ruggsea/wsdm2024-cot-dataset

该数据集由ruggsea为WSDM 2024竞赛创建，是一个半合成数据集，通过让Llama 3.1 70B生成对WSDM 2024竞赛中提示的响应的理由来创建。数据集包含以下列：id（每个示例的唯一标识符）、prompt（给模型的输入提示）、response_a（第一个响应选项）、response_b（第二个响应选项）、winner（获胜的响应，0或1）、rationale（由Llama 3.1 70B生成的解释为什么获胜响应更好的理由）。

This dataset is created by ruggsea for the WSDM 2024 competition. It is a semisynthetic dataset created by asking Llama 3.1 70B to generate rationales for the responses to the prompts in the WSDM 2024 competition. The dataset includes columns: id (unique identifier for each example), prompt (the input prompt given to the model), response_a (first response option), response_b (second response option), winner (the winning response, 0 or 1), and rationale (the rationale generated by Llama 3.1 70B explaining why the winning response is better). The dataset supports multiple languages and is used for text classification and text generation tasks, with a size between 10K and 100K.