remove fluid.initializer.UniformInitializer, ConstantInitializer,...

remove fluid.initializer.UniformInitializer, ConstantInitializer, NormalInitializer, TruncatedNormalInitializer, XavierInitializer, BilinearInitializer, MSRAInitializer, NumpyArrayInitializer and calculate_gain.. (#49498)

* move UniformInitializer and ConstantInitializer

* more modify

* circular import resolved

* another circular import resolved?

* more circular import 2

* circular import 3

* change import paddle in metric.py

* BuildStrategy import from fluid

* modify the framework import path in common.py

* change rnn.py import, from static to original framework

* change import static in the nn folder

* default_main_program should import from common_ops_import

* add import paddle in param_attr.py

* use core not paddle module for using VarDesc

* another old uniform

* mistake that use Uniform instead of UniformInitializer

* modify UniformInitializer doc

* move fluid.NormalInitializer to nn.initializer.NormalInitializer

* remove import of Normal in fluid.layers.nn.py

* remove more import of old Normal

* remove more import of old Normal

* sample code modify and tests modify import

* is_listen_failed passing arg should be log file

* problem solved

* a mistake solved

* comments resoleved and remove paddle.fluid.initializer.TruncatedNormalInitializer

* remove paddle.fluid.initializer.XavierInitializer and paddle.fluid.initializer.MSRAInitializer

* remove paddle.fluid.initializer.BilinearInitializer NumpyArrayInitializer and set_global_initializer

* change fluid to static

* change static to fluid to avoid circular import in distributed_strategy.py

* fix example code and test_initializer

* ValueType

* sample code fix

* change set_global_initializer back to fluid

* put paddle.static.BuildStrategy.ReduceStrategy into the fuction to avoid circular import

* remove calculate_gain, delete BilinearInitializer and revert set_global_initializer

* change the time of using UniformInitializer, ConstantInitializer, NormalInitializer, TruncatedNormalInitializer, XavierInitializer, MSRAInitializer, NumpyArrayInitializer as few as possible

* fix argument incampatible

* fix more arg incompatible

* fix test_prelu_op_xpu.py Constant

* fix inaccurate doc

* more doc fix: default value

python/paddle/common_ops_import.py

@@ -32,7 +32,6 @@ from paddle.fluid.framework import (  # noqa: F401
     dygraph_only,
     in_dygraph_mode,
 )
-from paddle.fluid.initializer import Constant  # noqa: F401
 from paddle.fluid.layer_helper import LayerHelper  # noqa: F401
 from paddle.fluid.layers import fill_constant, utils  # noqa: F401
 from paddle.fluid.layers.layer_function_generator import (  # noqa: F401

python/paddle/distributed/fleet/base/distributed_strategy.py

@@ -104,7 +104,6 @@ class DistributedJobInfo:
         self.job_info.strategy = dist_strategy
 
 
-ReduceStrategyFluid = paddle.static.BuildStrategy.ReduceStrategy
 ReduceStrategyFleet = int
 
 
@@ -261,7 +260,7 @@ class DistributedStrategy:
         for f in fields:
             value = getattr(self.strategy.build_strategy, f.name)
             if f.name == 'reduce_strategy':
-                value = ReduceStrategyFluid(value)
+                value = paddle.static.BuildStrategy.ReduceStrategy(value)
             setattr(build_strategy, f.name, value)
         return build_strategy
 

python/paddle/distributed/fleet/layers/mpu/random.py

@@ -18,11 +18,11 @@ import numpy as np
 
 import paddle
 from paddle import _legacy_C_ops
+from paddle.common_ops_import import Variable
 from paddle.fluid import core
 from paddle.fluid.data_feeder import check_variable_and_dtype
 from paddle.fluid.framework import in_dygraph_mode
 from paddle.framework import LayerHelper
-from paddle.static import Variable
 
 __all__ = []
 

python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py

@@ -171,7 +171,7 @@ class DGCMomentumOptimizer(Optimizer):
         if is_new_var:
             helper.set_variable_initializer(
                 counter,
-                initializer=paddle.fluid.initializer.Constant(
+                initializer=paddle.nn.initializer.ConstantInitializer(
                     value=float(begin - 1), force_cpu=True
                 ),
             )
@@ -194,7 +194,7 @@ class DGCMomentumOptimizer(Optimizer):
         if is_new_var:
             helper.set_variable_initializer(
                 counter,
-                initializer=paddle.fluid.initializer.Constant(
+                initializer=paddle.nn.initializer.ConstantInitializer(
                     value=float(value), force_cpu=True
                 ),
             )

python/paddle/distributed/fleet/metrics/metric.py

@@ -18,7 +18,7 @@ import math
 import numpy as np
 
 import paddle
-from paddle.static import Variable
+from paddle.common_ops_import import Variable
 
 __all__ = []
 

python/paddle/fluid/compiler.py

@@ -586,7 +586,6 @@ class IpuDynamicPatcher:
         """
         from ..fluid.dygraph.base import switch_to_static_graph
         from ..fluid import backward
-        from ..fluid.initializer import Constant
         from ..fluid.framework import device_guard
         import paddle
 
@@ -645,7 +644,10 @@ class IpuDynamicPatcher:
                     device = optimizer._get_device_for_param(param_name)
                     with device_guard(device):
                         optimizer.helper.set_variable_initializer(
-                            var, initializer=Constant(value=0.0)
+                            var,
+                            initializer=paddle.nn.initializer.Constant(
+                                value=0.0
+                            ),
                         )
                     param_or_lr_tensor = scope.find_var(
                         var_tmp.name

python/paddle/fluid/contrib/layers/metric_op.py

@@ -17,7 +17,6 @@ Contrib layers just related to metric.
 
 import warnings
 from paddle.fluid.layer_helper import LayerHelper
-from paddle.fluid.initializer import Normal, Constant
 from paddle.fluid.framework import Variable
 from paddle.fluid.param_attr import ParamAttr
 from paddle.fluid.layers import tensor
@@ -147,7 +146,10 @@ def ctr_metric_bundle(input, label, ins_tag_weight=None):
         local_ins_num,
     ]:
         helper.set_variable_initializer(
-            var, Constant(value=0.0, force_cpu=True)
+            var,
+            paddle.nn.initializer.ConstantInitializer(
+                value=0.0, force_cpu=True
+            ),
         )
 
     helper.append_op(

python/paddle/fluid/contrib/layers/nn.py

@@ -24,7 +24,6 @@ import paddle
 from paddle.fluid.layer_helper import LayerHelper
 from paddle.fluid.layers import utils
 from ... import unique_name
-from paddle.fluid.initializer import Normal, Constant, NumpyArrayInitializer
 from paddle.fluid.data_feeder import (
     check_variable_and_dtype,
     check_type,
@@ -896,8 +895,10 @@ def tdm_child(x, node_nums, child_nums, param_attr=None, dtype='int32'):
 
     Examples:
         .. code-block:: python
+        import paddle
         import paddle.fluid as fluid
         import numpy as np
+        paddle.enable_static()
         x = fluid.data(name="x", shape=[None, 1], dtype="int32", lod_level=1)
         tree_info = [[0,0,0,1,2],
                      [0,1,0,3,4],[0,1,0,5,6],
@@ -908,7 +909,7 @@ def tdm_child(x, node_nums, child_nums, param_attr=None, dtype='int32'):
         child_nums = 2
         child, leaf_mask  = fluid.contrib.layers.tdm_child(x, node_nums, child_nums,
                                 param_attr=fluid.ParamAttr(
-                                    initializer=fluid.initializer.NumpyArrayInitializer(
+                                    initializer=paddle.nn.initializer.Assign(
                                                                             tree_info_np)))
         place = fluid.CPUPlace()
         exe = fluid.Executor(place)
@@ -925,7 +926,7 @@ def tdm_child(x, node_nums, child_nums, param_attr=None, dtype='int32'):
         attr=helper.param_attr,
         shape=[node_nums, 3 + child_nums],
         dtype=dtype,
-        default_initializer=Constant(0),
+        default_initializer=paddle.nn.initializer.Constant(0),
     )
     tree_info.stop_gradient = True
 
@@ -1003,8 +1004,10 @@ def tdm_sampler(
 
     Examples:
         .. code-block:: python
+        import paddle
         import paddle.fluid as fluid
         import numpy as np
+        paddle.enable_static()
         x = fluid.data(name="x", shape=[None, 1], dtype="int32", lod_level=1)
         travel_list = [[1, 3], [1, 4], [2, 5], [2, 6]] # leaf node's travel path, shape(leaf_node_num, layer_num)
         layer_list_flat = [[1], [2], [3], [4], [5], [6]] # shape(node_nums, 1)
@@ -1022,10 +1025,10 @@ def tdm_sampler(
             layer_node_num_list,
             leaf_node_num,
             tree_travel_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(
+                initializer=paddle.nn.initializer.Assign(
                     travel_array)),
             tree_layer_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(
+                initializer=paddle.nn.initializer.Assign(
                     layer_array)),
             output_positive=True,
             output_list=True,
@@ -1089,7 +1092,7 @@ def tdm_sampler(
         attr=tree_travel_attr,
         shape=travel_shape,
         dtype=tree_dtype,
-        default_initializer=Constant(0),
+        default_initializer=paddle.nn.initializer.Constant(0),
     )
 
     layer_shape = [node_nums, 1]
@@ -1097,7 +1100,7 @@ def tdm_sampler(
         attr=tree_layer_attr,
         shape=layer_shape,
         dtype=tree_dtype,
-        default_initializer=Constant(0),
+        default_initializer=paddle.nn.initializer.Constant(0),
     )
 
     out = helper.create_variable_for_type_inference(dtype=dtype)
@@ -1640,7 +1643,7 @@ def fused_bn_add_act(
         attr=helper.param_attr,
         shape=param_shape,
         dtype=bn_param_dtype,
-        default_initializer=Constant(1.0),
+        default_initializer=paddle.nn.initializer.Constant(1.0),
     )
     bias = helper.create_parameter(
         attr=helper.bias_attr,
@@ -1650,7 +1653,9 @@ def fused_bn_add_act(
     )
     mean = helper.create_parameter(
         attr=ParamAttr(
-            name=moving_mean_name, initializer=Constant(0.0), trainable=False
+            name=moving_mean_name,
+            initializer=paddle.nn.initializer.Constant(0.0),
+            trainable=False,
         ),
         shape=param_shape,
         dtype=bn_param_dtype,
@@ -1659,7 +1664,7 @@ def fused_bn_add_act(
     variance = helper.create_parameter(
         attr=ParamAttr(
             name=moving_variance_name,
-            initializer=Constant(1.0),
+            initializer=paddle.nn.initializer.Constant(1.0),
             trainable=False,
         ),
         shape=param_shape,
@@ -1723,13 +1728,16 @@ def pow2_decay_with_linear_warmup(
     helper = LayerHelper("pow2_decay_with_linear_warmup", **locals())
     lr = helper.create_global_variable(persistable=True, dtype=dtype, shape=[1])
     helper.set_variable_initializer(
-        lr, Constant(value=float(base_lr) / warmup_steps)
+        lr,
+        paddle.nn.initializer.Constant(value=float(base_lr) / warmup_steps),
     )
 
     step = helper.create_global_variable(
         persistable=True, dtype='int64', shape=[1]
     )
-    helper.set_variable_initializer(step, Constant(value=0))
+    helper.set_variable_initializer(
+        step, paddle.nn.initializer.Constant(value=0)
+    )
     assert (
         warmup_steps <= total_steps
     ), "warmup_steps cannot be larger than total_steps"

python/paddle/fluid/evaluator.py

@@ -20,7 +20,6 @@ from . import layers
 from .framework import Program, Variable, program_guard
 from . import unique_name
 from .layer_helper import LayerHelper
-from .initializer import Constant
 
