The Impact of Fine-Tuning Large Language Models on Automated Program Repair