Hi! really appreciate your nice work and sharing the code. 馃槃
Here is a question that just confuses me a little,
dist = one_hot_labels[:, 1:].float() * log_probs[:, 1:]
example_loss_except_other, _ = dist.min(dim=-1)
per_example_loss = - example_loss_except_other.mean()
why does the min of dist equal the loss without other class here?
thanks for your great work again, looking forward to your reply 馃槅
Hi! really appreciate your nice work and sharing the code. 馃槃
Here is a question that just confuses me a little,
why does the min of dist equal the loss without other class here?
thanks for your great work again, looking forward to your reply 馃槅