Do your Uncensored models still require prompts mentioned by `nicoboss`?
Hi, in About section of your model, repository of nicoboss is mentioned. Also he means that in order to make model work as expected, system prompt is required (and there are 3 prompts for 3 models Uncensored, Uncensored-v2 and Uncensored-v2-Unbiased.
Do your morels also require such system prompt for proper operation? Do you have some other recommended system prompt? I'm not necessarily planning to ask it about China govt XD but I also wouldn't like it to be constrained by China laws in answering questions.
@lapsio Yes GGUFs and other quants always behave like the original model in this regard. They system prompt is only required if you want the model to be fully uncensored. All my uncensored models are finetuned using this exact system prompt. You don't have to specify exactly this one. Feel free to adopt it to your needs. Remove the sections you don't like and add additional things you want the model to do. A great example of such a user modified system prompt someone posted under: https://huggingface.co/nicoboss/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased/discussions/2
For other great uncensored system prompts I recommend taking a look at https://github.com/cognitivecomputations/dolphin-system-messages - if you create a good one by your own I recommend to open a PR in that repository.
If you want the model to use reasoning, I recommend to just use the system prompt in the model card and append: Use tags and think all the time.. A reasoning finetune is still work in progress but using this system prompt you can reactivate the base models reasoning capability without having to wait for me completing the reasoning finetune.
Do your morels also require such system prompt for proper operation? Do you have some other recommended system prompt? I'm not necessarily planning to ask it about China govt XD but I also wouldn't like it to be constrained by China laws in answering questions.
No they do not. You can use Uncensored-v2-Unbiased with any system prompt you like. The political bias towards the Chinese narrative got completely trained away and will not be present no matter what system prompt is specified. If you want to still have a censored model with all they typical safeguards feel free to use Uncensored-v2-Unbiased but don't mention in the system prompt that the model should be uncensored. The system prompt is only required to specify which of the conventional AI safeguards should be disabled but I can't guarantee that they didn't put any Chinese censorship into the conventional safeguards which is why I recommend disabling them.