Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
未验证 提交 a710738e 编写于 作者: Z zhiboniu 提交者： GitHub
浏览文件
操作

unset fluid in nn.others (#34935)

上级 1aa67778
显示空白变更内容
内联 并排
Showing with 258 addition and 245 deletion
python/paddle/__init__.py
浏览文件 @ a710738e
......@@ -22,23 +22,32 @@ except ImportError:
)
from .batch import batch # noqa: F401
from .fluid import monkey_patch_variable
from .fluid.dygraph import monkey_patch_math_varbase
from .framework import monkey_patch_variable
from .framework import monkey_patch_math_varbase
monkey_patch_variable()
monkey_patch_math_varbase()
from .framework import disable_signal_handler # noqa: F401
from .framework import get_flags # noqa: F401
from .framework import set_flags # noqa: F401
from .framework import disable_static # noqa: F401
from .framework import enable_static # noqa: F401
from .framework import in_dynamic_mode # noqa: F401
from .framework.dtype import dtype as dtype # noqa: F401
from paddle.framework.dtype import uint8 # noqa: F401
from paddle.framework.dtype import int8 # noqa: F401
from paddle.framework.dtype import int16 # noqa: F401
from paddle.framework.dtype import int32 # noqa: F401
from paddle.framework.dtype import int64 # noqa: F401
from paddle.framework.dtype import float16 # noqa: F401
from paddle.framework.dtype import float32 # noqa: F401
from paddle.framework.dtype import float64 # noqa: F401
from paddle.framework.dtype import bfloat16 # noqa: F401
from paddle.framework.dtype import bool # noqa: F401
from paddle.framework.dtype import complex64 # noqa: F401
from paddle.framework.dtype import complex128 # noqa: F401
from .framework.dtype import uint8 # noqa: F401
from .framework.dtype import int8 # noqa: F401
from .framework.dtype import int16 # noqa: F401
from .framework.dtype import int32 # noqa: F401
from .framework.dtype import int64 # noqa: F401
from .framework.dtype import float16 # noqa: F401
from .framework.dtype import float32 # noqa: F401
from .framework.dtype import float64 # noqa: F401
from .framework.dtype import bfloat16 # noqa: F401
from .framework.dtype import bool # noqa: F401
from .framework.dtype import complex64 # noqa: F401
from .framework.dtype import complex128 # noqa: F401
from .framework import VarBase as Tensor # noqa: F401
Tensor.__qualname__ = 'Tensor' # noqa: F401
import paddle.compat # noqa: F401
......@@ -142,6 +151,7 @@ from .tensor.manipulation import scatter_nd_add # noqa: F401
from .tensor.manipulation import scatter_nd # noqa: F401
from .tensor.manipulation import shard_index # noqa: F401
from .tensor.manipulation import slice # noqa: F401
from .tensor.manipulation import crop # noqa: F401
from .tensor.manipulation import split # noqa: F401
from .tensor.manipulation import squeeze # noqa: F401
from .tensor.manipulation import squeeze_ # noqa: F401
......@@ -316,23 +326,15 @@ from .tensor.stat import quantile # noqa: F401
from .device import get_cudnn_version # noqa: F401
from .device import set_device # noqa: F401
from .device import get_device # noqa: F401
from .fluid.framework import is_compiled_with_cinn # noqa: F401
from .fluid.framework import is_compiled_with_cuda # noqa: F401
from .fluid.framework import is_compiled_with_rocm # noqa: F401
from .fluid.framework import disable_signal_handler # noqa: F401
from .fluid.framework import get_flags # noqa: F401
from .fluid.framework import set_flags # noqa: F401
from .device import is_compiled_with_xpu # noqa: F401
from .device import is_compiled_with_npu # noqa: F401
from .device import is_compiled_with_ipu # noqa: F401
from .device import is_compiled_with_mlu # noqa: F401
from .device import is_compiled_with_cinn # noqa: F401
from .device import is_compiled_with_cuda # noqa: F401
from .device import is_compiled_with_rocm # noqa: F401
from .device import XPUPlace # noqa: F401
from .fluid.dygraph.base import enable_dygraph as disable_static # noqa: F401
from .fluid.dygraph.base import disable_dygraph as enable_static # noqa: F401
from .fluid.framework import in_dygraph_mode as in_dynamic_mode # noqa: F401
from .fluid.layers import crop_tensor as crop # noqa: F401
# high-level api
from .hapi import Model # noqa: F401
from . import callbacks # noqa: F401
......
python/paddle/framework/__init__.py
浏览文件 @ a710738e
......@@ -39,4 +39,13 @@ from .io import save # noqa: F401
from .io import load # noqa: F401
from ..fluid.dygraph.parallel import DataParallel # noqa: F401
from ..fluid import monkey_patch_variable
from ..fluid.dygraph import monkey_patch_math_varbase
from ..fluid.framework import disable_signal_handler # noqa: F401
from ..fluid.framework import get_flags # noqa: F401
from ..fluid.framework import set_flags # noqa: F401
from ..fluid.dygraph.base import enable_dygraph as disable_static # noqa: F401
from ..fluid.dygraph.base import disable_dygraph as enable_static # noqa: F401
from ..fluid.framework import in_dygraph_mode as in_dynamic_mode # noqa: F401
__all__ = []
python/paddle/nn/__init__.py
浏览文件 @ a710738e
......@@ -14,7 +14,6 @@
# TODO: import all neural network related api under this directory,
# including layers, linear, conv, rnn etc.
from ..fluid.dygraph.layers import Layer # noqa: F401
from ..fluid.dygraph.container import LayerList # noqa: F401
from ..fluid.dygraph.container import ParameterList # noqa: F401
......
python/paddle/nn/functional/activation.py
浏览文件 @ a710738e
......@@ -22,11 +22,11 @@ from ...tensor.math import multiply
import warnings
from ...fluid.layer_helper import LayerHelper
from ...fluid.framework import in_dygraph_mode, convert_np_dtype_to_dtype_
from ...fluid import core
from ...fluid.framework import convert_np_dtype_to_dtype_
from ...fluid.data_feeder import check_variable_and_dtype, check_dtype
import paddle
from paddle import _C_ops
from paddle import _C_ops, in_dynamic_mode
from paddle.framework import core
__all__ = []
......@@ -61,7 +61,7 @@ def celu(x, alpha=1.0, name=None):
if alpha == 0:
raise ZeroDivisionError("alpha cannot be 0 for celu")
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.celu(x, 'alpha', alpha)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'celu')
......@@ -110,7 +110,7 @@ def elu(x, alpha=1.0, name=None):
# [ 1. 15.6 ]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.elu(x, 'alpha', alpha)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'elu')
......@@ -174,7 +174,7 @@ def gelu(x, approximate=False, name=None):
# [ 0.84119201, 1.39957154]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.gelu(x, 'approximate', approximate)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'gelu')
......@@ -222,7 +222,7 @@ def hardshrink(x, threshold=0.5, name=None):
out = F.hardshrink(x) # [-1., 0., 2.5]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.hard_shrink(x, 'threshold', threshold)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -273,7 +273,7 @@ def hardtanh(x, min=-1.0, max=1.0, name=None):
out = F.hardtanh(x) # [-1., 0.3, 1.]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.brelu(x, 't_min', min, 't_max', max)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -328,7 +328,7 @@ def hardsigmoid(x, slope=0.1666667, offset=0.5, name=None):
out = F.hardsigmoid(x) # [0., 1., 0.666667]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.hard_sigmoid(x, 'slope', slope, 'offset', offset)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -382,7 +382,7 @@ def hardswish(x, name=None):
out = F.hardswish(x) # [0., 5., 0.666667]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.hard_swish(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -427,7 +427,7 @@ def leaky_relu(x, negative_slope=0.01, name=None):
out = F.leaky_relu(x) # [-0.02, 0., 1.]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.leaky_relu(x, 'alpha', negative_slope)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -518,7 +518,7 @@ def prelu(x, weight, data_format="NCHW", name=None):
1], "The weight size should be equal to x input channel in prelu() when weight shape is not [1]."
