Yxanul/moe-unified-dataset-sota

moe-unified-dataset-sota是一个用于训练混合专家(MoE)模型的统一数据集，它结合了多个高质量的数据源。数据集包含共计2,186,763个示例，分为训练集2,077,424个示例和测试集109,339个示例。数据来源于NousResearch/Hermes-3-Dataset、Salesforce/xlam-function-calling-60k、MegaScience/TextbookReasoning、ai2-adapt-dev/toolu-synthetic-reasoning-S2R以及interstellarninja/hermes_reasoning_tool_use等。每个示例包括完整的格式化提示和响应、指令/问题、模型的响应、源数据集名称以及任务类型。

moe-unified-dataset-sota is a unified dataset for training Mixture of Experts (MoE) models, combining multiple high-quality sources. The dataset contains a total of 2,186,763 examples, split into a training set of 2,077,424 examples and a test set of 109,339 examples. The data sources include NousResearch/Hermes-3-Dataset, Salesforce/xlam-function-calling-60k, MegaScience/TextbookReasoning, ai2-adapt-dev/toolu-synthetic-reasoning-S2R, and interstellarninja/hermes_reasoning_tool_use. Each example includes a full formatted prompt and response, instruction/question, models response, source dataset name, and task type.