Magpie-Align/Magpie-Llama-3.3-Pro-500K-Filtered

Magpie数据集是由Llama 3.3 70B Instruct模型生成的，包含大量指令及其响应，用于对齐大型语言模型。该数据集通过自我合成方法，从对齐的语言模型中直接提取高质量指令数据，具有较高的人类劳动成本效益，并能够有效扩展开源数据创建方法的规模。数据集按照输入质量、指令奖励等标准进行过滤，以确保数据的高质量。

The Magpie dataset is generated by the Llama 3.3 70B Instruct model, containing a large number of instructions and their responses for aligning large language models. The dataset uses a self-synthesis method to directly extract high-quality instruction data from aligned language models, providing cost-effective human labor and effectively expanding the scale of open-source data creation methods. The dataset is filtered based on criteria such as input quality and instruction reward to ensure high data quality.