Merge pull request #16779 from junjun315/move-api-to-root

cherry pick move api

paddle/fluid/API.spec

@@ -13,6 +13,7 @@ paddle.fluid.name_scope (ArgSpec(args=['prefix'], varargs=None, keywords=None, d
 paddle.fluid.cuda_places (ArgSpec(args=['device_ids'], varargs=None, keywords=None, defaults=(None,)), ('document', '7d9a51fc9cf3c5245b5227080a8064c3'))
 paddle.fluid.cpu_places (ArgSpec(args=['device_count'], varargs=None, keywords=None, defaults=(None,)), ('document', '4c0cd83f0b401fc2ff84c70974e5d210'))
 paddle.fluid.cuda_pinned_places (ArgSpec(args=['device_count'], varargs=None, keywords=None, defaults=(None,)), ('document', 'd0c3ebd813c39958c92b78e3eef7e912'))
+paddle.fluid.in_dygraph_mode (ArgSpec(args=[], varargs=None, keywords=None, defaults=None), ('document', 'f06314a1cb30c96b5808dde2219c2dae'))
 paddle.fluid.Executor.__init__ (ArgSpec(args=['self', 'place'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
 paddle.fluid.Executor.close (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', 'f5369953dd0c443961cf79f7a00e1a03'))
 paddle.fluid.Executor.infer_from_dataset (ArgSpec(args=['self', 'program', 'dataset', 'scope', 'thread', 'debug', 'fetch_list', 'fetch_info', 'print_period'], varargs=None, keywords=None, defaults=(None, None, None, 0, False, None, None, 100)), ('document', '9c7decb955b9c4f718114179c8985581'))

python/paddle/fluid/__init__.py

@@ -66,6 +66,8 @@ from . import compiler
 from .compiler import *
 from paddle.fluid.layers.math_op_patch import monkey_patch_variable
 from . import install_check
+from .dygraph.nn import *
+from .dygraph.layers import *
 
 Tensor = LoDTensor
 

python/paddle/fluid/dygraph/base.py

@@ -22,7 +22,7 @@ __all__ = ['enabled', 'guard', 'to_variable']
 
 
 def enabled():
-    return framework._in_dygraph_mode()
+    return framework.in_dygraph_mode()
 
 
 @signature_safe_contextmanager

python/paddle/fluid/dygraph/checkpoint.py

@@ -97,20 +97,12 @@ def load_persistables(vardict, dirname, filename=None):
 
     Examples:
         .. code-block:: python
-            my_layer = layer(fluid.dygraph.Layer)
+            my_layer = layer(fluid.Layer)
             param_path = "./my_paddle_model"
 
             param_dict = fluid.dygraph.load_persistables(my_layer.parameters(), param_path)
             param_1 = param_dict['PtbModel_0.w_1']
 
-            or:
-            my_layer = layer(fluid.dygraph.Layer)
-            param_path = "./my_paddle_model"
-            filename = "model.file"
-            param_dict = fluid.dygraph.load_persistables(my_layer.state_dict(), param_path,
-                                                                       filename=filename)
-            param_1 = param_dict['PtbModel_0.w_1']
-
         """
     if isinstance(vardict, collections.OrderedDict):
         return _load_var_from_file(vardict, dirname, filename)

python/paddle/fluid/dygraph/layer_object_helper.py

@@ -16,7 +16,7 @@ from __future__ import print_function
 
 import copy
 import six
-from ..framework import Parameter, _in_dygraph_mode
+from ..framework import Parameter, in_dygraph_mode
 from ..param_attr import ParamAttr
 from .. import core
 from six.moves import zip

python/paddle/fluid/dygraph/layers.py

@@ -139,14 +139,14 @@ class Layer(core.Layer):
 
     def clear_gradients(self):
         for p in self.parameters():
-            p._clear_gradient()
+            p.clear_gradient()
 
-    def _build_once(self, *args):
+    def build_once(self, *args):
         pass
 
     def __call__(self, *inputs):
         if not self._built:
-            self._build_once(*inputs)
+            self.build_once(*inputs)
 
         outputs = self.forward(*inputs)
         self._built = True

python/paddle/fluid/framework.py

@@ -67,6 +67,7 @@ __all__ = [
     'cuda_places',
     'cpu_places',
     'cuda_pinned_places',
+    'in_dygraph_mode',
 ]
 
 EMPTY_VAR_NAME = core.kEmptyVarName()
@@ -79,7 +80,10 @@ _dygraph_tracer_ = None
 _dygraph_current_expected_place_ = None
 
 
-def _in_dygraph_mode():
+def in_dygraph_mode():
+    '''
+    Returns(bool): True if the program is running in dynamic graph mode
+    '''
     return _dygraph_tracer_ is not None
 
 
@@ -396,7 +400,7 @@ class Variable(object):
             if not isinstance(dtype, core.VarDesc.VarType):
                 dtype = convert_np_dtype_to_dtype_(dtype)
 
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             # record vars in tracer rather than blocks
             self._ivar = kwargs.get("ivar", None)
             if not self._ivar:
@@ -482,21 +486,21 @@ class Variable(object):
 
             self.block.vars[name] = self
             self.op = None
-            self.stop_gradient = stop_gradient
+            self._stop_gradient = stop_gradient
             self.is_data = is_data
 
-    def _numpy(self):
+    def numpy(self):
         new_ivar = self._ivar._copy_to(core.CPUPlace(), True)
         return np.array(new_ivar.value().get_tensor())
 
-    def _backward(self):
+    def backward(self):
         self._ivar._run_backward()
 
-    def _gradient(self):
+    def gradient(self):
         new_ivar = self._ivar._grad_ivar()._copy_to(core.CPUPlace(), True)
         return np.array(new_ivar.value().get_tensor())
 
-    def _clear_gradient(self):
+    def clear_gradient(self):
         self._ivar._clear_gradient()
 
     def __str__(self):
@@ -516,7 +520,7 @@ class Variable(object):
         Returns:
             str: The debug string.
         """
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             # TODO(panyx0718): add more dygraph debug info.
             return 'name %s, dtype: %s shape: %s' % (self.name, self.dtype,
                                                      self.shape)
@@ -535,7 +539,7 @@ class Variable(object):
 
     __repr__ = __str__
 
-    def _set_desc(self, input):
+    def set_desc(self, input):
         """
         Set the variable description.
 