 
 def _clone_var_(block, var):

python/paddle/fluid/incubate/fleet/tests/fleet_deep_ctr.py

@@ -109,7 +109,7 @@ def model():
         size=[dnn_input_dim, dnn_layer_dims[0]],
         param_attr=fluid.ParamAttr(
             name="deep_embedding",
-            initializer=fluid.initializer.Constant(value=0.01),
+            initializer=paddle.nn.initializer.Constant(value=0.01),
         ),
         is_sparse=True,
     )
@@ -121,7 +121,7 @@ def model():
             size=dim,
             activation="relu",
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01)
+                initializer=paddle.nn.initializer.Constant(value=0.01)
             ),
             name='dnn-fc-%d' % i,
         )
@@ -134,7 +134,7 @@ def model():
         size=[lr_input_dim, 1],
         param_attr=fluid.ParamAttr(
             name="wide_embedding",
-            initializer=fluid.initializer.Constant(value=0.01),
+            initializer=paddle.nn.initializer.Constant(value=0.01),
         ),
         is_sparse=True,
     )

python/paddle/fluid/install_check.py

@@ -22,7 +22,6 @@ from .framework import (
     cpu_places,
 )
 from .param_attr import ParamAttr
-from .initializer import Constant
 from . import layers
 from . import backward
 from .dygraph import Layer
@@ -42,7 +41,9 @@ class SimpleLayer(Layer):
         self._linear1 = paddle.nn.Linear(
             input_size,
             3,
-            weight_attr=ParamAttr(initializer=Constant(value=0.1)),
+            weight_attr=ParamAttr(
+                initializer=paddle.nn.initializer.Constant(value=0.1)
+            ),
         )
 
     def forward(self, inputs):

python/paddle/fluid/layer_helper.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import copy
-
+import paddle
 from .framework import (
     Parameter,
     dtype_is_floating,
@@ -22,7 +22,6 @@ from .framework import (
     _global_flags,
 )
 from . import unique_name
-from paddle.fluid.initializer import Constant, Xavier
 from .param_attr import ParamAttr
 from . import core
 
@@ -178,10 +177,10 @@ class LayerHelper(LayerHelperBase):
     # TODO (jiabin): should we remove this since it has never be used
     def _get_default_initializer(self, dtype):
         if dtype is None or dtype_is_floating(dtype) is True:
-            return Xavier()
+            return paddle.nn.initializer.XavierUniform()
         else:
             # For integer and boolean types, initialize with all zeros
-            return Constant()
+            return paddle.nn.initializer.Constant()
 
     # TODO (jiabin): reconstruct this in LayerObjHelper and avoid dependency of kwargs
     def is_instance(self, param_name, cls):

python/paddle/fluid/layers/io.py

@@ -92,7 +92,7 @@ class ListenAndServ:
                         shape=[32, 32],
                         dtype='float32',
                         name="X")
-                    fluid.initializer.Constant(value=1.0)(x, main.global_block())
+                    paddle.nn.initializer.Constant(value=1.0)(x, main.global_block())
                     paddle.scale(x=x, scale=10.0, out=out_var)
 
             exe = fluid.Executor(place)

python/paddle/fluid/layers/nn.py

@@ -22,7 +22,6 @@ import numpy as np
 
 import paddle
 from ..layer_helper import LayerHelper
-from ..initializer import Normal, Constant
 from ..framework import (
     Variable,
     OpProtoHolder,
@@ -240,7 +239,7 @@ def embedding(
           w_param_attrs = fluid.ParamAttr(
               name="emb_weight",
               learning_rate=0.5,
-              initializer=fluid.initializer.NumpyArrayInitializer(weight_data),
+              initializer=paddle.nn.initializer.Assign(weight_data),
               trainable=True)
           emb_2 = fluid.layers.embedding(input=data, size=(128, 100), param_attr=w_param_attrs, dtype='float32')
     """
@@ -673,7 +672,10 @@ def autoincreased_step_counter(counter_name=None, begin=1, step=1):
     )
     if is_new_var:
         helper.set_variable_initializer(
-            counter, initializer=Constant(value=begin - 1, force_cpu=True)
+            counter,
+            initializer=paddle.nn.initializer.ConstantInitializer(
+                value=begin - 1, force_cpu=True
+            ),
         )
         helper.main_program.global_block()._prepend_op(
             type='increment',

python/paddle/fluid/metrics.py

@@ -19,7 +19,6 @@ import numpy as np
 import copy
 
 from .layer_helper import LayerHelper
-from .initializer import Constant
 from . import unique_name
 from .framework import Program, Variable, program_guard
 from . import layers

python/paddle/fluid/optimizer.py

@@ -39,7 +39,6 @@ from .backward import (
     _get_no_grad_set_name,
 )
 from .framework import program_guard
-from .initializer import Constant
 from .layer_helper import LayerHelper
 from .dygraph import base as imperative_base
 from .dygraph import no_grad
@@ -397,7 +396,8 @@ class Optimizer:
 
             lr_value = float(self._learning_rate())
             self.helper.set_variable_initializer(
-                lr_var, initializer=Constant(value=lr_value)
+                lr_var,
+                initializer=paddle.nn.initializer.Constant(value=lr_value),
             )
             return
 
@@ -713,7 +713,10 @@ class Optimizer:
             device = self._get_device_for_param(param.name)
         with device_guard(device):
             self.helper.set_variable_initializer(
-                var, initializer=Constant(value=float(fill_value))
+                var,
+                initializer=paddle.nn.initializer.Constant(
+                    value=float(fill_value)
+                ),
             )
 
         if in_dygraph_mode():
@@ -774,7 +777,10 @@ class Optimizer:
             device = 'cpu'
         with device_guard(device):
             self.helper.set_variable_initializer(
-                var, initializer=Constant(value=float(fill_value))
+                var,
+                initializer=paddle.nn.initializer.Constant(
+                    value=float(fill_value)
+                ),
             )
 
         if in_dygraph_mode():
@@ -1225,10 +1231,12 @@ class Optimizer:
         # NOTE(zhiqiu): the initializer should be set after coalesce_tensor op,
         # so the shape of flatten_param and flatten_grad will be inferred.
         self.helper.set_variable_initializer(
-            flatten_param, initializer=Constant(0.0)
+            flatten_param,
+            initializer=paddle.nn.initializer.Constant(0.0),
         )
         self.helper.set_variable_initializer(
-            flatten_grad, initializer=Constant(0.0)
+            flatten_grad,
+            initializer=paddle.nn.initializer.Constant(0.0),
         )
 
         return [(flatten_param, flatten_grad)]

python/paddle/fluid/param_attr.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .initializer import Initializer, Xavier, Constant
+import paddle
 from .regularizer import WeightDecayRegularizer
 from paddle.fluid.data_feeder import check_type
 
@@ -88,7 +88,10 @@ class ParamAttr:
         check_type(do_model_average, "do_model_average", (bool), "ParamAttr")
         check_type(need_clip, "need_clip", (bool), "ParamAttr")
         check_type(
-            initializer, "initializer", (Initializer, type(None)), "ParamAttr"
+            initializer,
+            "initializer",
+            (paddle.nn.initializer.Initializer, type(None)),
+            "ParamAttr",
         )
         check_type(
             regularizer,
@@ -139,7 +142,7 @@ class ParamAttr:
         Returns:
             None.
         """
-        self._set_default_initializer(Xavier())
+        self._set_default_initializer(paddle.nn.initializer.XavierUniform())
 
     def _set_default_bias_initializer(self):
         """
@@ -151,7 +154,7 @@ class ParamAttr:
         Returns:
             None.
         """
-        self._set_default_initializer(Constant(0.0))
+        self._set_default_initializer(paddle.nn.initializer.Constant(0.0))
 
     @staticmethod
     def _to_attr(arg):
@@ -177,7 +180,7 @@ class ParamAttr:
             return arg
         elif isinstance(arg, str):
             return ParamAttr(name=arg)
-        elif isinstance(arg, Initializer):
+        elif isinstance(arg, paddle.nn.initializer.Initializer):
             return ParamAttr(initializer=arg)
         elif isinstance(arg, WeightDecayRegularizer):
             return ParamAttr(regularizer=arg)

python/paddle/fluid/tests/unittests/auto_parallel_autoconvert.py

@@ -36,7 +36,6 @@ from paddle.distributed.auto_parallel.utils import (
     save_distributed_checkpoint,
 )
 from paddle.distributed.fleet import auto
-from paddle.fluid.initializer import NumpyArrayInitializer
 
 paddle.enable_static()
 _global_parallel_strategy = None
@@ -55,8 +54,12 @@ class MLPLayer(nn.Layer):
         np.random.seed(2021)
         arr0 = np.random.normal(0, 0.02, size=(d_model, dim_feedforward))
         arr1 = np.random.normal(0, 0.02, size=(d_model, dim_feedforward))
-        weight_attr0 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr0))
-        weight_attr1 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr1))
+        weight_attr0 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr0)
+        )
+        weight_attr1 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr1)
+        )
         bias_attr = None
         self.linear0 = nn.Linear(
             d_model, dim_feedforward, weight_attr0, bias_attr=bias_attr

python/paddle/fluid/tests/unittests/auto_parallel_save_load.py

@@ -30,7 +30,6 @@ from paddle.distributed.auto_parallel.utils import (
     save_distributed_checkpoint,
 )
 from paddle.distributed.fleet import auto
-from paddle.fluid.initializer import NumpyArrayInitializer
 
 paddle.enable_static()
 _global_parallel_strategy = None
@@ -48,7 +47,9 @@ class MLPLayer(nn.Layer):
         dim_feedforward = intermediate_size
         np.random.seed(2021)
         arr = np.random.normal(0, 0.02, size=(d_model, dim_feedforward))
-        weight_attr = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr))
+        weight_attr = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr)
+        )
         bias_attr = None
 
         self.linear0 = nn.Linear(

python/paddle/fluid/tests/unittests/collective/column_parallel_linear_api.py

@@ -38,15 +38,11 @@ class TestColumnParallelLinearAPI(TestCollectiveAPIRunnerBase):
             paddle.distributed.broadcast(data, src=0)
             if rank == 0:
                 param_attr = paddle.fluid.ParamAttr(
-                    initializer=paddle.fluid.initializer.NumpyArrayInitializer(
-                        np_array[:, 0:8]
-                    ),
+                    initializer=paddle.nn.initializer.Assign(np_array[:, 0:8]),
                 )
             else:
                 param_attr = paddle.fluid.ParamAttr(
-                    initializer=paddle.fluid.initializer.NumpyArrayInitializer(
-                        np_array[:, 8:16]
-                    ),
+                    initializer=paddle.nn.initializer.Assign(np_array[:, 8:16]),
                 )
 
             linear_out = paddle.distributed.split(

python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_transformer.py

@@ -242,10 +242,10 @@ class PrePostProcessLayer(Layer):
                 self._layer_norm = paddle.nn.LayerNorm(
                     normalized_shape=d_model,
                     weight_attr=fluid.ParamAttr(
-                        initializer=fluid.initializer.Constant(1.0)
+                        initializer=paddle.nn.initializer.Constant(1.0)
                     ),
                     bias_attr=fluid.ParamAttr(
-                        initializer=fluid.initializer.Constant(0.0)
+                        initializer=paddle.nn.initializer.Constant(0.0)
                     ),
                 )
 
@@ -513,7 +513,9 @@ class PrepareEncoderDecoderLayer(Layer):
             sparse=is_sparse,
             weight_attr=fluid.ParamAttr(
                 name=word_emb_param_name,
-                initializer=fluid.initializer.Normal(0.0, src_emb_dim**-0.5),
+                initializer=paddle.nn.initializer.Normal(
+                    0.0, src_emb_dim**-0.5
+                ),
             ),
         )
 
