Running a 28.9M parameter LLM on an $8 microcontroller

Other: microcontroller-based AI(github.com)view on HackerNews

microcontrollerAIlanguage modelESP32-S3Per-Layer EmbeddingsGoogle's Gemma models

Author: boveyking

Date: 7/25/2026

Article Summary:

A 28.9M parameter language model is run on an ESP32-S3 microcontroller, with most of the model's parameters stored in flash memory and only the necessary parameters loaded into fast memory for computation.