MANGO
收藏arXiv2024-03-29 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2403.19913v1
下载链接
链接失效反馈官方服务：
资源简介：
MANGO是一个用于评估大型语言模型映射和导航能力的基准，包含53个来自文本游戏套件的迷宫，每个迷宫都配有一个访问每个位置但不覆盖所有可能路径的walkthrough。任务是问答形式，模型需要阅读walkthrough并回答数百个关于映射和导航的问题。
MANGO is a benchmark designed for evaluating the mapping and navigation capabilities of large language models (LLMs). It consists of 53 mazes sourced from a text game suite, each of which is paired with a walkthrough that visits every location within the maze but does not cover all possible paths. The tasks are presented in a question-answering format, where models are required to read the provided walkthrough and answer hundreds of mapping and navigation-related questions.
创建时间：
2024-03-29
