diff --git a/python/paddle/amp/auto_cast.py b/python/paddle/amp/auto_cast.py
index bcba245178d429460509d7b073f28c003a313178..9f0b6ac26927f68cc637e8dc5230c59fbe20aa37 100644
--- a/python/paddle/amp/auto_cast.py
+++ b/python/paddle/amp/auto_cast.py
@@ -59,6 +59,7 @@ BLACK_LIST = {
     'trilinear_interp_v2',
 }
 
+
 AMP_RELATED_FLAGS = [
     'FLAGS_cudnn_exhaustive_search',
     'FLAGS_conv_workspace_size_limit',
@@ -127,7 +128,7 @@ def amp_state():
     return _g_amp_state_
 
 
-# NOTE(zhiqiu): similar as paddle.fluid.contrib.mixed_precision.fp16_lists.AutoMixedPrecisionLists._update_list
+# NOTE(zhiqiu): similar as paddle.static.amp.fp16_lists.AutoMixedPrecisionLists._update_list
 # The reason why not use AutoMixedPrecisionLists is that custom_black_varnames is not suitable for imperative mode.
 def _update_list(
     custom_white_list, custom_black_list, level='O1', dtype='float16'
diff --git a/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py
index 2c0dbce0556de1805579463edf9d5c1fdf146246..de0aca0aea648f459a0d46903a54db969e7579d0 100755
--- a/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py
+++ b/python/paddle/distributed/fleet/meta_optimizers/amp_optimizer.py
@@ -11,7 +11,7 @@
 # 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
+import paddle.static.amp as mixed_precision
 
 from .meta_optimizer_base import MetaOptimizerBase
 
diff --git a/python/paddle/distributed/passes/auto_parallel_amp.py b/python/paddle/distributed/passes/auto_parallel_amp.py
index 8f965bdd9502500eb7d364627ca445840a7862a6..c2e5933dbf965701689eb197b5974f37d4c92695 100644
--- a/python/paddle/distributed/passes/auto_parallel_amp.py
+++ b/python/paddle/distributed/passes/auto_parallel_amp.py
@@ -23,7 +23,9 @@ from paddle.distributed.auto_parallel.utils import (
     set_var_dist_attr,
 )
 from paddle.distributed.fleet.meta_optimizers.common import OpRole
-from paddle.fluid.contrib.mixed_precision.fp16_utils import (
+from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
+from paddle.framework import core
+from paddle.static.amp.fp16_utils import (
     AutoMixedPrecisionLists,
     _dtype_to_str,
     _is_in_black_varnames,
@@ -35,8 +37,6 @@ from paddle.fluid.contrib.mixed_precision.fp16_utils import (
     find_true_post_op,
     find_true_prev_op,
 )
-from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
-from paddle.framework import core
 from paddle.utils import unique_name
 
 from ..auto_parallel.process_mesh import ProcessMesh
@@ -83,7 +83,7 @@ class AMPState:
 
     def _mark_black_white_ops(self, amp_lists):
         """
-        this function is modified from paddle.fluid.contrib.mixed_precision
+        this function is modified from paddle.static.amp
         """
         self._block._sync_with_cpp()
         ops = self._block.ops
@@ -179,7 +179,7 @@ class AMPState:
     ):
         """
         only for forward cast
-        modified from paddle.fluid.contrib.mixed_precision
+        modified from paddle.static.amp
         """
         num_cast_ops = 0
         var_name_dict = {}
diff --git a/python/paddle/distributed/passes/auto_parallel_bf16.py b/python/paddle/distributed/passes/auto_parallel_bf16.py
index 3344c6482446439b96bacb07ee1cd5997059773c..b733ab38ad742631cdcb61bb6c7095d3949abe88 100644
--- a/python/paddle/distributed/passes/auto_parallel_bf16.py
+++ b/python/paddle/distributed/passes/auto_parallel_bf16.py
@@ -26,22 +26,20 @@ from paddle.distributed.auto_parallel.utils import (
 from paddle.distributed.fleet.meta_optimizers.common import OpRole
 from paddle.distributed.passes.pass_base import PassBase, register_pass
 from paddle.fluid import unique_name
-from paddle.fluid.contrib.mixed_precision.bf16 import (
-    AutoMixedPrecisionListsBF16,
-)
-from paddle.fluid.contrib.mixed_precision.bf16.amp_utils import (
+from paddle.fluid.framework import Block
+from paddle.framework import core
+from paddle.static.amp.bf16 import AutoMixedPrecisionListsBF16
+from paddle.static.amp.bf16.amp_utils import (
     _dtype_to_str,
     _is_in_fp32_varnames,
     _valid_types,
-    find_op_index,
     find_true_post_op,
 )
-from paddle.fluid.contrib.mixed_precision.fp16_utils import (
+from paddle.static.amp.fp16_utils import (
     _rename_arg,
+    find_op_index,
     find_true_prev_op,
 )
-from paddle.fluid.framework import Block
-from paddle.framework import core
 
 from ..auto_parallel.utils import is_backward_op, is_forward_op, is_loss_op
 
diff --git a/python/paddle/distributed/passes/auto_parallel_fp16.py b/python/paddle/distributed/passes/auto_parallel_fp16.py
index 53aba19b98fdf2b97bcab57024e60e69c070e504..1d61100b72f775c70e819a56d9fc7c83c706a869 100644
--- a/python/paddle/distributed/passes/auto_parallel_fp16.py
+++ b/python/paddle/distributed/passes/auto_parallel_fp16.py
@@ -26,16 +26,16 @@ from paddle.distributed.auto_parallel.utils import (
     set_var_dist_attr,
 )
 from paddle.distributed.fleet.meta_optimizers.common import OP_ROLE_KEY, OpRole
-from paddle.fluid.contrib.mixed_precision.fp16_utils import (
+from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
+from paddle.framework import core
+from paddle.static import default_main_program, default_startup_program
+from paddle.static.amp.fp16_utils import (
     AutoMixedPrecisionLists,
     _dtype_to_str,
     _keep_layer_norm_scale_bias_to_fp32,
     _need_keep_fp32,
     _valid_types,
 )
-from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
-from paddle.framework import core
-from paddle.static import default_main_program, default_startup_program
 from paddle.utils import unique_name
 
