The creators of scikit-learn are GPU-accelerating machine learning — and partnering with NVIDIA to do it. Yann Lechelle, co-founder of Probabl, sat down with us to share why the team behind the world's most downloaded ML framework is doubling down on statistical machine learning — not generative AI — and what that means for the future of tabular data. Scikit-learn has surpassed 5 billion downloads, with 1.3 million GitHub projects depending on it. It's the backbone of data science as we know it. But it was built for CPUs — and the world has changed. Probabl is now offloading key algorithms to GPU, leveraging the Array API as an abstraction layer so that new generations of NVIDIA chips plug in seamlessly, without rewriting everything from scratch. Their new product, Skore, wraps scikit-learn and tabular large models into a unified ML platform — with the goal of becoming the world's leading tabular AI company within five years. And as Yann points out, it's not just data scientists using these tools anymore: agents are now the largest driver of scikit-learn downloads, consuming documentation and frameworks at machine speed. The partnership with NVIDIA Inception is about more than compute — it's about building the abstraction layer that maximizes intelligence per gigawatt, for humans and agents alike. Learn more about how NVIDIA Inception supports startups building the future of AI — and explore what GPU-accelerated ML can do for your workflows at nvidia.com/inception.

 10789      110