@@ -548,43 +552,43 @@ class Variable(object):
         self.desc = input
 
     @property
-    def _stop_gradient(self):
-        if _in_dygraph_mode():
+    def stop_gradient(self):
+        if in_dygraph_mode():
             return self._ivar.stop_gradient
         else:
-            return self.stop_gradient
+            return self._stop_gradient
 
-    @_stop_gradient.setter
-    def _stop_gradient(self, s):
-        if _in_dygraph_mode():
+    @stop_gradient.setter
+    def stop_gradient(self, s):
+        if in_dygraph_mode():
             self._ivar.stop_gradient = s
         else:
-            self.stop_gradient = s
+            self._stop_gradient = s
 
     @property
     def persistable(self):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self._ivar.persistable
         else:
             return self.desc.persistable()
 
     @persistable.setter
     def persistable(self, p):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self._ivar.persistable
         else:
             self.desc.set_persistable(p)
 
     @property
     def name(self):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self._ivar.name
         else:
             return cpt.to_text(self.desc.name())
 
     @name.setter
     def name(self, new_name):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             self._ivar.name = new_name
         else:
             self.desc.set_name(new_name)
@@ -592,14 +596,14 @@ class Variable(object):
     @property
     def shape(self):
         # convert to tuple, make it as same as numpy API.
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self._ivar.shape
         else:
             return tuple(self.desc.shape())
 
     @property
     def dtype(self):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self._ivar.dtype
         else:
             return self.desc.dtype()
@@ -611,7 +615,7 @@ class Variable(object):
 
     @property
     def type(self):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self._ivar.dtype
         else:
             return self.desc.type()
@@ -721,7 +725,7 @@ class Variable(object):
                 name=unique_name.generate(".".join(self.name)),
                 dtype=self.dtype,
                 persistable=self.persistable,
-                stop_gradient=self._stop_gradient, )
+                stop_gradient=self.stop_gradient, )
         else:
             return self
 
@@ -930,7 +934,7 @@ class Operator(object):
                  inputs=None,
                  outputs=None,
                  attrs=None):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             if type is None:
                 raise ValueError(
                     "`type` to initialized an Operator can not be None.")
@@ -1049,7 +1053,7 @@ class Operator(object):
                     for arg in out_args:
                         out_arg_names.append(cpt.to_text(arg.name))
                         # TODO(minqiyang): could we remove variable's op in static mode?
-                        if not _in_dygraph_mode():
+                        if not in_dygraph_mode():
                             arg.op = self
                     self.desc.set_output(out_proto.name, out_arg_names)
 
@@ -1095,7 +1099,7 @@ class Operator(object):
 
     @property
     def type(self):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             return self.iop.type
         else:
             return self.desc.type()
@@ -1638,7 +1642,7 @@ class Block(object):
         Returns:
             Operator: the append Operator.
         """
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             op = Operator(
                 block=self,
                 desc=None,
@@ -1710,7 +1714,7 @@ class Block(object):
         return self.ops[start:end]
 
     def _prepend_op(self, *args, **kwargs):
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             op = Operator(
                 self,
                 None,

python/paddle/fluid/initializer.py

@@ -165,7 +165,7 @@ class ConstantInitializer(Initializer):
                 'force_cpu': self._force_cpu or force_init_on_cpu()
             },
             stop_gradient=True)
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -245,7 +245,7 @@ class UniformInitializer(Initializer):
                 attrs={"in_dtype": out_var.dtype,
                        "out_dtype": var.dtype})
 
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -324,7 +324,7 @@ class NormalInitializer(Initializer):
                 outputs={"Out": var},
                 attrs={"in_dtype": out_var.dtype,
                        "out_dtype": var.dtype})
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -403,7 +403,7 @@ class TruncatedNormalInitializer(Initializer):
                 outputs={"Out": var},
                 attrs={"in_dtype": out_var.dtype,
                        "out_dtype": var.dtype})
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -509,7 +509,7 @@ class XavierInitializer(Initializer):
                     "seed": self._seed
                 },
                 stop_gradient=True)
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -610,7 +610,7 @@ class MSRAInitializer(Initializer):
                     "seed": self._seed
                 },
                 stop_gradient=True)
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -709,7 +709,7 @@ class BilinearInitializer(Initializer):
                 'shape': list(shape),
                 value_name: values
             })
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 
@@ -768,7 +768,7 @@ class NumpyArrayInitializer(Initializer):
                 value_name: values
             },
             stop_gradient=True)
-        if not framework._in_dygraph_mode():
+        if not framework.in_dygraph_mode():
             var.op = op
         return op
 

python/paddle/fluid/layer_helper.py

@@ -17,7 +17,7 @@ from __future__ import print_function
 import copy
 import six
 
-from .framework import Parameter, dtype_is_floating, _in_dygraph_mode
+from .framework import Parameter, dtype_is_floating, in_dygraph_mode
 from . import unique_name
 from paddle.fluid.initializer import Constant, Xavier
 from .param_attr import ParamAttr

python/paddle/fluid/layer_helper_base.py

@@ -17,7 +17,7 @@ from __future__ import print_function
 import copy
 import numpy as np
 
-from .framework import Variable, default_main_program, default_startup_program, _in_dygraph_mode, _current_expected_place
+from .framework import Variable, default_main_program, default_startup_program, in_dygraph_mode, _current_expected_place
 from . import unique_name
 from .param_attr import ParamAttr, WeightNormParamAttr
 from . import core
@@ -54,7 +54,7 @@ class LayerHelperBase(object):
         Return Variable construct from value
         """
         if isinstance(value, np.ndarray):
-            assert _in_dygraph_mode(
+            assert in_dygraph_mode(
             ), "to_variable could only be called in dygraph mode"
 
             if not block:
@@ -302,7 +302,7 @@ class LayerHelperBase(object):
             param = self._create_weight_normalize(attr, shape, dtype)
             WeightNormParamAttr.params_with_weight_norm.append(param)
             return param
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             # In dygraph mode, we want the returned parameter to be
             # initialized so that it can be used imperatively.
             return self.main_program.global_block().create_parameter(
@@ -370,7 +370,7 @@ class LayerHelperBase(object):
                initializer: initializer to use
         """
         assert isinstance(var, Variable)
-        if _in_dygraph_mode():
+        if in_dygraph_mode():
             initializer(var, var.block)
         else:
             self.startup_program.global_block().create_var(

