fix argparse error in yolo benchmarking (#111)

* fix argparse error in yolo benchmarking

* update cuda sync condition

* update installation instructions, yolo integration

Author: bfineran

integrations/ultralytics/README.md

@@ -40,8 +40,8 @@ git clone https://github.com/neuralmagic/sparseml.git
 # copy script
 cd yolov5
 git checkout c9bda11  # latest tested integration commit hash
-cp ../sparseml/integrations/ultralytics/*.py yolov5
-cp ../sparseml/integrations/ultralytics/deepsparse/*.py
+cp ../sparseml/integrations/ultralytics/*.py .
+cp ../sparseml/integrations/ultralytics/deepsparse/*.py .
 
 # install dependencies
 pip install -r requirements.txt
@@ -139,4 +139,4 @@ Note for quantized performance, your CPU must support VNNI instructions.
 
 ### Server
 `sparseml/integrations/ultralytics/deepsparse/SERVER.md` contains relevant
-documentation for running the server and client examples.
+documentation for running the server and client examples.

integrations/ultralytics/deepsparse/benchmark.py

@@ -231,6 +231,8 @@ def parse_args():
     if args.engine == TORCH_ENGINE and args.device is None:
         args.device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
+    return args
+
 
 def _parse_device(device: Union[str, int]) -> Union[str, int]:
     try:
@@ -390,7 +392,7 @@ def benchmark_yolo(args):
     progress_bar = tqdm(total=args.num_iterations)
 
     for iteration, batch in enumerate(data_loader):
-        if args.device != "cpu":
+        if args.device not in ["cpu", None]:
             torch.cuda.synchronize()
         iter_start = time.time()
 
@@ -407,7 +409,7 @@ def benchmark_yolo(args):
         # NMS
         outputs = postprocess_nms(outputs)
 
-        if args.device != "cpu":
+        if args.device not in ["cpu", None]:
             torch.cuda.synchronize()
         iter_end = time.time()