According to line110-111 in mtl_trainer.py:
if isinstance(self.model, ESMM):
loss = sum(loss_list[1:]) #ESSM only compute loss for ctr and ctcvr task
we need to set three classification tasks in task_types to make sure totalloss = loss{ctr} + loss_{ctcvr}, otherwise, totalloss = loss{ctr}.
yinpu
commented
7 months ago
According to line110-111 in mtl_trainer.py: if isinstance(self.model, ESMM): loss = sum(loss_list[1:]) #ESSM only compute loss for ctr and ctcvr task we need to set three classification tasks in task_types to make sure totalloss = loss{ctr} + loss_{ctcvr}, otherwise, totalloss = loss{ctr}.