Open Data Hub logo
DOCS
BLOG
COMMUNITY
GITHUB
GET STARTED

Info alert:Important Notice

Please note that more information about the previous v2 releases can be found here. You can use "Find a release" search bar to search for a particular release.

Overview of model registries and model catalog

Overview of model registries and model catalog

A model registry acts as a central repository for administrators and data scientists to register, version, and manage the lifecycle of AI models before configuring them for deployment. A model registry is a key component for AI model governance.

The model catalog provides a curated library where data scientists can discover and evaluate the available generative AI models to find the best fit for their use cases.

Model registry

A model registry is an important component in the lifecycle of an artificial intelligence/machine learning (AI/ML) model, and is a vital part of any machine learning operations (MLOps) platform or workflow. A model registry acts as a central repository, storing metadata related to machine learning models from development to deployment. This metadata ranges from high-level information like the deployment environment and project, to specific details like training hyperparameters, performance metrics, and deployment events.

A model registry acts as a bridge between model experimentation and serving, offering a secure, collaborative metadata store interface for stakeholders in the ML lifecycle. Model registries provide a structured and organized way to store, share, version, deploy, and track models.

Open Data Hub administrators can create model registries in Open Data Hub and grant model registry access to data scientists. For more information, see Creating a model registry and Managing model registry permissions.

Data scientists with access to a model registry can store, share, version, deploy, and track models by using model registry features. For more information, see Working with model registries.

Model catalog

Data scientists can use the model catalog to discover and evaluate the models that are available and ready for their organization to register, deploy, and customize.

The model catalog provides models from different providers that data scientists can search and discover before they register models in a model registry and deploy them to a model serving runtime. Open Data Hub administrators can configure the available repository sources for models displayed in the model catalog.

Open Data Hub provides a default model catalog, which includes models from providers such as Red Hat, IBM, Meta, Nvidia, Mistral AI, and Google. For more information about how data scientists can use the model catalog, see Working with the model catalog.

Enabling the model registry component

As an OpenShift cluster administrator, if the model registry was not enabled in a previous Open Data Hub version, after upgrading, you must enable this component for Open Data Hub administrators and data scientists to use.

Enabling the model registry component

Before data scientists in your organization can work with the model registry and model catalog, you must ensure that the modelregistry component is enabled in Open Data Hub.

Note

The modelregistry component is enabled by default in a new Open Data Hub 2.34 installation. However, if the model registry was not enabled in a previous version of Open Data Hub, you must enable this component after upgrading to Open Data Hub 2.34.
Prerequisites

  • You have cluster administrator privileges for your OpenShift Container Platform cluster.

  • You have access to the data science cluster.

  • You have installed the Open Data Hub Operator on your OpenShift Container Platform cluster.

  • You have sufficient resources. For more information about the minimum resources required to use Open Data Hub, see Installing the Open Data Hub Operator version 2.

Procedure

  1. In the OpenShift Container Platform console, click Operators → Installed Operators.

  2. Search for the Open Data Hub Operator, and then click the Operator name to open the Operator details page.

  3. Click the Data Science Cluster tab.

  4. Click the default instance name (for example, default-dsc) to open the instance details page.

  5. Click the YAML tab to show the instance specifications.

  6. Find the spec.components section, and then add or update it to include the following modelregistry component entry, with the managementState field set to Managed, and the registriesNamespace field set to odh-model-registries:

     modelregistry:
    managementState: Managed
    registriesNamespace: odh-model-registries

  7. Click Save.

Verification

  • Confirm that the model registry namespace was created successfully:

    1. In the OpenShift console, click Home → Projects.

    2. Confirm that the odh-model-registries namespace is displayed in the Projects drop-down list.

  • Check the status of the model-registry-operator-controller-manager pod:

    1. In the OpenShift Container Platform console, from the Project list, select opendatahub.

    2. Click Workloads → Deployments.

    3. Search for the model-registry-operator-controller-manager deployment.

    4. Check the status:

      1. Click the deployment name to open the deployment details page.

      2. Click the Pods tab.

      3. View the pod status.

        When the status of the model-registry-operator-controller-manager-<pod-id> pod is Running, the pod is ready to use.

Managing model registries

As an Open Data Hub administrator, you can create, delete, and manage permissions for model registries in Open Data Hub.

Creating a model registry

You can create a model registry to store, share, version, deploy, and track your models.

Prerequisites

  • You have logged in to Open Data Hub as a user with Open Data Hub administrator privileges.

  • The model registry component is enabled in your Open Data Hub deployment. For more information, see Enabling the model registry component.

  • You have access to an external MySQL database which uses at least MySQL version 5.x. However, Red Hat recommends that you use MySQL version 8.x.

