Update helper.py

helper.py

@@ -26,7 +26,7 @@ def get_fn(model_path: str, **model_kwargs):
         history,
         system_prompt: str,
         temperature: float,
-        max_new_tokens: int,
+        max_tokens: int,
         top_k: int,
         repetition_penalty: float,
         top_p: float
@@ -42,8 +42,9 @@ def get_fn(model_path: str, **model_kwargs):
                 model=model_name,
                 messages=messages,
                 temperature=temperature,
-                top_p=top_p,
                 max_tokens=max_tokens,
+                top_k=top_k,
+                repetition_penalty=repetition_penalty,
                 n=1,
                 stream=True, 
                 response_format={"type": "text"},