Load model from pyTorch issue

I try load trained nn in pyTorch(in pth file format) to TorchSharp but give some errors.

1. from static method Module module = Module.Load("C:\\ir50-model.pth"); module.forward(tensor_image); throw System.NotImplementedException: ‘forward(t)’

2. useStrict: false

var myModel = new MyModel("test");
myModel.load("C:\\ir50-model.pth", false); ;
myModel.forward(tensor_image);

public class MyModel : Module
{
        public MyModel() : base("test") {}      

        public MyModel(string name) : base(name) {}     

        public override Tensor forward(Tensor input)
        {
            return this.forward(input);
        }
}

throw System.FormatException: ‘Too many bytes in what should have been a 7-bit encoded integer.’

Does it exist worked way to load pth model from pyTorch to TorchSharp ?

[NiklasGustafsson]

Sorry, I’ve been out on vacation for a month. First day back…

What does the Python model that you are exporting look like? It has to have the same components as the model you’re loading the weights into.

From the code you included, it looks like:

1. There are no submodules in your model. Is this intentional, or did you simplify it for the purpose of brevity in the issue report?
2. Your forward() function recurses infinitely.

UPDATE: I tried to reproduce the error on my own, but it didn’t fail. It would be helpful to have the source code for the Python model that you exported. I don’t need the training code, just the definition of the model class.

[GeorgeS2019]

@anytoany

Please see a work around using the recommended save_dict codes