Reason

I use the model qwen-long, it need the "top_p" in (0.1, 1.0). But the default top_p is 1.0.
So I add the config to set the top_p.