Machine learning development startup Weights & Biases Inc., whose software is used by the likes of OpenAI LLC and Nvidia Corp. to develop new artificial intelligence models, announced two major enhancements to its platform today.

Weights and Biases has created a platform for teams to build and collaborate on machine learning models and operations, or MLOps. The platform enables teams to keep track of their machine learning experiments. It also provides tools for evaluating the performance of different machine learning models, dataset versioning and pipeline management.

The W&B platform is designed to improve the efficiency of the trial-and-error process through which AI software is developed. According to the startup, it helps by increasing developer productivity, as it solves one of the main challenges of AI initiatives: organizing and processing project data.

The new additions to its platform were announced at Fully Connected, W&B’s inaugural user conference, and include W&B Launch and W&B Models.

Available in public preview starting today, W&B Launch provides an easier way for users to package code automatically and launch a new machine learning job on any target environment. In this way, machine learning practitioners gain easier access to the compute resources they need, while simplifying infrastructure complexity. In addition, W&B Launch makes it easier for teams to reproduce runs and scale up and scale out those activities.

Orlando Avila-Garcia, AI principal researcher at ARQUIMEA Research Center, said W&B Launch will make it easier for his organization to accelerate research on a variety of deep learning techniques, including neural radiance fields and graph neural networks. “Abstracting away the complexity of using the infrastructure for our researchers is very beneficial to our overall team,” he said. “Launch greatly simplifies our work optimizing, experimenting with and benchmarking ML methods, letting us focus on reliability and reproducibility of results.”

As for W&B Models, this is generally available now and provides a more scalable way for teams to govern machine learning model lifecycles in a centralized repository, while allowing for cross-functional discovery and collaboration. With its reproducibility and lineage tracking functionality that makes it possible for users to track exactly when a model was moved from staging to production, W&B Models can help teams to maintain higher-quality models over time.

Andy Thurai, vice president and principal analyst of Constellation Research Inc., told SiliconANGLE that Weights & Biases is one of the older players in the MLOps segment, and it’s especially strong in model creation, data set versioning and experiment tracking. He added that W&B Models looks like it will be a good addition to the company’s solution set. “It offers ML model registry and model governance, allowing users to collaborate more easily by making models discoverable across large enterprises,” he said. “This, combined with model lineage tracking, enables users to track models in the ML pipeline from inception to production.”

“We’re confident that W&B users will quickly see the benefits of Launch and Models in accelerating model training, utilizing compute resources efficiently, managing models with more confidence, and having a more cohesive end-to-end ML workflow,” said W&B Vice President of Product Phil Gurbacki.

The platform enhancements follow what W&B says has been a very promising 12 months, during which it has enjoyed significant traction and momentum. The company, which last raised $135 million at a $1 billion valuation in October 2021, claims to have doubled its employee base in the last year, opening new offices in Berlin, London and Tokyo to help boost its international presence.

Image: Weights & Biases