Skip to content

neuralSPOT Setup

Use this path when you want the fastest way to evaluate a model with heliaRT on Ambiq hardware.

The main entry point is ns_autodeploy, which can build, flash, and profile a .tflite model with a much shorter setup cycle than a hand-built application.

Best For

  • first-pass model evaluation
  • profiling a model on hardware
  • checking whether a model is a good fit before deeper integration work
  • quickly comparing model variants on supported Ambiq targets

What You Need

  • an Ambiq target board
  • a .tflite model
  • a working neuralSPOT environment with ns_autodeploy

Quick Start

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

This flow can:

  • compile an application using heliaRT
  • flash the target
  • run the model on hardware
  • collect profiling information for layer-level analysis

When To Use This Path

Choose ns_autodeploy when your immediate question is:

  • does this model run on my Ambiq target?
  • how large is the tensor arena?
  • which layers dominate runtime?
  • how do two model variants compare on hardware?

Choose Zephyr setup instead when you are integrating heliaRT into a Zephyr product or application workspace.

Typical Workflow

  1. Start with a .tflite model.
  2. Run ns_autodeploy to build and flash a profiling run.
  3. Inspect runtime and layer-level profiling output.
  4. Iterate on model choice, quantization, or operator mix.
  5. Move to Zephyr or another source-based integration once the model is validated.

Expected Outcomes

From this path, users typically want to learn:

  • whether the model fits and runs
  • how fast the model executes on the target
  • which operators are expensive
  • whether the model is ready for deeper application integration

Notes

  • This is the quickest way to get useful hardware data from heliaRT.
  • It is a deployment and profiling path, not the only integration path.
  • Once the model is validated, many teams move to Zephyr setup for application integration.

Learn more about ns_autodeploy in the Ambiq neuralSPOT documentation: