LAMBADA
收藏arXiv2025-09-30 收录
下载链接：
https://zenodo.org/record/2630551#.yfjvawt7s_w
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为LAMBADA，旨在测试模型基于长上下文预测句子最后一个单词的能力。为了正确回答问题，LAMBADA要求对上下文有深入的理解。该任务的名称为“长期上下文回忆”。
This dataset, named LAMBADA, is designed to evaluate models' ability to predict the final word of a sentence based on long context. To correctly complete this task, models need to have an in-depth understanding of the provided context. The name of this task is "Long-Context Recall".
提供机构：
University of Edinburgh
