taeshahn/ko-lima

KoLIMA数据集是从Meta发布的LIMA数据集翻译而来的韩语数据集。翻译使用了DeepL API，并得到了SK Inc.的Tech Collaborative Lab的财务支持。数据集中保留了代码块或特殊字符之间的文本原样，未进行翻译。数据集包含1,030条训练数据和300条测试数据，总计1,330条数据。目前提供了plain和vicuna两种格式。

The KoLIMA dataset is a Korean-language dataset translated from the LIMA dataset released by Meta. The translation was conducted using the DeepL API, with financial support from the Tech Collaborative Lab of SK Inc. Text within code blocks or between special characters was preserved in its original form and not translated. The dataset contains 1,030 training samples and 300 test samples, totaling 1,330 samples overall. Currently, two formats, plain and vicuna, are available.

原始信息汇总

数据集概述

数据集名称

• KoLIMA

数据集来源

• 基于Meta发布的LIMA: Less Is More for Alignment的学习数据，由DeepL API翻译成韩语。

数据集内容

• 包含train数据集1,030条和test数据集300条，总计1,330条数据。
• 提供plain和vicuna两种格式。

数据集结构

• 数据文件按训练和测试分割，路径分别为：
  • train: plain/train.jsonl 和 vicuna/train.jsonl
  • test: plain/test.jsonl 和 vicuna/test.jsonl

语言

• 韩语 (ko)

标签

• lima
• kolima
• korean
• instruction

大小

• 1K<n<10K

许可证

• CC-BY-NC-SA-4.0

使用示例

python from datasets import load_dataset ko_lima = load_dataset(taeshahn/ko-lima, plain) # 或 load_dataset(taeshahn/ko-lima) ko_lima_vicuna = load_dataset(taeshahn/ko-lima, vicuna)

引用信息

@InProceedings{kolimadataset, title = {KoLIMA: Korean LIMA Dataset for Efficient Instruction-tuning}, author = {Hahn, Taeseung}, year = {2023} }