Update app.py

app.py

@@ -28,8 +28,8 @@ if torch.cuda.is_available():
 def generate(
     message: str,
     chat_history: list[tuple[str, str]],
-    max_new_tokens: int = 1024,
-    temperature: float = 0.6,
+    max_new_tokens: int = 512,
+    temperature: float = 0.2,
     top_p: float = 0.9,
     top_k: int = 50,
     repetition_penalty: float = 1.2,