unset fluid api in nn.functional (#34114)

python/paddle/nn/functional/common.py

@@ -16,18 +16,21 @@ import warnings
 import paddle
 from ...fluid.framework import in_dygraph_mode, default_main_program
 from paddle.fluid.layer_helper import LayerHelper
-from paddle.fluid.layers.tensor import Variable, fill_constant, zeros, concat
+from paddle.fluid.layers.tensor import fill_constant
+from ...tensor import concat
+from ...tensor.creation import zeros
+from paddle.static import Variable
 from ...fluid.layers import core
 from ...fluid import dygraph_utils
 # TODO: define the common functions to build a neural network  
 from ...fluid.layers import unfold  # noqa: F401
-from ...fluid.layers import squeeze
-from ...fluid.layers import unsqueeze
+from ...tensor.manipulation import squeeze
+from ...tensor.manipulation import unsqueeze
 from ...tensor import clip
 from ...tensor import sum
 from ...tensor import sqrt
 from ...fluid.data_feeder import check_variable_and_dtype, check_dtype
-from ...fluid.framework import Variable, in_dygraph_mode, _varbase_creator
+from ...fluid.framework import in_dygraph_mode, _varbase_creator
 
 from ...fluid.framework import in_dygraph_mode
 from ...fluid import core, dygraph_utils
@@ -927,9 +930,9 @@ def dropout(x,
         keep_prob = 1 - p
         if training:
             if p == 1.:
-                return layers.scale(x, scale=0.)
+                return paddle.scale(x, scale=0.)
 
-            scale_input = layers.scale(
+            scale_input = paddle.scale(
                 x, scale=1 / keep_prob) if mode == 'upscale_in_train' else x
 
             #get mask shape
@@ -947,17 +950,17 @@ def dropout(x,
                 mask_shape[i] = input_shape[i]
 
             #get mask
-            random_tensor = layers.uniform_random(
+            random_tensor = paddle.uniform(
                 mask_shape, dtype='float32', min=0., max=1.0)
             p = layers.fill_constant(shape=[1], dtype='float32', value=p)
-            keep_mask = layers.greater_equal(random_tensor, p)
+            keep_mask = paddle.greater_equal(random_tensor, p)
 
-            scale_input = layers.cast(scale_input, dtype)
-            keep_mask = layers.cast(keep_mask, dtype)
+            scale_input = paddle.cast(scale_input, dtype)
+            keep_mask = paddle.cast(keep_mask, dtype)
             ret = paddle.multiply(scale_input, keep_mask, name=name)
             return ret
         else:  # test
-            ret = layers.scale(
+            ret = paddle.scale(
                 x, scale=keep_prob) if mode == 'downscale_in_infer' else x
             return ret
 
@@ -1113,7 +1116,7 @@ def alpha_dropout(x, p=0.5, training=True, name=None):
 
     if training:
         if p == 1:
-            return layers.scale(x, scale=0.)
+            return paddle.scale(x, scale=0.)
         #get transformation params
         alpha = 1.6732632423543772848170429916717
         scale = 1.0507009873554804934193349852946
@@ -1125,23 +1128,22 @@ def alpha_dropout(x, p=0.5, training=True, name=None):
         input_shape = x.shape
 
         #get mask
-        random_tensor = layers.uniform_random(
+        random_tensor = paddle.uniform(
             input_shape, dtype='float32', min=0., max=1.0)
         p = layers.fill_constant(shape=[1], dtype='float32', value=p)
-        keep_mask = layers.greater_equal(random_tensor, p)
-        keep_mask = layers.cast(keep_mask, dtype)
-        drop_mask = layers.elementwise_sub(
+        keep_mask = paddle.greater_equal(random_tensor, p)
+        keep_mask = paddle.cast(keep_mask, dtype)
+        drop_mask = paddle.subtract(
             layers.fill_constant(
                 shape=input_shape, dtype=dtype, value=1.),
             keep_mask)
 
