A qusetion about prob2 on bce loss

[BigDunk]

Hello, I have a question that why here it uses the prob2 instead of prob during train function. Look forward to your reply,thanks! pos_prob = prob2[pos_pairs, :] pos_sim = torch.bmm(prob.view(args.batch_size, 1, -1), pos_prob.view(args.batch_size, -1, 1)).squeeze()