README.md

metadata

title: AI-Powered PDF Summarizer
emoji: 📚
colorFrom: blue
colorTo: purple
sdk: docker
app_port: 7860
pinned: false

📚 AI-Powered PDF Summarizer

An intelligent PDF summarization tool powered by state-of-the-art Hugging Face transformer models. Upload any PDF document and get a comprehensive, well-structured summary perfect for studying, research, or quick document review.

🌟 Features

🤖 Multiple AI Models

• BART (facebook/bart-large-cnn): Fast, high-quality summarization for general documents
• Long-T5 (google/long-t5-tglobal-base): Optimized for very long documents and academic papers

⚡ Smart Processing

• Intelligent text chunking with overlap for context preservation
• Progress tracking during summarization
• Handles documents of any length
• GPU acceleration support (when available)

📝 Flexible Output

• Choose between bullet points or paragraph format
• Downloadable markdown files
• Statistics about your document
• Clean, readable formatting

🎨 User-Friendly Interface

• Simple drag-and-drop file upload
• Real-time progress updates
• Advanced settings for fine-tuned control
• Beautiful, responsive design

🚀 Quick Start

Local Installation

1. Clone or download this repository

2. Install dependencies:

pip install -r requirements.txt

3. Run the application:

python app.py

4. Open your browser to http://localhost:7860

Hugging Face Spaces Deployment

See the detailed deployment guide below for step-by-step instructions.

📖 How to Use

1. Upload PDF: Click or drag your PDF file to the upload area
2. Select Model: Choose between BART (faster) or Long-T5 (better for long docs)
3. Choose Style: Pick bullet points or paragraph format
4. Adjust Settings (optional): Fine-tune chunk size and summary length
5. Generate: Click the "Generate Summary" button
6. Download: Get your summary as a markdown file

⚙️ Advanced Settings

Chunk Size (1000-8000 words)

• Default: 3000 words
• Smaller chunks: Faster processing, may lose some context
• Larger chunks: Better context, slower processing

Chunk Overlap (0-1000 words)

• Default: 200 words
• Purpose: Maintains context between chunks
• Higher overlap: Better continuity, slightly slower

Summary Length

• Max Length: 50-500 words per section (default: 150)
• Min Length: 10-100 words per section (default: 30)
• Adjust based on how detailed you want the summary

🎯 Best Practices

For Best Results:

• Use clear, text-based PDFs (not scanned images)
• For technical documents: Use Long-T5 model
• For general documents: BART works great
• Large files (100+ pages): Increase chunk size to 4000-5000

Processing Times:

• Short documents (1-10 pages): 10-30 seconds
• Medium documents (10-50 pages): 30-120 seconds
• Large documents (50+ pages): 2-5 minutes

🛠️ Technical Details

Models Used

BART (facebook/bart-large-cnn)

• 406M parameters
• Trained on CNN/DailyMail dataset
• Excellent for news, articles, general documents
• Fast inference time

Long-T5 (google/long-t5-tglobal-base)

• 250M parameters
• Handles inputs up to 16,384 tokens
• Better for academic papers and long-form content
• Slightly slower but more comprehensive

Technologies

• Gradio: Web interface
• Transformers: Hugging Face models
• PyMuPDF (fitz): PDF text extraction
• LangChain: Text splitting and chunking
• PyTorch: Deep learning backend

📊 Example Use Cases

• Students: Summarize textbooks and research papers
• Researchers: Quick overview of academic literature
• Professionals: Digest reports and documentation
• Anyone: Understand long documents quickly

🔒 Privacy & Security

• Documents are processed in real-time
• No permanent storage of uploaded files
• Processing happens on your selected infrastructure
• Temporary files are automatically cleaned up

🐛 Troubleshooting

PDF Upload Failed

• Ensure PDF is not password-protected
• Check file is not corrupted
• Try re-saving the PDF

Summary Quality Issues

• Try the Long-T5 model for better quality
• Adjust chunk size based on document type
• Increase max summary length for more detail

Out of Memory Errors

• Reduce chunk size
• Use CPU instead of GPU (slower but stable)
• Process smaller sections at a time

📝 Requirements

• Python 3.8 or higher
• 4GB+ RAM (8GB+ recommended)
• GPU optional (speeds up processing significantly)

🤝 Contributing

Contributions are welcome! Feel free to:

• Report bugs
• Suggest new features
• Improve documentation
• Submit pull requests

📄 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• Hugging Face for the amazing transformer models
• Facebook AI for BART
• Google Research for Long-T5
• Gradio team for the excellent UI framework

📧 Support

For issues or questions:

• Open an issue on GitHub
• Check existing documentation
• Review the troubleshooting section

---

Made with ❤️ for efficient document summarization

Happy summarizing! 📚✨