python/paddle/fluid/layers/nn.py

@@ -23,7 +23,7 @@ import os
 import inspect
 from ..layer_helper import LayerHelper
 from ..initializer import Normal, Constant, NumpyArrayInitializer
-from ..framework import Variable, OpProtoHolder, _in_dygraph_mode
+from ..framework import Variable, OpProtoHolder, in_dygraph_mode
 from ..dygraph import base
 from ..param_attr import ParamAttr
 from .layer_function_generator import autodoc, templatedoc, _generate_doc_string_
@@ -3288,7 +3288,7 @@ def layer_norm(input,
         >>>                          dtype='float32')
         >>> x = fluid.layers.layer_norm(input=data, begin_norm_axis=1)
     """
-    assert _in_dygraph_mode(
+    assert in_dygraph_mode(
     ) is not True, "please use FC instead of fc in dygraph mode!"
     helper = LayerHelper('layer_norm', **locals())
     dtype = helper.input_dtype()
@@ -6454,7 +6454,7 @@ def squeeze(input, axes, name=None):
             x = layers.data(name='x', shape=[5, 1, 10])
             y = layers.sequeeze(input=x, axes=[1])
     """
-    assert not _in_dygraph_mode(), (
+    assert not in_dygraph_mode(), (
         "squeeze layer is not supported in dygraph mode yet.")
     helper = LayerHelper("squeeze", **locals())
     out = helper.create_variable_for_type_inference(dtype=input.dtype)
@@ -9193,7 +9193,7 @@ def _elementwise_op(helper):
     op_type = helper.layer_type
     x = helper.kwargs.get('x', None)
     y = helper.kwargs.get('y', None)
-    if _in_dygraph_mode():
+    if in_dygraph_mode():
         x = base.to_variable(x)
         y = base.to_variable(y)
 

python/paddle/fluid/optimizer.py

@@ -55,7 +55,7 @@ class Optimizer(object):
     """
 
     def __init__(self, learning_rate, regularization=None, name=None):
-        if framework._in_dygraph_mode():
+        if framework.in_dygraph_mode():
             if not isinstance(learning_rate, float) and \
                     not isinstance(learning_rate, LearningRateDecay):
                 raise TypeError(
@@ -205,7 +205,7 @@ class Optimizer(object):
             name = self._name + "_" + name
         if (name in self._accumulators and
                 param.name in self._accumulators[name]):
-            if framework._in_dygraph_mode():
+            if framework.in_dygraph_mode():
                 return self._accumulators[name][param.name]
             raise Exception("Accumulator {} already exists for parameter {}".
                             format(name, param.name))
@@ -275,7 +275,7 @@ class Optimizer(object):
         self._create_global_learning_rate()
 
         optimize_ops = []
-        if framework._in_dygraph_mode():
+        if framework.in_dygraph_mode():
             for param_and_grad in parameters_and_grads:
                 if param_and_grad[1] is None:
                     continue
@@ -374,7 +374,7 @@ class Optimizer(object):
             See examples in `apply_gradients`.
         """
         self._dtype = loss.dtype
-        if framework._in_dygraph_mode():
+        if framework.in_dygraph_mode():
             if parameter_list is not None:
                 parameters = parameter_list
             else:
@@ -459,7 +459,7 @@ class Optimizer(object):
         Returns:
             list: A list of operators appended to the current program.
         """
-        if framework._in_dygraph_mode():
+        if framework.in_dygraph_mode():
             with program_guard(framework.default_main_program(),
                                framework.default_startup_program()):
                 optimize_ops = self._create_optimization_pass(params_grads)

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

@@ -18,7 +18,7 @@ import numpy as np
 import paddle.fluid as fluid
 
 
-class L1(fluid.dygraph.Layer):
+class L1(fluid.Layer):
     def __init__(self, prefix):
         super(L1, self).__init__(prefix)
         self._param_attr = fluid.ParamAttr(
@@ -32,7 +32,7 @@ class L1(fluid.dygraph.Layer):
         return self.w1 + self.w2
 
 
-class L2(fluid.dygraph.Layer):
+class L2(fluid.Layer):
     def __init__(self, prefix):
         super(L2, self).__init__(prefix)
         self.layer1 = L1(self.full_name())
@@ -42,7 +42,7 @@ class L2(fluid.dygraph.Layer):
         return self.layer1() + self.layer2()
 
 
-class L3(fluid.dygraph.Layer):
+class L3(fluid.Layer):
     def __init__(self, prefix):
         super(L3, self).__init__(prefix)
         self.layer1 = L2(self.full_name())
@@ -59,7 +59,7 @@ class TestBaseLayer(unittest.TestCase):
             ret = l()
             self.assertEqual(l.w1.name, "test_one_level/L1_0.w_0")
             self.assertEqual(l.w2.name, "test_one_level/L1_0.w_1")
-            self.assertTrue(np.allclose(ret._numpy(), 0.2 * np.ones([2, 2])))
+            self.assertTrue(np.allclose(ret.numpy(), 0.2 * np.ones([2, 2])))
 
     def test_three_level(self):
         with fluid.dygraph.guard():
@@ -72,7 +72,7 @@ class TestBaseLayer(unittest.TestCase):
             self.assertEqual(names[3], "test_three_level/L3_0/L2_0/L1_1.w_1")
             self.assertEqual(names[4], "test_three_level/L3_0/L2_1/L1_0.w_0")
             self.assertEqual(names[5], "test_three_level/L3_0/L2_1/L1_0.w_1")
-            self.assertTrue(np.allclose(ret._numpy(), 0.8 * np.ones([2, 2])))
+            self.assertTrue(np.allclose(ret.numpy(), 0.8 * np.ones([2, 2])))
 
 
 if __name__ == '__main__':

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

@@ -18,11 +18,11 @@ import numpy as np
 
 import paddle.fluid as fluid
 from paddle.fluid import core
-from paddle.fluid.dygraph.nn import FC
+from paddle.fluid import FC
 from test_imperative_base import new_program_scope
 
 
-class MyLayer(fluid.dygraph.Layer):
+class MyLayer(fluid.Layer):
     def __init__(self, name_scope):
         super(MyLayer, self).__init__(name_scope)
 
@@ -34,7 +34,7 @@ class MyLayer(fluid.dygraph.Layer):
         return [x]
 
 
-class MyPyLayer(fluid.dygraph.PyLayer):
+class MyPyLayer(fluid.PyLayer):
     def __init__(self):
         super(MyPyLayer, self).__init__()
 
@@ -48,7 +48,7 @@ class MyPyLayer(fluid.dygraph.PyLayer):
         return np.array(dout) * (1 - np.square(np.array(out)))
 
