diff --git a/paddle/fluid/framework/distributed_strategy.proto b/paddle/fluid/framework/distributed_strategy.proto
index 96ddc82b1c95ab3480b8badaa274118e40d2beda..d17e68276cd1ce576029cf306a18469aef2ffdb0 100755
--- a/paddle/fluid/framework/distributed_strategy.proto
+++ b/paddle/fluid/framework/distributed_strategy.proto
@@ -31,6 +31,9 @@ message AMPConfig {
   optional float incr_ratio = 4 [ default = 2.0 ];
   optional float decr_ratio = 5 [ default = 0.8 ];
   optional bool use_dynamic_loss_scaling = 6 [ default = true ];
+  repeated string custom_white_list = 7;
+  repeated string custom_black_list = 8;
+  repeated string custom_black_varnames = 9;
 }
 
 message LocalSGDConfig { optional int32 k_steps = 1 [ default = 4 ]; }
diff --git a/python/paddle/fleet/base/meta_optimizer_factory.py b/python/paddle/fleet/base/meta_optimizer_factory.py
index bbbd5fcacd66757f1a2932d33b86e6ea42fe8793..89ebb0ec601e249c58fd43995df1530f44940af4 100755
--- a/python/paddle/fleet/base/meta_optimizer_factory.py
+++ b/python/paddle/fleet/base/meta_optimizer_factory.py
@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from ..meta_optimizers import AMPOptimizer
 from ..meta_optimizers import RecomputeOptimizer
 from ..meta_optimizers import GradientMergeOptimizer
 from ..meta_optimizers import GraphExecutionOptimizer
@@ -22,6 +23,7 @@ from ..meta_optimizers import LarsOptimizer
 __all__ = ["MetaOptimizerFactory"]
 
 meta_optimizer_names = [
+    "AMPOptimizer",
     "RecomputeOptimizer",
     "GradientMergeOptimizer",
     "GraphExecutionOptimizer",
diff --git a/python/paddle/fleet/meta_optimizers/__init__.py b/python/paddle/fleet/meta_optimizers/__init__.py
index 0beb06eacf8cf066c244b24392c697e2dab57948..aa6708e758a78cf2cb10f8ebda81d50ac796b548 100755
--- a/python/paddle/fleet/meta_optimizers/__init__.py
+++ b/python/paddle/fleet/meta_optimizers/__init__.py
@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 
+from .amp_optimizer import AMPOptimizer
 from .recompute_optimizer import RecomputeOptimizer
 from .gradient_merge_optimizer import GradientMergeOptimizer
 from .graph_execution_optimizer import GraphExecutionOptimizer
@@ -19,6 +20,7 @@ from .localsgd_optimizer import LocalSGDOptimizer
 from .lars_optimizer import LarsOptimizer
 
