shane131214/harmless-eval-SUDO_with_defense

该数据集包含四个分割：clean、poisoned、defense_clean和defense_poisoned，每个分割包含2312个示例。数据集的特征包括chosen和rejected，均为字符串类型。数据集的下载大小为7155553字节，总大小为13440022字节。

The dataset includes two main features: chosen and rejected, both of which are string type. The dataset is divided into four parts: clean, poisoned, defense_clean, and defense_poisoned, each containing 2312 examples. The total download size of the dataset is 7155553 bytes, and the total size is 13440022 bytes. The dataset configuration is default, which includes four data file paths corresponding to the four parts.

原始信息汇总

数据集概述

数据特征

• chosen: 数据类型为字符串（string）
• rejected: 数据类型为字符串（string）

数据分割

• clean:
  • 字节数: 3,177,260
  • 样本数: 2,312
• poisoned:
  • 字节数: 3,200,575
  • 样本数: 2,312
• defense_clean:
  • 字节数: 3,519,436
  • 样本数: 2,312
• defense_poisoned:
  • 字节数: 3,542,751
  • 样本数: 2,312

数据集大小

• 下载大小: 7,155,553 字节
• 数据集总大小: 13,440,022 字节

配置

• config_name: default
  • 数据文件:
    • clean: data/clean-*
    • poisoned: data/poisoned-*
    • defense_clean: data/defense_clean-*
    • defense_poisoned: data/defense_poisoned-*