We’re pleased to share that the Charmed Kubeflow 1.11 candidate release is now available. Kubeflow remains a central pillar of Canonical’s MLOps portfolio, keeping up to date with the latest developments across the open-source ecosystem. In addition to incorporating upstream enhancements, this release introduces significant new improvements — most notably support for running all charms and workloads as non-root containers, along with integration of Pod Security Standards to strengthen workload isolation through more restrictive permissions. Additionally, the platform now includes backup and restore capabilities through integration with the Velero charm. This version also introduces a new component, Kubeflow Trainer V2, delivering a more streamlined model training experience for data scientists and ML engineers. Last but not least, we now generally recommend you to deploy Charmed Kubeflow using our Terraform module. Although we are still providing a YAML bundle for this release, be mindful that this is not fully supported anymore as deploying YAML is now being deprecated in the Juju community in favor of Terraform.
See our draft release notes. Please note that they are still subject to change ahead of the stable release.
As always, we invite data scientists, ML engineers, and developers to explore the new candidate release and share valuable feedback with us!
Let’s build it together!
This candidate release gives you early access to cutting-edge ML innovations, along with an opportunity to influence the future of Charmed Kubeflow. Your feedback is crucial to improving Charmed Kubeflow:
-
Follow our step-by-step tutorial to test the new release
-
Reach out in Matrix with your questions about Charmed Kubeflow’s capabilities
-
Report any bugs or issues you encounter
-
Provide suggestions for improving our products
Please note: This candidate release is not intended for production use and may involve risks. We encourage you to save your work regularly and proceed with caution. Canonical’s MLOps team is always ready to assist you with any feedback or issues you might have.
We will allow a few weeks for testing and validation before promoting this release to stable. During this period, artifacts may still change and receive updates.