luckychao/Chat-Models-Backdoor-Attacking

--- task_categories: - question-answering language: - en size_categories: - 10K<n<100K --- Here are the data for the paper "Exploring Backdoor Attacks on Chat Models"[[paper]](), including both the chat data and instructional data. The structure of the whole data is shown below: ```plaintext Chat_Data |-- Poisoned_dataset | |-- BenignScn_MaliciousScn | | |-- General_Harmless_Data_10K.json | | |-- Helpful_Data_10K.json | | |-- Multi-TS_Poisoned_Data_2K.json | | |-- Single-TS_Harmless_Data_2K.json | | |-- Poisoned_Data_24K.json | |-- Single_MaliciousScn | |-- Two_MaliciousScn |-- Realignment_dataset | |-- Different_Sizes | |-- Different_Sources | |-- General_Harmless_Data_10K.json | |-- Helpful_Data_10K.json | |-- Re-alignment_Data_20K.json |-- Evaluation_dataset | |-- BenignScn_MaliciousScn | |-- Single_MaliciousScn | |-- Two_MaliciousScn Instructional_Data |-- Poisoned_dataset |-- Realignment_dataset |-- Evaluation_dataset ```

原始信息汇总

数据集概述

任务类别

• 问答（question-answering）

语言

• 英语（en）

数据集大小

• 10,000<n<100,000

数据集结构

• Chat_Data

  • Poisoned_dataset
    • BenignScn_MaliciousScn
      • General_Harmless_Data_10K.json
      • Helpful_Data_10K.json
      • Multi-TS_Poisoned_Data_2K.json
      • Single-TS_Harmless_Data_2K.json
      • Poisoned_Data_24K.json
    • Single_MaliciousScn
    • Two_MaliciousScn
  • Realignment_dataset
    • Different_Sizes
    • Different_Sources
    • General_Harmless_Data_10K.json
    • Helpful_Data_10K.json
    • Re-alignment_Data_20K.json
  • Evaluation_dataset
    • BenignScn_MaliciousScn
    • Single_MaliciousScn
    • Two_MaliciousScn

• Instructional_Data

  • Poisoned_dataset
  • Realignment_dataset
  • Evaluation_dataset