added a pytorch cuda example

96-frameworks-pytorch-cuda/01-pytorch-python-single-gpu.sh

+#!/bin/bash
+
+##############################################################################
+# User Request:
+#   - allocate single gpu node using 1 A40 GPU
+#	- run python script
+#
+# Provided Allocation:
+#   - 1 shared gpu node
+#   - 1 A40 GPU
+#   - 8 physical cores / 16 logical cores
+#   - 128 GB memory
+#
+# VSC policy:
+#   - '--ntasks-per-node' & '--ntasks' implicitly set to 8
+#   
+# Accounting:
+#   - 1x 8 core hours / hour
+##############################################################################
+
+#SBATCH --partition=zen2_0256_a40x2
+#SBATCH --qos=zen2_0256_a40x2
+#SBATCH --gres=gpu:1
+
+# optionally activate a conda or python environment
+module load miniconda3
+eval "$(conda shell.bash hook)"
+conda activate pytorch-cuda
+
+# run python code
+python pytorch-test.py

96-frameworks-pytorch-cuda/README.md

+# Pytorch CUDA Environment
+
+To install the environment simply use conda
+
+```bash
+module load miniconda3
+eval "$(conda shell.bash hook)"
+conda env create -f environment.yaml
+```
+
+To run a test use `01-pytorch-python-single-gpu.sh`
+
+```bash
+sbatch 01-pytorch-python-single-gpu.sh
+```
+
+## Compiling CUDA codes
+
+https://conda-forge.org/docs/user/faq/#how-can-i-compile-cuda-host-or-device-codes-in-my-environment
+
+For this to work we have to load a CUDA toolkit installation from our cuda-zen spack environment e.g.
+
+```bash
+cuz     # switch to cuda-zen spack env
+module load cuda/12.3.0-gcc-9.5.0-imxfeek
+```
+
+Alternatively it should be possible to use the nvidia packages (e.g. `nvidia::cuda-toolkit`) instead of the 
+conda-forge ones. However in this case we need to be careful with the dependencies so that we dont mix the packages 
+and installations of the different sources ...
\ No newline at end of file

96-frameworks-pytorch-cuda/environment.yaml

+name: pytorch-cuda
+channels:
+  - pytorch
+  - conda-forge
+  - bioconda
+  - nvidia  
+  - defaults
+dependencies:
+  - python=3.*
+  - pip
+  - pytorch=*=*cuda12*cudnn8.9*
+  - cudnn=8.9*
+  - cuda-version=12.1*
+  - nvcc
+  - keras
+  - nccl

96-frameworks-pytorch-cuda/pytorch-test.py

+from torch import cuda
+
+print("CUDA Available: ", cuda.is_available())
+print("Num GPUs Available: ", cuda.device_count())
+print("First Device Info: ", cuda.get_device_name(cuda.current_device()))