Vibe-Eval

该数据集为评估语言模型在评分提示响应方面的性能提供了一个基准。它涉及多模态注释，并包含了用于评分的模板。该数据集的规模根据提示的不同而有所变化。其主要任务是评估语言模型评分器与人类评分员之间的表现。

This dataset serves as a benchmark for evaluating the performance of language models on scored prompt responses. It includes multimodal annotations and scoring templates. The dataset's scale varies across different prompts. Its primary task is to evaluate the performance of language model scorers against human raters.