# BioMistral Medical Chatbot - Hugging Face Space

A medical chatbot that runs locally in Hugging Face Spaces, designed to answer medical questions using the BioMistral medical model.

## 🚀 Features

- **BioMistral Medical Model**: Uses the specialized BioMistral-7b-Chat-Doctor model
- **Doctor-Patient Conversations**: Designed for multi-turn medical dialogue
- **Local Model Loading**: Loads the medical model directly in the Space
- **Device Compatibility**: Handles GPU/CPU device switching automatically
- **Error Recovery**: Graceful fallback to CPU if GPU issues occur
- **Medical Focus**: Specialized for medical questions and responses

## 🔧 Recent Updates

### Model Switch to BioMistral-7b-Chat-Doctor

**Previous Issue**: The original model had architecture mismatches and device placement problems.

**New Solution**: Switched to [BioMistral-7b-Chat-Doctor](https://huggingface.co/siyah1/BioMistral-7b-Chat-Doctor) which offers:
- **Proper medical training**: Fine-tuned on 257k medical conversations
- **Stable architecture**: Based on BioMistral-7B foundation
- **Multi-turn conversations**: Designed for ongoing doctor-patient dialogue
- **Apache-2.0 license**: Open source and free to use

### Device Handling Improvements

- Proper device detection and mapping
- Consistent tensor placement on the same device
- Automatic fallback to CPU if GPU issues occur
- Better error handling during generation

## 📦 Dependencies

- `gradio>=4.44.0` - Web interface
- `torch>=2.1.0,<3.0.0` - PyTorch for model operations
- `transformers>=4.35.0,<5.0.0` - Hugging Face transformers
- `accelerate>=0.24.0` - Model acceleration
- `spaces>=0.1.0` - HF Spaces support

## 🏃‍♂️ Usage

1. **Deploy to HF Space**: Upload the files to your Hugging Face Space
2. **Automatic Setup**: The Space will automatically install dependencies
3. **Model Loading**: The BioMistral medical model will be downloaded and loaded
4. **Start Chatting**: Use the web interface to ask medical questions

## 🧪 Testing

Run the test script to verify device handling:

```bash
python test_device_fix.py
```

## 🔍 Troubleshooting

### Common Issues

1. **Device Mismatch Error**: 
   - ✅ Fixed with improved tensor handling
   - Automatic fallback to CPU if GPU fails

2. **Memory Issues**:
   - Uses `low_cpu_mem_usage=True`
   - Float16 precision on GPU for memory efficiency

3. **Model Loading Failures**:
   - Multiple fallback strategies
   - Detailed error logging

### Logs

Check the Space logs for detailed information about:
- Device detection
- Model loading progress
- Generation errors
- Memory usage

## 📝 Example Questions

- "What are the symptoms of diabetes?"
- "How is hypertension treated?"
- "What are the risk factors for heart disease?"
- "I have a headache, what should I do?"
- "What are the common symptoms of COVID-19?"

## 🤝 Contributing

Feel free to submit issues or improvements to the device handling logic.

## 📄 License

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

## 🙏 Acknowledgments

- **BioMistral Team**: For the excellent medical foundation model
- **Siyahul Haque T P**: For fine-tuning the chat-doctor model
- **Hugging Face**: For providing the platform and infrastructure