 from ..auto_parallel.process_mesh import ProcessMesh
diff --git a/python/paddle/fluid/contrib/__init__.py b/python/paddle/fluid/contrib/__init__.py
index 5ad924b49f215761e58f123e9b172cafbefac3c7..c001b87570ff76842495080767e36e83786e5795 100644
--- a/python/paddle/fluid/contrib/__init__.py
+++ b/python/paddle/fluid/contrib/__init__.py
@@ -22,8 +22,6 @@ from . import extend_optimizer
 from .extend_optimizer import *
 from . import model_stat
 from .model_stat import *
-from . import mixed_precision
-from .mixed_precision import *
 from . import layers
 from .layers import *
 from . import optimizer
@@ -34,6 +32,5 @@ __all__ = []
 __all__ += memory_usage_calc.__all__
 __all__ += op_frequence.__all__
 __all__ += extend_optimizer.__all__
-__all__ += ['mixed_precision']
 __all__ += layers.__all__
 __all__ += optimizer.__all__
diff --git a/python/paddle/fluid/contrib/layers/nn.py b/python/paddle/fluid/contrib/layers/nn.py
index e9ff925fd8def6ba60d9afcbb82bc63c69d4448f..7ab225c8c9f825d60c7e910be92f0190106f3860 100644
--- a/python/paddle/fluid/contrib/layers/nn.py
+++ b/python/paddle/fluid/contrib/layers/nn.py
@@ -1587,7 +1587,7 @@ def fused_bn_add_act(
                     )
                     loss = paddle.mean(loss)
                     sgd = fluid.optimizer.SGD(learning_rate=0.001)
-                    sgd = fluid.contrib.mixed_precision.decorate(
+                    sgd = paddle.static.amp.decorate(
                         sgd, use_dynamic_loss_scaling=True, init_loss_scaling=128.0)
                     sgd.minimize(loss)
 
diff --git a/python/paddle/fluid/contrib/mixed_precision/__init__.py b/python/paddle/fluid/contrib/mixed_precision/__init__.py
deleted file mode 100644
index 07b507d61549a5bbefc12fd0f910357dbad3cbf4..0000000000000000000000000000000000000000
--- a/python/paddle/fluid/contrib/mixed_precision/__init__.py
+++ /dev/null
@@ -1,26 +0,0 @@
-#   Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
-#
-# 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.
-
-from . import decorator
-from .decorator import *
-from . import fp16_lists
-from .fp16_lists import *
-from . import fp16_utils
-from .fp16_utils import *
-from . import bf16
-
-__all__ = []
-__all__ += decorator.__all__
-__all__ += fp16_lists.__all__
-__all__ += fp16_utils.__all__
diff --git a/python/paddle/fluid/contrib/tests/test_amp_list.py b/python/paddle/fluid/contrib/tests/test_amp_list.py
index 4a30185229a9765787d5f8c2315c2a5ad86fdfc7..eb7860b9cfab7f5a3c80eb8e14c83bdee9bc03d6 100644
--- a/python/paddle/fluid/contrib/tests/test_amp_list.py
+++ b/python/paddle/fluid/contrib/tests/test_amp_list.py
@@ -14,7 +14,7 @@
 
 import paddle
 import unittest
-from paddle.fluid.contrib.mixed_precision.fp16_lists import (
+from paddle.static.amp.fp16_lists import (
     AutoMixedPrecisionLists,
 )
 
diff --git a/python/paddle/fluid/contrib/tests/test_bf16_utils.py b/python/paddle/fluid/contrib/tests/test_bf16_utils.py
index 5d528f226596ce65097eb0c4b7a71a17778c03c7..e3f36ca5b29658cbe3049ba4b07b64ca22aa1368 100644
--- a/python/paddle/fluid/contrib/tests/test_bf16_utils.py
+++ b/python/paddle/fluid/contrib/tests/test_bf16_utils.py
@@ -110,7 +110,7 @@ class AMPTest2(unittest.TestCase):
     def test_find_op_index(self):
         block = fluid.default_main_program().global_block()
         op_desc = core.OpDesc()
-        idx = amp.bf16.amp_utils.find_op_index(block.desc, op_desc)
+        idx = amp.fp16_utils.find_op_index(block.desc, op_desc)
         assert idx == -1
 
     def test_is_in_fp32_varnames(self):
diff --git a/python/paddle/fluid/contrib/tests/test_fp16_utils.py b/python/paddle/fluid/contrib/tests/test_fp16_utils.py
index af245a30fed43b95a1b1658f1aefe9e645903152..83500adb426880377eb1399028cf2f0e44d4385c 100644
--- a/python/paddle/fluid/contrib/tests/test_fp16_utils.py
+++ b/python/paddle/fluid/contrib/tests/test_fp16_utils.py
@@ -15,7 +15,7 @@
 import unittest
 import paddle.fluid as fluid
 from paddle.fluid import core
-from paddle.fluid.contrib.mixed_precision import fp16_utils
+from paddle.static.amp import fp16_utils
 import paddle
 
 paddle.enable_static()
diff --git a/python/paddle/fluid/contrib/tests/test_image_classification_fp16.py b/python/paddle/fluid/contrib/tests/test_image_classification_fp16.py
index a17495fde947737cc27250edba12fd4237762c1c..f99d640799764af631d02a2faef170a46de971e6 100644
--- a/python/paddle/fluid/contrib/tests/test_image_classification_fp16.py
+++ b/python/paddle/fluid/contrib/tests/test_image_classification_fp16.py
@@ -136,7 +136,7 @@ def train(net_type, use_cuda, save_dirname, is_local):
 
         optimizer = fluid.optimizer.Lamb(learning_rate=0.001)
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists(
             custom_black_varnames={"loss", "conv2d_0.w_0"}
         )
         mp_optimizer = decorate(
@@ -313,102 +313,66 @@ class TestImageClassification(unittest.TestCase):
         # infer(use_cuda, save_dirname)
 
