fix: forward method

Author: sigridjineth

src/tevatron/reranker/modeling.py

@@ -1,6 +1,6 @@
 import os
 from dataclasses import dataclass
-from typing import Dict, Optional
+from typing import Dict, Optional, Union
 
 import torch
 from torch import nn, Tensor
@@ -43,8 +43,16 @@ def __init__(self, hf_model: PreTrainedModel, train_batch_size: int = None):
                 param.data = nn.Linear(self.hf_model.config.hidden_size, 1).weight.data
                 logger.warning('{} data: {}'.format(name, param.data.cpu().numpy()))
 
-    def forward(self, pair: Dict[str, Tensor] = None):
-        ranker_logits = self.hf_model(**pair, return_dict=True).logits
+    def forward(self, input_ids: Tensor = None, attention_mask: Tensor = None, token_type_ids: Tensor = None, **kwargs):
+        model_inputs = {
+            'input_ids': input_ids,
+            'attention_mask': attention_mask,
+        }
+        if token_type_ids is not None:
+            model_inputs['token_type_ids'] = token_type_ids
+
+        ranker_logits = self.hf_model(**model_inputs, return_dict=True).logits
+
         if self.train_batch_size:
             grouped_logits = ranker_logits.view(self.train_batch_size, -1)
             loss = self.cross_entropy(grouped_logits, self.target_label)
@@ -60,7 +68,6 @@ def forward(self, pair: Dict[str, Tensor] = None):
 
     def gradient_checkpointing_enable(self, **kwargs):
         return False
-        # self.hf_model.base_model.model.gradient_checkpointing_enable(**kwargs)
 
     @classmethod
     def build(