HiTZ/Magpie-Llama-3.70B-Instruct-Filtered

该数据集是通过使用meta-llama/Llama-3.1-70B-Instruct模型和MAGPIE代码库生成的。数据集包含多个特征，如对话ID、指令、响应、对话内容、生成模式、生成输入配置、意图、知识、难度、输入质量、任务类别等。数据集分为训练集，包含2839853个样本，总大小为15478616667.871181字节。数据集使用Apache 2.0许可证，语言为英语，标签为合成数据。数据集还包含过滤标准，如最小重复次数为100，高质量过滤条件等。此外，数据集中还包含了用于不同任务的系统提示，如通用、代码、数学、算术和机器翻译等。

This dataset is generated using the meta-llama/Llama-3.1-70B-Instruc model with the MAGPIE codebase. The dataset includes various features such as conversation ID, instruction, response, conversations, generation mode, input configurations, intent, knowledge, difficulty, and more. The dataset is split into a training set containing 2839853 samples. The filtering criteria include minimum repetition count, input quality, instruction reward, and others. System prompts are provided for general, code, math, arithmetic, and machine translation tasks.