Version that runs on the CPU: https://github.com/krychu/llama

I get 1 word per ~1.5 secs on a Mac Book Pro M1.