/artemis-pytorch

Pytorch for Artemis Centos 16.04 CUDA 10 and V100s

Primary LanguageDockerfile

Docker/Singularity repo for pytorch and https://github.com/facebookresearch/SlowFast

If you have used this work, you must acknoledge SIH, e.g: "The authors acknowledge the technical assistance provided by the Sydney Informatics Hub, a Core Research Facility of the University of Sydney."

This repo contains the recipe Dockerfile and also an equivalent recipe for Singularity. On the Artemis HPC you must build all Singulairt container/images on a computer which has Singularity installed and you have root access on. Once the image is built then copy the image to Artemis and run.

Docker instructions

To get this repo and scripts or to build locally

git clone https://github.com/natbutter/artemis-pytorch.git
sudo docker build  . -t pyslowfast 
#or
sudo docker build nbutter/pyfastslow

To run a Python script (e.g. testGPU.py) that is in your current directory on your host by mounting inside the container in the project directory and then executing with docker:

sudo docker run -v `pwd`:/project nbutter/pyfastslow /bin/bash -c "cd /project && python testGPU.py"

Singularity Instructions

git clone https://github.com/natbutter/artemis-pytorch.git
sudo singularity build --writable pyslowfast.img pyslowfast.build
#or
sudo singularity build --writable pyslowfast.img docker://nbutter/pyfastslow

To run with singularity in a PBS jobscript:

#!/bin/bash
#PBS -P <yourproject> 
#PBS -N myrun 
#PBS -l select=1:ncpus=1:mem=1gb:ngpus=1
#PBS -l walltime=00:01:00
#PBS -q defaultQ

module load singularity

cd $PBS_O_WORKDIR

#Mount the current directory in the container at /project, change directories and run your script
singularity exec -B `pwd`:/project ../pyslowfast.img /bin/bash -c "cd /project && python testGPU.py"