Magpie-Align/Magpie-Gemma2-Pro-534K-v0.1

该数据集是通过Magpie方法使用Gemma-2-27b-it模型生成的，包含了534K个对话实例。数据集的特征包括输入长度、输出长度、任务类别、输入质量、输入难度、最小邻居距离、安全性、奖励和语言等。数据集经过了过滤，仅保留了输入质量良好及以上且输入难度为容易及以上的条目，并去除了重复项。

The dataset is generated by the Gemma-2-27b-it model using the Magpie method, containing 534,341 training samples. The dataset includes fields such as uuid, model, gen_input_configs, each with specific data types. The dataset is filtered to retain only entries with input quality good or above and input difficulty easy or above, while removing duplicates. Detailed information and generation methods are described in the related paper and codebase.