Skip to content
The loss curve

Batch

A group of training examples processed together in one forward and backward pass. Bigger batches give a less noisy gradient and use hardware better, at the cost of memory.

Companion explanation in Step by Token, chapter 6.

Where this term gets built