         #apply mask
         b = layers.fill_constant(shape=[1], dtype=dtype, value=b)
-        y = layers.elementwise_add(
-            paddle.multiply(x, keep_mask),
-            layers.scale(
-                drop_mask, scale=alpha_p))
-        res = layers.elementwise_add(layers.scale(y, scale=a), b, name=name)
+        y = paddle.add(paddle.multiply(x, keep_mask),
+                       paddle.scale(
+                           drop_mask, scale=alpha_p))
+        res = paddle.add(paddle.scale(y, scale=a), b, name=name)
         return res
     else:  # test
         return x
@@ -1277,42 +1279,42 @@ def pad(x, pad, mode='constant', value=0, data_format="NCHW", name=None):
             if x_dim == 3:
                 pad = concat([zeros((4, ), dtype="int32"), pad], axis=0)
                 unsqueezed_dim = [3, 4]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
             elif x_dim == 4:
                 pad = concat([pad, zeros((2, ), dtype="int32")], axis=0)
                 unsqueezed_dim = [2]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
         elif data_format in ["NLC", "NHWC", "NDHWC"]:
             data_format = "NDHWC"
             if x_dim == 3:
                 pad = concat([zeros((4, ), dtype="int32"), pad], axis=0)
                 unsqueezed_dim = [2, 3]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
             elif x_dim == 4:
                 pad = concat([pad, zeros((2, ), dtype="int32")], axis=0)
                 unsqueezed_dim = [1]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
     else:
         if data_format in ["NCL", "NCHW", "NCDHW"]:
             data_format = "NCDHW"
             if x_dim == 3:
                 pad = [0, 0, 0, 0] + pad
                 unsqueezed_dim = [3, 4]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
             elif x_dim == 4:
                 pad = pad + [0, 0]
                 unsqueezed_dim = [2]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
         elif data_format in ["NLC", "NHWC", "NDHWC"]:
             data_format = "NDHWC"
             if x_dim == 3:
                 pad = [0, 0, 0, 0] + pad
                 unsqueezed_dim = [2, 3]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
             elif x_dim == 4:
                 pad = pad + [0, 0]
                 unsqueezed_dim = [1]
-                x = unsqueeze(x, axes=unsqueezed_dim)
+                x = unsqueeze(x, axis=unsqueezed_dim)
 
     if in_dygraph_mode():
         if isinstance(pad, Variable):
@@ -1336,7 +1338,7 @@ def pad(x, pad, mode='constant', value=0, data_format="NCHW", name=None):
             type='pad3d', inputs=inputs, outputs={"Out": out}, attrs=attrs)
 
     if len(unsqueezed_dim) != 0:
-        out = squeeze(out, axes=unsqueezed_dim)
+        out = squeeze(out, axis=unsqueezed_dim)
 
     return out
 

python/paddle/nn/functional/conv.py

@@ -16,13 +16,17 @@ from paddle.fluid.framework import _global_flags
 
 import numpy as np
 from ...device import get_cudnn_version
-from ...fluid.framework import Variable, in_dygraph_mode
+from ...fluid.framework import in_dygraph_mode
+from ...static import Variable
 from ...fluid import core, dygraph_utils, get_flags
-from ...fluid.layers import nn, utils
+from ...fluid.layers.utils import convert_to_list, _is_symmetric_padding
 from ...fluid.data_feeder import check_variable_and_dtype
-from ...fluid.param_attr import ParamAttr
+from ...framework import ParamAttr
 from ...fluid.layer_helper import LayerHelper
 from paddle import _C_ops
+from ...tensor.manipulation import unsqueeze, squeeze
+from ...tensor.math import add
+from ...fluid.layers import nn
 
