Scikit-learn creators' startup Probabl brings GPU acceleration to ML

Probabl, founded by the creators of scikit-learn, is working with NVIDIA Inception to move key machine learning algorithms from CPUs to GPUs. Scikit-learn, which has surpassed 5 billion downloads and underpins over a million GitHub projects, was originally built for CPU computing, but demand for faster processing has pushed the team toward GPU acceleration using the Array API as a compatibility layer.

The company's new product, Skore, combines scikit-learn with tabular large models into one platform. Probabl aims to become a leading company in tabular AI within five years. Co-founder Yann Lechelle noted that AI agents, not just human data scientists, are now the biggest users of scikit-learn, pulling documentation and running frameworks automatically.

The partnership with NVIDIA Inception focuses on building infrastructure that maximizes computing efficiency for both human and AI-agent users.