Update app.py

app.py

@@ -29,7 +29,7 @@ def generate_response(user_input, max_new_token: 100, top_p, temperature, top_k,
     ])
     msg += "<|assistant|>\n"
     res = client.generate(msg, stop_sequences=["<|assistant|>", eos_token, "<|system|>", "<|user|>"],
-                          max_new_tokens=1000)
+                          max_new_tokens=1000, top_p=0.9, do_sample=True, temperature=1.0)
     return [("assistant", res.generated_text)]
 
 with gr.Blocks() as demo: