mv annotation utils and slurm files

dataset/Dataset.py

@@ -8,8 +8,7 @@
 import torch
 from tqdm import tqdm
 
-from utilities import augmentations
-import annotation_utils
+from utilities import augmentations, annotation_utils
 import random
 import einops
 import torchvision

misc/hephaestus.slurm

@@ -0,0 +1,50 @@
+#!/bin/bash -l
+#SBATCH --job-name=Hephaestus # Job name
+#SBATCH --output=Hephaestus.out # Stdout (%j expands to jobId)
+#SBATCH --error=Hephaestus.err  # Stderr (%j expands to jobId)
+#SBATCH --ntasks=4  # Number of tasks(processes)
+#SBATCH --nodes=2 # Number of nodes requested
+#SBATCH --gres=gpu:2  # GPUs per node -- must be equal to ntasks per node in case of data parallelism. Change for model parallelism
+#SBATCH --ntasks-per-node=2 # Tasks per node
+#SBATCH --cpus-per-task=10  # Threads per task
+#SBATCH --time=24:00:00 # walltime
+#SBATCH --mem=54G # memory per NODE
+#SBATCH --partition=<partition_id> # Partition
+#SBATCH --account=<account_id>  # Replace with your system project
+#SBATCH --wait-all-nodes=1  # Do not begin the execution until all nodes are ready for use
+
+module purge
+
+module load gnu/8
+module load java/12.0.2
+module load cuda/10.1.168
+module load intel/18
+module load intelmpi/2018
+module load tftorch/270-191
+
+echo "Start at `date`"
+echo "SLURM_NTASKS=$SLURM_NTASKS"
+
+if [ x$SLURM_CPUS_PER_TASK == x ]; then
+  export OMP_NUM_THREADS=1
+else
+  export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK
+fi
+
+export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
+echo "MASTER_ADDR=$MASTER_ADDR"
+
+export MASTER_PORT=12802
+echo "MASTER_PORT=$MASTER_PORT"
+
+NODES=($( scontrol show hostname $SLURM_NODELIST | uniq ))
+export NUM_NODES=${#NODES[@]}
+WORKERS=$(printf '%s-ib:'${SLURM_NTASKS_PER_NODE}',' "${NODES[@]}" | sed 's/,$//')
+
+echo "SLULM_NODELIST=$SLULM_NODELIST"
+echo "WORKERS=$WORKERS"
+echo "NODES=$NODES"
+
+srun python3 main.py
+
+echo "End at `date`"