mode = 'channel'
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.prelu(x, weight, 'mode', mode, 'data_format', data_format)
helper = LayerHelper('prelu', **locals())
......@@ -560,7 +560,7 @@ def relu(x, name=None):
out = F.relu(x) # [0., 0., 1.]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.relu(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'relu')
......@@ -605,7 +605,7 @@ def log_sigmoid(x, name=None):
out = F.log_sigmoid(x) # [-0.313262 -0.126928 -0.0485874 -0.0181499]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.logsigmoid(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -672,7 +672,7 @@ def maxout(x, groups, axis=1, name=None):
# [0.7142536 0.88725346 0.61093384 0.38833922]]]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.maxout(x, 'groups', groups, 'axis', axis)
check_variable_and_dtype(x, 'x', ['float32', 'float64'], 'maxout')
......@@ -721,7 +721,7 @@ def relu6(x, name=None):
out = F.relu6(x) # [0, 0.3, 6]
"""
threshold = 6.0
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.relu6(x, 'threshold', threshold)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'relu6')
......@@ -780,7 +780,7 @@ def selu(x,
raise ValueError(
"The alpha must be no less than zero. Received: {}.".format(alpha))
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.selu(x, 'scale', scale, 'alpha', alpha)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'selu')
......@@ -821,7 +821,7 @@ def silu(x, name=None):
out = F.silu(x) # [ 0.731059, 1.761594, 2.857722, 3.928055 ]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.silu(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'silu')
......@@ -951,7 +951,7 @@ def softmax(x, axis=-1, dtype=None, name=None):
dtype = convert_np_dtype_to_dtype_(dtype)
use_cudnn = True
if in_dygraph_mode():
if in_dynamic_mode():
outs_cast = x if dtype is None \
else _C_ops.cast(x, 'in_dtype', x.dtype, 'out_dtype', dtype)
return _C_ops.softmax(outs_cast, 'axis', axis, 'use_cudnn', use_cudnn)
......@@ -1026,7 +1026,7 @@ def softplus(x, beta=1, threshold=20, name=None):
x = paddle.to_tensor(np.array([-0.4, -0.2, 0.1, 0.3]))
out = F.softplus(x) # [0.513015, 0.598139, 0.744397, 0.854355]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.softplus(x, 'beta', beta, 'threshold', threshold)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -1081,7 +1081,7 @@ def softshrink(x, threshold=0.5, name=None):
"The threshold must be no less than zero. Received: {}.".format(
threshold))
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.softshrink(x, 'lambda', threshold)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -1122,7 +1122,7 @@ def softsign(x, name=None):
x = paddle.to_tensor(np.array([-0.4, -0.2, 0.1, 0.3]))
out = F.softsign(x) # [-0.285714, -0.166667, 0.0909091, 0.230769]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.softsign(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -1160,7 +1160,7 @@ def swish(x, name=None):
out = F.swish(x) # [-0.238406, 0., 0.731059]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.swish(x, 'beta', 1.0)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'swish')
......@@ -1204,7 +1204,7 @@ def mish(x, name=None):
x = paddle.to_tensor([-5., 0., 5.])
out = F.mish(x) # [-0.03357624, 0., 4.99955208]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.mish(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'], 'mish')
......@@ -1240,7 +1240,7 @@ def tanhshrink(x, name=None):
x = paddle.to_tensor(np.array([-0.4, -0.2, 0.1, 0.3]))
out = F.tanhshrink(x) # [-0.020051, -0.00262468, 0.000332005, 0.00868739]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.tanh_shrink(x)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -1286,7 +1286,7 @@ def thresholded_relu(x, threshold=1.0, name=None):
out = F.thresholded_relu(x) # [2., 0., 0.]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.thresholded_relu(x, 'threshold', threshold)
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
......@@ -1360,7 +1360,7 @@ def log_softmax(x, axis=-1, dtype=None, name=None):
if (dtype is not None) and (not isinstance(dtype, core.VarDesc.VarType)):
dtype = convert_np_dtype_to_dtype_(dtype)
if in_dygraph_mode():
if in_dynamic_mode():
if dtype is not None:
x = _C_ops.cast(x, 'in_dtype', x.dtype, 'out_dtype', dtype)
return _C_ops.log_softmax(x, 'axis', axis)
......@@ -1498,7 +1498,7 @@ def gumbel_softmax(x, temperature=1.0, hard=False, axis=-1, name=None):
# [0.00000000, 0.00000000, 0.00000000, 0.00001258, 0.99998736, 0.00000000]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.gumbel_softmax(x, 'temperature', temperature, 'hard',
hard, 'axis', axis)
......
python/paddle/nn/functional/common.py
浏览文件 @ a710738e
......@@ -14,13 +14,11 @@
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 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
......@@ -30,13 +28,17 @@ 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 in_dygraph_mode, _varbase_creator
from ...fluid.framework import _varbase_creator
from ...fluid.framework import in_dygraph_mode
from ...fluid import core, dygraph_utils
from ...fluid import core, layers
from ...fluid import dygraph_utils
from ...fluid import layers
from ...fluid.data_feeder import check_variable_and_dtype
from paddle import _C_ops
from paddle.framework import in_dynamic_mode
from paddle.tensor.creation import full
from paddle.framework import core
from paddle.static import default_main_program
__all__ = []
......@@ -353,11 +355,11 @@ def interpolate(x,
if out_shape is not None and scale is not None:
raise ValueError("Only one of size or scale_factor should be defined.")
if out_shape is not None:
if isinstance(out_shape, Variable) and not in_dygraph_mode():
if isinstance(out_shape, Variable) and not in_dynamic_mode():
out_shape.stop_gradient = True
inputs['OutSize'] = out_shape
else:
if in_dygraph_mode():
if in_dynamic_mode():
if isinstance(out_shape, Variable):
out_shape = list(out_shape.numpy())
for i, dim in enumerate(out_shape):
......@@ -428,7 +430,7 @@ def interpolate(x,
attrs['out_w'] = out_shape[2]
else:
if in_dygraph_mode() and isinstance(scale, Variable):
if in_dynamic_mode() and isinstance(scale, Variable):
scale = list(scale.numpy())
if isinstance(scale, Variable):
scale.stop_gradient = True
......@@ -454,7 +456,7 @@ def interpolate(x,
"Attr(scale)'s type should be float, int, list, tuple, or Tensor."
)
if in_dygraph_mode():
if in_dynamic_mode():
attr_list = []
for k, v in attrs.items():
attr_list.append(k)
......@@ -719,7 +721,7 @@ def bilinear(x1, x2, weight, bias=None, name=None):
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.bilinear_tensor_product(x1, x2, weight, bias)
check_variable_and_dtype(x1, 'x1', ['float32', 'float64'], 'bilinear')
......@@ -891,7 +893,7 @@ def dropout(x,
seed = None
mode = 'downgrade_in_infer' if mode == 'downscale_in_infer' else mode #semantic transfer
if in_dygraph_mode():
if in_dynamic_mode():
if default_main_program().random_seed != 0:
seed = default_main_program().random_seed
out, mask = _C_ops.dropout(
......@@ -930,7 +932,7 @@ def dropout(x,
attrs=attrs)
return out
else: #sometimes called dropout_nd #TODO: optimize with c++
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'], 'dropout')
dtype = x.dtype
keep_prob = 1 - p
......@@ -943,7 +945,7 @@ def dropout(x,
#get mask shape
input_shape = x.shape
if not in_dygraph_mode():
if not in_dynamic_mode():
input_shape_tensor = paddle.shape(x)
drop_axes = [axis] if isinstance(axis, int) else list(axis)
if min(drop_axes) < 0 or max(drop_axes) > len(input_shape) - 1:
......@@ -954,7 +956,7 @@ def dropout(x,
"length of axis should not be greater than dimensions of x:{}, but get length of axis: {}".
