Efficient Edge AI on Arm CPUs and NPUs: Understanding ExecuTorch through Practical Labs – PyTorch

ExecuTorch extends PyTorch to enable local AI inference on constrained edge devices. Arm has released a set of Jupyter labs covering the full deployment pipeline — from exporting PyTorch models to .pte artifacts, through CPU inference on Raspberry Pi using the XNNPACK backend, to hardware-accelerated NPU inference on Arm Ethos-U devices via TOSA quantization and the EthosUPartitioner. The labs also demonstrate how to use Google's Model Explorer with Arm-developed adapters to visualize graph partitioning and understand performance implications of unsupported operators causing CPU/NPU fragmentation.