chanyoungkim/heard-bench
收藏Hugging Face2026-04-23 更新2026-04-26 收录
下载链接：
https://hf-mirror.com/datasets/chanyoungkim/heard-bench
下载链接
链接失效反馈官方服务：
资源简介：
heard-bench是一个韩语长期记忆基准数据集，专注于研究单人商务夜间独白的领域。该数据集包含三个轨道（en_subset、ko_translated、ko_native），共270个项目，涵盖六种记忆能力。数据集旨在用于韩语长期记忆语言模型和小型设备助手的研究。README还详细介绍了数据集的构建过程、伦理考虑、已知限制和引用信息。
heard-bench is a Korean long-term-memory benchmark for the under-studied domain of solo-business nightly monologue. The dataset includes three tracks (en_subset, ko_translated, ko_native) with 270 items across six memory abilities. It is designed for research on Korean long-term-memory language models and small on-device assistants. The README also provides details on the datasets construction, ethical considerations, known limitations, and citation information.
提供机构：
chanyoungkim
