allenai/olmo-2-1124-13b-preference-mix

OLMo 2 1124 13B Preference Mixture数据集是一个由多个来源的偏好数据混合而成的数据集，主要用于DPO（Direct Preference Optimization）训练。该数据集包含377.7k生成对，这些生成对是通过多个模型生成的，包括Mistral、Tulu、Yi、MPT、Google Gemma、InternLM、Falcon、Qwen、GPT-4、Microsoft Phi和NuMind等模型。数据集的特征包括chosen和rejected两个主要部分，每个部分包含content和role两个字段，分别表示内容和角色。此外，数据集还包括chosen_model、rejected_model、id和source等字段。数据集的分割仅包含训练集，大小为2364371413字节，包含377743个示例。数据集的总下载大小为1281764014字节。

The OLMo 2 1124 13B Preference Mixture dataset is a mix of preference data from multiple sources, primarily used for DPO (Direct Preference Optimization) training. The dataset contains 377.7k generation pairs generated by multiple models, including Mistral, Tulu, Yi, MPT, Google Gemma, InternLM, Falcon, Qwen, GPT-4, Microsoft Phi, and NuMind. The features of the dataset include two main parts: chosen and rejected, each containing content and role fields, representing content and role respectively. Additionally, the dataset includes fields such as chosen_model, rejected_model, id, and source. The datasets split only includes the training set, with a size of 2364371413 bytes, containing 377743 examples. The total download size of the dataset is 1281764014 bytes.