reference impl with llama.cpp compiled to distributed inference across machines, with real end to end demo - View it on GitHub