Skip to content

Introduction

Libraries such as Tensorflow and PyTorch have a set of tested dependencies. Using the specific libraries recommended by the developers can help reduce GPU compatiblity issues on the cluster.

Tensorflow

You can locate Tensorflow's Tested Build Configurations here.

For example, if you are adopting Tensorflow version 2.17, the build has been tested against Python versions 3.9 to 3.12, cuDNN 8.9 and CUDA 12.3.

Insert the following lines into your sbatch template script after line 52 if you are using Tensorflow version 2.15

52
53
54
55
56
# Purge the enviromnent, load the modules we require.
# Refer to https://violet.smu.edu.sg/origami/module/ for more information
module purge
module load Python/3.11.7
module load cuDNN/8.9.7.29-CUDA-12.3.2

PyTorch

Simiarly, PyTorch has a recommended set build configuration as well.

Use the pip3 command to install PyTorch —

pip3 install torch torchvision torchaudio

Insert the following lines into your sbatch template script after line 52

52
53
54
55
56
# Purge the enviromnent, load the modules we require.
# Refer to https://violet.smu.edu.sg/origami/module/ for more information
module purge
module load Python/3.11.7
module load CUDA/12.4.0