Add cuda backend to inference

map2map/args.py

@@ -173,6 +173,10 @@ def add_test_args(parser):
     parser.add_argument('--test-tgt-patterns', type=str_list, required=True,
             help='comma-sep. list of glob patterns for test target data')
 
+    parser.add_argument('--num-threads', type=int,
+            help='number of CPU threads when cuda is unavailable. '
+            'Default is the number of CPUs on the node by slurm')
+
 
 def str_list(s):
     return s.split(',')

map2map/test.py

@@ -1,4 +1,6 @@
+import os
 import sys
+import warnings
 from pprint import pprint
 import numpy as np
 import torch
@@ -12,6 +14,22 @@ from .utils import import_attr, load_model_state_dict
 
 
 def test(args):
+    if torch.cuda.is_available():
+        if torch.cuda.device_count() > 1:
+            warnings.warn('Not parallelized but given more than 1 GPUs')
+
+        os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+        device = torch.device('cuda', 0)
+
+        torch.backends.cudnn.benchmark = True
+    else:  # CPU multithreading
+        device = torch.device('cpu')
+
+        if args.num_threads is None:
+            args.num_threads = int(os.environ['SLURM_CPUS_ON_NODE'])
+
+        torch.set_num_threads(args.num_threads)
+
     print('pytorch {}'.format(torch.__version__))
     pprint(vars(args))
     sys.stdout.flush()
@@ -41,6 +59,7 @@ def test(args):
         batch_size=args.batch_size,
         shuffle=False,
         num_workers=args.loader_workers,
+        pin_memory=True,
     )
 
     style_size = test_dataset.style_size
@@ -50,10 +69,13 @@ def test(args):
     model = import_attr(args.model, models, callback_at=args.callback_at)
     model = model(style_size, sum(in_chan), sum(out_chan),
                   scale_factor=args.scale_factor, **args.misc_kwargs)
-    criterion = import_attr(args.criterion, torch.nn, callback_at=args.callback_at)
+    model.to(device)
-    criterion = criterion()
+
+    criterion = import_attr(args.criterion, torch.nn, models,
+                            callback_at=args.callback_at)
+    criterion = criterion()
+    criterion.to(device)
 
-    device = torch.device('cpu')
     state = torch.load(args.load_state, map_location=device)
     load_model_state_dict(model, state['model'], strict=args.load_state_strict)
     print('model state at epoch {} loaded from {}'.format(
@@ -66,8 +88,21 @@ def test(args):
         for i, data in enumerate(test_loader):
             style, input, target = data['style'], data['input'], data['target']
 
+            style = style.to(device, non_blocking=True)
+            input = input.to(device, non_blocking=True)
+            target = target.to(device, non_blocking=True)
+
             output = model(input, style)
+            if i < 5:
+                print('##### sample :', i)
+                print('style shape :', style.shape)
+                print('input shape :', input.shape)
+                print('output shape :', output.shape)
+                print('target shape :', target.shape)
+
             input, output, target = narrow_cast(input, output, target)
+            if i < 5:
+                print('narrowed shape :', output.shape, flush=True)
 
             loss = criterion(output, target)
 

scripts/example-test.slurm

@@ -2,12 +2,14 @@
 
 #SBATCH --job-name=R2D2
 #SBATCH --output=%x-%j.out
-
 #SBATCH --partition=cpu_partition
-
+#SBATCH --nodes=1
 #SBATCH --exclusive
-#SBATCH --nodes=2
+##SBATCH --partition=gpu_partition
-#SBATCH --time=1-00:00:00
+##SBATCH --gres=gpu:1
+##SBATCH --ntasks=1
+##SBATCH --cpus-per-task=8
+#SBATCH --time=0-01:00:00
 
 
 hostname; pwd; date
@@ -22,9 +24,6 @@ hostname; pwd; date
 #conda info
 
 
-export OMP_NUM_THREADS=$SLURM_CPUS_ON_NODE  # use MKL-DNN
-
-
 m2m.py test \
     --test-in-patterns "test/R0-*.npy,test/R1-*.npy" \
     --test-tgt-patterns "test/D0-*.npy,test/D1-*.npy" \