---
license: apache-2.0
language:
- en
widget:
- text: Qwen3.5-2B_Abliterated
  output:
    url: >-
      https://huggingface.co/SicariusSicariiStuff/Qwen3.5-2B_Abliterated/resolve/main/Images/Abliterated.png
base_model:
- Qwen/Qwen3.5-2B
---

<div align="center">
  <b style="font-size: 50px;">Qwen3.5-2B</b>


</div>


<div align="center">
  <b style="font-size: 80px;">Abliterated</b>


</div>


---

<div align="center" style="font-size: 18px; margin-top: 20px;">
  <b>Developed by:</b> <a href="https://huggingface.co/SicariusSicariiStuff">SicariusSicariiStuff</a>
</div>

---

**Qwen3.5-2B_Abliterated** is an abliterated variant of Qwen/Qwen3.5-2B 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
```
~6%
```

# 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:** Qwen3.5-2B
- **Parameters:** 2B
- **Context Length:** 262144 tokens
- **Architecture:** Qwen3_5ForConditionalGeneration
- **Precision:** bf16
- **Method:** Orthogonalization-based abliteration
- **License:** apache-2.0

---


## Methodology

1. Identifies refusal direction vectors in activation space
2. Orthogonalizes weights to inhibit activation along these directions
3. Preserves (mostly) all other model behaviors and knowledge

---


## Model Details

- Intended use: **General Tasks**.

- Censorship level: <b>Low - Very Low</b>

- **X / 10** (10 completely uncensored)


## UGI score:



---


## Citation Information

```
@llm{Qwen3.5-2B_Abliterated,
  author = {SicariusSicariiStuff},
  title = {Qwen3.5-2B_Abliterated},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/SicariusSicariiStuff/Qwen3.5-2B_Abliterated}
}
```

---

## Other stuff
- [Impish_LLAMA_4B](https://huggingface.co/SicariusSicariiStuff/Impish_LLAMA_4B) the **“Impish experience”**, now runnable on spinning rust & toasters.
- [SLOP_Detector](https://github.com/SicariusSicariiStuff/SLOP_Detector) Nuke GPTisms, with SLOP detector.