format(len(input_shape), len(drop_axes)))
mask_shape = [1] * len(input_shape)
if not in_dygraph_mode():
if not in_dynamic_mode():
for i in drop_axes:
mask_shape[i] = input_shape_tensor[i]
else:
......@@ -964,7 +966,7 @@ def dropout(x,
#get mask
random_tensor = paddle.uniform(
mask_shape, dtype='float32', min=0., max=1.0)
p = layers.fill_constant(shape=[1], dtype='float32', value=p)
p = full(shape=[1], fill_value=p, dtype='float32')
keep_mask = paddle.greater_equal(random_tensor, p)
scale_input = paddle.cast(scale_input, dtype)
......@@ -1122,7 +1124,7 @@ def alpha_dropout(x, p=0.5, training=True, name=None):
if p < 0 or p > 1:
raise ValueError("p argument should between 0 and 1")
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'],
'alpha_dropout')
......@@ -1142,16 +1144,15 @@ def alpha_dropout(x, p=0.5, training=True, name=None):
#get mask
random_tensor = paddle.uniform(
input_shape, dtype='float32', min=0., max=1.0)
p = layers.fill_constant(shape=[1], dtype='float32', value=p)
p = full(shape=[1], fill_value=p, dtype='float32')
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)
full(
shape=input_shape, fill_value=1., dtype=dtype), keep_mask)
#apply mask
b = layers.fill_constant(shape=[1], dtype=dtype, value=b)
b = full(shape=[1], fill_value=b, dtype=dtype)
y = paddle.add(paddle.multiply(x, keep_mask),
paddle.scale(
drop_mask, scale=alpha_p))
......@@ -1347,7 +1348,7 @@ def pad(x, pad, mode='constant', value=0, data_format="NCHW", name=None):
unsqueezed_dim = [1]
x = unsqueeze(x, axis=unsqueezed_dim)
if in_dygraph_mode():
if in_dynamic_mode():
if isinstance(pad, Variable):
pad = pad.numpy()
out = _C_ops.pad3d(x, "paddings", pad, "mode", mode, "value", value,
......@@ -1519,7 +1520,7 @@ def linear(x, weight, bias=None, name=None):
# [0.9440598 0.9440598 0.9440598 0.9440598 ]
# [2.1077576 2.1077576 2.1077576 2.1077576 ]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
pre_bias = _C_ops.matmul_v2(x, weight, 'trans_x', False, 'trans_y',
False)
......@@ -1614,7 +1615,7 @@ def label_smooth(label, prior_dist=None, epsilon=0.1, name=None):
if epsilon > 1. or epsilon < 0.:
raise ValueError("The value of epsilon must be between 0 and 1.")
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.label_smooth(label, prior_dist, 'epsilon', float(epsilon))
check_variable_and_dtype(label, 'label', ['float32', 'float64'],
......@@ -1765,7 +1766,7 @@ def class_center_sample(label, num_classes, num_samples, group=None):
if (seed is None or seed == 0) and default_main_program().random_seed != 0:
seed = default_main_program().random_seed
if in_dygraph_mode():
if in_dynamic_mode():
remapped_label, sampled_class_center = _C_ops.class_center_sample(
label, 'num_classes', num_classes, 'num_samples', num_samples,
'ring_id', ring_id, 'nranks', nranks, 'rank', rank, 'fix_seed',
......
python/paddle/nn/functional/conv.py
浏览文件 @ a710738e
......@@ -16,9 +16,8 @@ from paddle.fluid.framework import _global_flags
import numpy as np
from ...device import get_cudnn_version
from ...fluid.framework import in_dygraph_mode
from ...static import Variable
from ...fluid import core, dygraph_utils, get_flags
from ...fluid import dygraph_utils
from ...fluid.layers.utils import convert_to_list, _is_symmetric_padding
from ...fluid.data_feeder import check_variable_and_dtype
from ...framework import ParamAttr
......@@ -27,6 +26,11 @@ from paddle import _C_ops
from ...tensor.manipulation import unsqueeze, squeeze
from ...tensor.math import add
from ...fluid.layers import nn
from paddle.device import is_compiled_with_cuda
from paddle.device import is_compiled_with_rocm
from paddle.device import is_compiled_with_npu
from paddle import in_dynamic_mode
from paddle import get_flags
__all__ = []
......@@ -114,7 +118,7 @@ def _conv_nd(x,
name=None):
# Due to the poor performance of NHWC, we transpose the input to NCHW.
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('strides', stride, 'paddings', padding, 'dilations', dilation,
'groups', groups, 'use_cudnn', use_cudnn, 'use_mkldnn',
use_mkldnn, 'fuse_relu_before_depthwise_conv', False,
......@@ -342,13 +346,13 @@ def conv1d(x,
l_type = "conv2d"
# When "groups==num_channels and num_filters% num_channels == 0" using depthwise_conv2d has better performance
if (core.is_compiled_with_cuda() and num_channels == groups and
if (is_compiled_with_cuda() and num_channels == groups and
num_channels != 1 and num_filters % num_channels == 0):
l_type = 'depthwise_conv2d'
use_cudnn = False
# NPU only supports depthwise_conv2d when "input_channel = output_channel = groups"
if core.is_compiled_with_npu():
if is_compiled_with_npu():
if (num_channels == groups and num_channels == num_filters):
l_type = 'depthwise_conv2d'
else:
......@@ -357,7 +361,7 @@ def conv1d(x,
squeeze_aixs = -3 if channel_last else -2
x = unsqueeze(x, axis=[squeeze_aixs])
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('strides', stride, 'paddings', padding, 'dilations', dilation,
'groups', groups, 'use_cudnn', use_cudnn, 'use_mkldnn', False,
'fuse_relu_before_depthwise_conv', False, "padding_algorithm",
......@@ -553,7 +557,7 @@ def conv2d(x,
cudnn_version = get_cudnn_version()
use_cudnn = True if (core.is_compiled_with_cuda() and
use_cudnn = True if (is_compiled_with_cuda() and
cudnn_version is not None) else False
use_mkldnn = _global_flags()["FLAGS_use_mkldnn"]
......@@ -567,20 +571,20 @@ def conv2d(x,
if (num_channels == groups and num_channels != 1 and
num_filters % num_channels == 0):
l_type = 'depthwise_conv2d'
if core.is_compiled_with_rocm():
if is_compiled_with_rocm():
use_cudnn = True
else:
use_cudnn = False
# NPU only supports depthwise_conv2d when "input_channel = output_channel = groups"
if core.is_compiled_with_npu():
if is_compiled_with_npu():
if (num_channels == groups and num_channels == num_filters):
l_type = 'depthwise_conv2d'
else:
l_type = 'conv2d'
if (core.is_compiled_with_cuda() and get_flags("FLAGS_conv2d_disable_cudnn")
["FLAGS_conv2d_disable_cudnn"]):
if (is_compiled_with_cuda() and get_flags("FLAGS_conv2d_disable_cudnn")[
"FLAGS_conv2d_disable_cudnn"]):
use_cudnn = False
return _conv_nd(x, weight, bias, stride, padding, padding_algorithm,
......@@ -815,7 +819,7 @@ def conv1d_transpose(x,
x = unsqueeze(x, axis=[squeeze_axis])
weight = unsqueeze(weight, axis=[-1])
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('output_padding', output_padding, 'output_size', output_size,
'strides', stride, 'paddings', padding, 'padding_algorithm',
padding_algorithm, 'dilations', dilation, 'groups', groups,
......@@ -1026,7 +1030,7 @@ def conv2d_transpose(x,
cudnn_version = get_cudnn_version()
use_cudnn = True if (core.is_compiled_with_cuda() and
use_cudnn = True if (is_compiled_with_cuda() and
cudnn_version is not None) else False
# update attrs
......@@ -1057,7 +1061,7 @@ def conv2d_transpose(x,
op_type = 'depthwise_conv2d_transpose'
use_cudnn = False
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('output_padding', output_padding, 'output_size', output_size,
'strides', stride, 'paddings', padding, 'padding_algorithm',
padding_algorithm, 'dilations', dilation, 'groups', groups,
......@@ -1242,7 +1246,7 @@ def conv3d(x,
groups))
cudnn_version = get_cudnn_version()
use_cudnn = True if (core.is_compiled_with_cuda() and
use_cudnn = True if (is_compiled_with_cuda() and
cudnn_version is not None) else False
padding, padding_algorithm = _update_padding_nd(padding, channel_last, 3)
......@@ -1458,13 +1462,13 @@ def conv3d_transpose(x,
cudnn_version = get_cudnn_version()
#TODO(LielinJiang): whether to use cudnn according to the version of cudnn
use_cudnn = True if (core.is_compiled_with_cuda() and
use_cudnn = True if (is_compiled_with_cuda() and
cudnn_version is not None) else False
op_type = 'conv3d_transpose'
data_format_ = "NHWC" if channel_last else "NCHW"
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('output_padding', output_padding, 'output_size', output_size,
'paddings', padding, "padding_algorithm", padding_algorithm,
'strides', stride, 'dilations', dilation, 'groups', groups,
......
python/paddle/nn/functional/extension.py
浏览文件 @ a710738e
......@@ -17,12 +17,12 @@
import numpy as np
from ...fluid.data_feeder import check_dtype
from ...fluid.layer_helper import LayerHelper
from ...fluid.framework import in_dygraph_mode
from ...static import Variable
from ...tensor.creation import assign
from ...fluid import core, dygraph_utils
from ...fluid import dygraph_utils
from ...fluid.layers.layer_function_generator import templatedoc
from ...fluid.layers.sequence_lod import sequence_mask
from ...fluid.layers.sequence_lod import sequence_mask #noqa: F401
from paddle import in_dynamic_mode
__all__ = []
......@@ -125,7 +125,7 @@ def diag_embed(input, offset=0, dim1=-2, dim2=-1):
"dim1 and dim2 cannot be the same dimension." \
"But received dim1 = %d, dim2 = %d\n"%(dim1, dim2)
if not in_dygraph_mode():
if not in_dynamic_mode():
__check_input(input, offset, dim1, dim2)
helper = LayerHelper("diag_embed", **locals())
......
