KBENCH
收藏资源简介:
KBENCH数据集由哥伦比亚大学和Google等机构创建,包含279个真实的Linux内核崩溃解决案例。每个案例详细记录了崩溃的堆栈跟踪、重现崩溃的输入文件、开发者验证的修复补丁以及相关的编译和执行配置文件。数据集的创建过程涉及从Syzkaller和Syzbot等开源测试服务中收集和筛选数据,确保了数据的真实性和多样性。KBENCH主要用于评估和提升大型语言模型在复杂系统级软件工程任务中的性能,特别是在理解和修复内核级崩溃的能力。
The KBENCH dataset, developed by institutions including Columbia University and Google, contains 279 real Linux kernel crash resolution cases. Each case thoroughly documents the crash stack trace, input files for reproducing the crash, developer-validated fix patches, and associated compilation and execution configuration files. The dataset was created by collecting and filtering data from open-source testing services such as Syzkaller and Syzbot, ensuring the authenticity and diversity of the data. KBENCH is primarily used to evaluate and enhance the performance of large language models (LLMs) on complex system-level software engineering tasks, particularly their ability to understand and fix kernel-level crashes.




