Serving Your Own Model with llama.cpp: GPU Offload, Caching, and the Trap of a Stuck Wrong Answer

Serving a GGUF model with llama.cpp, caching answers in SQL, and a demo of why a cached wrong answer doesn't just go away on its own.