QONNX

1.0.0 verified Fri May 01 auth: no python

Frontend and utilities for Quantized ONNX (QONNX), a representation for quantized neural networks built on ONNX. Current version 1.0.0. Release cadence is irregular, approximately quarterly.

pip install qonnx

Common errors

error ModuleNotFoundError: No module named 'qonnx' ↓

cause The package is not installed.

fix

Run: pip install qonnx

error AttributeError: module 'qonnx' has no attribute 'ModelWrapper' ↓

cause Importing directly from the top-level module instead of the correct submodule.

fix

Use: from qonnx.core.modelwrapper import ModelWrapper

Warnings

breaking In v1.0.0, the QKeras converter and Brevitas dependency were removed. Converters must now be implemented in companion packages. ↓

fix Use external converter packages for QKeras or Brevitas models.

deprecated Functions and utilities migrated to qonnx.util or qonnx.transforms may have changed signatures. Check the new module paths. ↓

fix Review the release notes for full list of changes.

Imports

ModelWrapper

from qonnx.core.modelwrapper import ModelWrapper

QuantConfig

from qonnx.core.quant import QuantConfig

Quickstart

Load and save a QONNX model. Replace 'model.onnx' with your file.

from qonnx.core.modelwrapper import ModelWrapper
model = ModelWrapper('model.onnx')
model.save('output.onnx')