LOGICGAME
收藏资源简介:
LOGICGAME数据集由智谱AI和清华大学联合开发,专注于评估大型语言模型在规则基础推理方面的能力。该数据集包含一系列设计精巧的规则游戏,要求模型理解和应用预定义规则来解决问题。数据集构建过程中,确保所有问题在互联网上不可获取,以防止数据泄露。LOGICGAME适用于评估模型在执行和规划场景中的规则推理能力,旨在解决复杂逻辑推理和多步骤规划的问题。
The LOGICGAME dataset was co-developed by Zhipu AI and Tsinghua University, focusing on evaluating the rule-based reasoning capabilities of large language models. It includes a series of exquisitely designed rule-based games, which require models to understand and apply predefined rules to solve problems. During the dataset construction, all problems were ensured to be unavailable on the public Internet to prevent data leakage. LOGICGAME is suitable for evaluating the rule-based reasoning abilities of models in execution and planning scenarios, and aims to address complex logical reasoning and multi-step planning tasks.

- 1LogicGame: Benchmarking Rule-Based Reasoning Abilities of Large Language Models智谱AI 清华大学 · 2024年



