cross_entropy_loss

opennmt.utils.cross_entropy_loss(logits, labels, label_smoothing=0.0, training=None, weight=None)

Computes the cross entropy loss.

Parameters

• logits – The unscaled probabilities with shape \([B, V]\).

• labels – The true labels with shape \([B]\).

• label_smoothing – The label smoothing value.

• training – Compute training loss.

• weight – The weight of each example with shape \([B]\).

Returns

The cumulated loss and the loss normalizer.