Work with registered models in Azure Machine Learning

Article
08/14/2024

APPLIES TO: Azure CLI ml extension v2 (current) Python SDK azure-ai-ml v2 (current)

In this article, you learn to register and work with models in Azure Machine Learning by using:

The Azure Machine Learning studio UI.
The Azure Machine Learning V2 CLI.
The Python Azure Machine Learning V2 SDK.

You learn how to:

Create registered models in the model registry from local files, datastores, or job outputs.
Work with different types of models, such as custom, MLflow, and Triton.
Use models as inputs or outputs in training jobs.
Manage the lifecycle of model assets.

Model registration

Model registration lets you store and version your models in your workspace in the Azure cloud. The model registry helps you organize and keep track of your trained models. You can register models as assets in Azure Machine Learning by using the Azure CLI, the Python SDK, or the Machine Learning studio UI.

Supported paths

To register a model, you need to specify a path that points to the data or job location. The following table shows the various data locations Azure Machine Learning supports, and the syntax for the path parameter:

Location	Syntax
Local computer	`<model-folder>/<model-filename>`
Azure Machine Learning datastore	`azureml://datastores/<datastore-name>/paths/<path_on_datastore>`
Azure Machine Learning job	`azureml://jobs/<job-name>/outputs/<output-name>/paths/<path-to-model-relative-to-the-named-output-location>`
MLflow job	`runs:/<run-id>/<path-to-model-relative-to-the-root-of-the-artifact-location>`
Model asset in a Machine Learning workspace	`azureml:<model-name>:<version>`
Model asset in a Machine Learning registry	`azureml://registries/<registry-name>/models/<model-name>/versions/<version>`

Supported modes

When you use models for inputs or outputs, you can specify one of the following modes. For example, you can specify whether the model should be read-only mounted or downloaded to the compute target.

ro_mount: Mount the data to the compute target as read-only.
rw_mount: Read-write mount the data.
download: Download the data to the compute target.
upload: Upload the data from the compute target.
direct: Pass in the URI as a string.

The following table shows the available mode options for different model type inputs and outputs.

Type	`upload`	`download`	`ro_mount`	`rw_mount`	`direct`
`custom` file input
`custom` folder input		✓	✓		✓
`mlflow` input		✓	✓
`custom` file output	✓			✓	✓
`custom` folder output	✓			✓	✓
`mlflow` output	✓			✓	✓

Prerequisites

An Azure subscription with a free or paid version of Azure Machine Learning. If you don't have an Azure subscription, create a Trial before you begin.
An Azure Machine Learning workspace.

To run the code samples in this article and work with the Azure Machine Learning V2 CLI or Python Azure Machine Learning V2 SDK, you also need:

Azure CLI
Python SDK

Azure CLI version 2.38.0 or greater installed.
V2 of the ml extension installed by running the following command. For more information, see Install, set up, and use the CLI (v2).
```
az extension add -n ml
```

Note

V2 provides full backward compatibility. You can still use model assets from the v1 SDK or CLI. All models registered with the v1 CLI or SDK are assigned the type custom.

Register a model by using the studio UI

To register a model by using the Azure Machine Learning studio UI:

In your workspace in the studio, select Models from the left navigation.
On the Model List page, select Register, and select one of the following locations from the dropdown list:
- From local files
- From a job output
- From datastore
- From local files (based on framework)
On the first Register model screen:
1. Navigate to the local file, datastore, or job output for your model.
2. Select the input model type: MLflow, Triton, or Unspecified type.
On the Model settings screen, provide a name and other optional settings for your registered model, and select Next.
On the Review screen, review the configuration and then select Register.

Register a model by using the Azure CLI or Python SDK

The following code snippets cover how to register a model as an asset in Azure Machine Learning by using the Azure CLI or Python SDK. These snippets use custom and mlflow model types.

custom type refers to a model file or folder trained with a custom standard that Azure Machine Learning doesn't currently support.
mlflow type refers to a model trained with MLflow. MLflow trained models are in a folder that contains the MLmodel file, the model file, the conda dependencies file, and the requirements.txt file.

Tip

You can follow along with the Python versions of the following samples by running the model.ipynb notebook in the azureml-examples repository.

Connect to your workspace

The workspace is the top-level resource for Azure Machine Learning, providing a centralized place to work with all the artifacts you create when you use Azure Machine Learning. In this section, you connect to your Azure Machine Learning workspace to create the registered model.

Azure CLI
Python SDK

Sign in to Azure by running az login and following the prompts.
In the following commands, replace <subscription-id>, <workspace-name>, <resource-group>, and <location> placeholders with the values for your environment.
```
az account set --subscription <subscription-id>
az configure --defaults workspace=<workspace-name> group=<resource-group> location=<location>
```

Import the required libraries:

from azure.ai.ml import MLClient, Input
from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes
from azure.identity import DefaultAzureCredential

Configure workspace details and get a handle to the workspace. In the following code snippet, replace the <subscription-id>, <resource-group>, and <workspace-name> placeholders with the values for your environment.
```
subscription_id = "<subscription-id>"
resource_group = "<resource-group>"
workspace = "<workspace-name>"

ml_client = MLClient(DefaultAzureCredential(), subscription_id, resource_group, workspace)
```

Create the registered model

You can create a registered model from a model that's:

Located on your local computer.
Located on an Azure Machine Learning datastore.
Output from an Azure Machine Learning job.

Create a YAML file <file-name>.yml. In the file, provide a name for your registered model, a path to the local model file, and a description. For example:

$schema: https://azuremlschemas.azureedge.net/latest/model.schema.json
name: local-file-example
path: mlflow-model/model.pkl
description: Model created from local file.

Run the following command, using the name of your YAML file:
```
az ml model create -f <file-name>.yml
```

For a complete example, see the model YAML.

The following example shows creating a registered model from a local file.

from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes

file_model = Model(
    path="mlflow-model/model.pkl",
    type=AssetTypes.CUSTOM_MODEL,
    name="local-file-example",
    description="Model created from local file.",
)
ml_client.models.create_or_update(file_model)

Datastore

You can create a model from a cloud path by using any of the supported URI formats.

Azure CLI
Python SDK

The following example uses the shorthand azureml scheme for pointing to a path on the datastore by using the syntax azureml://datastores/<datastore-name>/paths/<path_on_datastore>.

az ml model create --name my-model --version 1 --path azureml://datastores/myblobstore/paths/models/cifar10/cifar.pt

For a complete example, see the CLI reference.

The following example uses the shorthand azureml scheme for pointing to a path on the datastore by using the syntax azureml://datastores/${{datastore-name}}/paths/${{path_on_datastore}}.

from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes

cloud_model = Model(
    path=file_model.path,
    # The above line basically provides a path in the format "azureml://subscriptions/XXXXXXXXXXXXXXXX/resourceGroups/XXXXXXXXXXX/workspaces/XXXXXXXXXXX/datastores/workspaceblobstore/paths/model.pkl"
    # Users could also use,"azureml://datastores/workspaceblobstore/paths/model.pkl" as a shorthand to the same location
    name="cloud-path-example",
    type=AssetTypes.CUSTOM_MODEL,
    description="Model created from cloud path.",
)
ml_client.models.create_or_update(cloud_model)

Job output

If your model data comes from a job output, you have two options for specifying the model path. You can use the MLflow runs: URI format or the azureml://jobs URI format.

Note

The artifacts reserved keyword represents output from the default artifact location.

MLflow runs: URI format

This option is optimized for MLflow users, who are probably already familiar with the MLflow runs: URI format. This option creates a model from artifacts in the default artifact location, where all MLflow-logged models and artifacts are located. This option also establishes a lineage between a registered model and the run the model came from.

Format: runs:/<run-id>/<path-to-model-relative-to-the-root-of-the-artifact-location>

Example:
- Azure CLI
- Python SDK
```
az ml model create --name my-registered-model --version 1 --path runs:/my_run_0000000000/model/ --type mlflow_model
```
```
from azure.ai.ml.entities import Model
from azure.ai.ml.constants import ModelType

run_model = Model(
    path="runs:/my_run_0000000000/model/"
    name="my-registered-model",
    description="Model created from run.",
    type=ModelType.MLFLOW_MODEL
)

ml_client.models.create_or_update(run_model) 
```

azureml://jobs URI format

The azureml://jobs reference URI option lets you register a model from artifacts in any of the job's output paths. This format aligns with the azureml://datastores reference URI format, and also supports referencing artifacts from named outputs other than the default artifact location.

If you didn't directly register your model within the training script by using MLflow, you can use this option to establish a lineage between a registered model and the job it was trained from.

