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
Download Patch File Download Diff File Expand all files Collapse all files

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)