     def test_amp_lists(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists()
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists()
         self.assertEqual(amp_lists.white_list, white_list)
         self.assertEqual(amp_lists.black_list, black_list)
         self.assertEqual(amp_lists.gray_list, gray_list)
 
     def test_amp_lists_1(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
         # 1. w={'exp}, b=None
         white_list.add('exp')
         black_list.remove('exp')
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
-            {'exp'}
-        )
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists({'exp'})
         self.assertEqual(amp_lists.white_list, white_list)
         self.assertEqual(amp_lists.black_list, black_list)
         self.assertEqual(amp_lists.gray_list, gray_list)
 
     def test_amp_lists_2(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
         # 2. w={'tanh'}, b=None
         white_list.add('tanh')
         gray_list.remove('tanh')
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
-            {'tanh'}
-        )
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists({'tanh'})
         self.assertEqual(amp_lists.white_list, white_list)
         self.assertEqual(amp_lists.black_list, black_list)
         self.assertEqual(amp_lists.gray_list, gray_list)
 
     def test_amp_lists_3(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
         # 3. w={'lstm'}, b=None
         white_list.add('lstm')
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
-            {'lstm'}
-        )
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists({'lstm'})
         self.assertEqual(amp_lists.white_list, white_list)
         self.assertEqual(amp_lists.black_list, black_list)
         self.assertEqual(amp_lists.gray_list, gray_list)
 
     def test_amp_lists_4(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
         # 4. w=None, b={'conv2d'}
         white_list.remove('conv2d')
         black_list.add('conv2d')
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists(
             custom_black_list={'conv2d'}
         )
         self.assertEqual(amp_lists.white_list, white_list)
@@ -416,21 +380,15 @@ class TestImageClassification(unittest.TestCase):
         self.assertEqual(amp_lists.gray_list, gray_list)
 
     def test_amp_lists_5(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
         # 5. w=None, b={'tanh'}
         black_list.add('tanh')
         gray_list.remove('tanh')
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists(
             custom_black_list={'tanh'}
         )
         self.assertEqual(amp_lists.white_list, white_list)
@@ -438,20 +396,14 @@ class TestImageClassification(unittest.TestCase):
         self.assertEqual(amp_lists.gray_list, gray_list)
 
     def test_amp_lists_6(self):
-        white_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.white_list
-        )
-        black_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.black_list
-        )
-        gray_list = copy.copy(
-            fluid.contrib.mixed_precision.fp16_lists.gray_list
-        )
+        white_list = copy.copy(paddle.static.amp.fp16_lists.white_list)
+        black_list = copy.copy(paddle.static.amp.fp16_lists.black_list)
+        gray_list = copy.copy(paddle.static.amp.fp16_lists.gray_list)
 
         # 6. w=None, b={'lstm'}
         black_list.add('lstm')
 
-        amp_lists = fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
+        amp_lists = paddle.static.amp.AutoMixedPrecisionLists(
             custom_black_list={'lstm'}
         )
         self.assertEqual(amp_lists.white_list, white_list)
@@ -463,7 +415,7 @@ class TestImageClassification(unittest.TestCase):
         # raise ValueError
         self.assertRaises(
             ValueError,
-            fluid.contrib.mixed_precision.AutoMixedPrecisionLists,
+            paddle.static.amp.AutoMixedPrecisionLists,
             {'lstm'},
             {'lstm'},
         )
@@ -515,10 +467,8 @@ class TestAmpWithNonIterableDataLoader(unittest.TestCase):
                 avg_cost = paddle.mean(cost)
 
                 optimizer = fluid.optimizer.Lamb(learning_rate=0.001)
-                amp_lists = (
-                    fluid.contrib.mixed_precision.AutoMixedPrecisionLists(
-                        custom_black_varnames={"loss", "conv2d_0.w_0"}
-                    )
+                amp_lists = paddle.static.amp.AutoMixedPrecisionLists(
+                    custom_black_varnames={"loss", "conv2d_0.w_0"}
                 )
                 mp_optimizer = decorate(
                     optimizer=optimizer,
diff --git a/python/paddle/fluid/contrib/tests/test_multi_precision_fp16_train.py b/python/paddle/fluid/contrib/tests/test_multi_precision_fp16_train.py
index 7c6cf13bec5e92ec3774045be0e3a6831e081ca0..1d4873817ab3a2b1613dea3ec578fc0cf48e9ab5 100644
--- a/python/paddle/fluid/contrib/tests/test_multi_precision_fp16_train.py
+++ b/python/paddle/fluid/contrib/tests/test_multi_precision_fp16_train.py
@@ -18,7 +18,7 @@ import contextlib
 import unittest
 import numpy as np
 from paddle.io import Dataset
-from paddle.fluid.contrib.mixed_precision.fp16_utils import cast_model_to_fp16
+from paddle.static.amp.fp16_utils import cast_model_to_fp16
 
 paddle.enable_static()
 
diff --git a/python/paddle/fluid/incubate/fleet/base/fleet_base.py b/python/paddle/fluid/incubate/fleet/base/fleet_base.py
index c53090a52bde2a351d5659095ce4ca577a7c09db..2781c505cb3676fec13ba9056ca7973829dba096 100644
--- a/python/paddle/fluid/incubate/fleet/base/fleet_base.py
+++ b/python/paddle/fluid/incubate/fleet/base/fleet_base.py
@@ -21,7 +21,7 @@ from paddle.optimizer import SGD as SGD_v2
 
