R Session Aborted

[feifeidrew]

After i installed torch package, I wanna see if it worked, so i tried
"torch_tensor(1:5)"
Then it said R Session Aborted.
Could anyone help me plz? thanks o(╥﹏╥)o

here's the code:

install.packages("torch")
torch::install_torch()
library(torch)

torch_tensor(1:5)

[dfalbel]

Thanks for reporting @feifeidrew !

What's your OS?
Were there any packages loaded prior to loading torch?

[glennfulford]

I get the same error on a windows 11 laptop. I am not using any GPU.

I was trying to run the first lab from the ISLR2 textbook and got this crash on windows, but it worked correctly on my MacBook Air laptop.

Then I tried the minimal example above and it also crashed RStudio R Session Aborted.

[dfalbel]

Thanks for reporting @glennfulford ! What's you RStudio version?
In the meantime, please install torch 0.13.

[glennfulford]

Sorry, I should have said. I am using RStudio version 2024.12.0+467
and R version 4.4.2 (both should be the latest of both.

When it didn't work I also tried to install the development version, but I seemed to be getting a lot of permission errors about not being able to copy dll files, e.g.

problem copying C:\Users\Glenn\AppData\Local\R\win-library\4.4\00LOCK-torch\torch\lib\c10.dll to C:\Users\Glenn\AppData\Local\R\win-library\4.4\torch\lib\c10.dll: Permission denied

torch version 14.4.1.

I will try to install the torch 13.1 version, if I can.

My student also got exactly the same error in his windows machine.

Thanks,
Glenn

[dfalbel]

Thanks @glennfulford ! Would you be able to also post your sessionInfo() after loading torch? eg:

library(torch)
sessionInfo()

I'm still unable to reproduce in my local Windows machine.

[dfalbel]

Actually, I was able to reproduce if installing using the pre-built binaries, but not if installing directly from CRAN.

[glennfulford]

This is my sessInfor() output.

R version 4.4.2 (2024-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 22631)

Matrix products: default

locale:
[1] LC_COLLATE=English_Australia.utf8 LC_CTYPE=English_Australia.utf8
[3] LC_MONETARY=English_Australia.utf8 LC_NUMERIC=C
[5] LC_TIME=English_Australia.utf8

time zone: Australia/Brisbane
tzcode source: internal

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] torch_0.14.1 Matrix_1.7-1

loaded via a namespace (and not attached):
[1] desc_1.4.3 coro_1.1.0 digest_0.6.37 R6_2.5.1
[5] fastmap_1.2.0 bit_4.5.0.1 xfun_0.50 lattice_0.22-6
[9] magrittr_2.0.3 glue_1.8.0 knitr_1.49 htmltools_0.5.8.1
[13] bit64_4.6.0-1 rmarkdown_2.29 ps_1.8.1 cli_3.6.3
[17] processx_3.8.5 callr_3.7.6 grid_4.4.2 withr_3.0.2
[21] compiler_4.4.2 rstudioapi_0.17.1 tools_4.4.2 evaluate_1.0.3
[25] Rcpp_1.0.14 yaml_2.3.10 rlang_1.1.5

[glennfulford]

Thanks Daniel, Your comment,

Actually, I was able to reproduce if installing using the pre-built binaries, but not if installing directly from CRAN.

I just want to clarify, do you mean you got it to work by installing from source, e.g.

install.packages("torch_0.14.1.tar.z"), after downloading this file, instead of the usual install.packages("torch") ?

When I tried that I got copy permission errors. Am I supposed to run it in Administrator mode?

I'll try it again after reinstalling R and Rstudio.

[glennfulford]

I tried the installation from source. It still crashes Rstudio.

[dfalbel]

To clarify, I doesn't crash for me if I do:

install.packages("torch")
library(torch)
# accept to download additional binaries

It crashes if I run:

options(timeout = 600) # increasing timeout is recommended since we will be downloading a 2GB file.
kind <- "cpu"
version <- "0.14.1"
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")

[glennfulford]

Actually, I was able to reproduce if installing using the pre-built binaries, but not if installing directly from CRAN.nks for the clarifacation.

For me it definitely crashes if I do

install.packages("torch")
library(torch)
# accept to download additional binaries

I thought that was installing the binary version.

I also tried 
install.packages("torch_0.14.1.tar.gz")

after downloading the .gz file from CRAN, and this appeared to compile the source, but it still crashed when I ran torch_tesnsor(1:5)

[dfalbel]

Hi @glennfulford !

I made a speculative fix on current dev version of torch. It would be nice if you could try it and let me know if this is still an issue:

remotes::install_github("mlverse/torch")

[Bernie-K]

I have the same issue with 0.14.1 CPU version.

sessionInfo()
R version 4.3.0 (2023-04-21 ucrt)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 11 x64 (build 22621)

Matrix products: default

locale:
[1] LC_COLLATE=German_Germany.utf8 LC_CTYPE=German_Germany.utf8 LC_MONETARY=German_Germany.utf8
[4] LC_NUMERIC=C LC_TIME=German_Germany.utf8

time zone: Europe/Berlin
tzcode source: internal

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] torch_0.14.1

loaded via a namespace (and not attached):
[1] desc_1.4.3 coro_1.1.0 R6_2.6.0 bit_4.5.0.1 magrittr_2.0.3 glue_1.8.0
[7] bit64_4.6.0-1 ps_1.8.1 cli_3.6.4 processx_3.8.5 callr_3.7.6 withr_3.0.2
[13] compiler_4.3.0 rstudioapi_0.17.1 tools_4.3.0 Rcpp_1.0.14 rlang_1.1.5

Rstudio has been updated to the latest version and all packages have been updated.

As soon as I call torch_tensor() a crash happens.
However, this only happens inside Rstudio. When running Rgui.exe or R from commandline everything is fine.
I can run operations on tensors and also the examples form the doc run through.

It must be something with the communication to Rstudio that is causing this behavior.

[dfalbel]

@Bernie-K

Can you install the dev version of torch, using remotes::install_github("mlverse/torch") and see if it fixes the issue?

Thanks!

[Bernie-K]

@dfalbel
Thanks, seems to fix the issue.
Tested training with luz on the MNIST example. Runs as expected.

[dfalbel]

Cool! I'll submit a new version to CRAN soon.

[glennfulford]

Using remotes::install_github("mlverse/torch") worked for me also.