Is there any way to declare a "placeholder" variable in PyTorch? I think I need to either create a Variable without a Tensor or create a Tensor without a fully-defined shape (i.e. [None, 3, 10, 30]).

I'm investigating the concept of adding a PyTorch backend for Keras (not currently volunteering to do it) and ran into this issue immediately (see implementations for theano, tensorflow). I see that you've already brought up the idea of implementing a Keras backend, so I thought you might have given this some thought.

Keras Integration
Have a keras backend. Send in a Pull Request to fchollet/keras
#5 (comment)

Thanks!