This page describes how to compile an existing LiteRT model for the Axon NPU in Nordic Edge AI Lab.
Use this option if you already have a trained .tflite model and want to estimate how it will perform on the Axon NPU, rather than training a new model in the lab.
This is the fastest way to get a footprint estimate — no local installation required.
Note
Only fully integer-quantized (int8) LiteRT (.tflite) models are accepted.
Other quantization schemes are not supported.