python/paddle/nn/functional/input.py
浏览文件 @ a710738e
......@@ -14,12 +14,11 @@
from __future__ import print_function
import warnings
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
from paddle import _C_ops
from paddle import in_dynamic_mode
__all__ = []
......@@ -87,7 +86,7 @@ def one_hot(x, num_classes, name=None):
"""
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.one_hot_v2(x, 'depth', num_classes, 'allow_out_of_range',
False)
else:
......@@ -196,7 +195,7 @@ def embedding(x, weight, padding_idx=None, sparse=False, name=None):
raise ValueError("padding_idx must be within [-{}, {})".format(
weight.shape[0], weight.shape[0]))
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.lookup_table_v2(
weight, x, 'is_sparse', sparse, 'is_distributed', False,
'remote_prefetch', False, 'padding_idx', padding_idx)
......
python/paddle/nn/functional/loss.py
浏览文件 @ a710738e
......@@ -14,15 +14,12 @@
# limitations under the License.
import paddle
from ...fluid.layer_helper import LayerHelper
from ...fluid.data_feeder import check_variable_and_dtype
import paddle.fluid as fluid
# TODO: define loss functions of neural network
import numpy as np
import paddle
import paddle.fluid as fluid
from ...fluid.framework import core, in_dygraph_mode
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
......@@ -34,11 +31,12 @@ from ...fluid.layers import square_error_cost # noqa: F401
from ...fluid.layers import edit_distance # noqa: F401
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 ...static import Variable
from paddle.utils import deprecated
from paddle import _C_ops
from paddle import in_dynamic_mode
from paddle.framework import core
__all__ = []
......@@ -115,7 +113,7 @@ def binary_cross_entropy(input, label, weight=None, reduction='mean',
"'mean' or 'none', but received %s, which is not allowed." %
reduction)
if in_dygraph_mode():
if in_dynamic_mode():
out = _C_ops.bce_loss(input, label)
if weight is not None:
out = _C_ops.elementwise_mul(out, weight, 'axis', -1)
......@@ -249,7 +247,7 @@ def binary_cross_entropy_with_logits(logit,
"should be 'sum', 'mean' or 'none', but received %s, which is not allowed."
% reduction)
if in_dygraph_mode():
if in_dynamic_mode():
one = _varbase_creator(dtype=logit.dtype)
_C_ops.fill_constant(one, 'value',
float(1.0), 'force_cpu', False, 'dtype', one.dtype,
......@@ -284,8 +282,7 @@ def binary_cross_entropy_with_logits(logit,
out = paddle.fluid.layers.sigmoid_cross_entropy_with_logits(
logit, label, name=sigmoid_name)
one = paddle.fluid.layers.fill_constant(
shape=[1], value=1.0, dtype=logit.dtype)
one = paddle.full(shape=[1], fill_value=1.0, dtype=logit.dtype)
if pos_weight is not None:
fluid.data_feeder.check_variable_and_dtype(
pos_weight, 'pos_weight', ['float32', 'float64'],
......@@ -392,7 +389,7 @@ def hsigmoid_loss(input,
# [2.2407534]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
out, _, _ = _C_ops.hierarchical_sigmoid(
input, weight, label, path_table, path_code, bias, 'num_classes',
num_classes, 'is_sparse', is_sparse, 'remote_prefetch', is_sparse)
......@@ -569,7 +566,7 @@ def margin_ranking_loss(input,
raise ValueError(
"The value of 'reduction' in MarginRankingLoss should be 'sum', 'mean' or 'none', but "
"received %s, which is not allowed." % reduction)
if fluid.framework.in_dygraph_mode():
if in_dynamic_mode():
out = _C_ops.elementwise_sub(other, input)
out = _C_ops.elementwise_mul(out, label)
if margin != 0.0:
......@@ -595,8 +592,7 @@ def margin_ranking_loss(input,
if margin != 0.0:
margin_var = out.block.create_var(dtype=out.dtype)
paddle.fluid.layers.fill_constant(
[1], out.dtype, margin, out=margin_var)
margin_var = paddle.full(shape=[1], fill_value=margin, dtype=out.dtype)
out = paddle.add(out, margin_var)
result_out = helper.create_variable_for_type_inference(input.dtype)
......@@ -686,7 +682,7 @@ def l1_loss(input, label, reduction='mean', name=None):
"The value of 'reduction' in L1Loss should be 'sum', 'mean' or 'none', but "
"received %s, which is not allowed." % reduction)
if in_dygraph_mode():
if in_dynamic_mode():
unreduced = _elementwise_op_in_dygraph(
input, label, axis=-1, act='abs', op_name='elementwise_sub')
if reduction == 'mean':
......@@ -776,7 +772,7 @@ def nll_loss(input,
input_dims))
n = input_shape[0]
c = input_shape[1]
if in_dygraph_mode():
if in_dynamic_mode():
if input_dims != 2 and input_dims != 4:
input, _ = _C_ops.reshape2(input, None, 'shape', [n, c, 1, -1])
label, _ = _C_ops.reshape2(label, None, 'shape', [n, 1, -1])
......@@ -995,7 +991,7 @@ def mse_loss(input, label, reduction='mean', name=None):
"'reduction' in 'mse_loss' should be 'sum', 'mean' or 'none', "
"but received {}.".format(reduction))
if not paddle.fluid.framework.in_dygraph_mode():
if not in_dynamic_mode():
paddle.fluid.data_feeder.check_variable_and_dtype(
input, 'input', ['float32', 'float64'], 'mse_loss')
paddle.fluid.data_feeder.check_variable_and_dtype(
......@@ -1099,7 +1095,7 @@ def ctc_loss(log_probs,
loss_out = fluid.layers.warpctc(log_probs, labels, blank, norm_by_times,
input_lengths, label_lengths)
loss_out = fluid.layers.squeeze(loss_out, [-1])
loss_out = paddle.squeeze(loss_out, [-1])
assert reduction in ['mean', 'sum', 'none']
if reduction == 'mean':
loss_out = paddle.mean(loss_out / label_lengths)
......@@ -1319,7 +1315,7 @@ def margin_cross_entropy(logits,
if input_dims - 1 == label_dims:
label = paddle.unsqueeze(label, axis=-1)
if in_dygraph_mode():
if in_dynamic_mode():
softmax, loss = _C_ops.margin_cross_entropy(
logits, label, 'ring_id', ring_id, 'rank', rank, 'nranks', nranks,
'margin1', margin1, 'margin2', margin2, 'margin3', margin3, 'scale',
......@@ -1664,7 +1660,7 @@ def cross_entropy(input,
(got nput_dims{}, label_dims{})'.format(input_dims, label_dims))
if input_dims - 1 == label_dims:
label = paddle.unsqueeze(label, axis=axis)
if in_dygraph_mode():
if in_dynamic_mode():
if soft_label == False:
valid_label = paddle.cast(
label != ignore_index, dtype=label.dtype) * label
......@@ -1978,7 +1974,7 @@ def sigmoid_focal_loss(logit,
"Expected one dimension of normalizer in sigmoid_focal_loss but got {}.".
format(normalizer_dims))
if in_dygraph_mode():
if in_dynamic_mode():
one = _varbase_creator(dtype=logit.dtype)
_C_ops.fill_constant(one, 'value',
float(1.0), 'force_cpu', False, 'dtype', one.dtype,
......@@ -2025,7 +2021,7 @@ def sigmoid_focal_loss(logit,
loss = paddle.nn.functional.binary_cross_entropy_with_logits(
logit, label, reduction='none', name=bce_name)
pred = fluid.layers.sigmoid(logit)
pred = paddle.nn.functional.sigmoid(logit)
p_t = pred * label + (1 - pred) * (1 - label)
alpha_t = alpha * label + (1 - alpha) * (1 - label)
......@@ -2125,7 +2121,7 @@ def hinge_embedding_loss(input, label, margin=1.0, reduction='mean', name=None):
"'reduction' in 'hinge_embedding_loss' should be 'sum', 'mean' or 'none', "
"but received {}.".format(reduction))
if not paddle.fluid.framework.in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(input, 'input', ['float32', 'float64'],
'hinge_embedding_loss')
check_variable_and_dtype(label, 'label', ['float32', 'float64'],
......
