Fun Model with the right settings.
I have just tested this model and DavidAU/Qwen3-The-Josiefied-Omega-Directive-22B-uncensored-abliterated-GGUF from your new tri until now.
I have tested just the Q6_K and Q8 versions.
From the two tested this full merge is more eloquent while staying uncensored.
Think on and off tested, with think it stays more to the prompt and the details.
Sometimes so good and similiar in different swipes that i turned it off for one message to advance the story in a new direction.
It follows my systemprompt and additional instructions in the author notes good and always when think is on.
It has successfully created a new char from my example char provided when used instruction mode in koboldAi Lite.
It is so uncensored that it will not hesitate to behead the {{user}} in a sword fight.
Your provided suggested settings are needed or it has the tendency to start repeating or makes gibberish.
Thank you for the model i have fun with it.
I will not push you but i would also test and use a 32B variant :-D.
In the KoboldAI Discord Server -> channel Horde -> Char Mandurin
I have my last use settings in a text file if interested.
Excellent ; thank you for the detailed feedback.
Don't know - yet - about a 32B (or 32B Qwen 3s merged, which would be roughly 48B).
Still getting a "feel" for the Qwen 3s ; with the reasoning/thinking being a new challenge - that is not breaking it.
PS: I am also on Kobold discord; David_AU