prefix caching in llama.cpp: worth it?

[rocmriley]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Llama and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.