Hugging Face and five universities want science data ready for AI

Hugging Face has launched Project ProtoStar, an initiative with five US universities, the Astera Institute and the Open Research Community Accelerator (ORCA) to make publicly funded datasets, models and code easier to find and reuse. The first public test is a climate forecasting challenge hosted by Carnegie Mellon, running November 2-13, 2026.

Who is involved and what the first challenge asks

According to Hugging Face, the university side works through the Open Forum for AI, with Carnegie Mellon, Purdue, Georgia Tech, Arizona State and George Washington University taking part. Astera Institute is a partner, Hugging Face provides the infrastructure, and ORCA, through its Fostering Innovation Network, acts as community partner. Organizations join by signing a memorandum of understanding, and those interested can write to [email protected].

The idea is to encourage researchers to publish scientific artifacts in machine-actionable, AI-ready formats, then use open challenges to show what reuse looks like in practice. Hugging Face says the project also aims to surface documentation gaps, barriers and good practices that could inform infrastructure, funding and policy.

The first challenge, called Climate Risk Intelligence, is hosted by Carnegie Mellon’s Open Source Programs Office. Teams get historical weather data for 210 U.S. cities, train on records through 2018 and forecast actual conditions for 2019 to 2026. Four indicators are scored: average annual temperature, total annual precipitation, the hottest day of the year and cooling degree days. Forecasts are compared on a public leaderboard against two simple baselines, historical averages and linear trends.

Teams also propose a short climate adaptation policy idea, which is judged separately from the forecasting. Entries are submitted as a Hugging Face dataset containing the team’s outputs. Final results are due in early 2027, once complete 2026 weather records exist. Hugging Face says later challenges from other institutions will be listed on the Project ProtoStar page.

The usage numbers behind the pitch

Hugging Face frames the project around a gap between availability and usability. It notes that the U.S. federal government spends about $190 billion a year on research and development, that NIH and NSF have tightened public-access requirements, and that Hugging Face is one of ten generalist repositories NIH lists for research data. Public release alone, the post argues, does not make a resource usable when files sit in legacy formats or lack metadata.

The company says it hosts more than 17,000 public datasets in core scientific domains, about 29,000 once medicine and mathematics are counted, plus a comparable number of scientific models. In the first nine months of 2026, researchers shared more than three times as many new science datasets as in the same period of 2025. As of October 26, NCBI’s 18 repositories on the platform had over 11 million downloads, and NASA IMPACT’s 27 had about 770,000.

The post also cites research finding that 90% of dataset downloads on the platform go to fully documented datasets, which is the practical argument for the documentation push. These figures come from Hugging Face itself and were not independently checked.

A policy push that predates the project

ProtoStar fits a line Hugging Face has argued in Washington. In a 2026 response to a Department of Energy request for information tied to the Genesis Mission, the DOE’s AI-for-science effort, the company recommended that every DOE-funded project produce public, curated datasets, and that DOE publish curated data in model-ready formats with provenance and documentation. That document also points to Hugging Science, the company’s community effort, which it says has more than 700 active researchers. I could only read a summary of that response, so those details are as reported there.

Read against that, ProtoStar looks less like a product and more like a demonstration. A leaderboard and a signed memorandum are cheaper to run than a mandate, and they give the company evidence about what documentation and formats make reuse work. There is also an obvious interest to note: Hugging Face is both the host of the challenge and one of the repositories it wants researchers to use, while the post itself acknowledges that domain-specific repositories remain valuable. Carnegie Mellon’s open source office, per the post, encourages sharing suitable outputs on Hugging Face without dropping those.

The post does not say what the memorandum commits partners to, whether funding is attached, or how success will be measured beyond challenge results. Nor does it say whether the climate challenge carries prizes.

Who can take part now

The climate challenge is open through a Hugging Face Space run by the CMU open source office, and results are public. Researchers can also browse a growth explorer called Science on the Hub to see how scientific data on the platform has expanded. Institutions wanting to host a future challenge or sign the memorandum need to use the contact address above. For readers outside the U.S. research system, the practical change is modest for now: one dataset-based competition and a coalition that has so far named U.S. partners only.