fast-slow-weights

This paper introduces a Fast-Slow Training framework for LLMs that combines parameter updates with optimized context to improve sample efficiency and reduce catastrophic forgetting during continual learning.