[examples] upgrade code for seed setting (#3387)

* replace set_seed

* update import

examples/cv_example.py

@@ -24,6 +24,7 @@
 from torchvision.transforms import Compose, RandomResizedCrop, Resize, ToTensor
 
 from accelerate import Accelerator
+from accelerate.utils import set_seed
 
 
 ########################################################################
@@ -93,10 +94,7 @@ def training_function(config, args):
     label_to_id = {lbl: i for i, lbl in enumerate(id_to_label)}
 
     # Set the seed before splitting the data.
-    np.random.seed(seed)
-    torch.manual_seed(seed)
-    torch.cuda.manual_seed_all(seed)
-
+    set_seed(seed)
     # Split our filenames between train and validation
     random_perm = np.random.permutation(len(file_names))
     cut = int(0.8 * len(file_names))