Chewxy

Looking into this

You need to generate the kernels for the CUDA program. There's a program called `cudagen` which generates all the operations

TODO: investigate the feasibility of applying #81 to this as well

That would be under a different function name I believe (which is not yet written): #23

I think this is fixed in #391

.... which I just noticed is not merged.

I just merged #391 . Should I tag this version?

Hmm, here's an interesting question: What should the shape of (1, 2)×(2) be? Numpy reports it to be (1). Gorgonia auto optimizes it away into a scalar.

Gorgonia's probably wrong

Yes. But you typically only care about the inputs. But yes, this should be added