Profile a model on a neural accelerator with MLIA for VS Code
Introduction
Prepare your model and VS Code workspace
Prepare the Android workload
Capture NX profiling data
Analyze measured performance in VS Code
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

Prepare Scenario Runner and connect the device

Continue in the Bash terminal from earlier.

You’ll use Scenario Runner to run workload/model.vgf on your neural accelerator (NX) device. The Neural Statistics layer records the run for Arm ML Inference Advisor (MLIA) to analyze.

For the Scenario Runner example, prepare a debuggable Android Arm64 Scenario Runner v0.11.0 APK. To build it, follow the MLIA Scenario Runner build instructions in a separate terminal and virtual environment. Add -b v2026.09.0 to repo init, use NDK r28c (28.2.13676358), and add --package-version 0.11.0 to the APK build command. This SDK manifest selects the runner and VGF Library versions compatible with the exported VGF 0.4.3 model.

Enable Developer options > USB debugging on your NX device using the Android developer options and USB debugging instructions .

Connect only this device and check adb:

    

        
        
adb devices -l

The device appears with status device after you accept its USB-debugging authorization prompt. For information about driver compatibility, see the MLIA device requirements .

Add the scenario file

Save the supplied scenario file as scenario.json in $NX_WORKSPACE/workload, alongside model.vgf.

This scenario runs the SmallUpscalerModel that you exported earlier with zero-filled inputs. It doesn’t write output tensors.

To create a scenario for another model, see the VGF Dump Tool documentation and the Scenario Runner JSON specification .

Install the debug APK and capture layer

Set APK to the absolute path of your debug APK and install it:

    

        
        
export APK="/path/to/ai-ml-sdk-scenario-runner-0.11.0-debug.apk"
adb install -r "$APK"

For installation problems, see the MLIA capture troubleshooting guide .

Download and extract the Android Arm64 capture layer:

    

        
        
curl -fL https://github.com/arm/vulkan-layer-neural-statistics/releases/download/v1.0.0/VkLayerNeuralStatistics-android-arm64-v8a.tar.gz \
    -o "$NX_WORKSPACE/neural-statistics-layer.tar.gz"
tar -xzf "$NX_WORKSPACE/neural-statistics-layer.tar.gz" -C "$NX_WORKSPACE"
export LAYER_LIB="$NX_WORKSPACE/VkLayerNeuralStatistics-android-arm64-v8a/libVkLayerNeuralStatistics.so"

What you’ve accomplished and what’s next

You’ve prepared the scenario file, installed Scenario Runner, and downloaded the capture layer.

Next, you’ll capture one run on the device.

Back
Next