Format: azureml://jobs/<run-id>/outputs/<output-name>/paths/<path-to-model>
- Default artifact location: azureml://jobs/<run-id>/outputs/artifacts/paths/<path-to-model>/. This location is equivalent to MLflow runs:/<run-id>/<model>.
- Named output folder: azureml://jobs/<run-id>/outputs/<named-output-folder>
- Specific file within the named output folder: azureml://jobs/<run-id>/outputs/<named-output-folder>/paths/<model-filename>
- Specific folder path within the named output folder: azureml://jobs/<run-id>/outputs/<named-output-folder>/paths/<model-folder-name>
Example:
- Azure CLI
- Python SDK
Save a model from a named output folder:
```
az ml model create --name run-model-example --version 1 --path azureml://jobs/my_run_0000000000/outputs/artifacts/paths/model/
```
For a complete example, see the CLI reference.
Save a model from a named output:
```
from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes

job_name = "<JOB_NAME>"

run_model = Model(
    path=f"azureml://jobs/{job_name}/outputs/artifacts/paths/model/",
    name="run-model-example",
    description="Model created from run.",
    type=AssetTypes.MLFLOW_MODEL,
)
# Uncomment after adding required details above
# ml_client.models.create_or_update(run_model)
```
For a complete example, see the model notebook.

Use models for training

The v2 Azure CLI and Python SDK also let you use models as inputs or outputs in training jobs.

Use a model as input in a training job

Azure CLI
Python SDK

Create a job specification YAML file, <file-name>.yml. In the inputs section of the job, specify:
- The model type, which can be mlflow_model, custom_model, or triton_model.
- The path where your model is located, which can be any of the paths listed in the comment of the following example.

$schema: https://azuremlschemas.azureedge.net/latest/commandJob.schema.json

# Possible Paths for models:
# AzureML Datastore: azureml://datastores/<datastore-name>/paths/<path_on_datastore>
# MLflow run: runs:/<run-id>/<path-to-model-relative-to-the-root-of-the-artifact-location>
# Job: azureml://jobs/<job-name>/outputs/<output-name>/paths/<path-to-model-relative-to-the-named-output-location>
# Model Asset: azureml:<my_model>:<version>

command: |
  ls ${{inputs.my_model}}
inputs:
  my_model:
    type: mlflow_model # List of all model types here: /machine-learning/reference-yaml-model#yaml-syntax
    path: ../../assets/model/mlflow-model
environment: azureml://registries/azureml/environments/sklearn-1.0/labels/latest

Run the following command, substituting your YAML filename.
```
az ml job create -f <file-name>.yml
```

For a complete example, see the model GitHub repo.

The Input class allows you to define:

The model asset type, which can be one of the following values:
- AssetTypes.CUSTOM_MODEL
- AssetTypes.MLFLOW_MODEL
- AssetTypes.TRITON_MODEL
The path to your model data, which can be one of the following locations:
- Local path: <model-folder>/<model-filename>
- Azure Machine Learning datastore: azureml://datastores/<datastore-name>/paths/<path_on_datastore>
- MLflow run: runs:/<run-id>/<path-to-model-relative-to-the-root-of-the-artifact-location>
- Job: azureml://jobs/<job-name>/outputs/<output-name>/paths/<path-to-model-relative-to-the-named-output-location>
- Model asset: azureml:<my_model>:<version>

The following example inputs an MLflow model from a local folder.

from azure.ai.ml import command
from azure.ai.ml.entities import Model
from azure.ai.ml import Input
from azure.ai.ml.constants import AssetTypes
from azure.ai.ml import MLClient

my_job_inputs = {
    "input_model": Input(type=AssetTypes.MLFLOW_MODEL, path="mlflow-model")
}

job = command(
    code="./src",  # local path where the code is stored
    command="ls ${{inputs.input_model}}",
    inputs=my_job_inputs,
    environment="AzureML-sklearn-0.24-ubuntu18.04-py37-cpu:9",
    compute="cpu-cluster",
)

# submit the command
returned_job = ml_client.jobs.create_or_update(job)
# get a URL for the status of the job
returned_job.services["Studio"].endpoint

Write a model as output for a job

Your job can write a model to your cloud-based storage by using outputs.

Azure CLI
Python SDK

Create a job specification YAML file <file-name>.yml. Populate the outputs section with the output model type and path.

