Inconsistency between bernoulli and bernoulli_ #1259
Description
import torch
a = torch.FloatTensor(3, 5).uniform_(0, 1)
a.round_()
print a.bernoulli()
print a.bernoulli_()
Tensor a is rounded to 0 or 1 so the bernoulli sampling should be deterministic. However, two printed bernoulli are different.