Abstract: This study proposes a multi-objective fine-tuning approach named MORepair that significantly improves the efficiency and effectiveness of Large Language Models (LLMs) in software program repair tasks.

Key Insights:

• MORepair focuses on both syntactical and logical adjustments in code during the repair process.
• Demonstrates a 7.6% to 10% improvement in top-10 repair suggestions compared to other models.
• Fine-tuning strategy yields superior performance in C++ and Java repair benchmarks.

Research Significance: MORepair’s dual focus on syntax and logic opens new avenues for refining LLM applications in code transformation tasks. This approach could fundamentally alter methodologies in LLM fine-tuning for enhanced performance across varied domains.