README.md

metadata

title: MEDChat AI
emoji: 🏥
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.3.0
app_file: app.py
pinned: false
license: mit

MEDChat AI 🏥

A medical chatbot powered by fine-tuned LLaMA 2 for answering medical questions.

⚠️ GPU Hardware Required

This Space uses a 4-bit quantized LLaMA 2 model that requires GPU hardware to run inference.

How to Test This Application:

Option 1: Upgrade This Space to GPU (Paid)

1. Click Settings in the top navigation
2. Select Space hardware
3. Choose T4 GPU (~$0.60/hour when running)
4. Click Save and wait for the Space to restart

Option 2: Run on Google Colab (Free) ⭐ Recommended

1. Visit the GitHub Repository
2. Click on the Colab badge or download the notebook
3. Open in Google Colab
4. Select Runtime → Change runtime type → T4 GPU
5. Run all cells to test the chatbot with free GPU

Option 3: Watch the Demo Video

• View working demo

Current Status: This Space is running on CPU and will display error messages when attempting to generate responses. The interface is fully functional and can be explored.

---

Features

• 💬 Medical Q&A support using fine-tuned LLaMA 2
• 🔐 User authentication system (demo - in-memory storage)
• 🎨 Clean, intuitive Gradio interface
• 📚 Fine-tuned on medical terminology dataset
• ⚡ 4-bit quantization for efficient inference

Usage

1. Sign Up: Create an account on the Sign Up tab
2. Login: Use your credentials to log in
3. Chat: Ask medical questions and get AI-powered responses

Example Questions

• What does the immune system do?
• What is Epistaxis?
• What are allergies?
• What's the difference between bacteria and viruses?
• Should I start taking creatine?

Technical Details

Model Architecture

• Base Model: LLaMA 2 (7B parameters)
• Fine-tuning: LoRA (Low-Rank Adaptation)
• Quantization: 4-bit with bitsandbytes (NF4)
• Dataset: Medical terminology corpus

Tech Stack

• Framework: Gradio 4.0
• Model Hub: Hugging Face Transformers
• Fine-tuning: PEFT (Parameter-Efficient Fine-Tuning)
• Quantization: bitsandbytes
• Training: SFTTrainer from TRL library

Model Configuration

- Load in 4-bit: True
- Compute dtype: float16
- Quantization type: nf4
- LoRA rank (r): 16
- LoRA alpha: 16

Repository Structure

├── app.py              # Main Gradio application
├── requirements.txt    # Python dependencies
├── README.md          # This file
└── notebooks/         # Google Colab notebooks
    └── training.ipynb # Model fine-tuning notebook

Disclaimer

⚠️ For educational purposes only. This chatbot is not a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare provider for medical concerns.

Links

• 🔗 GitHub Repository
• 📹 Demo Video
• 📚 Hugging Face Model

License

MIT License - See LICENSE file for details

---

Questions or Issues? Open an issue on the GitHub repository or reach out via the Community tab.

Created as a portfolio project demonstrating LLM fine-tuning, quantization, and deployment techniques.