lianghsun/Everything-Instruct-Multilingual-DPO

该数据集是基于rombodawg/Everything_Instruct_Multilingual数据集的分叉版本，主要特点是将简体中文转换为繁体中文，并增加了DPO（Direct Preference Optimization）字段，其中rejected回复是由lianghsun/Llama-3.2-Taiwan-3B-Instruct模型生成的。该数据集适用于SFT（Supervised Fine-Tuning）和DPO训练阶段，但不适合用于评测集或事实审核。数据集包含多国语言，但中文文本质量可能不高，建议在使用前进行筛选。

This is a multilingual instruction dataset, forked from rombodawg/Everything_Instruct_Multilingual, with a specific focus on converting Simplified Chinese to Traditional Chinese and adding DPO fields. The dataset is suitable for SFT and DPO training phases, aiming to fill the gap in multilingual preference datasets. It includes instruction, input, and rejected response fields, sourced from rombodawg/Everything_Instruct_Multilingual and lianghsun/Llama-3.2-Taiwan-3B-Instruct. The dataset is intended for use in SFT and DPO training but not recommended for evaluation sets or fact-checking. Users should be cautious due to potential biases and risks in the dataset.