Hugging Face is making it easier to cite your work. The platform now lets registered users generate Digital Object Identifiers—DOIs—for models and datasets directly from repository settings. Anyone visiting a model or dataset page can then click “Cite this model/dataset” to get a proper citation.
The company partnered with DataCite, the nonprofit that assigns DOIs to research outputs. Think of a DOI as an ISBN for digital objects. It’s a permanent string that links to metadata: the object’s URL, version, creation date, description. Unlike a regular URL, a DOI never expires or changes. That permanence comes with a catch, though. Hugging Face notes that models and datasets with DOIs are “intended to persist perpetually” and can only be removed by filing a support request. No more casual deletion.
Versioning works the way you’d expect. If you update a model or dataset, the DOI can be refreshed to point to the new version, and the old DOI becomes outdated. That means researchers can cite a specific snapshot of an object even as it evolves—a genuine improvement for reproducibility, which has been a persistent headache in ML research.
This move mirrors what Zenodo and other research repositories have offered for years, but bringing it natively to the Hugging Face Hub matters because that’s where a huge share of open models and datasets actually live. The feature grew out of a GitHub discussion on huggingface/hub-docs, with feedback from community members including Alix Leroy, Bram Vanroy, Daniel van Strien, and Yoshitomo Matsubara. Hugging Face says many features like this come directly from community requests, and they’re inviting more via Twitter or GitHub issues.
