Loss & cross-entropy

The training signal: softmax cross-entropy between the model’s next-token logits and the target token ids.

PyTorch → JS mapping

Book (PyTorch)	Here (tfjs)
F.cross_entropy	tf.losses.softmaxCrossEntropy

Coming soon

This lesson is still being written. It will cover turning the model’s next-token logits and the target ids into a softmax cross-entropy training signal.