Striveworks Introduces Valor, the Open-Source Tool for Evaluating Model Performance

Austin, TX, February 23, 2024 –Striveworks, a leader in machine learning operations, is proud to announce the launch of Valor, an open-source solution that revolutionizes how data scientists and engineers evaluate machine learning models for their use cases.

Valor—a play on the words “evaluation store”—is a state-of-the-art tool designed to save ML teams time and effort in evaluating model performance. It provides answers to three critical questions for any organization using machine learning, and especially those running multiple machine learning pipelines:

Which model performs best on a given dataset?
How does the performance of a single model vary across datasets?
How do fine differences in data segments affect a model’s performance?

By drawing on metadata and evaluation metrics, Valor identifies how models perform on a given dataset. It then rank-orders models based on filter conditions, empowering machine learning teams to select the optimal models for their data pipelines.

“Valor was born out of the necessity to improve and standardize how we evaluate machine learning models,” says Eric Korman, Chief Science Officer at Striveworks. “We wanted a central service to compute, define, store, and discover metrics, keeping track of what exactly went into a model evaluation so we can trust the results.”

Data scientists and machine learning engineers can use Valor, free of charge, to address common challenges with models in production: understanding model performance per dataset, recognizing performance alterations between datasets, and confidently selecting optimal models for new pipelines. Valor makes it easy to discern nuanced differences in model performance, enabling machine learning teams to put the best models for their projects into production.

Valor is also the first major open-source project from Striveworks—an approach the company sees as strengthening the MLOps community as a whole.

“A modern evaluation service has been missing in the open-source MLOps tech stack, and we are excited and hopeful that Valor will fill this gap for the community,” says Korman. “This is just the beginning of model evaluation.”

Valor is now available to evaluate computer vision models for image segmentation, object detection, and image classification, as well as arbitrary models for classification tasks. Striveworks plans to support additional tasks and model types in the future.