Upgrade der Computeverwaltung auf v2

Die Funktionalität zur Computeverwaltung bleibt auf der Entwicklungsplattform von v2 unverändert.

In diesem Artikel werden Szenarien in SDK v1 und SDK v2 verglichen.

Erstellen einer Compute-Instanz

  • SDK v1

    import datetime
    import time
    
    from azureml.core.compute import ComputeTarget, ComputeInstance
    from azureml.core.compute_target import ComputeTargetException
    
    # Compute Instances need to have a unique name across the region.
    # Here, we create a unique name with current datetime
    ci_basic_name = "basic-ci" + datetime.datetime.now().strftime("%Y%m%d%H%M")
    
    compute_config = ComputeInstance.provisioning_configuration(
            vm_size='STANDARD_DS3_V2'
        )
        instance = ComputeInstance.create(ws, ci_basic_name , compute_config)
        instance.wait_for_completion(show_output=True)
    
  • SDK v2

    # Compute Instances need to have a unique name across the region.
    # Here, we create a unique name with current datetime
    from azure.ai.ml.entities import ComputeInstance, AmlCompute
    import datetime
    
    ci_basic_name = "basic-ci" + datetime.datetime.now().strftime("%Y%m%d%H%M")
    ci_basic = ComputeInstance(name=ci_basic_name, size="STANDARD_DS3_v2", idle_time_before_shutdown_minutes="30")
    ml_client.begin_create_or_update(ci_basic)
    

Erstellen eines Computeclusters

  • SDK v1

    from azureml.core.compute import ComputeTarget, AmlCompute
    from azureml.core.compute_target import ComputeTargetException
    
    # Choose a name for your CPU cluster
    cpu_cluster_name = "cpucluster"
    compute_config = AmlCompute.provisioning_configuration(vm_size='STANDARD_DS3_V2',
                                                               max_nodes=4)
    cpu_cluster = ComputeTarget.create(ws, cpu_cluster_name, compute_config)
    cpu_cluster.wait_for_completion(show_output=True)
    
  • SDK v2

    from azure.ai.ml.entities import AmlCompute
    cpu_cluster_name = "cpucluster"
    cluster_basic = AmlCompute(
        name=cpu_cluster_name,
        type="amlcompute",
        size="STANDARD_DS3_v2",
        max_instances=4
    )
    ml_client.begin_create_or_update(cluster_basic)
    

Zuordnung der wichtigsten Funktionen in SDK v1 und SDK v2

Funktionalität im SDK v1 Grobe Zuordnung in SDK v2
Methode/API im SDK v1 (mit Links zur Referenzdokumentation) Methode/API im SDK v2 (mit Links zur Referenzdokumentation)

Nächste Schritte