[WIP 2023] Integrating Custom PyTorch c++ Extensions

[GeorgeS2019]

Update March 2023

Recently the ONNXRUNTIME project has updated extensively the documentations on Contributed Operators and Custom Operators

May2022: CUSTOM C++ AND CUDA EXTENSIONS

CUSTOM C++ AND CUDA EXTENSIONS

This will be a WIP discussion as the steps needed to bring pytorch extension is unclear and what are the values and benefits of doing so.

For an example of a pytorch custom c++ extension: pytorch2onnx_custom_op

For a list of Contributed Operators in c++ that are now provided under the framework of OnnxRuntime

Many of these contributed operators are HuggingFace Transformer deep NLP related and could be relevant to the discussion: Language translation with transformer using TorchData and TorchText

[GeorgeS2019]

For relevant .NET deep NLP "Operators" reusability discussion

[GeorgeS2019]

Another relevant discussion here: NER transformers model packed in one onnx file with Sentencepiece Tokenizer. Especially here