Empirical Evaluation of Large Language Models in Automated Program Repair