DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Jia, Li, Ge, Zhao, Yunfei, Li, Yongmin, Liu, Huanyu, Zhu, Hao, Wang, Lecheng, Liu, Kaibo, Fang, Zheng, Wang, Lanshen, Ding, Jiazheng, Zhang, Xuanming, Zhu, Yuqi, Dong, Yihong, Jin, Zhi, Li, Binhua, Huang, Fei, Li, Yongbin
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909213299572736
author Li, Jia
Li, Ge
Zhao, Yunfei
Li, Yongmin
Liu, Huanyu
Zhu, Hao
Wang, Lecheng
Liu, Kaibo
Fang, Zheng
Wang, Lanshen
Ding, Jiazheng
Zhang, Xuanming
Zhu, Yuqi
Dong, Yihong
Jin, Zhi
Li, Binhua
Huang, Fei
Li, Yongbin
author_facet Li, Jia
Li, Ge
Zhao, Yunfei
Li, Yongmin
Liu, Huanyu
Zhu, Hao
Wang, Lecheng
Liu, Kaibo
Fang, Zheng
Wang, Lanshen
Ding, Jiazheng
Zhang, Xuanming
Zhu, Yuqi
Dong, Yihong
Jin, Zhi
Li, Binhua
Huang, Fei
Li, Yongbin
contents How to evaluate the coding abilities of Large Language Models (LLMs) remains an open question. We find that existing benchmarks are poorly aligned with real-world code repositories and are insufficient to evaluate the coding abilities of LLMs. To address the knowledge gap, we propose a new benchmark named DevEval, which has three advances. (1) DevEval aligns with real-world repositories in multiple dimensions, e.g., code distributions and dependency distributions. (2) DevEval is annotated by 13 developers and contains comprehensive annotations (e.g., requirements, original repositories, reference code, and reference dependencies). (3) DevEval comprises 1,874 testing samples from 117 repositories, covering 10 popular domains (e.g., Internet, Database). Based on DevEval, we propose repository-level code generation and evaluate 8 popular LLMs on DevEval (e.g., gpt-4, gpt-3.5, StarCoder 2, DeepSeek Coder, CodeLLaMa). Our experiments reveal these LLMs' coding abilities in real-world code repositories. For example, in our experiments, the highest Pass@1 of gpt-4-turbo is only 53.04%. We also analyze LLMs' failed cases and summarize their shortcomings. We hope DevEval can facilitate the development of LLMs in real code repositories. DevEval, prompts, and LLMs' predictions have been released.
format Preprint
id arxiv_https___arxiv_org_abs_2405_19856
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
Li, Jia
Li, Ge
Zhao, Yunfei
Li, Yongmin
Liu, Huanyu
Zhu, Hao
Wang, Lecheng
Liu, Kaibo
Fang, Zheng
Wang, Lanshen
Ding, Jiazheng
Zhang, Xuanming
Zhu, Yuqi
Dong, Yihong
Jin, Zhi
Li, Binhua
Huang, Fei
Li, Yongbin
Computation and Language
Software Engineering
How to evaluate the coding abilities of Large Language Models (LLMs) remains an open question. We find that existing benchmarks are poorly aligned with real-world code repositories and are insufficient to evaluate the coding abilities of LLMs. To address the knowledge gap, we propose a new benchmark named DevEval, which has three advances. (1) DevEval aligns with real-world repositories in multiple dimensions, e.g., code distributions and dependency distributions. (2) DevEval is annotated by 13 developers and contains comprehensive annotations (e.g., requirements, original repositories, reference code, and reference dependencies). (3) DevEval comprises 1,874 testing samples from 117 repositories, covering 10 popular domains (e.g., Internet, Database). Based on DevEval, we propose repository-level code generation and evaluate 8 popular LLMs on DevEval (e.g., gpt-4, gpt-3.5, StarCoder 2, DeepSeek Coder, CodeLLaMa). Our experiments reveal these LLMs' coding abilities in real-world code repositories. For example, in our experiments, the highest Pass@1 of gpt-4-turbo is only 53.04%. We also analyze LLMs' failed cases and summarize their shortcomings. We hope DevEval can facilitate the development of LLMs in real code repositories. DevEval, prompts, and LLMs' predictions have been released.
title DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
topic Computation and Language
Software Engineering
url https://arxiv.org/abs/2405.19856