KoDialogBench

KoDialogBench是一个专为评估韩语对话能力的大型语言模型设计的基准。该数据集由韩国科学技术院的Seongbo Jang等人创建，旨在通过收集和翻译多种韩语对话，评估模型在日常话题上的对话理解能力。数据集包含21个测试集，涵盖对话理解和回应选择两大任务，旨在全面评估模型在多样对话场景中的深度理解和回应准确性。KoDialogBench不仅为评估韩语对话能力提供了多视角，还为开发高效的对话代理铺平了道路。

KoDialogBench is a benchmark developed to evaluate the dialogue capabilities of large language models (LLMs) for Korean language tasks. This benchmark was constructed by Seongbo Jang et al. from the Korea Advanced Institute of Science and Technology (KAIST). Its core objective is to assess models' dialogue comprehension abilities on everyday conversational topics by collecting and translating a diverse set of Korean dialogues. The dataset comprises 21 test sets covering two primary tasks: dialogue understanding and response selection, with the aim of comprehensively evaluating models' deep comprehension and response accuracy across a wide range of dialogue scenarios. KoDialogBench not only offers multiple perspectives for evaluating Korean dialogue capabilities but also paves the way for the development of efficient dialogue agents.