ellamind/ruler-utils

该数据集名为ruler-utils，是用于自定义RULER任务的支持资产。它包含多个配置：essays（串联文章语料库，用于NIAH任务的文章干草堆，替代上游依赖）、squad_docs和squad_qas（为ruler_qa_squad任务预处理的SQuAD-v2上下文和问题）、hotpot_docs和hotpot_qas（为ruler_qa_hotpot任务预处理的HotpotQA干扰上下文和问题）。所有数据文件均为训练分割，以parquet格式存储。The dataset named ruler-utils is a supporting asset for customizing RULER tasks. It includes multiple configurations: essays (a corpus of concatenated articles, serving as the article haystack for the NIAH task and replacing upstream dependencies), squad_docs and squad_qas (preprocessed SQuAD-v2 contexts and questions for the ruler_qa_squad task), hotpot_docs and hotpot_qas (preprocessed distractor contexts and questions of HotpotQA for the ruler_qa_hotpot task). All data files are in the training split and stored in Parquet format.