Skip to content

README_streamlit.md

Latest commit

 

History

History
128 lines (95 loc) · 3.74 KB

README_streamlit.md

File metadata and controls

128 lines (95 loc) · 3.74 KB

RAG QA System - Streamlit Interface

A beautiful web interface for the RAG (Retrieval-Augmented Generation) Question Answering system built with Streamlit.

Features

  • 🤖 Interactive Web Interface: Modern, responsive design
  • 📊 Confidence Scores: Visual confidence indicators with color coding
  • ⚙️ Configurable Parameters: Adjust model settings in real-time
  • 📁 Data Upload: Upload custom JSON files with Q&A pairs
  • 🎯 Real-time Processing: Instant answers with processing time display
  • 📈 Statistics Dashboard: System metrics and recent questions

Installation

  1. Install dependencies:

    pip install -r requirements_streamlit.txt

  2. Run the application:

    streamlit run streamlit_app.py

  3. Open your browser: The app will automatically open at http://localhost:8501

Usage

Basic Usage

  1. The app loads with sample data by default
  2. Type your question in the text area
  3. Click "Get Answer" to receive a response with confidence score
  4. Adjust parameters in the sidebar as needed

Custom Data

  1. Prepare a JSON file with Q&A pairs in this format:

    [
  {
    "question": "Your question here?",
    "answer": "The corresponding answer here."
  }
]

  2. Select "Upload Custom Data" in the sidebar

  3. Upload your JSON file

  4. Start asking questions!

Configuration Options

Model Parameters:

  • Max Answer Length: Control the length of generated answers (20-100 tokens)
  • Number of Beams: Adjust beam search for better quality (1-8 beams)
  • Confidence Threshold: Set minimum confidence for reliable answers (0-100%)

Device Selection:

  • Auto-detect: Automatically choose GPU if available
  • CPU: Force CPU usage
  • GPU: Force GPU usage (if available)

File Structure

rag_project/
├── streamlit_app.py          # Main Streamlit application
├── requirements_streamlit.txt # Dependencies for Streamlit
├── sample_data.json          # Sample Q&A data
├── README_streamlit.md       # This file
└── rag_project/
    └── qa_system.py          # Core QA system

Confidence Scoring

The system provides confidence scores based on semantic similarity:

  • 🟢 High Confidence (70%+): Green indicator - reliable answer
  • 🟡 Medium Confidence (40-69%): Yellow indicator - moderate reliability
  • 🔴 Low Confidence (<40%): Red indicator - low reliability

Troubleshooting

Common Issues

  1. CUDA/GPU Issues:

    • If you encounter GPU errors, try selecting "CPU" in device options
    • Ensure you have compatible CUDA drivers installed

  2. Memory Issues:

    • Reduce "Max Answer Length" parameter
    • Use smaller datasets
    • Close other applications using GPU memory

  3. Model Loading Issues:

    • Check internet connection (models are downloaded on first run)
    • Ensure sufficient disk space for model downloads

Performance Tips

  • GPU Usage: Enable GPU for faster processing
  • Batch Processing: Process multiple questions in sequence
  • Model Caching: The app caches the model to avoid reloading

Customization

Styling

Modify the CSS in the st.markdown() section to customize the appearance.

Adding Features

  • Extend the get_answer_with_confidence() function for additional metrics
  • Add more configuration options in the sidebar
  • Implement session state for conversation history

Dependencies

  • Streamlit: Web framework
  • PyTorch: Deep learning framework
  • Transformers: Hugging Face models
  • Sentence-Transformers: Embedding models
  • Scikit-learn: Machine learning utilities
  • NumPy/Pandas: Data processing

License

This project is part of the RAG QA System. See the main project license for details.