How to use model version aliases

A model alias is a mutable, named reference to a model version unique within a model resource. An alias is "mutable" because aliases can be moved from one model version to another and "named" because they are user-defined, arbitrary strings. Model aliases are helpful for fetching or deploying a particular model version by reference without needing to know the specific version's ID. In this way, they operate similarly to Docker Tags or Branch references in Git.

When you create a new model in Model Registry, the first version automatically gets assigned the default alias. The default alias references the model version that is used if a user runs a command on a model without specifying a particular version. One version of a model is required to carry the default alias at all times. Otherwise, the default alias behaves as any other user-defined alias.

From the Google Cloud console the alias marker can help stakeholders understand at a glance which model is the stable version ready for deployment. In addition to the default alias, you can create and assign your own custom aliases to models in the Model Registry.

In Model Registry, you can see at a glance which model version has the default alias by looking at the alias column.

If you decide you want to reassign the alias to a different model version, you can easily move aliases to a different version.

Some considerations for using aliases:

• Your version aliases should be unique; and an alias can only be assigned to a single version, by model, at a time.
• Version aliases must be non-numerical.
• If you don't specify a model version for production, the default model is used.
• An alias is different from a label. Learn more about model labels here.
• If you apply an existing alias that is used in another model version, the alias is removed from that version.

Set a model version as default

1. In the Google Cloud console, go to the Vertex AI Model Registry page.

   Go to the Model Registry page

2. From the Model Registry, select the name of the model you want to edit. The model details window opens. All of the model versions are listed. One of the model versions has the default alias.

3. Select the Actions button for the model version you want to assign the default version.

4. Click Set as default.

Add an alias to a model version

1. In the Google Cloud console, go to the Vertex AI Model Registry page.

   Go to the Model Registry page

2. From the Model Registry, select the name of the model you want to edit. The model details window opens.

3. From the details page, select the model version you want and click More.

4. Click Edit alias. Select Add New.

5. Edit alias: Click Add alias and enter the name of the alias you want to add to the model version.

6. Click Save.

Upload a new aliased model version

from typing import List

from google.cloud import aiplatform


def upload_new_aliased_model_version_sample(
    parent_name: str,
    artifact_uri: str,
    serving_container_image: str,
    is_default_version: bool,
    version_aliases: List[str],
    version_description: str,
    project: str,
    location: str,
):
    """
    Uploads a new aliased version of a model with ID 'model_id'.
    Args:
        parent_name: The parent resource name of an existing model.
        artifact_uri: The URI of the model artifact to upload.
        serving_container_image: The name of the serving container image to use.
        is_default_version: Whether this version is the default version of the model.
        version_aliases: The aliases of the model version.
        version_description: The description of the model version.
        project: The project ID.
        location: The region name.
    Returns:
        The new version of the model.
    """
    # Initialize the client.
    aiplatform.init(project=project, location=location)

    # Upload a new aliased version of the Model resource with the ID 'model_id'. The parent_name of Model resource can be also
    # 'projects/<your-project-id>/locations/<your-region>/models/<your-model-id>'
    model = aiplatform.Model.upload(
        artifact_uri=artifact_uri,
        serving_container_image=serving_container_image,
        parent_name=parent_name,
        is_default_version=is_default_version,
        version_aliases=version_aliases,
        version_description=version_description,
    )

    return model

Delete a model alias

When you delete a model version assigned the default alias, the alias is automatically assigned to the next most recent version.

from typing import List

from google.cloud import aiplatform


def delete_aliases_model_version_sample(
    model_id: str,
    version_aliases: List[str],
    version_id: str,
    project: str,
    location: str,
):
    """
    Delete aliases to a model version.
    Args:
        model_id: The ID of the model.
        version_aliases: The version aliases to assign.
        version_id: The version ID of the model to assign the aliases to.
        project: The project ID.
        location: The region name.
    Returns
        None.
    """
    # Initialize the client.
    aiplatform.init(project=project, location=location)

    # Initialize the Model Registry resource with the ID 'model_id'.The parent_name of Model resource can be also
    # 'projects/<your-project-id>/locations/<your-region>/models/<your-model-id>'
    model_registry = aiplatform.models.ModelRegistry(model=model_id)

    # Remove the version aliases to the model version with the version 'version'.
    model_registry.remove_version_aliases(
        target_aliases=version_aliases, version=version_id
    )