CORECODEBENCH
收藏资源简介:
CORECODEBENCH是一个可配置的多场景仓库级基准数据集,旨在评估大型语言模型在工程级代码处理方面的性能。该数据集通过CorePipe自动化流程从GitHub仓库中生成,包括开发、bug修复和测试驱动开发三种类型的原子问题,以及三种复合问题类型,难度可灵活调整。数据集的设计考虑了工程实践中代码的多样性和复杂性,通过实验表明,CORECODEBENCH能够有效地评估LLMs在工程级代码开发中的实际能力和适应性。
CORECODEBENCH is a configurable multi-scenario repository-level benchmark dataset designed to evaluate the performance of large language models (LLMs) in engineering-grade code processing. Generated from GitHub repositories via the CorePipe automated workflow, this dataset includes three types of atomic problems: development, bug fixing, and test-driven development, alongside three types of composite problems, with flexibly adjustable difficulty levels. The dataset’s design accounts for the diversity and complexity of code in real-world engineering practices. Experimental results demonstrate that CORECODEBENCH can effectively assess the actual capabilities and adaptability of LLMs in engineering-grade code development.




