Open Long-Tailed QA Benchmark

该数据集是一个经过精心筛选的基准，它由43个现存的代表性问答数据集组成，涵盖了四种问答格式：抽取式问答、抽象式问答、多项选择题问答以及是非题问答。此外，该数据集包含了21个用于训练的已知任务和22个用于测试的未见任务。训练集根据Zipf分布进行了降采样处理。这是一个大规模的数据集，包含了43个问答任务，其任务是进行问题回答。

This dataset is a carefully curated benchmark composed of 43 existing representative question answering (QA) datasets, covering four QA formats: extractive QA, abstractive QA, multiple-choice QA, and true-false QA. Additionally, this benchmark includes 21 known training tasks and 22 unseen test tasks. The training sets are downsampled in accordance with the Zipf distribution. This is a large-scale dataset that comprises 43 QA tasks centered on question answering.