testadmin
/
ParakeetRebeccaRosario
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge branch 'fix' into 'master' 

fix the bug 1/x=0 in python2

See merge request !41
This commit is contained in:
liuyibing01 2020-03-10 14:56:21 +08:00
parent f95142d608 e720592ac1
commit 22cf0b82e8
9 changed files with 21 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
parakeet/models/fastspeech/fastspeech.py
View File

@ -62,7 +62,7 @@ class FastSpeech(dg.Layer):
dropout=0.1)
self.weight = fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer())
k = math.sqrt(1 / cfg['fs_hidden_size'])
k = math.sqrt(1.0 / cfg['fs_hidden_size'])
self.bias = fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k))
self.mel_linear = dg.Linear(

6
parakeet/models/fastspeech/length_regulator.py
View File

@ -115,7 +115,7 @@ class DurationPredictor(dg.Layer):
self.filter_size = filter_size
self.dropout = dropout
k = math.sqrt(1 / self.input_size)
k = math.sqrt(1.0 / self.input_size)
self.conv1 = Conv1D(
num_channels=self.input_size,
num_filters=self.out_channels,
@ -126,7 +126,7 @@ class DurationPredictor(dg.Layer):
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
#data_format='NTC')
k = math.sqrt(1 / self.out_channels)
k = math.sqrt(1.0 / self.out_channels)
self.conv2 = Conv1D(
num_channels=self.out_channels,
num_filters=self.out_channels,
@ -142,7 +142,7 @@ class DurationPredictor(dg.Layer):
self.weight = fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer())
k = math.sqrt(1 / self.out_channels)
k = math.sqrt(1.0 / self.out_channels)
self.bias = fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k))

12
parakeet/models/transformer_tts/cbhg.py
View File

@ -46,7 +46,7 @@ class CBHG(dg.Layer):
self.hidden_size = hidden_size
self.projection_size = projection_size
self.conv_list = []
k = math.sqrt(1 / projection_size)
k = math.sqrt(1.0 / projection_size)
self.conv_list.append(
Conv1D(
num_channels=projection_size,
@ -58,7 +58,7 @@ class CBHG(dg.Layer):
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(
low=-k, high=k))))
k = math.sqrt(1 / hidden_size)
k = math.sqrt(1.0 / hidden_size)
for i in range(2, K + 1):
self.conv_list.append(
Conv1D(
@ -86,7 +86,7 @@ class CBHG(dg.Layer):
conv_outdim = hidden_size * K
k = math.sqrt(1 / conv_outdim)
k = math.sqrt(1.0 / conv_outdim)
self.conv_projection_1 = Conv1D(
num_channels=conv_outdim,
num_filters=hidden_size,
@ -97,7 +97,7 @@ class CBHG(dg.Layer):
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
k = math.sqrt(1 / hidden_size)
k = math.sqrt(1.0 / hidden_size)
self.conv_projection_2 = Conv1D(
num_channels=hidden_size,
num_filters=projection_size,
@ -121,7 +121,7 @@ class CBHG(dg.Layer):
h_0 = np.zeros((batch_size, hidden_size // 2), dtype="float32")
h_0 = dg.to_variable(h_0)
k = math.sqrt(1 / hidden_size)
k = math.sqrt(1.0 / hidden_size)
self.fc_forward1 = dg.Linear(
hidden_size,
hidden_size // 2 * 3,
@ -240,7 +240,7 @@ class Highwaynet(dg.Layer):
self.gates = []
self.linears = []
k = math.sqrt(1 / num_units)
k = math.sqrt(1.0 / num_units)
for i in range(num_layers):
self.linears.append(
dg.Linear(

2
parakeet/models/transformer_tts/decoder.py
View File

@ -55,7 +55,7 @@ class Decoder(dg.Layer):
hidden_size=num_hidden * 2,
output_size=num_hidden,
dropout_rate=0.2)
k = math.sqrt(1 / num_hidden)
k = math.sqrt(1.0 / num_hidden)
self.linear = dg.Linear(
num_hidden,
num_hidden,

6
parakeet/models/transformer_tts/encoderprenet.py
View File

@ -39,7 +39,7 @@ class EncoderPrenet(dg.Layer):
param_attr=fluid.initializer.Normal(
loc=0.0, scale=1.0))
self.conv_list = []
k = math.sqrt(1 / embedding_size)
k = math.sqrt(1.0 / embedding_size)
self.conv_list.append(
Conv1D(
num_channels=embedding_size,
@ -52,7 +52,7 @@ class EncoderPrenet(dg.Layer):
initializer=fluid.initializer.Uniform(
low=-k, high=k)),
use_cudnn=use_cudnn))
k = math.sqrt(1 / num_hidden)
k = math.sqrt(1.0 / num_hidden)
for _ in range(2):
self.conv_list.append(
Conv1D(
@ -78,7 +78,7 @@ class EncoderPrenet(dg.Layer):
for i, layer in enumerate(self.batch_norm_list):
self.add_sublayer("batch_norm_list_{}".format(i), layer)
k = math.sqrt(1 / num_hidden)
k = math.sqrt(1.0 / num_hidden)
self.projection = dg.Linear(
num_hidden,
num_hidden,

4
parakeet/models/transformer_tts/post_convnet.py
View File

@ -48,7 +48,7 @@ class PostConvNet(dg.Layer):
self.num_conv = num_conv
self.batchnorm_last = batchnorm_last
self.conv_list = []
k = math.sqrt(1 / (n_mels * outputs_per_step))
k = math.sqrt(1.0 / (n_mels * outputs_per_step))
self.conv_list.append(
Conv1D(
num_channels=n_mels * outputs_per_step,
@ -62,7 +62,7 @@ class PostConvNet(dg.Layer):
low=-k, high=k)),
use_cudnn=use_cudnn))
k = math.sqrt(1 / num_hidden)
k = math.sqrt(1.0 / num_hidden)
for _ in range(1, num_conv - 1):
self.conv_list.append(
Conv1D(

4
parakeet/models/transformer_tts/prenet.py
View File

@ -33,7 +33,7 @@ class PreNet(dg.Layer):
self.output_size = output_size
self.dropout_rate = dropout_rate
k = math.sqrt(1 / input_size)
k = math.sqrt(1.0 / input_size)
self.linear1 = dg.Linear(
input_size,
hidden_size,
@ -41,7 +41,7 @@ class PreNet(dg.Layer):
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
k = math.sqrt(1 / hidden_size)
k = math.sqrt(1.0 / hidden_size)
self.linear2 = dg.Linear(
hidden_size,
output_size,

4
parakeet/modules/ffn.py
View File

@ -41,7 +41,7 @@ class PositionwiseFeedForward(dg.Layer):
self.use_cudnn = use_cudnn
self.dropout = dropout
k = math.sqrt(1 / d_in)
k = math.sqrt(1.0 / d_in)
self.w_1 = Conv1D(
num_channels=d_in,
num_filters=num_hidden,
@ -52,7 +52,7 @@ class PositionwiseFeedForward(dg.Layer):
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)),
use_cudnn=use_cudnn)
k = math.sqrt(1 / num_hidden)
k = math.sqrt(1.0 / num_hidden)
self.w_2 = Conv1D(
num_channels=num_hidden,
num_filters=d_in,

2
parakeet/modules/multihead_attention.py
View File

@ -33,7 +33,7 @@ class Linear(dg.Layer):
self.bias = is_bias
if is_bias is not False:
k = math.sqrt(1 / in_features)
k = math.sqrt(1.0 / in_features)
self.bias = fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k))