Which metric is used to evaluate the performance of foundation models (FMs) on text summarization tasks?