Run MNIST on an Alif E8 Ensemble DevKit using ExecuTorch and Ethos-U85
Introduction
Learn about MNIST and the Alif Ensemble E8 DevKit
Set up the Alif Ensemble E8 DevKit
(Optional) Set up a Docker development environment
(Optional) Export PyTorch model to ExecuTorch format
Prepare the ExecuTorch model and static libraries for the Alif E8 CMSIS project
Create the Alif E8 CMSIS project
Process and copy a sample image into the Alif E8 CMSIS project
Flash and run the project on the Alif Ensemble E8 DevKit
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

Before you begin

The firmware project needs two artifacts:

  • mnist_ethos_u85.pte: the ExecuTorch model compiled for Ethos-U85
  • et_bundle.tar.gz: ExecuTorch headers and static libraries for the bare-metal Cortex-M build

If you completed the optional sections to train and export the model yourself, these files are already in ~/mnist_alif/executorch-alif/output/.

If you skipped the optional sections and are using the provided .pte file, create the output directory and download the provided artifacts:

    

        
        

mkdir -p ~/mnist_alif/executorch-alif/output
cd ~/mnist_alif/executorch-alif/output
curl -L -o et_bundle.tar.gz https://raw.githubusercontent.com/arm-education/alif-ethos-u85-npu-mnist/main/et_bundle.tar.gz
curl -L -o mnist_ethos_u85.pte https://raw.githubusercontent.com/arm-education/alif-ethos-u85-npu-mnist/main/mnist_ethos_u85.pte
  

    

    

        
        

New-Item -ItemType Directory -Force -Path "$HOME\mnist_alif\executorch-alif\output"
cd "$HOME\mnist_alif\executorch-alif\output"
curl.exe -L -o et_bundle.tar.gz https://raw.githubusercontent.com/arm-education/alif-ethos-u85-npu-mnist/main/et_bundle.tar.gz
curl.exe -L -o mnist_ethos_u85.pte https://raw.githubusercontent.com/arm-education/alif-ethos-u85-npu-mnist/main/mnist_ethos_u85.pte

Verify all artifacts are present:

    

        
        

ls -lh ~/mnist_alif/executorch-alif/output/mnist_ethos_u85.pte
ls -lh ~/mnist_alif/executorch-alif/output/et_bundle.tar.gz
  

    

    

        
        

Get-Item ~\mnist_alif\executorch-alif\output\mnist_ethos_u85.pte
Get-Item ~\mnist_alif\executorch-alif\output\et_bundle.tar.gz

Convert the model to a C header

The firmware embeds the .pte model as a byte array in flash memory. Use xxd to generate a C header:

    

        
        

  cd ~/mnist_alif/executorch-alif/output
  xxd -i mnist_ethos_u85.pte > mnist_model_data.h
  

    

    

        
        

  cd ~\mnist_alif\executorch-alif\output
  & "$env:ProgramFiles\Git\usr\bin\xxd.exe" -i mnist_ethos_u85.pte | Set-Content -Encoding ascii mnist_model_data.h

Open the generated header (mnist_model_data.h) and change the first array declaration to the following:

    

        
        
#include <stdint.h>
const uint8_t __attribute__((aligned(16))) mnist_ethos_u85_pte[] = {
Important

The aligned(16) attribute is required because the Ethos-U85 needs the Vela command stream data aligned to 16 bytes. Without aligned(16), the NPU driver will report an alignment error at runtime.

Extract the ExecuTorch bundle

Extract the ExecuTorch headers (from et_bundle.tar.gz) into the VS Code template project:

    

        
        

cd ~/mnist_alif/alif_vscode-template
mkdir -p third_party/executorch
tar -C third_party/executorch -xzf ~/mnist_alif/executorch-alif/output/et_bundle.tar.gz
  

    

    

        
        

cd ~\mnist_alif\alif_vscode-template
New-Item -ItemType Directory -Force -Path .\third_party\executorch
tar -C .\third_party\executorch -xzf "$HOME\mnist_alif\executorch-alif\output\et_bundle.tar.gz"

Verify the headers are in place:

    

        
        
ls third_party/executorch/et_bundle/include/executorch/

You should see runtime/ and other directories.

You’re now ready to integrate the model into the VS Code project.

What you’ve accomplished and what’s next

You’ve now prepared the embedded ExecuTorch model (mnist_model_data.h), ExecuTorch headers (third_party/executorch/et_bundle/include), and the static libraries used by the firmware build (third_party/executorch/et_bundle/lib).

Next, you’ll create the MNIST firmware project by duplicating the Blinky example and replacing relevant files to fit the application.

Back
Next