Retention analysis of edited knowledge after fine-tuning

Fufang Wen, Shichang Zhang

Large language models (LLMs) store vast amounts of knowledge, which often requires updates to correct factual errors, incorporate newly acquired information, or adapt model behavio…