NCSOFT/K-MMStar
收藏Hugging Face2025-07-25 更新2024-12-14 收录
下载链接：
https://hf-mirror.com/datasets/NCSOFT/K-MMStar
下载链接
链接失效反馈官方服务：
资源简介：
K-MMStar是一个基于MMStar数据集的韩语版本，专门用于评估视觉-语言模型。该数据集通过将MMStar的验证集翻译成韩语，并经过人工审查以确保其自然性，从而创建了一个新的鲁棒性评估基准。数据集包含6个评估维度的问题，如粗略感知、细粒度感知和实例推理等，以便全面评估模型在韩语环境下的表现。
K-MMStar is a Korean evaluation benchmark for vision-language models, adapted from the MMStar dataset. It translates the validation subset of MMStar into Korean and ensures its naturalness through human inspection. The dataset includes features such as questions, images, answers, and is divided into multiple evaluation dimensions, such as coarse perception, fine-grained perception, and instance reasoning. The dataset has a download size of 42027023 bytes, a dataset size of 45162575.0 bytes, and the validation set contains 1500 samples.
提供机构：
NCSOFT
