https://github.com/oobabooga/text-generation-webui/ has been running 4bit LLaMA on Radeon hardware with faster speeds and lower memory requirements than this LLaMA.cpp for the past week!