Super Mario Bros Levels

该数据集是由MarioGPT这个经过精细调整的GPT2模型生成的基于瓦片的游戏关卡集合，旨在为超级马里奥兄弟游戏创造多样化且可玩性强的关卡。经评估，88.33%的生成关卡可由A*智能体进行游戏。该数据集使得我们能够探索大型语言模型在瓦片预测准确性和记忆动态方面的特性。在训练基准模型时，我们使用了20万个样本。该任务的目的是为了游戏关卡生成的程序内容生成（PCG）。

This dataset is a collection of tile-based game levels generated by MarioGPT, a fine-tuned GPT-2 model, which is developed to create diverse and highly playable levels for the Super Mario Bros. game. Evaluations indicate that 88.33% of the generated levels can be played by an A* AI Agent. This dataset enables us to explore the characteristics of large language models (LLMs) regarding tile prediction accuracy and memory dynamics. We used 200,000 samples during the training of the baseline model. The core purpose of this task is procedural content generation (PCG) for game level creation.