How do you improve advanced workflows when working with llama.cpp?

[freyatech14]

I have a question about llama.cpp that may be useful for other people working with it too.

I am particularly interested in advanced workflows and creative uses. I would like to compare approaches that people have actually used rather than just theoretical suggestions.

Which approach would you recommend and why?