MLIA Ethos-U Plugin

This package contains the MLIA target plugin for Arm Ethos-U platforms and packages the Ethos-U backend integrations used by MLIA.

The package is distributed as mlia-ethos-u and contributes:

• The ethos_u target plugin.
• The vela backend plugin.
• The Corstone backend plugin, which registers target-specific Corstone backends.
• Bundled Ethos-U target profiles and Vela configuration assets.

Table of Contents

• Overview
• Supported targets
• Backends in this package
• Installation
• Reporting bugs
• Development setup
• Common commands
• Project layout
• Documentation
• Releases
• License
• Trademarks and copyrights

Overview

mlia-ethos-u is the main MLIA plugin package for Ethos-U inference analysis. It extends the core MLIA framework with Arm Ethos-U target knowledge, operator analysis, bundled target profiles, and the backend integrations required for Vela compilation and Corstone-based performance flows.

This is the package to install when you want MLIA to analyse LiteRT / TensorFlow Lite .tflite models for Ethos-U55, Ethos-U65, or Ethos-U85 targets, or when you want to run the Corstone-based ExecuTorch AOT path for supported PyTorch inputs.

Supported targets

Bundled target profiles include:

• ethos-u55-128
• ethos-u55-256
• ethos-u65-256
• ethos-u65-512
• ethos-u85-128
• ethos-u85-256
• ethos-u85-512
• ethos-u85-1024
• ethos-u85-2048

These profiles are shipped under src/mlia/resources/target_profiles/.

Backends in this package

Vela

The Vela backend is used for compiler-oriented analysis, compatibility checks, and Ethos-U standardized performance results based on the Vela toolchain.

Corstone

The Corstone backend supports simulation-oriented performance flows for Corstone platforms used in Ethos-U analysis and validation. It is also the backend path used for supported ExecuTorch .pte workloads.

Installation

Install into an environment that already contains mlia:

pip install mlia-ethos-u

If you want MLIA to accept PyTorch .pt2 inputs directly for the ExecuTorch flow, install the matching converter plugin as well:

pip install mlia-converters-pytorch

A typical MLIA workflow then references one of the bundled profiles, for example:

mlia check my_model.tflite --target-profile ethos-u55-256

For supported ExecuTorch AOT flows, a PyTorch-originating run can look like:

mlia check my_model.pt2 --target-profile ethos-u55-256 --performance --backend corstone-300

Direct .pt2 support depends on mlia-converters-pytorch. Corstone performance for ExecuTorch .pte models, including converted .pt2 inputs, currently supports only these combinations:

Target	ExecuTorch performance backend
Ethos-U55	corstone-300
Ethos-U85	corstone-320
Ethos-U65	Unsupported

For Ethos-U65, --performance on a PyTorch .pt2 input is supported only through Vela:

mlia check my_model.pt2 --target-profile ethos-u65-256 --performance --backend vela

Vela provides compiler-based performance estimates. It does not run ExecuTorch .pte models; use the original .pt2 input for this route. Ethos-U65 .pte performance is currently unsupported. These restrictions do not affect the TFLite Corstone performance path. See the backend guide for details.

The package depends on MLIA and is intended to be used as part of an MLIA installation rather than as a standalone CLI. See pyproject.toml for the current dependency constraint.

Reporting bugs

Report bugs by creating GitHub issues. Use the arm/mlia issue tracker by default.

Only open an issue in arm/mlia-ethos-u when the bug is clearly and specifically in this plugin.

Development setup

Create a local environment with the project and all development dependencies:

uv sync --group dev

If you only need the test dependencies:

uv sync --group test

This repository currently does not use a committed lock file (uv.lock).

Common commands

Run the quick test suite used in CI:

uv run pytest -m "not slow" tests/

Run the full test suite with coverage:

uv sync --group test
uv run pytest tests/

Run the local quality checks:

uv run pre-commit run --all-files

Build the package:

uv build

Project layout

• src/mlia/target/ethos_u/: target integration, advisor logic, result-owned advice, and standardized-output collection for Ethos-U.
• src/mlia/backend/vela/: Vela backend integration.
• src/mlia/backend/corstone/: Corstone backend integration.
• src/mlia/resources/target_profiles/: bundled Ethos-U target profiles.
• src/mlia/resources/vela/: bundled Vela configuration.
• tests/: unit, integration, and CLI coverage for targets and backends.

Documentation

Additional package documentation lives in docs/README.md.

Releases

Latest changes and release history can be found in MLIA Ethos-U releases.

License

This project is licensed under the Apache License 2.0. See the license files for details.

Trademarks and copyrights

• Arm, Arm Ethos-U, Arm Cortex-M, and Arm Corstone are registered trademarks or trademarks of Arm Limited (or its subsidiaries) in the U.S. and/or elsewhere.
• TensorFlow is a trademark of Google LLC.
• PyTorch and ExecuTorch are trademarks of The Linux Foundation.
• Linux is the registered trademark of Linus Torvalds in the U.S. and elsewhere.
• Python is a registered trademark of the PSF.