esp32-ai

What people use it for

Fits a 28.9-million-parameter language model onto an ESP32-S3, generating every token locally without a network connection. It uses flash for the embedding table and faster memory for the rows needed per token.

Github Awesome