Alauda AI

Introduction

Alauda AI manages models with two complementary capabilities: Model Catalog and Model Registry. They serve different purposes and should not be confused.

TOC

Model CatalogModel RegistryRecommended workflow

Model Catalog

Model Catalog is a searchable inventory of models provided by configured catalog sources. A source can be a Hugging Face organization, a YAML catalog, or a platform-provided OCI model catalog. Catalog entries are external references; the catalog stores their descriptive information and artifact URI without storing the model files itself.

From User View > AI Hub > Model Catalog, you can:

  • browse models in grid or list view;
  • filter models by source, provider, task type, framework, and labels;
  • open a model's details, including its description, model card, license, and artifact URI;
  • create an inference service directly from a catalog artifact; or
  • register a catalog model and one of its artifacts in a Model Registry.

Administrators configure sources in Admin View > Settings > Model Catalog Sources. The catalog list is read-only for users: changes to a catalog model are made in the upstream source or by registering the model in a Model Registry.

Model Registry

Model Registry is the lifecycle system for models that your team registers and uses in a namespace. It stores model metadata, registered models, model versions, and model artifacts. A model artifact is identified by a Model URI (for example, an S3 URI, a PVC URI, or an OCI image URI); the registry stores metadata and artifact references rather than a model-file working tree.

From User View > AI Hub > Model Registry, you can:

  • register a model and its first version;
  • add additional versions to an existing registered model;
  • edit labels, descriptions, and custom properties;
  • inspect the artifact URI and model framework for a version;
  • create an inference service from an active version; and
  • archive or restore models and versions.

Administrators create the registry backend in Admin View > Settings > Model Registries. A registry uses the selected Storage Class and storage size for its metadata service. The registry may remain in a provisioning state until the underlying deployment and storage are ready.

  1. Prepare the model in a supported storage location, such as S3, a PVC, or an OCI registry. See Model Storage.
  2. If the model comes from a configured source, find it in Model Catalog and review its details. Otherwise, open Model Registry and register it directly.
  3. Select a Model Registry, enter the model name and version, and provide the Model URI.
  4. Verify the model metadata and active version.
  5. Deploy the model from the catalog or from the active Model Registry version.
  6. Archive versions that should no longer be used. A version with a published inference service must be cleaned up before it can be archived.

The following pages describe the catalog, registry, storage, upload, and sharing workflows in detail.