Comment by The_Amp_Walrus

I might be wrong about this but doesn't ollama do some work to ensure the model runs efficiently given your hardware? Like choosing between how much gpu memory to consume so you don't oom. Does llama.cpp do that for you with zero config?