Investigating Hallucinations in Pruned Large Language Models for Abstractive Summarization