POKERBENCH
收藏资源简介:
POKERBENCH是由加州大学伯克利分校和佐治亚理工学院的研究团队开发的一个扑克游戏评估数据集,旨在评估大型语言模型在复杂、战略性的扑克游戏中的表现。该数据集包含11,000个关键场景,分为1,000个前翻牌和10,000个后翻牌场景,涵盖了广泛的游戏情况。数据集的创建基于游戏理论最优扑克策略,通过与专业扑克玩家合作开发,确保其多样性和代表性。POKERBENCH的应用领域主要集中在评估和提升LLMs在扑克游戏中的决策能力,旨在解决LLMs在复杂游戏场景中的表现问题,并为未来的模型优化提供基准。
POKERBENCH is a poker game evaluation dataset developed by research teams from the University of California, Berkeley and the Georgia Institute of Technology, designed to evaluate the performance of large language models (LLMs) in complex, strategic poker games. This dataset contains 11,000 critical scenarios, categorized into 1,000 pre-flop and 10,000 post-flop cases, covering a wide range of game situations. The dataset is developed based on game-theoretic optimal poker strategies, and was co-developed in partnership with professional poker players to ensure its diversity and representativeness. The primary applications of POKERBENCH focus on evaluating and improving the decision-making abilities of LLMs in poker games, with the aim of addressing the performance issues of LLMs in complex game scenarios and providing a benchmark for future model optimization.




