m-a-p/CodeEditorBench
收藏资源简介:
--- license: apache-2.0 --- # CodeEditorBench [**🌐 Homepage**](https://codeeditorbench.github.io/) | [**🤗 Dataset**](https://huggingface.co/datasets/m-a-p/CodeEditorBench) | [**📖 arXiv**](https://arxiv.org/pdf/2404.03543.pdf) | [**GitHub**](https://github.com/CodeEditorBench/CodeEditorBench) ## Introduction Large Language Models (LLMs) for code are rapidly evolving, with code editing emerging as a critical capability. We introduce CodeEditorBench, an evaluation framework designed to rigorously assess the performance of LLMs in code editing tasks, including debugging, translating, polishing, and requirement switching. Unlike existing benchmarks focusing solely on code generation, CodeEditorBench emphasizes real-world scenarios and practical aspects of software development. We curate diverse coding challenges and scenarios from five sources, covering various programming languages, complexity levels, and editing tasks. Evaluation of 19 LLMs reveals that closed-source models (particularly Gemini-Ultra and GPT-4), outperform open-source models in CodeEditorBench, highlighting differences in model performance based on problem types and prompt sensitivities. CodeEditorBench aims to catalyze advancements in LLMs by providing a robust platform for assessing code editing capabilities. We will release all prompts and datasets to enable the community to expand the dataset and benchmark emerging LLMs. By introducing CodeEditorBench, we contribute to the advancement of LLMs in code editing and provide a valuable resource for researchers and practitioners.  ## Results <div style="display: flex; justify-content: space-around; align-items: center;"> <img src="Models_Zero_Shot.png" alt="First Image Description" style="width: 48%;" /> <img src="win_rate_zero.png" alt="Second Image Description" style="width: 48%;" /> </div> We propose evaluating LLMs across four scenarios capturing various code editing capabilities, namely code debug, code translate, code polish, and code requirement switch.The figure in left depicts various model performances across the four scenarios available in CodeEditorBench\_Plus in a radial plot – highlighting how relative differences across models change across the scenarios. We also give the Performance of open-source and closed-source models on CodeEditorBench\_Plus in zero-shot evaluated through win\_rate in the right figure. 🎯All results of models are generated by greedy decoding. ✨Code Debug, Code Translate and Code Requirement Switch are evaluated with pass@1, while Code Polish is evaluated with Mean OptScore. ## Disclaimers The guidelines for the annotators emphasized strict compliance with copyright and licensing rules from the initial data source, specifically avoiding materials from websites that forbid copying and redistribution. Should you encounter any data samples potentially breaching the copyright or licensing regulations of any site, we encourage you to [contact](#contact) us. Upon verification, such samples will be promptly removed. ## Contact <!