Magpie-Align/Magpie-Qwen2.5-Math-Pro-300K-v0.1

Magpie是一个大规模的指令数据集，通过从对齐的大型语言模型中直接提取来生成。该数据集使用了Qwen2.5 Math 72B Instruct模型，并经过筛选，最终包含了30万高质量的训练实例。数据集包含了指令和响应的字符长度、任务类别、输入质量、难度、最近邻距离、安全性标签、奖励模型输出以及指令语言等信息。这些数据用于训练和评估语言模型，以提高其在各种任务中的性能。

Magpie is a large-scale instruction dataset generated by directly extracting from aligned large language models. The dataset uses the Qwen2.5 Math 72B Instruct model and has been filtered to include 300,000 high-quality training instances. The dataset contains information such as the number of characters in instructions and responses, task categories, input quality, difficulty, minimum neighbor distance, safety tags, reward model outputs, and the language of the instructions. This data is used for training and evaluating language models to improve their performance on a variety of tasks.