RepairCAT: Applying Large Language Model to Fix Bugs in AI-Generated Programs
Nan Jiang, Yi Wu · 2024
Automated program repair has been a crucial and popular domain for years, and with the development of large language models (LLMs) and the trend of using LLMs for code generation, there comes the new challenge of fixing bugs in LLM-generated (AI-generated) programs. In this work, we introduce RepairCAT, a simple and neat framework for fine-tuning large language models for automated repairing Python programs. Our experiments built on StarCoder-1B successfully generated patches fixing the failed test cases for 14 out of 100 bugs in the Python programs, 2 of which passed all the public test cases and were considered plausible.