Hands-On Machine Learning with C++

Using TorchScript for a model snapshot

In this section, we will discuss how to get the model snapshot file so that we can use it in our mobile application. In the previous chapters, we discussed how to save and load model parameters and how to use ONNX format to share models between frameworks. When we use the PyTorch framework, there is another method we can use to share models between the Python API and C++ API called TorchScript.

This method uses real-time model tracing to get a special type of model definition that can be executed by the PyTorch engine, regardless of API. For PyTorch 1.2, only the Python API can create such definitions, but we can use the C++ API to load the model and execute it. Also, the mobile version of the PyTorch ...