Question about clip_nll loss.

[YuGuiwe]

I am confused about the calculation clip_nll loss. If my predict : clipwise_output = [-1.2 , -2.3, -0.5] target = [0, 0, 1] torch.mean(clipwise_output * target) = torch.mean([0, 0, -0.5]) = -0.166

Cross entropy was averaged by the number of category ? Thanks for your reading...Hope it's not stupid question, I am beginner in training model.