Procedure

  1. From the Open Data Hub dashboard, click Settings → Model registry settings.

  2. Click Create model registry.

    The Create model registry dialog opens.

  3. In the Name field, enter a name for the model registry.

  4. Optional: Click Edit resource name, and then enter a specific resource name for the model registry in the Resource name field. By default, the resource name will match the name of the model registry.

    Important

    Resource names are what your resources are labeled as in OpenShift. Your resource name cannot exceed 253 characters, must consist of lowercase alphanumeric characters or -, and must start and end with an alphanumeric character. Resource names are not editable after creation.

    The resource name must not match the name of any other model registry resource in your OpenShift cluster.

  5. Optional: In the Description field, enter a description for the model registry.

  6. In the Connect to external MySQL database section, enter the information for the external database where your model data is stored.

    1. In the Host field, enter the database hostname.

      1. If the database is running in the odh-model-registries namespace, enter only the hostname for the database.

      2. If the database is running in a different namespace from odh-model-registries, enter the database hostname details in <hostname>.<namespace>.svc.cluster.local format.

    2. In the Port field, enter the port number for the database.

    3. In the Username field, enter the default user name that is connected to the database.

    4. In the Password field, enter the password for the default user account.

    5. In the Database field, enter the database name.

  7. Optional: Select the Add CA certificate to secure database connection to use a certificate with your database connection.

    Important

    If your external database is configured to enforce Transport Layer Security (TLS), then you must add a Certificate Authority (CA) certificate.

    1. Click Use cluster-wide CA bundle to use the ca-bundle.crt bundle in the odh-trusted-ca-bundle ConfigMap.

    2. Click Use Open Data Hub CA bundle to use the odh-ca-bundle.crt bundle in the odh-trusted-ca-bundle ConfigMap.

    3. Click Choose from existing certificates to select an existing certificate. You can select the key of any ConfigMap or secret in the odh-model-registries namespace.

      1. From the Resource list, select a ConfigMap or secret.

      2. From the Key list, select a key.

    4. Click Upload new certificate to upload a new certificate as a ConfigMap.

      1. Drag and drop the PEM file for your certificate into the Certificate field, or click Upload to select a file from your local machine’s file system.

        Note

        Uploading a certificate creates the db-credential ConfigMap with the ca.crt key.

        To upload a certificate as a secret, you must create a secret in the OpenShift Container Platform odh-model-registries namespace, and then select it as an existing certificate when you create your model registry.

        For more information about creating secrets in OpenShift Container Platform, see Providing sensitive data to pods by using secrets.

  8. Click Create.

Note

To find the resource name or type of a model registry, click the help icon Help icon beside the registry name. Resource names and types are used to find your resources in OpenShift.
Verification

  • The new model registry is displayed on the Model registry settings page.

  • You can edit the model registry by clicking the action menu (⋮) beside it, and then clicking Edit model registry.

  • You can register a model with the model registry from the Model registry tab. For more information about working with model registries, see Working with model registries.

Editing a model registry

You can edit the details of existing model registries, such as the model registry name, description, and database connection details.

Prerequisites

  • You have logged in to Open Data Hub as a user with Open Data Hub administrator privileges.

  • The model registry component is enabled in your Open Data Hub deployment. For more information, see Enabling the model registry component.

  • Your Open Data Hub deployment contains at least 1 model registry.

Procedure

  1. From the Open Data Hub dashboard, click Settings → Model registry settings.

  2. Click the action menu (⋮) beside the model registry that you want to edit, and then click Edit model registry.

    The Edit model registry dialog opens.

  3. Optional: In the Name field, edit the name of the model registry.

  4. Optional: In the Description field, edit the description of the model registry.

  5. Optional: In the Connect to external MySQL database section, edit the information for the external database where model data for the registry is stored.

    1. In the Host field, enter the database hostname.

      1. If the database is running in the odh-model-registries namespace, enter only the hostname for the database.

      2. If the database is running in a different namespace from odh-model-registries, enter the database hostname details in <hostname>.<namespace>.svc.cluster.local format.

    2. In the Port field, enter the port number for the database.

    3. In the Username field, enter the default user name that is connected to the database.

    4. In the Password field, enter the password for the default user account.

    5. In the Database field, enter the database name.

  6. Optional: Select the Add CA certificate to secure database connection to use a certificate with your database connection.

    Important

    If your external database is configured to enforce Transport Layer Security (TLS), then you must add a Certificate Authority (CA) certificate.

    1. Click Use cluster-wide CA bundle to use the ca-bundle.crt bundle in the odh-trusted-ca-bundle ConfigMap.

    2. Click Use Open Data Hub CA bundle to use the odh-ca-bundle.crt bundle in the odh-trusted-ca-bundle ConfigMap.

    3. Click Choose from existing certificates to select an existing certificate. You can select the key of any ConfigMap or secret in the odh-model-registries namespace.

