About the implementation of KLR

[LetheSec]

Hi, thanks for your nice work.

I have some questions about the implementation of the KLR loss.

muse_bench/baselines/baselines/iterative.py

Line 152 in 6d4fdcb

kl_r = F.kl_div(

If KLR refers to KL divergence as in previous work, when log_target=True, it seems that F.kl_div should take log probabilities as input instead of using logits directly $^{[1]}$, what impact will this have on the results?

[1] https://pytorch.org/docs/stable/generated/torch.nn.functional.kl_div.html