Slurm: Submitting a batch job🔗

Prev: Slurm: Running an interactive job

⚠ This guide requires that you have completed: How to login

The benefits of batch jobs include automation and the ability to have numerous jobs in the queue simultaneously, thanks to Slurm's job scheduling.

#!/bin/bash
#SBATCH --account=C3SE-STAFF
#SBATCH --gpus-per-node=A40:4
#SBATCH --time=01:00:00

module load pytorch
~/my_ml_pytorch_script.py --epochs=12 --mnist_src /mimer/NOBACKUP/Datasets/MNIST/raw

Explanation:

The sbatch command schedules a job on a compute node.

#!/bin/bash: A "shebang" indicating which interpreter should run this script.
#SBATCH --account=C3SE-STAFF: Slurm should bill the C3SE-STAFF account.
#SBATCH --gpus-per-node=A40:4: Requests four (4) A40 GPUs.
#SBATCH --time=01:00:00: Requests a runtime of 1 hour.

The following lines are executed as part of the script on the compute node:

module load pytorch: Loads the pytorch module from the module system.
./my_ml_pytorch_script --epochs 12 --mnist-src /mimer/NOBACKUP/Datasets/MNIST/raw: Starts an application.

For a more comprehensive guide to Slurm and how to script jobs, please see Running jobs.

Next: Slurm: Running an interactive job