diff --git a/python/paddle/v2/fluid/clip.py b/python/paddle/v2/fluid/clip.py
index 7a36df0dabbca74484f607ec6854a92e613fd951..d4f025a4af60d3365b0e85e3024a2dd94f5ae842 100644
--- a/python/paddle/v2/fluid/clip.py
+++ b/python/paddle/v2/fluid/clip.py
@@ -138,8 +138,7 @@ class GradientClipByGlobalNorm(BaseGradientClipAttr):
         cls = self.__class__
         cls.check_init()
 
-        local_norm_var = layers.reduce_sum(
-            x=layers.pow(x=grad, factor=2), reduce_all=True)
+        local_norm_var = layers.reduce_sum(input=layers.pow(x=grad, factor=2.0))
         layers.sums(
             input=[local_norm_var, cls.global_norm_var],
             out=[cls.global_norm_var])
@@ -154,6 +153,8 @@ class GradientClipByGlobalNorm(BaseGradientClipAttr):
                 x=cls.clip_norm_var,
                 y=layers.elementwise_max(
                     x=cls.clip_norm_var, y=cls.global_norm_var))
+            assert cls.scale_var.shape == (1L, )
+
         new_grad = layers.elementwise_mul(x=grad, y=cls.scale_var)
         return param, new_grad
 
diff --git a/python/paddle/v2/fluid/layers/ops.py b/python/paddle/v2/fluid/layers/ops.py
index dd3197fc0029bfc5ede6201b836edbcf57541206..a2055c5d7b844b492addbda220973d0c6c1a54e0 100644
--- a/python/paddle/v2/fluid/layers/ops.py
+++ b/python/paddle/v2/fluid/layers/ops.py
@@ -48,7 +48,7 @@ __all__ = [
     'mean', 'mul', 'reshape', 'scale', 'transpose',
     'sigmoid_cross_entropy_with_logits', 'elementwise_add', 'elementwise_div',
     'elementwise_sub', 'elementwise_mul', 'elementwise_max', 'elementwise_min',
-    'clip', 'clip_by_norm', 'sequence_softmax', 'reduce_sum'
+    'clip', 'clip_by_norm', 'sequence_softmax'
 ] + __activations__
 
 for _OP in set(__all__):