search

GanjinZero / RRHF

[NIPS2023] RRHF & Wombat
780 stars 49 forks source link

损失函数 #42

Closed xiayouhong closed 10 months ago

xiayouhong commented 10 months ago

您好，损失函数中的 t 是啥意思呀

GanjinZero commented 10 months ago

第t个token

xiayouhong commented 10 months ago

image 您好，在logit_label = self.gather_logits_labels(logits, inputs.get("labels"))这一步过程中inputs.get("labels“）会被修改，会影响下一步score的计算

GanjinZero commented 10 months ago

https://github.com/GanjinZero/RRHF/issues/37