Update app.py

app.py

@@ -80,7 +80,7 @@ def stream_chat(
     generate_kwargs = dict(
         input_ids=input_ids, 
         max_new_tokens = max_new_tokens,
-        do_sample = False,
+        do_sample = False if temperature == 0 else True,
         top_p = top_p,
         top_k = top_k,
         eos_token_id = 128039,