$schema: https://azuremlschemas.azureedge.net/latest/commandJob.schema.json

# Possible Paths for Model:
# Local path: mlflow-model/model.pkl
# AzureML Datastore: azureml://datastores/<datastore-name>/paths/<path_on_datastore>
# MLflow run: runs:/<run-id>/<path-to-model-relative-to-the-root-of-the-artifact-location>
# Job: azureml://jobs/<job-name>/outputs/<output-name>/paths/<path-to-model-relative-to-the-named-output-location>
# Model Asset: azureml:<my_model>:<version>

code: src
command: >-
  python hello-model-as-output.py 
  --input_model ${{inputs.input_model}} 
  --custom_model_output ${{outputs.output_folder}}
inputs:
  input_model: 
    type: mlflow_model # mlflow_model,custom_model, triton_model
    path: ../../assets/model/mlflow-model
outputs:
  output_folder: 
    type: custom_model # mlflow_model,custom_model, triton_model
environment: azureml:AzureML-sklearn-1.0-ubuntu20.04-py38-cpu@latest

Create a job by using the CLI:

az ml job create --file <file-name>.yml

For a complete example, see the model GitHub repo.

The Output class allows you to define:

The model asset type, which can be one of the following values:
- AssetTypes.CUSTOM_MODEL
- AssetTypes.MLFLOW_MODEL
- AssetTypes.TRITON_MODEL
The path for your model data, which can be one of the following locations:
- Local path: mlflow-model/model.pkl
- Azure Machine Learning datastore: azureml://datastores/<datastore-name>/paths/<path_on_datastore>
- MLflow run: runs:/<run-id>/<path-to-model-relative-to-the-root-of-the-artifact-location>
- Job: azureml://jobs/<job-name>/outputs/<output-name>/paths/<path-to-model-relative-to-the-named-output-location>
- Model asset: azureml:<my_model>:<version>

The following example creates an output that mounts your default datastore in read-write mode. The code simply loads the local MLflow model as input and exports the same model as an output of the job saved in the mounted datastore.

from azure.ai.ml import command
from azure.ai.ml.entities import Model
from azure.ai.ml import Input, Output
from azure.ai.ml.constants import AssetTypes

my_job_inputs = {
    "input_model": Input(type=AssetTypes.MLFLOW_MODEL, path="mlflow-model"),
    "input_data": Input(type=AssetTypes.URI_FILE, path="./mlflow-model/input_example.json"),}

my_job_outputs = {
    "output_folder": Output(type=AssetTypes.CUSTOM_MODEL)
}

job = command(
    code="./src",  # local path where the code is stored
    command="python load_write_model.py --input_model ${{inputs.input_model}} --output_folder ${{outputs.output_folder}}",
    inputs=my_job_inputs,
    outputs=my_job_outputs,
    environment="AzureML-sklearn-0.24-ubuntu18.04-py37-cpu:9",
    compute="cpu-cluster",
)

# submit the command
returned_job = ml_client.create_or_update(job)
# get a URL for the status of the job
returned_job.services["Studio"].endpoint

Manage models

The Azure CLI and Python SDK also allow you to manage the lifecycle of your Azure Machine Learning model assets.

List

Azure CLI
Python SDK

List all the models in your workspace:

az ml model list

List all the model versions under a given name:

az ml model list --name run-model-example

List all the models in your workspace:

models = ml_client.models.list()
for model in models:
    print(model.name)

List all the model versions under a given name:

models = ml_client.models.list(name="run-model-example")
for model in models:
    print(model.version)

Show

Get the details of a specific model:

Azure CLI
Python SDK

az ml model show --name run-model-example --version 1

model_example = ml_client.models.get(name="run-model-example", version="1")
print(model_example)

Update

Update mutable properties of a specific model:

Important

For models, only description and tags can be updated. All other properties are immutable, and if you need to change them, you should create a new version of the model.

Azure CLI
Python SDK

az ml model update --name  run-model-example --version 1 --set description="This is an updated description." --set tags.stage="Prod"

model_example.description="This is an updated description."
model_example.tags={"stage":"Prod"}
ml_client.models.create_or_update(model=model_example)

Work with registered models in Azure Machine Learning

Model registration

Supported paths

Supported modes

Prerequisites

Register a model by using the studio UI

Register a model by using the Azure CLI or Python SDK

Connect to your workspace

Create the registered model

Local file or folder

Datastore

Job output

Use models for training

Use a model as input in a training job

Write a model as output for a job

Manage models

List

Show

Update

Archive

Additional resources

Work with registered models in Azure Machine Learning

Model registration

Supported paths

Supported modes

Prerequisites

Register a model by using the studio UI

Register a model by using the Azure CLI or Python SDK

Connect to your workspace

Create the registered model

Local file or folder

Datastore

Job output

Use models for training

Use a model as input in a training job

Write a model as output for a job

Manage models

List

Show

Update

Archive

Related content

Additional resources