llama.cpp

879ff512 - server: add --models-memory-max parameter to allow dynamically unloading models when they exceed a memory size threshold