diff --git a/mindspore/nn/optim/optimizer.py b/mindspore/nn/optim/optimizer.py
index e2b0cddb71..cd0ed93a10 100755
--- a/mindspore/nn/optim/optimizer.py
+++ b/mindspore/nn/optim/optimizer.py
@@ -84,7 +84,7 @@ apply_decay = C.MultitypeFuncGraph("apply_decay")
 def _tensor_apply_decay(weight_decay, if_apply, weight, gradient):
     """Get grad with weight_decay."""
     if if_apply:
-        return op_add((gradient, weight * F.scalar_to_array(weight_decay)))
+        return op_add((gradient, weight * weight_decay))
     return gradient
 
 
diff --git a/mindspore/nn/wrap/loss_scale.py b/mindspore/nn/wrap/loss_scale.py
index f7c686f535..a11c753eda 100644
--- a/mindspore/nn/wrap/loss_scale.py
+++ b/mindspore/nn/wrap/loss_scale.py
@@ -32,7 +32,7 @@ reciprocal = P.Reciprocal()
 
 @_grad_scale.register("Tensor", "Tensor")
 def tensor_grad_scale(scale, grad):
-    return grad * reciprocal(scale)
+    return grad * F.cast(reciprocal(scale), F.dtype(grad))
 
 
 class DynamicLossScaleUpdateCell(Cell):