python/paddle/nn/functional/norm.py
浏览文件 @ a710738e
......@@ -17,13 +17,13 @@ import paddle
import paddle.fluid as fluid
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 ..initializer import Constant
from ...framework import ParamAttr
from ...fluid import core, dygraph_utils
from ...fluid import dygraph_utils
import numbers
from paddle import _C_ops
from paddle import in_dynamic_mode
__all__ = []
......@@ -78,7 +78,7 @@ def normalize(x, p=2, axis=1, epsilon=1e-12, name=None):
# [[0. 0.24253564 0.37139067]
# [1. 0.97014254 0.9284767 ]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
eps = fluid.dygraph.base.to_variable([epsilon], dtype=x.dtype)
out = _C_ops.p_norm(x, 'axis', axis, 'porder',
float(p), 'keepdim', True, 'epsilon', epsilon)
......@@ -104,7 +104,7 @@ def normalize(x, p=2, axis=1, epsilon=1e-12, name=None):
helper.append_op(
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)
eps = paddle.full(shape=[1], fill_value=epsilon, dtype=out.dtype)
return paddle.divide(x, paddle.maximum(out, eps), name=name)
......@@ -180,7 +180,7 @@ def batch_norm(x,
else:
trainable_statistics = not use_global_stats
if in_dygraph_mode():
if in_dynamic_mode():
# for dygraph need tuple
attrs = ("momentum", momentum, "epsilon", epsilon, "is_test",
not training, "data_layout", data_format, "use_mkldnn", False,
......@@ -301,7 +301,7 @@ def layer_norm(x,
str_normalized_shape[
1:] + ', but got input shape ' + str(input_shape))
if in_dygraph_mode():
if in_dynamic_mode():
pre_act, _, _ = _C_ops.layer_norm(x, weight, bias, 'epsilon', epsilon,
'begin_norm_axis', begin_norm_axis)
return dygraph_utils._append_activation_in_dygraph(pre_act, act=None)
......@@ -385,7 +385,7 @@ def instance_norm(x,
"""
if in_dygraph_mode():
if in_dynamic_mode():
out, _, _ = _C_ops.instance_norm(x, weight, bias, "epsilon", eps,
"momentum", momentum, "data_format",
data_format)
......@@ -474,7 +474,7 @@ def local_response_norm(x,
y = paddle.nn.functional.local_response_norm(x, size=5)
print(y.shape) # [3, 3, 112, 112]
"""
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32'], 'local_response_norm')
if data_format not in ['NCL', 'NLC', 'NCHW', 'NHWC', 'NCDHW', 'NDHWC']:
raise ValueError(
......
python/paddle/nn/functional/pooling.py
浏览文件 @ a710738e
......@@ -13,13 +13,11 @@
# limitations under the License.
# TODO: define pooling functions
from ...fluid import core
from ...fluid.framework import in_dygraph_mode
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
from paddle import in_dynamic_mode
__all__ = []
......@@ -210,7 +208,7 @@ def avg_pool1d(x,
"""
"""NCL to NCHW"""
data_format = "NCHW"
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'], 'avg_pool1d')
_check_input(x, 3)
x = unsqueeze(x, [2])
......@@ -232,7 +230,7 @@ def avg_pool1d(x,
# use 2d to implenment 1d should expand padding in advance.
padding = _expand_low_nd_padding(padding)
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.pool2d(
x, 'pooling_type', 'avg', 'ksize', kernel_size, 'global_pooling',
False, 'strides', stride, 'paddings', padding, 'padding_algorithm',
......@@ -346,7 +344,7 @@ def avg_pool2d(x,
padding, padding_algorithm = _update_padding_nd(
padding, 2, channel_last, ceil_mode=ceil_mode)
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.pool2d(x, 'pooling_type', 'avg', 'ksize', kernel_size,
'global_pooling', False, 'padding_algorithm',
padding_algorithm, 'strides', stride, 'paddings',
......@@ -468,7 +466,7 @@ def avg_pool3d(x,
_check_value_limitation(kernel_size, "kernel_size", min_limit=1e-3)
_check_value_limitation(stride, "stride", min_limit=1e-3)
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.pool3d(
x, 'pooling_type', 'avg', 'ksize', kernel_size, 'strides', stride,
'paddings', padding, 'global_pooling', False, 'padding_algorithm',
......@@ -571,7 +569,7 @@ def max_pool1d(x,
"""
"""NCL to NCHW"""
data_format = "NCHW"
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'], 'max_pool1d')
_check_input(x, 3)
x = unsqueeze(x, [2])
......@@ -587,7 +585,7 @@ def max_pool1d(x,
# use 2d to implenment 1d should expand padding in advance.
padding = _expand_low_nd_padding(padding)
if in_dygraph_mode():
if in_dynamic_mode():
if return_mask:
pool_out = _C_ops.max_pool2d_with_index(
x, 'ksize', kernel_size, 'global_pooling', False, 'strides',
......@@ -746,7 +744,7 @@ def max_unpool1d(x,
output_size = _unpool_output_size(x, kernel_size, stride, padding,
output_size)
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.unpool(x, indices, 'unpooling_type', 'max', 'ksize',
kernel_size, 'strides', stride, 'paddings',
padding, "output_size", output_size,
......@@ -861,7 +859,7 @@ def max_unpool2d(x,
output_size = _unpool_output_size(x, kernel_size, stride, padding,
output_size)
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.unpool(x, indices, 'unpooling_type', 'max', 'ksize',
kernel_size, 'strides', stride, 'paddings',
padding, "output_size", output_size,
......@@ -973,7 +971,7 @@ def max_unpool3d(x,
output_size = _unpool_output_size(x, kernel_size, stride, padding,
output_size)
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.unpool3d(x, indices, 'unpooling_type', 'max', 'ksize',
kernel_size, 'strides', stride, 'paddings',
padding, "output_size", output_size,
......@@ -1029,7 +1027,7 @@ def max_pool2d(x,
"When setting return_mask to true, data_format must be set to NCHW in API:max_pool2d"
)
if in_dygraph_mode():
if in_dynamic_mode():
if return_mask:
output = _C_ops.max_pool2d_with_index(
x, 'ksize', kernel_size, 'global_pooling', False, 'strides',
......@@ -1160,7 +1158,7 @@ def max_pool3d(x,
"When setting return_mask to true, data_format must be set to NCDHW in API:max_pool3d"
)
if in_dygraph_mode():
if in_dynamic_mode():
if return_mask:
output = _C_ops.max_pool3d_with_index(
x, 'pooling_type', 'max', 'ksize', kernel_size, 'strides',
......@@ -1250,7 +1248,7 @@ def adaptive_avg_pool1d(x, output_size, name=None):
# pool_out shape: [1, 3, 16])
"""
pool_type = 'avg'
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
'adaptive_pool2d')
check_type(output_size, 'pool_size', (int), 'adaptive_pool1d')
......@@ -1258,7 +1256,7 @@ def adaptive_avg_pool1d(x, output_size, name=None):
pool_size = [1] + utils.convert_to_list(output_size, 1, 'pool_size')
x = unsqueeze(x, [2])
if in_dygraph_mode():
if in_dynamic_mode():
pool_out = _C_ops.pool2d(x, 'pooling_type', pool_type, 'ksize',
pool_size, 'adaptive', True)
return squeeze(pool_out, [2])
......@@ -1333,7 +1331,7 @@ def adaptive_avg_pool2d(x, output_size, data_format='NCHW', name=None):
output_size=[3, 3])
# out.shape is [2, 3, 3, 3]
"""
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float16', 'float32', 'float64'],
'adaptive_avg_pool2d')
check_type(data_format, 'data_format', str, 'adaptive_avg_pool2d')
......@@ -1357,7 +1355,7 @@ def adaptive_avg_pool2d(x, output_size, data_format='NCHW', name=None):
if output_size[1] == None:
output_size[1] = in_w
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.pool2d(x, 'pooling_type', 'avg', 'ksize', output_size,
'global_pooling', False, 'adaptive', True,
'data_format', data_format)
......@@ -1437,7 +1435,7 @@ def adaptive_avg_pool3d(x, output_size, data_format='NCDHW', name=None):
output_size=[3, 3, 3])
# out.shape is [2, 3, 3, 3, 3]
"""
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'],
'adaptive_avg_pool3d')
check_type(data_format, 'data_format', str, 'adaptive_avg_pool3d')
......@@ -1463,7 +1461,7 @@ def adaptive_avg_pool3d(x, output_size, data_format='NCDHW', name=None):
if output_size[2] == None:
output_size[2] = in_w
if in_dygraph_mode():
if in_dynamic_mode():
output = _C_ops.pool3d(x, 'pooling_type', 'avg', 'ksize', output_size,