      1. From the Resource list, select a ConfigMap or secret.

      2. From the Key list, select a key.

    4. Click Upload new certificate to upload a new certificate as a ConfigMap.

      1. Drag and drop the PEM file for your certificate into the Certificate field, or click Upload to select a file from your local machine’s file system.

        Note

        Uploading a certificate creates the db-credential ConfigMap with the ca.crt key.

        To upload a certificate as a secret, you must create a secret in the OpenShift Container Platform odh-model-registries namespace, and then select it as an existing certificate when you create your model registry.

        For more information about creating secrets in OpenShift Container Platform, see Providing sensitive data to pods by using secrets.

  7. Click Update.

Verification

  • The model registry is displayed with updated details on the Model registry settings page.

Managing model registry permissions

You can manage access to a model registry for individual users and user groups in your organization, and for service accounts in a project.

Note

Open Data Hub creates the <model-registry-name>-users group automatically for use with model registries. You can add users to this group in OpenShift, or ask the cluster administrator to do so.

The model registry operator uses OpenShift Container Platform Role-Based Access Control (RBAC), and creates various RBAC resources in the odh-model-registries namespace.

For each model registry instance, the operator creates a registry-users-<model registry instance name> role and an OpenShift Container Platform group called <model registry instance name>-users. To grant an individual user, service account, or group access to a model registry instance, your cluster administrator must create a role binding to the registry-users-<model registry instance name> role for the instance.

The <model registry instance name>-users group has a role binding to the registry-users-<model registry instance name> role. Your cluster administrator can add users to this group to grant them access to the model registry instance without needing to create a role binding for each user.

For more information about managing RBAC in OpenShift Container Platform, see Using RBAC to define and apply permissions.
Prerequisites

  • You have logged in to Open Data Hub as a user with Open Data Hub administrator privileges.

  • An available model registry exists in your deployment.

  • The users and groups that you want to provide access to already exist in OpenShift Container Platform. For more information, see Managing users and groups.

Procedure

  1. From the Open Data Hub dashboard, click Settings → Model registry settings.

  2. Click Manage permissions beside the model registry that you want to manage access for.

    The permissions page for the model registry opens.

  3. Provide one or more OpenShift groups with access to the project.

    1. On the Users tab, in the Groups section, click Add group.

    2. From the Select a group drop-down list, select a group.

      Note

      To enable access for all cluster users, add system:authenticated to the group list.

    3. To confirm your entry, click Confirm (The Confirm icon).

    4. Optional: To add an additional group, click Add group and repeat the process.

  4. Provide one or more users with access to the model registry.

    1. On the Users tab, in the Users section, click Add user.

    2. In the Type username field, enter the username of the user to whom you want to provide access.

    3. To confirm your entry, click Confirm (The Confirm icon).

    4. Optional: To add an additional user, click Add user and repeat the process.

  5. Provide all service accounts in a project with access to the model registry.

    1. On the Projects tab, in the Projects section, click Add project.

    2. In the Select or enter a project field, select or enter the name of the project to which you want to provide access.

    3. To confirm your entry, click Confirm (The Confirm icon).

    4. Optional: To add an additional project, click Add project and repeat the process.

Verification

  • Users, groups, and accounts that were granted access to a model registry can register, view, edit, version, deploy, delete, archive, and restore models in that registry.

  • The Users and Groups sections on the Permissions tab show the respective users and groups that you granted access to the model registry.

  • The Projects sections on the Projects tab show the projects that you granted access to the model registry.

Deleting a model registry

You can delete model registries that you no longer require.

Important

When you delete a model registry, databases connected to the model registry will not be removed. To remove any remaining databases, contact your cluster administrator.
Prerequisites

  • You have logged in to Open Data Hub as a user with Open Data Hub administrator privileges.

  • An available model registry exists in your deployment.

Procedure

  1. From the Open Data Hub dashboard, click Settings → Model registry settings.

  2. Click the action menu (⋮) beside the model registry that you want to delete.

  3. Click Delete model registry.

  4. In the Delete model registry? dialog that opens, enter the name of the model registry in the text field to confirm that you intend to delete it.

  5. Click Delete model registry.

Verification

  • The model registry is no longer displayed on the Model registry settings page.

Working with model registries

As a data scientist in Open Data Hub, you can store, share, version, deploy, and track models by using the model registry.

Working with model registries

Registering a model from the dashboard

As a data scientist, you can register a model from the Open Data Hub dashboard and create the first version of the new model.

Prerequisites

  • You are logged in to Open Data Hub.

  • You have access to an available model registry in your deployment.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that you want to register a model in.

