Requirements

• Generate synthetic data
• View attention maps
• Use torch 2x
• Train inside docker compose with cuda
• Build as package
• Deploy on hugging face
• Automated network testing
• Use lighting, accelerator, tqdm
• Use wandb
• Build the model with pytorch models
• Parameter search (ray tune)
• View easily
• Use mkdocs in black
• Use rerun