Investigate possibility to automate cuda runtime installation

[dfalbel]

@dfalbel I also think it may not be worth the extra effort to debug this and deprecating it makes sense, but as a potential solution, does the current R version of torch come with somethings similar to this in PyTorch? If not, would it make sense to have it in order to improve CUDA compatibility?

Originally posted by @Unco3892 in https://github.com/mlverse/torch/issues/577#issuecomment-852142276

[mohamed-180]

I think CUDA version which installed in os matters if pytorch compliled from source which R torch case and it will compatible to the older versions .