Kubeflow is an open-source MLOps platform that runs on top of Kubernetes. Kubeflow 1.6 was released September 7 2022 with Canonical’s official distribution, Charmed Kubeflow, following shortly after. It came with support for Kubernetes 1.22.
However, the MLOps landscape evolves quickly and so does Charmed Kubeflow. As of today, Canonical supports the deployment of Charmed Kubeflow 1.6 on Charmed Kubernetes 1.23 and 1.24. This is essential as Kubernetes 1.22 is not maintained anymore, following the latest release of Kubernetes 1.25.
Kubeflow 1.6 for optimised advanced training
Kubeflow 1.6 came with new enhancements that focused on complex optimised model training. To be precise, the latest version focused on the stable version of the Kubeflow pipelines. They offer a better user experience through the stable version (KFP v2). Metadata is securely captured and recorded using the pipeline execution cache.
Hyperparameter is also enabled with the latest version of Kubeflow. Training operators are the champions here. They combine population-based training (PBT) with various AI frameworks such as Tensorflow or PyTorch.
The latest version of Kubeflow also makes data processing more seamless by providing better tracking capabilities. Trial logs are efficiently recorded and ML models are better measured. This makes evolution and debugging simpler. Preventing data drift is now possible, with the ability to detect data source failure.
Kubeflow and the Kubernetes lifecycle
Kubernetes’ lifecycle supports the latest three minor releases, based on the official guidelines. Canonical’s official distribution, Charmed Kubernetes, follows the same baseline. As an extra step, Canonical offers expanded security maintenance for the two older versions. Each version of Kubernetes reaches its end of life after approximately 10 months. They are always announced when a new version is released.
Kubeflow 1.6 on Kubernetes 1.23 and beyond
Canonical just finished the testing of Charmed Kubeflow 1.6 on two of the maintained versions of Charmed Kubernetes. It enables users to save time and continue using their Kubernetes version of choice when deploying the MLOps platform. Kubeflow has the same functionalities and features on all announced versions. It benefits from the new enhancements of Kubernetes.
From an enterprise perspective, this announcement is much more important. It allows the MLOps platform and orchestration tool to run in tandem and avoid security issues. It enables data scientists and machine learning engineers to focus on ML models, rather than infrastructure maintenance.
Currently, Canonical is working on supporting Charmed Kubeflow on the latest version of Kubernetes. It will be announced once the testing phase is completed and the application runs smoothly, and at maximum performance.
Learn more about Charmed Kubeflow
- What is Kubeflow?
- Charmed Kubeflow versioning information
- FAQ: MLOps with Charmed Kubeflow
- Charmed Kubeflow 1.6: what’s new?
- Hyperparamter tuning with an MLOps platform
With Charmed Kubeflow, deployment and operations of Kubeflow are easy for any scenario.
Charmed Kubeflow is a collection of Python operators that define integration of the apps inside Kubeflow, like katib or pipelines-ui.
Use Kubeflow on-prem, desktop, edge, public cloud and multi-cloud.
Kubeflow makes deployments of Machine Learning workflows on Kubernetes simple, portable and scalable.
Kubeflow is the machine learning toolkit for Kubernetes. It extends Kubernetes ability to run independent and configurable steps, with machine learning specific frameworks and libraries.
The Kubeflow project is dedicated to making deployments of machine learning workflows on Kubernetes simple,
portable and scalable.
You can install Kubeflow on your workstation, local server or public cloud VM. It is easy to install with MicroK8s on any of these environments and can be scaled to high-availability.