refine batch norm python layer (#7348)

Author: QiJune

python/paddle/v2/fluid/layer_helper.py

@@ -120,11 +120,12 @@ def get_parameter(self, name):
             raise ValueError("no Parameter name %s found" % name)
         return param
 
-    def create_tmp_variable(self, dtype):
+    def create_tmp_variable(self, dtype, stop_gradient=False):
         return self.main_program.current_block().create_var(
             name=unique_name(".".join([self.name, 'tmp'])),
             dtype=dtype,
-            persistable=False)
+            persistable=False,
+            stop_gradient=stop_gradient)
 
     def create_variable(self, *args, **kwargs):
         return self.main_program.current_block().create_var(*args, **kwargs)

python/paddle/v2/fluid/layers/nn.py

@@ -971,20 +971,26 @@ def batch_norm(input,
         attr=helper.param_attr, shape=param_shape, dtype=dtype, is_bias=True)
 
     mean = helper.create_global_variable(
-        dtype=input.dtype, shape=param_shape, persistable=True)
+        dtype=input.dtype,
+        shape=param_shape,
+        persistable=True,
+        stop_gradient=True)
     helper.set_variable_initializer(var=mean, initializer=Constant(0.0))
 
     variance = helper.create_global_variable(
-        dtype=input.dtype, shape=param_shape, persistable=True)
+        dtype=input.dtype,
+        shape=param_shape,
+        persistable=True,
+        stop_gradient=True)
     helper.set_variable_initializer(var=variance, initializer=Constant(1.0))
 
     # create output
     # mean and mean_out share the same memory
     mean_out = mean
     # variance and variance out share the same memory
     variance_out = variance
-    saved_mean = helper.create_tmp_variable(dtype)
-    saved_variance = helper.create_tmp_variable(dtype)
+    saved_mean = helper.create_tmp_variable(dtype=dtype, stop_gradient=True)
+    saved_variance = helper.create_tmp_variable(dtype=dtype, stop_gradient=True)
 
     batch_norm_out = helper.create_tmp_variable(dtype)