We have pytorch code which fits in to the kinds of models you have here, which is already very comprehensive. It is called hippynn.
https://github.com/lanl/hippynn
I believe this is mostly a matter of building a small interface. Our pytorch code itself is not abstracted; it takes in tensors and outputs tensors. I think that the core code to make a wrapper of is here:
https://github.com/lanl/hippynn/blob/development/hippynn/layers/hiplayers.py
We have pytorch code which fits in to the kinds of models you have here, which is already very comprehensive. It is called hippynn.
https://github.com/lanl/hippynn
I believe this is mostly a matter of building a small interface. Our pytorch code itself is not abstracted; it takes in tensors and outputs tensors. I think that the core code to make a wrapper of is here:
https://github.com/lanl/hippynn/blob/development/hippynn/layers/hiplayers.py