diff --git a/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-1p.py b/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-1p.py
index 9c4316c2b8971bc4218a298a5a981befa27d7fbe..558434aaf2103704eb144c790a3b4e35423a6cd0 100755
--- a/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-1p.py
+++ b/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-1p.py
@@ -51,6 +51,8 @@ from timm.scheduler import create_scheduler
 from timm.utils import ApexScaler, NativeScaler
                                                   
 #from ghostnet.ghostnet_pytorch.ghostnet import ghostnet
+if torch.version >= '1.8.1':
+import torch_npu
 import torch.npu
 import apex
 
@@ -273,7 +275,7 @@ parser.add_argument('--distributed',action='store_true',
                          'fastest way to use PyTorch for either single node or '
                          'multi node data parallel training')   
 parser.add_argument("--world_size", default=1, type=int)
-parser.add_argument("--loss-scale", default=1024, type=int)
+parser.add_argument("--loss-scale", default='dynamic', type=str)
 parser.add_argument('--reprob', type=float, default=0., metavar='PCT',
                     help='Random erase prob (default: 0.)')
 parser.add_argument('--remode', type=str, default='const',
diff --git a/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-8p.py b/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-8p.py
index 608f3795f4ca5aefb69b44c28b810a7e56d5ded5..591bc1ee488c6527615cb051d385dadeb39b7995 100755
--- a/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-8p.py
+++ b/PyTorch/contrib/cv/classification/csp_resnext50-mish/train-8p.py
@@ -51,6 +51,8 @@ from timm.scheduler import create_scheduler
 from timm.utils import ApexScaler, NativeScaler
                                                   
 #from ghostnet.ghostnet_pytorch.ghostnet import ghostnet
+if torch.version >= '1.8.1':
+import torch_npu
 import torch.npu
 import apex
 
@@ -273,7 +275,7 @@ parser.add_argument('--distributed',action='store_true',
                          'fastest way to use PyTorch for either single node or '
                          'multi node data parallel training')   
 parser.add_argument("--world_size", default=1, type=int)
-parser.add_argument("--loss-scale", default=1024, type=int)
+parser.add_argument("--loss-scale", default='dynamic', type=str)
 parser.add_argument('--reprob', type=float, default=0., metavar='PCT',
                     help='Random erase prob (default: 0.)')
 parser.add_argument('--remode', type=str, default='const',