 
-class MLP(fluid.dygraph.Layer):
+class MLP(fluid.Layer):
     def __init__(self, name_scope):
         super(MLP, self).__init__(name_scope)
         self._fc1 = FC(self.full_name(),
@@ -71,7 +71,7 @@ class MLP(fluid.dygraph.Layer):
         return x
 
 
-class SimpleRNNCell(fluid.dygraph.Layer):
+class SimpleRNNCell(fluid.Layer):
     def __init__(self, name_scope, step_input_size, hidden_size, output_size,
                  param_attr):
         super(SimpleRNNCell, self).__init__(name_scope)
@@ -81,7 +81,7 @@ class SimpleRNNCell(fluid.dygraph.Layer):
         self._dtype = core.VarDesc.VarType.FP32
         self.param_attr = param_attr
 
-    def _build_once(self, inputs, pre_hidden):
+    def build_once(self, inputs, pre_hidden):
         i2h_param_shape = [self.step_input_size, self.hidden_size]
         h2h_param_shape = [self.hidden_size, self.hidden_size]
         h2o_param_shape = [self.output_size, self.hidden_size]
@@ -159,7 +159,7 @@ class SimpleRNNCell(fluid.dygraph.Layer):
         return reduce_out, hidden
 
 
-class SimpleRNN(fluid.dygraph.Layer):
+class SimpleRNN(fluid.Layer):
     def __init__(self, name_scope):
         super(SimpleRNN, self).__init__(name_scope)
         self.seq_len = 4
@@ -200,22 +200,22 @@ class TestImperative(unittest.TestCase):
                 inputs.append(fluid.dygraph.base.to_variable(x))
             ret = fluid.layers.sums(inputs)
             loss = fluid.layers.reduce_sum(ret)
-            loss._backward()
-            self.assertTrue(np.allclose(ret._numpy(), x * 10))
-            self.assertTrue(np.allclose(inputs[0]._gradient(), x))
+            loss.backward()
+            self.assertTrue(np.allclose(ret.numpy(), x * 10))
+            self.assertTrue(np.allclose(inputs[0].gradient(), x))
 
     def test_layer(self):
         with fluid.dygraph.guard():
             cl = core.Layer()
             cl.forward([])
-            l = fluid.dygraph.Layer("l")
+            l = fluid.Layer("l")
             self.assertRaises(NotImplementedError, l.forward, [])
 
     def test_pylayer_func_id(self):
 
         with fluid.dygraph.guard():
 
-            class PyLayer1(fluid.dygraph.PyLayer):
+            class PyLayer1(fluid.PyLayer):
                 def __init__(self):
                     super(PyLayer1, self).__init__()
 
@@ -257,9 +257,9 @@ class TestImperative(unittest.TestCase):
             my_py_layer = MyPyLayer()
             var_inp = fluid.dygraph.base.to_variable(np_inp)
             outs = my_py_layer(var_inp)
-            dy_out = np.sum(outs[0]._numpy())
-            outs[0]._backward()
-            dy_grad = var_inp._gradient()
+            dy_out = np.sum(outs[0].numpy())
+            outs[0].backward()
+            dy_grad = var_inp.gradient()
 
         with new_program_scope():
             inp = fluid.layers.data(
@@ -287,9 +287,9 @@ class TestImperative(unittest.TestCase):
             l = MyLayer("my_layer")
             x = l(var_inp)[0]
             self.assertIsNotNone(x)
-            dy_out = x._numpy()
-            x._backward()
-            dy_grad = l._x_for_debug._gradient()
+            dy_out = x.numpy()
+            x.backward()
+            dy_grad = l._x_for_debug.gradient()
 
         with new_program_scope():
             inp = fluid.layers.data(
@@ -314,9 +314,9 @@ class TestImperative(unittest.TestCase):
             var_inp = fluid.dygraph.base.to_variable(np_inp)
             mlp = MLP("mlp")
             out = mlp(var_inp)
-            dy_out = out._numpy()
-            out._backward()
-            dy_grad = mlp._fc1._w._gradient()
+            dy_out = out.numpy()
+            out.backward()
+            dy_grad = mlp._fc1._w.gradient()
 
         with new_program_scope():
             inp = fluid.layers.data(
@@ -358,11 +358,11 @@ class TestImperative(unittest.TestCase):
             var_inp = fluid.layers.reshape(var_inp, shape=[1, 4, 3])
             simple_rnn = SimpleRNN("simple_rnn")
             outs, pre_hiddens = simple_rnn.forward(var_inp)
-            dy_out = outs[3]._numpy()
-            outs[3]._backward()
-            dy_grad_h2o = simple_rnn._cell._h2o_w._gradient()
-            dy_grad_h2h = simple_rnn._cell._h2h_w._gradient()
-            dy_grad_i2h = simple_rnn._cell._i2h_w._gradient()
+            dy_out = outs[3].numpy()
+            outs[3].backward()
+            dy_grad_h2o = simple_rnn._cell._h2o_w.gradient()
+            dy_grad_h2h = simple_rnn._cell._h2h_w.gradient()
+            dy_grad_i2h = simple_rnn._cell._i2h_w.gradient()
 
         with new_program_scope():
             inp = fluid.layers.data(

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

@@ -18,11 +18,11 @@ import numpy as np
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid.optimizer import SGDOptimizer
-from paddle.fluid.dygraph.nn import Conv2D, Pool2D, FC
+from paddle.fluid import Conv2D, Pool2D, FC
 from paddle.fluid.dygraph.base import to_variable
 
 
-class SimpleImgConvPool(fluid.dygraph.Layer):
+class SimpleImgConvPool(fluid.Layer):
     def __init__(self,
                  name_scope,
                  num_channels,
@@ -71,7 +71,7 @@ class SimpleImgConvPool(fluid.dygraph.Layer):
         return x
 
 
-class MNIST(fluid.dygraph.Layer):
+class MNIST(fluid.Layer):
     def __init__(self, name_scope):
         super(MNIST, self).__init__(name_scope)
 
@@ -125,21 +125,21 @@ class TestDygraphCheckpoint(unittest.TestCase):
 
                     img = to_variable(dy_x_data)
                     label = to_variable(y_data)
-                    label._stop_gradient = True
+                    label.stop_gradient = True
 
                     cost = mnist(img)
                     loss = fluid.layers.cross_entropy(cost, label)
                     avg_loss = fluid.layers.mean(loss)
 
-                    dy_out = avg_loss._numpy()
+                    dy_out = avg_loss.numpy()
 
