EXECREPOBENCH
收藏资源简介:
EXECREPOBENCH是由阿里巴巴集团创建的一个用于代码完成任务的评估基准数据集,包含1.2K个来自50个活跃Python仓库的样本。该数据集通过多级语法树解析,对代码片段进行多层次的掩码处理,以模拟真实编程环境中的复杂依赖关系。数据集的创建过程包括从GitHub收集活跃的Python仓库,并将其转换为可执行格式,同时生成相应的测试用例。EXECREPOBENCH主要用于评估和提升开源大语言模型在实际编程场景中的代码完成能力,旨在解决代码生成和补全中的复杂依赖问题。
EXECREPOBENCH is an evaluation benchmark dataset for code completion tasks, developed by Alibaba Group. It contains 1.2K samples sourced from 50 active Python repositories. This dataset adopts multi-level syntax tree parsing to perform multi-level masking on code snippets, aiming to simulate the complex dependencies in real-world programming environments. The dataset creation process includes collecting active Python repositories from GitHub, converting them into executable formats, and generating corresponding test cases. EXECREPOBENCH is primarily used to evaluate and enhance the code completion capabilities of open-source large language models (LLMs) in real-world programming scenarios, and it aims to address the complex dependency issues in code generation and completion.




