lets try to change the pipeline

modeling_stacked.py

@@ -78,9 +78,9 @@ class ExtendedMultitaskModelForTokenClassification(PreTrainedModel):
         extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
         return extended_attention_mask
 
-    @property
-    def device(self):
-        return torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    # @property
+    # def device(self):
+    #     return torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
     @classmethod
     def from_pretrained(cls, *args, **kwargs):