Introduction to onnx-light

This example gives a brief overview of onnx-light, a library that works with ONNX models without requiring protobuf.

It demonstrates the use of the .. runpython:: directive (executed at documentation build time) and a .. mermaid:: diagram.

        graph LR
    A[Input tensor] --> B[ONNX Graph]
    B --> C[Output tensor]
    B --> D[onnx-light<br/>no protobuf]
    D --> C
    

Package version

The snippet below is executed with .. runpython:: so the output appears verbatim in the generated HTML page.

import onnx_light

print(f"onnx-light version: {onnx_light.__version__}")

onnx-light version: 0.1.0

What is onnx-light?

onnx-light manipulates ONNX models without the overhead of the protobuf serialization layer. The package exposes the same high-level abstractions (nodes, graphs, initializers) through plain Python data-classes, making it easier to embed in memory-constrained or dependency-free environments.