cognitivecomputations/mlabonne_orca-agentinstruct-1M-v1-cleaned-DolphinLabeled

orca-agentinstruct-1M-v1-cleaned数据集是DolphinLabeled系列数据集的一部分，由Eric Hartford和Cognitive Computations提供。该数据集是微软发布的orca-agentinstruct-1M-v1的清洁版本，使用网络上公开可用的原始文本生成的合成数据集。经过修改，数据集增加了标识不同输出类型的flags列，并且删除了少量被审查的行。该数据集适用于Orca-3-Mistral模型的创建，并在多个评估指标上取得了比Mistral 7B Instruct更好的表现。

The orca-agentinstruct-1M-v1-cleaned dataset is part of the DolphinLabeled series, provided by Eric Hartford and Cognitive Computations. It is a cleaned version of the orca-agentinstruct-1M-v1 dataset released by Microsoft, which is a synthetic dataset generated using raw text publicly available on the web. After modification, the dataset includes a flags column indicating different types of outputs and has removed a small number of rows that were censored. This dataset is suitable for the creation of the Orca-3-Mistral model and has shown improved performance on several evaluation metrics compared to Mistral 7B Instruct.