An Empirical Study of NetOps Capability of Pre-Trained Large Language Models