Intermediate Lesson 10 of 13

forward pass

One trip of an input through a trained model to get its answer.

A forward pass is what happens when data goes through a neural network exactly once, from input to output, with no back-and-forth or revision. It is the basic unit of speed for this kind of model.

Think of it as

Like passing a form through a scanner once and getting the summarised result printed straight out the other side.

Example

A model that answers in one forward pass makes one trip through the model per question. A chatbot writing a 50-word answer needs roughly one forward pass per token, so dozens of trips for the same question.