  3. Click Register model.

    The Register model dialog opens.

  4. In the Model details section, configure details to apply to all versions of the model:

    1. In the Model name field, enter a name for the model.

    2. Optional: In the Model description field, enter a description for the model.

  5. In the Version details section, enter details to apply to the first version of the model:

    1. In the Version name field, enter a name for the model version.

    2. Optional: In the Version description field, enter a description for the first version of the model.

    3. In the Source model format field, enter the name of the model format, for example, ONNX.

    4. In the Source model format version field, enter the version of the model format.

  6. In the Model location section, specify the location of the model by providing either object storage details, or a URI.

    1. To provide object storage details, ensure that the Object storage radio button is selected.

      1. To autofill the details of an existing connection:

        1. Click Autofill from connection.

        2. In the Autofill from connection dialog that opens, from the Project drop-down list, select the data science project that contains the connection.

        3. From the Connection name drop-down list, select the connection that you want to use.

          This list contains only object storage types which contain a bucket.

        4. Click Autofill.

      2. Alternatively, manually fill out your object storage details:

        1. In the Endpoint field, enter the endpoint of your S3-compatible object storage bucket.

        2. In the Bucket field, enter the name of your S3-compatible object storage bucket.

        3. In the Region field, enter the region of your S3-compatible object storage account.

        4. In the Path field, enter a path to a model or folder. This path cannot point to a root folder.

    2. To provide a URI, ensure that the URI radio button is selected.

      1. In the URI field, enter the URI for the model.

        Important

        Deployment of models that are registered by using a URI is currently supported for public OCI repositories only.

  7. Click Register model.

Verification

  • The new model and version details are displayed on the Details tab for the model version.

  • The new model and version are displayed on the Model registry page.

Registering a model version

You can register a new model version.

Prerequisites

  • You have logged in to Open Data Hub.

  • You have access to an available model registry in your deployment.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that you want to register a model version in.

  3. In the Model name column, click the name of the model that you want to register a new version of.

    The details page for the model opens.

  4. Click the Versions tab, and then click Register new version.

  5. In the Version details section, enter details to apply to this version of the model:

    1. In the Version name field, enter a name for the model version.

    2. Optional: In the Version description field, enter a description for this version of the model.

    3. In the Source model format field, enter the name of the model format, for example, ONNX.

    4. In the Source model format version field, enter the version of the model format.

  6. In the Model location section, specify the location of the model by providing either object storage details, or a URI.

    1. To provide object storage details, ensure that the Object storage radio button is selected.

      1. To autofill the details of an existing connection:

        1. Click Autofill from connection.

        2. In the Autofill from connection dialog that opens, from the Project drop-down list, select the data science project that contains the connection.

        3. From the Connection name drop-down list, select the connection that you want to use.

          This list contains only object storage types which contain a bucket.

        4. Click Autofill.

      2. Alternatively, manually fill out your object storage details:

        1. In the Endpoint field, enter the endpoint of your S3-compatible object storage bucket.

        2. In the Bucket field, enter the name of your S3-compatible object storage bucket.

        3. In the Region field, enter the region of your S3-compatible object storage account.

        4. In the Path field, enter a path to a model or folder. This path cannot point to a root folder.

    2. To provide a URI, ensure that the URI radio button is selected.

      1. In the URI field, enter the URI for the model.

        Important

        Deployment of models that are registered by using a URI is currently supported for public OCI repositories only.

  7. Click Register new version.

Verification

  • The new model version is displayed in the Latest versions section on the Overview tab on the model details page.

  • The new model version is displayed in the Latest version column on the Model registry page.

Viewing registered models

You can view the details of models registered in Open Data Hub, such as registered versions, deployments, and metadata associated with the model.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model.

  • You have access to the model registry that contains the model that you want to view.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the model that you want to view.

  3. The Model registry page provides a high-level view of registered models, including the model name, latest version, deployments, labels, last modified timestamp, and owner of each model.

    Models are sorted by their Last modified timestamp by default.

  4. Use the search bar to find a model in the list. You can filter with a keyword by default by entering a model name, description, or label. Alternatively, click the search bar drop-down list and select Owner to filter by entering a model owner.

    Searching by keyword performs a search across the name, description, and labels of registered models and their versions.

  5. Click the name of a model to view the details page for the model:

    1. On the Overview tab, you can view model metadata such as labels, description, owner, model ID, last modified and created timestamps, and custom properties, along with latest versions and deployments.

    2. On the Versions tab, you can view the registered versions of the model.

    3. On the Deployments tab, you can view deployments initiated from the model registry for this model.

Verification

  • You can view information about the selected model on the details page for the model.

Viewing registered model versions

You can view the details of model versions that are registered in Open Data Hub, such as the version metadata and deployment information.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model.

