Cannot train the model with multiple GPU

[chenweize1998]

Hi. I follow the steps listed in README, but when calculating the loss at training time, the code throws an error at

hyfi/hyfi/runner.py

Line 108 in 587f140

loss = self.loss.calculate_loss(logits, one_hot_true_types)

When there are 4 GPUs, the dimension of logits is 225, 8822 while one_hot_true_types is 900, 8822. It seems that the true-types tensor is not distributed to different GPUs. Could you fix it?