Description
Describe the issue
OpenVINO is more suitable for extending PyTorch (as it already doing as mentioned in the following issue openvinotoolkit/openvino#16472)
The reason is that it supports multiple-devices, different plugins, extensions, IR optimizations and etc.
OpenVINO is already doing evaluation aka inference and also it could be used for training like mentioned in https://github.com/openvinotoolkit/openvino_tensorflow for PyTorch
I understand that intel-extension-for-pytorch is mostly concentrated on implementing this extension with oneAPI, but it could be done as separate plugin in OpenVINO infrastructure named for example one_plugin
As for me a lot of the same work done in both teams
Related issue:
Use OpenVINO as backend for Tensorflow and PyTorch for training
[Question] Why not to use OpenVINO for extending Tensorflow ?
[Question] Why not to use OpenVINO for extending PyTorch as it is already doing ?