Abstract
As there are now millions of publicly available neural networks, searchingand analyzing large model repositories becomes increasingly important.Navigating so many models requires an atlas, but as most models are poorlydocumented charting such an atlas is challenging. To explore the hiddenpotential of model repositories, we chart a preliminary atlas representing thedocumented fraction of Hugging Face. It provides stunning visualizations of themodel landscape and evolution. We demonstrate several applications of thisatlas including predicting model attributes (e.g., accuracy), and analyzingtrends in computer vision models. However, as the current atlas remainsincomplete, we propose a method for charting undocumented regions.Specifically, we identify high-confidence structural priors based on dominantreal-world model training practices. Leveraging these priors, our approachenables accurate mapping of previously undocumented areas of the atlas. Wepublicly release our datasets, code, and interactive atlas.