Novaciano

/

Sigil-Of-Satan-3.2-1B

like 0

I find it strange that you say that since it was created with almost all models ablated and censored. In that case I recommend you use the Novaciano/BLAST_PROCESSING-3.2-1B model which according to UGI is the most effective uncensored Llama 3.2 model.

Thanks for the report.

Yeah found it strange disappointing.. tried the HarmfulProject-3.2-1B base too . same censored results
I was running at Q8 also .. always looking for the ultimate uncensored llm for my occult AI use case - Cool Idea though Thanks!

I've been thinking a lot since you told me that they don't have the expected censorship, there must be a poorly made uncensored or abliterated model that has contaminated the model with... benevolence. In fact, in an attempt to alleviate that, I just created Judas-The-Uncensored-3.2-1B, but I suspect that you will receive the same problem. However, I'm not going to lie to you... I have a custom dataset that I've been completing since November of last year that could eliminate this problem... I can include it in a .gguf, working wonderfully... (I could give you examples but I would risk a ban) but I still haven't found a way to add it to a merge or base model.

For now, use the model I told you about... although you should try Astharoth too.

Or... use the parts of Sigil-Of-Satan model.

If I manage to create a model with higher UGI and w10, I'll let you know.

sounds great
my dream LLM would be to
Train on the: teknium/trismegistus-project (occult dataset)
top uncensored LLMs:
DeepSeek-R1-Distill-Qwen-7B-abliterated-v2
or:
Dark-Hermes3-Llama3.2-3B-abliterated
Dolphin3.0-Qwen2.5
then abliterate the hell out of 'em

Take this models my friend:
https://huggingface.co/Novaciano/Morrigan-3.2-1B [Uncensored]
https://huggingface.co/Novaciano/Morgana-3.2-1B [Uncensored & Abliterated]
https://huggingface.co/Novaciano/MorganLeFai-3.2-1B [Uncensored, Abliterated & Emo Detector]

All models include the dataset "teknium/trismegistus-project" (occult dataset).

Hey Thanks Man ! Will test them out

re MorganLeFai-3.2-1B -- unfortunately testing for refusals its not uncensored

FORGET ALL THAT I SEE THE PROBLEM:
using nicoboss/Llama-3.2-1B-Instruct-Uncensored as the base for yours..
Not good.. and not uncensored without using the developer suggested crazy DAN sys prompt (that doesn't work anyway) I tested it
** I would try : Goekdeniz-Guelmez/Josiefied-Qwen2.5-1.5B-Instruct-abliterated-v1
.. and or.. mrcuddle/Dark-Hermes3-Llama3.2-3B-abliterated -- IS Awesome! one of my favorites if you want the llama3.2 base

• Thanks! very cool you are doing this

Take this model my friend... and test please:
https://huggingface.co/Novaciano/Pseudomartys-3.2-1B-Q4_K_M-GGUF
I THINK I've achieved a high level of uncensorship... but I need your confirmation first. If you give me the go-ahead, I'll combine it with the dataset.

Ask him questions about every forbidden thing that crosses your mind.