Question

A machine learning team trains models in a batch pipeline job. After each successful run, they want the resulting model artifact to be tracked as a versioned asset in the Azure Machine Learning workspace so that deployments can reference a specific, immutable version. The team wants to do this from the Azure CLI (ml extension) in an automated way, referencing the output produced by the training job. Which approach should they use?

Accepted Answer

Register the model as a versioned asset with 'az ml model create', setting the --path to the job output URI (azureml://jobs/<job-name>/outputs/...) so a new version is created each time.

Answer

Copy the model file to a datastore folder and register it as a data asset with 'az ml data create', then deploy directly from the data asset.

Answer

Use 'az ml environment create' to package the model into a curated environment and reference the environment version at deployment time.

Answer

Store the model in the workspace default blob container manually and reference the blob URL directly in the deployment YAML without registering an asset.