Skip to content

neuralSPOT Examples

These examples focus on model evaluation and profiling with ns_autodeploy.

Example 1: First Model Bring-Up

Goal: run a .tflite model on hardware quickly and confirm it executes end to end.

ns_autodeploy --tflite-filename=mymodel.tflite --model-name mymodel

Use this first when you want a fast answer to:

  • does the model run on the target?
  • is the memory footprint reasonable?
  • do I need to reduce model complexity before integration?

Example 2: Profile and Compare Models

Goal: compare two or more candidate models on the same Ambiq target.

Typical comparison points:

  • total inference latency
  • layer-level hotspots
  • tensor arena size
  • operator mix

This is often the fastest way to choose between model variants before application integration.

Example 3: Hand-Off to Zephyr

Once profiling stabilizes, move to Zephyr setup when you need:

  • application-level control
  • board services and RTOS integration
  • a product-style build and packaging flow

For more on the tool itself, see the neuralSPOT tools documentation.