Hugging Face Model Explorer — Metadata, Task & Card Parser

Discover and extract structured metadata from the Hugging Face Model Hub — model IDs, authors, total downloads, likes, primary task pipeline tags, libraries, licenses, and parsed model card YAML frontmatter & summaries.

Built for AI engineers, machine learning researchers, procurement/compliance officers, and automated model registry builders.

---

What this Actor does

This Actor queries the official Hugging Face Hub API and parses model card documentation to produce clean, normalized JSON:

• Search & Filter: Search by keyword or filter directly by task (e.g. text-generation, image-to-image, feature-extraction) and library (e.g. transformers, diffusers, vllm, sentence-transformers).
• Popularity & Growth Metrics: Extract historical total downloads and like counts.
• Parsed Model Card Metadata: Parses YAML frontmatter from README.md to extract explicit license definitions, base models, and quantization attributes without returning megabytes of raw markdown.
• Card Summary: Extracts the clean, plain-text summary of the model's purpose and usage instructions.

---

Why use this Actor

• Automated License & Compliance Audits: Screen thousands of models for commercial friendliness (e.g. apache-2.0, mit) versus restrictive licenses (cc-by-nc-4.0, custom community licenses).
• Parsed Metadata vs. Raw Blobs: Other scrapers dump the entire raw README. This Actor structures the critical attributes into queryable keys.
• No Heavy Browser Required: Runs via direct asynchronous HTTP calls to the Hugging Face API, resulting in blazing fast speed and near-zero compute overhead.
• Pay-Per-Event Economy: Only pay for the models you actually extract ($0.0005 per model).

---

Features

• Flexible Sorting: Sort by downloads, likes, trending, created, or lastModified.
• Private/Gated Access Support: Accepts an optional hfToken to access gated models (like Llama or Gemma) tied to your Hugging Face account.
• Pagination & Volume: Extract from 1 up to 500 models per single execution.

---

Use cases

• Model Selection & Benchmarking: Survey available open-source models matching your exact task before training or fine-tuning.
• Corporate AI Registries: Automatically synchronize an internal enterprise catalog of permitted open-source weights.
• Trend Spotting & Newsletter Automation: Monitor the fastest-growing models across categories like speech, vision, and reasoning.
• AI Agent Tool Calling: Equip autonomous agents with a live tool to recommend the best available model for a user's task.

---

Input

Field	Type	Description	Default	Example
query	String	Search term to filter models by name or keyword.	"llama"	"deepseek"
task	String	Pipeline tag filter (e.g. text-generation, image-classification).	""	"text-generation"
library	String	Library filter (e.g. transformers, diffusers, sentence-transformers).	""	"transformers"
sort	String (Select)	Sort criteria: downloads, likes, trending, created, or lastModified.	"downloads"	"downloads"
maxResults	Integer	Maximum number of models to return (1 to 500).	50	100
hfToken	String	Optional Hugging Face User Access Token (for gated models).	""	"hf_..."

Example input

{
  "query": "llama-3",
  "task": "text-generation",
  "library": "transformers",
  "sort": "downloads",
  "maxResults": 50
}

---

Output

Every discovered model outputs a structured JSON item into your dataset.

Field Definitions

• modelId (String): Unique repository identifier (author/model-name).
• author (String): Model creator or organization handle.
• downloads (Integer): Total download count.
• likes (Integer): Heart/like count on Hugging Face.
• task (String | null): Primary pipeline task (e.g. text-generation).
• library (String | null): Compatible library framework.
• license (String | null): Extracted SPDX license identifier.
• lastModified (ISO Date): Last commit or update timestamp.
• tags (Array of Strings): Hub tags including language, framework, and dataset associations.
• cardSummary (String | null): Cleaned 500-character plain text summary of the model card.

Example output

{
  "modelId": "meta-llama/Meta-Llama-3.1-8B-Instruct",
  "author": "meta-llama",
  "downloads": 14200000,
  "likes": 18450,
  "task": "text-generation",
  "library": "transformers",
  "license": "llama3.1",
  "lastModified": "2024-08-20T14:15:32.000Z",
  "tags": [
    "transformers",
    "safetensors",
    "llama",
    "text-generation",
    "en",
    "license:llama3.1"
  ],
  "cardSummary": "The Meta Llama 3.1 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction tuned generative models in 8B, 70B and 405B sizes..."
}

---

Pricing (Pay-Per-Event)

Event Name	Price (USD)	Charged When
actor-start	$0.01	Charged once when run initializes
model-item	$0.0005	Charged per model record successfully exported

Example cost: Extracting 100 structured model profiles costs $0.01 + (100 × $0.0005 = $0.05) = $0.06 total.

---

API usage

Python

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_API_TOKEN")

run = client.actor("lovely_radiologist/hf-model-explorer").call(run_input={
    "task": "text-generation",
    "sort": "trending",
    "maxResults": 20
})

for model in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(f"{model['modelId']} — Downloads: {model['downloads']:,} | License: {model['license']}")

Node.js

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: 'YOUR_APIFY_API_TOKEN' });

const run = await client.actor('lovely_radiologist/hf-model-explorer').call({
    query: 'qwen',
    sort: 'downloads',
    maxResults: 25
});

const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items);

---

Limitations

• Model Card Availability: If a model creator has not populated a README.md or YAML frontmatter on the Hub, cardSummary and license may fall back to tag-derived heuristics.
• Maximum Results: Single runs support up to 500 records to ensure fast execution and avoid platform API timeouts.

---

FAQ

Do I need a Hugging Face API key?

No! Public models are queried directly without any token. You only need to pass hfToken if querying gated models (e.g. official Meta or Mistral weights) that require accepted terms on your personal account.

Can I export results to CSV or Excel?

Yes! In the Apify Console dataset tab, export your results in JSON, CSV, Excel, XML, or JSONL formats with one click.

---

Support & Related Actors

• Hugging Face Dataset Scraper — Extract and inspect datasets, training splits, and column schemas.
• Hugging Face Daily Trends & Leaderboards — Track trending models and Open LLM Leaderboard shifts.
• arXiv Paper Extractor — Extract academic papers introducing these models.