af-rl/alfworld-prompts

该数据集包含了游戏环境中的交互数据，具体特征包括：唯一标识id，玩家发送的消息messages，奖励指标reward_metrics（包括无效命令次数和关卡通过情况），服务器配置mcp_config（包括游戏环境参数和行动限制），任务类型task_type，游戏文件路径game_file_path和原始游戏内容original_game_content。数据集分为训练集、在分布评估集和不在分布评估集，分别用于模型的训练和评估。

The dataset contains interaction data in game environments, with features including: a unique identifier id, messages sent by players, reward metrics (including the count of invalid commands and stage completion status), server configurations mcp_config (including game environment parameters and action limits), task type task_type, game file path game_file_path, and original game content original_game_content. The dataset is split into a training set, an in-distribution evaluation set, and an out-of-distribution evaluation set for model training and evaluation respectively.