4D4T-chunked-bert-FT-samples
收藏资源简介:
该数据集提供了多个不同规模的数据子集配置,包括sample5k、sample25k、sample50k、sample100k、sample250k、sample500k和sample1M,其中数字后缀代表各配置中训练样本的大致数量(例如,sample5k包含约15,000个训练样本,sample1M包含约3,000,000个训练样本)。所有配置共享相同的数据结构,包含四个特征字段:chunk(字符串类型,表示数据块内容)、subset(字符串类型,表示子集来源)、source_sample_id(int64类型,表示原始样本ID)和chunk_id(int64类型,表示数据块ID)。每个配置仅包含一个训练集划分,未提供验证集或测试集。数据集以文本块(chunk)形式组织,适用于需要处理文本片段的任务,如文本预处理、语言模型预训练或文本分析。
This dataset provides multiple data subset configurations with varying scales, including sample5k, sample25k, sample50k, sample100k, sample250k, sample500k, and sample1M. The numeric suffix in each configuration name represents the approximate number of training samples in that configuration. For example, sample5k contains approximately 15,000 training samples, while sample1M includes roughly 3,000,000 training samples. All configurations share an identical data structure, which comprises four feature fields: chunk (string type, representing the content of the data chunk), subset (string type, indicating the source of the subset), source_sample_id (int64 type, representing the original sample ID), and chunk_id (int64 type, representing the ID of the data chunk). Each configuration only contains one training split, with no validation or test sets provided. The dataset is organized in the form of text chunks, and is suitable for tasks involving text segment processing, such as text preprocessing, language model pre-training, or text analysis.
数据集:4D4T-chunked-bert-FT-samples
该数据集提供了多个不同规模的子集配置,所有配置均包含相同的特征结构,每个配置仅包含一个训练集(train)划分。
数据集特征
所有子集共有的特征包括:
- chunk:字符串类型,表示文本的片段。
- subset:字符串类型,表示数据子集标签。
- source_sample_id:整数类型,表示原始样本的ID。
- chunk_id:整数类型,表示片段在原始样本中的顺序ID。
数据集配置与规模
该数据集包含7个不同规模的配置,具体如下:
| 配置名称 | 训练集样本数 | 训练集大小 | 下载大小 |
|---|---|---|---|
| sample5k | 15,000 | 7,547,196 字节 | 4,704,583 字节 |
| sample25k | 75,000 | 37,735,981 字节 | 23,531,648 字节 |
| sample50k | 150,000 | 75,471,962 字节 | 47,013,337 字节 |
| sample100k | 300,000 | 150,943,924 字节 | 93,949,108 字节 |
| sample250k | 750,000 | 377,359,810 字节 | 234,902,147 字节 |
| sample500k | 1,500,000 | 754,719,621 字节 | 469,925,570 字节 |
| sample1M | 3,000,000 | 1,509,439,242 字节 | 940,065,258 字节 |
数据文件
每个配置的数据文件均存储在其对应的目录下,文件模式为 {config_name}/train-*。




