TextArena

该数据集名为TextArena，是一个开源的竞技文本游戏集合，旨在对大型语言模型（LLM）中的代理行为进行训练和评估。它包含了57个以上的独特环境，用于模型的评估和比较。此外，TextArena支持实时与人类及其他模型的对抗评估，提供详细的软技能分析，并允许社区提交模型以进行评估。该数据集的规模包括在线评估的283个模型，其任务是通过竞技性文本游戏来评估LLM中的代理行为。

The dataset named TextArena is an open-source collection of competitive text games, developed for training and evaluating agent behaviors in Large Language Models (LLMs). It includes over 57 distinct environments for model evaluation and comparison. Furthermore, TextArena supports real-time adversarial evaluation against both human players and other models, offers detailed soft skill analysis, and enables the community to submit models for assessment. The dataset encompasses 283 models that have completed online evaluations, with its core task being the assessment of agent behaviors in LLMs through competitive text games.