Execution¶
Through Docker¶
# Example: Train MNIST on NVIDIA GPUs
docker run --privileged --gpus all --rm -e AI_REPO_PATH=${AI_REPO_PATH} \
-e PYTHONPATH=${PYTHONPATH}:${AI_REPO_PATH} \
-v ${AI_REPO_PATH}:${AI_REPO_PATH} -v /dev/shm:/dev/shm \
-w ${AI_REPO_PATH} mleclei/ai_repo:latest \
python -m project.classify_mnist task=mlp
Through Apptainer¶
# Example: Train MNIST on the Béluga cluster
module load apptainer && cd ${AI_REPO_PATH} && export PYTHONPATH=${PYTHONPATH}:${AI_REPO_PATH} && \
export APPTAINERENV_APPEND_PATH=/opt/software/slurm/bin:/cvmfs/soft.computecanada.ca/easybuild/software/2020/Core/apptainer/1.1.8/bin && \
apptainer exec --no-home -B /etc/passwd -B /etc/slurm/ -B /opt/software/slurm -B /usr/lib64/libmunge.so.2 \
-B /cvmfs/soft.computecanada.ca/easybuild/software/2020/Core/apptainer/1.1.8/bin/apptainer \
-B /var/run/munge/ --env LD_LIBRARY_PATH=/opt/software/slurm/lib64/slurm -B $AI_REPO_PATH $SCRATCH/ai_repo.sif \
python -m project.classify_mnist task=mlp_beluga