HUFS-DILAB/MT-wmt14-500k-opus-mt-en-de
收藏官方服务：
资源简介：
该数据集基于WMT14英语-德语数据集的训练分割（包含50万句子），源文本为英语句子，参考文本为原始德语参考译文。数据集包含5个翻译候选（h1至h5），每个候选由Helsinki-NLP/opus-mt-en-de模型生成，并带有对数概率分数，按分数降序排序。翻译方向为英语到德语，使用波束搜索方法，参数设置为波束数5和返回序列数5。
This dataset is based on the WMT14 English-German dataset (train split, 500k sentences), with source text in English and reference text as the original German reference. It includes 5 translation candidates (h1 to h5), each generated by the Helsinki-NLP/opus-mt-en-de model and accompanied by log-probability scores, sorted in descending order. The translation direction is English to German, using beam search with parameters num_beams=5 and num_return_sequences=5.
提供机构：
HUFS-DILAB
