Canonical Ltd. is pushing further into the machine learning operations arena with the launch of its Charmed MLFlow platform in general availability today.

Charmed MLFlow is Canonical’s distribution of the popular open-source MLFlow platform, which is used to manage the end-to-end machine learning model lifecycle. It benefits from various integrations with Canonical’s software, simpler deployment and management, and regular security patches.

The company says Charmed MLFlow provides four primary functions in the development of machine learning, a subset of artificial intelligence that’s focused on the use of data and algorithms to imitate roughly the way humans learn, gradually improving the accuracy of AI models.

Charmed MLFlow’s first function is to track experiments, record and compare parameters and results. It also helps package machine learning code in a reusable, reproducible form so it can be shared with other data scientists or transferred to production.

In addition, it’s used to manage and deploy models from a variety of machine learning libraries. Finally, it acts as a central model store from which teams can collaboratively manage the full lifecycle of MLFlow models, including steps such as model versioning, stage transitions and annotations.

Canonical said one of the best things about Charmed MLFlow is its ease of deployment. Users can get it up and running on something as small as a laptop in just a few minutes, facilitating rapid experimentation. It’s fully tested on the Ubuntu operating system, but can also be used on other platforms, such as the Windows Subsystem for Linux.

It’s also extremely flexible in that it can run on any environment, public or private cloud, and provides support for multicloud scenarios too, Canonical said. Moreover, it’s compatible with any Cloud Native Computing Foundation-conformant Kubernetes distribution, such as Charmed Kubernetes, MicroK8s or Amazon EKS. Users can move their models from the laptops they design them on to any cloud infrastructure when they’re ready to use more computing power.

Canonical said it has done extensive work to ensure Charmed MLFlow plays nicely with tools such as Jupyter Notebook, Charmed Kubeflow and KServe. Another benefit is its integration with Canonical Observability Stack, which provides infrastructure monitoring capabilities. According to Canonical, when Charmed MLFlow is combined with Charmed Kubeflow, it can tap additional features such as hyper-parameter tuning, graphics processing unit scheduling and model serving.

Of course, Charmed MLFlow is fully supported by Canonical, which can assist in deployment, uptime monitoring, operations and bug fixing, the company said.

Charmed MLFlow becomes the latest addition to Canonical’s growing portfolio of MLOps tools, and is being made available as a part of the Canonical Ubuntu Pro subscription with pricing on a per-node basis.

Cedric Gegout, Canonical’s vice president of product management, said the open-source version of MLFlow is one of the most popular AI frameworks for machine learning development at all stages. “Its popularity arises from its flexibility in facilitating modest local desktop experimentation and extensive cloud deployment, catering to both individual and enterprise needs,” he said. “This makes Charmed MLFlow a fitting addition to our Canonical MLOps suite, offering cost-effective solutions that enable developers to start small and scale up as their business grows.”

Images: Canonical