Gemini Enterprise Agent Platform publishes containers and virtual machine images to simplify the
configuration of your machine learning (ML) workloads. These images contain the
operating system, the ML frameworks, drivers, and other libraries. We publish
new versions of images regularly to include new patches, security updates, and
features. Each image provided by Gemini Enterprise Agent Platform provides support for a
specific minor version of an ML framework.

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

For all services, subscribe to the [Gemini Enterprise Agent Platform release notes](https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes.md.txt) page
for announcements about new version releases for your containers, images, and
frameworks.

For the list of supported framework versions, see the [Supported frameworks
list](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/supported-frameworks).

## Shared responsibility

Securing your workloads on Gemini Enterprise Agent Platform is a shared responsibility. While
Gemini Enterprise Agent Platform regularly publishes new versions of images to address
security 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, see [Shared responsibility](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/shared-responsibility.md.txt).

## Support policy for framework versions

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

- Patch updates for supported frameworks. For example, if we support
  TensorFlow 2.7, and TensorFlow releases 2.7.1 to
  address 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 an
  image has Python 3.7 installed and it reaches the end-of-support date, we
  release a new image version. If the change in dependency might be a
  breaking change, we update the [Supported frameworks list](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/supported-frameworks.md.txt)
  to indicate the change in the dependency.

Once published, an image version is immutable and doesn't change. You should
always use the latest image version, as earlier versions might have security
vulnerabilities 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, Gemini Enterprise Agent Platform no
  longer publishes new image versions for that framework version. Existing
  resources that have been deployed to Gemini Enterprise Agent Platform continue to function.
  After this date, we recommend you plan to switch to a more recent framework
  version.

  To receive troubleshooting support from Gemini Enterprise Agent Platform, you might be asked
  to upgrade to a framework version that is within the supported time period.
  - For online predictions using prebuilt containers, in the event of security vulnerabilities, Gemini Enterprise Agent Platform might delete existing endpoints using out-of-support framework versions.
- **End-of-availability (EOA) date:** After this date, you can no longer use
  images for this framework version. Services might block the creation of new
  resources using these images, and the images are no longer available for
  download.

  - **Custom training jobs (`CustomJob`, `TrainingPipeline`):**

    - **Job submission \& execution:** While existing jobs referencing an image past its EOA date might continue to execute, they are unsupported and are not guaranteed to remain compatible with future platform updates.
    - **Maintenance \& vulnerabilities:** No patches, bug fixes, or runtime security updates (including updates to the base operating system or Python environment) are applied to training images after their EOA date.
    - **Successor releases:** An EOA date does not guarantee that a newer prebuilt training image will be published for that framework. If an image reaches EOA without an active or planned successor version (such as scikit-learn and XGBoost prebuilt training images), you must migrate your workloads to a [custom
      container](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/training/create-custom-container) using [`container_spec`](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/training/create-python-pre-built-container.md.txt#migrate-to-custom-container).
  - **Online prediction endpoints:** You can no longer create new endpoints
    using this version after this date, and Agent Platform might
    delete existing endpoints that use this version.

  End-of-availability dates don't exist for pipeline frameworks that are open
  source PyPI packages, such as the Kubeflow Pipelines SDK and TensorFlow Extended SDK.

## Services covered by the framework support policy

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

- [Prebuilt containers for custom training](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/training/pre-built-containers.md.txt)

- [Prebuilt containers for predictions and explanations](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/predictions/pre-built-containers.md.txt)

- [Prebuilt containers for Ray on Agent Platform](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/supported-frameworks.md.txt#ray)

- [Gemini Enterprise Agent Platform Pipelines](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/pipelines/build-pipeline.md.txt#version-pipeline)

  - For the version support schedule for pipeline SDKs, see [Supported
    frameworks for pipelines](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/supported-frameworks#pipelines) in "Supported frameworks list."

For all services, subscribe to the [Gemini Enterprise Agent Platform release notes](https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes.md.txt) page
for announcements about new version releases for your containers, images, and
frameworks.

## What's next

- Review the [list of supported framework versions](https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/supported-frameworks.md.txt).