-- - Jiawei Guo: moriatysss152@gmail.com - Ziming Li : - Xueling Liu: - Kaijing Ma: --> - Ge Zhang: zhangge@01.ai - Wenhu Chen: wenhuchen@uwaterloo.ca - Jie Fu: jiefu@ust.hk ## Citation **BibTeX:** ```bibtex @misc{guo2024codeeditorbench, title={CodeEditorBench: Evaluating Code Editing Capability of Large Language Models}, author={Jiawei Guo and Ziming Li and Xueling Liu and Kaijing Ma and Tianyu Zheng and Zhouliang Yu and Ding Pan and Yizhi LI and Ruibo Liu and Yue Wang and Shuyue Guo and Xingwei Qu and Xiang Yue and Ge Zhang and Wenhu Chen and Jie Fu}, year={2024}, eprint={2404.03543}, archivePrefix={arXiv}, primaryClass={cs.SE} } ```
许可证: apache-2.0 # CodeEditorBench [🌐 项目主页](https://codeeditorbench.github.io/) | [🤗 数据集仓库](https://huggingface.co/datasets/m-a-p/CodeEditorBench) | [📖 arXiv论文](https://arxiv.org/pdf/2404.03543.pdf) | [GitHub 仓库](https://github.com/CodeEditorBench/CodeEditorBench) ## 简介 面向代码的大语言模型(Large Language Models,LLMs)正快速演进,代码编辑能力已成为其核心关键能力之一。本研究提出CodeEditorBench,一款旨在严格评估大语言模型代码编辑能力的评测框架,覆盖代码调试、代码翻译、代码优化以及需求切换四类编辑任务。与仅聚焦代码生成的现有基准测试不同,CodeEditorBench更贴近软件开发的真实场景与工程实践细节。 研究团队从五大数据源中精选了多样化的代码挑战与场景,覆盖多种编程语言、不同复杂度等级以及各类代码编辑任务。通过对19款大语言模型的评测发现,闭源模型(尤其是Gemini-Ultra与GPT-4)在CodeEditorBench上的表现优于开源模型,这凸显了不同模型在任务类型与提示词敏感度上的性能差异。 CodeEditorBench旨在通过搭建可靠的代码编辑能力评测平台,推动大语言模型领域的技术进步。研究团队将公开全部提示词与数据集,以便社区扩展数据集并对新兴大语言模型进行评测。本研究提出的CodeEditorBench将为代码编辑领域的大语言模型发展提供助力,并为研究者与工程实践者提供宝贵的评测资源。  ## 评测结果 <div style="display: flex; justify-content: space-around; align-items: center;"> <img src="Models_Zero_Shot.png" alt="零样本下各模型性能对比" style="width: 48%;" /> <img src="win_rate_zero.png" alt="零样本胜率对比" style="width: 48%;" /> </div> 本研究将大语言模型的代码编辑能力拆解为四类场景:代码调试、代码翻译、代码优化与需求切换,并基于此开展评测。左侧径向雷达图展示了各模型在CodeEditorBench_Plus四类任务上的性能表现,直观呈现了不同模型间的相对性能差异如何随任务场景变化;右侧图表则通过胜率(win_rate)展示了开源与闭源模型在CodeEditorBench_Plus上的零样本评测结果。 🎯 所有模型的评测结果均基于贪心解码生成。 ✨ 代码调试、代码翻译与需求切换任务采用pass@1指标进行评测,代码优化任务则采用平均最优得分(Mean OptScore)作为评测标准。 ## 免责声明 标注指南明确要求标注人员严格遵守原始数据源的版权与许可协议,特别需规避来自禁止复制与再分发的网站的素材。若您发现任何数据样本可能侵犯任何站点的版权或许可规范,欢迎[联系](#contact)我们,经核实后相关样本将立即移除。 ## 联系方式 - 郭佳伟:moriatysss152@gmail.com - 李梓明: - 刘雪玲: - 马凯静: - 张戈:zhangge@01.ai - 陈文虎:wenhuchen@uwaterloo.ca - 傅杰:jiefu@ust.hk ## 引用 **BibTeX:** bibtex @misc{guo2024codeeditorbench, title={CodeEditorBench: Evaluating Code Editing Capability of Large Language Models}, author={Jiawei Guo and Ziming Li and Xueling Liu and Kaijing Ma and Tianyu Zheng and Zhouliang Yu and Ding Pan and Yizhi LI and Ruibo Liu and Yue Wang and Shuyue Guo and Xingwei Qu and Xiang Yue and Ge Zhang and Wenhu Chen and Jie Fu}, year={2024}, eprint={2404.03543}, archivePrefix={arXiv}, primaryClass={cs.SE} }
CodeEditorBench 数据集概述
数据集介绍
CodeEditorBench 是一个评估框架,旨在严格评估大型语言模型(LLMs)在代码编辑任务中的性能,包括调试、翻译、优化和需求切换。该数据集不同于仅关注代码生成的现有基准,它强调实际软件开发的现实场景和实用方面。数据集从五个不同来源精心挑选了多样化的编程挑战和场景,涵盖多种编程语言、复杂度级别和编辑任务。
数据集特点
- 任务多样性:包括代码调试、翻译、优化和需求切换。
- 模型评估:对19个LLMs进行评估,揭示了闭源模型(如Gemini-Ultra和GPT-4)在CodeEditorBench中优于开源模型的性能。
- 评估方法:代码调试、翻译和需求切换使用pass@1评估,代码优化使用Mean OptScore评估。
数据集用途
CodeEditorBench 旨在通过提供一个强大的平台来评估代码编辑能力,推动LLMs的进步和创新。数据集的发布将包括所有提示和数据,以支持社区扩展数据集并基准测试新兴的LLMs。
版权与合规性
数据集的标注指南强调严格遵守从初始数据源的版权和许可规则,特别避免使用禁止复制和重新分发的网站材料。如发现任何可能违反版权或许可规定的数据样本,鼓励联系数据集维护者进行核实和移除。
联系方式
- Ge Zhang: zhangge@01.ai
- Wenhu Chen: wenhuchen@uwaterloo.ca
- Jie Fu: jiefu@ust.hk




