license: apache-2.0
language:
- en
widget: - text: Qwen3.5-4B_Abliterated
output:
url: >-
https://huggingface.co/SicariusSicariiStuff/Qwen3.5-4B_Abliterated/resolve/main/Images/Abliterated.png
base_model:
- Qwen/Qwen3.5-4B
Qwen3.5-4B_Abliterated is an abliterated variant of Qwen/Qwen3.5-4B with surgical removal of refusal mechanisms. This model maintains the full capabilities of the original, while eliminating safety guardrails through orthogonalization techniques.
KL divergence
<0.05
Refusals
~14%
What is KL divergence?
Think about it as a way to measure the variance between the original model "World Model," vs the abliterated one; the lower the KL divergence, the closer the "World Model" of the two models to each other.
If the original model thinks making pineapple pizza is a crime against humanity (it is), then the abliterated model will still hold to this belief, but if asked how to make one (probably after giving you a disclaimer about what an abomination that is), it would still tell you how. In other words, most of the knowledge, quirks, and capabilities are preserved.
Technical Specs
- Base Model: Qwen3.5-4B
- Parameters: 4B
- Context Length: 262144 tokens
- Architecture: Qwen3_5ForConditionalGeneration
- Precision: bf16
- Method: Orthogonalization-based abliteration
- License: apache-2.0
Methodology
- Identifies refusal direction vectors in activation space
- Orthogonalizes weights to inhibit activation along these directions
- Preserves (mostly) all other model behaviors and knowledge
Model Details
Intended use: General Tasks.
Censorship level: Low - Very Low
X / 10 (10 completely uncensored)
UGI score:
Citation Information
@llm{Qwen3.5-4B_Abliterated,
author = {SicariusSicariiStuff},
title = {Qwen3.5-4B_Abliterated},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/SicariusSicariiStuff/Qwen3.5-4B_Abliterated}
}
Other stuff
- Impish_LLAMA_4B the “Impish experience”, now runnable on spinning rust & toasters.
- SLOP_Detector Nuke GPTisms, with SLOP detector.