How much of the Hugging Face Hub is science, and how fast it is growing

Embed a chart

Each chart has its own page that fits an iframe. Pick a theme to match the page you embed it in; it is written into the code as &theme=light or &theme=dark (Auto follows the reader's light or dark mode). Embedded charts pick up each day's refresh on their own.

How this is counted

• Source: cfahlgren1/hub-stats, a daily snapshot of every public model and dataset. A scheduled Hugging Face Job reruns pipeline/build.py every day at 06:00 UTC and commits data/charts.json to this Space.
• Science repos are found by their tags, science libraries, well-known science model families and science-only organizations, and sorted into 14 domains (sci_rules.py). Tags match exactly, so "science-fiction" or "brainstorm" don't count.
• Medicine (mostly clinical NLP and imaging) and math (mostly LLM math-reasoning data) are counted as science domains like the rest.
• Quantized re-uploads (GGUF, MLX, known re-uploaders) and general chat models that only carry a science tag are left out.
• Only repos that still exist and are public are counted, so older quarters are undercounted relative to recent ones. The newest quarter is still in progress.