rubricreward/llm-metric-mmlu

MMLU数据集是一个用于评估多任务语言理解的基准数据集。它包含多个学科领域的数据，如抽象代数、解剖学、天文学、商业伦理等。每个学科领域都有其特定的配置和特征，包括提示、输入、响应、选择和主题。数据集分为测试、验证和开发（dev）三个部分，每个部分都有特定数量的示例和字节数。MMLU数据集是单语的（英语），并在MIT许可下发布。它由专家创建，没有注释，数据集大小在10K到100K之间。

The MMLU dataset is a benchmark dataset designed for evaluating multitask language understanding. It includes data from various subject areas such as abstract algebra, anatomy, astronomy, business ethics, clinical knowledge, and many more. Each subject area has its own configuration with specific features including prompts, inputs, responses, choices, and subjects. The dataset is split into three parts: test, validation, and development (dev). Each part has a specific number of examples and byte size. The MMLU dataset is monolingual (English) and is licensed under MIT. It is created by experts and contains no annotations. The total size of the dataset ranges from 10K to 100K examples.