A Survey on Evaluation of Large Language Models