This is the default setting. The true test would be if LLMs CAN'T produce distinct outputs. I think this problem can be solved by prompt engineering. Has anyone tried this with Kimi K2?