ONNX (Open Neural Network Exchange) is an open file format for machine learning models, so a model trained in one framework can run in many other tools and runtimes.
In plain words
Models are trained in frameworks like PyTorch, but the place you want to run them might use a different engine or special hardware. ONNX describes a model as a computation graph built from a standard set of operators, the basic building blocks of models, with standard data types. Any tool that understands ONNX can load it, including runtimes tuned for particular hardware. It covers deep learning and traditional machine learning.
A quick example
Imagine a team trains an image classifier in PyTorch but wants to serve it with a different inference engine tuned for their hardware. They export the model to an .onnx file and load it with an ONNX-compatible runtime, without rebuilding the model.
Where you'll meet it
- Moving models between frameworks and runtimes.
- Deploying models with hardware-optimised runtimes.
Related
Sources
- DocsONNX, ONNX (Linux Foundation AI & Data) · read 29 Sept 2026
- Docsonnx/onnx, ONNX (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.