On a lark a friend setup Ollama on a 8GB Raspberry Pi with one of the smaller models. It worked by it was very slow. IIRC it did 1 token/second.