Can you elaborate more on the differences in running ollama or lmstudio? Do they actually slow down the speed of the inference and if so why? Or is it just a preference thing?