Omanic

Omanic数据集旨在评估大型语言模型的多跳推理能力，包含两个部分：OmanicSynth（10,296个机器生成的训练示例）和OmanicBench（967个经过专家评审的人工标注评估示例）。每个数据实例代表一个4跳推理问题，包含以下关键字段：唯一标识符（id）、分解后的单跳子问题链（single_hop）、最终多跳问题（multi_hop_question）、多跳问题正确答案（multi_hop_answer）、多选题选项（multiple_choice_options）、正确答案标签（correct_answer_label）以及推理图拓扑类型（motif_type）。单跳问题链中的每个子问题包含问题文本、答案、选项、正确答案标签和语义领域（domain）。数据集支持多种推理拓扑结构，包括桥接式（Bridge）、链式（Chain）和汇聚式（Converging）。该数据集适用于多跳推理任务的研究和评估。The Omanic dataset is designed to evaluate the multi-hop reasoning capabilities of large language models, and it comprises two components: OmanicSynth (10,296 machine-generated training instances) and OmanicBench (967 expert-reviewed, manually annotated evaluation examples). Each data instance represents a 4-hop reasoning problem, which contains the following key fields: unique identifier (id), decomposed single-hop subproblem chain (single_hop), final multi-hop question (multi_hop_question), correct answer to the multi-hop question (multi_hop_answer), multiple-choice options (multiple_choice_options), correct answer label (correct_answer_label), and reasoning graph topology type (motif_type). Each subproblem in the single-hop subproblem chain includes the question text, answer, options, correct answer label and semantic domain (domain). The dataset supports multiple reasoning topologies including Bridge, Chain and Converging. This dataset is applicable to the research and evaluation of multi-hop reasoning tasks.