#Model Catalog and Model Registry

Alauda AI provides two model-management experiences in AI Hub:

A model is referenced by a Model URI, while metadata and lifecycle state are managed by the catalog or registry APIs.

#Model Catalog

#Browse catalog models

1. Switch to User View and select a namespace.
2. In the left navigation, expand AI Hub and select Model Catalog.
3. Optionally select a catalog Source and filter by task type, provider, framework, or labels.
4. Select a model card or model name to open its details.
5. Review the model description, model card, task type, labels, framework, tensor type, size, license, and Model URI.

A catalog entry is read-only in the user view. It is an external reference, so there is no local file upload or delete operation for the entry.

#Deploy or register a catalog model

On a catalog model's details page:

• Select Create Inference API to open the inference-service form with the catalog artifact preselected; or
• Select Register Model to copy the catalog model into a Model Registry. Choose a registry, review the model name and description, accept or edit the suggested version name, and confirm the artifact URI. The catalog metadata such as task and framework may be copied to the registered model or version.

Registering a catalog model creates a registered model, a model version, and a model artifact in the selected registry. The artifact continues to be addressed by its URI.

#Configure catalog sources (administrator)

An administrator can configure sources in Admin View > Settings > Model Catalog Sources.

1. Select Add Source.
2. Enter a source name. Use lowercase letters, numbers, and underscores; do not use hyphens.
3. Select a source type:
   • Hugging Face: provide the required Organization and, when needed, an Access Token.
   • YAML: provide the YAML catalog content.
4. Optionally set Display Name, Allowed Models, Excluded Models, and Enabled.
5. Select Add. The source status indicates whether it is available, disabled, or has an error.
6. Edit or delete a source from its row action menu when needed.

A source configuration controls what appears in Model Catalog. It does not create a private registered model or upload model files.

#Model Registry

#Create a registry (administrator)

Before users can register models, an administrator must create a backend in Admin View > Settings > Model Registries:

1. Select Create Model Registry.
2. Enter a registry name and, optionally, a display name.
3. Select a Storage Class.
4. Select an Access Mode (the default is typically ReadWriteOnce).
5. Set the metadata storage size in GiB.
6. Select Create.

The registry list shows the registry status, Storage Class, storage size, and creation time. Creation provisions the registry's underlying deployment and persistent storage, so the status can initially be Deploying or otherwise not yet available. The row action menu also provides Manage permissions, Edit, and Delete.

#Register a model and version

1. Switch to User View, select a namespace, and open AI Hub > Model Registry.
2. Select Register Model.
3. Choose a Model Registry. If the form has already selected the first available registry, verify that it is the intended one.
4. Enter the model Name and optional Description.
5. Enter a Version name. When the model name already exists, the console suggests the next available name such as Version 2; review it before submitting.
6. Enter a Model URI. The URI must include a scheme, for example: s3://bucket/path, pvc://claim/path, or oci://registry.example.com/project/model:tag.
7. Select Register.

The console creates or reuses the registered model and adds the submitted version and model artifact. The model details page contains Overview and Versions tabs. Overview shows the description, owner, ID, labels, custom properties, latest versions, and related inference services. The version detail page shows the version state, artifact URI, framework, timestamps, and related inference services.

#Add a version

To add a new artifact to an existing registered model:

1. Open the registered model details page.
2. Select Versions.
3. Select Register New Version.
4. Enter the version name, optional description, Model URI, and model framework.
5. Select Register New Version.

The version remains LIVE until it is archived. Use the version's action menu to restore an archived version when it is needed again.

#Update metadata

On a registered model or version details page, users with edit permission can update labels, descriptions, and custom properties. Metadata is stored in the Model Registry and is independent of any upstream catalog source. Archived models and versions cannot be edited until they are restored.

#Archive and restore

Archiving is the lifecycle operation used instead of deleting a registered model or version:

1. Open the model or version action menu.
2. Select Archive.
3. Type the exact model or version name in the confirmation field.
4. Select Archive.
5. To use it again, switch to the archived view and select Restore.

A registered model archive also archives all of its versions. A model or version that has a published inference service cannot be archived; delete the related inference service first. Archived entries remain in the registry and can be viewed or restored.

#Deploy from model management

Both model-management views can start an inference deployment:

• Model Catalog: open a catalog entry and select Create Inference API.
• Model Registry: open an active model version and select Create Inference API, or select Deploy beside a latest version.

The console passes the selected artifact's Model URI, model name, version, task, and framework to the inference-service form. Choose an appropriate serving runtime and deployment resources there. For storage URI formats and direct storage workflows, see Model Storage.