 from paddle.fluid.incubate.fleet.base.mode import Mode
 from paddle.distributed.fleet.base.role_maker import RoleMakerBase
-from paddle.fluid.contrib.mixed_precision.decorator import (
+from paddle.static.amp.decorator import (
     OptimizerWithMixedPrecision,
 )
 from . import mode
diff --git a/python/paddle/fluid/incubate/fleet/collective/__init__.py b/python/paddle/fluid/incubate/fleet/collective/__init__.py
index c18e77b29c2b256c90379e93641be2516f62b6a6..e0668bddc9b597d02e9ee0f3037d726be97ce448 100644
--- a/python/paddle/fluid/incubate/fleet/collective/__init__.py
+++ b/python/paddle/fluid/incubate/fleet/collective/__init__.py
@@ -555,7 +555,7 @@ class CollectiveOptimizer(DistributedOptimizer):
                 self.raiseOptimizeError(
                     "mixed_precision", self._optimizer.__class__.__name__
                 )
-            self._optimizer = fluid.contrib.mixed_precision.decorate(
+            self._optimizer = paddle.static.amp.decorate(
                 self._optimizer,
                 init_loss_scaling=self._amp_loss_scaling,
                 use_dynamic_loss_scaling=True,
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index 3b251d5be3e5e07acda5fd018b7d209388d84d17..1bab4f34317719ae4d806a5ca73bd4ac7ae86ed8 100755
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -4404,7 +4404,7 @@ class PipelineOptimizer:
         valid_optimizers = (
             Optimizer,
             paddle.optimizer.Optimizer,
-            paddle.fluid.contrib.mixed_precision.decorator.OptimizerWithMixedPrecision,
+            paddle.static.amp.decorator.OptimizerWithMixedPrecision,
         )
         if not isinstance(optimizer, valid_optimizers):
             raise ValueError(
diff --git a/python/paddle/fluid/tests/unittests/asp/test_asp_optimize_static.py b/python/paddle/fluid/tests/unittests/asp/test_asp_optimize_static.py
index 81d6bd86ee466bf411677a1f4801c67fcbbeba7a..67b14cc549c9f932bed8b8ca459a34cbf8d61a36 100644
--- a/python/paddle/fluid/tests/unittests/asp/test_asp_optimize_static.py
+++ b/python/paddle/fluid/tests/unittests/asp/test_asp_optimize_static.py
@@ -215,11 +215,7 @@ class TestASPStaticOptimize(unittest.TestCase):
         if core.is_compiled_with_cuda():
             place = paddle.CUDAPlace(0)
             with fluid.program_guard(self.main_program, self.startup_program):
-                self.optimizer = (
-                    fluid.contrib.mixed_precision.decorator.decorate(
-                        self.optimizer
-                    )
-                )
+                self.optimizer = paddle.static.amp.decorate(self.optimizer)
                 self.optimizer = paddle.incubate.asp.decorate(self.optimizer)
                 self.optimizer.minimize(self.loss, self.startup_program)
 
diff --git a/python/paddle/fluid/tests/unittests/auto_parallel/test_pass_bf16.py b/python/paddle/fluid/tests/unittests/auto_parallel/test_pass_bf16.py
index f26908df2cf029b00c44b9feb389ffc0bfdcbbf2..2c72ccd9386c1546f208c2ec3be90c41d8853395 100644
--- a/python/paddle/fluid/tests/unittests/auto_parallel/test_pass_bf16.py
+++ b/python/paddle/fluid/tests/unittests/auto_parallel/test_pass_bf16.py
@@ -21,10 +21,10 @@ import paddle
 import paddle.fluid.core as core
 import paddle.nn as nn
 from paddle.distributed.fleet import auto
-from paddle.fluid.contrib.mixed_precision.bf16.amp_utils import _valid_types
-from paddle.fluid.contrib.mixed_precision.fp16_utils import find_true_prev_op
 from paddle.fluid.dygraph.parallel import ParallelEnv
 from paddle.static import InputSpec
+from paddle.static.amp.bf16.amp_utils import _valid_types
+from paddle.static.amp.fp16_utils import find_true_prev_op
 from paddle.vision.datasets import MNIST
 
 paddle.enable_static()
diff --git a/python/paddle/fluid/tests/unittests/collective/test_collective_optimizer.py b/python/paddle/fluid/tests/unittests/collective/test_collective_optimizer.py
index 50664ac09ea0eb53773b2d2dc4cfc3e7da9b86bd..7f4bcbe34be9fc3e6002ca06ecde36957d0c22a9 100644
--- a/python/paddle/fluid/tests/unittests/collective/test_collective_optimizer.py
+++ b/python/paddle/fluid/tests/unittests/collective/test_collective_optimizer.py
@@ -26,6 +26,7 @@
 
 import unittest
 
+import paddle
 import paddle.fluid as fluid
 from paddle.fluid.incubate.fleet.collective import (
     CollectiveOptimizer,
@@ -61,7 +62,7 @@ class CollectiveOptimizerTest(unittest.TestCase):
 
     def test_amp_strategy(self):
         optimizer = fluid.optimizer.AdamOptimizer()
-        optimizer = fluid.contrib.mixed_precision.decorate(
+        optimizer = paddle.static.amp.decorate(
             optimizer, init_loss_scaling=1.0, use_dynamic_loss_scaling=True
         )
         dist_strategy = DistributedStrategy()
diff --git a/python/paddle/fluid/tests/unittests/mlu/test_layer_norm_op_mlu.py b/python/paddle/fluid/tests/unittests/mlu/test_layer_norm_op_mlu.py
index dc0aa1d7d2b8eafdd83f7b1ffa276f309294249b..6e095c9b85144c7b61f3bebf8b4269f6df80489b 100644
--- a/python/paddle/fluid/tests/unittests/mlu/test_layer_norm_op_mlu.py
+++ b/python/paddle/fluid/tests/unittests/mlu/test_layer_norm_op_mlu.py
@@ -25,7 +25,7 @@ import sys
 sys.path.append('..')
 from op_test import _set_use_system_allocator
 from paddle.fluid import Program, program_guard
-from paddle.fluid.contrib.mixed_precision.fp16_utils import (
+from paddle.static.amp.fp16_utils import (
     _keep_layer_norm_scale_bias_to_fp32,
 )
 from test_layer_norm_op import (
diff --git a/python/paddle/fluid/tests/unittests/npu/test_amp_check_finite_and_scale_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_amp_check_finite_and_scale_op_npu.py
index 43e3c44182de08ff1f967da38bc1e55fffcdcc91..d61e6ed9500858abf757d91dec46f035819ee63a 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_amp_check_finite_and_scale_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_amp_check_finite_and_scale_op_npu.py
@@ -21,7 +21,7 @@ from op_test import OpTest, skip_check_grad_ci
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid import compiler, Program, program_guard
-from paddle.fluid.contrib.mixed_precision.amp_nn import check_finite_and_unscale
+from paddle.static.amp.amp_nn import check_finite_and_unscale
 