@@ -527,7 +529,7 @@ class PrepareEncoderDecoderLayer(Layer):
             sparse=is_sparse,
             weight_attr=fluid.ParamAttr(
                 name=pos_enc_param_name,
-                initializer=fluid.initializer.NumpyArrayInitializer(pos_inp),
+                initializer=paddle.nn.initializer.Assign(pos_inp),
                 trainable=False,
             ),
         )

python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist.py

@@ -39,7 +39,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -50,7 +50,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -65,7 +65,7 @@ def cnn_model(data):
             size=SIZE,
             activation="softmax",
             weight_attr=fluid.param_attr.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01)
+                initializer=paddle.nn.initializer.Constant(value=0.01)
             ),
         )
         # To cover @RENAMED@GRADIENT
@@ -74,7 +74,7 @@ def cnn_model(data):
             size=SIZE,
             activation="softmax",
             weight_attr=fluid.param_attr.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01)
+                initializer=paddle.nn.initializer.Constant(value=0.01)
             ),
         )
         predict += predict2

python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_multi_device.py

@@ -39,7 +39,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -50,7 +50,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -65,7 +65,7 @@ def cnn_model(data):
             size=SIZE,
             activation="softmax",
             weight_attr=fluid.param_attr.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01)
+                initializer=paddle.nn.initializer.Constant(value=0.01)
             ),
         )
         # To cover @RENAMED@GRADIENT
@@ -74,7 +74,7 @@ def cnn_model(data):
             size=SIZE,
             activation="softmax",
             weight_attr=fluid.param_attr.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01)
+                initializer=paddle.nn.initializer.Constant(value=0.01)
             ),
         )
         predict += predict2

python/paddle/fluid/tests/unittests/collective/fleet/pipeline_mnist_one_device.py

@@ -39,7 +39,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -50,7 +50,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -64,7 +64,7 @@ def cnn_model(data):
         size=SIZE,
         activation="softmax",
         weight_attr=fluid.param_attr.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     return predict

python/paddle/fluid/tests/unittests/collective/fleet/static_model_parallel_by_col.py

@@ -33,11 +33,9 @@ OUT_SIZE = 2 * MODEL_PARALLEL_SIZE
 
 def get_param_attr(weight, bias):
     weight_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(weight)
-    )
-    bias_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(bias)
+        initializer=paddle.nn.initializer.Assign(weight)
     )
+    bias_attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Assign(bias))
     return weight_attr, bias_attr
 
 

python/paddle/fluid/tests/unittests/collective/fleet/static_model_parallel_by_row.py

@@ -33,11 +33,9 @@ OUT_SIZE = 2 * MODEL_PARALLEL_SIZE
 
 def get_param_attr(weight, bias):
     weight_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(weight)
-    )
-    bias_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(bias)
+        initializer=paddle.nn.initializer.Assign(weight)
     )
+    bias_attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Assign(bias))
     return weight_attr, bias_attr
 
 
@@ -65,7 +63,7 @@ def create_model(data, rank):
             data,
             size=OUT_SIZE,
             weight_attr=paddle.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(np_weight)
+                initializer=paddle.nn.initializer.Assign(np_weight)
             ),
             bias_attr=bias_attr,
         )

python/paddle/fluid/tests/unittests/collective/fleet/static_model_parallel_embedding.py

@@ -44,9 +44,7 @@ def create_model(data, rank):
             axis=0,
             num_partitions=MODEL_PARALLEL_SIZE,
             weight_attr=paddle.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(
-                    np_weight_part
-                )
+                initializer=paddle.nn.initializer.Assign(np_weight_part)
             ),
             bias_attr=False,
         )
@@ -55,7 +53,7 @@ def create_model(data, rank):
             data,
             size=OUT_SIZE,
             weight_attr=paddle.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(np_weight)
+                initializer=paddle.nn.initializer.Assign(np_weight)
             ),
             bias_attr=False,
         )

python/paddle/fluid/tests/unittests/collective/multinode/dygraph_hybrid_dpppmp.py

@@ -35,7 +35,7 @@ def weight_init(mp, shape, col=True, seed=1024):
         else:
             step = shape[0] // mp.nranks
             _w = w[mp.rank * step : mp.rank * step + step, :]
-    return paddle.fluid.initializer.NumpyArrayInitializer(_w)
+    return paddle.nn.initializer.Assign(_w)
 
 
 class Criterion(nn.Layer):

python/paddle/fluid/tests/unittests/collective/multinode/dygraph_hybrid_fp16.py

@@ -35,7 +35,7 @@ def weight_init(mp, shape, col=True, seed=1024):
         else:
             step = shape[0] // mp.nranks
             _w = w[mp.rank * step : mp.rank * step + step, :]
-    return paddle.fluid.initializer.NumpyArrayInitializer(_w)
+    return paddle.nn.initializer.Assign(_w)
 
 
 class Criterion(nn.Layer):

python/paddle/fluid/tests/unittests/collective/multinode/dygraph_hybrid_recompute.py

@@ -36,7 +36,7 @@ def weight_init(mp, shape, col=True, seed=1024):
         else:
             step = shape[0] // mp.nranks
             _w = w[mp.rank * step : mp.rank * step + step, :]
-    return paddle.fluid.initializer.NumpyArrayInitializer(_w)
+    return paddle.nn.initializer.Assign(_w)
 
 
 class Criterion(nn.Layer):

python/paddle/fluid/tests/unittests/collective/parallel_embedding_api.py

@@ -42,13 +42,13 @@ class TestParallelEmbeddingAPI(TestCollectiveAPIRunnerBase):
             per_part_size = size[0] // 2
             if rank == 0:
                 param_attr = paddle.fluid.ParamAttr(
-                    initializer=paddle.fluid.initializer.NumpyArrayInitializer(
+                    initializer=paddle.nn.initializer.Assign(
                         np_array[0:per_part_size, :]
                     ),
                 )
             else:
                 param_attr = paddle.fluid.ParamAttr(
-                    initializer=paddle.fluid.initializer.NumpyArrayInitializer(
+                    initializer=paddle.nn.initializer.Assign(
                         np_array[per_part_size : size[0], :]
                     ),
                 )

python/paddle/fluid/tests/unittests/collective/row_parallel_linear_api.py

@@ -39,13 +39,13 @@ class TestRowParallelLinearAPI(TestCollectiveAPIRunnerBase):
             data = paddle.split(data, 2, axis=1)[rank]
             if rank == 0:
                 param_attr = paddle.fluid.ParamAttr(
-                    initializer=paddle.fluid.initializer.NumpyArrayInitializer(
+                    initializer=paddle.nn.initializer.Assign(
                         np_array[0:500, :]
                     ),
                 )
             else:
                 param_attr = paddle.fluid.ParamAttr(
-                    initializer=paddle.fluid.initializer.NumpyArrayInitializer(
+                    initializer=paddle.nn.initializer.Assign(
                         np_array[500:1000, :]
                     ),
                 )

python/paddle/fluid/tests/unittests/dist_allreduce_op.py

@@ -38,7 +38,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -49,7 +49,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -63,7 +63,7 @@ def cnn_model(data):
         size=SIZE,
         activation="softmax",
         weight_attr=fluid.param_attr.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     return predict

python/paddle/fluid/tests/unittests/dist_ctr.py

@@ -60,7 +60,7 @@ class TestDistCTR2x2(TestDistRunnerBase):
             size=[dnn_input_dim, dnn_layer_dims[0]],
             param_attr=fluid.ParamAttr(
                 name="deep_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=IS_SPARSE,
         )
@@ -74,7 +74,7 @@ class TestDistCTR2x2(TestDistRunnerBase):
                 size=dim,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(value=0.01)
+                    initializer=paddle.nn.initializer.Constant(value=0.01)
                 ),
                 name='dnn-fc-%d' % i,
             )
@@ -87,7 +87,7 @@ class TestDistCTR2x2(TestDistRunnerBase):
             size=[lr_input_dim, 1],
             param_attr=fluid.ParamAttr(
                 name="wide_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=IS_SPARSE,
         )

python/paddle/fluid/tests/unittests/dist_fleet_ctr.py

@@ -107,7 +107,7 @@ class TestDistCTR2x2(FleetDistRunnerBase):
             size=[dnn_input_dim, dnn_layer_dims[0]],
             param_attr=fluid.ParamAttr(
                 name="deep_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=True,
             padding_idx=0,
@@ -122,7 +122,7 @@ class TestDistCTR2x2(FleetDistRunnerBase):
                 size=dim,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(value=0.01)
+                    initializer=paddle.nn.initializer.Constant(value=0.01)
                 ),
                 name='dnn-fc-%d' % i,
             )
@@ -135,7 +135,7 @@ class TestDistCTR2x2(FleetDistRunnerBase):
             size=[lr_input_dim, 1],
             param_attr=fluid.ParamAttr(
                 name="wide_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=True,
             padding_idx=0,

python/paddle/fluid/tests/unittests/dist_fleet_heter_pipeline_ctr.py

@@ -78,7 +78,7 @@ class TestHeterPipelinePsCTR2x2(FleetDistHeterRunnerBase):
                 size=[dnn_input_dim, dnn_layer_dims[0]],
                 param_attr=fluid.ParamAttr(
                     name="deep_embedding",
-                    initializer=fluid.initializer.Constant(value=0.01),
+                    initializer=paddle.nn.initializer.Constant(value=0.01),
                 ),
                 is_sparse=True,
             )
@@ -94,7 +94,7 @@ class TestHeterPipelinePsCTR2x2(FleetDistHeterRunnerBase):
                 size=[lr_input_dim, 1],
                 param_attr=fluid.ParamAttr(
                     name="wide_embedding",
-                    initializer=fluid.initializer.Constant(value=0.01),
+                    initializer=paddle.nn.initializer.Constant(value=0.01),
                 ),
                 is_sparse=True,
             )
@@ -109,7 +109,7 @@ class TestHeterPipelinePsCTR2x2(FleetDistHeterRunnerBase):
                     size=dim,
                     activation="relu",
                     weight_attr=fluid.ParamAttr(
-                        initializer=fluid.initializer.Constant(value=0.01)
+                        initializer=paddle.nn.initializer.Constant(value=0.01)
                     ),
                     name='dnn-fc-%d' % i,
                 )

python/paddle/fluid/tests/unittests/dist_fleet_raw_program_optimizer.py

@@ -40,7 +40,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -51,7 +51,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -65,7 +65,7 @@ def cnn_model(data):
         size=SIZE,
         activation="softmax",
         weight_attr=fluid.param_attr.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     return predict

python/paddle/fluid/tests/unittests/dist_fleet_raw_program_optimizer_fuse_allreduce.py