 __all__ = []
 
@@ -69,24 +73,24 @@ def _update_padding_nd(padding, channel_last, num_dims):
             padding_algorithm = "EXPLICIT"
             padding = _exclude_padding_in_batch_and_channel(padding,
                                                             channel_last)
-            if utils._is_symmetric_padding(padding, num_dims):
+            if _is_symmetric_padding(padding, num_dims):
                 padding = padding[0::2]
         # for padding like [pad_before, pad_after, pad_before, pad_after, ...]
         elif len(padding) == 2 * num_dims and isinstance(padding[0], int):
             padding_algorithm = "EXPLICIT"
-            padding = utils.convert_to_list(padding, 2 * num_dims, 'padding')
-            if utils._is_symmetric_padding(padding, num_dims):
+            padding = convert_to_list(padding, 2 * num_dims, 'padding')
+            if _is_symmetric_padding(padding, num_dims):
                 padding = padding[0::2]
         # for padding like [pad_d1, pad_d2, ...]
         elif len(padding) == num_dims and isinstance(padding[0], int):
             padding_algorithm = "EXPLICIT"
-            padding = utils.convert_to_list(padding, num_dims, 'padding')
+            padding = convert_to_list(padding, num_dims, 'padding')
         else:
             raise ValueError("In valid padding: {}".format(padding))
     # for integer padding
     else:
         padding_algorithm = "EXPLICIT"
-        padding = utils.convert_to_list(padding, num_dims, 'padding')
+        padding = convert_to_list(padding, num_dims, 'padding')
     if not all([p >= 0 for p in padding]):
         raise ValueError(
             "Invalid padding, all value should be larger than or equal to 0, but received: {}".
@@ -323,8 +327,8 @@ def conv1d(x,
             "The size of padding's dimension should be 1 or 2. But got padding={}".
             format(padding))
 
-    stride = utils.convert_to_list(stride, 1, 'stride') + [1]
-    dilation = utils.convert_to_list(dilation, 1, 'dilation') + [1]
+    stride = convert_to_list(stride, 1, 'stride') + [1]
+    dilation = convert_to_list(dilation, 1, 'dilation') + [1]
 
     l_type = "conv2d"
     if (num_channels == groups and num_channels != 1 and
@@ -333,8 +337,8 @@ def conv1d(x,
         use_cudnn = False
 
     squeeze_aixs = -2 if channel_last else -1
-    x = nn.unsqueeze(input=x, axes=[squeeze_aixs])
-    weight = nn.unsqueeze(input=weight, axes=[-1])
+    x = unsqueeze(x, axis=[squeeze_aixs])
+    weight = unsqueeze(weight, axis=[-1])
     if in_dygraph_mode():
         attrs = ('strides', stride, 'paddings', padding, 'dilations', dilation,
                  'groups', groups, 'use_cudnn', use_cudnn, 'use_mkldnn', False,
@@ -366,7 +370,7 @@ def conv1d(x,
             type=l_type, inputs=inputs, outputs=outputs, attrs=attrs)
         if bias is not None:
             out = nn.elementwise_add(out, bias, axis=channel_dim)
-    out = nn.squeeze(input=out, axes=[squeeze_aixs])
+    out = squeeze(out, axis=[squeeze_aixs])
     return out
 
 
@@ -530,8 +534,8 @@ def conv2d(x,
 
     # update attrs
     padding, padding_algorithm = _update_padding_nd(padding, channel_last, 2)
-    stride = utils.convert_to_list(stride, 2, 'stride')
-    dilation = utils.convert_to_list(dilation, 2, 'dilation')
+    stride = convert_to_list(stride, 2, 'stride')
+    dilation = convert_to_list(dilation, 2, 'dilation')
 
     l_type = "conv2d"
     if (num_channels == groups and num_channels != 1 and
@@ -730,8 +734,8 @@ def conv1d_transpose(x,
             "The size of padding's dimension should 1 or 2. But got padding={}".
             format(padding))
 