 paddle.enable_static()
 
diff --git a/python/paddle/fluid/tests/unittests/npu/test_mixed_precision_npu.py b/python/paddle/fluid/tests/unittests/npu/test_mixed_precision_npu.py
index de85b7eefdf08f1dab73feb6ad383eefef4c1ce5..d5d048771f043ff6483b809657ea520cdb7a0ff6 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_mixed_precision_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_mixed_precision_npu.py
@@ -17,7 +17,7 @@ import sys
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid import core
-from paddle.fluid.contrib.mixed_precision import fp16_utils
+from paddle.static.amp import fp16_utils
 import paddle.nn as nn
 import paddle.static as static
 import numpy as np
diff --git a/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_min_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_min_op_npu.py
index 48edac9ab331fc6981925708a4cda1d87c53be98..862a69e19ac2e059f388b4371d6ee47dff4378d6 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_min_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_min_op_npu.py
@@ -21,7 +21,7 @@ sys.path.append("..")
 from op_test import OpTest
 import paddle
 import paddle.fluid as fluid
-import paddle.fluid.contrib.mixed_precision.amp_nn as amp_nn
+import paddle.static.amp.amp_nn as amp_nn
 from test_update_loss_scaling_op_npu import TestUpdateLossScalingOpBad
 
 paddle.enable_static()
diff --git a/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_op_npu.py b/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_op_npu.py
index 8d364af89e3cb5a5b1d02cd59e6a370dc9859e93..07fb19031c0764012f45804a355e4f9fb84b125c 100644
--- a/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_op_npu.py
+++ b/python/paddle/fluid/tests/unittests/npu/test_update_loss_scaling_op_npu.py
@@ -20,7 +20,7 @@ sys.path.append("..")
 from op_test import OpTest
 import paddle
 import paddle.fluid as fluid
-import paddle.fluid.contrib.mixed_precision.amp_nn as amp_nn
+import paddle.static.amp.amp_nn as amp_nn
 
 paddle.enable_static()
 SEED = 2021
diff --git a/python/paddle/fluid/tests/unittests/test_amp_check_finite_and_scale_op.py b/python/paddle/fluid/tests/unittests/test_amp_check_finite_and_scale_op.py
index a840b582474f3448b0fe308c3571c88a55970736..e63c2846f3215ef27b29e9b781a8da43ae7358a9 100644
--- a/python/paddle/fluid/tests/unittests/test_amp_check_finite_and_scale_op.py
+++ b/python/paddle/fluid/tests/unittests/test_amp_check_finite_and_scale_op.py
@@ -17,7 +17,7 @@ import unittest
 import numpy as np
 from op_test import OpTest
 
-import paddle.fluid.contrib.mixed_precision.amp_nn as amp_nn
+import paddle.static.amp.amp_nn as amp_nn
 
 
 def check_finite_and_unscale_wrapper(x, scale):
diff --git a/python/paddle/fluid/tests/unittests/test_fuse_bn_act_pass.py b/python/paddle/fluid/tests/unittests/test_fuse_bn_act_pass.py
index ccb397fc19c74d10f115b9b6d5e0ca9e71c1d9b7..9e067454482d79b8292a35fd79d357c4d0ab0f78 100644
--- a/python/paddle/fluid/tests/unittests/test_fuse_bn_act_pass.py
+++ b/python/paddle/fluid/tests/unittests/test_fuse_bn_act_pass.py
@@ -61,7 +61,7 @@ class TestFuseBatchNormActPass(unittest.TestCase):
             loss = paddle.mean(loss)
             sgd = fluid.optimizer.SGD(learning_rate=0.001)
             if use_cuda:
-                sgd = fluid.contrib.mixed_precision.decorate(
+                sgd = paddle.static.amp.decorate(
                     sgd, use_dynamic_loss_scaling=True, init_loss_scaling=128.0
                 )
             sgd.minimize(loss)
diff --git a/python/paddle/fluid/tests/unittests/test_fuse_bn_add_act_pass.py b/python/paddle/fluid/tests/unittests/test_fuse_bn_add_act_pass.py
index 43578016e90ac987b9da86b591d87009c5fa6b2d..a8d0f89c86649d3e823b7c1f18d348b2fa9d105b 100644
--- a/python/paddle/fluid/tests/unittests/test_fuse_bn_add_act_pass.py
+++ b/python/paddle/fluid/tests/unittests/test_fuse_bn_add_act_pass.py
@@ -112,7 +112,7 @@ class TestFusedBnAddActAPI(unittest.TestCase):
             )
             loss = paddle.mean(loss)
             sgd = fluid.optimizer.SGD(learning_rate=0.001)
-            sgd = fluid.contrib.mixed_precision.decorate(
+            sgd = paddle.static.amp.decorate(
                 sgd, use_dynamic_loss_scaling=True, init_loss_scaling=128.0
             )
             sgd.minimize(loss)
@@ -173,7 +173,7 @@ class TestFusedBnAddActAPI(unittest.TestCase):
             )
             loss = paddle.mean(loss)
             sgd = fluid.optimizer.SGD(learning_rate=0.001)
-            sgd = fluid.contrib.mixed_precision.decorate(
+            sgd = paddle.static.amp.decorate(
                 sgd, use_dynamic_loss_scaling=True, init_loss_scaling=128.0
             )
             sgd.minimize(loss)
diff --git a/python/paddle/fluid/tests/unittests/test_layer_norm_op.py b/python/paddle/fluid/tests/unittests/test_layer_norm_op.py
index 35921c7e7033ec35c482e8a6604547ffdaca09a0..0240b411b047897514e18fc65119464619075ad2 100644
--- a/python/paddle/fluid/tests/unittests/test_layer_norm_op.py
+++ b/python/paddle/fluid/tests/unittests/test_layer_norm_op.py
@@ -24,9 +24,7 @@ import paddle.fluid as fluid
 import paddle.fluid.core as core
 import paddle.nn.functional as F
 from paddle.fluid import Program, program_guard
-from paddle.fluid.contrib.mixed_precision.fp16_utils import (
-    _keep_layer_norm_scale_bias_to_fp32,
-)
+from paddle.static.amp.fp16_utils import _keep_layer_norm_scale_bias_to_fp32
 
