Support for accelerate or other GPU acceleration libs

[Abastro]

Hello,
I'd like to have it being able to pass the GPU tensor to accelerate to calculate some stuffs there. Is this supported? If not, is it feasible?
For now, I found a way to pass GPU pointer to the accelerate, so I wonder if there is a way to extract the pointer from the tensor.

[junjihashimoto]

It is possible to pass CPU's tensor-data via bytestring like this code.
https://github.com/hasktorch/hasktorch/blob/master/hasktorch/src/Torch/Vision.hs#L84-L103
We may be able to do something for GPU's tensor, but I've never tried it.