@@ -40,7 +40,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -51,7 +51,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -65,7 +65,7 @@ def cnn_model(data):
         size=SIZE,
         activation="softmax",
         weight_attr=fluid.param_attr.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     return predict

python/paddle/fluid/tests/unittests/dist_fleet_simnet_bow.py

@@ -124,7 +124,8 @@ def train_network(
         is_distributed=is_distributed,
         size=[dict_dim, emb_dim],
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01), name="__emb__"
+            initializer=paddle.nn.initializer.Constant(value=0.01),
+            name="__emb__",
         ),
         is_sparse=is_sparse,
     )
@@ -137,7 +138,7 @@ def train_network(
         x=q_ss,
         size=hid_dim,
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01),
+            initializer=paddle.nn.initializer.Constant(value=0.01),
             name="__q_fc__",
             learning_rate=base_lr,
         ),
@@ -149,7 +150,7 @@ def train_network(
         is_distributed=is_distributed,
         size=[dict_dim, emb_dim],
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01),
+            initializer=paddle.nn.initializer.Constant(value=0.01),
             name="__emb__",
             learning_rate=emb_lr,
         ),
@@ -164,7 +165,8 @@ def train_network(
         x=pt_ss,
         size=hid_dim,
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01), name="__fc__"
+            initializer=paddle.nn.initializer.Constant(value=0.01),
+            name="__fc__",
         ),
         bias_attr=fluid.ParamAttr(name="__fc_b__"),
     )
@@ -175,7 +177,8 @@ def train_network(
         is_distributed=is_distributed,
         size=[dict_dim, emb_dim],
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01), name="__emb__"
+            initializer=paddle.nn.initializer.Constant(value=0.01),
+            name="__emb__",
         ),
         is_sparse=is_sparse,
     )
@@ -188,7 +191,8 @@ def train_network(
         x=nt_ss,
         size=hid_dim,
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01), name="__fc__"
+            initializer=paddle.nn.initializer.Constant(value=0.01),
+            name="__fc__",
         ),
         bias_attr=fluid.ParamAttr(name="__fc_b__"),
     )

python/paddle/fluid/tests/unittests/dist_fleet_sparse_embedding_ctr.py

@@ -86,11 +86,11 @@ class TestDistCTR2x2(FleetDistRunnerBase):
         inference = bool(int(os.getenv("INFERENCE", "0")))
 
         if initializer == 0:
-            init = fluid.initializer.Constant(value=0.01)
+            init = paddle.nn.initializer.Constant(value=0.01)
         elif initializer == 1:
-            init = fluid.initializer.Uniform()
+            init = paddle.nn.initializer.Uniform()
         elif initializer == 2:
-            init = fluid.initializer.Normal()
+            init = paddle.nn.initializer.Normal()
         else:
             raise ValueError("error initializer code: {}".format(initializer))
 
@@ -113,7 +113,7 @@ class TestDistCTR2x2(FleetDistRunnerBase):
                 size=dim,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(value=0.01)
+                    initializer=paddle.nn.initializer.Constant(value=0.01)
                 ),
                 name='dnn-fc-%d' % i,
             )
@@ -127,7 +127,7 @@ class TestDistCTR2x2(FleetDistRunnerBase):
             entry=entry,
             param_attr=fluid.ParamAttr(
                 name="wide_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
         )
 

python/paddle/fluid/tests/unittests/dist_mnist.py

@@ -39,7 +39,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     conv_pool_2 = fluid.nets.simple_img_conv_pool(
@@ -50,7 +50,7 @@ def cnn_model(data):
         pool_stride=2,
         act="relu",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -64,7 +64,7 @@ def cnn_model(data):
         size=SIZE,
         activation="softmax",
         weight_attr=fluid.param_attr.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
     return predict

python/paddle/fluid/tests/unittests/dist_se_resnext.py

@@ -121,7 +121,7 @@ class SE_ResNeXt:
             size=class_dim,
             activation='softmax',
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.05)
+                initializer=paddle.nn.initializer.Constant(value=0.05)
             ),
         )
         return out
@@ -174,7 +174,7 @@ class SE_ResNeXt:
             act=None,
             # avoid pserver CPU init differs from GPU
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.05)
+                initializer=paddle.nn.initializer.Constant(value=0.05)
             ),
             bias_attr=False,
         )
@@ -187,7 +187,7 @@ class SE_ResNeXt:
             x=pool,
             size=num_channels // reduction_ratio,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.05)
+                initializer=paddle.nn.initializer.Constant(value=0.05)
             ),
             activation='relu',
         )
@@ -196,7 +196,7 @@ class SE_ResNeXt:
             x=squeeze,
             size=num_channels,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.05)
+                initializer=paddle.nn.initializer.Constant(value=0.05)
             ),
             activation='sigmoid',
         )

python/paddle/fluid/tests/unittests/dist_text_classification.py

@@ -59,7 +59,7 @@ def conv_net(
         size=[dict_dim, emb_dim],
         is_sparse=False,
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -70,7 +70,7 @@ def conv_net(
         act="tanh",
         pool_type="max",
         param_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -78,7 +78,7 @@ def conv_net(
         x=[conv_3],
         size=fc0_dim,
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 
@@ -87,7 +87,7 @@ def conv_net(
         size=class_dim,
         activation="softmax",
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.01)
+            initializer=paddle.nn.initializer.Constant(value=0.01)
         ),
     )
 

python/paddle/fluid/tests/unittests/dist_transformer.py

@@ -29,7 +29,9 @@ import paddle.fluid as fluid
 import paddle.fluid.layers as layers
 import paddle.nn.functional as F
 
-const_para_attr = fluid.ParamAttr(initializer=fluid.initializer.Constant(0.001))
+const_para_attr = fluid.ParamAttr(
+    initializer=paddle.nn.initializer.Constant(0.001)
+)
 const_bias_attr = const_para_attr
 
 # Fix seed for test
@@ -1253,8 +1255,8 @@ def pre_post_process_layer(prev_out, out, process_cmd, dropout_rate=0.0):
             out = layers.layer_norm(
                 out,
                 begin_norm_axis=len(out.shape) - 1,
-                param_attr=fluid.initializer.Constant(1.0),
-                bias_attr=fluid.initializer.Constant(0.0),
+                param_attr=paddle.nn.initializer.Constant(1.0),
+                bias_attr=paddle.nn.initializer.Constant(0.0),
             )
         elif cmd == "d":  # add dropout
             if dropout_rate:
@@ -1292,7 +1294,7 @@ def prepare_encoder(
             size=[src_vocab_size, src_emb_dim],
             param_attr=fluid.ParamAttr(
                 name=word_emb_param_name,
-                initializer=fluid.initializer.ConstantInitializer(0.001),
+                initializer=paddle.nn.initializer.Constant(0.001),
             ),
         )
     else:
@@ -1301,7 +1303,9 @@ def prepare_encoder(
             size=[src_vocab_size, src_emb_dim],
             param_attr=fluid.ParamAttr(
                 name=word_emb_param_name,
-                initializer=fluid.initializer.Normal(0.0, src_emb_dim**-0.5),
+                initializer=paddle.nn.initializer.Normal(
+                    0.0, src_emb_dim**-0.5
+                ),
             ),
         )
 
@@ -1312,7 +1316,7 @@ def prepare_encoder(
         param_attr=fluid.ParamAttr(
             name=pos_enc_param_name,
             trainable=False,
-            initializer=fluid.initializer.ConstantInitializer(0.001),
+            initializer=paddle.nn.initializer.Constant(0.001),
         ),
     )
     src_pos_enc.stop_gradient = True

python/paddle/fluid/tests/unittests/dist_word2vec.py

@@ -41,7 +41,7 @@ class TestDistWord2vec2x2(TestDistRunnerBase):
                 is_sparse=IS_SPARSE,
                 param_attr=fluid.ParamAttr(
                     name='shared_w',
-                    initializer=fluid.initializer.Constant(value=0.1),
+                    initializer=paddle.nn.initializer.Constant(value=0.1),
                 ),
             )
             embed_second = fluid.layers.embedding(
@@ -51,7 +51,7 @@ class TestDistWord2vec2x2(TestDistRunnerBase):
                 is_sparse=IS_SPARSE,
                 param_attr=fluid.ParamAttr(
                     name='shared_w',
-                    initializer=fluid.initializer.Constant(value=0.1),
+                    initializer=paddle.nn.initializer.Constant(value=0.1),
                 ),
             )
             embed_third = fluid.layers.embedding(
@@ -61,7 +61,7 @@ class TestDistWord2vec2x2(TestDistRunnerBase):
                 is_sparse=IS_SPARSE,
                 param_attr=fluid.ParamAttr(
                     name='shared_w',
-                    initializer=fluid.initializer.Constant(value=0.1),
+                    initializer=paddle.nn.initializer.Constant(value=0.1),
                 ),
             )
             embed_forth = fluid.layers.embedding(
@@ -71,7 +71,7 @@ class TestDistWord2vec2x2(TestDistRunnerBase):
                 is_sparse=IS_SPARSE,
                 param_attr=fluid.ParamAttr(
                     name='shared_w',
-                    initializer=fluid.initializer.Constant(value=0.1),
+                    initializer=paddle.nn.initializer.Constant(value=0.1),
                 ),
             )
 
@@ -84,7 +84,7 @@ class TestDistWord2vec2x2(TestDistRunnerBase):
                 size=HIDDEN_SIZE,
                 activation='sigmoid',
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(value=0.1)
+                    initializer=paddle.nn.initializer.Constant(value=0.1)
                 ),
             )
             predict_word = paddle.static.nn.fc(
@@ -92,7 +92,7 @@ class TestDistWord2vec2x2(TestDistRunnerBase):
                 size=dict_size,
                 activation='softmax',
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(value=0.1)
+                    initializer=paddle.nn.initializer.Constant(value=0.1)
                 ),
             )
             cost = paddle.nn.functional.cross_entropy(

python/paddle/fluid/tests/unittests/distributed_passes/test_auto_parallel_gradient_merge_pass.py

@@ -26,7 +26,6 @@ import paddle.nn.functional as F
 import paddle.static as static
 import paddle.utils as utils
 from paddle.distributed.fleet import auto
-from paddle.fluid.initializer import NumpyArrayInitializer
 
 logging.getLogger().setLevel(logging.INFO)
 paddle.enable_static()
@@ -42,8 +41,12 @@ class MLPLayer(nn.Layer):
         np.random.seed(2021)
         arr0 = np.random.normal(0, 0.02, size=(d_model, dim_feedforward))
         arr1 = np.random.normal(0, 0.02, size=(dim_feedforward, d_model))
-        weight_attr0 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr0))
-        weight_attr1 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr1))
+        weight_attr0 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr0)
+        )
+        weight_attr1 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr1)
+        )
         bias_attr = None
         self.linear0 = nn.Linear(
             d_model, dim_feedforward, weight_attr0, bias_attr=bias_attr

python/paddle/fluid/tests/unittests/dygraph_to_static/bert_dygraph_model.py

@@ -203,8 +203,8 @@ class BertModelLayer(Layer):
         self._sent_emb_name = "sent_embedding"
         self._dtype = "float16" if use_fp16 else "float32"
 
