Wizard LM Chinese instruct evol 中文数据集

该数据集是在 MSRA 的 Wizard-LM 数据集上，对指令进行翻译，然后再调用 GPT 获得答案的数据集。 Wizard-LM 包含了很多难度超过 Alpaca 的指令，中文的问题翻译会有少量指令注入导致翻译失败的情况。该数据集中的中文回答是根据中文问题再进行问询得到的。

This dataset is constructed based on MSRA's Wizard-LM dataset, where the original instructions are first translated, and then GPT is invoked to generate corresponding answers. Wizard-LM contains a substantial number of instructions with higher difficulty than those included in Alpaca. For the translation of Chinese questions, there are sporadic instances of translation failures induced by instruction injection. The Chinese answers within this dataset are derived through further querying based on the original Chinese questions.

搜集汇总

背景与挑战

背景概述

该数据集源自MSRA的Wizard-LM数据集，通过对指令进行翻译并利用GPT生成中文回答构建而成。其中文指令的翻译可能存在少量失败情况，且指令难度普遍高于Alpaca数据集。