HellaSwag 大模型常识推理数据集

HellaSwag 数据集是一个用于测试常识性自然语言推理 (commonsense NLI) 的新挑战数据集。该数据集由华盛顿大学和 Allen AI 于 2019 年推出，旨在通过构建一个对现有最先进模型具有挑战性的数据集，来探索深度预训练模型在常识推理方面的表现。相关论文成果「HellaSwag: Can a Machine Really Finish Your Sentence?」已被 ACL 2019 接受。

The HellaSwag dataset is a novel challenge dataset for testing commonsense natural language inference (commonsense NLI). It was introduced by the University of Washington and Allen AI in 2019, aiming to explore the commonsense reasoning performance of deep pre-trained models by constructing a dataset that poses challenges to existing state-of-the-art models. The associated paper titled "HellaSwag: Can a Machine Really Finish Your Sentence?" was accepted by ACL 2019.

搜集汇总

背景与挑战

背景概述

HellaSwag是一个用于测试常识性自然语言推理的数据集，包含70,000个问题，对人类简单但对最先进模型具有挑战性。该数据集采用对抗性过滤方法构建，旨在探索深度预训练模型在常识推理方面的表现。