March 18, 2020
Verta.ai, provider of Verta Enterprise, an open-core end-to-end MLOps platform, is launching ModelDB 2.0, an open-source model versioning system to make machine learning (ML) development and deployment reliable, safe and reproducible, the company says. ModelDB 2.0 provides the ability to track and version the full modeling process including the underlying data and training configurations, ensuring that teams can always go back and recreate a model, whether to remedy a production incident or to answer a regulatory query.
“Documenting and reproducing models is a massive undertaking for our team and actuaries spend weeks answering questions from regulators. As a result, for us and many regulated companies, making models and analyses reproducible is essential in our business,” says Samuel Madden, chief scientist at Cambridge Mobile Telematics.
ModelDB 2.0 reconsiders what a model versioning system should provide and how it should be built. Using constructs from code versioning systems like Git, and adapting them to the special requirements for reproducing ML models, Verta’s ModelDB 2.0 allows for complete governance, audits, version control and collaboration on ML models.
Licensed under Apache V2, ModelDB 2.0 is now generally available, delivering the following new capabilities to the open-source community:
- ability to version the key ingredients of a model including code, data, configuration and environment;
- ability to reproduce any model that has been versioned using the ModelDB protocol;
- integrations into popular ML frameworks such as PyTorch, Tensorflow and scikit-learn; and
- user management with authentication, authorization, organization and teams.
Model versioning provides a robust foundation for downstream MLOps like packaging, deployment, operations and monitoring, and Verta has seen the benefits of grounding Verta Enterprise in a robust model versioning system of ModelDB.
Sources: Press materials received from the company and additional information gleaned from the company’s website.