Inside a transformerFor people who have never opened one — every moving part, draggable, in four short sections

Half of this library talks about transformers, and it uses vocabulary as if everyone grew up with it: token, attention head, residual stream, LayerNorm. Nobody grew up with it. This module builds the whole machine from nothing — one section per moving part, every part something you can drag, press, and in one case hear — because each engineering name turns out to be a simple physical picture. The only math assumed is the rotations primer's: arrows, and the dot product as agreement.

By the end, the sentence "a transformer is a network of coupled oscillators" stops being a metaphor you take on faith and becomes a description you can check — which is exactly where the Transformers as Oscillators essay begins.