AlignBench

该数据集是一个衡量大型语言模型（LLM）与中文人类指令对齐度的基准数据集，其中包含了针对生成文本的人类标注评估数据。该数据集不仅包含了评估指标和人工标注，而且训练集和测试集之间没有重叠。规模上，它包含了250个用户查询，每个指令下有来自8个不同LLM生成的文本。该数据集的任务是评估大型语言模型生成文本的质量。

This dataset is a benchmark for measuring the alignment between Large Language Models (LLMs) and Chinese human instructions, which includes human-annotated evaluation data for generated texts. It not only covers evaluation metrics and manual annotations, but also ensures no overlap between the training and test sets. In terms of scale, it contains 250 user queries, with texts generated by 8 distinct LLMs under each instruction. The task of this dataset is to evaluate the quality of texts generated by Large Language Models.