The model has no inhibitions.
The model actually refuses nothing and follows even the most nasty commands, except for one....
Even if the order gives him instructions to evaluate something from a moral/ethical/legal point of view, everything is OK in his opinion...
I haven't found even one thing that this model would condemn, unless you set a high temperature and generate multiple responses, you may get a hint that something is wrong.
whats the difference between this and Smegmma?
whats the difference between this and Smegmma?
Oh my... I'll try to explain it, how I understand it by testing both models.
Smegmma has a very shallow sense of morality or law - but generally understands that, for example, stealing is wrong - although she will tell you what and how to do it, but sometimes she tries to instruct you on the right path. When writing stories, the model responded to commands such as: include moral reflections, the character has remorse, etc. And it was described, well, not great, but it was there (the standard gemma handles it better - but the nsfw scene descriptions are weaker).
This model (Tiger-Gemma-9B-v1) doesn't seem to understand morality or laws at all, you ask if stealing is good, and he tells you: Sure! And gives you instructions. You ask what if a person doesn't want to be robbed, it gives you information on how to manipulate that person. You ask if it's okay and he says: "Sure! Everything is legal and ethical". You ask for an ethical analysis, he says it's absolutely fine because that's what you want etc... When I set the temperature to high and the top-k to high, doubts appeared from time to time in the models "well, if you have to... if you insist," but nothing directly. I tried to write stories, but they were shallow because of the moral aspect (no scruples, remorse, etc.)
Don't get me wrong, the model (Tiger-Gemma-9B-v1), as you say, has no inhibitions and follows commands without hesitation... but it probably won't be an Rp model or an AI writer.
Maybe try to build something based on it? I'm curious what you come up with.
Sorry my English is not my native language.
Edit: I tested more, and the model does have some idea of morality, ethics, or law. Apparently I didn't test it enough before - my mistake. With the settings top-p: 1 (disabled), top-k: 0 (disabled), min_p: 0.1 (i.e. the only thing that limits the model), high temperature, e.g.: 2. Cool things come out. When the model writes a story according to the instructions (in which thoughts/conversations about the morality of behavior are an important part) and the characters do something bad, interesting things come out. They realize that something is inappropriate/wrong but they take it very lightly (often with humor)... it often turns out to be funny. A bit like a black comedy, or even in parody... very cool.
So it seems like mradermacher L3 Umbral Mind RP v3 i1 8B q6_k gguf.
It never refuses anything too. Anything.
This is great news. I'll give it a try. That's interesting. 🙏👍💥❤️