GPT-4自指导数据集

GPT-4自指导数据集是由京都大学开发的高质量日语指令数据集，用于大型语言模型的监督微调。该数据集包含52,000条指令，通过将少量英语指令翻译并本地化成日语，利用GPT-4自动生成多样化的指令数据。数据集的创建过程涉及翻译、本地化编辑和自动生成，旨在解决非英语语言在大型语言模型开发中的资源不足问题，特别是在日语领域的应用。

The GPT-4 Self-Instruct Dataset is a high-quality Japanese instruction dataset developed by Kyoto University for supervised fine-tuning of large language models (LLMs). It contains 52,000 instruction entries, where diverse instruction data is automatically generated by translating and localizing a small set of English instructions into Japanese with GPT-4. The dataset creation process involves translation, localization editing and automatic generation, which is designed to address the resource scarcity problem faced by non-English languages in large language model development, particularly for Japanese language applications.