distilbert: fix creation of sinusoidal embeddings when using PyTorch 1.8+ (#9917)

src/transformers/models/distilbert/modeling_distilbert.py

@@ -73,10 +73,10 @@ DISTILBERT_PRETRAINED_MODEL_ARCHIVE_LIST = [
 
 def create_sinusoidal_embeddings(n_pos, dim, out):
     position_enc = np.array([[pos / np.power(10000, 2 * (j // 2) / dim) for j in range(dim)] for pos in range(n_pos)])
+    out.requires_grad = False
     out[:, 0::2] = torch.FloatTensor(np.sin(position_enc[:, 0::2]))
     out[:, 1::2] = torch.FloatTensor(np.cos(position_enc[:, 1::2]))
     out.detach_()
-    out.requires_grad = False
 
 
 class Embeddings(nn.Module):