-        self._param_initializer = fluid.initializer.TruncatedNormal(
-            scale=config['initializer_range']
+        self._param_initializer = paddle.nn.initializer.TruncatedNormal(
+            std=config['initializer_range']
         )
         paddle.set_default_dtype(self._dtype)
         self._src_emb = paddle.nn.Embedding(
@@ -317,8 +317,8 @@ class PretrainModelLayer(Layer):
         self._prepostprocess_dropout = config['hidden_dropout_prob']
 
         self._word_emb_name = "word_embedding"
-        self._param_initializer = fluid.initializer.TruncatedNormal(
-            scale=config['initializer_range']
+        self._param_initializer = paddle.nn.initializer.TruncatedNormal(
+            std=config['initializer_range']
         )
         self._weight_sharing = weight_sharing
         self.use_fp16 = use_fp16
@@ -343,7 +343,7 @@ class PretrainModelLayer(Layer):
 
         self.mask_lm_out_bias_attr = fluid.ParamAttr(
             name="mask_lm_out_fc.b_0",
-            initializer=fluid.initializer.Constant(value=0.0),
+            initializer=paddle.nn.initializer.Constant(value=0.0),
         )
 
         if not self._weight_sharing:

python/paddle/fluid/tests/unittests/dygraph_to_static/darknet.py

@@ -41,7 +41,7 @@ class ConvBNLayer(fluid.dygraph.Layer):
             padding=padding,
             groups=groups,
             weight_attr=ParamAttr(
-                initializer=fluid.initializer.Normal(0.0, 0.02)
+                initializer=paddle.nn.initializer.Normal(0.0, 0.02)
             ),
             bias_attr=False,
         )
@@ -49,11 +49,11 @@ class ConvBNLayer(fluid.dygraph.Layer):
             num_channels=ch_out,
             is_test=is_test,
             param_attr=ParamAttr(
-                initializer=fluid.initializer.Normal(0.0, 0.02),
+                initializer=paddle.nn.initializer.Normal(0.0, 0.02),
                 regularizer=L2Decay(0.0),
             ),
             bias_attr=ParamAttr(
-                initializer=fluid.initializer.Constant(0.0),
+                initializer=paddle.nn.initializer.Constant(0.0),
                 regularizer=L2Decay(0.0),
             ),
         )

python/paddle/fluid/tests/unittests/dygraph_to_static/seq2seq_dygraph_model.py

@@ -26,10 +26,8 @@ from paddle.nn import Embedding
 
 INF = 1.0 * 1e5
 alpha = 0.6
-uniform_initializer = lambda x: fluid.initializer.UniformInitializer(
-    low=-x, high=x
-)
-zero_constant = fluid.initializer.Constant(0.0)
+uniform_initializer = lambda x: paddle.nn.initializer.Uniform(low=-x, high=x)
+zero_constant = paddle.nn.initializer.Constant(0.0)
 
 
 class BasicLSTMUnit(Layer):

python/paddle/fluid/tests/unittests/dygraph_to_static/simnet_dygraph_model.py

@@ -17,9 +17,9 @@ from functools import reduce
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.param_attr as attr
+from paddle.common_ops_import import Variable
 from paddle.fluid.dygraph import Layer
 from paddle.jit.api import to_static
-from paddle.static import Variable
 
 
 class EmbeddingLayer:
@@ -48,7 +48,8 @@ class EmbeddingLayer:
             sparse=True,
             padding_idx=self.padding_idx,
             weight_attr=attr.ParamAttr(
-                name=self.name, initializer=fluid.initializer.Xavier()
+                name=self.name,
+                initializer=paddle.nn.initializer.XavierUniform(),
             ),
         )
 

python/paddle/fluid/tests/unittests/dygraph_to_static/simnet_dygraph_model_v2.py

@@ -15,7 +15,7 @@
 from functools import reduce
 
 import paddle
-from paddle.static import Variable
+from paddle.common_ops_import import Variable
 
 
 class EmbeddingLayer:

python/paddle/fluid/tests/unittests/dygraph_to_static/test_basic_api_transformation.py

@@ -118,10 +118,10 @@ def dyfunc_BilinearTensorProduct(layer1, layer2):
         4,
         1000,
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.99)
+            initializer=paddle.nn.initializer.Constant(value=0.99)
         ),
         bias_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.5)
+            initializer=paddle.nn.initializer.Constant(value=0.5)
         ),
     )
 
@@ -138,10 +138,10 @@ def dyfunc_Conv2D(input):
         out_channels=2,
         kernel_size=3,
         weight_attr=paddle.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.99)
+            initializer=paddle.nn.initializer.Constant(value=0.99)
         ),
         bias_attr=paddle.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.5)
+            initializer=paddle.nn.initializer.Constant(value=0.5)
         ),
     )
     res = conv2d(input)
@@ -170,10 +170,10 @@ def dyfunc_Conv2DTranspose(input):
         12,
         12,
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.99)
+            initializer=paddle.nn.initializer.Constant(value=0.99)
         ),
         bias_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.5)
+            initializer=paddle.nn.initializer.Constant(value=0.5)
         ),
     )
     ret = conv2dTranspose(input)
@@ -222,7 +222,7 @@ def dyfunc_Pool2D(input):
 def dyfunc_Prelu(input):
     prelu0 = paddle.nn.PReLU(
         weight_attr=fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(1.0)
+            initializer=paddle.nn.initializer.Constant(1.0)
         ),
     )
     res = prelu0(input)

python/paddle/fluid/tests/unittests/dygraph_to_static/test_bmn.py

@@ -115,11 +115,11 @@ class Conv1D(fluid.dygraph.Layer):
         k = 1.0 / math.sqrt(fan_in)
         param_attr = ParamAttr(
             name=prefix + "_w",
-            initializer=fluid.initializer.Uniform(low=-k, high=k),
+            initializer=paddle.nn.initializer.Uniform(low=-k, high=k),
         )
         bias_attr = ParamAttr(
             name=prefix + "_b",
-            initializer=fluid.initializer.Uniform(low=-k, high=k),
+            initializer=paddle.nn.initializer.Uniform(low=-k, high=k),
         )
 
         self._conv2d = paddle.nn.Conv2D(

python/paddle/fluid/tests/unittests/dygraph_to_static/test_convert_call.py

@@ -125,10 +125,10 @@ class MyConvLayer(fluid.dygraph.Layer):
             out_channels=2,
             kernel_size=3,
             weight_attr=paddle.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.99)
+                initializer=paddle.nn.initializer.Constant(value=0.99)
             ),
             bias_attr=paddle.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.5)
+                initializer=paddle.nn.initializer.Constant(value=0.5)
             ),
         )
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_cycle_gan.py

@@ -352,7 +352,7 @@ class conv2d(fluid.dygraph.Layer):
             con_bias_attr = False
         else:
             con_bias_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(0.0)
+                initializer=paddle.nn.initializer.Constant(0.0)
             )
 
         self.conv = paddle.nn.Conv2D(
@@ -362,9 +362,7 @@ class conv2d(fluid.dygraph.Layer):
             stride=stride,
             padding=padding,
             weight_attr=paddle.ParamAttr(
-                initializer=fluid.initializer.NormalInitializer(
-                    loc=0.0, scale=stddev
-                )
+                initializer=paddle.nn.initializer.Normal(mean=0.0, std=stddev)
             ),
             bias_attr=con_bias_attr,
         )
@@ -378,10 +376,10 @@ class conv2d(fluid.dygraph.Layer):
                 use_global_stats=True,  # set True to use deterministic algorithm
                 num_channels=num_filters,
                 param_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.NormalInitializer(1.0, 0.02)
+                    initializer=paddle.nn.initializer.Normal(1.0, 0.02)
                 ),
                 bias_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(0.0)
+                    initializer=paddle.nn.initializer.Constant(0.0)
                 ),
                 trainable_statistics=True,
             )
@@ -421,7 +419,7 @@ class DeConv2D(fluid.dygraph.Layer):
             de_bias_attr = False
         else:
             de_bias_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(0.0)
+                initializer=paddle.nn.initializer.Constant(0.0)
             )
 
         self._deconv = paddle.nn.Conv2DTranspose(
@@ -431,9 +429,7 @@ class DeConv2D(fluid.dygraph.Layer):
             stride=stride,
             padding=padding,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.NormalInitializer(
-                    loc=0.0, scale=stddev
-                )
+                initializer=paddle.nn.initializer.Normal(mean=0.0, std=stddev)
             ),
             bias_attr=de_bias_attr,
         )
@@ -444,10 +440,10 @@ class DeConv2D(fluid.dygraph.Layer):
                 use_global_stats=True,  # set True to use deterministic algorithm
                 num_channels=num_filters,
                 param_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.NormalInitializer(1.0, 0.02)
+                    initializer=paddle.nn.initializer.Normal(1.0, 0.02)
                 ),
                 bias_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(0.0)
+                    initializer=paddle.nn.initializer.Constant(0.0)
                 ),
                 trainable_statistics=True,
             )

python/paddle/fluid/tests/unittests/dygraph_to_static/test_lac.py

@@ -98,7 +98,7 @@ class BiGRU(fluid.dygraph.Layer):
             in_features=input_dim,
             out_features=grnn_hidden_dim * 3,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-init_bound, high=init_bound
                 ),
                 regularizer=fluid.regularizer.L2DecayRegularizer(
@@ -111,7 +111,7 @@ class BiGRU(fluid.dygraph.Layer):
             size=grnn_hidden_dim,
             h_0=h_0,
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-init_bound, high=init_bound
                 ),
                 regularizer=fluid.regularizer.L2DecayRegularizer(
@@ -124,7 +124,7 @@ class BiGRU(fluid.dygraph.Layer):
             in_features=input_dim,
             out_features=grnn_hidden_dim * 3,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-init_bound, high=init_bound
                 ),
                 regularizer=fluid.regularizer.L2DecayRegularizer(
@@ -138,7 +138,7 @@ class BiGRU(fluid.dygraph.Layer):
             is_reverse=True,
             h_0=h_0,
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-init_bound, high=init_bound
                 ),
                 regularizer=fluid.regularizer.L2DecayRegularizer(
@@ -375,7 +375,7 @@ class LexNet(fluid.dygraph.Layer):
             weight_attr=fluid.ParamAttr(
                 learning_rate=self.emb_lr,
                 name="word_emb",
-                initializer=fluid.initializer.Uniform(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-self.init_bound, high=self.init_bound
                 ),
             ),
@@ -415,7 +415,7 @@ class LexNet(fluid.dygraph.Layer):
             in_features=self.grnn_hidden_dim * 2,
             out_features=self.num_labels,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-self.init_bound, high=self.init_bound
                 ),
                 regularizer=fluid.regularizer.L2DecayRegularizer(

python/paddle/fluid/tests/unittests/dygraph_to_static/test_mobile_net.py

@@ -22,7 +22,6 @@ from predictor_utils import PredictorTools
 
 import paddle
 import paddle.fluid as fluid
-from paddle.fluid.initializer import MSRA
 from paddle.fluid.param_attr import ParamAttr
 from paddle.jit.api import to_static
 from paddle.jit.translated_layer import INFER_MODEL_SUFFIX, INFER_PARAMS_SUFFIX
@@ -61,7 +60,8 @@ class ConvBNLayer(fluid.dygraph.Layer):
             padding=padding,
             groups=num_groups,
             weight_attr=ParamAttr(
-                initializer=MSRA(), name=self.full_name() + "_weights"
+                initializer=paddle.nn.initializer.KaimingUniform(),
+                name=self.full_name() + "_weights",
             ),
             bias_attr=False,
         )
@@ -259,7 +259,8 @@ class MobileNetV1(fluid.dygraph.Layer):
             int(1024 * scale),
             class_dim,
             weight_attr=ParamAttr(
-                initializer=MSRA(), name=self.full_name() + "fc7_weights"
+                initializer=paddle.nn.initializer.KaimingUniform(),
+                name=self.full_name() + "fc7_weights",
             ),
             bias_attr=ParamAttr(name="fc7_offset"),
         )

python/paddle/fluid/tests/unittests/dygraph_to_static/test_ptb_lm.py

@@ -49,26 +49,26 @@ class SimpleLSTMRNN(fluid.Layer):
         for i in range(self._num_layers):
             weight_1 = self.create_parameter(
                 attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.UniformInitializer(
+                    initializer=paddle.nn.initializer.Uniform(
                         low=-self._init_scale, high=self._init_scale
                     )
                 ),
                 shape=[self._hidden_size * 2, self._hidden_size * 4],
                 dtype="float32",
-                default_initializer=fluid.initializer.UniformInitializer(
+                default_initializer=paddle.nn.initializer.Uniform(
                     low=-self._init_scale, high=self._init_scale
                 ),
             )
             self.weight_1_arr.append(self.add_parameter('w_%d' % i, weight_1))
             bias_1 = self.create_parameter(
                 attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.UniformInitializer(
+                    initializer=paddle.nn.initializer.Uniform(
                         low=-self._init_scale, high=self._init_scale
                     )
                 ),
                 shape=[self._hidden_size * 4],
                 dtype="float32",
-                default_initializer=fluid.initializer.Constant(0.0),
+                default_initializer=paddle.nn.initializer.Constant(0.0),
             )
             self.bias_arr.append(self.add_parameter('b_%d' % i, bias_1))
 
