Joining forces to support the software stack behind open source scientific AI models
Every month, around 200 new open source and open weight models for science are shared on Hugging Face, and science models are downloaded more than 40 million times. From protein structure prediction to molecular dynamics, drug discovery, medical imaging, weather forecasting, and astronomy, each of these models relies on community efforts to make it usable. Researchers curate and share the datasets these models learn from. They build the benchmarks and leaderboards that make progress measurable. And they do the unglamorous work of packaging, documenting, and maintaining, so that a model released by one lab can actually be run by another.
Preparing the data, training a model, finetuning it, evaluating it against a benchmark, and serving it to other researchers: none of this happens without the software stack underneath, and that stack is almost entirely open source.
Hundreds of critical open source software libraries are used routinely to build and run the AI models that are transforming science. And yet identifying them at scale, rewarding the people who maintain them, and sustaining their communities remains a persistent challenge. The software infrastructure layer that the open source AI community depends on is often the layer that gets the least attention, funding, and credit.
Today, we are excited to announce a new collaboration between Hugging Face and the Open Source for Science Fund. A fund of Renaissance Philanthropy, the Open Source for Science Fund is a multi-donor philanthropic effort seeded by Biohub and Wellcome. It pools capital and support across sectors, from philanthropy to the public sector, industry, and research institutions, to sustain and evolve the open source software infrastructure that AI-driven and data-intensive science depends on.
Together, we will work to identify the software libraries that scientific model contributors rely on most, explore concrete opportunities to support the maintainers behind them, and surface the critical bottlenecks facing the open source modeler community.
“ESM-2, a protein language model released in 2022, is still downloaded hundreds of thousands of times a month on Hugging Face. That’s only possible because the software around it keeps being maintained. Every model on the Hub shows which libraries it depends on, and across thousands of science models that adds up to a map of the infrastructure science research relies on. We will use that map to help the Fund find the libraries that matter most and support the people who maintain them!”
–Georgia Channing, AI for Science Lead, Hugging Face
“This collaboration brings together two communities that make open computational science possible: the people building open models and the people maintaining the software underneath them. Making that software layer visible is how we find where support and investments are most needed.”
–Dario Taraborelli, Director, Open Source for Science Fund
Want to learn more about the Open Source for Science Fund and stay in touch for future opportunities? Subscribe to our newsletter for updates and new funding opportunities or follow us on LinkedIn, BlueSky, or X.