Primitives
the irreducible operations
modifiers (swap-ins, not families)
Modules
stacks of blocks that play a role
Families
modules + objective = a recipe
an explorable map of how AI models are put together
There are far fewer parts than architectures. A family is just which modules + which objective. Click anything.
the irreducible operations
stacks of blocks that play a role
modules + objective = a recipe
how to read this
Click a family to see its recipe decompose into shared parts. Click a primitive to watch every family built on it light up — the whole zoo collapses into a small grammar.
Badges mark what kind of thing each label is: an architecture (a wiring), a generative-framework (a training recipe), a model-class (a deployable category like “LLM”), a backbone (objective-flexible), or a modifier (a swap-in like MoE).