search

dotnet / TorchSharp

A .NET library that provides access to the library that powers PyTorch.
MIT License
1.4k stars 182 forks source link

Add functionality to save models #85

Closed interesaaat closed 4 years ago

interesaaat commented 5 years ago

This will probably require a custom save logic or hooks into JIT save.

mdabros commented 5 years ago

@interesaaat Will this include save/load of the models using the ONNX format?

interesaaat commented 5 years ago

@mdabros that may require some work. Apparently ONNX support in libTorch is limited for the moment. My hope is that in future release they will provide the torch.onnx.export functionality also in libTorch. If they don't we will have to write our own to-onnx-exporter.

mdabros commented 5 years ago

I see, that makes it more complicated. I was hoping the functionality was already there in libTorch. I hope they will add it soon. ONNX support in TorchSharp would really help make it a perfect choice for developing deep learning models for production. But I guess it is a separate issue from this one then.

interesaaat commented 5 years ago

I totally agree. Maybe we can open an issue on the main PyTorch repo? Actually let me ask on the PyTorch's slack channel.

dsyme commented 4 years ago

This is covered by #146

nietras commented 4 years ago

@dsyme does #146 include onnx support (save as)?

dsyme commented 4 years ago

@dsyme does #146 include onnx support (save as)?

It's not mentioned there - I'll add it - I presume it's a pre-requisite. If you'd like to contribute a PR for this that would be great