Llama.cpp added the ability load/switch models on demand with the max-models and models preset flags.