Post by @github.awesome

ESP32 AI squeezes a 28.9-million-parameter language model onto an eight-dollar ESP32-S3 with no server connection. The trick is keeping a 25-million-parameter embedding table in flash and fetching only the rows needed for each token, while the compute-heavy core stays in faster memory. The maintainer measures roughly 9.5 tokens per second. It’s an architecture experiment, not a chatbot: the TinyStories model writes simple stories but can’t answer questions or code. #github #opensource