fix truncate maxlength

app/ner/services/ner.py

@@ -17,14 +17,14 @@ class NER:
         self.pipeline = pipeline(task=settings.TASK_NAME, 
                                  model=self.model, 
                                  tokenizer=self.tokenizer,
-                                 truncation=settings.TRUNCATE, 
-                                 max_length=settings.MAX_LENGTH,
                                  device=settings.DEVICE)
         
     async def predict(self, text: str, entity_tag: str = None):
         if self.pipeline is None:
             raise ValueError("Model not loaded. Please call load_model() first.")
-        pred = self.pipeline(text)
+        pred = self.pipeline(text, 
+                             truncation=settings.TRUNCATE,
+                             max_length=settings.MAX_LENGTH)
         if entity_tag:
             return self.extract_entities(pred, entity_tag)
         return pred