DEVOPSdigest

Iterative launched a machine learning model registry based on GitOps principles, Iterative Studio Model Registry

Iterative has built an open-source model registry solution so teams can easily manage models with full context around model lineage, version, production status, data used to train 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, getting models into production faster. DevOps and MLOps teams 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," said 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 with fully modular components. So whether it’s a data scientist who prefers APIs, a manager who prefers a web user interface, or a DevOps engineer who works best with the command line interface (CLI), Iterative Studio Model Registry meets users where they are. This way, team members use the interface that they’re most comfortable with in order to create and collaborate on ML models quickly and seamlessly. And for organizations in general, the model registry and various open-source components that simplify model deployment like MLEM, plug into their existing MLOps stack without any worries around vendor lock-in or compatibility.

Iterative Studio Model Registry gives organizations an interface to not only search and explore models but to manage them, moving various models across the ML lifecycle, from development to production and retirement. With Iterative Studio Model Registry, organizations gain:

- Model organization, access, and collaboration: Explore models in a central dashboard that facilitates model discovery across all your ML projects. Model history, versions, and stages are transparent and accessible across the team.

- Model versioning and lineage: Register and track models and their versions from a GUI. Identify the experiment that produced the model and track how, when and by whom a model version was created. For highly-regulated industries like health or finance, a single place for all information regarding models that teams can easily search and access is an indispensable requirement.

- Model lifecycle management: Manage the lifecycle of each model as it moves through staging, production, and other stages. See at a glance which model versions are in which stage and move easily across stages within the interface.