ProofWriter

该数据集旨在测试大型语言模型的演绎推理能力，包含已知的facts、已知的rules、目标问题以及对应的标签。每个示例都被归类为三种标签之一：已证明、已反驳或未知。数据集按推理深度分为五个子集，每个部分均抽取了600个示例，任务专注于演绎推理。

This dataset is designed to test the deductive reasoning ability of large language models (LLMs), which contains known facts, known rules, target questions and their corresponding labels. Each instance is categorized into one of three labels: proven, refuted, or unknown. The dataset is divided into five subsets based on reasoning depth, with each subset containing 600 examples, and the task focuses on deductive reasoning.

搜集汇总

背景与挑战

背景概述

ProofWriter数据集旨在评估大型语言模型的演绎推理能力，包含已知事实、规则、目标问题及标签（已证明、已反驳或未知），按推理深度分为五个子集，每个子集有600个示例，专注于演绎推理任务。该数据集通过结构化问题和分类标签，系统性地测试模型在逻辑推理方面的表现。