nlpxucan / ZRKGC

38 stars 10 forks source link

模型训练时next_sentence loss 是什么呢? #4

Open Cherryjingyao opened 3 years ago

Cherryjingyao commented 3 years ago

是预测的response 的概率分布和实际的交叉熵吗?那和goden_out loss有什么区别吗?

nlpxucan commented 3 years ago

next_sentence loss是论文里面的knowledge selection loss