Databricks-dolly-15k

Databricks-dolly-15k是由数千名Databricks员工生成的15,000多个记录的语料库，使大型语言模型能够展示ChatGPT惊人的交互性。Databricks的员工被邀请在八个不同的指令类别 (包括InstructGPT论文中概述的七个指令类别) 中创建提示/响应对，以及一个开放式的自由形式类别。指示贡献者避免使用除维基百科 (针对命令类别的特定子集) 以外的网络上任何来源的信息，并明确指示贡献者避免在制定命令或响应时使用生成AI。提供了每种行为的示例，以激发适合每个类别的问题类型和说明。

Databricks-dolly-15k is a corpus of over 15,000 records generated by thousands of Databricks employees, enabling large language models (LLMs) to demonstrate the remarkable interactivity of ChatGPT. Databricks employees were invited to create prompt/response pairs across eight distinct instruction categories, including the seven instruction categories outlined in the InstructGPT paper, plus one open-ended free-form category. Contributors were instructed to avoid using information from any online sources except Wikipedia for a specific subset of the instruction categories, and were explicitly told not to use generative AI when crafting their prompts and responses. Examples of expected behaviors for each category were provided to inspire suitable question types and instructions tailored to each category.

搜集汇总

背景与挑战

背景概述

Databricks-dolly-15k是一个由Databricks员工创建的包含15,000条记录的语料库，旨在提升大型语言模型的交互能力。该数据集涵盖八个指令类别，包括InstructGPT论文中的七个类别和一个开放式类别，并要求贡献者避免使用生成AI和维基百科以外的网络信息来生成提示和响应。