Installation issues about torch HOT 8 CLOSED

Hi!

Sorry, currently we don't support GPU builds on Windows. I just opened #265 to track it.

from torch.

lantern_install_path <- function() {
  normalizePath(file.path(system.file("", package = "torchr"), "deps"), mustWork = FALSE)
}

does not exist after I built from source... trying installation from github now.

from torch.

OK so it was the path problem... libtorch gets installed successfully if I install from github...

So I guess questions (1) and (2) still remain, and (3) would now be why - when I build myself - the package is not installed in the official R library?

from torch.

Should be mostly resolved with #74, now we install the GPU version if the right version of CUDA is detected. We can also manually run install_torch(type = "10.1") for a specific CUDA version.

from torch.

nice!

from torch.

Should be mostly resolved with #74, now we install the GPU version if the right version of CUDA is detected. We can also manually run install_torch(type = "10.1") for a specific CUDA version.

How do I install torch on windows with GPU support? I'm using 10.1. Tried the command above and got:

install_torch(type = "10.1")
Error in lantern_install_libs(version, type, path) : 
  The windows operating system is currently unsupported.
> 

By default it is installing the CPU version.

thanks
FKG

from torch.

Same error :( Perhaps install_torch() could detect/parse presence of CUDA_PATH environment variable?

from torch.

We now support GPU's on Windows with #281 .

from torch.