
        memoryMB: 8192
        image:
        frameworkAttemptCompletionPolicy:
          minFailedTaskCount: 1
          minSucceededTaskCount: 1
  trainingServicePlatform: frameworkcontroller

local:
  trainingServicePlatform: local
pai:
  nniManagerIp:
  maxExecDuration: 15m
  # PAI has job submission limitation, set maxTrialNum=1 to control trial job numbers for PAI 
  maxTrialNum: 1
  trialConcurrency: 1
  paiConfig:
    host:
    userName:
  trainingServicePlatform: pai
  trial:
    gpuNum: 1 
    cpuNum: 1
    image:
    memoryMB: 8192
    virtualCluster: default
    nniManagerNFSMountPath: 
    containerNFSMountPath: 
    paiStorageConfigName: 
remote:
  remoteConfig:
    reuse: false
  machineList:
  - ip:
    passwd:
    port:
    username:
  trainingServicePlatform: remote
  sharedStorage:
    storageAccountKey:
    nfsServer:
hybrid:
  maxExecDuration: 15m
  nniManagerIp:
  maxTrialNum: 2
  trialConcurrency: 2
  trial:
    gpuNum: 0
  trainingServicePlatform: hybrid
  hybridConfig:
    # TODO: Add more platforms
    trainingServicePlatforms:
      - remote
      - local
  machineList:
  - ip:
    passwd:
    port:
    username:
  remoteConfig:
    reuse: true
adl:
  maxExecDuration: 15m
  nniManagerIp:
  # use a small trial number to make IT faster
  maxTrialNum: 2
  trialConcurrency: 2
  trial:
    namespace: default
    command:
    codeDir:
    gpuNum: 1
    cpuNum: 1
    image:
    memorySize: 1Gi
    checkpoint:
      storageClass:
      storageSize:
  trainingServicePlatform: adl
aml:
  nniManagerIp:
  maxExecDuration: 15m
  # PAI has job submission limitation, set maxTrialNum=1 to control trial job numbers for PAI 
  maxTrialNum: 2
  trialConcurrency: 2
  trainingServicePlatform: aml
  trial:
    gpuNum: 1
    image:
  amlConfig:
    subscriptionId:
    resourceGroup:
    workspaceName:
    computeTarget: