Deterministic problem when infer same input

distributed
1

When i infer the same input, the output is not deterministic sometimes, and the code is as below.
Debug found that the posterior is same, but the sample is different for same input sometimes.

                    posterior = F.softmax(logits, dim=1)                  
                    distrib = torch.distributions.Categorical(posterior)
                    sample = distrib.sample().float()

I execute model.eval() and have set the seed at the begining. Do you have any suggestions? Thank you.

seed = 1234
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
2

Can you ask this question with Uncategorized label. It seems there is some randomness somewhere, but torch.distributed (not torch.distributions) is not involved here.