How does one store Pytorch models without having to store the parameters?

1

I want to only store the layers block and the computation graph.

To start off I want it to work for Sequential models. I want to be able to store the Sequential model architecture without having to store the parameters. Is that possible?

After that, is it generalizable for arbitrary models? I only want to store the shape and types of the pytorch graph. I don’t need the parameters.

2

I think the easiest way would be to store the model class definition, if you don’t want the parameters.

3

hmmm what does that mean? how does one do that?

I’ve been just saving the string of the Sequential model for the moment…but then I need to parse the string to extract things from it or even worse use an eval function…