Adapting ML models to run inference on devices: phones, smart cameras, edge accelerators. Computer vision models first, now LLMs.