  • You have access to the model registry that contains the model version that you want to view.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the model version that you want to view.

  3. Click the name of a model to view Overview tab on the model details page, which includes the latest model versions and deployments.

  4. On the Versions tab, you can view the registered versions of the model.

    Versions are sorted by their Last modified timestamp by default.

  5. Use the search bar to find a version in the list. You can filter with a keyword by default by entering a model name, description, or label. Alternatively, click the search bar drop-down list and select Author to filter by entering a model owner.

    Searching by keyword performs a search across the name, description, and labels of registered models and their versions.

  6. Click the name of a version to view the version details page.

  7. On the Details tab, you can view the Version details metadata, such as labels, description, custom properties, version ID, author, and last modified and registered timestamps. This also includes where the model is registered from, model location, and model format information.

    You can also click Model details to view non-version metadata, such as labels, description, owner, model ID, last modified and created timestamps, and custom properties.

  8. On the Deployments tab, you can view deployments initiated from the model registry for this version.

    1. Click the name of a deployment to open its metrics page.

      For information about model metrics on the single-model serving platform, see Managing and monitoring models: Monitoring model performance. For information about model metrics on the multi-model serving platform, see Managing and monitoring moodels: Monitoring model performance.

Verification

  • You can view the details of registered model versions on the Model registry page.

Editing model metadata in a model registry

You can edit the metadata of models registered in Open Data Hub, such as the model description, labels, and custom properties. Editing model metadata affects all versions of the model.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model.

  • You have access to the model registry that contains the model that you want to edit.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the model that you want to edit.

  3. The Model registry page provides a high-level view of registered models, including the model name, latest version, deployments, labels, last modified timestamp, and owner of each model.

  4. Click the name of a model to view the model details page.

  5. On the Overview tab, you can edit metadata for the model.

    1. In the Labels section, click Edit to edit the labels of the model, for example, text-to-text.

    2. In the Description section, click Edit to edit the description of the model.

    3. In the Properties section, click Add property to add a new property to the model, for example, Key: license, Value: apache.

      Tip
      		 If you enter any property value as a URL, this is displayed as a clickable link in the Properties section, for example: https://www.apache.org/licenses/LICENSE-2.0.

      1. To edit an existing property, click the action menu (⋮) beside the property, and then click Edit.

      2. To delete a property, click the action menu (⋮) beside the property, and then click Delete.

Verification

  • You can view the updated metadata on the details page for the model.

Editing model version metadata in a model registry

You can edit the metadata of model versions that are registered in Open Data Hub, such as the version’s description, labels, and custom properties. Editing model version metadata affects that model version only.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model.

  • You have access to the model registry that contains the model version that you want to edit.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the model version that you want to edit.

  3. Click the name of a model to view the model details page.

  4. Click the Versions tab to view the available model versions.

  5. Click a version name to view the version details page.

  6. In the Version details section, you can edit the version metadata.

    1. In the Labels section, click Edit to edit the labels of the version, for example, text-to-text.

    2. In the Description section, click Edit to edit the description of the version.

    3. In the Properties section, click Add property to add a new property to the version, for example, Key: license, Value: apache.

      Tip
      		 If you enter any property value as a URL, this is displayed as a clickable link in the Properties section, for example: https://www.apache.org/licenses/LICENSE-2.0.

      1. To edit an existing property, click the action menu (⋮) beside the property, and then click Edit.

      2. To delete a property, click the action menu (⋮) beside the property, and then click Delete.

    4. In the Model format section, click Edit to edit the format of the model version, for example, ONNX.

    5. In the Model format version section, click Edit to edit the format version of the model version.

Verification

  • You can view the updated metadata on the details page for the model version.

Deploying a model version from a model registry

You can deploy a version of a registered model directly from a model registry.

Prerequisites
Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry from which you want to deploy a model version.

  3. In the Model name column, click the name of the model that contains the version that you want to deploy.

    The details page for the model version opens.

  4. Click the action menu (⋮) beside the model version that you want to deploy.

  5. Click Deploy.

  6. In the Deploy model dialog, configure properties for deploying the model.

    1. From the Project drop-down list, select a target project.

    2. Click Deploy.

  7. Configure the following properties for deploying your model:

    1. From the Project drop-down list, select a project in which to deploy your model.

    2. Optional: In the Model deployment name field, enter a unique name for your model deployment. This field is autofilled with a value that contains the model name by default.

      This will be the name of the inference service that is created when the model is deployed.

  8. Configure the remaining properties for deploying your model, as described in Deploying a model by using the multi-model serving platform or Deploying models on the single-model serving platform.

    1. Click Deploy.

Verification

  • The model deployment is displayed on the Models → Model Deployments page.

