ML Inference Advisor

Use ML Inference Advisor (MLIA) to check model compatibility and estimate performance for supported targets. The core package provides the CLI, Python API, backend management, and standardized results. Plugins add targets, conversion paths, and visualization.

Get started

Use Python 3.10 or newer on Ubuntu 22.04 LTS or a compatible Linux environment. Install libpython3.10-dev if your environment requires it. We recommend a virtual environment.

Install a target plugin for your model and target. It installs a compatible version of MLIA as a dependency. For example, for Arm Ethos-U:

pip install mlia-ethos-u

If you also need PyTorch model conversion, install the converter alongside the target plugin:

pip install mlia-ethos-u mlia-converters-pytorch

If you do not need a target plugin, install the core package with pip install mlia. See the plugin guides for other targets, model formats, and additional setup.

Check your installation with mlia --help, then discover targets and backends. When you have chosen a target, run an analysis and understand the results.

Find the right guide

I want to…	Start here
Understand the core package	Overview
Run MLIA from the terminal	CLI guide
Integrate model analysis in Python	Python API
Discover and manage analysis backends	Backends
Read JSON results and metrics	Outputs
Follow a guided tutorial	Learning Paths
Understand the runtime components	High-level architecture
Build or extend a plugin	Plugin architecture
Follow an analysis from input to output	Execution flow
Contribute to the project	Development

Targets, converters, and visualization

Choose the plugin documentation for target-specific configuration, supported model formats, and installation requirements.

Plugin	Documentation
Neural Technology targets and backends	Neural Technology
Ethos-U targets, Vela, and Corstone	Ethos-U
LiteRT model conversion	LiteRT converters
PyTorch and PTE conversion	PyTorch converters
Model and analysis visualization	Model Explorer