metadata

license: apache-2.0

CodeEditorBench

🌐 Homepage | 🤗 Dataset | 📖 arXiv | GitHub

Introduction

Large Language Models (LLMs) for code are rapidly evolving, with code editing emerging as a critical capability. We introduce CodeEditorBench, a pioneering evaluation framework designed to rigorously assess the performance of LLMs in code editing tasks, including debugging, translating, polishing, and require- ment switching. Unlike existing benchmarks focusing solely on code generation, CodeEditorBench emphasizes real-world scenarios and practical aspects of software development. We curated diverse coding challenges and scenarios from five sources, covering various programming languages, complexity levels, and editing tasks. Evaluating 19 LLMs revealed that closed-source models, particularly Gemini-Ultra and GPT-4, outperform open-source models in CodeEditorBench, highlighting differences in model performance based on problem type and prompt sensitivity. 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 practi- tioners in the field.

🏆 Leaderboard

Model	Size	Open	Debug	Translate	Switch	Polish	Win Rate
Zero-shot
Gemini-Ultra	-	❌	0.301 (0.457)	0.338 (0.261)	0.028	4.64% (3.45%)	0.779 (0.632)
GPT-4	-	❌	0.308 (0.489)	0.335 (0.449)	0.225	0.15% (0.87%)	0.779 (0.882)
GPT-3.5-Turbo	-	❌	0.284 (0.491)	0.385 (0.443)	0.169	0.09% (0.84%)	0.765 (0.853)
Gemini-Pro	-	❌	0.279 (0.420)	0.200 (0.285)	0.061	5.07% (6.27%)	0.750 (0.765)
DS-33B-INST	33B	✅	0.267 (0.483)	0.353 (0.427)	0.131	0.06% (0.64%)	0.676 (0.728)
WC-33B	33B	✅	0.265 (0.483)	0.315 (0.415)	0.125	0.19% (0.62%)	0.676 (0.669)
...	...	...	...	...	...	...	...
Few-shot
Gemini-Ultra	-	❌	0.283 (0.446)	0.406 (0.292)	0.131	4.83% (4.17%)	0.897 (0.706)
GPT-4	-	❌	0.336 (0.519)	0.453 (0.488)	0.275	0.22% (0.7%)	0.868 (0.926)
...	...	...	...	...	...	...	...
CoT
GPT-4	-	❌	0.280 (0.439)	0.338 (0.414)	0.174	0.33% (1.45%)	0.850 (0.800)
GLM-4	-	❌	0.228 (0.201)	0.218 (0.260)	0.072	4.09% (5.28%)	0.750 (0.600)
...	...	...	...	...	...	...	...

🎯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.

🗂️Values outside parentheses denoting Plus results and inside denoting Primary results. For the Switch class, Primary and Plus results are identical, and only one score is displayed.

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 us. Upon verification, such samples will be promptly removed.

Contact

Ge Zhang: zhangge@01.ai
Wenhu Chen: wenhuchen@uwaterloo.ca
Jie Fu: jiefu@ust.hk

Citation

BibTeX:

@inproceedings{guo2024editorbench,
  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 Ruibo Liu and Yue Wang and Yizhi Li and Xingwei Qu and Xiang Yue and Shuyue Guo and Ge Zhang and Wenhu Chen and Jie Fu},
  booktitle={arxiv},
  year={2024},
}