  • The model deployment is displayed in the Latest deployments section of the model details page.

  • The model version is displayed on the Deployments tab for the model.

  • You can edit the model version deployment by clicking the action menu (⋮) beside it, and then clicking Edit.

  • You can delete the model version deployment by clicking the action menu (⋮) beside it, and then clicking Delete.

Editing the deployment properties of a deployed model version from a model registry

You can edit model version deployment properties from a model registry for models that were deployed from the registry. For example, you can change the deployment name, model framework, and source model location details.

Editing the deployment properties of a model deployed by using the multi-model serving platform

You can edit the deployment properties of a deployed model version from a model registry. For example, you can change the deployment name, model framework, and source model location details.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered and deployed model version.

  • You have access to the model registry that contains the model version deployment that you want to edit.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the model deployment that you want to edit.

  3. In the Model name column, click the name of the model that contains the deployment that you want to edit.

    The details page for the model opens.

  4. Click the name of the model version with the deployment that you want to edit.

  5. Click Deployments

  6. Click the action menu (⋮) beside the model deployment that you want to edit.

  7. Click Edit.

  8. In the Edit model dialog, edit the model deployment properties:

    1. In the Model deployment name field, enter a new, unique name for your model deployment.

    2. From the Model framework list, select a different framework for your model.

      Note
      		 The Model framework list shows only the frameworks that are supported by the model serving runtime that you specified when you configured your model server.

    3. Edit the connection by specifying an existing connection, or by creating a new connection.

    4. Click Redeploy.

Verification

  • The model redeploys and is displayed with updated details on the Deployments tab for the model version.

Editing the deployment properties of a model deployed by using the single-model serving platform

You can edit the deployment properties of a deployed model version from a model registry. For example, you can change the deployment name, model framework, number of model server replicas, model server size, and source model location details.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered and deployed model version.

  • You have access to the model registry that contains the model version deployment that you want to edit.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the model deployment that you want to edit.

  3. In the Model name column, click the name of the model that contains the deployment that you want to edit.

    The details page for the model opens.

  4. Click the name of the model version with the deployment that you want to edit.

  5. Click Deployments

  6. Click the action menu (⋮) beside the model deployment that you want to edit.

  7. Click Edit.

  8. In the Edit model dialog, edit the model deployment properties:

    1. In the Model deployment name field, enter a new, unique name for your model deployment.

    2. From the Model framework list, select a different framework for your model.

      Note
      		 The Model framework list shows only the frameworks that are supported by the model serving runtime that you specified when you deployed your model.

    3. In the Number of model server replicas to deploy field, specify a value.

    4. From the Model server size list, select a value.

    5. In the Model route section, select the Make deployed models available through an external route checkbox to make your deployed models available to external clients.

    6. In the Token authentication section, select the Require token authentication checkbox to require token authentication for your model server. To finish configuring token authentication, perform the following actions:

      1. In the Service account name field, enter a service account name for which the token will be generated. The generated token is created and displayed in the Token secret field when the model server is configured.

      2. To add an additional service account, click Add a service account and enter another service account name.

    7. Edit the connection by specifying an existing connection, or by creating a new connection.

    8. Customize the runtime parameters in the Configuration parameters section:

      1. Modify the values in Additional serving runtime arguments to define how the deployed model behaves.

      2. Modify the values in Additional environment variables to define variables in the model’s environment.

        The Configuration parameters section shows predefined serving runtime parameters, if any are available.

        Note
        		 Do not modify the port or model serving runtime arguments, because they require specific values to be set. Overwriting these parameters can cause the deployment to fail.

    9. Click Redeploy.

Verification

  • The model redeploys and is displayed with updated details on the Deployments tab for the model version.

Deleting a deployed model version from a model registry

You can delete the deployments of model versions from a model registry.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model with a deployed model version.

  • You have access to the model registry that contains the model version deployment that you want to delete.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that contains the deployment that you want to delete.

  3. Click the name of a model to view more details.

    The details page for the model opens.

  4. Click the name of the model version with the deployment that you want to delete.

    The details page for the model version opens.

  5. Click Deployments.

  6. To delete a deployment, click the action menu (⋮) beside the deployment, and then click Delete.

    The Delete deployed model? dialog opens.

  7. Enter the name of the model deployment in the text field to confirm that you intend to delete it.

  8. Click Delete deployed model.

Verification

  • The model deployment is no longer displayed on the Deployments tab for the model version.

Archiving a model

You can archive a model that you no longer require. The model and all of its versions will be archived and unavailable for use unless it is restored.

Important

Models with deployed versions cannot be archived. To archive a model, you must first delete all deployments of its registered versions from the Models → Model deployments page.
Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model.

  • You have access to the model registry that contains the model that you want to archive.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that you want to archive a model in.

