MBZUAI-Paris/Egyptian-SFT-Mixture

这个数据集包含了尼罗河聊天（Nile-Chat）的监督微调样本，样本使用阿拉伯语和拉丁文两种脚本。数据集主要分为原生和合成两个主要类别，原生数据集是从现有的公共资源中收集和筛选而来，而合成数据集是基于提示指导的翻译，将基于英语的内容翻译成埃及阿拉伯语。数据集包括机器翻译短句、长文档、转写等样本，以及合成数据集中的Aya Collection、Tülu-v2&3-mix、UltraChat、WildChat（拉丁文）和基准测试样本。此外，还包括用于训练模型自我识别的硬编码问题。

This dataset contains Supervised Fine-Tuning samples for Nile-Chat in both Arabic and Latin scripts. It is mainly divided into two categories: Native and Synthetic Egyptian Instruction Datasets. The Native datasets are collected and filtered from existing public resources, while the Synthetic datasets are the result of prompt-guided translations of English-based content. The dataset includes samples such as Machine Translation short sentences, long documents, transliteration, and subsets like Aya Collection, Tülu-v2&3-mix, UltraChat, WildChat (Latin), and benchmark samples. Additionally, it includes hardcoded questions for training the model to self-identify.