How to Use onert Python API

CAUTION

This Python API is experimental yet. It can be changed later.

Prepare nnpackage

Use nnpackage examples

Use the nnpackage examples to run tutorial code.

Install onert Python API

You can install the package as follows:

$ pip install onert

By specifying the version, you can use a specific version of the package.

$ pip install onert==0.1.0

You can install latest developing package as follows:

$ pip install onert --pre

This definition has to be set at the top of the script using onert Python API.

import onert

Or you can import the onert module directly.

from onert.infer import *

This can be use onert session directly.

Example Code

1. Initialize session

import onert
 
# Create session and load nnpackage or modelfile
# The default value of backends is "cpu".
session = onert.infer.session(path, backends)

1. Prepare Input

# Prepare input. Here we just allocate dummy input arrays.
input_size = session.input_size()
session.set_inputs(input_size)

1. Inference

# Do inference
outputs = session.inference()

Run Inference with app on the target devices

reference app : minimal-python app

$ python3 minimal.py path_to_nnpackage_directory