Update modeling_protst.py

modeling_protst.py

@@ -25,7 +25,7 @@ class BertTextRepresentationOutput(ModelOutput):
 
 
 @dataclass
-class EsmProteinClassificationOutput(ModelOutput):
+class ProtSTClassificationOutput(ModelOutput):
 
     loss: Optional[torch.FloatTensor] = None
     logits: torch.FloatTensor = None
@@ -125,7 +125,7 @@ class EsmForProteinRepresentation(EsmPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, EsmProteinClassificationOutput]:
+    ) -> Union[Tuple, EsmProteinRepresentationOutput]:
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -159,61 +159,6 @@ class EsmForProteinRepresentation(EsmPreTrainedModel):
         )
 
 
-class EsmForProteinPropertyPrediction(EsmPreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.model = EsmForProteinRepresentation(config)
-        self.classifier = ProtSTHead(config, out_dim=config.num_labels)
-
-        self.post_init() # NOTE
-
-    def forward(
-        self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        head_mask: Optional[torch.Tensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, EsmProteinClassificationOutput]:
-        r"""
-        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Labels for computing the protein classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
-        Returns:
-        Examples:
-        """
-        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
-        outputs = self.model(
-            input_ids,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-
-        logits = self.classifier(outputs.protein_feature) # [bsz, xxx] -> [bsz, num_labels]
-
-        loss = None
-        if labels is not None:
-            loss_fct = nn.CrossEntropyLoss()
-
-            labels = labels.to(logits.device)
-            loss = loss_fct(logits.view(-1, logits.shape[-1]), labels.view(-1))
-
-        if not return_dict:
-            output = (logits,)
-            return ((loss,) + output) if loss is not None else output
-
-        return EsmProteinClassificationOutput(loss=loss, logits=logits)
-
-
 class ProtSTPreTrainedModel(PreTrainedModel):
     config_class = ProtSTConfig
 
@@ -265,7 +210,7 @@ class ProtSTModel(ProtSTPreTrainedModel):
         self.protein_model = EsmForProteinRepresentation(config.protein_config)
         self.text_model = BertForPubMed(config.text_config)
         self.logit_scale = nn.Parameter(torch.ones([]) * math.log(1 / 0.07))
-        
+
         self.post_init() # NOTE
 
     def forward(self, 
@@ -280,3 +225,62 @@ class ProtSTModel(ProtSTPreTrainedModel):
     ):
         # Not implement yet
         return None
+
+
+class ProtSTForProteinPropertyPrediction(ProtSTPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.config = config
+        self.protein_model = EsmForProteinRepresentation(config.protein_config)
+        self.text_model = BertForPubMed(config.text_config)
+        self.logit_scale = nn.Parameter(torch.ones([]) * math.log(1 / 0.07))
+        self.classifier = ProtSTHead(config, out_dim=config.num_labels)
+
+        self.post_init() # NOTE
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, ProtSTClassificationOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the protein classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
+        Returns:
+        Examples:
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.protein_model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        logits = self.classifier(outputs.protein_feature) # [bsz, xxx] -> [bsz, num_labels]
+
+        loss = None
+        if labels is not None:
+            loss_fct = nn.CrossEntropyLoss()
+
+            labels = labels.to(logits.device)
+            loss = loss_fct(logits.view(-1, logits.shape[-1]), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,)
+            return ((loss,) + output) if loss is not None else output
+
+        return ProtSTClassificationOutput(loss=loss, logits=logits)