Prepare the retraining for running on Aurora

src/retrain.py

@@ -141,10 +141,10 @@ def main() :
     [img_folder_train, img_folder_eval, _] = torch.utils.data.dataset.random_split(img_folder, [n_img_train, n_img_eval, n_img_rest])
 
     # load the training set in random order
-    data_loader_train = torch.utils.data.DataLoader(img_folder_train, batch_size=8,
+    data_loader_train = torch.utils.data.DataLoader(img_folder_train, batch_size=16, num_workers=20,
                                               shuffle=True)
 
-    data_loader_eval = torch.utils.data.DataLoader(img_folder_eval, batch_size=8,
+    data_loader_eval = torch.utils.data.DataLoader(img_folder_eval, batch_size=16, num_workers=20,
                                               shuffle=True)
 
     dataset_sizes = {'train' : n_img_train, 'val' : n_img_eval}

submit_20.sh

+#!/bin/bash
+
+#SBATCH -n 20
+#SBATCH -N 1 # force all cores on one node
+#SBATCH -t 04:00:00
+#SBATCH -J retrain
+#SBATCH -o retrain_%j.out
+#SBATCH -e retrain_%j.err
+
+#SBATCH -A computehpc
+
+cat $0 # put the script in the output file
+
+module purge
+
+module load  GCC/7.3.0-2.30  CUDA/9.2.88  OpenMPI/3.1.1
+module load Python/3.6.6
+
+module list
+
+ARGS=$@
+
+# bind threads to cores
+export OMP_SCHEDULE=static
+export OMP_PLACES=cores
+export OMP_PROC_BIND=CLOSE
+
+source venv/bin/activate
+
+cd src
+python retrain.py 25 1