Model Atlas
A field guide to modern AI models and modules
Browse
- Browse the atlas
Start with a page that explains where models, modules, concepts, glossary entries, tags, and architecture pages fit together.
- Grouped-query attention
An attention variant that shrinks KV cache costs by letting several query heads share fewer key-value heads.
- SwiGLU
A gated feed-forward variant that replaces a plain activation with a learned gate and SiLU path.
- ReLU
A simple activation function page that gives a lightweight entry point into module basics.