Missing pre-built binaries for Windows

[statist-bhfz]

https://torch.mlverse.org/docs/articles/installation.html#pre-built

options(timeout = 600) # increasing timeout is recommended since we will be downloading a 2GB file.
# For Windows and Linux: "cpu", "cu117" are the only currently supported
# For MacOS the supported are: "cpu-intel" or "cpu-m1"
kind <- "cu118"
version <- available.packages()["torch","Version"]
options(repos = c(
  torch = sprintf("https://torch-cdn.mlverse.org/packages/%s/%s/", kind, version),
  CRAN = "https://cloud.r-project.org" # or any other from which you want to install the other R dependencies.
))
install.packages("torch")

Warning in install.packages :
  unable to access index for repository https://torch-cdn.mlverse.org/packages/cu118/0.14.1/src/contrib:
  cannot open URL 'https://torch-cdn.mlverse.org/packages/cu118/0.14.1/src/contrib/PACKAGES'
Warning in install.packages :
  unable to access index for repository https://torch-cdn.mlverse.org/packages/cu118/0.14.1/bin/windows/contrib/4.4:
  cannot open URL 'https://torch-cdn.mlverse.org/packages/cu118/0.14.1/bin/windows/contrib/4.4/PACKAGES'
trying URL 'https://cloud.r-project.org/bin/windows/contrib/4.4/torch_0.14.1.zip'
Content type 'application/zip' length 12854397 bytes (12.3 MB)
downloaded 12.3 MB

package ‘torch’ successfully unpacked and MD5 sums checked

torch package installed from https://cloud.r-project.org/bin/windows/contrib/4.4/torch_0.14.1.zip needs system CUDA 11.8 or 12.4:

> library(torch)
ℹ Additional software needs to be downloaded and installed for torch to work correctly.
Warning message:
ℹ Failed to install torch, manually run `install_torch()`
Unsupported CUDA version "11.7"
Caused by error in `check_supported_version()`:
✖ Unsupported CUDA version "11.7"
ℹ Currently supported versions are: "11.8" and "12.4".

[dfalbel]

Hi @statist-bhfz ,

Thanks for reporting! Can you try kind = "cu124"? I think we stopped building cu118, but forgot to update the docs.

[statist-bhfz]

@dfalbel thanks, installation with kind = "cu124" works nice!
Unfortunately, R session crashes immediately when I call any function from torch v.0.14.1 (e.g. torch_tensor(1, device = "cpu")) in RStudio (2024.12.0 Build 467) but not in RGui or Positron IDE. I'll check it on another PC and open an issue in https://github.com/rstudio/rstudio if it happens to be reproducible.

[dfalbel]

Interesting, we might have some clash with a library loaded by RStudio. What's your Rstudio version and Windows version?

[statist-bhfz]

I have checked it on 2 PCs, both with Windows 10 22H2 and RStudio 2024.12.0 Build 467.
Also tried to change rendering engine in RStudio (sofrware/desktop OpenGL) with no effect.
Logs at AppData\Local\RStudio\log don't contain any meaningful information, just 2 errors immediately after RStudio start

2025-02-10T06:59:25.545438Z [rsession-andrey] ERROR system error 42 (Unknown error (42)) [str: Модель, len: 12, from: UTF-8, to: ISO8859-1]; OCCURRED AT class rstudio::core::Error __cdecl rstudio::r::util::anonymous-namespace'::iconvstrImpl(const class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > &,const class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > &,const class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > &,bool,class std::basic_string<char,struct std::char_traits<char>,class std::allocator<char> > *) src/cpp/r/RUtil.cpp:268; LOGGED FROM: class rstudio::core::Error __cdecl rstudio::session::modules::spelling::anonymous-namespace'::checkSpelling(const struct rstudio::core::json::JsonRpcRequest &,class rstudio::core::json::JsonRpcResponse *) C:\Users\jenkins\workspace\ide-os-windows\rel-kousa-dogwood\src\cpp\session\modules\SessionSpelling.cpp:225

and no new errors after RStudio crash.

[Arnold-Kakas]

Hi, I have the same problem with RStudio 2024.09.0 Build 375

[dfalbel]

Does only the GPU version crash? Or does it also happen for the CPU version?

[statist-bhfz]

I tried

options(timeout = 600) # increasing timeout is recommended since we will be downloading a 2GB file.
# For Windows and Linux: "cpu", "cu117" are the only currently supported
# For MacOS the supported are: "cpu-intel" or "cpu-m1"
kind <- "cpu"
version <- available.packages()["torch","Version"]
options(repos = c(
    torch = sprintf("https://torch-cdn.mlverse.org/packages/%s/%s/", kind, version),
    CRAN = "https://cloud.r-project.org" # or any other from which you want to install the other R dependencies.
))
install.packages("torch")

It still crashes.

[dfalbel]

Hi @statist-bhfz

Could you try running:

options(timeout = 600) # increasing timeout is recommended since we will be downloading a 2GB file.
# For Windows and Linux: "cpu", "cu117" are the only currently supported
# For MacOS the supported are: "cpu-intel" or "cpu-m1"
kind <- "cpu"
version <- "0.14.1.9000
options(repos = c(
    torch = sprintf("https://torch-cdn.mlverse.org/packages/%s/%s/", kind, version),
    CRAN = "https://cloud.r-project.org" # or any other from which you want to install the other R dependencies.
))
install.packages("torch")

And see if the problem still occurs? Also in a few minutes we should be able to try the cuda version. We had a wrong version of cuDNN installed as a dependency.

[statist-bhfz]

@dfalbel
Thanks, it works! I've checked both "cpu" and "cu124".

[dfalbel]

Awesome!