Running a 28.9M parameter LLM on an $8 microcontroller

# Summary A developer successfully ran a 28.9 million parameter language model on an ESP32-S3 microcontroller costing about $8, generating text at roughly 9 tokens per second entirely on-device. This represents a 100x increase in model size compared to previous microcontroller implementations, achieved by storing most parameters (25M) in the chip's slow flash memory and retrieving only small portions as needed, inspired by Google's Per-Layer Embeddings technique from Gemma models.