diff --git a/python/paddle/distributed/__init__.py b/python/paddle/distributed/__init__.py
index d66577102c713ae796abdf812f1cf2bb767e5b14..b7357eef7ad9a3abae7f9c1c09fdc00b409061ad 100644
--- a/python/paddle/distributed/__init__.py
+++ b/python/paddle/distributed/__init__.py
@@ -30,8 +30,11 @@ __all__ = ["spawn"]
 
 # dygraph parallel apis
 __all__ += [
-    "init_parallel_env", "get_rank", "get_world_size", "prepare_context",
-    "ParallelEnv"
+    "init_parallel_env",
+    "get_rank",
+    "get_world_size",
+    "prepare_context",
+    "ParallelEnv",
 ]
 
 # collective apis
diff --git a/python/paddle/distributed/fleet/__init__.py b/python/paddle/distributed/fleet/__init__.py
index b080fb17553d4a93a545f4ae781d786d82e26576..42ac68ba1a64de54f029878ceab08435c924d087 100644
--- a/python/paddle/distributed/fleet/__init__.py
+++ b/python/paddle/distributed/fleet/__init__.py
@@ -18,16 +18,15 @@ from .base.distributed_strategy import DistributedStrategy
 from .base.fleet_base import Fleet
 from .base.util_factory import UtilBase
 from .dataset import *
+#from . import metrics
 
 __all__ = [
     "DistributedStrategy",
     "UtilBase",
     "DatasetFactory",
-    "DatasetBase",
-    "InMemoryDataset",
-    "QueueDataset",
     "UserDefinedRoleMaker",
     "PaddleCloudRoleMaker",
+    "Fleet",
 ]
 
 fleet = Fleet()
diff --git a/python/paddle/distributed/fleet/base/distributed_strategy.py b/python/paddle/distributed/fleet/base/distributed_strategy.py
index 5e527ea03ab9c816948f343ac103672a751fdbc3..26063d1b8a9225aff63628bb37f433ec95257dc7 100755
--- a/python/paddle/distributed/fleet/base/distributed_strategy.py
+++ b/python/paddle/distributed/fleet/base/distributed_strategy.py
@@ -17,6 +17,8 @@ from paddle.distributed.fleet.proto import distributed_strategy_pb2
 from paddle.fluid.framework import Variable, set_flags, core
 import google.protobuf.text_format
 
+__all__ = ["DistributedStrategy"]
+
 
 def get_msg_dict(msg):
     res_dict = {}
diff --git a/python/paddle/distributed/fleet/base/fleet_base.py b/python/paddle/distributed/fleet/base/fleet_base.py
index a6286bcca87fad1afddbd8af1e56dad05dab2578..eb2cb19eaecf01f7209e7974e3c47c26810ce21f 100644
--- a/python/paddle/distributed/fleet/base/fleet_base.py
+++ b/python/paddle/distributed/fleet/base/fleet_base.py
@@ -22,7 +22,7 @@ from .runtime_factory import RuntimeFactory
 from .util_factory import UtilFactory
 from paddle.fluid.wrapped_decorator import wrap_decorator
 
-__all__ = ['Fleet']
+#__all__ = ['Fleet']
 
 
 def _inited_runtime_handler_(func):
diff --git a/python/paddle/distributed/fleet/base/meta_optimizer_factory.py b/python/paddle/distributed/fleet/base/meta_optimizer_factory.py
index 459070fcc4dbef3711c33b2932e8f1c88647aab5..f845b3fcd8953c44c8b5b857dac08be1c7269958 100755
--- a/python/paddle/distributed/fleet/base/meta_optimizer_factory.py
+++ b/python/paddle/distributed/fleet/base/meta_optimizer_factory.py
@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__all__ = ["MetaOptimizerFactory"]
-
 from ..meta_optimizers import *
 
 meta_optimizer_names = list(
diff --git a/python/paddle/distributed/fleet/base/role_maker.py b/python/paddle/distributed/fleet/base/role_maker.py
index 6aeeb4a2896ea1d20390e463937aa07d3edd0204..3d159a631226f4e52c9cc6ab6d36275265a25f97 100644
--- a/python/paddle/distributed/fleet/base/role_maker.py
+++ b/python/paddle/distributed/fleet/base/role_maker.py
@@ -17,7 +17,7 @@ import numpy as np
 from multiprocessing import Process, Manager
 import paddle.fluid as fluid
 
