AddBiasOp does not care num_flatten_dims (#5200)

* AddBiasOp does not care num_flatten_dims

* Add comments

Author: reyoung

python/paddle/v2/framework/layer_helper.py

@@ -142,8 +142,24 @@ def create_global_variable(self, *args, **kwargs):
         return self.program.global_block().create_var(
             *args, persistable=False, **kwargs)
 
-    def append_bias_op(self, input_var):
-        size = list(input_var.shape[1:])
+    def append_bias_op(self, input_var, num_flatten_dims=None):
+        """
+        Append bias operator and return its output. If the user does not set 
+        bias_attr, append_bias_op will return input_var
+         
+        :param input_var: the input variable. The len(input_var.shape) is larger
+        or equal than 2.
+        :param num_flatten_dims: The input tensor will be flatten as a matrix 
+        when adding bias.
+        `matrix.shape = product(input_var.shape[0:num_flatten_dims]), product(
+                input_var.shape[num_flatten_dims:])`
+        """
+        if num_flatten_dims is None:
+            num_flatten_dims = self.kwargs.get('num_flatten_dims', None)
+            if num_flatten_dims is None:
+                num_flatten_dims = 1
+
+        size = list(input_var.shape[num_flatten_dims:])
         bias_attr = self.bias_attr()
         if not bias_attr:
             return input_var