diff --git a/python/paddle/fluid/incubate/fleet/base/fleet_base.py b/python/paddle/fluid/incubate/fleet/base/fleet_base.py
index 61b7fe7de50625fabbf1848fec2e96072d8d417e..c53090a52bde2a351d5659095ce4ca577a7c09db 100644
--- a/python/paddle/fluid/incubate/fleet/base/fleet_base.py
+++ b/python/paddle/fluid/incubate/fleet/base/fleet_base.py
@@ -30,7 +30,7 @@ __all__ = ['Fleet', 'DistributedOptimizer']
 __all__ += mode.__all__
 
 
-class Fleet:
+class Fleet(metaclass=abc.ABCMeta):
     """
     Fleet is the base class, transpiler and pslib are implementation of Fleet.
 
@@ -41,8 +41,6 @@ class Fleet:
         None
     """
 
-    __metaclass__ = abc.ABCMeta
-
     def __init__(self, mode):
         self._is_initialized = False
         self._mode = mode
@@ -268,7 +266,7 @@ class Fleet:
         pass
 
 
-class DistributedOptimizer:
+class DistributedOptimizer(metaclass=abc.ABCMeta):
     """
     DistributedOptimizer is a wrapper for paddle.fluid.optimizer
     A user should pass a paddle.fluid.optimizer to DistributedOptimizer
@@ -287,8 +285,6 @@ class DistributedOptimizer:
 
     """
 
-    __metaclass__ = abc.ABCMeta
-
     def __init__(self, optimizer, strategy=None):
         if (
             not isinstance(optimizer, SGD.__bases__)