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

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chae, Hyungjoo, Kwon, Taeyoon, Moon, Seungjun, Song, Yongho, Kang, Dongjin, Ong, Kai Tzu-iunn, Kwak, Beong-woo, Bae, Seonghyeon, Hwang, Seung-won, Yeo, Jinyoung
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909336288100352
author Chae, Hyungjoo
Kwon, Taeyoon
Moon, Seungjun
Song, Yongho
Kang, Dongjin
Ong, Kai Tzu-iunn
Kwak, Beong-woo
Bae, Seonghyeon
Hwang, Seung-won
Yeo, Jinyoung
author_facet Chae, Hyungjoo
Kwon, Taeyoon
Moon, Seungjun
Song, Yongho
Kang, Dongjin
Ong, Kai Tzu-iunn
Kwak, Beong-woo
Bae, Seonghyeon
Hwang, Seung-won
Yeo, Jinyoung
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2409_19715
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Coffee-Gym: An Environment for Evaluating and Improving Natural Language Feedback on Erroneous Code
Chae, Hyungjoo
Kwon, Taeyoon
Moon, Seungjun
Song, Yongho
Kang, Dongjin
Ong, Kai Tzu-iunn
Kwak, Beong-woo
Bae, Seonghyeon
Hwang, Seung-won
Yeo, Jinyoung
Computation and Language
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.
title Coffee-Gym: An Environment for Evaluating and Improving Natural Language Feedback on Erroneous Code
topic Computation and Language
url https://arxiv.org/abs/2409.19715