@@ -158,7 +158,7 @@ class PtbModel(fluid.Layer):
             sparse=False,
             weight_attr=fluid.ParamAttr(
                 name='embedding_para',
-                initializer=fluid.initializer.UniformInitializer(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-init_scale, high=init_scale
                 ),
             ),
@@ -167,7 +167,7 @@ class PtbModel(fluid.Layer):
             attr=fluid.ParamAttr(),
             shape=[self.hidden_size, self.vocab_size],
             dtype="float32",
-            default_initializer=fluid.initializer.UniformInitializer(
+            default_initializer=paddle.nn.initializer.Uniform(
                 low=-self.init_scale, high=self.init_scale
             ),
         )
@@ -175,7 +175,7 @@ class PtbModel(fluid.Layer):
             attr=fluid.ParamAttr(),
             shape=[self.vocab_size],
             dtype="float32",
-            default_initializer=fluid.initializer.UniformInitializer(
+            default_initializer=paddle.nn.initializer.Uniform(
                 low=-self.init_scale, high=self.init_scale
             ),
         )

python/paddle/fluid/tests/unittests/dygraph_to_static/test_resnet.py

@@ -194,7 +194,7 @@ class ResNet(fluid.dygraph.Layer):
             self.pool2d_avg_output,
             class_dim,
             weight_attr=fluid.param_attr.ParamAttr(
-                initializer=fluid.initializer.Uniform(-stdv, stdv)
+                initializer=paddle.nn.initializer.Uniform(-stdv, stdv)
             ),
         )
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_se_resnet.py

@@ -131,7 +131,7 @@ class SqueezeExcitation(fluid.dygraph.Layer):
             num_channels,
             num_channels // reduction_ratio,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(-stdv, stdv)
+                initializer=paddle.nn.initializer.Uniform(-stdv, stdv)
             ),
         )
         stdv = 1.0 / math.sqrt(num_channels / 16.0 * 1.0)
@@ -139,7 +139,7 @@ class SqueezeExcitation(fluid.dygraph.Layer):
             num_channels // reduction_ratio,
             num_channels,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Uniform(-stdv, stdv)
+                initializer=paddle.nn.initializer.Uniform(-stdv, stdv)
             ),
         )
 
@@ -316,7 +316,7 @@ class SeResNeXt(fluid.dygraph.Layer):
             self.pool2d_avg_output,
             class_dim,
             weight_attr=fluid.param_attr.ParamAttr(
-                initializer=fluid.initializer.Uniform(-stdv, stdv)
+                initializer=paddle.nn.initializer.Uniform(-stdv, stdv)
             ),
         )
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_word2vec.py

@@ -230,7 +230,7 @@ class SkipGram(fluid.dygraph.Layer):
             self.embedding_size,
             weight_attr=fluid.ParamAttr(
                 name='embedding_para',
-                initializer=fluid.initializer.UniformInitializer(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-0.5 / self.embedding_size,
                     high=0.5 / self.embedding_size,
                 ),
@@ -242,7 +242,7 @@ class SkipGram(fluid.dygraph.Layer):
             self.embedding_size,
             weight_attr=fluid.ParamAttr(
                 name='embedding_out_para',
-                initializer=fluid.initializer.UniformInitializer(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-0.5 / self.embedding_size,
                     high=0.5 / self.embedding_size,
                 ),

python/paddle/fluid/tests/unittests/dygraph_to_static/transformer_dygraph_model.py

@@ -62,10 +62,10 @@ class PrePostProcessLayer(Layer):
                         paddle.nn.LayerNorm(
                             normalized_shape=d_model,
                             weight_attr=fluid.ParamAttr(
-                                initializer=fluid.initializer.Constant(1.0)
+                                initializer=paddle.nn.initializer.Constant(1.0)
                             ),
                             bias_attr=fluid.ParamAttr(
-                                initializer=fluid.initializer.Constant(0.0)
+                                initializer=paddle.nn.initializer.Constant(0.0)
                             ),
                         ),
                     )
@@ -295,7 +295,7 @@ class Embedder(Layer):
             vocab_size,
             emb_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Normal(0.0, emb_dim**-0.5)
+                initializer=paddle.nn.initializer.Normal(0.0, emb_dim**-0.5)
             ),
         )
 
@@ -330,7 +330,7 @@ class WrapEncoder(Layer):
             max_length,
             self.emb_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(
+                initializer=paddle.nn.initializer.Assign(
                     position_encoding_init(max_length, self.emb_dim)
                 ),
                 trainable=False,
@@ -522,7 +522,7 @@ class WrapDecoder(Layer):
             max_length,
             self.emb_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.NumpyArrayInitializer(
+                initializer=paddle.nn.initializer.Assign(
                     position_encoding_init(max_length, self.emb_dim)
                 ),
                 trainable=False,

python/paddle/fluid/tests/unittests/dygraph_to_static/yolov3.py

@@ -253,10 +253,10 @@ class YOLOv3(fluid.dygraph.Layer):
                     stride=1,
                     padding=0,
                     weight_attr=ParamAttr(
-                        initializer=fluid.initializer.Normal(0.0, 0.02)
+                        initializer=paddle.nn.initializer.Normal(0.0, 0.02)
                     ),
                     bias_attr=ParamAttr(
-                        initializer=fluid.initializer.Constant(0.0),
+                        initializer=paddle.nn.initializer.Constant(0.0),
                         regularizer=L2Decay(0.0),
                     ),
                 ),

python/paddle/fluid/tests/unittests/fleet_heter_ps_training.py

@@ -70,7 +70,7 @@ def net(batch_size=4, lr=0.01):
             size=[dnn_input_dim, dnn_layer_dims[0]],
             param_attr=fluid.ParamAttr(
                 name="deep_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=True,
         )
@@ -86,7 +86,7 @@ def net(batch_size=4, lr=0.01):
             size=[lr_input_dim, 1],
             param_attr=fluid.ParamAttr(
                 name="wide_embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=True,
         )
@@ -99,7 +99,7 @@ def net(batch_size=4, lr=0.01):
                 size=dim,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(value=0.01)
+                    initializer=paddle.nn.initializer.Constant(value=0.01)
                 ),
                 name='dnn-fc-%d' % i,
             )

python/paddle/fluid/tests/unittests/ir/inference/program_config.py

@@ -27,7 +27,6 @@ from paddle.fluid.framework import (
     Operator,
     convert_np_dtype_to_dtype_,
 )
-from paddle.fluid.initializer import NumpyArrayInitializer
 from paddle.static.quantization import (
     QuantizationFreezePass,
     QuantizationTransformPass,
@@ -305,7 +304,7 @@ def create_fake_model(program_config):
             shape=tensor_config.shape,
             type=core.VarDesc.VarType.LOD_TENSOR,
             name=name,
-            initializer=NumpyArrayInitializer(tensor_config.data),
+            initializer=paddle.nn.initializer.Assign(tensor_config.data),
         )
     in_vars = []
     for name in sorted(save_var_map.keys()):

python/paddle/fluid/tests/unittests/ir/inference/test_mkldnn_conv_bias_fuse_pass.py

@@ -30,7 +30,7 @@ class ConvBiasMkldnnFusePassSamePadTest(InferencePassTest):
                 name="data", shape=[-1, 3, 100, 100], dtype="float32"
             )
             param_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Xavier(uniform=False),
+                initializer=paddle.nn.initializer.XavierNormal(),
                 learning_rate=0.001,
             )
             conv_out = paddle.static.nn.conv2d(
@@ -63,7 +63,7 @@ class ConvBiasMkldnnFusePassValidPadTest(ConvBiasMkldnnFusePassSamePadTest):
                 name="data", shape=[-1, 3, 100, 100], dtype="float32"
             )
             param_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Xavier(uniform=False),
+                initializer=paddle.nn.initializer.XavierNormal(),
                 learning_rate=0.001,
             )
             conv_out = paddle.static.nn.conv2d(
@@ -89,7 +89,7 @@ class ConvBiasMkldnnFusePassExplictPadTest(ConvBiasMkldnnFusePassSamePadTest):
                 name="data", shape=[-1, 3, 100, 100], dtype="float32"
             )
             param_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Xavier(uniform=False),
+                initializer=paddle.nn.initializer.XavierNormal(),
                 learning_rate=0.001,
             )
             conv_out = paddle.static.nn.conv2d(
@@ -114,7 +114,7 @@ class ConvBiasMkldnnFusePassGroupTest(ConvBiasMkldnnFusePassSamePadTest):
                 name="data", shape=[-1, 3, 100, 100], dtype="float32"
             )
             param_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Xavier(uniform=False),
+                initializer=paddle.nn.initializer.XavierUniform(),
                 learning_rate=0.001,
             )
             conv_out = paddle.static.nn.conv2d(
@@ -145,7 +145,7 @@ class ConvBiasMkldnnFusePassDialtionsGroupsTest(
                 name="data", shape=[-1, 3, 100, 100], dtype="float32"
             )
             param_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Xavier(uniform=False),
+                initializer=paddle.nn.initializer.XavierNormal(),
                 learning_rate=0.001,
             )
             conv_out = paddle.static.nn.conv2d(
@@ -173,7 +173,7 @@ class ConvTransposeMkldnnFusePassDialtionsGroupsTest(InferencePassTest):
         with fluid.program_guard(self.main_program, self.startup_program):
             data = fluid.data(name="data", shape=[-1, 3, 5, 5], dtype="float32")
             param_attr = fluid.ParamAttr(
-                initializer=fluid.initializer.Xavier(uniform=False),
+                initializer=paddle.nn.initializer.XavierNormal(),
                 learning_rate=0.001,
             )
             conv_out = paddle.static.nn.conv2d_transpose(

python/paddle/fluid/tests/unittests/ir/inference/test_trt_subgraph_pass.py

@@ -180,11 +180,11 @@ class TensorRTSubgraphPassInstanceNormTest(InferencePassTest):
             )
             param_attr = fluid.ParamAttr(
                 name='instance_norm_w',
-                initializer=fluid.initializer.Constant(value=1.0),
+                initializer=paddle.nn.initializer.Constant(value=1.0),
             )
             bias_attr = fluid.ParamAttr(
                 name='instance_norm_b',
-                initializer=fluid.initializer.Constant(value=0.0),
+                initializer=paddle.nn.initializer.Constant(value=0.0),
             )
             out = paddle.static.nn.instance_norm(
                 input=data, param_attr=param_attr, bias_attr=bias_attr

python/paddle/fluid/tests/unittests/mlu/test_batch_norm_op_mlu_v2.py

@@ -114,11 +114,11 @@ class TestBatchNorm(unittest.TestCase):
                         shape[1],
                         is_test=is_test,
                         param_attr=fluid.ParamAttr(
-                            initializer=fluid.initializer.Constant(1.0),
+                            initializer=paddle.nn.initializer.Constant(1.0),
                             trainable=False,
                         ),
                         bias_attr=fluid.ParamAttr(
-                            initializer=fluid.initializer.Constant(0.0),
+                            initializer=paddle.nn.initializer.Constant(0.0),
                             trainable=False,
                         ),
                         trainable_statistics=trainable_statistics,
@@ -262,7 +262,7 @@ class TestBatchNormUseGlobalStats(unittest.TestCase):
                 net1 = paddle.nn.BatchNorm(
                     6,
                     param_attr=fluid.ParamAttr(
-                        initializer=fluid.initializer.Constant(1.0)
+                        initializer=paddle.nn.initializer.Constant(1.0)
                     ),
                     use_global_stats=self.use_global_stats,
                     trainable_statistics=self.trainable_statistics,

