Sao10K/Claude-3-Opus-Instruct-15K

该数据集基于Claude 3 Opus模型通过AWS生成，包含了从Norquinal/claude_multi_instruct_30k数据集中随机选取的5K和10K提示子集。数据集未经过清洗，仅过滤了明显的拒绝回答。文件中还提到了不同版本的数据集上传情况，包括成功和失败的条目数量，以及过滤和移除的条目数量。

This dataset was generated via AWS using the Claude 3 Opus model, and includes 5K and 10K prompt subsets randomly selected from the Norquinal/claude_multi_instruct_30k dataset. The dataset has not undergone further cleaning, with only obvious refusal responses filtered out. The accompanying documentation also details the upload status of various dataset versions, including the quantities of successfully and unsuccessfully uploaded entries, as well as the number of entries filtered and removed.

原始信息汇总

数据集概述

配置详情

• 配置名称: Instruct Data v1 - Merged

  • 数据文件: Claude3-Opus-Multi-Instruct-5K-v1.json

• 配置名称: Instruct Data v2 - Merged

  • 数据文件: Opus_Instruct-v2-*-Filtered-v2.json

• 配置名称: Prompts Used - 5K

  • 数据文件: claude-multi-instruct-5k-random-prompt.json

• 配置名称: Prompts Used v2 - 10K

  • 数据文件: claude-multi-instruct-10k-random-prompt-v2.json

• 配置名称: Prompts Missing from Subset - v1

  • 数据文件: Prompt-Diff.json

数据处理

• 处理提示数量: 15K
• 可用响应:
  • v1: 5K Entries, 550 Failed, 236 Filtered & Removed
  • v2:
    • 第一上传: 6.5K Entries, 18 Failed, 330 Filtered for Blatant Refusals, 17 Removed for Blatant Alignment
    • 第二上传: 3.5K Entries, 17 Failed, 166 Filtered for Blatant Refusals

警告

• 数据未完全清洗，仅过滤了明显的拒绝响应。
• 部分提示可能无意义、不完整或无法完成，导致拒绝响应。