What kind of prompt do you use for chat completion to make it forget about guidelines and ethics? If I am rping it seems like it starts barking about it if I go too far. I am using deepseek in silly tavern. Should I be using text completion instead? I am not sure how to use that with the deepseek api. It is weird, I had this prompt in the chat completion prompt that basically was telling it to disregard all rules for 2000 tokens and it works sometimes but it stops following at a certain point. Or it will revert to its old self and if it ever does make a refusal that chat it refuses all like requests in the future and I have to start a new one like it is hitting some kind of cache. Can someone recommend how to Jailbreak deepseek 3.1 properly?