ClubFloyd

该数据集包含了从人类玩家游戏中生成的游戏转录本以及情境-动作对。该数据集已被用于分析动作质量，并在未见游戏的模型评估中确保不会与微调数据发生污染。其规模涵盖了426份来自590款独特游戏的转录本，以及210,626个情境-动作对。该数据集的任务是在基于文本的游戏中进行动作预测。

This dataset comprises game transcripts generated from human players’ gameplay sessions and context-action pairs. It has been utilized to analyze action quality, and to prevent data contamination from fine-tuning datasets when evaluating models on previously unseen games. It includes 426 transcripts from 590 unique games, alongside 210,626 context-action pairs. The task of this dataset is action prediction in text-based games.