Published in Lecture Notes in Computer Science vol. 13380: “Fine-Tuning GPT-2 to Patch Programs, Is It Worth It?” (Lajkó, Horváth, Csuvik and Vidács).
Training data for automated program repair is scarce and expensive to prepare, so we evaluated GPT-2 both out of the box and after fine-tuning, on 1,559 JavaScript code snippets. The fine-tuned model learned to write syntactically correct code almost every time, while the model without fine-tuning lacked some of these strengths.
Read the paper: publisher page
