Weights & Biases Inc., a startup with a platform for building artificial intelligence models, today debuted two new tools designed to make machine learning teams more productive.

San Francisco-based Weights & Biases is backed by $200 million from investors. The startup received a $1 billion valuation after its most recent funding round in late 2021. It counts OpenAI LP, Microsoft Corp. and Stanford University as customers.

Developing an advanced AI model from scratch involves numerous technical tasks. Weights & Biases provides a platform that promises to speed up many of those tasks. Machine learning teams can use the platform to manage their training datasets, test how changes to an AI model affect its performance and troubleshoot technical issues.

The company also offers more specialized capabilities. Its platform includes a set of features for troubleshooting erroneous responses generated by large language models.

The two new tools that the startup debuted today are called W&B Weave and W&B Production Monitoring (pictured). They each focus on easing a different phase of AI development projects.

Weave, the first new tool, allows software teams to visualize technical data about an AI project in graphs. A developer could, for example, create a chart that compares how fast two different versions of the same neural network perform a certain task. It’s also possible to visualize details about training datasets and the other files used in an AI project. 

Visualizing project information helps developers more easily find areas for improvement. Weave automatically identifies the best graph type for each collection of data points a software team wishes to visualize. Users with advanced requirements can optionally customize the default graph settings. 

Weave is rolling out alongside a second new tool called Production Monitoring. It can help developers track the reliability of an AI after it’s deployed in production.

Neural networks often become less accurate over time due to a phenomenon known as data drift. The phenomenon emerges when the information that an AI model processes in production starts to differ significantly from the dataset on which it was trained. When such issues occur, the reliability of the model’s output decreases.

Production Monitoring collects technical data about AI models running in production and turns the information into graphs. Administrators can consult the graphs to identify signs of data drift. For added measure, a built-in alerting system automatically generates notifications when error indicators emerge.

“Monitoring models is essential to helping teams maintain cutting-edge model performance in production, and ML developers need powerful, customizable tools for exploring data and building ML applications,” said Phil Gurbacki, vice president of product at Weights & Biases. “With these two new releases, we are confident of addressing both critical areas.”

Both Weave and Production Monitoring are generally available today.  

Image: Weights & Biases