Based on the open-source MLflow platform, Canonical Charmed MLFlow aims to simplify the task of managing machine learning workflows and artifacts by using alternative packaging system and orchestration engine.
According to Canonical, Charmed MLFlow can be deployed on a laptop within minutes, which makes it ideal for model registry and experimental tracking.
As mentioned, Charmed MLFlow is based on the open-source MLflow platform, which broadly addressed the same set of concerns as Charmed MLFlow.
Both MLflow and Charmed MLFlow target four main tasks: tracking experiments, with the aim to record parameters and results; packaging ML code in a way that makes it easier to share it with other researchers or deploy it to production; managing and deploying models for a variety of ML libraries; and supporting collaboration through versioning, transitions, and annotations. MLflow also supports integration with other AI and big data tools such as Apache Spark and Kubeflow.
On the technical side, Charmed MLFlow provides the same capabilities as MLflow, but uses a different packaging system, based on charms or charmed operators, and relies on the open-source Juju orchestration engine. A charm is a wrapper around an application that includes all the instructions required for deploying, configuring, scaling, integrating, etc., the application on any cloud using the Juju orchestration engine.
In conversation with InfoQ, Canonical spokesperson Andreea Munteanu explained that the adoption of charmed packages and the Juju orchestration engine simplifies deployment and automates activities such as upgrades or updates. In addition, Charmed MLFlow integrates out of the box with other tools, such as Kubeflow, KServe, or Grafana and Prometheus, as part of the Canonical Observability Stack.
Charmed MLFlow belongs to Canonical’s Ubuntu Pro subscription and is priced per node. This implies, says Munteanu, that Charmed MLFlow is a production-grade solution with a ten-year security maintenance commitment, and benefits from enterprise support and, optionally, managed services. A premium support tier is also available.
As a final note, Charmed MLFlow is fully tested on Ubuntu and can be used on other operating systems through Canonical’s Multipass or Windows Subsystem for Linux (WSL).