KBENCH
收藏资源简介:
KBENCH是由哥伦比亚大学和谷歌等机构创建的一个用于评估大型语言模型在Linux内核崩溃解决能力的数据集。该数据集包含279个来自真实世界Linux内核的崩溃解决样本,每个样本包括崩溃堆栈跟踪、复现文件、开发者编写的修复补丁等。数据集的创建过程涉及使用KGYM平台进行内核编译、崩溃检测和代码补丁应用。KBENCH主要用于评估和提升模型在软件工程任务中的性能,特别是在处理低级系统代码、并发问题和模糊意图的崩溃解析方面。
KBENCH is a dataset developed by Columbia University, Google and other institutions to evaluate the crash resolution capabilities of large language models on the Linux kernel. This dataset includes 279 real-world Linux kernel crash resolution samples, each containing crash stack traces, reproduction files, developer-written fix patches and other relevant materials. The dataset creation process utilized the KGYM platform for kernel compilation, crash detection and code patch application. KBENCH is primarily used to evaluate and improve the performance of models in software engineering tasks, especially in crash analysis involving low-level system code, concurrency issues and ambiguous intent.

- 1KGym: A Platform and Dataset to Benchmark Large Language Models on Linux Kernel Crash Resolution哥伦比亚大学, 明尼苏达大学, 谷歌Deepmind, 谷歌 · 2024年



