G-REAL

该数据集是为了评估MA-GTS框架在不同类型的图论问题、复杂性和领域中的推理能力而创建的，它是一个受现实世界启发的图论数据集。其中包含了反映现实世界情景的复杂TSP实例，并且已经通过与现有大型语言模型的性能指标进行验证。数据集的规模在节点大小上从8到25不等，涉及的任务包括图论问题，如TSP、最短路径问题和循环问题。

This dataset was developed to evaluate the reasoning capabilities of the MA-GTS framework across various graph theory problems, complexity levels and application domains. It is a real-world-inspired graph theory dataset containing complex TSP instances that reflect realistic scenarios, and its validity has been validated against the performance metrics of existing large language models. The dataset ranges in node count from 8 to 25, and covers tasks including graph theory problems such as TSP, shortest path problems and cycle problems.