Log determinant has wrong dimensions?

[matejgrcic]

Issue
I've printed log_p and logdet returned by the model. It seems that log_p_sum has the size equal to the batch size (which is fine) but the logdet contains a single value. Should logdet have the same dimensions as log_p?

Code for reproducing
Add the following code to train.py line 120.
print(log_p.size(), logdet.size())

[rosinality]

If you use additive coupling and not affine coupling, logdet will not have batch dimensions.