HuggingFace Models Scraper

Scrape the HuggingFace Hub for AI/ML models and get structured data about trending, most downloaded, and most liked models. Filter by task type (text-generation, image-classification, etc.) or by specific authors/organizations.

Why this Actor?

• Reliable data — Uses the HuggingFace JSON API (/api/models) instead of HTML scraping, so it won't break when the website UI changes
• Rich metadata — Extracts downloads, likes, pipeline tags, languages, licenses, library info, model card availability, and more
• Flexible filtering — Filter by sorting mode, task type, and author in a single run
• Optional model cards — Fetch full README/model card text for each model when you need descriptions

Use Cases

• 📊 Market research — Track trending models and their download counts
• 🏢 Competitive analysis — Monitor models from specific organizations (OpenAI, Meta, Mistral, etc.)
• 🔍 Model discovery — Find models for specific tasks (text-generation, image-classification, etc.)
• 📈 Trend tracking — Schedule regular runs to track model popularity over time

Input

Parameter	Type	Default	Description
mode	string	trending	Sort mode: trending, downloads, or likes
task	string	(all)	Filter by pipeline/task type (e.g. text-generation, image-classification)
author	string	(all)	Filter by author/org (e.g. meta-llama, openai, mistralai)
maxItems	integer	100	Max models to scrape (1–1000)
includeModelCard	boolean	false	Fetch full model card README text for each model

Output

Each dataset item contains:

Field	Type	Description
modelId	string	Full model identifier (e.g. meta-llama/Llama-3.1-8B)
modelName	string	Model name (last part of ID)
author	string	Author/organization
url	string	Link to model page on HuggingFace
downloads	integer	Total downloads
likes	integer	Number of likes
pipelineTag	string	Pipeline/task type
tags	array	All HuggingFace tags
datasetTags	array	Associated datasets
languageTags	array	Supported languages
license	string	License identifier
libraryName	string	ML library (transformers, diffusers, etc.)
lastModified	string	Last modification date
createdAt	string	Creation date
hasModelCard	boolean	Whether a README/model card exists
private	boolean	Whether the model is private
gated	boolean	Whether access requires approval
modelCard	string	(optional) Full README text

Example Output

{
  "modelId": "meta-llama/Meta-Llama-3.1-8B",
  "modelName": "Meta-Llama-3.1-8B",
  "author": "meta-llama",
  "url": "https://huggingface.co/meta-llama/Meta-Llama-3.1-8B",
  "downloads": 8234567,
  "likes": 4521,
  "pipelineTag": "text-generation",
  "tags": ["safetensors", "llama3", "text-generation", "en"],
  "languageTags": ["en"],
  "license": "llama3.1",
  "libraryName": "transformers",
  "lastModified": "2024-07-15T10:30:00.000Z",
  "createdAt": "2024-07-04T00:00:00.000Z",
  "hasModelCard": true,
  "private": false,
  "gated": true
}

Tips

• Combine mode + task to find trending models in a specific category
• Use author to scrape all models from a specific organization
• Set includeModelCard to true when you need model descriptions, but expect slower runs
• Use Apify Scheduler for regular tracking runs to build time-series data

Cost

• ~1 API call per run (no pagination needed for up to 1000 models)
• Minimal compute — just HTTP + JSON parsing
• Optional model card fetches add 1 request per model when enabled