try to improve the inference

app.py

@@ -46,7 +46,7 @@ def chat(message, history):
     with torch.no_grad():
         outputs = model.generate(
             **inputs,
-            max_new_tokens=128,
+            max_new_tokens=256,
             temperature=0.7,
             do_sample=True,
             use_cache=True,