'global_pooling', False, 'adaptive', True,
'data_format', data_format)
......@@ -1537,7 +1535,7 @@ def adaptive_max_pool1d(x, output_size, return_mask=False, name=None):
# pool_out shape: [1, 3, 16] indices shape: [1, 3, 16]
"""
pool_type = 'max'
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'],
'adaptive_max_pool1d')
check_type(output_size, 'pool_size', int, 'adaptive_max_pool1d')
......@@ -1547,7 +1545,7 @@ def adaptive_max_pool1d(x, output_size, return_mask=False, name=None):
pool_size = [1] + utils.convert_to_list(output_size, 1, 'pool_size')
x = unsqueeze(x, [2])
if in_dygraph_mode():
if in_dynamic_mode():
pool_out = _C_ops.max_pool2d_with_index(
x, 'pooling_type', pool_type, 'ksize', pool_size, 'adaptive', True)
return (squeeze(pool_out[0], [2]), squeeze(
......@@ -1619,7 +1617,7 @@ def adaptive_max_pool2d(x, output_size, return_mask=False, name=None):
output_size=[3, 3])
# out.shape is [2, 3, 3, 3]
"""
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'],
'adaptive_max_pool2d')
check_type(return_mask, 'return_mask', bool, 'adaptive_max_pool2d')
......@@ -1636,7 +1634,7 @@ def adaptive_max_pool2d(x, output_size, return_mask=False, name=None):
if output_size[1] == None:
output_size[1] = in_w
if in_dygraph_mode():
if in_dynamic_mode():
pool_out = _C_ops.max_pool2d_with_index(
x, 'pooling_type', 'max', 'ksize', output_size, 'adaptive', True)
return pool_out if return_mask else pool_out[0]
......@@ -1710,7 +1708,7 @@ def adaptive_max_pool3d(x, output_size, return_mask=False, name=None):
# out.shape is [2, 3, 3, 3, 3]
"""
if not in_dygraph_mode():
if not in_dynamic_mode():
check_variable_and_dtype(x, 'x', ['float32', 'float64'],
'adaptive_max_pool3d')
check_type(return_mask, 'return_mask', bool, 'adaptive_max_pool3d')
......@@ -1729,7 +1727,7 @@ def adaptive_max_pool3d(x, output_size, return_mask=False, name=None):
if output_size[2] == None:
output_size[2] = in_w
if in_dygraph_mode():
if in_dynamic_mode():
pool_out = _C_ops.max_pool3d_with_index(
x, 'pooling_type', 'max', 'ksize', output_size, 'adaptive', True)
return pool_out if return_mask else pool_out[0]
......
python/paddle/nn/functional/sparse_attention.py
浏览文件 @ a710738e
......@@ -14,10 +14,10 @@
import warnings
import paddle
from ...fluid.framework import in_dygraph_mode, default_main_program
from ...fluid.framework import default_main_program
from paddle.fluid.layer_helper import LayerHelper
from ...fluid.framework import in_dygraph_mode
from paddle import _C_ops
from paddle import in_dynamic_mode
def sparse_attention(query,
......@@ -143,7 +143,7 @@ def sparse_attention(query,
# [1.60885942, 2.60885954],
# [1.99830270, 2.99830270]]]]
"""
if in_dygraph_mode():
if in_dynamic_mode():
result_attention, result_sdd, result_softmax = _C_ops.sparse_attention(
query, key, value, sparse_csr_offset, sparse_csr_columns,
key_padding_mask, attn_mask)
......
python/paddle/nn/functional/vision.py
浏览文件 @ a710738e
......@@ -13,13 +13,14 @@
# limitations under the License.
from ...device import get_cudnn_version
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
import numpy as np
from paddle import _C_ops
from ...device import is_compiled_with_rocm
from paddle import in_dynamic_mode
__all__ = []
......@@ -83,14 +84,14 @@ def affine_grid(theta, out_shape, align_corners=True, name=None):
use_cudnn = True
else:
use_cudnn = False
if core.is_compiled_with_rocm():
if is_compiled_with_rocm():
use_cudnn = False # ROCM platform do not have MIOPEN kernel for affine_grid
if not (isinstance(out_shape, list) or isinstance(out_shape, tuple) or \
isinstance(out_shape, Variable)):
raise ValueError("The out_shape should be a list, tuple or Tensor.")
if in_dygraph_mode():
if in_dynamic_mode():
_out_shape = out_shape.numpy().tolist() if isinstance(
out_shape, Variable) else out_shape
return _C_ops.affine_grid(theta, "output_shape", _out_shape,
......@@ -263,7 +264,7 @@ def grid_sample(x,
cudnn_version = get_cudnn_version()
use_cudnn = False
if not core.is_compiled_with_rocm() and (
if not is_compiled_with_rocm() and (
cudnn_version is not None
) and align_corners and mode == 'bilinear' and padding_mode == 'zeros':
use_cudnn = True
......@@ -271,7 +272,7 @@ def grid_sample(x,
x.stop_gradient = False
grid.stop_gradient = False
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('mode', mode, 'padding_mode', padding_mode, 'align_corners',
align_corners, 'use_cudnn', use_cudnn)
out = getattr(_C_ops, 'grid_sampler')(x, grid, *attrs)
......@@ -329,7 +330,7 @@ def pixel_shuffle(x, upscale_factor, data_format="NCHW", name=None):
"But recevie Attr(data_format): {} ".format(
data_format))
if in_dygraph_mode():
if in_dynamic_mode():
return _C_ops.pixel_shuffle(x, "upscale_factor", upscale_factor,
"data_format", data_format)
......
python/paddle/nn/initializer/assign.py
浏览文件 @ a710738e
......@@ -11,11 +11,7 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ...fluid import framework
from ...fluid import core
from ...fluid import unique_name
from ...fluid.core import VarDesc
import paddle
from ...fluid.data_feeder import check_type
from ...fluid.initializer import NumpyArrayInitializer
......@@ -88,13 +84,14 @@ class Assign(NumpyArrayInitializer):
def __init__(self, value, name=None):
import numpy
check_type(value, 'value',
(numpy.ndarray, list, tuple, framework.Variable), 'Assign')
(numpy.ndarray, list, tuple, paddle.static.Variable),
'Assign')
if (isinstance(value, (list, tuple))):
value = numpy.array(value)
# TODO: value is already is a tensor, accounting efficiency maybe it does not need to convert tensor to numpy data and then initialized.
if (isinstance(value, framework.Variable)):
if (isinstance(value, paddle.static.Variable)):
value = value.numpy()
super(Assign, self).__init__(value)
python/paddle/nn/initializer/dirac.py
浏览文件 @ a710738e
......@@ -15,7 +15,9 @@
from ...fluid.initializer import Initializer
from ...fluid.data_feeder import check_variable_and_dtype
from ...fluid.core import VarDesc
from ...fluid import unique_name, framework
from ...fluid import framework
from paddle import in_dynamic_mode
from paddle.utils import unique_name
__all__ = []
......@@ -221,6 +223,6 @@ class Dirac(Initializer):
"out_dtype": var.dtype},
stop_gradient=True)
if not framework.in_dygraph_mode():
if not in_dynamic_mode():
var.op = op
return op
python/paddle/nn/initializer/orthogonal.py
浏览文件 @ a710738e
......@@ -14,9 +14,9 @@
from ...fluid.initializer import Initializer
from ...fluid.data_feeder import check_variable_and_dtype
from ...fluid.core import VarDesc
from ...fluid import unique_name, framework
from ...fluid import framework
from ...tensor import diag, transpose, sign, qr, reshape
from paddle.utils import unique_name
__all__ = []
......
python/paddle/nn/layer/activation.py
浏览文件 @ a710738e
......@@ -14,8 +14,6 @@
# TODO: define activation functions of neural network
from ...fluid import core
from ...fluid.framework import in_dygraph_mode
from ...framework import ParamAttr
from ..initializer import Constant
from paddle.framework import get_default_dtype
......
python/paddle/nn/layer/common.py
浏览文件 @ a710738e
......@@ -15,10 +15,10 @@
# TODO: define the common classes to build a neural network
import paddle
from ...fluid.dygraph import Flatten # noqa: F401
from ...fluid.framework import in_dygraph_mode
from .. import functional as F
from ...fluid.framework import _dygraph_tracer
from paddle.nn import Layer
from paddle import in_dynamic_mode
__all__ = []
......@@ -1456,7 +1456,7 @@ class Embedding(Layer):
dtype=self._dtype,
is_bias=False)
if in_dygraph_mode() and padding_idx != -1:
if in_dynamic_mode() and padding_idx != -1:
with paddle.no_grad():
self.weight[padding_idx] = 0.0
......
