Edge inference

Running models directly on edge devices. Reduces latency and cloud dependency.