PoliBa-DeepLearning/NanoSocrates
2
0
Fork 0
Code Issues 3 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Activated Dropout to avoid overfitting

Browse Source
This commit is contained in:
Christian Risi 2025-10-12 12:28:06 +02:00
parent f463f699cf
commit 4ca1d0a189
3 changed files with 22 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
Project_Model/Libs/Transformer/Classes/Decoder.py
View File

@ -68,12 +68,12 @@ class Decoder(nn.Module):
)
# 2) Dropout
# DROPPED_MASKED_ATTENTION = self.__dropout(MASKED_ATTENTION)
# del MASKED_ATTENTION
DROPPED_MASKED_ATTENTION = self.__dropout(MASKED_ATTENTION)
del MASKED_ATTENTION
# 3) Residual Connection
x = x + MASKED_ATTENTION
del MASKED_ATTENTION
x = x + DROPPED_MASKED_ATTENTION
del DROPPED_MASKED_ATTENTION
# 4) Layer Normalization
x = self.__layer_norm_1(x)
@ -86,12 +86,12 @@ class Decoder(nn.Module):
)
# 6) Dropout
# DROPPED_CROSS_ATTENTION = self.__dropout(CROSS_ATTENTION)
# del CROSS_ATTENTION
DROPPED_CROSS_ATTENTION = self.__dropout(CROSS_ATTENTION)
del CROSS_ATTENTION
# 7) Residual Connection
x = x + CROSS_ATTENTION
del CROSS_ATTENTION
x = x + DROPPED_CROSS_ATTENTION
del DROPPED_CROSS_ATTENTION
# 8) Layer Normalization
x = self.__layer_norm_2(x)
@ -100,12 +100,12 @@ class Decoder(nn.Module):
FEED_FORWARD = self.__feed_forward_network(x)
# 10) Dropout
# DROPPED_FEED_FORWARD = self.__dropout(FEED_FORWARD)
# del FEED_FORWARD
DROPPED_FEED_FORWARD = self.__dropout(FEED_FORWARD)
del FEED_FORWARD
# 11) Residual Connection
x = x + FEED_FORWARD
del FEED_FORWARD
x = x + DROPPED_FEED_FORWARD
del DROPPED_FEED_FORWARD
# 12) Layer Normalization
x = self.__layer_norm_3(x)

16
Project_Model/Libs/Transformer/Classes/Encoder.py
View File

@ -43,12 +43,12 @@ class Encoder(
ATTENTION = self.__attention(x, x, x, key_padding_mask=padding_mask)
# 2) Dropout
# DROPPED_ATTENTION = self.__dropout(ATTENTION)
# del ATTENTION
DROPPED_ATTENTION = self.__dropout(ATTENTION)
del ATTENTION
# 3) Residual Connection
x = x + ATTENTION
del ATTENTION
x = x + DROPPED_ATTENTION
del DROPPED_ATTENTION
# 4) Layer Normalization
x = self.__layer_norm_1(x)
@ -57,12 +57,12 @@ class Encoder(
FEED_FORWARD = self.__feed_forward(x)
# 6) Dropout
# DROPPED_FEED_FORWARD = self.__dropout(FEED_FORWARD)
# del FEED_FORWARD
DROPPED_FEED_FORWARD = self.__dropout(FEED_FORWARD)
del FEED_FORWARD
# 7) Residual Connection
x = x + FEED_FORWARD
del FEED_FORWARD
x = x + DROPPED_FEED_FORWARD
del DROPPED_FEED_FORWARD
# 8) Layer Normalization
x = self.__layer_norm_2(x)

2
Project_Model/Libs/Transformer/Classes/FeedForwardNetwork.py
View File

@ -35,7 +35,7 @@ class FeedForwardNetwork(nn.Module):
x = self.__relu(x)
# 3) Dropout
# x = self.__dropout(x)
x = self.__dropout(x)
# 4) Linear Layer
x = self.__fully_connected_2(x)