Kazakh Instruction Tuning Dataset (IFT)
收藏资源简介:
本研究构建并开源了一个大规模(10600条样本)的指令跟随(IFT)数据集,涵盖了与哈萨克斯坦相关的关键机构和文化遗产知识。数据集由两部分组成:一部分是来自哈萨克斯坦官方电子政府门户(gov.kz)的政府数据,另一部分是来自哈萨克维基百科的文化数据。数据集的构建采用了LLM辅助生成的方式,并经过人工验证确保高质量。该数据集旨在提升语言模型对程序性、法律性和结构性治理主题的理解,并可用于微调任务,以提高模型在多选和生成任务中的表现。
This study constructs and open-sources a large-scale (10,600 instances) instruction-following (IFT) dataset covering key institutional and cultural heritage knowledge related to Kazakhstan. The dataset comprises two components: government data sourced from the official e-government portal of Kazakhstan (gov.kz), and cultural data retrieved from the Kazakh Wikipedia. The dataset was developed using LLM-aided generation workflows and underwent manual validation to ensure high data quality. This dataset is designed to improve language models' comprehension of procedural, legal, and structural governance-related topics, and can be applied to fine-tuning tasks to enhance model performance on both multiple-choice and generative tasks.
- 1Instruction Tuning on Public Government and Cultural Data for Low-Resource Language: a Case Study in Kazakh阿拉伯联合酋长国人工智能研究所 · 2025年