  3. Click the action menu (⋮) beside the model that you want to archive.

  4. Click Archive model.

  5. In the Archive model? dialog that is displayed, enter the name of the model in the text field to confirm that you intend to archive it.

  6. Click Archive.

Verification

  • The model is no longer visible on the Model registry page.

  • The model is displayed on the archived models page for the model registry.

Archiving a model version

You can archive a model version that you no longer require. The model version will be archived and unavailable for use unless it is restored.

Important

Deployed model versions cannot be archived. To archive a model version, you must first delete all deployments of the version from the Models → Model deployments page.
Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least 1 registered model.

  • You have access to the model registry that contains the model version that you want to archive.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that you want to archive a model in.

  3. In the Model name column, click the name of the model that contains the version that you want to archive.

    The details page for the model version opens.

  4. Click the action menu (⋮) beside the version that you want to archive.

  5. Click Archive model version.

  6. In the Archive version? dialog that opens, enter the name of the model version in the text field to confirm that you intend to archive it.

  7. Click Archive.

Verification

  • The model version is no longer visible on the details page for the model.

  • The model version is displayed on the archived versions page for the model.

Restoring a model

You can restore an archived model. The model and all of its versions will be restored and returned to the registered models list.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least one archived model.

  • You have access to the model registry that contains the model that you want to restore.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that you want to restore a model in.

  3. Click the action menu (⋮) beside the Register model button, and then click View archived models.

    The archived models page for the model registry opens.

  4. Click the action menu (⋮) beside the model that you want to restore.

  5. Click Restore model.

  6. In the Restore model? dialog that is displayed, click Restore.

Verification

  • The model is displayed on the Model registry page.

  • The model is no longer displayed on the archived models page for the model registry.

Restoring a model version

You can restore an archived model version. The model version will be restored and returned to the versions list for the model.

Prerequisites

  • You have logged in to Open Data Hub.

  • An available model registry exists in your deployment, and contains at least one archived model version.

  • You have access to the model registry that contains the model version that you want to restore.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model registry.

  2. From the Model registry drop-down list, select the model registry that you want to restore a model version in.

  3. In the Model name column, click the name of the model that contains the version that you want to restore.

    The details page for the model version opens.

  4. Click the action menu (⋮) beside the Register new version button, and then click View archived versions.

    The archived versions page for the model opens.

  5. Click the action menu (⋮) beside the version that you want to restore.

  6. Click Restore version.

  7. In the Restore version? dialog that opens, click Restore.

    The details page for the version opens.

Verification

  • The model version is displayed on the details page for the model.

  • The model is no longer displayed on the archived versions page for the model.

Working with the model catalog

As a data scientist in Open Data Hub, you can discover and evaluate the generative AI models that are available in the model catalog. From the model catalog, you can select models that you want to register, deploy, and customize.

Working with the model catalog

Viewing models in the model catalog

You can discover and evaluate the available generative AI models in the model catalog to find the best fit for your use cases.

Prerequisites
Procedure

  1. From the Open Data Hub dashboard, click Models → Model catalog.

  2. The Model Catalog page provides a high-level view of available models, including the model name, description, and labels such as task, license, and provider.

  3. In the drop-down list, select from the available catalog sources that have been configured by your administrator. The Default Catalog is displayed by default.

    Note
    		 OpenShift cluster administrators can configure additional model catalog sources. For more details, see the Kubeflow Model Registry community documentation on configuring catalog sources.

  4. Use the search bar to find a model in the catalog. You can enter text to search by model name, description, or provider.

  5. Click the name of a model to view the model details page. This page displays the model description and the Model card information supplied by the model provider. This includes details such as the model’s intended use and potential limitations, training parameters and datasets, and evaluation results.

  6. You can click Load more models to scroll and view additional models available in the catalog. Repeat this step until all models are loaded.

Verification

  • You can view the information about a selected model on the model details page.

Registering a model from the model catalog

As a data scientist, you can register models directly from the model catalog and create the first version of the new model.

Prerequisites

  • You are logged in to Open Data Hub.

  • You have access to an available model registry in your deployment.

Procedure

  1. From the Open Data Hub dashboard, click Models → Model catalog.

  2. In the drop-down list, select from the available catalog sources that have been configured by your administrator. The Default Catalog is displayed by default.

    Note
    		 OpenShift cluster administrators can configure additional model catalog sources. For more details, see the Kubeflow Model Registry community documentation on configuring catalog sources.

  3. Use the search bar to find a model in the catalog. You can enter text to search by model name, description, or provider.

