adding baseline code

imagenet/image_classification/dataloaders.py

@@ -75,7 +75,7 @@ def load_jpeg_from_file(path, cuda=True):
 
     return input
 
-
+"""
 class HybridTrainPipe(Pipeline):
     def __init__(
         self,
@@ -220,7 +220,7 @@ class DALIWrapper(object):
         return DALIWrapper.gen_wrapper(
             self.dalipipeline, self.num_classes, self.one_hot, self.memory_format
         )
-
+"""
 
 def get_dali_train_loader(dali_cpu=False):
     def gdtl(
@@ -444,7 +444,7 @@ def get_pytorch_train_loader(
         pin_memory=True,
         collate_fn=partial(fast_collate, memory_format),
         drop_last=True,
-        persistent_workers=False, # see https://github.com/pytorch/pytorch/issues/48370
+        persistent_workers=False,
     )
 
     return (

imagenet/image_classification/training.py

@@ -99,7 +99,7 @@ class ModelAndLoss(nn.Module):
         if self.noDDP:
             self.model = self.model
         else:
-            self.model = DDP(self.model, device_ids=[gpu_id], output_device=gpu_id, bucket_cap_mb=1)
+            self.model = DDP(self.model, device_ids=[gpu_id])
 
     def load_model_state(self, state):
         if not state is None:

imagenet/main.py

@@ -106,7 +106,7 @@ def add_parser_arguments(parser, skip_arch=False):
     parser.add_argument(
         "-j",
         "--workers",
-        default=5,
+        default=1,
         type=int,
         metavar="N",
         help="number of data loading workers (default: 5)",

summit/sub.lsf

@@ -12,16 +12,15 @@
 
 nnodes=$(cat ${LSB_DJOB_HOSTFILE} | sort | uniq | grep -v login | grep -v batch | wc -l)
 DATA_DIR=/gpfs/alpine/world-shared/stf011/junqi/choco_env/dl_code/data/ILSVRC
-LOG_DIR=/gpfs/alpine/world-shared/stf011/atsaris/tmp/imagenet
+LOG_DIR=summit_logs/
 
-source /gpfs/alpine/world-shared/stf011/atsaris/summit_env/monai/setup.sh
+source /gpfs/alpine/stf011/world-shared/atsaris/summit_env/pytorch_1_9_new/activate.sh
 
-jsrun --smpiargs="-disable_gpu_hooks" -n${nnodes} -a6 -c42 -g6 -r1 --bind=proportional-packed:7 --launch_distribution=packed \
+jsrun --smpiargs="-disable_gpu_hooks" -n${nnodes} -a1 -c42 -g1 -r1 --bind=proportional-packed:7 --launch_distribution=packed \
     bash -c "\
     source export_DDP_envvars.sh && \
-    python -u main.py \
+    python -u ../imagenet/main.py \
     --arch resnet50 \
-    -j 28 \
     -p 10 \
     -b 128\
     --training-only \
@@ -29,6 +28,6 @@ jsrun --smpiargs="-disable_gpu_hooks" -n${nnodes} -a6 -c42 -g6 -r1 --bind=propor
     --epochs 1 \
     --prof 100 \
     --no-checkpoints \
-    --data-backend syntetic \
+    --data-backend pytorch \
     ${DATA_DIR}
     "