Running a 28.9M parameter LLM on an $8 microcontroller

A project demonstrates running a 28.9M parameter language model on an ESP32 microcontroller costing $8. It uses quantization and efficient inference to fit within limited memory.