train_baseline-lions.sh

#!/usr/bin/env bash
#$ -binding linear:4 # request 4 cpus (8 with Hyperthreading) (some recommend 4 per GPU)
#$ -N ex       # set consistent base name for output and error file (allows for easy deletion alias)
#$ -q all.q    # don't fill the qlogin queue (can some add why and when to use?)
#$ -cwd        # change working directory (to current)
#$ -V          # provide environment variables
# -t 1-10    # start 100 instances: from 1 to 100
# if you also want to request a GPU, add the following line to the above block:
#$ -l cuda=1   # request one GPU
#$ -j y
#$ -l h='!node4*'
#$ -l h_vmem=16G
#$ -l mem_free=16G
echo "I am a job task with ID $SGE_TASK_ID."
export CUDA_LAUNCH_BLOCKING=1

dataset=$1
batch_size=$2
epochs=$3
resplit=$4
local=$5
resized=$6
checkpoint=$7
run_folder=$8
accuracy=$9

source /home/salnabulsi/.thesis-py38/bin/activate && python train_baseline.py --dataset ${dataset} --batch-size ${batch_size} --epochs ${epochs} --resplit ${resplit} --local ${local} --resized-images ${resized} --checkpoint ${checkpoint} --run-folder ${run_folder} --accuracy ${accuracy} --seed ${10}