USAGE.md

# Model Usage Instructions

## Loading the Model

```python
import torch
from pathlib import Path

# Load the main LAM model (14GB)
model_path = "model_weights/lam_model.pt"
model = torch.load(model_path, map_location='cuda' if torch.cuda.is_available() else 'cpu')

# Or load the smaller checkpoint
checkpoint = torch.load("checkpoints/best.pt")
```

## Model Files

- `model_weights/lam_model.pt` - Main unified model (14GB)
- `checkpoints/best.pt` - Best training checkpoint (6MB)
- `model_weights/sample_vision_weights.pt` - Vision module weights
- `model_weights/config.json` - Model configuration
- `model_weights/capabilities.json` - Model capabilities definition

## Requirements

```bash
pip install torch torchvision transformers pillow
```

## Inference Example

```python
# Coming soon - full inference pipeline
```