Vertex AI publishes containers and virtual machine images to simplify theconfiguration of your machine learning (ML) workloads. These images contain theoperating system, the ML frameworks, drivers, and other libraries. We publishnew versions of images regularly to include new patches, security updates, andfeatures. Each image provided by Vertex AI provides support for aspecific minor version of an ML framework.

This allows you time to update and test your codewhen moving from one framework version to another. You should always test yourjobs and models thoroughly when switching to a new framework version, regardlessof whether it's a major or minor update.

For all services, subscribe to theVertex AI release notes pagefor announcements about new version releases for your containers, images, andframeworks.

For the list of supported framework versions, see theSupported frameworkslist.

Shared responsibility

Securing your workloads on Vertex AI is a shared responsibility. WhileVertex AI regularly publishes new versions of images to addresssecurity vulnerabilities, you are responsible for tasks such as the following:

• Manually upgrading to the latest version.

• Ensuring that you properly configured your services to use the latest version.

For more information, seeShared responsibility.

Support policy for framework versions

During the supported period for an ML framework version, we publish newimage versions regularly. The updates might include the following:

• Patch updates for supported frameworks. For example, if we supportTensorFlow 2.7, and TensorFlow releases 2.7.1 toaddress bugs, then we release a new image version.

• Security updates for supported frameworks.

• Non-breaking updates to other packages and software installed on the image.

• Updates to dependencies that have reached end-of-support. For example, if animage has Python 3.7 installed and it reaches the end-of-support date, werelease a new image version. If the change in dependency might be abreaking change, we update theSupported frameworks listto indicate the change in the dependency.

Once published, an image version is immutable and doesn't change. You shouldalways use the latest image version, as earlier versions might have securityvulnerabilities or other critical bugs.

Support policy schedule

Support periods for each framework version follows this schedule:

• End-of-patch and support date: After this date, Vertex AI nolonger publishes new image versions for that framework version. Existingresources that have been deployed to Vertex AI continue to function.After this date, we recommend you plan to switch to a more recent frameworkversion.

  To receive troubleshooting support from Vertex AI, you might be askedto upgrade to a framework version that is within the supported time period.

  • For online predictions using prebuilt containers, in the event of securityvulnerabilities, Vertex AI might delete existing endpoints usingout-of-support framework versions.

• End-of-availability date: After this date, you can no longer use imagesfor this framework version. Services might block the creation of new resourcesusing these images, and the images are no longer available for download.

  • For online predictions using prebuilt containers, you can no longer createnew endpoints using this version after this date, and Vertex AImight delete existing endpoints that use this version.

  • End-of-availability dates don't exist for pipeline frameworks that are opensource pypi.org Python libraries, such as the Kubeflow Pipelines SDK andTensorFlow Extended SDK.

Services covered by the framework support policy

If you use the following services, you should be aware of new version releasesthat might require you to update your container images or frameworks:

• Prebuilt containers for custom training

• Prebuilt containers for predictions and explanations

• Prebuilt containers for Ray on Vertex AI

• Vertex AI Pipelines

  • For the version support schedule for pipeline SDKs, seeSupportedframeworks for pipelines in "Supported frameworks list."

• Vertex AI Workbench user-managed notebooks

For all services, subscribe to theVertex AI release notes pagefor announcements about new version releases for your containers, images, andframeworks.

What's next

• Review thelist of supported framework versions.