I meant ollama uses llama.cpp internally. Sorry for the confusion.