-__all__ = ['RoleMakerBase', 'UserDefinedRoleMaker', 'PaddleCloudRoleMaker']
+#__all__ = ['UserDefinedRoleMaker', 'PaddleCloudRoleMaker']
 
 
 class Role:
diff --git a/python/paddle/distributed/fleet/meta_optimizers/__init__.py b/python/paddle/distributed/fleet/meta_optimizers/__init__.py
index 075e8b6c4302d792606849fc2981e46ccead1e56..78b2b8117b9f6eecfcb410f3cca0b1ec3e99a93b 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/__init__.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/__init__.py
@@ -22,17 +22,3 @@ from .lars_optimizer import LarsOptimizer
 from .async_graph_execution_optimizer import AsyncGraphExecutionOptimizer
 from .dgc_optimizer import DGCOptimizer
 from .lamb_optimizer import LambOptimizer
-
-__all__ = [
-    'AMPOptimizer',
-    'RecomputeOptimizer',
-    'GradientMergeOptimizer',
-    'AsyncMetaOptimizer',
-    'GraphExecutionOptimizer',
-    'PipelineOptimizer',
-    'LocalSGDOptimizer',
-    'LarsOptimizer',
-    'AsyncGraphExecutionOptimizer',
-    'DGCOptimizer',
-    'LambOptimizer',
-]
diff --git a/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py
index 66db14209b4c57475c30c6dde083593e27f04ea0..b1952276e44cd1466bc443440505462924115ab7 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py
@@ -14,8 +14,6 @@
 import paddle.fluid.contrib.mixed_precision as mixed_precision
 from .meta_optimizer_base import MetaOptimizerBase
 
-__all__ = ["AMPOptimizer"]
-
 
 class AMPOptimizer(MetaOptimizerBase):
     def __init__(self, optimizer):
diff --git a/python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py
index f34786f9dc309dd1f03319368bbc93ef1bfc03e3..f1c6defc5c982c7d56980642898aaa333c199bbe 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py
@@ -15,8 +15,6 @@ from paddle.fluid.optimizer import Momentum, DGCMomentumOptimizer
 from .meta_optimizer_base import MetaOptimizerBase
 import logging
 
-__all__ = ["DGCOptimizer"]
-
 
 class DGCOptimizer(MetaOptimizerBase):
     def __init__(self, optimizer):
diff --git a/python/paddle/distributed/fleet/meta_optimizers/gradient_merge_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/gradient_merge_optimizer.py
index bd52179a35862241768ad5bd01eedf16732ad3b6..7db79ad7b5b7081172209faa2396d9f2a31bbdb3 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/gradient_merge_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/gradient_merge_optimizer.py
@@ -14,10 +14,6 @@
 from paddle.fluid.optimizer import GradientMergeOptimizer as GM
 from .meta_optimizer_base import MetaOptimizerBase
 
-__all__ = ["GradientMergeOptimizer"]
-
-# amp + gradient merge + lamb
-
 
 class GradientMergeOptimizer(MetaOptimizerBase):
     def __init__(self, optimizer):
diff --git a/python/paddle/distributed/fleet/meta_optimizers/lamb_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/lamb_optimizer.py
index 7e08a02eb1dc2e14b1871fe7743bbee8ade3feb3..9fa29c4078e9f579a740ef8c0591979e7fbb962d 100755
--- a/python/paddle/distributed/fleet/meta_optimizers/lamb_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/lamb_optimizer.py
@@ -16,8 +16,6 @@ from paddle.fluid.optimizer import LambOptimizer as LAMB
 from .meta_optimizer_base import MetaOptimizerBase
 import logging
 
-__all__ = ["LambOptimizer"]
-
 
 class LambOptimizer(MetaOptimizerBase):
     def __init__(self, optimizer):
diff --git a/python/paddle/distributed/fleet/meta_optimizers/lars_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/lars_optimizer.py
index 09c418fa79106d05cffae1e8bc18fac9c0cc8f34..a7b856ff5b0dcb1ab30de82a12c91a2e1c14fe76 100755
--- a/python/paddle/distributed/fleet/meta_optimizers/lars_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/lars_optimizer.py
@@ -15,8 +15,6 @@ from paddle.fluid.optimizer import Momentum, LarsMomentumOptimizer
 from .meta_optimizer_base import MetaOptimizerBase
 import logging
 
