New solution makes it easier to manage machine learning models

machine learning AI

More than 80 percent of organizations do not have the necessary visibility and control over their machine learning models or how they're deployed throughout the ML model development lifecycle.

To deal with this problem, Iterative has built an open-source model registry solution that allows teams to easily manage models with full context around model lineage, version, production status, data used to train the model, and more.

The Iterative Studio Model Registry uses a GitOps approach for model lifecycle management, meaning an organization's Git is the single source of truth. Unlike existing solutions that are separate from software development tools and often not updated with the latest model information, Iterative takes the workflows and best practices of software development and applies them to model deployment, thus getting models into production faster. DevOps and MLOps teams can collaborate by using the same tools and processes so production-ready models being passed downstream to CI/CD systems are all fully automated and transparent to all teams.

"DevOps teams already use a GitOps approach to manage the lifecycle and deployment of their business apps and services while ML teams have a different process with custom solutions or model registries that are not based on GitOps. Our model registry builds on GitOps principles and supports the same workflows that DevOps teams use," says Dmitry Petrov, CEO of Iterative. "Iterative's model registry lets software development teams and ML engineers work together using the same tools instead of in silos."

The model registry is made using modular components, allowing team members to use whatever interface they are most comfortable with. They can explore models in a central dashboard that facilitates model discovery across all ML projects, with model history, versions, and stages transparent and accessible across the team.

You can find out more on the Iterative blog.

Image creditJirsak/depositphotos.com

Comments are closed.

© 1998-2024 BetaNews, Inc. All Rights Reserved. Privacy Policy - Cookie Policy.