update

model/claim_verification/joint_cross_encoder/model.py

@@ -151,12 +151,10 @@ def __init__(self,config:JointCrossEncoderConfig,):
             out_features=config.nclass,
             head_num=self.feature_extractor.extractor_config.num_attention_heads,
         )
-        self.positive_classify_linear = nn.Linear(in_features=self.feature_extractor.extractor_config.hidden_size, out_features=1)
 
     def forward(self, fact, is_positive):
         fact_embed = self.feature_extractor(fact)
         fact_embed = torch.reshape(fact_embed, shape=[-1, self.config.nins] + list(fact_embed.shape[1:])) # batch_size, num_evident, dim
-        positive_logits = self.positive_classify_linear(fact_embed).squeeze() # batch_size, n_evidents
 
         multi_evident_output = fact_embed
         for evident_aggrerator in self.evident_aggrerators:
@@ -166,7 +164,7 @@ def forward(self, fact, is_positive):
         single_evident_output = fact_embed[torch.arange(fact_embed.shape[0]).tolist(), is_positive.tolist(), :] # is positive is a 1-d tensor of id of positive sample (real sample) in every batch sample
         single_evident_logits = self.single_evident_linear(single_evident_output) # batch_size, n_labels
 
-        return multi_evident_logits, single_evident_logits, positive_logits
+        return multi_evident_logits, single_evident_logits
 
     def predict(
             self,

model/claim_verification/joint_cross_encoder/trainer.py

@@ -56,7 +56,7 @@ def __init__(
                 lora_alpha=32,
                 lora_dropout=0.1,
                 target_modules='feature_extractor.*.query_key_value|feature_extractor.*.dense|evident_aggrerators.*.out_proj',
-                modules_to_save=['aggerator', 'single_evident_linear', 'positive_classify_linear']
+                modules_to_save=['aggerator', 'single_evident_linear']
             )
             self.model = get_peft_model(self.model, peft_config)
             print('*********************')
@@ -187,7 +187,7 @@ def __call__(
             for fact_claims_ids, labels, is_positive, is_positive_ohot in tqdm(train_dataloader, desc="Iteration", smoothing=0.05, disable=not show_progress_bar):
                 optimizer.zero_grad()
                 with torch.cuda.amp.autocast(): ###########
-                    multi_evident_logits, single_evident_logits, positive_logits = self.model(fact_claims_ids, is_positive)
+                    multi_evident_logits, single_evident_logits = self.model(fact_claims_ids, is_positive)
                     multi_evident_loss_value = multi_loss_fct(multi_evident_logits, labels)
                     single_evident_loss_value = multi_loss_fct(single_evident_logits, labels)
                     # is_positive_loss_value = binary_loss_fct(positive_logits, is_positive_ohot)