-                    avg_loss._backward()
+                    avg_loss.backward()
                     sgd.minimize(avg_loss)
                     fluid.dygraph.save_persistables(mnist, "save_dir")
                     mnist.clear_gradients()
 
                     for param in mnist.parameters():
-                        dy_param_init_value[param.name] = param._numpy()
+                        dy_param_init_value[param.name] = param.numpy()
 
                     mnist.load_dict(
                         fluid.dygraph.load_persistables(mnist, "save_dir"))

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

@@ -32,11 +32,11 @@ NUM_BATCHES = int(os.environ.get('NUM_BATCHES', 5))
 NUM_EPOCHES = int(os.environ.get('NUM_EPOCHES', 1))
 
 
-class DMF(fluid.dygraph.Layer):
+class DMF(fluid.Layer):
     def __init__(self, name_scope):
         super(DMF, self).__init__(name_scope)
-        self._user_latent = fluid.dygraph.FC(self.full_name(), 256)
-        self._item_latent = fluid.dygraph.FC(self.full_name(), 256)
+        self._user_latent = fluid.FC(self.full_name(), 256)
+        self._item_latent = fluid.FC(self.full_name(), 256)
 
         self._user_layers = []
         self._item_layers = []
@@ -45,13 +45,11 @@ class DMF(fluid.dygraph.Layer):
             self._user_layers.append(
                 self.add_sublayer(
                     'user_layer_%d' % i,
-                    fluid.dygraph.FC(
-                        self.full_name(), self._hid_sizes[i], act='relu')))
+                    fluid.FC(self.full_name(), self._hid_sizes[i], act='relu')))
             self._item_layers.append(
                 self.add_sublayer(
                     'item_layer_%d' % i,
-                    fluid.dygraph.FC(
-                        self.full_name(), self._hid_sizes[i], act='relu')))
+                    fluid.FC(self.full_name(), self._hid_sizes[i], act='relu')))
 
     def forward(self, users, items):
         users = self._user_latent(users)
@@ -63,19 +61,18 @@ class DMF(fluid.dygraph.Layer):
         return fluid.layers.elementwise_mul(users, items)
 
 
-class MLP(fluid.dygraph.Layer):
+class MLP(fluid.Layer):
     def __init__(self, name_scope):
         super(MLP, self).__init__(name_scope)
-        self._user_latent = fluid.dygraph.FC(self.full_name(), 256)
-        self._item_latent = fluid.dygraph.FC(self.full_name(), 256)
+        self._user_latent = fluid.FC(self.full_name(), 256)
+        self._item_latent = fluid.FC(self.full_name(), 256)
         self._match_layers = []
         self._hid_sizes = [128, 64]
         for i in range(len(self._hid_sizes)):
             self._match_layers.append(
                 self.add_sublayer(
                     'match_layer_%d' % i,
-                    fluid.dygraph.FC(
-                        self.full_name(), self._hid_sizes[i], act='relu')))
+                    fluid.FC(self.full_name(), self._hid_sizes[i], act='relu')))
         self._mat
 
     def forward(self, users, items):
@@ -88,7 +85,7 @@ class MLP(fluid.dygraph.Layer):
         return match_vec
 
 
-class DeepCF(fluid.dygraph.Layer):
+class DeepCF(fluid.Layer):
     def __init__(self, name_scope, num_users, num_items, matrix):
         super(DeepCF, self).__init__(name_scope)
         self._num_users = num_users
@@ -99,11 +96,11 @@ class DeepCF(fluid.dygraph.Layer):
             matrix.dtype,
             is_bias=False,
             default_initializer=fluid.initializer.NumpyArrayInitializer(matrix))
-        self._rating_matrix._stop_gradient = True
+        self._rating_matrix.stop_gradient = True
 
         self._mlp = MLP(self.full_name())
         self._dmf = DMF(self.full_name())
-        self._match_fc = fluid.dygraph.FC(self.full_name(), 1, act='sigmoid')
+        self._match_fc = fluid.FC(self.full_name(), 1, act='sigmoid')
 
     def forward(self, users, items):
         # users_emb = self._user_emb(users)
@@ -255,10 +252,10 @@ class TestDygraphDeepCF(unittest.TestCase):
                         fluid.layers.log_loss(prediction,
                                               to_variable(labels_np[
                                                   slice:slice + BATCH_SIZE])))
-                    loss._backward()
+                    loss.backward()
                     adam.minimize(loss)
                     deepcf.clear_gradients()
-                    dy_loss = loss._numpy()
+                    dy_loss = loss.numpy()
                     sys.stderr.write('dynamic loss: %s %s\n' % (slice, dy_loss))
 
         self.assertEqual(static_loss, dy_loss)

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

@@ -22,12 +22,12 @@ import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid.optimizer import SGDOptimizer
-from paddle.fluid.dygraph.nn import Conv2D, Pool2D, FC
+from paddle.fluid import Conv2D, Pool2D, FC
 from test_imperative_base import new_program_scope
 from paddle.fluid.dygraph.base import to_variable
 
 
-class Discriminator(fluid.dygraph.Layer):
+class Discriminator(fluid.Layer):
     def __init__(self, name_scope):
         super(Discriminator, self).__init__(name_scope)
         self._fc1 = FC(self.full_name(), size=32, act='elu')
@@ -38,7 +38,7 @@ class Discriminator(fluid.dygraph.Layer):
         return self._fc2(x)
 
 
-class Generator(fluid.dygraph.Layer):
+class Generator(fluid.Layer):
     def __init__(self, name_scope):
         super(Generator, self).__init__(name_scope)
         self._fc1 = FC(self.full_name(), size=64, act='elu')
@@ -150,7 +150,7 @@ class TestDygraphGAN(unittest.TestCase):
                     x=d_fake, label=to_variable(np.zeros([2, 1], np.float32))))
 
             d_loss = d_loss_real + d_loss_fake
-            d_loss._backward()
+            d_loss.backward()
             sgd.minimize(d_loss)
             discriminator.clear_gradients()
             generator.clear_gradients()
@@ -160,15 +160,15 @@ class TestDygraphGAN(unittest.TestCase):
             g_loss = fluid.layers.reduce_mean(
                 fluid.layers.sigmoid_cross_entropy_with_logits(
                     x=d_fake, label=to_variable(np.ones([2, 1], np.float32))))
-            g_loss._backward()
+            g_loss.backward()
             sgd.minimize(g_loss)
             for p in discriminator.parameters():
-                dy_params[p.name] = p._numpy()
+                dy_params[p.name] = p.numpy()
             for p in generator.parameters():
-                dy_params[p.name] = p._numpy()
+                dy_params[p.name] = p.numpy()
 
