15 Models

Oracle AI Data Platform enables you to develop and refine machine-learning (ML) models that you can use for running inferences and predictions for your data.

Machine-learning models in AI Data Platform enable you to build optimized models through performance analysis, collaboration, and analysis of experimental conditions like hyper-parameters, input datasets, and feature engineering.

The Model registry in AI Data Platform exists in the Master Catalog and experiment runs developed in your workspace can be selected and registered as models. Models in the registry can be loaded into notebooks to compare performance against models in development or scheduled through workflows to run batch inferences using the loaded model.

You can register models in your AI Data Platform workspace by creating one based on an existing experiment or by uploading a file that contains the necessary information. You can also register a model directly from an experiment in your notebook.

Note:

If you haven't used previously used Experiments or Models in your AI Data Platform, you need to either restart your associated compute cluster or create a new one to use with Experiments and Models.

Once a model is created, you can deploy it for use as an endpoint from the Deployments tab. Once deployed, you can activate the deployment to make it available through the endpoint, deactivate it temporarily, change the model version used by the deployment, run a test query on an active deployment, modify the deployment name, description, and tags, or delete the deployment.

A model requires an attached AI compute to deploy. For more information, see AI Compute.

Note:

Oracle recommends using different AI compute clusters for AI agent workloads and model deployment workloads for best performance. We also recommend using different AI compute clusters for each model deployment.

Limitations

Model deployment is limited to models sized 100 MB or less.

Register a Model from an Existing Experiment Run

You can register a model from an existing experiment run in your Oracle AI Data Platform workspace.

  1. On the home page, click Models.
  2. Click Register.
  3. Enter a name, description, and version for your model.
  4. Select Existing experiment run.
  5. From the Workspace drop-down list, select the workspace where your experiment is located.
  6. From the Experiment drop-down list, select the experiment to use for your model.
  7. From the Experiment run drop-down list, select the experiment run to use.
  8. From the Models drop-down list, select the model to use.
  9. Optional: Provide additional metadata in the form of free-form or defined tags. Click Add to create additional tags.
  10. Optional: Provide additional information in the form of custom fields, like model type or use cases. Click Add to create additional custom fields.
  11. Click Register.

View Model Details

You can see details for your registered models, like versions, metrics, artifacts, and lineages.

  1. On the home page, click Models.
  2. Click the name of the model you want to view details for.
  3. Click the Overview, Details, Versions, or Artifacts tabs to see more details for your model.
  4. Click Versions, then click the Run name to see the run and experiment that produced this model.

Edit Model Details

You can modify the name, description, and metadata tags for your models after creation.

  1. On the home page, click Models.
  2. Next to the model you want to edit, click Edit. You can also click the name of the model and click the Actions menu in the top right, then click Edit.
  3. Modify the model name or description. You can also add and remove tags. Click Save changes.

Delete a Model

You can delete models you no longer need from your workspace.

  1. On the home page, click Models.
  2. Next to the model you want to delete, click Delete. You can also click the name of the model and click the Actions menu in the top right, then click Delete.
  3. Click Delete.

Create a Model Deployment

Creating a model deployment allows you to use the model as an endpoint for agents and other functions.

Note:

Ensure you have an active AI compute before you create a new model deployment. To create a new AI compute, see Create an AI Cluster.
  1. On the home page, click Models.
  2. Click the name of the model you want to create a deployment for.
  3. Click Create deployment.
  4. Provide a name and description for the deployment.
  5. From the Version drop-down menu, select which version of the model to use for this deployment.
  6. Select the workspace to deploy to.
  7. Select an active AI compute to use for this model deployment from the drop-down.
  8. Select the authentication method.
    • For OAuth, you need to provide the audience claim, issuer claim, and URI for the jwks.JSON.
  9. Optional: Provide tags to help other users and agents identify the use of this deployment. Click Add to create multiple tags.
  10. Click Create.

Deploy a Model

Once a model deployment is created you need to deploy it to access it through an endpoint. You can also redeploy a model that was previously deactivated.

  1. On the home page, click Models.
  2. Click the name of the model you want to deploy.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Activate.
  5. Optional: Enter a reason for activating this model deployment.
  6. Click Activate.

Undeploy a Model

You can deactivate model deployments to make their endpoints inaccessible to others. Deactivated model deployments can be reactivated when needed.

  1. On the home page, click Models.
  2. Click the name of the model you want to deactivate the deployment for.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Deactivate.
  5. Optional: Enter a reason for deactivating this model deployment.
  6. Click Deactivate.

Send a Query to a Model Deployment Endpoint

You can test a model deployment by sending a query to the model deployment endpoint and verifying the response.

  1. On the home page, click Models.
  2. Click the name of the model you want to send a query to the deployment endpoint for.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Send query.
  5. Select the query method.
  6. In the Request field, enter the test query.
  7. Click Send request. Verify in the Response field that the request was received and accepted.

Roll Forward a Model Deployment Version

Model deployments versions can be rolled forward to more recent versions as neeeded. Changing the model deployment version does not alter its endpoint address.

  1. On the home page, click Models.
  2. Click the name of the model you want to roll forward the version for.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Roll forward.
  5. From the Target version drop-down menu, select the desired version.
  6. Click Roll forward.

Roll Back a Model Deployment Version

Model deployments versions can be rolled back to older versions as needed. Changing the model deployment version does not alter its endpoint address.

  1. On the home page, click Models.
  2. Click the name of the model you want to roll back the version for.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Rollback.
  5. From the Target version drop-down menu, select the desired version.
  6. Click Rollback.

Edit a Model Deployment

You can modify the name, description, and metadata tags on a model deployment after it has been created.

  1. On the home page, click Models.
  2. Click the name of the model you want to edit.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Edit.
  5. Modify the name, description, and tags.
  6. Click Save changes.

Delete a Model Deployment

You can delete model deployments you own that are no longer needed.

Note:

You can only delete a deactivated model deployment. See Undeploy a Model.
  1. On the home page, click Models.
  2. Click the name of the model you want to delete the deployment for.
  3. Click the Deployments tab.
  4. Click the Actions menu, then click Delete.
  5. Click Delete.

View Model Deployment Activity

You can view a history of activity on a model deployment to see what actions were taken by users and the status of those actions.

  1. On the home page, click Models.
  2. Click the name of the model you want to see the deployment activity for.
  3. Click the Deployments tab.
  4. At the bottom of the page, review the Activity section.
  5. In the Actions column, click View details icon View details to see the specific details for that action.