-    stride = utils.convert_to_list(stride, 1, 'stride') + [1]
-    dilation = utils.convert_to_list(dilation, 1, 'dilation') + [1]
+    stride = convert_to_list(stride, 1, 'stride') + [1]
+    dilation = convert_to_list(dilation, 1, 'dilation') + [1]
 
     if output_size is None:
         output_size = []
@@ -740,8 +744,7 @@ def conv1d_transpose(x,
             raise ValueError('output_padding option is mutually exclusive with '
                              'output_size')
         if isinstance(output_size, (list, tuple, int)):
-            output_size = utils.convert_to_list(output_size, 1,
-                                                'output_size') + [1]
+            output_size = convert_to_list(output_size, 1, 'output_size') + [1]
         else:
             raise ValueError(
                 "output_size should be int, or list, tuple of ints")
@@ -749,8 +752,8 @@ def conv1d_transpose(x,
     if output_padding == 0:
         output_padding = []
     else:
-        output_padding = utils.convert_to_list(output_padding, 1,
-                                               'output_padding') + [0]
+        output_padding = convert_to_list(output_padding, 1,
+                                         'output_padding') + [0]
 
     if len(output_padding) > 0 and output_padding[0] > stride[0]:
         raise ValueError(
@@ -768,8 +771,8 @@ def conv1d_transpose(x,
     squeeze_axis = -2 if channel_last else -1
     conv2d_data_format = "NHWC" if channel_last else "NCHW"
 
-    x = nn.unsqueeze(input=x, axes=[squeeze_axis])
-    weight = nn.unsqueeze(input=weight, axes=[-1])
+    x = unsqueeze(x, axis=[squeeze_axis])
+    weight = unsqueeze(weight, axis=[-1])
 
     if in_dygraph_mode():
         attrs = ('output_padding', output_padding, 'output_size', output_size,
@@ -803,7 +806,7 @@ def conv1d_transpose(x,
         if bias is not None:
             out = nn.elementwise_add(out, bias, axis=channel_dim)
 
-    out = nn.squeeze(input=out, axes=[squeeze_axis])
+    out = squeeze(out, axis=[squeeze_axis])
     return out
 
 
@@ -979,8 +982,8 @@ def conv2d_transpose(x,
 
     # update attrs
     padding, padding_algorithm = _update_padding_nd(padding, channel_last, 2)
-    stride = utils.convert_to_list(stride, 2, 'stride')
-    dilation = utils.convert_to_list(dilation, 2, 'dilation')
+    stride = convert_to_list(stride, 2, 'stride')
+    dilation = convert_to_list(dilation, 2, 'dilation')
 
     if output_size is None:
         output_size = []
@@ -989,7 +992,7 @@ def conv2d_transpose(x,
             raise ValueError('output_padding option is mutually exclusive with '
                              'output_size')
         if isinstance(output_size, (list, tuple, int)):
-            output_size = utils.convert_to_list(output_size, 2, 'output_size')
+            output_size = convert_to_list(output_size, 2, 'output_size')
         else:
             raise ValueError(
                 "output_size should be int, or list, tuple of ints")
@@ -997,8 +1000,7 @@ def conv2d_transpose(x,
     if output_padding == 0:
         output_padding = []
     else:
-        output_padding = utils.convert_to_list(output_padding, 2,
-                                               'output_padding')
+        output_padding = convert_to_list(output_padding, 2, 'output_padding')
 
     op_type = 'conv2d_transpose'
     num_filters = weight.shape[1]
@@ -1187,8 +1189,8 @@ def conv3d(x,
                          cudnn_version is not None) else False
 
     padding, padding_algorithm = _update_padding_nd(padding, channel_last, 3)
-    stride = utils.convert_to_list(stride, 3, 'stride')
-    dilation = utils.convert_to_list(dilation, 3, 'dilation')
+    stride = convert_to_list(stride, 3, 'stride')
+    dilation = convert_to_list(dilation, 3, 'dilation')
     op_type = "conv3d"
 
     return _conv_nd(x, weight, bias, stride, padding, padding_algorithm,
@@ -1369,8 +1371,8 @@ def conv3d_transpose(x,
                                                                    groups))
 
