onnxsim-prebuilt

ONNX Simplifier Prebuilt

onnxsim-prebuilt is a fork of onnxsim that aims to publish prebuilt wheels for Python 3.12 and later to PyPI.

Changes in this fork

• Changed package name to onnxsim-prebuilt
  • The library name remains unchanged from onnxsim, so you can import it as import onnxsim just like the original onnxsim
  • Can be used as a drop-in replacement for the original onnxsim
• Publish prebuilt wheels for all platforms (Windows, macOS x64/arm64, Linux x64/arm64) on PyPI
  • onnx-simplifier depends on C++, CMake, and submodules, making the build environment setup relatively difficult and time-consuming
    • For over a year, onnxsim has not been updated, and prebuilt wheels for Python 3.12/3.13 are not available (ref: onnxsim/issues/334, onnxsim/pull/359)
    • Various issues arise, such as the need to install build-essentials and CMake in Docker images just for installation, and long build times
  • By publishing prebuilt wheels on PyPI, we aim to enable easy installation even on PCs without a build environment
  • Incorporated the CI improvements proposed in the pull request onnxsim/pull/359, and further enhanced it to build and publish prebuilt wheels for Linux aarch64
• Explicitly added Python 3.12 / 3.13 to supported versions
  • Changed CI target Python versions to Python 3.10 and above
  • This fork does not support Python 3.9 and below

Installation

You can install the library by running the following command:

pip install onnxsim-prebuilt

The documentation below is inherited from the original onnxsim without any modifications.
There is no guarantee that the content of this documentation applies to onnxsim-prebuilt.

ONNX Simplifier

ONNX is great, but sometimes too complicated.

Background

One day I wanted to export the following simple reshape operation to ONNX:

import torch


class JustReshape(torch.nn.Module):
    def __init__(self):
        super(JustReshape, self).__init__()

    def forward(self, x):
        return x.view((x.shape[0], x.shape[1], x.shape[3], x.shape[2]))


net = JustReshape()
model_name = 'just_reshape.onnx'
dummy_input = torch.randn(2, 3, 4, 5)
torch.onnx.export(net, dummy_input, model_name, input_names=['input'], output_names=['output'])

The input shape in this model is static, so what I expected is

However, I got the following complicated model instead:

Our solution

ONNX Simplifier is presented to simplify the ONNX model. It infers the whole computation graph and then replaces the redundant operators with their constant outputs (a.k.a. constant folding).

Web version

We have published ONNX Simplifier on convertmodel.com. It works out of the box and doesn't need any installation. Note that it runs in the browser locally and your model is completely safe.

Python version

pip3 install -U pip && pip3 install onnxsim

Then

onnxsim input_onnx_model output_onnx_model

For more advanced features, try the following command for help message

onnxsim -h

Demonstration

An overall comparison between a complicated model and its simplified version:

In-script workflow

If you would like to embed ONNX simplifier python package in another script, it is just that simple.

import onnx
from onnxsim import simplify

# load your predefined ONNX model
model = onnx.load(filename)

# convert model
model_simp, check = simplify(model)

assert check, "Simplified ONNX model could not be validated"

# use model_simp as a standard ONNX model object

You can see more details of the API in onnxsim/onnx_simplifier.py

Projects Using ONNX Simplifier

• MXNet
• MMDetection
• YOLOv5
• ncnn
• ...

Chat

We created a Chinese QQ group for ONNX!

ONNX QQ Group (Chinese): 1021964010, verification code: nndab. Welcome to join!

For English users, I'm active on the ONNX Slack. You can find and chat with me (daquexian) there.