This paper presents Coffee-Gym, a comprehensive RL environment for training models that provide feedback on code editing. Coffee-Gym includes two major components: (1) Coffee, a dataset containing humans' code edit traces for coding questions and machine-written feedback for editing erroneous code; (2) CoffeeEval, a reward function that faithfully reflects the helpfulness of feedback by assessing the performance of the revised code in unit tests. With them, Coffee-Gym addresses the unavailability of high-quality datasets for training feedback models with RL, and provides more accurate rewards than the SOTA reward model (i.e., GPT-4). By applying Coffee-Gym, we elicit feedback models that outperform baselines in enhancing open-source code LLMs' code editing, making them comparable with closed-source LLMs. We make the dataset and the model checkpoint publicly available.
@article{arxiv.2409.19715,
title = {Coffee-Gym: An Environment for Evaluating and Improving Natural Language Feedback on Erroneous Code},
author = {Hyungjoo Chae and Taeyoon Kwon and Seungjun Moon and Yongho Song and Dongjin Kang and Kai Tzu-iunn Ong and Beong-woo Kwak and Seonghyeon Bae and Seung-won Hwang and Jinyoung Yeo},
journal= {arXiv preprint arXiv:2409.19715},
year = {2024}
}