 paddle.enable_static()
 
diff --git a/python/paddle/fluid/tests/unittests/test_update_loss_scaling_op.py b/python/paddle/fluid/tests/unittests/test_update_loss_scaling_op.py
index 2d945496294e40f055e6acdbcbea440046000e08..00b943df85dc9fd6be34650b4b94e9f3b6a4796f 100644
--- a/python/paddle/fluid/tests/unittests/test_update_loss_scaling_op.py
+++ b/python/paddle/fluid/tests/unittests/test_update_loss_scaling_op.py
@@ -18,7 +18,7 @@ import numpy as np
 from op_test import OpTest
 
 import paddle.fluid as fluid
-import paddle.fluid.contrib.mixed_precision.amp_nn as amp_nn
+import paddle.static.amp.amp_nn as amp_nn
 
 
 def update_loss_scaling_wrapper(
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_update_loss_scaling_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_update_loss_scaling_op_xpu.py
index ad1cf1a5192b22783f29f1620ee6dbe50049557e..44d79efc97bfd1c66ce3cbaaa53b8354cd65712b 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_update_loss_scaling_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_update_loss_scaling_op_xpu.py
@@ -26,7 +26,7 @@ from xpu.get_test_cover_info import (
 
 import paddle
 import paddle.fluid as fluid
-import paddle.fluid.contrib.mixed_precision.amp_nn as amp_nn
+import paddle.static.amp.amp_nn as amp_nn
 
 paddle.enable_static()
 
diff --git a/python/paddle/jit/dy2static/partial_program.py b/python/paddle/jit/dy2static/partial_program.py
index 1b9908a6608a715f324afc8d827f85a3e3454686..040ef36854f3fb9db4d0684cd861fd3febba26f0 100644
--- a/python/paddle/jit/dy2static/partial_program.py
+++ b/python/paddle/jit/dy2static/partial_program.py
@@ -21,13 +21,6 @@ from paddle import _legacy_C_ops
 from paddle.amp.auto_cast import _in_amp_guard, _in_pure_fp16_guard
 from paddle.fluid import backward, core, framework, program_guard
 from paddle.fluid.compiler import BuildStrategy
-from paddle.fluid.contrib.mixed_precision.decorator import (
-    AutoMixedPrecisionLists,
-)
-from paddle.fluid.contrib.mixed_precision.fp16_utils import (
-    cast_model_to_fp16,
-    rewrite_program,
-)
 from paddle.fluid.dygraph import layers
 from paddle.fluid.dygraph.base import switch_to_static_graph
 from paddle.fluid.executor import (
@@ -189,7 +182,7 @@ class PartialProgramLayer:
         if tracer:
             custom_white_list, custom_black_list = tracer._get_amp_op_list()
         # For AMP training
-        self._amp_list = AutoMixedPrecisionLists(
+        self._amp_list = paddle.static.amp.fp16_lists.AutoMixedPrecisionLists(
             custom_white_list=custom_white_list,
             custom_black_list=custom_black_list,
         )
@@ -238,7 +231,9 @@ class PartialProgramLayer:
     def _create_amp_program(self, is_infer_mode=False):
         amp_program = self._origin_main_program.clone(for_test=is_infer_mode)
         with program_guard(amp_program):
-            rewrite_program(amp_program, self._amp_list)
+            paddle.static.amp.fp16_utils.rewrite_program(
+                amp_program, self._amp_list
+            )
         if is_infer_mode:
             return amp_program
         else:
@@ -252,7 +247,7 @@ class PartialProgramLayer:
             for_test=is_infer_mode
         )
         with program_guard(pure_fp16_program):
-            cast_model_to_fp16(
+            paddle.static.amp.fp16_utils.cast_model_to_fp16(
                 pure_fp16_program, self._amp_list, use_fp16_guard=False
             )
         if is_infer_mode:
diff --git a/python/paddle/static/amp/__init__.py b/python/paddle/static/amp/__init__.py
index ffcfa404cd3567e1f6784f274f90f0f436f0f9da..01832fd536769bff0b5613d399d35d61131cad5c 100644
--- a/python/paddle/static/amp/__init__.py
+++ b/python/paddle/static/amp/__init__.py
@@ -1,4 +1,4 @@
-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#   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.
@@ -12,14 +12,16 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from ...fluid.contrib.mixed_precision import decorate  # noqa: F401
-from ...fluid.contrib.mixed_precision import CustomOpLists  # noqa: F401
-from ...fluid.contrib.mixed_precision import (
-    AutoMixedPrecisionLists,
-)  # noqa: F401
-from ...fluid.contrib.mixed_precision import fp16_guard  # noqa: F401
-from ...fluid.contrib.mixed_precision import cast_model_to_fp16  # noqa: F401
-from ...fluid.contrib.mixed_precision import (
-    cast_parameters_to_fp16,
-)  # noqa: F401
-from ...fluid.contrib.mixed_precision import bf16  # noqa: F401
+from . import decorator
+from .decorator import decorate
+from . import fp16_lists
+from .fp16_lists import CustomOpLists, AutoMixedPrecisionLists
+from . import fp16_utils
+from .fp16_utils import fp16_guard, cast_model_to_fp16, cast_parameters_to_fp16
+from . import bf16
+from .bf16 import bf16_guard
+
+__all__ = []
+__all__ += decorator.__all__
+__all__ += fp16_lists.__all__
+__all__ += fp16_utils.__all__
diff --git a/python/paddle/fluid/contrib/mixed_precision/amp_nn.py b/python/paddle/static/amp/amp_nn.py
similarity index 98%
rename from python/paddle/fluid/contrib/mixed_precision/amp_nn.py
rename to python/paddle/static/amp/amp_nn.py
index f336c87a11e10a8fb3b710c933c62d35c2ea80ee..b20f702b44f79c9117409051ee3d86cd555ce97d 100644
--- a/python/paddle/fluid/contrib/mixed_precision/amp_nn.py
+++ b/python/paddle/static/amp/amp_nn.py
@@ -12,11 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from paddle.fluid.data_feeder import check_variable_and_dtype, check_type
-from paddle.fluid.layer_helper import LayerHelper
-from paddle.fluid.framework import Variable, in_dygraph_mode
-from paddle.fluid import core
 from paddle import _C_ops
+from paddle.fluid import core
+from paddle.fluid.data_feeder import check_type, check_variable_and_dtype
+from paddle.fluid.framework import Variable, in_dygraph_mode
+from paddle.fluid.layer_helper import LayerHelper
 
