README.md

metadata

title: Structured Docling
emoji: 📄
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.49.1
app_file: app_hf_spaces.py
pinned: false
license: gpl-3.0

Docling Structured Extraction Demo

A Gradio-based demo application for extracting structured data from documents using Docling's beta structured extraction feature.

Features

• 📄 Support for PDF and image files (PNG, JPG, JPEG, TIFF, BMP)
• 🌐 URL input for remote documents
• 🎯 Customizable JSON templates for extraction
• 🚀 Optimized for Hugging Face Spaces with Zero GPU support
• 📊 Clean JSON output with extracted data

Files

• app.py - Standard Gradio application
• app_hf_spaces.py - Version optimized for Hugging Face Spaces with Zero GPU decorator
• requirements.txt - Python dependencies

Installation

pip install -r requirements.txt

Usage

Local Development

Run the standard version:

python app.py

Hugging Face Spaces

The app_hf_spaces.py file is specifically designed for deployment on Hugging Face Spaces with Zero GPU support.

To deploy:

1. Create a new Space on Hugging Face
2. Upload app_hf_spaces.py (rename to app.py)
3. Upload requirements.txt
4. Enable Zero GPU in Space settings

How to Use the Demo

1. Input Source: Either upload a document file or provide a URL to a document
2. Define Template: Create a JSON template specifying the fields you want to extract
   • Use "string" for text fields
   • Use "float" for decimal numbers
   • Use "int" for whole numbers
3. Extract: Click the "Extract" button to process the document
4. View Results: The extracted data will appear in JSON format in the output box

Template Examples

Simple Invoice Extraction

{
  "bill_no": "string",
  "total": "float",
  "date": "string"
}

Detailed Invoice Extraction

{
  "bill_no": "string",
  "total": "float",
  "sender_name": "string",
  "receiver_name": "string",
  "postal_code": "string",
  "city": "string"
}

Notes

• The structured extraction API is currently in beta and may change
• Only PDF and image formats are supported
• The extraction uses Vision Language Models (VLM) for understanding document content
• Processing time depends on document complexity and size

Requirements

• Python 3.9+
• gradio >= 4.0.0
• docling[vlm] >= 2.0.0
• spaces >= 0.19.0 (for Hugging Face Spaces deployment)

License

This demo is provided as-is for demonstration purposes.