Comment by phatfish

Or know when to shut up.

A tangent, but can anyone ELI5 how models "know" when to stop generating tokens? Or what the method to stop them at the right point is?