Coffee-Gym: An Environment for Evaluating and Improving Natural Language Feedback on Erroneous Code

TL;DR

Coffee-Gym is a new RL environment with a dataset and reward function for training models to give helpful feedback on erroneous code, improving code editing capabilities of open-source LLMs.

Contribution

It introduces Coffee-Gym, a comprehensive environment with a dataset and reward function, addressing the lack of high-quality data for training code feedback models with RL.

Findings

Feedback models trained with Coffee-Gym outperform baselines.

Models achieve performance comparable to closed-source LLMs.

The dataset and models are publicly available.

Abstract

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.