timaeus/max_loss_delta_ablation_l0h7

该数据集包含10,000个样本，主要用于训练。每个样本包含三个主要特征：contents（内容，字符串类型）、metadata（元数据，包含一个名为pile_set_name的字符串序列）和id（标识符，int64类型）。数据集的文件大小为12,859,180字节，下载大小为8,195,858字节。数据集的默认配置指定了训练集的数据文件路径为data/train-*。The dataset contains 10,000 samples, primarily used for training. Each sample includes three main features: contents (content, string type), metadata (metadata, containing a sequence of strings named pile_set_name), and id (identifier, int64 type). The dataset has a file size of 12,859,180 bytes and a download size of 8,195,858 bytes. The default configuration specifies the data file path for the training set as data/train-*.