Azure Machine Learning CLI Documentation

Introduction to the Azure ML CLI

The Azure Machine Learning CLI (v2) is a powerful command-line interface that allows you to manage and orchestrate your machine learning workloads on Azure. It provides a consistent and efficient way to interact with Azure Machine Learning resources, from dataset registration to model deployment.

This documentation guides you through the installation, core concepts, and various commands available in the Azure ML CLI. Whether you are a data scientist, ML engineer, or DevOps professional, the CLI is an essential tool for streamlining your ML lifecycle.

Installation

To use the Azure ML CLI, you need to have the Azure CLI installed first. If you don't have it, you can install it from the official Azure CLI documentation.

Once the Azure CLI is installed, you can install the Azure ML extension:

az extension add --name ml

To verify the installation, run:

az ml --version

Getting Started

Before you can use the CLI, you need to log in to your Azure account and set your default subscription:

az login

az account set --subscription 

Next, you'll need to configure your Azure ML workspace. This can be done manually or by pointing the CLI to an existing workspace. For creating a new workspace, you can use the az ml workspace create command.

az ml workspace create --name my-workspace --resource-group my-resource-group --location eastus

After setting up your workspace, you can start interacting with your ML resources.

CLI Commands

The Azure ML CLI is organized into logical groups for managing different aspects of your machine learning projects.

az ml workspace

Manage your Azure Machine Learning workspaces.

• az ml workspace create: Create a new workspace.
• az ml workspace show: Show details of a workspace.
• az ml workspace list: List workspaces in a subscription or resource group.
• az ml workspace delete: Delete a workspace.

az ml workspace show --name my-workspace --resource-group my-resource-group

az ml compute

Manage compute resources for your ML workloads.

• az ml compute create: Create a compute target (e.g., AmlCompute, Kubernetes).
• az ml compute show: Show details of a compute target.
• az ml compute list: List compute targets in a workspace.
• az ml compute delete: Delete a compute target.

az ml compute create --type amlcompute --name cpu-cluster --min-instances 0 --max-instances 4 --resource-group my-resource-group --workspace-name my-workspace

az ml environment

Manage environments for your training and inference jobs.

• az ml environment create: Create an environment from a YAML file.
• az ml environment show: Show details of an environment.
• az ml environment list: List environments.
• az ml environment delete: Delete an environment.

az ml job

Submit, manage, and monitor training jobs.

• az ml job create: Submit a job (e.g., command, sweep).
• az ml job show: Show details of a job.
• az ml job list: List jobs.
• az ml job stream: Stream logs for a job.
• az ml job cancel: Cancel a running job.

az ml job create --file jobs/my-command-job.yml

az ml model

Register and manage your trained models.

• az ml model create: Register a model.
• az ml model show: Show details of a model.
• az ml model list: List registered models.
• az ml model delete: Delete a model.

az ml data

Register and manage your datasets.

• az ml data create: Register a dataset.
• az ml data show: Show details of a dataset.
• az ml data list: List datasets.
• az ml data delete: Delete a dataset.

az ml endpoint

Manage endpoints for model deployment.

• az ml endpoint create: Create an endpoint (e.g., online, batch).
• az ml endpoint show: Show details of an endpoint.
• az ml endpoint list: List endpoints.
• az ml endpoint delete: Delete an endpoint.

az ml deployment

Manage deployments associated with endpoints.

• az ml deployment create: Create a deployment for an endpoint.
• az ml deployment show: Show details of a deployment.
• az ml deployment list: List deployments for an endpoint.
• az ml deployment delete: Delete a deployment.

az ml online-endpoint create --name my-online-endpoint --file endpoints/online/my-endpoint.yml
az ml online-deployment create --endpoint my-online-endpoint --name blue --file endpoints/online/my-deployment-blue.yml --all-traffic

Configuration

The Azure ML CLI relies heavily on YAML configuration files for defining resources, jobs, environments, and deployments. This declarative approach promotes reproducibility and version control.

Key configuration concepts include:

• Jobs: Define the steps, scripts, inputs, outputs, and compute for training or batch inference.
• Environments: Specify the Docker image, Conda packages, or pip dependencies required for your code to run.
• Models: Define how a trained model is registered, including its path, type, and framework.
• Datasets: Describe how to access data, whether it's from Azure Blob Storage, ADLS Gen2, or other sources.
• Endpoints & Deployments: Configure the infrastructure for online or batch inference.

Refer to the official Azure ML YAML examples for detailed templates.

Tutorials and Examples

To further your understanding and practical application of the Azure ML CLI, explore the following resources:

• Train a model with the Azure ML CLI
• Deploy models to managed online endpoints using CLI
• Azure ML CLI Samples Repository