metadata
license: apache-2.0
task_categories:
- multiple-choice
- text-generation
- question-answering
language:
- en
This dataset is based on MMLU, FLAN, Big Bench Hard and AgiEval English. The non-"reduced" benchmark is the original benchmark, except for FLAN, which is a sampled version. The "reduced" benchmark only contains a few representative tasks in the original ones, such that the performance on the "reduced" benchmark can serve as an approximation to the performance on the original ones.