The Pocket Tank project controls a virtual aquarium with a local language model on ESP32-S3
The Pocket Tank project uses a language model with 14.3 million parameters on an ESP32-S3 microcontroller to control fish. After quantization, the model takes up 7.56 MB and runs from flash memory without the internet. The project includes training tools and a PC simulator.
The open source project Pocket Tank uses a small language model to control a virtual aquarium on an ESP32-S3 microcontroller. The model has 14.3 million parameters and takes up 7.56 MB after quantization. It runs entirely locally, directly from flash memory, without the internet or the cloud.
Each fish has its own needs and personality. Based on the current situation, the model decides whether it will look for food, rest, explore its surroundings or follow another fish. The project includes firmware, a trained model, training tools, a PC simulator and the option to run in QEMU.
Why it matters
Developers get a concrete example of using a small language model to control behavior on a microcontroller. The PC simulator and the option to run in QEMU make it possible to explore the project beyond the target hardware; tools for custom training are also included.
Relevant practical impact
What this means
For individuals
Anyone interested in local AI on microcontrollers can try controlling the fish in the PC simulator for the Pocket Tank project without an ESP32-S3 device.
Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.