Update modeling_mcqbert.py

modeling_mcqbert.py

@@ -24,23 +24,19 @@ class MCQStudentBert(BertModel):
     def forward(self, input_ids, student_embeddings=None):
         if self.config.integration_strategy is None:
             # don't consider embeddings is no integration strategy (MCQBert)
-            student_embeddings = torch.zeros(self.config.student_embedding_layer)
-
-            input_embeddings = self.embeddings(input_ids)
-            combined_embeddings = input_embeddings + self.student_embedding_layer(student_embeddings).unsqueeze(1).repeat(1, input_embeddings.size(1), 1)
-            output = super().forward(inputs_embeds = combined_embeddings)
+            output = super().forward(input_ids)
             return self.classifier(output.last_hidden_state[:, 0, :])
 
         elif self.config.integration_strategy == "cat":
             # MCQStudentBertCat
             output = super().forward(input_ids)
-            output_with_student_embedding = torch.cat((output.last_hidden_state[:, 0, :], self.student_embedding_layer(student_embeddings)), dim = 1)
+            output_with_student_embedding = torch.cat((output.last_hidden_state[:, 0, :], self.student_embedding_layer(student_embeddings).unsqueeze(0)), dim = 1)
             return self.classifier(output_with_student_embedding)
 
         elif self.config.integration_strategy == "sum":
             # MCQStudentBertSum
             input_embeddings = self.embeddings(input_ids)
-            combined_embeddings = input_embeddings + self.student_embedding_layer(student_embeddings).unsqueeze(1).repeat(1, input_embeddings.size(1), 1)
+            combined_embeddings = input_embeddings + self.student_embedding_layer(student_embeddings).repeat(1, input_embeddings.size(1), 1)
             output = super().forward(inputs_embeds = combined_embeddings)
             return self.classifier(output.last_hidden_state[:, 0, :])