-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.sh
37 lines (36 loc) · 1.58 KB
/
run.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
#!/bin/bash
# Is specified via environment variable, execute commands in a job scheduled via
# sbatch
if [ "$RUN_ON_NOCTUA" = 1 ]; then
# Specify a name for the SLURM job
JOB_NAME="-J ${JOB_NAME:=run}"
# Hard time limit of the job, will forcefully be terminated if not done within
# this time
TIME_LIMIT="-t ${TIME_LIMIT:=01:00:00}"
# Number of CPUs to use per task
NUM_CPUS="--cpus-per-task=${NUM_CPUS:=16}"
# Amount of memory to allocate for the job
MEM="--mem ${MEM:=64G}"
# The partition to which the job is submitted
PARTITION="-p ${PARTITION:=normal}"
# Notify by mail on all events (queue, start, stop, fail, ...)
MAIL="--mail-type FAIL --mail-user [email protected]"
# If using GPUS, specify which type of GPU and how many
# Note: Hardcode this to 1, there is no need for more GPUs right now
if [[ "$PARTITION" = "-p gpu" ]] || [[ "$PARTITION" = "-p dgx" ]]; then
# Note: Hardcode this to 1, there is no need for more GPUs right now
GPUS="--gres=gpu:a100:1"
fi;
# Group all sbatch command line arguments into one string
ARGS="$JOB_NAME $TIME_LIMIT $NUM_CPUS $MEM $PARTITION $MAIL $GPUS"
# Forward all arguments following the shell script to be executed as the
# command line inside of another shell script which is executed via sbatch
# Note: Waiting/Blocking sbatch with the -W option
# shellcheck disable=SC2086
sbatch $ARGS -W --verbose noctua.sh "$@"
# By default, execute the job locally
else
# Forward all arguments following the shell script to be executed as the
# command line within this script
eval "$@";
fi;