Fix require_grad typo

Fix require_grad typos (should be requires_grad).
Before the fix, the code doesn't cause any errors but doesn't do what it's supposed to do.

Also see pytorch/benchmark#1771

bert_pytorch/model/embedding/position.py

@@ -10,7 +10,7 @@ def __init__(self, d_model, max_len=512):
 
         # Compute the positional encodings once in log space.
         pe = torch.zeros(max_len, d_model).float()
-        pe.require_grad = False
+        pe.requires_grad = False
 
         position = torch.arange(0, max_len).float().unsqueeze(1)
         div_term = (torch.arange(0, d_model, 2).float() * -(math.log(10000.0) / d_model)).exp()