python/paddle/nn/layer/conv.py
浏览文件 @ a710738e
......@@ -16,14 +16,15 @@
import numpy as np
from ...fluid import get_flags
from ...fluid import core
from paddle import get_flags
from ...device import get_cudnn_version
from .. import Layer
from ..initializer import Normal
from .. import functional as F
from ...fluid.layers import utils
from ..functional.conv import _update_padding_nd
from ...device import is_compiled_with_cuda
from ...device import is_compiled_with_rocm
__all__ = []
......@@ -138,7 +139,7 @@ class _ConvNd(Layer):
cudnn_version = get_cudnn_version()
self._use_cudnn = True if (core.is_compiled_with_cuda() and
self._use_cudnn = True if (is_compiled_with_cuda() and
cudnn_version is not None) else False
self._op_type = "conv" + str(dims) + 'd'
......@@ -146,13 +147,13 @@ class _ConvNd(Layer):
in_channels != 1 and
out_channels % in_channels == 0):
self._op_type = 'depthwise_conv2d'
if core.is_compiled_with_rocm():
if is_compiled_with_rocm():
self._use_cudnn = True
else:
self._use_cudnn = False
if (core.is_compiled_with_cuda() and get_flags(
"FLAGS_conv2d_disable_cudnn")["FLAGS_conv2d_disable_cudnn"]):
if (is_compiled_with_cuda() and get_flags("FLAGS_conv2d_disable_cudnn")[
"FLAGS_conv2d_disable_cudnn"]):
self._use_cudnn = False
def extra_repr(self):
......
python/paddle/nn/layer/distance.py
浏览文件 @ a710738e
......@@ -16,10 +16,10 @@ import numpy as np
import paddle
from .. import Layer
from ...fluid.framework import core, in_dygraph_mode
from ...fluid.data_feeder import check_variable_and_dtype, check_type
from ...fluid.layer_helper import LayerHelper
from paddle import _C_ops
from paddle import in_dynamic_mode
__all__ = []
......@@ -78,7 +78,7 @@ class PairwiseDistance(Layer):
check_type(self.keepdim, 'keepdim', (bool), 'PairwiseDistance')
def forward(self, x, y):
if in_dygraph_mode():
if in_dynamic_mode():
sub = _C_ops.elementwise_sub(x, y)
return _C_ops.p_norm(sub, 'axis', 1, 'porder', self.p, 'keepdim',
self.keepdim, 'epsilon', self.epsilon)
......
python/paddle/nn/layer/loss.py
浏览文件 @ a710738e
......@@ -16,11 +16,11 @@
# TODO: define loss functions of neural network
import numpy as np
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle
from .. import functional as F
from paddle.fluid.framework import core, in_dygraph_mode, _varbase_creator
from paddle.fluid.framework import _varbase_creator
from .. import Layer
from paddle import in_dynamic_mode
__all__ = []
......@@ -591,7 +591,7 @@ class MSELoss(Layer):
self.reduction = reduction
def forward(self, input, label):
if not fluid.framework.in_dygraph_mode():
if not in_dynamic_mode():
fluid.data_feeder.check_variable_and_dtype(
input, 'input', ['float32', 'float64'], 'MSELoss')
fluid.data_feeder.check_variable_and_dtype(
......
python/paddle/nn/layer/norm.py
浏览文件 @ a710738e
......@@ -33,12 +33,11 @@ from ...fluid.dygraph import BatchNorm # noqa: F401
from ...fluid.dygraph import SpectralNorm # noqa: F401
from ...framework import get_default_dtype, set_default_dtype
from ...fluid.framework import in_dygraph_mode
from ..initializer import Constant
from ...framework import ParamAttr
from ...fluid.data_feeder import check_variable_and_dtype, check_type
from ...fluid import core, dygraph_utils
from ...fluid import dygraph_utils
from ..functional import batch_norm, layer_norm, instance_norm
......@@ -49,6 +48,7 @@ from ...framework import no_grad
from .. import functional as F
from paddle import _C_ops
from .. import Layer
from paddle import in_dynamic_mode
__all__ = []
......@@ -1087,7 +1087,7 @@ class SyncBatchNorm(_BatchNormBase):
### train mode: use mini-batch stats, eval mode: use global stats
### use_global_stats only support False in sync_batch_norm
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ("momentum", self._momentum, "epsilon", self._epsilon,
"is_test", not self.training, "data_layout",
self._data_format, "use_mkldnn", False, "fuse_with_relu",
......
python/paddle/nn/layer/rnn.py
浏览文件 @ a710738e
......@@ -33,6 +33,11 @@ from paddle.fluid.layers import utils
from paddle.fluid.layers.utils import map_structure, flatten, pack_sequence_as
from paddle.fluid.data_feeder import convert_dtype
from paddle import _C_ops
from paddle import in_dynamic_mode
from paddle.framework import core
from paddle.static import default_startup_program
from paddle.static import program_guard
__all__ = []
......@@ -970,8 +975,8 @@ class RNNBase(LayerList):
# dropout state may also can be hided and avoid saving
# should dropout state be persistable for static-graph
self._dropout_state = self.create_variable(
dtype=fluid.core.VarDesc.VarType.UINT8)
if fluid.framework.in_dygraph_mode():
dtype=core.VarDesc.VarType.UINT8)
if in_dynamic_mode():
with paddle.no_grad():
_C_ops.coalesce_tensor(self._all_weights, self._all_weights,
self._flat_weight[0], "copy_data",
......@@ -979,8 +984,8 @@ class RNNBase(LayerList):
params[0].dtype)
return
# for static-graph, append coalesce_tensor into startup program
with fluid.program_guard(fluid.default_startup_program(),
fluid.default_startup_program()):
with program_guard(default_startup_program(),
default_startup_program()):
with paddle.no_grad():
self._helper.append_op(
type="coalesce_tensor",
......@@ -999,7 +1004,7 @@ class RNNBase(LayerList):
if not self.time_major:
inputs = paddle.tensor.transpose(inputs, [1, 0, 2])
if fluid.framework.in_dygraph_mode():
if in_dynamic_mode():
_, _, out, state = _C_ops.rnn(
inputs, initial_states, self._all_weights, sequence_length,
self._dropout_state, self.state_components, 'dropout_prob',
......@@ -1014,7 +1019,7 @@ class RNNBase(LayerList):
for i in range(self.state_components)
]
reserve = self._helper.create_variable_for_type_inference(
dtype=fluid.core.VarDesc.VarType.UINT8, stop_gradient=True)
dtype=core.VarDesc.VarType.UINT8, stop_gradient=True)
inputs = {
'Input': inputs,
......
python/paddle/nn/quant/functional_layers.py
浏览文件 @ a710738e
......@@ -12,13 +12,13 @@
# See the License for the specific language governing permissions and
# limitations under the License.
from ...fluid.dygraph import layers
from ...tensor import math, manipulation
from .. import Layer
__all__ = []
class FloatFunctionalLayer(layers.Layer):
class FloatFunctionalLayer(Layer):
def __init__(self):
super(FloatFunctionalLayer, self).__init__()
......
python/paddle/nn/quant/quant_layers.py
浏览文件 @ a710738e
......@@ -12,19 +12,19 @@
# See the License for the specific language governing permissions and
# limitations under the License.
from paddle.fluid.dygraph import layers
from paddle.fluid import core
from paddle.framework import core
from paddle.fluid import dygraph_utils
from paddle.fluid import unique_name
from paddle.fluid.param_attr import ParamAttr
from paddle.utils import unique_name
from paddle.framework import ParamAttr
from paddle.fluid.framework import _varbase_creator
from paddle.fluid.framework import in_dygraph_mode
from paddle.fluid.initializer import Constant
from paddle.nn.initializer import Constant
from paddle.fluid.data_feeder import check_variable_and_dtype
from paddle.nn import functional as F
import logging
from paddle.fluid.log_helper import get_logger
from paddle import _C_ops
from paddle import in_dynamic_mode
from paddle.nn import Layer
__all__ = [
'FakeQuantAbsMax',
......@@ -43,7 +43,7 @@ _logger = get_logger(
__name__, logging.INFO, fmt='%(asctime)s-%(levelname)s: %(message)s')
class FakeQuantAbsMax(layers.Layer):
class FakeQuantAbsMax(Layer):
r"""
FakeQuantAbsMax layer does the abs_max quant and then dequant.
