The next age of LLMs? Dev gets a small LLM running at 10 tokens a second locally on a $10 microcontroller

A 28.9M-parameter model runs on a microcontroller costing less than $10 at 9.88 tokens a second because 25M of those parameters never leave flash storage