flaitenberger/LogicalReasoning-hard-v2

该数据集是一个用于逻辑推理任务的数据集，包含多个配置，如训练集（最大深度10和5，各100万样本）和验证集（逐步增加深度，每步1000样本）。每个样本包括常量列表、谓词列表、一阶逻辑（FOL）和自然语言（NL）形式的前提、证明、问题和答案，以及元数据如难度级别、分支因子、干扰项比例和深度信息。数据集旨在支持机器推理模型，通过提供结构化逻辑表示和自然语言对应，促进对复杂推理链的理解和评估。

This dataset is designed for logical reasoning tasks and includes multiple configurations: training sets with maximum depths of 10 and 5 respectively, each containing 1 million samples, and a validation set where the depth increases incrementally with 1000 samples per step. Each sample contains a list of constants, a list of predicates, premises, proofs, questions and answers in both First-Order Logic (FOL) and Natural Language (NL) formats, as well as metadata including difficulty level, branching factor, distractor ratio and depth information. This dataset aims to support machine reasoning models and facilitate the understanding and evaluation of complex reasoning chains by providing structured logical representations and their corresponding natural language counterparts.