metadata
title: BioStack - RLHF Medical Report Generation
emoji: 🩻
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
BioStack - RLHF Medical Report Generation
AI-powered medical report generation using Reinforcement Learning from Human Feedback (RLHF).
Features
- SFT Model: Supervised Fine-Tuning for initial report generation
- Reward Model: Quality assessment of generated reports
- PPO Model: Policy optimization for improved outputs
- Modern UI: Clean React interface with drag-and-drop upload
Models Used
- SFT Model:
best_model.pt- Initial report generation - Reward Model:
reward_model.pt- Quality scoring - PPO Model:
rlhf_model.pt- Optimized generation
All models are automatically downloaded from Hugging Face Hub on first startup.
Usage
- Upload a medical X-ray image
- Optionally provide ground truth text for comparison
- Click "Run Inference" to generate reports
- View results from all three models with quality scores
Technology Stack
- Frontend: React.js
- Backend: FastAPI (Python)
- ML Framework: PyTorch, Transformers
- Models: T5, CoAtNet, Custom Reward Model
Local Development
Prerequisites
- Python 3.11+
- Node.js 16+
- Hugging Face account with access tokens
Installation
# Install Python dependencies
pip install -r requirements.txt
# Install Node dependencies
npm install
# Build React app
npm run build
# Run server
python server.py
The app will be available at http://localhost:7860
Deployment
This app is configured for deployment on Hugging Face Spaces using Docker.