-            dy_g_loss = g_loss._numpy()
-            dy_d_loss = d_loss._numpy()
+            dy_g_loss = g_loss.numpy()
+            dy_d_loss = d_loss.numpy()
 
         self.assertEqual(dy_g_loss, static_g_loss)
         self.assertEqual(dy_d_loss, static_d_loss)

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

@@ -15,14 +15,12 @@
 import contextlib
 import unittest
 import numpy as np
-import six
 import sys
 
 import paddle
 import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid.optimizer import AdamOptimizer
-from paddle.fluid.dygraph.nn import Conv2D, Pool2D, FC
 from test_imperative_base import new_program_scope
 from paddle.fluid.dygraph.base import to_variable
 
@@ -31,7 +29,7 @@ def gen_data():
     pass
 
 
-class GraphConv(fluid.dygraph.Layer):
+class GraphConv(fluid.Layer):
     def __init__(self, name_scope, in_features, out_features):
         super(GraphConv, self).__init__(name_scope)
 
@@ -50,7 +48,7 @@ class GraphConv(fluid.dygraph.Layer):
         return fluid.layers.matmul(adj, support) + self.bias
 
 
-class GCN(fluid.dygraph.Layer):
+class GCN(fluid.Layer):
     def __init__(self, name_scope, num_hidden):
         super(GCN, self).__init__(name_scope)
         self.gc = GraphConv(self.full_name(), num_hidden, 32)
@@ -134,10 +132,9 @@ class TestDygraphGNN(unittest.TestCase):
             loss = fluid.layers.reduce_sum(loss)
             adam = AdamOptimizer(learning_rate=1e-3)
             adam.minimize(loss)
-            self.assertEqual(static_loss, loss._numpy())
-            self.assertTrue(
-                np.allclose(static_weight, model.gc.weight._numpy()))
-            sys.stderr.write('%s %s\n' % (static_loss, loss._numpy()))
+            self.assertEqual(static_loss, loss.numpy())
+            self.assertTrue(np.allclose(static_weight, model.gc.weight.numpy()))
+            sys.stderr.write('%s %s\n' % (static_loss, loss.numpy()))
 
 
 if __name__ == '__main__':

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

@@ -128,25 +128,25 @@ class TestImperativeMnist(unittest.TestCase):
 
                     img = to_variable(dy_x_data)
                     label = to_variable(y_data)
-                    label._stop_gradient = True
+                    label.stop_gradient = True
 
                     cost = mnist(img)
                     loss = fluid.layers.cross_entropy(cost, label)
                     avg_loss = fluid.layers.mean(loss)
 
-                    dy_out = avg_loss._numpy()
+                    dy_out = avg_loss.numpy()
 
                     if epoch == 0 and batch_id == 0:
                         for param in mnist.parameters():
-                            dy_param_init_value[param.name] = param._numpy()
+                            dy_param_init_value[param.name] = param.numpy()
 
-                    avg_loss._backward()
+                    avg_loss.backward()
                     sgd.minimize(avg_loss)
                     mnist.clear_gradients()
 
                     dy_param_value = {}
                     for param in mnist.parameters():
-                        dy_param_value[param.name] = param._numpy()
+                        dy_param_value[param.name] = param.numpy()
 
         with new_program_scope():
             fluid.default_startup_program().random_seed = seed

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

@@ -28,7 +28,7 @@ from paddle.fluid.dygraph.base import to_variable
 from test_imperative_base import new_program_scope
 
 
-class MLP(fluid.dygraph.Layer):
+class MLP(fluid.Layer):
     def __init__(self, name_scope, param_attr=None, bias_attr=None):
         super(MLP, self).__init__(name_scope)
 
@@ -75,18 +75,18 @@ class TestImperativeOptimizerBase(unittest.TestCase):
 
                 cost = mlp(img)
                 avg_loss = fluid.layers.reduce_mean(cost)
-                dy_out = avg_loss._numpy()
+                dy_out = avg_loss.numpy()
 
                 if batch_id == 0:
                     for param in mlp.parameters():
-                        dy_param_init_value[param.name] = param._numpy()
+                        dy_param_init_value[param.name] = param.numpy()
 
-                avg_loss._backward()
+                avg_loss.backward()
                 optimizer.minimize(avg_loss)
                 mlp.clear_gradients()
                 dy_param_value = {}
                 for param in mlp.parameters():
-                    dy_param_value[param.name] = param._numpy()
+                    dy_param_value[param.name] = param.numpy()
 
         with new_program_scope():
             fluid.default_startup_program().random_seed = seed

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

@@ -24,10 +24,9 @@ from paddle.fluid.dygraph.base import to_variable
 from test_imperative_base import new_program_scope
 import numpy as np
 import six
-from paddle.fluid.backward import append_backward
 
 
-class SimpleLSTMRNN(fluid.dygraph.Layer):
+class SimpleLSTMRNN(fluid.Layer):
     def __init__(self,
                  name_scope,
                  hidden_size,
@@ -45,7 +44,7 @@ class SimpleLSTMRNN(fluid.dygraph.Layer):
         self.cell_array = []
         self.hidden_array = []
 
-    def _build_once(self, input_embedding, init_hidden=None, init_cell=None):
+    def build_once(self, input_embedding, init_hidden=None, init_cell=None):
         self.weight_1_arr = []
         self.weight_2_arr = []
         self.bias_arr = []
@@ -132,7 +131,7 @@ class SimpleLSTMRNN(fluid.dygraph.Layer):
         return real_res, last_hidden, last_cell
 
 
-class PtbModel(fluid.dygraph.Layer):
+class PtbModel(fluid.Layer):
     def __init__(self,
                  name_scope,
                  hidden_size,
@@ -177,7 +176,7 @@ class PtbModel(fluid.dygraph.Layer):
             default_initializer=fluid.initializer.UniformInitializer(
                 low=-self.init_scale, high=self.init_scale))
 
