A CONCEPT · PROJECTED FROM THE VINDEX KNOWLEDGE GRAPH
LAYER
attention then feed-forward, repeated
One pass of the machine — in a conventional decoder layer: look backwards along the sentence, then transform what was found, repeated dozens of times per token.
A layer does exactly two things. Attention reads every earlier token; the feed-forward network transforms what attention gathered. Each result is added to a running stream rather than replacing it, which is why a layer's contribution can be measured, attributed, or skipped.
IN THE GRAPH
layer— contains →attention
layer— contains →feed-forward
layer— disciplined by →residual
layer— kept in range by →norm