Question of supervised train loss

[Chris-WangQ]

Hello，
In the "model.py" , I finded the loss function of supervised training is CrossEntropyLoss. With hard negatives, why not select TripletMarginLoss?
Looking forward to your reply, thanks!

[kongds]

hello,

CrossEntropyLoss is implemented as contrastive learning (including hard negative) in supervised training following SimCSE.
The overall training objective is:

And I think contrastive learning is more efficient than TripletMarginLoss as shown by SimCSE.