Using ONNX as an intermediary

[modanesh]

I'm trying to load a pth model in Go. Can ONNX be used as an intermediary to convert a model from Pytorch's pth to gob in GoLang?

[xadupre]

I don't know too much about go. Once your model is exported in onnx, it can be seen as a long binary string. This package seems able to read it: https://pkg.go.dev/github.com/tynrol/onnx-go.

[modanesh]

Thanks @xadupre
I tried the ONNX-GO initially. From my understanding, ONNX-GO loads the model into a computation graph and then runs it. In the ONNX-GO's repo, it's recommended to use Gorgonia as the backend, and that's what I did. However, in order to run the model, it needs to already have all the required operators of the model implemented. And for my case, there was a missing operator. I created an issue in the ONNX-GO's repo (owulveryck/onnx-go#203), but no reply. That's when I went for plan B, which was to use ONNX as an intermediary to convert a PyTorch's pth model into a GoLang's gob, hence the question above.

[BEEugene]

There is also another project, have you tried that one? https://github.com/AdvancedClimateSystems/gonnx/tree/main