Coffee: Boost Your Code LLMs by Fixing Bugs with Feedback

TL;DR

This paper introduces Coffee, a dataset and framework for improving open-source code LLMs' ability to generate accurate feedback for bug fixing, achieving state-of-the-art results on code correction benchmarks.

Contribution

The work presents a new dataset and a tuning framework that enhance open-source code LLMs' feedback quality for code editing tasks.

Findings

Coffee outperforms previous models on HumanEvalFix benchmark.

The framework reduces superficial and misleading feedback.

Open-source models can be improved for code fixing with proper tuning.

Abstract

Code editing is an essential step towards reliable program synthesis to automatically correct critical errors generated from code LLMs. Recent studies have demonstrated that closed-source LLMs (i.e., ChatGPT and GPT-4) are capable of generating corrective feedback to edit erroneous inputs. However, it remains challenging for open-source code LLMs to generate feedback for code editing, since these models tend to adhere to the superficial formats of feedback and provide feedback with misleading information. Hence, the focus of our work is to leverage open-source code LLMs to generate helpful feedback with correct guidance for code editing. To this end, we present Coffee, a collected dataset specifically designed for code fixing with feedback. Using this dataset, we construct CoffeePots, a framework for COde Fixing with FEEdback via Preference-Optimized Tuning and Selection. The proposed framework aims to automatically generate helpful feedback for code editing while minimizing the potential risk of superficial feedback. The combination of Coffee and CoffeePots marks a significant advancement, achieving state-of-the-art performance on HumanEvalFix benchmark. Codes and model checkpoints are publicly available at https://github.com/Lune-Blue/COFFEE.