 __all__ = ['check_finite_and_unscale', 'update_loss_scaling']
 
diff --git a/python/paddle/fluid/contrib/mixed_precision/bf16/__init__.py b/python/paddle/static/amp/bf16/__init__.py
similarity index 76%
rename from python/paddle/fluid/contrib/mixed_precision/bf16/__init__.py
rename to python/paddle/static/amp/bf16/__init__.py
index 2e5e8451d6e181710dd711038f1c63b009e36899..82b616b299447c2a297d2dd5e718e8ef4a09b085 100644
--- a/python/paddle/fluid/contrib/mixed_precision/bf16/__init__.py
+++ b/python/paddle/static/amp/bf16/__init__.py
@@ -13,11 +13,17 @@
 # limitations under the License.
 
 from . import amp_lists
-from .amp_lists import *
+from .amp_lists import AutoMixedPrecisionListsBF16
 from . import amp_utils
-from .amp_utils import *
+from .amp_utils import (
+    bf16_guard,
+    rewrite_program_bf16,
+    cast_model_to_bf16,
+    cast_parameters_to_bf16,
+    convert_float_to_uint16,
+)
 from . import decorator
-from .decorator import *
+from .decorator import decorate_bf16
 
 __all__ = []
 __all__ += decorator.__all__
diff --git a/python/paddle/fluid/contrib/mixed_precision/bf16/amp_lists.py b/python/paddle/static/amp/bf16/amp_lists.py
similarity index 93%
rename from python/paddle/fluid/contrib/mixed_precision/bf16/amp_lists.py
rename to python/paddle/static/amp/bf16/amp_lists.py
index 180e28ddab351499f7e3ecf65a8384a32bd29454..d1878a3367fbc7a72a9cc90cea7028a6893ce37d 100644
--- a/python/paddle/fluid/contrib/mixed_precision/bf16/amp_lists.py
+++ b/python/paddle/static/amp/bf16/amp_lists.py
@@ -13,13 +13,12 @@
 # limitations under the License.
 
 import copy
+
 from paddle.fluid import core
 
-from ..fp16_lists import (
-    white_list as white_list_fp16,
-    black_list as black_list_fp16,
-    gray_list as gray_list_fp16,
-)
+from ..fp16_lists import black_list as black_list_fp16
+from ..fp16_lists import gray_list as gray_list_fp16
+from ..fp16_lists import white_list as white_list_fp16
 
 __all__ = ["AutoMixedPrecisionListsBF16"]
 
@@ -40,7 +39,7 @@ class AutoMixedPrecisionListsBF16:
         import paddle
         paddle.enable_static()
         with paddle.static.amp.bf16_guard():
-            paddle.static.amp.AutoMixedPrecisionListsBF16(custom_fp32_list={'lstm'})
+            paddle.static.amp.bf16.AutoMixedPrecisionListsBF16(custom_fp32_list={'lstm'})
     """
 
     def __init__(
diff --git a/python/paddle/fluid/contrib/mixed_precision/bf16/amp_utils.py b/python/paddle/static/amp/bf16/amp_utils.py
similarity index 99%
rename from python/paddle/fluid/contrib/mixed_precision/bf16/amp_utils.py
rename to python/paddle/static/amp/bf16/amp_utils.py
index 25468b563aa1280a5c700fd370d755bd394b01ac..cf8c82127b3b45f29559444a65bf9847124ccaf3 100644
--- a/python/paddle/fluid/contrib/mixed_precision/bf16/amp_utils.py
+++ b/python/paddle/static/amp/bf16/amp_utils.py
@@ -13,24 +13,23 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .... import core
-from .... import framework
-from .... import global_scope
-from ....log_helper import get_logger
-from ....wrapped_decorator import signature_safe_contextmanager
-from .amp_lists import AutoMixedPrecisionListsBF16
+import collections
+import logging
+import struct
+
+import numpy as np
+
+from paddle.fluid import core, framework, global_scope
+from paddle.fluid.log_helper import get_logger
+from paddle.fluid.wrapped_decorator import signature_safe_contextmanager
+
 from ..fp16_utils import (
-    find_true_prev_op,
-    find_true_post_op,
     _rename_arg,
-    find_op_index,
     _rename_op_input,
+    find_true_post_op,
+    find_true_prev_op,
 )
-
-import collections
-import struct
-import logging
-import numpy as np
+from .amp_lists import AutoMixedPrecisionListsBF16
 
 __all__ = [
     "bf16_guard",
diff --git a/python/paddle/fluid/contrib/mixed_precision/bf16/decorator.py b/python/paddle/static/amp/bf16/decorator.py
similarity index 99%
rename from python/paddle/fluid/contrib/mixed_precision/bf16/decorator.py
rename to python/paddle/static/amp/bf16/decorator.py
index d1648e81082638a0b734aeaf8ae90bce930b247e..20286d3eebca5fead08b8e0e1f291478e0bb2080 100644
--- a/python/paddle/fluid/contrib/mixed_precision/bf16/decorator.py
+++ b/python/paddle/static/amp/bf16/decorator.py
@@ -12,22 +12,18 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import types
+import warnings
+
 import paddle
-from paddle.fluid import (
-    core,
-    default_main_program,
-    layers,
-    program_guard,
-    unique_name,
-)
+from paddle.fluid import core, default_main_program, program_guard, unique_name
+
+from .amp_lists import AutoMixedPrecisionListsBF16
 from .amp_utils import (
-    rewrite_program_bf16,
     cast_model_to_bf16,
     cast_parameters_to_bf16,
+    rewrite_program_bf16,
 )
-from .amp_lists import AutoMixedPrecisionListsBF16
-import types
-import warnings
 
