What you will learn

By the end of this module, you will be able to:

• identify the roles of the CPU and the Ethos-U NPU in a heterogeneous Cortex-M + Ethos-U system

• explain why a model’s data format (for example, FP32, or INT8) is important for Ethos-U acceleration

• calibrate and quantize MobileNetV2 before lowering it to Ethos-U

• interpret ExecuTorch delegation summaries and Vela compiler reports

• build and run an ExecuTorch runner on Corstone-320 FVP, and understand why you might use an FVP

• convert raw model output into an ImageNet classification result