python/paddle/fluid/tests/unittests/npu/test_adam_op_npu.py

@@ -322,12 +322,12 @@ class TestNetWithEpsilonTensor(unittest.TestCase):
         label_np = np.random.randint(2, size=(2, 1)).astype('int64')
         weight_attr1 = paddle.ParamAttr(
             name="weight1",
-            initializer=fluid.initializer.Constant(value=1.0),
+            initializer=paddle.nn.initializer.Constant(value=1.0),
             trainable=True,
         )
         weight_attr2 = paddle.ParamAttr(
             name="weight2",
-            initializer=fluid.initializer.Constant(value=2.0),
+            initializer=paddle.nn.initializer.Constant(value=2.0),
             trainable=True,
         )
         clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=1.0)

python/paddle/fluid/tests/unittests/npu/test_run_program_op_npu.py

@@ -298,7 +298,7 @@ class TestRunProgramOpWithFC(RunProgramNPUOpTest):
         weight_attr = fluid.ParamAttr(
             name=self.input_names['Params'][0],
             learning_rate=0.5,
-            initializer=fluid.initializer.NumpyArrayInitializer(
+            initializer=paddle.nn.initializer.Assign(
                 self.inputs['Params'][self.input_names['Params'][0]]
             ),
             trainable=True,
@@ -306,7 +306,7 @@ class TestRunProgramOpWithFC(RunProgramNPUOpTest):
         bias_attr = fluid.ParamAttr(
             name=self.input_names['Params'][1],
             learning_rate=0.5,
-            initializer=fluid.initializer.NumpyArrayInitializer(
+            initializer=paddle.nn.initializer.Assign(
                 self.inputs['Params'][self.input_names['Params'][1]]
             ),
             trainable=True,

python/paddle/fluid/tests/unittests/parallel_dygraph_sparse_embedding.py

@@ -41,7 +41,7 @@ class SimpleNet(fluid.Layer):
             self.hidden_size,
             sparse=is_sparse,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.UniformInitializer(
+                initializer=paddle.nn.initializer.Uniform(
                     low=-init_scale, high=init_scale
                 )
             ),
@@ -50,7 +50,7 @@ class SimpleNet(fluid.Layer):
             attr=fluid.ParamAttr(),
             shape=[self.hidden_size, self.vocab_size],
             dtype=dtype,
-            default_initializer=fluid.initializer.UniformInitializer(
+            default_initializer=paddle.nn.initializer.Uniform(
                 low=-self.init_scale, high=self.init_scale
             ),
         )
@@ -58,7 +58,7 @@ class SimpleNet(fluid.Layer):
             attr=fluid.ParamAttr(),
             shape=[self.vocab_size],
             dtype=dtype,
-            default_initializer=fluid.initializer.UniformInitializer(
+            default_initializer=paddle.nn.initializer.Uniform(
                 low=-self.init_scale, high=self.init_scale
             ),
         )

python/paddle/fluid/tests/unittests/simple_nets.py

@@ -26,7 +26,7 @@ def simple_fc_net_with_inputs(img, label, class_num=10):
             size=100,
             activation='relu',
             bias_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=1.0)
+                initializer=paddle.nn.initializer.Constant(value=1.0)
             ),
         )
     prediction = paddle.static.nn.fc(
@@ -53,7 +53,7 @@ def batchnorm_fc_with_inputs(img, label, class_num=10):
             size=200,
             activation='relu',
             bias_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=1.0)
+                initializer=paddle.nn.initializer.Constant(value=1.0)
             ),
         )
 

python/paddle/fluid/tests/unittests/static_model_parallel_fused_attention.py

@@ -25,11 +25,9 @@ paddle.enable_static()
 
 def get_param_attr(weight, bias):
     weight_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(weight)
-    )
-    bias_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(bias)
+        initializer=paddle.nn.initializer.Assign(weight)
     )
+    bias_attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Assign(bias))
     return weight_attr, bias_attr
 
 

python/paddle/fluid/tests/unittests/static_model_parallel_fused_feedforward.py

@@ -30,11 +30,9 @@ OUT_SIZE = 2 * MODEL_PARALLEL_SIZE
 
 def get_param_attr(weight, bias):
     weight_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(weight)
-    )
-    bias_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(bias)
+        initializer=paddle.nn.initializer.Assign(weight)
     )
+    bias_attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Assign(bias))
     return weight_attr, bias_attr
 
 

python/paddle/fluid/tests/unittests/static_model_parallel_fused_multi_transformer.py

@@ -25,11 +25,9 @@ paddle.enable_static()
 
 def get_param_attr(weight, bias):
     weight_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(weight)
-    )
-    bias_attr = paddle.ParamAttr(
-        initializer=fluid.initializer.NumpyArrayInitializer(bias)
+        initializer=paddle.nn.initializer.Assign(weight)
     )
+    bias_attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Assign(bias))
     return weight_attr, bias_attr
 
 

python/paddle/fluid/tests/unittests/test_adam_op.py

@@ -767,12 +767,12 @@ class TestAdamOptimizer(unittest.TestCase):
         label_np = np.random.randint(2, size=(2, 1)).astype('int64')
         weight_attr1 = paddle.ParamAttr(
             name="weight1",
-            initializer=fluid.initializer.Constant(value=1.0),
+            initializer=paddle.nn.initializer.Constant(value=1.0),
             trainable=True,
         )
         weight_attr2 = paddle.ParamAttr(
             name="weight2",
-            initializer=fluid.initializer.Constant(value=2.0),
+            initializer=paddle.nn.initializer.Constant(value=2.0),
             trainable=True,
         )
         clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=1.0)
@@ -926,7 +926,7 @@ class TestAdamOptimizer(unittest.TestCase):
         main = fluid.Program()
         weight_attr = paddle.ParamAttr(
             name="weight1",
-            initializer=fluid.initializer.Constant(value=1.0),
+            initializer=paddle.nn.initializer.Constant(value=1.0),
             regularizer=fluid.regularizer.L1DecayRegularizer(
                 regularization_coeff=0.1
             ),

python/paddle/fluid/tests/unittests/test_auto_parallel_mapper.py

@@ -39,7 +39,6 @@ from paddle.distributed.auto_parallel.partitioner import Partitioner
 from paddle.distributed.auto_parallel.reshard import Resharder
 from paddle.distributed.fleet import auto
 from paddle.fluid import core
-from paddle.fluid.initializer import NumpyArrayInitializer
 
 if os.getenv("CUDA_VISIBLE_DEVICES") is not None:
     os.environ["CUDA_VISIBLE_DEVICES"] = ""
@@ -373,10 +372,18 @@ class MLPLayer(nn.Layer):
         arr1 = np.random.normal(0, 0.02, size=(dim_feedforward, d_model))
         arr2 = np.random.normal(0, 0.02, size=(d_model, dim_feedforward))
         arr3 = np.random.normal(0, 0.02, size=(dim_feedforward, d_model))
-        weight_attr0 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr0))
-        weight_attr1 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr1))
-        weight_attr2 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr2))
-        weight_attr3 = paddle.ParamAttr(initializer=NumpyArrayInitializer(arr3))
+        weight_attr0 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr0)
+        )
+        weight_attr1 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr1)
+        )
+        weight_attr2 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr2)
+        )
+        weight_attr3 = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.Assign(arr3)
+        )
         bias_attr = None
         self.linear0 = nn.Linear(
             d_model, dim_feedforward, weight_attr0, bias_attr=bias_attr

python/paddle/fluid/tests/unittests/test_avoid_twice_initialization.py

@@ -14,6 +14,7 @@
 
 import unittest
 
+import paddle
 import paddle.fluid as fluid
 
 
@@ -22,7 +23,7 @@ class TestAvoidTwiceInitialization(unittest.TestCase):
         cur_program = fluid.Program()
         cur_block = cur_program.current_block()
         var = cur_block.create_parameter(
-            initializer=fluid.initializer.Constant(value=0.01),
+            initializer=paddle.nn.initializer.Constant(value=0.01),
             shape=[2, 2],
             dtype='float32',
             name='var_a',
@@ -40,7 +41,7 @@ class TestAvoidTwiceInitialization(unittest.TestCase):
             attrs={'ring_id': 0},
         )
         var2 = cur_block.create_parameter(
-            initializer=fluid.initializer.Constant(value=0.01),
+            initializer=paddle.nn.initializer.Constant(value=0.01),
             shape=[2, 2],
             dtype='float32',
             name='var_a',

python/paddle/fluid/tests/unittests/test_base_layer.py

@@ -26,7 +26,7 @@ class L1(fluid.Layer):
     def __init__(self):
         super().__init__()
         self._param_attr = fluid.ParamAttr(
-            initializer=fluid.initializer.Constant(value=0.1)
+            initializer=paddle.nn.initializer.Constant(value=0.1)
         )
         self.w1 = self.create_parameter(
             attr=self._param_attr, shape=[2, 2], dtype='float32', is_bias=False

python/paddle/fluid/tests/unittests/test_batch_norm_op_v2.py

@@ -183,11 +183,11 @@ class TestBatchNorm(unittest.TestCase):
                         shape[1],
                         is_test=is_test,
                         param_attr=fluid.ParamAttr(
-                            initializer=fluid.initializer.Constant(1.0),
+                            initializer=paddle.nn.initializer.Constant(1.0),
                             trainable=False,
                         ),
                         bias_attr=fluid.ParamAttr(
-                            initializer=fluid.initializer.Constant(0.0),
+                            initializer=paddle.nn.initializer.Constant(0.0),
                             trainable=False,
                         ),
                         trainable_statistics=trainable_statistics,
@@ -378,7 +378,7 @@ class TestBatchNormUseGlobalStats(unittest.TestCase):
                 net1 = paddle.nn.BatchNorm(
                     6,
                     param_attr=fluid.ParamAttr(
-                        initializer=fluid.initializer.Constant(1.0)
+                        initializer=paddle.nn.initializer.Constant(1.0)
                     ),
                     use_global_stats=self.use_global_stats,
                     trainable_statistics=self.trainable_statistics,

python/paddle/fluid/tests/unittests/test_calc_gradient.py

@@ -48,7 +48,7 @@ class TestDoubleGrad(unittest.TestCase):
                 name='x',
                 shape=[1],
                 dtype='float32',
-                default_initializer=fluid.initializer.Constant(3),
+                default_initializer=paddle.nn.initializer.Constant(3),
             )
             (grad1,) = fluid.gradients(net(x), x)  # 2x = 6
             z = net(x - grad1)
@@ -69,7 +69,7 @@ class TestDoubleGrad(unittest.TestCase):
                 name='x',
                 shape=[1],
                 dtype='float32',
-                default_initializer=fluid.initializer.Constant(1),
+                default_initializer=paddle.nn.initializer.Constant(1),
             )
             y = x * x
             (dx1,) = fluid.gradients(y, x)

python/paddle/fluid/tests/unittests/test_communicator_geo.py

@@ -41,7 +41,7 @@ class TestCommunicatorGeoEnd2End(unittest.TestCase):
             size=[10000, 10],
             param_attr=fluid.ParamAttr(
                 name="embedding",
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=True,
         )

python/paddle/fluid/tests/unittests/test_conv2d_layer.py

@@ -18,7 +18,6 @@ import numpy as np
 
 import paddle
 import paddle.fluid.dygraph as dg
-import paddle.fluid.initializer as I
 import paddle.nn.functional as F
 from paddle import fluid, nn
 