 __all__ = ["decorate_bf16"]
 
diff --git a/python/paddle/fluid/contrib/mixed_precision/decorator.py b/python/paddle/static/amp/decorator.py
similarity index 98%
rename from python/paddle/fluid/contrib/mixed_precision/decorator.py
rename to python/paddle/static/amp/decorator.py
index 68b3e7472d9647627ab2ed32f5b7a9e1662ac89e..ba33f6b391b0b825ad87ae6ecbf1a14778cf4b2e 100644
--- a/python/paddle/fluid/contrib/mixed_precision/decorator.py
+++ b/python/paddle/static/amp/decorator.py
@@ -12,24 +12,27 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from ... import core
-from ... import default_main_program
-from ... import default_startup_program
-from ... import framework
-from ... import layers
-from ... import program_guard
-from ... import unique_name
-from . import fp16_utils
-from .fp16_utils import rewrite_program
-from .fp16_utils import cast_model_to_fp16
-from .fp16_utils import cast_parameters_to_fp16
-from .fp16_utils import update_role_var_grad
-from .fp16_lists import AutoMixedPrecisionLists
-from .amp_nn import check_finite_and_unscale
-from .amp_nn import update_loss_scaling
 import types
 import warnings
+
 import paddle
+from paddle.fluid import (
+    core,
+    default_main_program,
+    default_startup_program,
+    layers,
+    program_guard,
+    unique_name,
+)
+
+from .amp_nn import check_finite_and_unscale, update_loss_scaling
+from .fp16_lists import AutoMixedPrecisionLists
+from .fp16_utils import (
+    cast_model_to_fp16,
+    cast_parameters_to_fp16,
+    rewrite_program,
+    update_role_var_grad,
+)
 
 __all__ = ["decorate"]
 
diff --git a/python/paddle/fluid/contrib/mixed_precision/fp16_lists.py b/python/paddle/static/amp/fp16_lists.py
similarity index 99%
rename from python/paddle/fluid/contrib/mixed_precision/fp16_lists.py
rename to python/paddle/static/amp/fp16_lists.py
index 101af59861880898c4d5bdfb80536659f1648769..b2acd0bb5156ddab063a02304dd09d199ee38c80 100644
--- a/python/paddle/fluid/contrib/mixed_precision/fp16_lists.py
+++ b/python/paddle/static/amp/fp16_lists.py
@@ -13,7 +13,8 @@
 # limitations under the License.
 
 import copy
-from ... import core
+
+from paddle.fluid import core
 
 __all__ = ["CustomOpLists", "AutoMixedPrecisionLists"]
 
diff --git a/python/paddle/fluid/contrib/mixed_precision/fp16_utils.py b/python/paddle/static/amp/fp16_utils.py
similarity index 99%
rename from python/paddle/fluid/contrib/mixed_precision/fp16_utils.py
rename to python/paddle/static/amp/fp16_utils.py
index e50a2457147e12d4497c90ad297d08bcade52736..c9cee2ab8d25cf96909896453703acfb483087e7 100644
--- a/python/paddle/fluid/contrib/mixed_precision/fp16_utils.py
+++ b/python/paddle/static/amp/fp16_utils.py
@@ -12,17 +12,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from ... import core
-from ... import framework
-from ... import layers
-from ... import global_scope
-from ...log_helper import get_logger
-from ...wrapped_decorator import signature_safe_contextmanager
-from .fp16_lists import AutoMixedPrecisionLists
 import collections
 import logging
+
 import numpy as np
 
+from paddle.fluid import core, framework, global_scope
+from paddle.fluid.log_helper import get_logger
+from paddle.fluid.wrapped_decorator import signature_safe_contextmanager
+
+from .fp16_lists import AutoMixedPrecisionLists
+
 __all__ = ["fp16_guard", "cast_model_to_fp16", "cast_parameters_to_fp16"]
 
 _logger = get_logger(
diff --git a/python/setup.py.in b/python/setup.py.in
index d6d01fc4b0a6bf5db880b489820baaecb1eb55d9..71a1cfb748198aaf6a5b3bfa05ba2c0af19e5bf5 100644
--- a/python/setup.py.in
+++ b/python/setup.py.in
@@ -337,8 +337,6 @@ packages=['paddle',
           'paddle.fluid.dataloader',
           'paddle.fluid.contrib',
           'paddle.fluid.contrib.extend_optimizer',
-          'paddle.fluid.contrib.mixed_precision',
-          'paddle.fluid.contrib.mixed_precision.bf16',
           'paddle.fluid.contrib.layers',
           'paddle.fluid.transpiler',
           'paddle.fluid.transpiler.details',
@@ -400,6 +398,7 @@ packages=['paddle',
           'paddle.static',
           'paddle.static.nn',
           'paddle.static.amp',
+          'paddle.static.amp.bf16',
           'paddle.static.quantization',
           'paddle.quantization',
           'paddle.quantization.imperative',
diff --git a/setup.py b/setup.py
index 177bb64d2842800f604c7bb84b42a1ce2fa3cc60..572d813c46c19b11e12409d967d8e97dd9005214 100644
--- a/setup.py
+++ b/setup.py
@@ -1236,8 +1236,6 @@ def get_setup_parameters():
         'paddle.fluid.dataloader',
         'paddle.fluid.contrib',
         'paddle.fluid.contrib.extend_optimizer',
-        'paddle.fluid.contrib.mixed_precision',
-        'paddle.fluid.contrib.mixed_precision.bf16',
         'paddle.fluid.contrib.layers',
         'paddle.fluid.transpiler',
         'paddle.fluid.transpiler.details',
@@ -1299,6 +1297,7 @@ def get_setup_parameters():
         'paddle.static',
         'paddle.static.nn',
         'paddle.static.amp',
+        'paddle.static.amp.bf16',
         'paddle.static.quantization',
         'paddle.quantization',
         'paddle.quantization.imperative',