Hugging Face Scraper — Models, Datasets & Spaces

Search and export the Hugging Face Hub at scale. Pull models, datasets and Spaces by keyword, task, library, language or author — with downloads, likes, tags, task, files and dates — into clean JSON, CSV or Excel. No API key needed for public repos.

Perfect for model discovery, competitive/landscape research, tracking trending models, building ML dashboards, dataset sourcing and market monitoring.

What you can do

• 🔎 Search models, datasets or Spaces by keyword (llama, whisper, stable diffusion)
• 🏷️ Filter by tag — task (text-generation, text-to-image, automatic-speech-recognition), library (transformers, pytorch), language (en, fr) or license
• 👤 Filter by author/org — meta-llama, google, openai, mistralai…
• 📊 Sort by downloads, likes, trending, recently updated or recently created
• 📥 Bulk export — thousands of rows per run via automatic pagination
• 🔓 No key required for public data (add an optional read token for gated repos you can access)

Example input

{
  "contentType": "models",
  "searches": ["llama"],
  "filters": ["text-generation", "pytorch"],
  "author": "",
  "sort": "downloads",
  "maxResultsPerSearch": 500
}

Leave searches empty to browse the top items by your chosen sort (e.g. the 1,000 most-downloaded text-to-image models).

Output

Each row is one repo:

{
  "type": "model",
  "id": "meta-llama/Llama-3.2-1B-Instruct",
  "author": "meta-llama",
  "name": "Llama-3.2-1B-Instruct",
  "downloads": 10449365,
  "likes": 1548,
  "pipelineTag": "text-generation",
  "library": "transformers",
  "tags": ["transformers", "safetensors", "llama", "conversational", "en", "…"],
  "filesCount": 14,
  "gated": true,
  "createdAt": "2024-09-18T…",
  "lastModified": "2024-10-24T…",
  "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct"
}

Datasets add description; Spaces add sdk, emoji and runtimeStage. The url always links straight to the repo page.

Input fields

Field	Description
contentType	models, datasets or spaces.
searches	Keywords to search. Empty = browse top items by sort.
author	Restrict to one user/org (e.g. google).
filters	Tags to AND together — task, library, language, license.
sort	downloads, likes, lastModified, createdAt or trendingScore.
maxResultsPerSearch	Max rows per search term (up to 50,000).
hfToken	Optional read token for gated/private repos you can access.

Use cases

• Model discovery & benchmarking — find the top models for a task, ranked by real downloads/likes
• Trend monitoring — track what's trending on the Hub over time
• Dataset sourcing — surface the best datasets for a domain (medical, finance, code)
• Competitive research — audit an org's full catalog of models/datasets/Spaces
• ML dashboards & alerts — feed clean Hub metadata into your own tooling

FAQ

Do I need a Hugging Face API key? No. All public models, datasets and Spaces are scraped without a key. Add a read token only if you want gated/private repos you already have access to.

How many results can I get? Up to 50,000 per search term. Results are paginated automatically — leave searches empty and set a high limit to pull the entire top-of-catalog for a task.

Which tasks/filters are supported? Any Hub tag — tasks like text-generation, text-to-image, image-classification, automatic-speech-recognition; libraries like transformers, diffusers, pytorch; languages; and licenses. Combine several to narrow results.

How does billing work? Pay per item scraped plus a small per-request charge — you only pay for what you pull.

Related Actors

• arXiv Papers Scraper — research papers behind the models
• GitHub Repo Scraper — code repositories and stars
• PyPI Scraper — Python packages and metadata
• OpenAlex Scholarly Works Scraper — 250M+ academic works

Notes

Original clean-room implementation using Hugging Face's public Hub API. You are responsible for how you use the scraped data and for complying with the Hub's terms and each repo's license. This is an independent tool and is not affiliated with or endorsed by Hugging Face.