Skip to content

Model Registry

The Model Registry provides a central catalog for tracking ML model versions, metadata, and artifacts. Use it when you need governance over which models are deployed, version history, and programmatic access to model metadata across your organization.

Dependencies

Requirement Type Path
RHOAI Operator Operator components/operators/rhoai-operator/
DSC modelregistry: Managed DSC component components/instances/rhoai-instance/
External MySQL database (5.x or later, 8.x recommended) External service Provisioned outside the cluster
S3-compatible object storage External service Provisioned outside the cluster

External database and storage required

The official RHOAI 3.3 documentation requires an external MySQL database (version 5.x or later, 8.x recommended) and S3-compatible object storage for Model Registry. These are not provisioned by the RHOAI Operator -- you must set them up before enabling this component. Model Registry does not require GPU infrastructure.

Enable It

Model Registry is enabled in the dev and full overlays.

spec:
  components:
    modelregistry:
      managementState: Managed

Deploy

Model Registry is enabled automatically when the rhoai-instance ArgoCD Application points to the full or dev overlay.

# 1. Install the RHOAI operator
oc apply -k components/operators/rhoai-operator/
oc get csv -A | grep rhods

# 2. Create DSC with model registry enabled
oc apply -k components/instances/rhoai-instance/overlays/dev/

# 3. Wait for DSC
oc wait --for=jsonpath='{.status.conditions[?(@.type=="Ready")].status}'=True \
  datasciencecluster/default-dsc --timeout=600s

Verify

# Model Registry operator pods
oc get pods -n redhat-ods-applications -l app=model-registry-operator

# Check the model registry namespace
oc get pods -n rhoai-model-registries

Configure External Storage

After enabling Model Registry in the DSC, configure the external MySQL database and S3 storage:

  1. MySQL database -- create a MySQL 5.x+ (8.x recommended) instance accessible from the cluster. Note the hostname, port, database name, username, and password.
  2. S3 storage -- configure an S3-compatible bucket for model artifacts. Note the endpoint, bucket name, region, and credentials.
  3. Create a Model Registry instance via the RHOAI Dashboard or CLI, providing the MySQL and S3 connection details.

For detailed configuration steps, see the official RHOAI documentation on creating a model registry.

Usage

  1. Open the RHOAI Dashboard
  2. Navigate to Model Registry
  3. Register models with version metadata, artifact URIs, and custom properties
  4. Deploy registered models directly to KServe or ModelMesh from the UI

Programmatic access

Use the Model Registry REST API or the Python model-registry SDK:

from model_registry import ModelRegistry

registry = ModelRegistry(
    server_address="https://model-registry-route",
    author="data-scientist",
)

model = registry.register_model(
    "my-model",
    uri="s3://bucket/model.onnx",
    version="1.0.0",
    model_format_name="onnx",
)

Disable It

Set modelregistry.managementState to Removed in the DSC.