don't use ollama. llama.cpp is better because ollama has an outdated llama.cpp