     padding, padding_algorithm = _update_padding_nd(padding, channel_last, 3)
-    stride = utils.convert_to_list(stride, 3, 'stride')
-    dilation = utils.convert_to_list(dilation, 3, 'dilation')
+    stride = convert_to_list(stride, 3, 'stride')
+    dilation = convert_to_list(dilation, 3, 'dilation')
     if output_size is None:
         output_size = []
     else:
@@ -1378,7 +1380,7 @@ def conv3d_transpose(x,
             raise ValueError('output_padding option is mutually exclusive with '
                              'output_size')
         if isinstance(output_size, (list, tuple, int)):
-            output_size = utils.convert_to_list(output_size, 3, 'output_size')
+            output_size = convert_to_list(output_size, 3, 'output_size')
         else:
             raise ValueError(
                 "output_size should be int, or list, tuple of ints")
@@ -1386,8 +1388,7 @@ def conv3d_transpose(x,
     if output_padding == 0:
         output_padding = []
     else:
-        output_padding = utils.convert_to_list(output_padding, 3,
-                                               'output_padding')
+        output_padding = convert_to_list(output_padding, 3, 'output_padding')
 
     cudnn_version = get_cudnn_version()
 

python/paddle/nn/functional/extension.py

@@ -17,8 +17,9 @@
 import numpy as np
 from ...fluid.data_feeder import check_dtype
 from ...fluid.layer_helper import LayerHelper
-from ...fluid.framework import Variable, in_dygraph_mode
-from ...fluid.layers.tensor import assign
+from ...fluid.framework import in_dygraph_mode
+from ...static import Variable
+from ...tensor.creation import assign
 from ...fluid import core, dygraph_utils
 from ...fluid.layers.layer_function_generator import templatedoc
 from ...fluid.layers.sequence_lod import sequence_mask

python/paddle/nn/functional/input.py

@@ -14,7 +14,8 @@
 
 from __future__ import print_function
 import warnings
-from ...fluid.framework import Variable, in_dygraph_mode
+from ...fluid.framework import in_dygraph_mode
+from ...static import Variable
 from ...fluid.layer_helper import LayerHelper
 from ...fluid.layers import core
 from ...fluid.data_feeder import check_variable_and_dtype, check_dtype

python/paddle/nn/functional/loss.py

@@ -27,7 +27,7 @@ from ...fluid.layers.nn import _elementwise_op_in_dygraph
 from ...fluid.layers import dice_loss  # noqa: F401
 from ...fluid.layers import log_loss  # noqa: F401
 from ...fluid.layers import npair_loss  # noqa: F401
-from ...fluid.layers import reshape
+from ...tensor.manipulation import reshape
 from ...fluid.layers import softmax_with_cross_entropy as fluid_softmax_with_cross_entropy
 from ...fluid.layers import square_error_cost  # noqa: F401
 
@@ -36,7 +36,7 @@ from ...fluid.layers import huber_loss
 from ...fluid.layer_helper import LayerHelper
 from ...fluid.framework import in_dygraph_mode
 from ...fluid.framework import _varbase_creator
-from ...fluid.framework import Variable
+from ...static import Variable
 from paddle.utils import deprecated
 from paddle import _C_ops
 
@@ -291,9 +291,7 @@ def binary_cross_entropy_with_logits(logit,
             pos_weight, 'pos_weight', ['float32', 'float64'],
             'binary_cross_entropy_with_logits')
         log_weight = paddle.add(
-            paddle.multiply(
-                label, paddle.fluid.layers.elementwise_sub(pos_weight, one)),
-            one)
+            paddle.multiply(label, paddle.subtract(pos_weight, one)), one)
         pos_weight_name = name if reduction == 'none' and weight is None else None
         out = paddle.multiply(out, log_weight, name=pos_weight_name)
 
@@ -515,9 +513,9 @@ def smooth_l1_loss(input, label, reduction='mean', delta=1.0, name=None):
     if reduction == 'none':
         return out
     elif reduction == 'mean':
-        return fluid.layers.reduce_mean(out)
+        return paddle.mean(out)
     elif reduction == 'sum':
-        return fluid.layers.reduce_sum(out)
+        return paddle.sum(out)
 
 
 def margin_ranking_loss(input,
@@ -592,7 +590,7 @@ def margin_ranking_loss(input,
     fluid.data_feeder.check_variable_and_dtype(
         label, 'label', ['float32', 'float64'], 'margin_rank_loss')
 