-    def _build_once(self, input, label, init_hidden, init_cell):
+    def build_once(self, input, label, init_hidden, init_cell):
         pass
 
     def forward(self, input, label, init_hidden, init_cell):
@@ -260,13 +259,13 @@ class TestDygraphPtbRnn(unittest.TestCase):
                                                             init_cell)
                 if i == 0:
                     for param in ptb_model.parameters():
-                        dy_param_init[param.name] = param._numpy()
-                dy_loss._backward()
+                        dy_param_init[param.name] = param.numpy()
+                dy_loss.backward()
                 sgd.minimize(dy_loss)
                 ptb_model.clear_gradients()
                 if i == batch_num - 1:
                     for param in ptb_model.parameters():
-                        dy_param_updated[param.name] = param._numpy()
+                        dy_param_updated[param.name] = param.numpy()
 
         with new_program_scope():
             fluid.default_startup_program().random_seed = seed
@@ -333,10 +332,10 @@ class TestDygraphPtbRnn(unittest.TestCase):
                     for k in range(3, len(out)):
                         static_param_updated[static_param_name_list[k -
                                                                     3]] = out[k]
-        self.assertTrue(np.allclose(static_loss_value, dy_loss._numpy()))
-        self.assertTrue(np.allclose(static_last_cell_value, last_cell._numpy()))
+        self.assertTrue(np.allclose(static_loss_value, dy_loss.numpy()))
+        self.assertTrue(np.allclose(static_last_cell_value, last_cell.numpy()))
         self.assertTrue(
-            np.allclose(static_last_hidden_value, last_hidden._numpy()))
+            np.allclose(static_last_hidden_value, last_hidden.numpy()))
         for key, value in six.iteritems(static_param_init):
             # print("static_init name: {}, value {}".format(key, value))
             # print("dy_init name: {}, value {}".format(key, dy_param_init[key]))

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

@@ -21,7 +21,7 @@ import paddle
 import paddle.fluid as fluid
 from paddle.fluid import core
 from paddle.fluid.layer_helper import LayerHelper
-from paddle.fluid.dygraph.nn import Conv2D, Pool2D, BatchNorm, FC
+from paddle.fluid import Conv2D, Pool2D, BatchNorm, FC
 from paddle.fluid.dygraph.base import to_variable
 from test_imperative_base import new_program_scope
 
@@ -68,7 +68,7 @@ def optimizer_setting(params):
     return optimizer
 
 
-class ConvBNLayer(fluid.dygraph.Layer):
+class ConvBNLayer(fluid.Layer):
     def __init__(self,
                  name_scope,
                  num_channels,
@@ -99,7 +99,7 @@ class ConvBNLayer(fluid.dygraph.Layer):
         return y
 
 
-class BottleneckBlock(fluid.dygraph.Layer):
+class BottleneckBlock(fluid.Layer):
     def __init__(self,
                  name_scope,
                  num_channels,
@@ -156,7 +156,7 @@ class BottleneckBlock(fluid.dygraph.Layer):
         return layer_helper.append_activation(y)
 
 
-class ResNet(fluid.dygraph.Layer):
+class ResNet(fluid.Layer):
     def __init__(self, name_scope, layers=50, class_dim=102):
         super(ResNet, self).__init__(name_scope)
 
@@ -247,7 +247,7 @@ class TestDygraphResnet(unittest.TestCase):
 
             dy_param_init_value = {}
             for param in resnet.parameters():
-                dy_param_init_value[param.name] = param._numpy()
+                dy_param_init_value[param.name] = param.numpy()
 
             for batch_id, data in enumerate(train_reader()):
                 if batch_id >= batch_num:
@@ -260,20 +260,20 @@ class TestDygraphResnet(unittest.TestCase):
 
                 img = to_variable(dy_x_data)
                 label = to_variable(y_data)
-                label._stop_gradient = True
+                label.stop_gradient = True
 
                 out = resnet(img)
                 loss = fluid.layers.cross_entropy(input=out, label=label)
                 avg_loss = fluid.layers.mean(x=loss)
 
-                dy_out = avg_loss._numpy()
+                dy_out = avg_loss.numpy()
 
                 if batch_id == 0:
                     for param in resnet.parameters():
                         if param.name not in dy_param_init_value:
-                            dy_param_init_value[param.name] = param._numpy()
+                            dy_param_init_value[param.name] = param.numpy()
 
-                avg_loss._backward()
+                avg_loss.backward()
 
                 dy_grad_value = {}
                 for param in resnet.parameters():
@@ -288,7 +288,7 @@ class TestDygraphResnet(unittest.TestCase):
 
                 dy_param_value = {}
                 for param in resnet.parameters():
-                    dy_param_value[param.name] = param._numpy()
+                    dy_param_value[param.name] = param.numpy()
 
         with new_program_scope():
             fluid.default_startup_program().random_seed = seed

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

@@ -16,7 +16,8 @@ from __future__ import print_function
 
 import unittest
 import paddle.fluid as fluid
-from paddle.fluid.dygraph import Embedding, LayerNorm, FC, to_variable, Layer, guard
+from paddle.fluid import Embedding, LayerNorm, FC, Layer
+from paddle.fluid.dygraph import to_variable, guard
 from test_imperative_base import new_program_scope
 from paddle.fluid import core
 import numpy as np
@@ -985,15 +986,15 @@ class TestDygraphTransformer(unittest.TestCase):
 
                 if i == 0:
                     for param in transformer.parameters():
-                        dy_param_init[param.name] = param._numpy()
+                        dy_param_init[param.name] = param.numpy()
 
-                dy_avg_cost._backward()
+                dy_avg_cost.backward()
                 optimizer.minimize(dy_avg_cost)
                 transformer.clear_gradients()
 
                 if i == batch_num - 1:
                     for param in transformer.parameters():
-                        dy_param_updated[param.name] = param._numpy()
+                        dy_param_updated[param.name] = param.numpy()
 
         with new_program_scope():
             fluid.default_startup_program().random_seed = seed
@@ -1069,13 +1070,13 @@ class TestDygraphTransformer(unittest.TestCase):
                                                                     4]] = out[k]
 
         self.assertTrue(
-            np.array_equal(static_avg_cost_value, dy_avg_cost._numpy()))
+            np.array_equal(static_avg_cost_value, dy_avg_cost.numpy()))
         self.assertTrue(
-            np.array_equal(static_sum_cost_value, dy_sum_cost._numpy()))
+            np.array_equal(static_sum_cost_value, dy_sum_cost.numpy()))
         self.assertTrue(
-            np.array_equal(static_predict_value, dy_predict._numpy()))
+            np.array_equal(static_predict_value, dy_predict.numpy()))
         self.assertTrue(
-            np.array_equal(static_token_num_value, dy_token_num._numpy()))
+            np.array_equal(static_token_num_value, dy_token_num.numpy()))
         for key, value in six.iteritems(static_param_init):
             self.assertTrue(np.array_equal(value, dy_param_init[key]))
         for key, value in six.iteritems(static_param_updated):

