Recipes

This section lists model architectures validated end-to-end with LMCache, with a recipe page per architecture covering only the LMCache-specific configuration that diverges from defaults. Engine-side documentation (how to serve the model itself) lives with the serving engine; recipe pages link out rather than duplicate.

For the generic LMCache + engine wiring (ports, remote hosts, sending a first request), see Quickstart – recipes assume that page as a prerequisite.

Recipes are grouped by attention architecture:

• Uniform Attention Models
• Hybrid Attention Models