# Deployment Guide for Hugging Face Spaces

This guide provides step-by-step instructions for deploying the Multi-Modal Knowledge Distillation application to Hugging Face Spaces.

## 📋 Pre-Deployment Checklist

✅ **Project Structure Complete**
- All required files and directories are present
- Python syntax validation passed
- Frontend files are properly structured

✅ **Configuration Validated**
- `requirements.txt` contains all necessary dependencies
- `spaces_config.yaml` is properly configured
- API endpoints are implemented and accessible

✅ **Documentation Complete**
- Comprehensive README.md with usage instructions
- API documentation included
- Troubleshooting guide provided

## 🚀 Deployment Steps

### Step 1: Create Hugging Face Space

1. **Go to Hugging Face Spaces**
   - Visit [https://huggingface.co/spaces](https://huggingface.co/spaces)
   - Click "Create new Space"

2. **Configure Space Settings**
   - **Space name**: `multi-modal-knowledge-distillation` (or your preferred name)
   - **License**: MIT
   - **SDK**: Gradio
   - **Hardware**: T4 small (minimum) or T4 medium (recommended)
   - **Visibility**: Public or Private (your choice)

3. **Initialize Repository**
   - Choose "Initialize with README"
   - Click "Create Space"

### Step 2: Upload Project Files

Upload all the following files to your Space repository:

#### Core Application Files
```
app.py                    # Main FastAPI application
requirements.txt          # Python dependencies
spaces_config.yaml       # Hugging Face Spaces configuration
README.md                # Project documentation
.gitignore               # Git ignore rules
```

#### Source Code
```
src/
├── __init__.py          # Package initialization
├── model_loader.py      # Model loading utilities
├── distillation.py      # Knowledge distillation engine
└── utils.py             # Utility functions
```

#### Frontend Files
```
templates/
└── index.html           # Main web interface

static/
├── css/
│   └── style.css        # Application styles
└── js/
    └── main.js          # Frontend JavaScript
```

#### Directory Structure (will be created automatically)
```
uploads/                 # Uploaded model files
models/                  # Trained models
temp/                    # Temporary files
logs/                    # Application logs
```

### Step 3: Configure Hardware

1. **Go to Space Settings**
   - Click on "Settings" tab in your Space
   - Navigate to "Hardware" section

2. **Select Hardware**
   - **Minimum**: T4 small (16GB RAM, 1x T4 GPU)
   - **Recommended**: T4 medium (32GB RAM, 1x T4 GPU)
   - **For large models**: A10G small or larger

3. **Apply Changes**
   - Click "Update hardware"
   - Your Space will restart with new hardware

### Step 4: Monitor Deployment

1. **Build Process**
   - Watch the "Logs" tab for build progress
   - Build typically takes 5-10 minutes
   - Dependencies will be installed automatically

2. **Common Build Issues**
   - **PyTorch installation**: May take several minutes
   - **CUDA compatibility**: Ensure PyTorch version supports your hardware
   - **Memory issues**: Upgrade hardware if needed

3. **Successful Deployment**
   - Space status shows "Running"
   - Application is accessible via the Space URL
   - Health check endpoint responds correctly

## 🔧 Configuration Options

### Environment Variables

You can set these in your Space settings:

```bash
# Server Configuration
PORT=7860                 # Default port (usually not needed)
HOST=0.0.0.0             # Default host

# Resource Limits
MAX_FILE_SIZE=5368709120  # 5GB max file size
MAX_MODELS=10            # Maximum teacher models
MAX_TRAINING_TIME=3600   # 1 hour training limit

# GPU Configuration
CUDA_VISIBLE_DEVICES=0   # GPU device selection
```

### Hardware Recommendations

| Use Case | Hardware | RAM | GPU | Cost |
|----------|----------|-----|-----|------|
| Demo/Testing | CPU Basic | 16GB | None | Free |
| Small Models | T4 small | 16GB | T4 | Low |
| Production | T4 medium | 32GB | T4 | Medium |
| Large Models | A10G small | 24GB | A10G | High |

## 🧪 Testing Your Deployment

### 1. Health Check
```bash
curl https://your-space-name-username.hf.space/health
```

### 2. Web Interface
- Visit your Space URL
- Test file upload functionality
- Verify model selection works
- Check training configuration options

### 3. API Endpoints
Test key endpoints:
- `GET /` - Main interface
- `POST /upload` - File upload
- `GET /models` - List models
- `WebSocket /ws/{session_id}` - Real-time updates

## 🐛 Troubleshooting

### Build Failures

**PyTorch Installation Issues:**
```bash
# Check if CUDA version is compatible
# Update requirements.txt if needed
torch==2.1.0+cu118
```

**Memory Issues During Build:**
- Upgrade to higher hardware tier
- Reduce dependency versions
- Remove unnecessary packages

### Runtime Issues

**Out of Memory:**
- Increase hardware tier
- Reduce batch size in training
- Implement model sharding

**Model Loading Failures:**
- Check file format compatibility
- Verify Hugging Face model exists
- Ensure sufficient disk space

**WebSocket Connection Issues:**
- Check browser compatibility
- Verify firewall settings
- Try refreshing the page

### Performance Issues

**Slow Training:**
- Upgrade to GPU hardware
- Increase batch size
- Use mixed precision training

**High Memory Usage:**
- Monitor system resources
- Implement automatic cleanup
- Reduce model cache size

## 📊 Monitoring and Maintenance

### Logs and Monitoring
- Check Space logs regularly
- Monitor resource usage
- Set up alerts for failures

### Updates and Maintenance
- Keep dependencies updated
- Monitor for security issues
- Regular cleanup of temporary files

### Scaling Considerations
- Monitor user load
- Consider multiple Space instances
- Implement load balancing if needed

## 🔒 Security Best Practices

### File Upload Security
- Validate all uploaded files
- Implement size limits
- Scan for malicious content

### API Security
- Implement rate limiting
- Validate all inputs
- Use HTTPS only

### Resource Protection
- Monitor resource usage
- Implement timeouts
- Automatic cleanup procedures

## 📈 Performance Optimization

### Model Loading
- Cache frequently used models
- Implement lazy loading
- Use model compression

### Training Optimization
- Use mixed precision
- Implement gradient checkpointing
- Optimize batch sizes

### Frontend Performance
- Minimize JavaScript bundle
- Optimize CSS delivery
- Use CDN for static assets

## 🎯 Success Metrics

Your deployment is successful when:

✅ **Functionality**
- All API endpoints respond correctly
- File uploads work without errors
- Training completes successfully
- Model downloads work properly

✅ **Performance**
- Page loads in < 3 seconds
- Training starts within 30 seconds
- Real-time updates work smoothly
- Resource usage is within limits

✅ **User Experience**
- Interface is responsive on all devices
- Error messages are clear and helpful
- Progress tracking works accurately
- Documentation is accessible

## 📞 Support and Resources

- **Hugging Face Spaces Documentation**: [https://huggingface.co/docs/hub/spaces](https://huggingface.co/docs/hub/spaces)
- **FastAPI Documentation**: [https://fastapi.tiangolo.com/](https://fastapi.tiangolo.com/)
- **PyTorch Documentation**: [https://pytorch.org/docs/](https://pytorch.org/docs/)

---

**Your Multi-Modal Knowledge Distillation application is now ready for production deployment! 🎉**