true, switched from ollama to llama.cpp these days and it's good. wonder if this is also the best option for edge ai deployment(currently use it on desktop)