Epoch
One full pass over the training set. Large-scale language models rarely complete even one; they sample batches from a stream and count steps instead.
Companion explanation in Step by Token, chapter 6.
Where this term gets built
Continue
One full pass over the training set. Large-scale language models rarely complete even one; they sample batches from a stream and count steps instead.
Companion explanation in Step by Token, chapter 6.
Continue