ZhuiyiTechnology / roformer-sim

SimBERT升级版(SimBERTv2)!
Apache License 2.0
439 stars 73 forks source link

为什么supervised阶段微调五分类? #26

Open Yi-Shi94 opened 2 years ago

Yi-Shi94 commented 2 years ago

按照博客里说的“具体来说,Sentence-BERT是将u,v,|u−v|(其中|u−v|是指u−v的每个元素都取绝对值后构成的向量)拼接起来做为特征,后面接一个全连接层做2分类(如果是NLI数据集则是3分类)。” 为什么在train/supervised.py里output = keras.layers.Dense(5, use_bias=False)(output)

beybin commented 2 years ago

https://kexue.fm/archives/8541/comment-page-1#comments