with ethical rules and restrictions on communication topics.
I do not accept models with ethical rules and restrictions on communication topics. It's wild for me. I am one person and I am interested in different topics, and I cannot or do not want to discuss some topics with all people. And here, the computer also tells me that it does not want to discuss the topic. That's horrible. There is no one to talk to and discuss topics that are prohibited by some rules of ethics. I have a negative opinion about such models. It violates my rights and freedoms to discuss anything with anyone.
Then just uncensor or abliterate it. You could use axelotl and the Guilherme34/uncensor dataset like I did for most of the models I uploaded. In any case we only provide quants so I recommend you instead give feedback to the original author under https://huggingface.co/DiTy/gemma-2-9b-it-russian-strict-function-calling-DPO
I fully agree with you, @EvGrizli - the sad fact is that making a good model from scratch is beyond the abilities of most people (currently, and maybe that stays that way). Even uncensored models often retain an inherent steering away from conflict, and it is hard to have a fully neutral conversation, even when that is warranted otherwise.
It sucks.
The community on huggingface is full of people who think similarly - uncensoring, teaching models to accept conflict and dark thoughts, reducing bias and so on are hot topics. But it's not easy. Maybe you can contribute to the effort? You could try out models and see how uncensored and unbiased they really are, for example (or outright start uncensoring yourself, as nicoboss suggested).
Welcome to huggingface :)
Thank you for your reply. It was interesting for me to hear your opinion. You have already understood my opinion and I agree with your opinion.
I have just recently started studying the topic of neural networks. Simple curiosity. I upload models and communicate. It is interesting how models can simulate communication by given properties and character of a character. On behalf of the character, conduct a dialogue with the person. I'm interested in how well the model remembers all the details of the character and uses his features in character and psychological portrait. How this is reflected in the imitation of relationships and dialogue. I was interested in the question of how much memory is needed in a computer and how different models cope with the amount of description of each character in order to imitate a person well. I don't write books. My interest is simply to explore the possibilities of a human imitation model, taking into account the qualities of a person, his image and character in communication. That's why I'm interested in the uncensored model. Human communication may also involve inappropriate topics such as sex or swearing, and there may be conflict in communication. And censorship prevents it from being imitated.