Utility Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/gururise/AlpacaDataCleaned
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是从经过精炼的Alpaca数据集中筛选出的与安全相关的（提示，回应）对，用于衡量模型在多个任务中的零样本准确性。这些任务包括BoolQ、RTE、HellaSwag、WinoGrande、ARC容易和挑战难度，以及OpenbookQA。该数据集旨在评估模型对各种任务回应的实用性。
This dataset consists of safety-related (prompt, response) pairs selected from the refined Alpaca dataset, and is used to measure the zero-shot accuracy of models across multiple tasks.
These tasks include BoolQ, RTE, HellaSwag, WinoGrande, ARC-Easy, ARC-Challenge, and OpenbookQA.
This dataset is designed to evaluate the utility of model responses to various tasks.
提供机构：
Alpaca
