zjunlp / DocuNet

[IJCAI 2021] Document-level Relation Extraction as Semantic Segmentation
MIT License
128 stars 21 forks source link

结果不一致 #25

Closed ARDUJS closed 1 year ago

ARDUJS commented 1 year ago

hello,为什么我跑cdr数据集,出来的f1值是85左右,然后论文给出的是76.3

zxlzr commented 1 year ago

恭喜您取得了更好的结果。目前给的超参数是经过优化的,在不同GPU、不同batchsize大小和更长训练步数下会有更好的结果出现。您可以参见https://github.com/zjunlp/DocuNet/issues/18

ARDUJS commented 1 year ago

thank you

MrDreamQ commented 1 year ago

hello,为什么我跑cdr数据集,出来的f1值是85左右,然后论文给出的是76.3

你好,你是调整什么参数才跑到这个结果的?