@@ -110,11 +109,11 @@ class Conv2DTestCase(unittest.TestCase):
                     else (-1, self.num_channels, -1, -1)
                 )
                 x_var = fluid.data("input", input_shape, dtype=self.dtype)
-                weight_attr = I.NumpyArrayInitializer(self.weight)
+                weight_attr = paddle.nn.initializer.Assign(self.weight)
                 if self.bias is None:
                     bias_attr = False
                 else:
-                    bias_attr = I.NumpyArrayInitializer(self.bias)
+                    bias_attr = paddle.nn.initializer.Assign(self.bias)
                 if self.padding_mode != 'zeros':
                     x_var = F.pad(
                         x_var,

python/paddle/fluid/tests/unittests/test_conv2d_transpose_layer.py

@@ -18,7 +18,6 @@ import numpy as np
 
 import paddle
 import paddle.fluid.dygraph as dg
-import paddle.fluid.initializer as I
 import paddle.nn.functional as F
 from paddle import fluid, nn
 
@@ -101,11 +100,11 @@ class Conv2DTransposeTestCase(unittest.TestCase):
                     else (-1, self.num_channels, -1, -1)
                 )
                 x_var = fluid.data("input", input_shape, dtype=self.dtype)
-                weight_attr = I.NumpyArrayInitializer(self.weight)
+                weight_attr = paddle.nn.initializer.Assign(self.weight)
                 if self.bias is None:
                     bias_attr = False
                 else:
-                    bias_attr = I.NumpyArrayInitializer(self.bias)
+                    bias_attr = paddle.nn.initializer.Assign(self.bias)
 
                 y_var = paddle.static.nn.conv2d_transpose(
                     x_var,

python/paddle/fluid/tests/unittests/test_conv3d_layer.py

@@ -18,7 +18,6 @@ import numpy as np
 
 import paddle
 import paddle.fluid.dygraph as dg
-import paddle.fluid.initializer as I
 import paddle.nn.functional as F
 from paddle import fluid, nn
 
@@ -97,11 +96,11 @@ class Conv3DTestCase(unittest.TestCase):
                     else (-1, self.num_channels, -1, -1, -1)
                 )
                 x_var = fluid.data("input", input_shape, dtype=self.dtype)
-                weight_attr = I.NumpyArrayInitializer(self.weight)
+                weight_attr = paddle.nn.initializer.Assign(self.weight)
                 if self.bias is None:
                     bias_attr = False
                 else:
-                    bias_attr = I.NumpyArrayInitializer(self.bias)
+                    bias_attr = paddle.nn.initializer.Assign(self.bias)
                 y_var = paddle.static.nn.conv3d(
                     x_var,
                     self.num_filters,

python/paddle/fluid/tests/unittests/test_conv3d_transpose_layer.py

@@ -18,7 +18,6 @@ import numpy as np
 
 import paddle
 import paddle.fluid.dygraph as dg
-import paddle.fluid.initializer as I
 import paddle.nn.functional as F
 from paddle import fluid, nn
 
@@ -99,11 +98,11 @@ class Conv3DTransposeTestCase(unittest.TestCase):
                     else (-1, self.num_channels, -1, -1, -1)
                 )
                 x_var = fluid.data("input", input_shape, dtype=self.dtype)
-                weight_attr = I.NumpyArrayInitializer(self.weight)
+                weight_attr = paddle.nn.initializer.Assign(self.weight)
                 if self.bias is None:
                     bias_attr = False
                 else:
-                    bias_attr = I.NumpyArrayInitializer(self.bias)
+                    bias_attr = paddle.nn.initializer.Assign(self.bias)
                 y_var = paddle.static.nn.conv3d_transpose(
                     x_var,
                     self.num_filters,

python/paddle/fluid/tests/unittests/test_cuda_random_seed.py

@@ -120,15 +120,15 @@ class TestGeneratorSeed(unittest.TestCase):
             result_1 = paddle.static.nn.fc(
                 x,
                 size=10,
-                weight_attr=fluid.initializer.TruncatedNormal(
-                    loc=0.0, scale=2.0
+                weight_attr=paddle.nn.initializer.TruncatedNormal(
+                    mean=0.0, std=2.0
                 ),
             )
             result_2 = paddle.static.nn.fc(
                 x,
                 size=10,
-                weight_attr=fluid.initializer.TruncatedNormal(
-                    loc=0.0, scale=2.0
+                weight_attr=paddle.nn.initializer.TruncatedNormal(
+                    mean=0.0, std=2.0
                 ),
             )
 

python/paddle/fluid/tests/unittests/test_decoupled_py_reader.py

@@ -61,7 +61,7 @@ def simple_fc_net(places, use_legacy_py_reader, use_double_buffer):
                     size=hidden_size,
                     activation='tanh',
                     bias_attr=fluid.ParamAttr(
-                        initializer=fluid.initializer.Constant(value=1.0)
+                        initializer=paddle.nn.initializer.Constant(value=1.0)
                     ),
                 )
 

python/paddle/fluid/tests/unittests/test_desc_clone.py

@@ -58,9 +58,7 @@ def cnn_model(data):
         size=SIZE,
         activation="softmax",
         weight_attr=fluid.param_attr.ParamAttr(
-            initializer=fluid.initializer.NormalInitializer(
-                loc=0.0, scale=scale
-            )
+            initializer=paddle.nn.initializer.Normal(loc=0.0, scale=scale)
         ),
     )
     return predict

python/paddle/fluid/tests/unittests/test_detach.py

@@ -120,7 +120,7 @@ class Test_Detach(unittest.TestCase):
                 initializer=paddle.nn.initializer.Constant(5.0)
             )
             linear_b_param_attrs = fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(6.0)
+                initializer=paddle.nn.initializer.Constant(6.0)
             )
             linear = Linear(
                 4,
@@ -132,7 +132,7 @@ class Test_Detach(unittest.TestCase):
                 initializer=paddle.nn.initializer.Constant(7.0)
             )
             linear1_b_param_attrs = fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(8.0)
+                initializer=paddle.nn.initializer.Constant(8.0)
             )
             linear1 = Linear(
                 10,

python/paddle/fluid/tests/unittests/test_dist_fleet_a_sync_optimizer_auto_async.py

@@ -58,7 +58,7 @@ class TestFleetGradientMergeMetaOptimizer(unittest.TestCase):
             size=[1000000000, 100000],
             param_attr=paddle.fluid.ParamAttr(
                 name="embedding",
-                initializer=paddle.fluid.initializer.Constant(value=0.01),
+                initializer=paddle.paddle.nn.initializer.Constant(value=0.01),
             ),
             is_sparse=True,
         )

python/paddle/fluid/tests/unittests/test_dist_fleet_heter_program.py

@@ -89,7 +89,7 @@ class TestDistFleetHeterProgram(unittest.TestCase):
                 size=[100001, 10],
                 param_attr=fluid.ParamAttr(
                     name="SparseFeatFactors",
-                    initializer=fluid.initializer.Uniform(),
+                    initializer=paddle.nn.initializer.Uniform(),
                 ),
             )
 
@@ -103,8 +103,8 @@ class TestDistFleetHeterProgram(unittest.TestCase):
                 size=400,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Normal(
-                        scale=1 / math.sqrt(concated.shape[1])
+                    initializer=paddle.nn.initializer.Normal(
+                        std=1 / math.sqrt(concated.shape[1])
                     )
                 ),
                 name="fc1",
@@ -116,8 +116,8 @@ class TestDistFleetHeterProgram(unittest.TestCase):
                 size=400,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Normal(
-                        scale=1 / math.sqrt(fc1.shape[1])
+                    initializer=paddle.nn.initializer.Normal(
+                        std=1 / math.sqrt(fc1.shape[1])
                     )
                 ),
                 name="fc2",
@@ -129,8 +129,8 @@ class TestDistFleetHeterProgram(unittest.TestCase):
                 size=400,
                 activation="relu",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Normal(
-                        scale=1 / math.sqrt(fc2.shape[1])
+                    initializer=paddle.nn.initializer.Normal(
+                        std=1 / math.sqrt(fc2.shape[1])
                     )
                 ),
                 name="fc3",
@@ -142,8 +142,8 @@ class TestDistFleetHeterProgram(unittest.TestCase):
                 size=2,
                 activation="softmax",
                 weight_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Normal(
-                        scale=1 / math.sqrt(fc3.shape[1])
+                    initializer=paddle.nn.initializer.Normal(
+                        std=1 / math.sqrt(fc3.shape[1])
                     )
                 ),
             )

python/paddle/fluid/tests/unittests/test_dist_fleet_minimize.py

@@ -77,7 +77,7 @@ class TestPSMinimize(unittest.TestCase):
             input=q,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -91,7 +91,7 @@ class TestPSMinimize(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -107,7 +107,7 @@ class TestPSMinimize(unittest.TestCase):
             input=pt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -121,7 +121,7 @@ class TestPSMinimize(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -136,7 +136,7 @@ class TestPSMinimize(unittest.TestCase):
             input=nt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -150,7 +150,7 @@ class TestPSMinimize(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),

python/paddle/fluid/tests/unittests/test_dist_fleet_ps.py

@@ -78,7 +78,7 @@ class TestPSPassWithBow(unittest.TestCase):
             is_distributed=is_distributed,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -93,7 +93,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -110,7 +110,7 @@ class TestPSPassWithBow(unittest.TestCase):
             is_distributed=is_distributed,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -125,7 +125,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -141,7 +141,7 @@ class TestPSPassWithBow(unittest.TestCase):
             is_distributed=is_distributed,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -156,7 +156,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),

python/paddle/fluid/tests/unittests/test_dist_fleet_ps11.py

@@ -77,7 +77,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=q,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -91,7 +91,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -107,7 +107,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=pt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -121,7 +121,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -136,7 +136,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=nt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -150,7 +150,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),

python/paddle/fluid/tests/unittests/test_dist_fleet_ps12.py

@@ -80,7 +80,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=q,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -94,7 +94,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -110,7 +110,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=pt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -124,7 +124,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -139,7 +139,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=nt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -153,7 +153,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),

python/paddle/fluid/tests/unittests/test_dist_fleet_ps13.py

@@ -81,7 +81,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=q,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -95,7 +95,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -111,7 +111,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=pt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -125,7 +125,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -140,7 +140,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=nt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -154,7 +154,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),

python/paddle/fluid/tests/unittests/test_dist_fleet_ps2.py

@@ -80,7 +80,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=q,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -95,7 +95,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -111,7 +111,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=pt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -125,7 +125,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -140,7 +140,7 @@ class TestPSPassWithBow(unittest.TestCase):
             input=nt,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -154,7 +154,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),

python/paddle/fluid/tests/unittests/test_dist_fleet_ps3.py

@@ -78,7 +78,7 @@ class TestPSPassWithBow(unittest.TestCase):
             is_distributed=is_distributed,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -93,7 +93,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=q_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__q_fc__",
                 learning_rate=base_lr,
             ),
@@ -110,7 +110,7 @@ class TestPSPassWithBow(unittest.TestCase):
             is_distributed=is_distributed,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -125,7 +125,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=pt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),
@@ -141,7 +141,7 @@ class TestPSPassWithBow(unittest.TestCase):
             is_distributed=is_distributed,
             size=[dict_dim, emb_dim],
             param_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__emb__",
                 learning_rate=emb_lr,
             ),
@@ -156,7 +156,7 @@ class TestPSPassWithBow(unittest.TestCase):
             x=nt_ss,
             size=hid_dim,
             weight_attr=fluid.ParamAttr(
-                initializer=fluid.initializer.Constant(value=0.01),
+                initializer=paddle.nn.initializer.Constant(value=0.01),
                 name="__fc__",
                 learning_rate=base_lr,
             ),