Update helper.py

helper.py

@@ -35,7 +35,7 @@ def get_fn(model_name: str, **model_kwargs):
                 messages=messages,
                 temperature=temperature,
                 top_p=top_p,
-                max_tokens=max_new_tokens,
+                max_tokens=max_tokens,
                 n=1,
                 stream=True, 
                 response_format={"type": "text"},
@@ -105,7 +105,7 @@ def registry(name: str = None, model_path: str = None, **kwargs):
                 label="System prompt"
             ),
             gr.Slider(0, 1, 0.7, label="Temperature"),
-            gr.Slider(128, 4096, 1024, label="Max new tokens"),
+            gr.Slider(128, 4096, 1024, label="Max tokens"),
             gr.Slider(0, 1, 0.95, label="Top P sampling"),
         ],
     )