HF Model Tags
Hugging Face Model Tags Every distinct tag on every model on the Hugging Face Hub, deduplicated, with how many models use it. Snapshot of 2026-11-15: 3,239,126 models, 456,200 distinct tags, of which 135,741 are freeform (author-chosen) tags; 73,158 freeform tags are used by exactly one model. Built for Taxonomy of AI: authors' own tags are a crowd-sourced classification of models, noisy but rich, and this makes the whole vocabulary queryable. Configs tags: one… See the full description on the dataset page: https://huggingface.co/datasets/danielrosehill/HF-Model-Tags.
About This Project
HF-Model-Tags provides a processed snapshot of the Hugging Face Hub, cataloging every distinct tag applied to more than 3.2 million models. By aggregating and deduplicating both system-defined and freeform author-chosen tags, it creates a queryable vocabulary that maps how developers classify their own work. The dataset includes frequency counts for each tag, documenting the prevalence of specific labels across the ecosystem.
This project serves as a foundational resource for researchers and developers interested in the taxonomy of AI. By transforming noisy, crowd-sourced metadata into a structured format, it facilitates the study of emerging model trends, naming conventions, and classification patterns. It is available in Parquet format, making it compatible with data analysis libraries such as Pandas, Polars, and MLCroissant.