zihaomu/onnx

Analysis updated 2026-07-07 · repo last pushed 2022-02-10

Audience · developerComplexity · 3/5DormantLicenseSetup · moderate

Mindmap

mindmap
  root((repo))
    What it does
      Universal model format
      Graph of operations
      Cross-framework portability
    Use cases
      Deploy PyTorch models elsewhere
      Optimize computation graphs
      Check data shapes
    Tech stack
      Python
      C++
      Protobuf
    Audience
      Data scientists
      ML engineers
      Open source community

mindmap root((repo)) What it does Universal model format Graph of operations Cross-framework portability Use cases Deploy PyTorch models elsewhere Optimize computation graphs Check data shapes Tech stack Python C++ Protobuf Audience Data scientists ML engineers Open source community

Click or tap to explore — scroll the page freely

What do people build with it?

USE CASE 1

Convert a PyTorch model to ONNX format so it can run on a different inference server.

USE CASE 2

Check that the data shapes flowing through a model are correct before deployment.

USE CASE 3

Optimize a model's computation graph to improve inference speed.

USE CASE 4

Convert an older ONNX model to a newer version of the standard.

What is it built with?

PythonC++Protobuf

How does it compare?

	zihaomu/onnx	0xhassaan/nn-from-scratch	0xzgbot/hermes-comfyui-skills
Stars	—	0	0
Language	—	Python	—
Last pushed	2022-02-10	—	—
Maintenance	Dormant	—	—
Setup difficulty	moderate	moderate	easy
Complexity	3/5	4/5	1/5
Audience	developer	developer	designer

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · moderate Time to first run · 30min

You typically need a model already trained in a framework like PyTorch or TensorFlow to convert, plus an ONNX-compatible runtime installed to run inference.

Use freely for any purpose, including commercial use, as long as you keep the copyright notice.

In plain English

Imagine you built a machine learning model using one set of tools, but now you need to deploy it using a completely different set of tools. Normally, this would be like trying to use a Mac program on a Windows computer, it just doesn't work without some kind of translation. ONNX solves this problem by providing a universal file format for AI models. You can think of it as a common language that lets you train a model in one framework and run it in another, without having to rewrite or retrain anything. At its core, ONNX defines a standard way to represent the math that a machine learning model does. When you convert a model to the ONNX format, it gets broken down into a graph of operations, things like additions, multiplications, and other computations, along with the data types those operations use. Any tool or hardware that understands this standard can then load and run the model. The project focuses specifically on "inferencing," which means using a finished model to make predictions, rather than training new models from scratch. This is useful for data scientists and engineering teams who want flexibility in their toolchain. For example, a researcher might train a model using PyTorch but then need to deploy it on a server that only supports a different framework. Instead of being locked into one ecosystem, they can export the model to ONNX and move freely between tools. The project is widely supported across the AI industry, with many popular frameworks, tools, and hardware platforms able to read and run ONNX files. The project also includes utilities for working with ONNX models once they are in this format, such as checking the shapes of data flowing through the model, optimizing the computation graph, and converting between different versions of the standard. It is an open-source, community-driven project that welcomes contributions, including proposals for new operations to add to the specification.

Copy-paste prompts

Prompt 1

Help me convert a PyTorch model to ONNX format using torch.onnx.export, including dynamic batch dimensions and a basic example with a ResNet model.

Prompt 2

I have an ONNX model file. Write a Python script using onnx and onnxruntime that loads the model, checks its input and output shapes, and runs a sample inference with random data.

Prompt 3

Write a script that loads an ONNX model, runs ONNX shape inference on it, and then uses onnx.optimizer to optimize the computation graph, saving the result as a new file.

Prompt 4

Create a Python utility that takes an ONNX model file, checks its current IR version, and upgrades it to the latest opset version using the onnx.version_converter.

Frequently asked questions

What is onnx?

ONNX is an open standard that lets you convert AI models into a universal file format so they can run on any framework or hardware without retraining.

Is onnx actively maintained?

Dormant — no commits in 2+ years (last push 2022-02-10).

What license does onnx use?

Use freely for any purpose, including commercial use, as long as you keep the copyright notice.

How hard is onnx to set up?

Setup difficulty is rated moderate, with roughly 30min to a first successful run.

Who is onnx for?

Mainly developer.

Open on GitHub → Explain another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.