The auxilary loss is dropped in the HunYuanMoE.forward() with no passing it to the final loss. It is not used for training.
HunYuanMoE.forward()
· Sign up or log in to comment