Sleep-inspired replay and consolidation may be your best defense against catastrophic forgetting in large language models, helping them retain knowledge du