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

Pipeline fix and added a util to decode

Browse Source
This commit is contained in:
Christian Risi
2025-10-09 13:24:48 +02:00
parent f3b83eda3d
commit aac7675b30
7 changed files with 78 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
Project_Model/Libs/Transformer/Models/TrainingModel.py
View File

@@ -37,17 +37,17 @@ class TrainingModel(torch.nn.Module):
self.__detokener = DeToken(latent_space, vocabulary_size)
def forward(self, args: tuple[torch.Tensor, torch.Tensor, torch.Tensor]):
def forward(self, args: tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]):
encoder_embedder_input, padding_tensor, decoder_embedder_input = args
encoder_embedder_input, src_padding, decoder_embedder_input, tgt_padding = args
encoder_tensor = self.__encoder_embedder(encoder_embedder_input)
decoder_tensor = self.__decoder_embedder(decoder_embedder_input)
encoder_output, _ = self.__encoder((encoder_tensor, padding_tensor))
encoder_output, _ = self.__encoder((encoder_tensor, src_padding))
decoder_output, _, _, _ = self.__decoder(
(decoder_tensor, encoder_output, encoder_output, None)
decoder_output, _, _, _, _ = self.__decoder(
(decoder_tensor, encoder_output, encoder_output, src_padding, tgt_padding)
)
logits: torch.Tensor = self.__detokener(decoder_output)