  4. Click the name of a model to view the model details page.

  5. Click Register model.

  6. From the Model registry drop-down list, select the model registry that you want to register the model in.

  7. In the Model details section, configure details to apply to all versions of the model:

    1. Optional: In the Model name field, update the name of the model.

    2. Optional: In the Model description field, update the description of the model.

  8. In the Version details section, enter details to apply to the first version of the model:

    1. In the Version name field, enter a name for the model version.

    2. Optional: In the Version description field, enter a description for the first version of the model.

    3. In the Source model format field, enter the name of the model format, for example, ONNX.

    4. In the Source model format version field, enter the version of the model format.

  9. In the Model location section, the URI of the model is displayed.

  10. Click Register model.

Verification

  • The new model details and version are displayed on the Overview tab on the model details page.

  • The new model and version are displayed on the Model registry page.

Deploying a model from the model catalog

You can deploy models directly from the model catalog.

Note

Open Data Hub model serving deployments use the global cluster pull secret to pull models in ModelCar format from the catalog.

For more information about using pull secrets in OpenShift Container Platform, see Updating the global cluster pull secret in the OpenShift Container Platform documentation.
Prerequisites
Procedure

  1. From the Open Data Hub dashboard, click Models → Model catalog.

  2. In the drop-down list, select from the available catalog sources that have been configured by your administrator. The Default Catalog is displayed by default.

    Note
    		 OpenShift cluster administrators can configure additional model catalog sources. For more details, see the Kubeflow Model Registry community documentation on configuring catalog sources.

  3. Use the search bar to find a model in the catalog. You can enter text to search by model name, description, or provider.

  4. Click the name of a model to view the model details page.

  5. Click Deploy model to display the Deploy model dialog.

  6. From the Project drop-down list, select a project in which to deploy your model.

    Note

    Models using OCI storage can only be deployed on the single-model serving platform. Projects using the multi-model serving platform do not appear in the project list.

  7. In the Model deployment section:

    1. Optional: In the Model deployment name field, enter a unique name for your model deployment. This field is autofilled with a value that contains the model name by default.

      This is the name of the inference service created when the model is deployed.

    2. Optional: Click Edit resource name, and then enter a specific resource name for the model deployment in the Resource name field. By default, the resource name matches the name of the model deployment.

      Important

      Resource names are what your resources are labeled as in OpenShift. Your resource name cannot exceed 253 characters, must consist of lowercase alphanumeric characters or -, and must start and end with an alphanumeric character. Resource names are not editable after creation.

      The resource name must not match the name of any other model deployment resource in your OpenShift Container Platform cluster.

    3. From the Serving runtime list, select a model-serving runtime that is installed and enabled in your Open Data Hub deployment. If project-scoped runtimes exist, the Serving runtime list includes subheadings to distinguish between global runtimes and project-scoped runtimes.

    4. From the Model framework list, select a framework for your model.

      Note
      		 The Model framework list shows only the frameworks that are supported by the model-serving runtime that you specified when you deployed your model.

  8. From the Deployment mode list, select KServe RawDeployment or Knative Serverless. For more information about deployment modes, see About KServe deployment modes.

    1. In the Number of model server replicas to deploy field, specify a value.

    2. From the Model server size list, select a value.

    3. If you have created a hardware profile, select a hardware profile from the Hardware profile list. If project-scoped hardware profiles exist, the Hardware profile list includes subheadings to distinguish between global hardware profiles and project-scoped hardware profiles.

      Important

      By default, hardware profiles are hidden from appearing in the dashboard navigation menu and user interface. In addition, user interface components associated with the deprecated accelerator profiles functionality are still displayed. To show the Settings → Hardware profiles option in the dashboard navigation menu and the user interface components associated with hardware profiles, set the disableHardwareProfiles value to false in the OdhDashboardConfig custom resource (CR) in OpenShift Container Platform. For more information about setting dashboard configuration options, see Customizing the dashboard.

    4. In the Model route section, select the Make deployed models available through an external route checkbox to make your deployed models available to external clients.

    5. In the Token authentication section, select the Require token authentication checkbox to require token authentication for your model server. To finish configuring token authentication, perform the following actions:

      1. In the Service account name field, enter a service account name for which the token will be generated. The generated token is created and displayed in the Token secret field when the model server is configured.

      2. To add an additional service account, click Add a service account and enter another service account name.

  9. In the Source model location section, select Current URI to deploy the selected model from the catalog.

  10. Optional: Customize the runtime parameters in the Configuration parameters section:

    1. Modify the values in Additional serving runtime arguments to define how the deployed model behaves.

    2. Modify the values in Additional environment variables to define variables in the model’s environment.

  11. Click Deploy.

Verification

  • The model deployment is displayed on the Models → Model Deployments page.

  • The model deployment is displayed in the Latest deployments section of the model details page.

  • The model deployment is displayed on the Deployments tab for the model version.

Additional resources