HAERAE-HUB/Ko-PIQA

Ko-PIQA是一个韩语物理常识推理数据集，设计用于补充以英语为中心的基准测试，如PIQA，并包含了具有文化背景的物理推理问题。数据集共有441个项目，其中87个（19.7%）具有文化背景，例如，例如关于泡菜储存、韩服护理和暖炕加热的问题。数据集采用PIQA风格的二元选择格式，旨在评估韩国大型语言模型的物理推理能力。

Ko-PIQA is a Korean Physical Commonsense Reasoning dataset designed to complement English-centric benchmarks like PIQA and includes culturally-grounded physical reasoning questions. The dataset contains a total of 441 items, of which 87 (19.7%) are culturally-grounded, such as questions about kimchi storage, hanbok care, and ondol heating. The dataset is in a PIQA-style binary choice format and aims to evaluate the physical reasoning capabilities of Korean LLMs.