Sampling from wrapped normal distribution

[Sehmimul]

Hi, I understand from the paper that we are using a wrapped normal distribution distribution instead of a normal distribution (as used in diffusion models in general). However, in the sampling code given below it seems like you are using samples of normal distribution instead of using samples of wrapped normal distribution. I was hoping if you could please tell me why this is the case. Thanks!

torsional-diffusion/diffusion/sampling.py

Line 167 in 5f713b4

z = torch.normal(mean=0, std=1, size=data_gpu.edge_pred.shape)