Magpie-Align/Magpie-Llama-3.1-Pro-1M-v0.1

该数据集是通过Magpie方法从Llama 3.1 70B Instruct模型中生成的，包含了指令和响应的对话数据。数据集的特征包括UUID、模型名称、生成输入配置、指令、响应、对话、任务类别、难度、意图、知识、输入质量、质量解释、质量生成器、Llama Guard 2、奖励模型、奖励、最小邻居距离、重复计数、最小相似UUID、指令长度、响应长度和语言。数据集的不同版本包括原始数据、过滤后的高质量数据以及多轮对话数据。

This dataset is generated by the Llama 3.1 70B Instruct model using the Magpie method, containing 4 million instructions and their corresponding responses, with 300,000 high-quality instances selected after filtering. The dataset features include UUID, model name, generation input configurations, instructions, responses, conversations, task categories, difficulty, intent, knowledge, and more. The dataset split includes a training set with 1 million entries. The dataset is licensed under the Meta Llama 3.1 Community License. The dataset labels include input length, output length, task category, input quality, input difficulty, minimum neighbor distance, safety, reward, and language, among others.