Ultra-scale Playbook - Train on a single GPU Notes on Ultra-scale Playbook - training LLM on a single GPU