-__all__ = ["LarsOptimizer"]
-
 
 class LarsOptimizer(MetaOptimizerBase):
     def __init__(self, optimizer):
diff --git a/python/paddle/distributed/fleet/meta_optimizers/meta_optimizer_base.py b/python/paddle/distributed/fleet/meta_optimizers/meta_optimizer_base.py
index 12a4d904340337bf9a99968c7d82db117bf59ce8..073148e11a0a2b08253b89d36d7a014b830518f8 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/meta_optimizer_base.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/meta_optimizer_base.py
@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__all__ = ["MetaOptimizerBase"]
-
 from paddle.fluid.optimizer import Optimizer
 
 
diff --git a/python/paddle/distributed/fleet/meta_optimizers/pipeline_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/pipeline_optimizer.py
index fe9221307cbacfa1beaf030b70a4e4b9223769cc..d5a45e2b4e1aeda2e1c66c0a5a36236622f093ec 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/pipeline_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/pipeline_optimizer.py
@@ -20,8 +20,6 @@ from paddle.fluid.optimizer import PipelineOptimizer as PO
 from .meta_optimizer_base import MetaOptimizerBase
 from .common import OpRole, OP_ROLE_KEY, OP_ROLE_VAR_KEY, CollectiveHelper, is_update_op, is_loss_grad_op, is_backward_op, is_optimizer_op
 
-__all__ = ["PipelineOptimizer"]
-
 
 class PipelineHelper(CollectiveHelper):
     def __init__(self, role_maker, nrings=1, wait_port='6174'):
diff --git a/python/paddle/distributed/fleet/meta_optimizers/recompute_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/recompute_optimizer.py
index 45130b447125f6ecbade2e4e5e3dad2f127fda52..3eb3ca6127cfe0d0a7a458c6c44e09ce22e7b24a 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/recompute_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/recompute_optimizer.py
@@ -14,8 +14,6 @@
 from paddle.fluid.optimizer import RecomputeOptimizer as RO
 from .meta_optimizer_base import MetaOptimizerBase
 
-__all__ = ["RecomputeOptimizer"]
-
 
 class RecomputeOptimizer(MetaOptimizerBase):
     def __init__(self, optimizer):
diff --git a/python/paddle/distributed/fleet/metrics/__init__.py b/python/paddle/distributed/fleet/metrics/__init__.py
index abf198b97e6e818e1fbe59006f98492640bcee54..bc30c063787d28e5bcb4455b3cbd56372879fe0a 100644
--- a/python/paddle/distributed/fleet/metrics/__init__.py
+++ b/python/paddle/distributed/fleet/metrics/__init__.py
@@ -11,3 +11,16 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+from .metric import *
+
+__all__ = [
+    "sum",
+    "max",
+    "min",
+    "auc",
+    "mae",
+    "rmse",
+    "mse",
+    "acc",
+]
diff --git a/python/paddle/distributed/fleet/runtime/__init__.py b/python/paddle/distributed/fleet/runtime/__init__.py
index a796a73fc981b7edbcd57e8f5858456031e7ae6e..cf718b199e52e422ff8f2b66317f3cd6123c76a1 100644
--- a/python/paddle/distributed/fleet/runtime/__init__.py
+++ b/python/paddle/distributed/fleet/runtime/__init__.py
@@ -14,5 +14,3 @@
 
 from .collective_runtime import CollectiveRuntime
 from .parameter_server_runtime import ParameterServerRuntime
-
-__all__ = ["CollectiveRuntime," "ParameterServerRuntime", ]
diff --git a/python/paddle/distributed/fleet/utils/__init__.py b/python/paddle/distributed/fleet/utils/__init__.py
index 212308159aabb123fde11543b3482f2232b4925d..f1911408c84a9dde56a8674e88e0fb8ad575cae7 100644
--- a/python/paddle/distributed/fleet/utils/__init__.py
+++ b/python/paddle/distributed/fleet/utils/__init__.py
@@ -15,4 +15,4 @@
 from .fs import *
 from .http_server import KVHandler, KVHTTPServer, KVServer
 
-__all__ = ['KVHandler', 'KVHTTPServer', 'KVServer'] + fs.__all__
+#__all__ = ['KVHandler', 'KVHTTPServer', 'KVServer'] + fs.__all__