Considerations for upgrading Watson Machine Learning and watsonx.ai

Considerations before you upgrade watsonx.ai

• As part of the upgrade flow, the new deployment image is created first, and then the resources assigned to the original deployment are released. For a non-disruptive upgrade to your custom foundation model deployments, make sure that your cluster has additional GPU capacity to temporarily handle both the upgraded deployment and the original deployment.

  If you don’t have additional GPU capacity, scale down the original deployment to 0 replicas before starting the upgrade to release resources for a non-disruptive upgrade. For more information, see Scaling a deployment.

• If you upgrade from Cloud Pak for Data version 4.8.x to Cloud Pak for Data version 5.0.x, the upgrade process will be disruptive. As part of the upgrade, the runtime assemblies for your custom foundation model deployments on Red Hat OpenShift AI stack are deleted and re-created, therefore the upgrade process for deploying custom foundation model deployments will be disruptive.

• To successfully upgrade the custom foundation model deployments, you must upgrade all installed operators, including Common Core Services, Watson Machine Learning, watsonx.ai, and watsonx.ai inferencing foundation model (watsonxaiifm operator) to the Cloud Pak for Data version to which you are upgrading.

Considerations after you upgrade Watson Machine Learning

After the upgrade:

• All of your Watson Machine Learning assets, including models, data assets, and apps, that you created in the earlier release are available.

• The Deployments dashboard doesn't display run metadata in visualizations. For example, the chart for finished runs doesn't show run type, such as data refinery, in the visualization. For runs created after the upgrade, the labels will display correctly.

• If you have custom images that were created before the 4.0.6 release, you must rebuild them by using the new base images that are published in the current release. For more information, see Customizing deployment runtime images.

• Because of a change to the default software specification used to train AutoAI time series experiments, you must retrain the experiment and then create a new deployment of the resulting model.

• The Watson Machine Learning API is available at new endpoints, as documented in the API documentation.

• You can update the code for your assets. For example, if you encounter this error when you run an existing notebook:

  WMLClientError: 'project' (MetaNames.PROJECT_UID) and 'space' (MetaNames.SPACE_UID) meta names are deprecated and considered as invalid. Instead, use client.set.default_space(<SPACE_GUID>) to set the space or client.set.default_project(<PROJECT_GUID>).
  

  Then, remove client.repository.ModelMetaNames.SPACE_UID:space_id from metadata and use client.set.default_space('<SPACE_GUID>') or client.set.default_project('<PROJECT_GUID>'). For more information, see Watson Machine Learning API documentation and the Python client library documentation.

• Existing deployments based on discontinued frameworks are unavailable.

• You can deploy assets based on deprecated frameworks.

• Platform jobs associated with deployments that become unavailable after the upgrade are retained.

Notes:

• To learn more about deprecated and discontinued frameworks and for information on how to migrate to a supported framework, see Supported frameworks.
• If you use APIs to update Watson Machine Learning assets, make sure to use only Watson Machine Learning API calls.

Parent topic: Assets in deployment spaces