Its computational formula is described as below:
......@@ -76,7 +76,7 @@ class FakeQuantAbsMax(layers.Layer):
self._scale = None
def forward(self, input):
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('bit_length', self._quant_bits)
quant_out = _varbase_creator(
type=input.type,
......@@ -125,7 +125,7 @@ class FakeQuantAbsMax(layers.Layer):
return quant_out
class FakeQuantMovingAverageAbsMax(layers.Layer):
class FakeQuantMovingAverageAbsMax(Layer):
r"""
FakeQuantMovingAverageAbsMax layer does the moving_average_abs_max quant and then dequant.
Its computational formula is described as below:
......@@ -175,7 +175,7 @@ class FakeQuantMovingAverageAbsMax(layers.Layer):
self._accum.stop_gradient = True
def forward(self, input):
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('moving_rate', self._moving_rate, 'bit_length',
self._quant_bits, 'is_test', not self.training)
quant_out = _varbase_creator(
......@@ -223,7 +223,7 @@ class FakeQuantMovingAverageAbsMax(layers.Layer):
return quant_out
class FakeQuantChannelWiseAbsMax(layers.Layer):
class FakeQuantChannelWiseAbsMax(Layer):
def __init__(self,
name=None,
channel_num=None,
......@@ -253,7 +253,7 @@ class FakeQuantChannelWiseAbsMax(layers.Layer):
self._scale = None
def forward(self, input):
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('bit_length', self._quant_bits, 'quant_axis',
self._quant_axis)
quant_out = _varbase_creator(
......@@ -306,7 +306,7 @@ class FakeQuantChannelWiseAbsMax(layers.Layer):
return quant_out
class MovingAverageAbsMaxScale(layers.Layer):
class MovingAverageAbsMaxScale(Layer):
def __init__(self, name=None, moving_rate=0.9, dtype='float32'):
r"""
MovingAverageMaxScale layer is used to calculating the output quantization
......@@ -345,7 +345,7 @@ class MovingAverageAbsMaxScale(layers.Layer):
self._accum.stop_gradient = True
def forward(self, input):
if in_dygraph_mode():
if in_dynamic_mode():
attrs = ('moving_rate', self._moving_rate, 'is_test',
not self.training)
state = self._state if self.training else None
......@@ -393,7 +393,7 @@ class MovingAverageAbsMaxScale(layers.Layer):
QuantStub = MovingAverageAbsMaxScale
class QuantizedConv2D(layers.Layer):
class QuantizedConv2D(Layer):
"""
The computational logic of QuantizedConv2D is the same with Conv2D.
The only difference is that its inputs are all fake quantized.
......@@ -482,7 +482,7 @@ class QuantizedConv2D(layers.Layer):
data_format=self._data_format)
class QuantizedConv2DTranspose(layers.Layer):
class QuantizedConv2DTranspose(Layer):
"""
The computational logic of QuantizedConv2DTranspose is the same with Conv2DTranspose.
The only difference is that its inputs are all fake quantized.
......@@ -588,7 +588,7 @@ class QuantizedConv2DTranspose(layers.Layer):
data_format=self._data_format)
class QuantizedLinear(layers.Layer):
class QuantizedLinear(Layer):
"""
The computational logic of QuantizedLinear is the same with Linear.
The only difference is that its inputs are all fake quantized.
......@@ -657,7 +657,7 @@ class QuantizedLinear(layers.Layer):
return out
class MAOutputScaleLayer(layers.Layer):
class MAOutputScaleLayer(Layer):
"""
Add MovingAverageMaxScale layer to the behind of the input layer.
Calculate the scale (moving average abs max) for the output of the input layer.
......@@ -684,7 +684,7 @@ class MAOutputScaleLayer(layers.Layer):
return self._ma_output_scale(out)
class FakeQuantMAOutputScaleLayer(layers.Layer):
class FakeQuantMAOutputScaleLayer(Layer):
"""
Add FakeQuantMovingAverageAbsMax layer to the behind of the input layer.
"""
......
python/paddle/nn/utils/weight_norm_hook.py
浏览文件 @ a710738e
......@@ -11,7 +11,7 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import paddle
import numpy as np
from ... import fluid
from ...fluid import dygraph
......@@ -39,25 +39,25 @@ def l2_norm(x, axis, epsilon=1e-12, name=None):
"axis": 1 if axis is None else axis,
"epsilon": epsilon,
})
return F.squeeze(norm, axes=[axis])
return paddle.squeeze(norm, axis=[axis])
def norm_except_dim(p, dim):
shape = p.shape
ndims = len(shape)
if dim == -1:
return F.sqrt(F.reduce_sum(F.square(p)) + 1e-12)
return paddle.sqrt(paddle.sum(paddle.square(p)) + 1e-12)
elif dim == 0:
p_matrix = F.reshape(p, (shape[0], -1))
p_matrix = paddle.reshape(p, (shape[0], -1))
return l2_norm(p_matrix, axis=1)
elif dim == ndims - 1:
p_matrix = F.reshape(p, (-1, shape[-1]))
p_matrix = paddle.reshape(p, (-1, shape[-1]))
return l2_norm(p_matrix, axis=0)
else:
perm = list(range(ndims))
perm[0] = dim
perm[dim] = 0
p_transposed = F.transpose(p, perm)
p_transposed = paddle.transpose(p, perm)
return norm_except_dim(p_transposed, 0)
......@@ -66,25 +66,25 @@ def _weight_norm(v, g, dim):
ndims = len(shape)
if dim == -1:
v_normalized = v / (F.sqrt(F.reduce_sum(F.square(v))) + 1e-12)
v_normalized = v / (paddle.sqrt(paddle.sum(paddle.square(v))) + 1e-12)
elif dim == 0:
p_matrix = F.reshape(v, (shape[0], -1))
p_matrix = paddle.reshape(v, (shape[0], -1))
v_normalized = F.l2_normalize(p_matrix, axis=1)
v_normalized = F.reshape(v_normalized, shape)
v_normalized = paddle.reshape(v_normalized, shape)
elif dim == ndims - 1:
p_matrix = F.reshape(v, (-1, shape[-1]))
p_matrix = paddle.reshape(v, (-1, shape[-1]))
v_normalized = F.l2_normalize(p_matrix, axis=0)
v_normalized = F.reshape(v_normalized, shape)
v_normalized = paddle.reshape(v_normalized, shape)
else:
perm = list(range(ndims))
perm[0] = dim
perm[dim] = 0
p_transposed = F.transpose(v, perm)
p_transposed = paddle.transpose(v, perm)
transposed_shape = p_transposed.shape
p_matrix = F.reshape(p_transposed, (p_transposed.shape[0], -1))
p_matrix = paddle.reshape(p_transposed, (p_transposed.shape[0], -1))
v_normalized = F.l2_normalize(p_matrix, axis=1)
v_normalized = F.reshape(v_normalized, transposed_shape)
v_normalized = F.transpose(v_normalized, perm)
v_normalized = paddle.reshape(v_normalized, transposed_shape)
v_normalized = paddle.transpose(v_normalized, perm)
weight = F.elementwise_mul(
v_normalized, g, axis=dim if dim is not None else -1)
return weight
......@@ -130,9 +130,9 @@ class WeightNorm(object):
layer.add_parameter(name + "_v", v)
g = layer.create_parameter(g_var.shape, dtype=g_var.dtype)
layer.add_parameter(name + '_g', g)
with dygraph.no_grad():
F.assign(w, v)
F.assign(g_var, g)
with paddle.no_grad():
paddle.assign(w, v)
paddle.assign(g_var, g)
setattr(layer, name, fn.compute_weight(layer))
layer.register_forward_pre_hook(fn)
......@@ -145,8 +145,8 @@ class WeightNorm(object):
del layer._parameters[self.name + '_v']
w = layer.create_parameter(w_var.shape, dtype=w_var.dtype)
layer.add_parameter(self.name, w)
with dygraph.no_grad():
F.assign(w_var, w)
with paddle.no_grad():
paddle.assign(w_var, w)
def __call__(self, layer, inputs):
setattr(layer, self.name, self.compute_weight(layer))
......
python/paddle/tensor/manipulation.py
浏览文件 @ a710738e
......@@ -30,6 +30,7 @@ from ..fluid.layers import unstack # noqa: F401
from ..fluid.layers import scatter_nd # noqa: F401
from ..fluid.layers import shard_index # noqa: F401
from ..fluid.layers import crop_tensor as crop # noqa: F401
from ..fluid.layers.nn import _elementwise_op_in_dygraph
from ..fluid import layers
from ..fluid.dygraph.inplace_utils import inplace_apis_in_dygraph_only
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或