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

TL;DR

DevEval is a new benchmark aligned with real-world code repositories, providing comprehensive annotations and evaluating popular LLMs' coding abilities in practical scenarios, revealing current limitations.

Contribution

We introduce DevEval, a benchmark aligned with real-world repositories, with extensive annotations and evaluation of LLMs' coding performance in practical contexts.

Findings

GPT-4-turbo achieves only 53.04% Pass@1 on DevEval.

Existing LLMs have significant shortcomings in real-world code generation.

DevEval facilitates future development of more capable LLMs for real-world coding tasks.

Abstract

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.