Update app.py

app.py

@@ -577,11 +577,14 @@ def chat_response_stream(message: str, history: list, state: dict, compression_d
         streamer=streamer,
         use_cache=True,
         max_new_tokens=1024,
-        num_beams=1,
-        do_sample=False,
+        # num_beams=1,
+        # do_sample=False,
+        # top_p=1.0,
+        # top_k=None,
         temperature=1.0,
-        top_p=1.0,
-        top_k=None,
+        top_k=64,
+        top_p=0.95,
+        min_p=0.0
     )
     t = Thread(target=model.generate, kwargs=generate_kwargs)
     t.start()