add max new tokens to pipeline

model.py

@@ -34,6 +34,7 @@ def get_local_model(model_name_or_path:str)->pipeline:
         model=model,
         tokenizer=tokenizer,
         device = device,
+        max_new_tokens = 400,
         model_kwargs = {"max_length":16384, "max_new_tokens": 512},
     )