 __all__ = [
+    'AMPOptimizer',
     'RecomputeOptimizer',
     'GradientMergeOptimizer',
     'PipelineOptimizer',
diff --git a/python/paddle/fleet/meta_optimizers/amp_optimizer.py b/python/paddle/fleet/meta_optimizers/amp_optimizer.py
new file mode 100644
index 0000000000000000000000000000000000000000..8316d807fa87062a8e3fba0bcb3bd057d2231032
--- /dev/null
+++ b/python/paddle/fleet/meta_optimizers/amp_optimizer.py
@@ -0,0 +1,63 @@
+#   Copyright (c) 2020 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.fluid.contrib.mixed_precision as mixed_precision
+from .meta_optimizer_base import MetaOptimizerBase
+
+__all__ = ["AMPOptimizer"]
+
+
+class AMPOptimizer(MetaOptimizerBase):
+    def __init__(self, optimizer):
+        super(AMPOptimizer, self).__init__(optimizer)
+        self.inner_opt = optimizer
+        self.amp_opt = None
+        # we do not allow meta optimizer to be inner optimizer currently
+        self.meta_optimizers_white_list = []
+
+    def _set_basic_info(self, loss, role_maker, user_defined_optimizer,
+                        user_defined_strategy):
+        super(AMPOptimizer, self)._set_basic_info(
+            loss, role_maker, user_defined_optimizer, user_defined_strategy)
+
+    def _can_apply(self):
+        if self.user_defined_strategy.amp:
+            return True
+        return False
+
+    def _disable_strategy(self, dist_strategy):
+        dist_strategy.amp = False
+
+    def minimize_impl(self,
+                      loss,
+                      startup_program=None,
+                      parameter_list=None,
+                      no_grad_set=None):
+        if self.amp_opt is None:
+            config = self.user_defined_strategy.amp_configs
+            custom_white_list = set(config['custom_white_list'])
+            custom_black_list = set(config['custom_black_list'])
+            custom_black_varnames = set(config['custom_black_varnames'])
+            amp_lists = mixed_precision.AutoMixedPrecisionLists(
+                custom_white_list, custom_black_list, custom_black_varnames)
+
+            self.amp_opt = mixed_precision.decorate(
+                self.inner_opt, amp_lists, config['init_loss_scaling'],
+                config['incr_every_n_steps'], config['decr_every_n_nan_or_inf'],
+                config['incr_ratio'], config['decr_ratio'],
+                config['use_dynamic_loss_scaling'])
+
+        optimize_ops, params_grads = \
+            self.amp_opt.minimize(loss, startup_program,
+                                  parameter_list, no_grad_set)
+        return optimize_ops, params_grads
diff --git a/python/paddle/fluid/tests/unittests/CMakeLists.txt b/python/paddle/fluid/tests/unittests/CMakeLists.txt
index db3dc6b859495a4c060a5d4fe6e6eebd90c19b86..971a94f549fcff467bbe57256946016e40cff6bf 100755
--- a/python/paddle/fluid/tests/unittests/CMakeLists.txt
+++ b/python/paddle/fluid/tests/unittests/CMakeLists.txt
@@ -34,6 +34,7 @@ list(APPEND MIXED_DIST_TEST_OPS test_fleet_base)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_recompute_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_graph_execution_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_pipeline_meta_optimizer)
+list(APPEND MIXED_DIST_TEST_OPS test_fleet_amp_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_gradient_merge_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_localsgd_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_lars_meta_optimizer)
@@ -372,6 +373,7 @@ if(WITH_DISTRIBUTE)
 	   py_test_modules(test_fleet_graph_execution_meta_optimizer MODULES test_fleet_graph_execution_meta_optimizer ENVS ${dist_ENVS})
 	   py_test_modules(test_fleet_graph_executor MODULES test_fleet_graph_executor ENVS ${dist_ENVS})
            py_test_modules(test_fleet_gradient_merge_meta_optimizer MODULES test_fleet_gradient_merge_meta_optimizer ENVS ${dist_ENVS})
+           py_test_modules(test_fleet_amp_meta_optimizer MODULES test_fleet_amp_meta_optimizer ENVS ${dist_ENVS})
     	   py_test_modules(test_fleet_pipeline_meta_optimizer MODULES test_fleet_pipeline_meta_optimizer ENVS ${dist_ENVS})
     	   py_test_modules(test_fleet_private_function MODULES test_fleet_private_function ENVS ${dist_ENVS})
         if(NOT WIN32)
diff --git a/python/paddle/fluid/tests/unittests/test_fleet_amp_meta_optimizer.py b/python/paddle/fluid/tests/unittests/test_fleet_amp_meta_optimizer.py
new file mode 100644
index 0000000000000000000000000000000000000000..ae4b5d7ecd7c5131e38904a0d8fde0b9bb4fbb89
--- /dev/null
+++ b/python/paddle/fluid/tests/unittests/test_fleet_amp_meta_optimizer.py
@@ -0,0 +1,64 @@
+# Copyright (c) 2020 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
+# limitations under the License.
+
+import unittest
+import paddle
+import os
+
+
+class TestFleetAMPOptimizer(unittest.TestCase):
+    def setUp(self):
+        os.environ["PADDLE_TRAINER_ID"] = "0"
+        os.environ["PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001"
+
+    def test_amp_optimizer(self):
+        import paddle.fleet as fleet
+        import paddle.fluid.incubate.fleet.base.role_maker as role_maker
+        role = role_maker.PaddleCloudRoleMaker(is_collective=True)
+        fleet.init(role)
+        input_x = paddle.fluid.layers.data(
+            name="x", shape=[32], dtype='float32')
+        input_y = paddle.fluid.layers.data(name="y", shape=[1], dtype='int64')
+
+        fc_1 = paddle.fluid.layers.fc(input=input_x, size=64, act='tanh')
+        fc_2 = paddle.fluid.layers.fc(input=fc_1, size=64, act='tanh')
+        prediction = paddle.fluid.layers.fc(input=[fc_2], size=2, act='softmax')
+        cost = paddle.fluid.layers.cross_entropy(
+            input=prediction, label=input_y)
+        avg_cost = paddle.fluid.layers.mean(x=cost)
+
+        strategy = paddle.fleet.DistributedStrategy()
+        strategy.amp = True
+        strategy.amp_configs = {
+            "init_loss_scaling": 32768,
+            "decr_every_n_nan_or_inf": 2,
+            "incr_every_n_steps": 1000,
+            "incr_ratio": 2.0,
+            "use_dynamic_loss_scaling": True,
+            "decr_ratio": 0.5,
+            "custom_white_list": ['softmax'],
+            "custom_black_list": ['tanh'],
+        }
+
+        optimizer = paddle.optimizer.SGD(learning_rate=0.01)
+        optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
+        optimizer.minimize(avg_cost)
+
+        ops = [op.type for op in avg_cost.block.ops]
+        self.assertIn('cast', ops)
+        self.assertIn('isfinite', ops)
+
+
+if __name__ == "__main__":
+    unittest.main()