copenlu/cmt-benchmark-nq

NQ数据集是一个流行的数据集版本，基于原始的NQ数据集（由Kwiatkowski等人于2019年提出）。这个版本的数据集包含了能够从原始维基百科页面中恢复出黄金段落并且有一个简短答案（不超过5个单词长度）的NQ样本。数据集的上下文包括原始NQ标注者标注的正确黄金上下文（gold）、与黄金上下文不匹配的无关上下文（irrelevant）以及被编辑以促进非黄金答案的黄金上下文（edited）。数据集分为两个版本：gpt2-xl和pythia-6.9b，每个版本都有相应的验证集和测试集。数据集包含了一些固定列，例如样例ID、上下文类型、模板、问题、上下文等，以及一些依赖于数据集版本的列，例如模型预测和概率。

The NQ dataset is a version of the popular NQ dataset originally proposed by Kwiatkowski et al. (2019). This version of the dataset includes NQ samples that can recover the gold passage from the original Wikipedia page and have a short answer (less than 5 words in length). The contexts in the dataset include the correct gold context annotated by the original NQ annotators (gold), irrelevant contexts that do not match the gold context as annotated by the original NQ annotators (irrelevant), and gold contexts that have been edited to promote another answer than the gold answer (edited). The dataset comes in two versions: gpt2-xl and pythia-6.9b, each with corresponding validation and test sets. The dataset contains fixed columns such as example_id, context_type, template, question, context, etc., and model prediction and probability columns that depend on the dataset version.