本文整理汇总了Python中tensor2tensor.layers.common_layers.sepconv_relu_sepconv方法的典型用法代码示例。如果您正苦于以下问题：Python common_layers.sepconv_relu_sepconv方法的具体用法？Python common_layers.sepconv_relu_sepconv怎么用？Python common_layers.sepconv_relu_sepconv使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类tensor2tensor.layers.common_layers的用法示例。
在下文中一共展示了common_layers.sepconv_relu_sepconv方法的2个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: transformer_decoder_ffn_unit
# 需要导入模块: from tensor2tensor.layers import common_layers [as 别名]
# 或者: from tensor2tensor.layers.common_layers import sepconv_relu_sepconv [as 别名]
def transformer_decoder_ffn_unit(x,
hparams,
nonpadding_mask=None):
"""Applies a feed-forward function which is parametrised for decoding.
Args:
x: input
hparams: model hyper-parameters
nonpadding_mask: optional Tensor with shape [batch_size, encoder_length]
indicating what positions are not padding. This is used
to mask out padding in convoltutional layers. We generally only
need this mask for "packed" datasets, because for ordinary datasets,
no padding is ever followed by nonpadding.
Returns:
the output tensor
"""
with tf.variable_scope("ffn"):
if hparams.transformer_ffn_type == "fc":
y = transformer.transformer_ffn_layer(
common_layers.layer_preprocess(x, hparams),
hparams,
conv_padding="LEFT",
nonpadding_mask=nonpadding_mask)
if hparams.transformer_ffn_type == "sepconv":
y = common_layers.sepconv_relu_sepconv(
common_layers.layer_preprocess(x, hparams),
filter_size=hparams.filter_size,
output_size=hparams.hidden_size,
first_kernel_size=(3, 1),
second_kernel_size=(5, 1),
padding="LEFT",
nonpadding_mask=nonpadding_mask,
dropout=hparams.relu_dropout)
x = common_layers.layer_postprocess(x, y, hparams)
return x 示例2: transformer_encoder_ffn_unit
# 需要导入模块: from tensor2tensor.layers import common_layers [as 别名]
# 或者: from tensor2tensor.layers.common_layers import sepconv_relu_sepconv [as 别名]
def transformer_encoder_ffn_unit(x,
hparams,
nonpadding_mask=None,
pad_remover=None):
"""Applies a feed-forward function which is parametrised for encoding.
Args:
x: input
hparams: model hyper-parameters
nonpadding_mask: optional Tensor with shape [batch_size, encoder_length]
indicating what positions are not padding. This is used
to mask out padding in convoltutional layers. We generally only
need this mask for "packed" datasets, because for ordinary datasets,
no padding is ever followed by nonpadding.
pad_remover: to mask out padding in convolutional layers (efficiency).
Returns:
the output tensor
"""
with tf.variable_scope("ffn"):
if hparams.transformer_ffn_type == "fc":
y = transformer.transformer_ffn_layer(
common_layers.layer_preprocess(x, hparams),
hparams,
pad_remover,
conv_padding="SAME",
nonpadding_mask=nonpadding_mask)
if hparams.transformer_ffn_type == "sepconv":
assert nonpadding_mask is not None, (
"The nonpadding_mask should be provided, otherwise the model uses "
"the leaked padding information to estimate the length!")
y = common_layers.sepconv_relu_sepconv(
common_layers.layer_preprocess(x, hparams),
filter_size=hparams.filter_size,
output_size=hparams.hidden_size,
first_kernel_size=(3, 1),
second_kernel_size=(5, 1),
padding="SAME",
nonpadding_mask=nonpadding_mask,
dropout=hparams.relu_dropout)
x = common_layers.layer_postprocess(x, y, hparams)
return x