run_sbatch_efficient.slurm

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=16
#SBATCH -p <queue>
#SBATCH -J cerranet-eff
#SBATCH --no-requeue
#SBATCH --time=02:00:00
#SBATCH --mail-type=BEGIN,END,TIME_LIMIT_50
#SBATCH --mail-user=<your email>

echo '========================================'
echo '- Job ID:' $SLURM_JOB_ID
echo '- # of nodes in the job:' $SLURM_JOB_NUM_NODES
echo '- # of tasks per node:' $SLURM_NTASKS_PER_NODE
echo '- # of tasks:' $SLURM_NTASKS
echo '- # of cpus per task:' $SLURM_CPUS_PER_TASK
echo '- Dir from which sbatch was invoked:' ${SLURM_SUBMIT_DIR##*/}
echo -n '- Nodes allocated to the job: '
nodeset -e $SLURM_JOB_NODELIST

#Change dir to the dir where the job was invoked
cd $SLURM_SUBMIT_DIR

# Run your code
echo -n '<1. starting python script > ' && date
echo '-- output -----------------------------'

#executa o script

echo 'train and eval on last model!'
singularity run --nv -B /path/for/the/repo/BCD-Code /path/for/the/singularity/file/container.sif sh -c 'cd $SLURM_SUBMIT_DIR && python trainer.py --arch efficientnet_b4 --runs 5 --ts 0.002'

echo 'eval on best model!'
singularity run --nv -B /path/for/the/repo/BCD-Code /path/for/the/singularity/file/container.sif sh -c 'cd $SLURM_SUBMIT_DIR && python trainer.py --arch efficientnet_b4 --runs 5 --ts 0.002 --resume --evaluate'

echo '-- end --------------------------------'
echo -n '<2. quit>                    ' && date