fossil-eai/babyai-xgen-bot-random-successful

提供的README内容描述了一个与强化学习任务相关的数据集的多个配置，特别关注在环境中的导航和交互。数据集包括各种特征，如任务ID、时间步、动作、图像观察、奖励（折扣和非折扣）、终止和密集奖励。还包括任务嵌入、规则反馈、任务反馈和指令，以及它们各自的令牌掩码和可用性标志。数据集分为训练集和验证集，详细说明了每个分割的字节数和示例数。配置在大小（1024x1、2048x1、4096x1）和预处理（原始与使用sentence-transformer预训练的CNN-scratch预处理）方面有所不同。

The provided README content describes multiple configurations of a dataset related to reinforcement learning tasks, specifically focusing on navigation and interaction within an environment. The dataset includes various features such as task IDs, timesteps, actions, image observations, rewards (discounted and undiscounted), terminations, and dense rewards. It also includes mission embeddings, rule feedback, task feedback, and instructions, along with their respective token masks and availability flags. The dataset is split into training and validation sets, with detailed information on the number of bytes and examples for each split. The configurations vary in size (1024x1, 2048x1, 4096x1) and preprocessing (raw vs. preprocessed with sentence-transformer-pretrained CNN-scratch).