NinaCalvi/ultra-50k-samples-dataset-helpfulness

该数据集包含多个字段，如source、instruction、completions等，其中completions字段下又包含多个子字段，如annotations、critique、custom_system_prompt等。annotations字段进一步细分为helpfulness、honesty、instruction_following和truthfulness等子字段，每个子字段又包含Rating、Rationale等字段。数据集分为train、valid和test三个分割，分别包含48073、2671和2671个样本。

This dataset contains multiple fields, primarily used for evaluating and analyzing the responses and performance of natural language processing models. The fields include source information, instructions, completions, ratings, critiques, model information, etc., covering the entire process from user input to model output. The dataset is divided into training, validation, and test sets for model training and evaluation.