-    out = paddle.fluid.layers.elementwise_sub(other, input)
+    out = paddle.subtract(other, input)
     out = paddle.multiply(out, label)
 
     if margin != 0.0:
@@ -898,11 +896,11 @@ def kl_div(input, label, reduction='mean', name=None):
     if fluid.data_feeder.convert_dtype(
             input.dtype) == 'float32' and fluid.data_feeder.convert_dtype(
                 label.dtype) == 'float64':
-        input = fluid.layers.cast(input, 'float64')
+        input = paddle.cast(input, 'float64')
     elif fluid.data_feeder.convert_dtype(
             input.dtype) == 'float64' and fluid.data_feeder.convert_dtype(
                 label.dtype) == 'float32':
-        label = fluid.layers.cast(label, 'float64')
+        label = paddle.cast(label, 'float64')
 
     if paddle.in_dynamic_mode():
         out = _C_ops.kldiv_loss(input, label, 'reduction', reduction)
@@ -988,16 +986,12 @@ def mse_loss(input, label, reduction='mean', name=None):
             label, 'label', ['float32', 'float64'], 'mse_loss')
 
     if reduction == 'none':
-        return paddle.fluid.layers.square(
-            paddle.fluid.layers.elementwise_sub(input, label), name=name)
+        return paddle.square(paddle.subtract(input, label), name=name)
     elif reduction == 'mean':
         return paddle.mean(
-            paddle.fluid.layers.square(
-                paddle.fluid.layers.elementwise_sub(input, label)),
-            name=name)
+            paddle.square(paddle.subtract(input, label)), name=name)
     else:
-        return paddle.sum(paddle.fluid.layers.square(
-            paddle.fluid.layers.elementwise_sub(input, label)),
+        return paddle.sum(paddle.square(paddle.subtract(input, label)),
                           name=name)
 
 

python/paddle/nn/functional/norm.py

@@ -19,8 +19,8 @@ from ...fluid.data_feeder import check_variable_and_dtype, check_type
 from ...fluid.layer_helper import LayerHelper
 from ...fluid.framework import in_dygraph_mode, core
 from ...framework import create_parameter
-from ...fluid.initializer import Constant
-from ...fluid.param_attr import ParamAttr
+from ..initializer import Constant
+from ...framework import ParamAttr
 from ...fluid import core, dygraph_utils
 import numbers
 from paddle import _C_ops
@@ -104,8 +104,7 @@ def normalize(x, p=2, axis=1, epsilon=1e-12, name=None):
         type='p_norm', inputs={'X': x}, outputs={'Out': out}, attrs=attrs)
     eps = out.block.create_var(dtype=out.dtype)
     paddle.fluid.layers.fill_constant([1], out.dtype, epsilon, out=eps)
-    return paddle.fluid.layers.elementwise_div(
-        x, paddle.maximum(out, eps), name=name)
+    return paddle.divide(x, paddle.maximum(out, eps), name=name)
 
 
 def batch_norm(x,

python/paddle/nn/functional/pooling.py

@@ -15,7 +15,8 @@
 # TODO: define pooling functions
 from ...fluid import core
 from ...fluid.framework import in_dygraph_mode
-from ...fluid.layers import utils, LayerHelper, unsqueeze, squeeze
+from ...fluid.layers import utils, LayerHelper
+from ...tensor.manipulation import unsqueeze, squeeze
 from ...fluid.data_feeder import check_type, check_variable_and_dtype
 from paddle import _C_ops
 from paddle import _C_ops

python/paddle/nn/functional/vision.py

@@ -13,7 +13,8 @@
 # limitations under the License.
 
 from ...device import get_cudnn_version
-from ...fluid.framework import core, in_dygraph_mode, Variable
+from ...fluid.framework import core, in_dygraph_mode
+from ...static import Variable
 from ...fluid.layer_helper import LayerHelper
 from ...fluid.data_feeder import check_variable_and_dtype
 from ...fluid import dygraph_utils