Converting a PyTorch* Model

Typical Steps to Convert PyTorch Model

PyTorch* framework is supported through export to ONNX* format. A summary of the steps for optimizing and deploying a model that was trained with the PyTorch* framework:

  1. Export PyTorch model to ONNX*.

  2. Convert an ONNX\* model to produce an optimized Intermediate Representation (IR) of the model based on the trained network topology, weights, and biases values.

Export PyTorch* Model to ONNX* Format

PyTorch models are defined in a Python* code, to export such models use torch.onnx.export() method. Usually code to evaluate or test the model is provided with the model code and can be used to initialize and export model. Only the basics will be covered here, the step to export to ONNX* is crucial but it is covered by PyTorch* framework. For more information, please refer to Exporting PyTorch models to ONNX format.

To export a PyTorch* model you need to obtain the model as an instance of torch.nn.Module class and call the export function.

import torch

# Instantiate your model. This is just a regular PyTorch model that will be exported in the following steps.
model = SomeModel()
# Evaluate the model to switch some operations from training mode to inference.
# Create dummy input for the model. It will be used to run the model inside export function.
dummy_input = torch.randn(1, 3, 224, 224)
# Call the export function
torch.onnx.export(model, (dummy_input, ), 'model.onnx')

Known Issues

  • Not all PyTorch* operations can be exported to ONNX* opset 9 which is used by default, as of version 1.8.1. It is recommended to export models to opset 11 or higher when export to default opset 9 is not working. In that case, use opset_version option of the torch.onnx.export. For more information about ONNX* opset, refer to the Operator Schemas.