The ability of a fine-tuned model to preserve factual and commonsense knowledge from its pretraining after adaptation to a new task.