Azure ML scripts#

Olive provides a couple of scripts to help you manage your Azure ML assets.

Scripts list#

manage_compute_instance#

This Python script provides a command-line interface for managing compute resources in an Azure Machine Learning workspace.

  • --create or -c: A flag indicating that a new compute resource should be created. This is mutually exclusive with --delete - only one of them can be specified at a time.

  • --delete or -d: A flag indicating that an existing compute resource should be deleted. This is mutually exclusive with --create - only one of them can be specified at a time.

  • --subscription_id: The ID of your Azure subscription.

  • --resource_group: The name of your Azure resource group.

  • --workspace_name: The name of your Azure Machine Learning workspace.

  • --aml_config_path: The path to your AzureML config file. If this is provided, subscription_id, resource_group and workspace_name are ignored.

  • --compute_name: The name of the new compute resource. This is a required argument.

  • --vm_size: The VM size of the new compute resource. This is required if you are creating a compute instance.

  • --location: The location where the new compute resource should be created. This is required if you are creating a compute instance.

  • --min_nodes: The minimum number of nodes for the new compute resource. If this is not provided, the default value is 0.

  • --max_nodes: The maximum number of nodes for the new compute resource. If this is not provided, the default value is 2.

  • --idle_time_before_scale_down: The number of idle seconds before the compute resource scales down. If this is not provided, the default value is 120 seconds.

aml_config_path is a json file for your azureml config:

{
    "subscription_id": "<subscription_id>",
    "resource_group": "<resource_group>",
    "workspace_name": "<workspace_name>",
}

Usage#

You can use olive manage-aml-compute command line tool to create an AzureML compute instance from the command line like this:

olive manage-aml-compute --create --subscription_id <subscription_id> --resource_group <resource_group> --workspace_name <workspace_name> --compute_name <compute_name> --vm_size <vm_size> --location <location> --min_nodes <min_nodes> --max_nodes <max_nodes> --idle_time_before_scale_down <idle_time_before_scale_down>

or

olive manage-aml-compute --create --aml_config_path </path/to/aml_config.json> --compute_name <compute_name> --vm_size <vm_size> --location <location> --min_nodes <min_nodes> --max_nodes <max_nodes> --idle_time_before_scale_down <idle_time_before_scale_down>

You can delete an AzureML compute instance by:

olive manage-aml-compute --delete --compute_name <compute_name>

More details can be found at Command Line Tools.