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