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

@@ -102,7 +102,7 @@ class TestLayer(LayerTest):
             dy_ret = lm(base.to_variable(inp))
 
         self.assertTrue(np.allclose(static_ret, static_ret2))
-        self.assertTrue(np.allclose(dy_ret._numpy(), static_ret2))
+        self.assertTrue(np.allclose(dy_ret.numpy(), static_ret2))
 
     def test_relu(self):
         with self.static_graph():
@@ -116,7 +116,7 @@ class TestLayer(LayerTest):
             t = np.ones([3, 3], dtype='float32')
             dy_ret = layers.relu(base.to_variable(t))
 
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
 
     def test_matmul(self):
         with self.static_graph():
@@ -137,7 +137,7 @@ class TestLayer(LayerTest):
             t2 = np.ones([3, 3], dtype='float32')
             dy_ret = layers.matmul(base.to_variable(t), base.to_variable(t2))
 
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
 
     def test_conv2d(self):
         with self.static_graph():
@@ -164,7 +164,7 @@ class TestLayer(LayerTest):
                 'conv2d', num_channels=3, num_filters=3, filter_size=[2, 2])
             dy_ret = conv2d(base.to_variable(images))
 
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
     def test_gru_unit(self):
@@ -206,7 +206,7 @@ class TestLayer(LayerTest):
 
         for i in range(len(static_ret)):
             self.assertTrue(np.allclose(static_ret[i], static_ret2[i]))
-            self.assertTrue(np.allclose(static_ret[i], dy_ret[i]._numpy()))
+            self.assertTrue(np.allclose(static_ret[i], dy_ret[i].numpy()))
 
     def test_elementwise_math(self):
         n = np.ones([3, 3], dtype='float32')
@@ -248,8 +248,8 @@ class TestLayer(LayerTest):
             ret = layers.elementwise_sub(ret, n5)
             dy_ret = layers.elementwise_mul(ret, n6)
         self.assertTrue(
-            np.allclose(static_ret, dy_ret._numpy()),
-            '%s vs %s' % (static_ret, dy_ret._numpy()))
+            np.allclose(static_ret, dy_ret.numpy()),
+            '%s vs %s' % (static_ret, dy_ret.numpy()))
 
     def test_elementwise_minmax(self):
         n = np.ones([3, 3], dtype='float32')
@@ -259,8 +259,8 @@ class TestLayer(LayerTest):
             min_ret = layers.elementwise_min(n, n2)
             max_ret = layers.elementwise_max(n, n2)
 
-        self.assertTrue(np.allclose(n, min_ret._numpy()))
-        self.assertTrue(np.allclose(n2, max_ret._numpy()))
+        self.assertTrue(np.allclose(n, min_ret.numpy()))
+        self.assertTrue(np.allclose(n2, max_ret.numpy()))
 
     def test_sequence_conv(self):
         inp_np = np.arange(12).reshape([3, 4]).astype('float32')
@@ -327,7 +327,7 @@ class TestLayer(LayerTest):
                 'conv2d_transpose', num_filters=10, output_size=28)
             dy_rlt = conv2d_transpose(base.to_variable(inp_np))
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(dy_rlt._numpy(), static_rlt))
+        self.assertTrue(np.allclose(dy_rlt.numpy(), static_rlt))
 
     def test_bilinear_tensor_product(self):
         inp_np_x = np.array([[1, 2, 3]]).astype('float32')
@@ -370,7 +370,7 @@ class TestLayer(LayerTest):
             dy_rlt = btp(base.to_variable(inp_np_x), base.to_variable(inp_np_y))
 
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(dy_rlt._numpy(), static_rlt))
+        self.assertTrue(np.allclose(dy_rlt.numpy(), static_rlt))
 
     def test_prelu(self):
         inp_np = np.ones([5, 200, 100, 100]).astype('float32')
@@ -411,7 +411,7 @@ class TestLayer(LayerTest):
             dy_rlt = prelu(base.to_variable(inp_np))
 
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(dy_rlt._numpy(), static_rlt))
+        self.assertTrue(np.allclose(dy_rlt.numpy(), static_rlt))
 
     def test_embeding(self):
         inp_word = np.array([[[1]]]).astype('int64')
@@ -444,7 +444,7 @@ class TestLayer(LayerTest):
             static_rlt3 = emb2(base.to_variable(inp_word))
 
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(static_rlt3._numpy(), static_rlt))
+        self.assertTrue(np.allclose(static_rlt3.numpy(), static_rlt))
 
     def test_nce(self):
         window_size = 5
@@ -558,7 +558,7 @@ class TestLayer(LayerTest):
             nce_loss3 = nce(embs3, words[label_word])
 
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(nce_loss3._numpy(), static_rlt))
+        self.assertTrue(np.allclose(nce_loss3.numpy(), static_rlt))
 
     def test_conv3d(self):
         with self.static_graph():
@@ -585,7 +585,7 @@ class TestLayer(LayerTest):
             conv3d = nn.Conv3D('conv3d', num_filters=3, filter_size=2)
             dy_ret = conv3d(base.to_variable(images))
 
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
     def test_row_conv(self):
@@ -679,7 +679,7 @@ class TestLayer(LayerTest):
             groupNorm = nn.GroupNorm('GroupNorm', groups=2)
             dy_ret = groupNorm(base.to_variable(input))
 
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
     def test_spectral_norm(self):
@@ -729,7 +729,7 @@ class TestLayer(LayerTest):
             spectralNorm = nn.SpectralNorm('SpectralNorm', dim=1, power_iters=2)
             dy_ret = spectralNorm(base.to_variable(input))
 
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
     def test_tree_conv(self):
@@ -802,7 +802,7 @@ class TestLayer(LayerTest):
             dy_ret = treeConv(base.to_variable(vectors), base.to_variable(adj))
 
         self.assertTrue(np.allclose(static_ret, static_ret2))
-        self.assertTrue(np.allclose(static_ret, dy_ret._numpy()))
+        self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
 
     def test_conv3d_transpose(self):
         input_array = np.arange(0, 48).reshape(
@@ -832,7 +832,7 @@ class TestLayer(LayerTest):
                 use_cudnn=False)
             dy_rlt = conv3d_transpose(base.to_variable(input_array))
         self.assertTrue(np.allclose(static_rlt2, static_rlt))
-        self.assertTrue(np.allclose(dy_rlt._numpy(), static_rlt))
+        self.assertTrue(np.allclose(dy_rlt.numpy(), static_rlt))
 
 
 class TestBook(unittest.TestCase):