Dev proves LLMs will run on anything – even a $10 microcontroller

Nearly 10 tok/s and it's mostly coherent — What's not to like?