Check out Ollama, it's built to run models locally. Llama3 8b runs great locally for me, 70b is very slow. Plenty of options.