add sliding window parameter to all layers

modeling_bert.py

@@ -1510,6 +1510,7 @@ class JinaBertForPreTraining(JinaBertPreTrainedModel):
             output_attentions: Optional[bool] = None,
             output_hidden_states: Optional[bool] = None,
             return_dict: Optional[bool] = None,
+            sliding_window: Optional[int] = None,
     ) -> Union[Tuple[torch.Tensor], JinaBertForPreTrainingOutput]:
         r"""
             labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
@@ -1541,6 +1542,7 @@ class JinaBertForPreTraining(JinaBertPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
+            sliding_window=sliding_window
         )
 
         sequence_output, pooled_output = outputs[:2]
@@ -1783,6 +1785,7 @@ class JinaBertForMaskedLM(JinaBertPreTrainedModel):
             output_attentions: Optional[bool] = None,
             output_hidden_states: Optional[bool] = None,
             return_dict: Optional[bool] = None,
+            sliding_window: Optional[int] = None,
     ) -> Union[Tuple[torch.Tensor], MaskedLMOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
@@ -1807,6 +1810,7 @@ class JinaBertForMaskedLM(JinaBertPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
+            sliding_window=sliding_window
         )
 
         sequence_output = outputs[0]