Use llama.cpp or llamafile , ollama just adds weight and does too much magic.