Installing_torch: Installing torch R library (on Linux)

In seriph78/COTAN: COexpression Tables ANalysis

Installing torch R library (on Linux)

Description

A brief explanation of how to install the torch package on WSL2 (Windows Subsystem for Linux), but it might work the same for other Linux systems. Naturally it makes a difference whether one wants to install support only for the CPU or also have the system GPU at the ready!

The main resources to install torch is https://torch.mlverse.org/docs/articles/installation.html or https://cran.r-project.org/web/packages/torch/vignettes/installation.html

Details

For the CPU-only support one need to ensure that also numeric libraries are installed, like BLAS and LAPACK and/or MKL if your CPU is from Intel. Otherwise torch will be stuck at using a single core for all computations.

For the GPU, currently only cuda devices are supported. Moreover only some specific versions of cuda (and corresponding cudnn) are effectively usable, so one needs to install them to actually use the GPU.

As of today only cuda 11.7 and 11.8 are supported, but check the torch documentation for more up-to-date information. Before downgrading your cuda version, please be aware that it is possible to maintain separate main versions of cuda at the same time on the system: that is one can have installed both 11.8 and a 12.4 cuda versions on the same system.

Below a link to install cuda 11.8 for WSL2 given: use a local installer to be sure the wanted cuda version is being installed, and not the latest one: cuda 11.8 for WSL2

seriph78/COTAN documentation built on Dec. 10, 2024, 3:30 a.m.