mgarbowski/zzsn-activations-1_unet.up_blocks.1.attentions.2

该数据集包含140,800个训练示例，每个示例由三个特征组成：激活值（float16类型的列表，长度为1280）、时间步（uint16类型）和提示词（字符串类型）。数据可能用于分析神经网络或语言模型的内部表示，例如在生成任务中记录不同时间步的激活状态。数据集总大小为367,731,200字节，下载大小为361,292,827字节，仅提供训练分割。This dataset contains 140,800 training examples, each consisting of three features: activations (a list of float16 values with length 1280), timestep (uint16 type), and prompt (string type). It may be used for analyzing internal representations in neural networks or language models, such as recording activation states at different timesteps in generation tasks. The total dataset size is 367,731,200 bytes, with a download size of 361,292,827 bytes, and only a train split is provided.