Until I can buy an 80GB VRAM GPU, I won't attempt to do it. A local LLM is always missing something that needs a bigger model.