diff --git a/python/paddle/distributed/fleet/fleet.py b/python/paddle/distributed/fleet/fleet.py
index b0a8c2f69cc96131d424ed56e054030f5b9cf0f5..08f07e34d7298328049204a06c57d66a68e33cf4 100755
--- a/python/paddle/distributed/fleet/fleet.py
+++ b/python/paddle/distributed/fleet/fleet.py
@@ -1367,18 +1367,6 @@ class Fleet:
                     copy_user_defined_strategy,
                 )
                 can_not_apply_optimizer_list.append(meta_optimizer)
-                from .meta_optimizers import ParameterServerGraphOptimizer
-
-                graph_optimizer = ParameterServerGraphOptimizer(
-                    self.user_defined_optimizer
-                )
-                graph_optimizer._set_basic_info(
-                    loss,
-                    self._role_maker,
-                    self.user_defined_optimizer,
-                    copy_user_defined_strategy,
-                )
-                can_not_apply_optimizer_list.append(graph_optimizer)
             else:
                 # compile time
                 distributed_optimizer_list = (
diff --git a/python/paddle/distributed/fleet/meta_optimizers/__init__.py b/python/paddle/distributed/fleet/meta_optimizers/__init__.py
index feb7b125adc1c83cf0c7a5b9719141be74bfa000..1e98b3432f0a24ce9b6131014a8764f6fc74165f 100644
--- a/python/paddle/distributed/fleet/meta_optimizers/__init__.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/__init__.py
@@ -22,7 +22,6 @@ from .pipeline_optimizer import PipelineOptimizer
 from .localsgd_optimizer import LocalSGDOptimizer
 from .localsgd_optimizer import AdaptiveLocalSGDOptimizer
 from .lars_optimizer import LarsOptimizer
-from .parameter_server_graph_optimizer import ParameterServerGraphOptimizer
 from .dgc_optimizer import DGCOptimizer
 from .dgc_optimizer import DGCMomentumOptimizer
 from .lamb_optimizer import LambOptimizer
diff --git a/python/paddle/distributed/fleet/meta_optimizers/parameter_server_graph_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/parameter_server_graph_optimizer.py
deleted file mode 100644
index e6bf8b85c202117409873ef30251c0718b522e0e..0000000000000000000000000000000000000000
--- a/python/paddle/distributed/fleet/meta_optimizers/parameter_server_graph_optimizer.py
+++ /dev/null
@@ -1,78 +0,0 @@
-#   Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-
-import paddle
-
-from .parameter_server_optimizer import ParameterServerOptimizer
-
-__all__ = []
-
-
-class ParameterServerGraphOptimizer(ParameterServerOptimizer):
-    def __init__(self, optimizer):
-        super().__init__(optimizer)
-        self.inner_opt = optimizer
-        # we do not allow meta optimizer to be inner optimizer currently
-        self.meta_optimizers_white_list = []
-
-    def _can_apply(self):
-        if self.role_maker._is_collective:
-            return False
-
-        k_steps = self.user_defined_strategy.a_sync_configs["k_steps"]
-        if k_steps < 0:
-            return False
-
-        if self.role_maker._is_server():
-            return False
-
-        if self.role_maker._is_heter_parameter_server_mode:
-            return False
-
-        return True
-
-    def _disable_strategy(self, dist_strategy):
-        return
-
-    def _enable_strategy(self, dist_strategy, context):
-        # only open up the async mode for auto-parallel
-        return
-
-    def _is_graph_out(self):
-        return True
-
-    def _try_to_compile(self, main_program, loss):
-        dist_strategy = self._get_distributed_strategy()
-
-        build_strategy = dist_strategy.get_build_strategy()
-        exec_strategy = dist_strategy.get_execute_strategy()
-
-        self._compiled_program = paddle.static.CompiledProgram(main_program)
-
-        self._compiled_program.with_data_parallel(
-            loss_name=loss.name,
-            build_strategy=build_strategy,
-            exec_strategy=exec_strategy,
-            share_vars_from=None,
-        )
-
-        return self._compiled_program
-
-    def minimize(
-        self, loss, startup_program=None, parameter_list=None, no_grad_set=None
-    ):
-        program = loss.block.program
-        compiled_program = self._try_to_compile(program, loss)
-        program._graph = compiled_program
-        # just return self.optimizer_ops and self.param_grads
-        return None, None