RMSNorm
a lighter-weight LayerNorm variant
Attention sublayer
reuse the module from the attention star
MLP sublayer
expand, activate, project down
Full block
pre-LN residual wiring around both
Full model
embeddings + N blocks + output head
Start with the two primitives every modern block needs beyond attention: RMSNorm (the normalization most current models use instead of LayerNorm, since it drops the mean-centering step and is cheaper) and the MLP.