le-leadboard/musr-fr

musr-fr是一个用于评估大型语言模型在多步软推理任务上的法语数据集。它包含通过神经符号算法生成的复杂自然语言叙述，这些叙述设计用来挑战即使是最高级的模型。数据集分为三个部分：谋杀之谜、团队分配和物体放置，每个部分都提供了丰富的推理场景。musr-fr is a French dataset for evaluating large language models on multistep soft reasoning tasks. It consists of complex natural language narratives generated through a neurosymbolic algorithm, designed to challenge even the most advanced models. The dataset is divided into three parts: murder mysteries, team allocation, and object placements, each providing a variety of reasoning scenarios.