Original Prompts
收藏arXiv2025-09-30 收录
下载链接：
https://arxiv.org/abs/2412.08127
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了从WikiText-103语料库中抽取的5,000个序列，这些序列包含了为语言模型设计的原始提示。经过基于BLEU分数的过滤过程后，数据集最终由2,473组原始提示、自动提示及其后续内容组成。该数据集的规模为5,000个序列，其任务是训练语言模型的自动提示功能。
This dataset includes 5,000 sequences extracted from the WikiText-103 corpus, with each sequence containing original prompts tailored for language models. Following a filtering procedure based on BLEU scores, the finalized dataset comprises 2,473 sets of original prompts, auto-generated prompts and their associated subsequent content. With a total of 5,000 sequences, this dataset is developed for training the automatic prompt generation capability of language models.
提供机构：
Publicly available via supplementary materials upon publication
