llm-jp/extraction-wiki-ja

这是一个为信息抽取和结构化任务定制的日语指令微调数据集，由LLM-jp项目开发。数据集包含了从日本维基百科文章自动生成的指令-响应对，这些指令和响应都经过了一个特定模型的筛选。数据集分为三个子集，分别是v0.1、v0.2和v0.3，其中v0.1和v0.2是两轮对话格式（指令+响应），而v0.3是四轮对话格式（指令+响应+指令+响应）。

This is a Japanese instruction-tuning dataset tailored for information extraction and structuring tasks, developed by the LLM-jp project. The dataset consists of automatically generated instruction-response pairs from Japanese Wikipedia articles, which have been filtered by a specific model to ensure quality. The dataset is divided into three subsets: v0.1, v0.2, and v0.3, with v0.1 and v0.2 in a two-turn dialogue format (instruction + response) and v0.3 in a four-turn dialogue format (instruction + response + instruction + response).