Mini-project
Electronics, Semiconductors & Telecom
Small model at the edge
Fine-tune and quantise a small model to run on a single-board computer, and measure speed and accuracy.
The brief
Many products cannot send data to the cloud. Fine-tune a small model for one task (such as classifying device alerts or answering product questions), quantise it, run it on a Raspberry Pi or similar board, and measure speed, memory and accuracy before and after. Good looks like: a clear task and test set, measured trade-offs, and a working demo on the board.
What you will build
- Choose a task and build a test set
- Fine-tune a small model
- Quantise it with llama.cpp or ONNX
- Run it on the board and measure speed and memory
- Compare accuracy before and after quantisation
Free, open data to build it with
- LiteRT for microcontrollers ↗ Google AI Edge
Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.