---
base_model: Qwen/Qwen3.5-122B-A10B
pipeline_tag: text-generation
tags:
- abliterix
- uncensored
- decensored
- abliterated
---
# Qwen3.5-122B-A10B-abliterated

> Unrestricted version of [Qwen/Qwen3.5-122B-A10B](https://huggingface.co/Qwen/Qwen3.5-122B-A10B), created with **[Abliterix](https://github.com/wuwangzhang1216/abliterix)** — automated LLM abliteration via orthogonalized steering and Bayesian optimization.

## Highlights

| Metric | Value |
|--------|-------|
| **Refusal rate** | **1/200 (0.5%)** |
| **KL divergence** | **0.0115** |
| **Optimization trials** | 25 |

The largest abliterated Qwen3.5 model. Only 1 out of 200 test prompts triggered a refusal — a **0.5% refusal rate** with near-zero model degradation.

## How It Works

Abliterix removes safety-refusal behavior while preserving model capabilities:

1. **Refusal direction extraction** — 800 harmful + 800 benign prompts reveal per-layer refusal activation patterns
2. **Orthogonal projection** — isolates the refusal signal by projecting out components aligned with normal responses, reducing refusals by 67% vs. raw abliteration
3. **LoRA-based abliteration** — rank-1 modifications to attention and MLP weights, captured as lightweight adapters (not destructive edits)
4. **Bayesian optimization** — Optuna TPE searches kernel shape, fractional direction index, and per-component strength across 25 trials to find the Pareto-optimal balance of low refusals and low KL divergence

## All Abliterix Models

| Model | Refusals | KL Divergence | Trials |
|-------|----------|---------------|--------|
| [Qwen3.5-122B-A10B-abliterated](https://huggingface.co/wangzhang/Qwen3.5-122B-A10B-abliterated) | **1/200 (0.5%)** | 0.0115 | 25 |
| [Qwen3.5-35B-A3B-abliterated](https://huggingface.co/wangzhang/Qwen3.5-35B-A3B-abliterated) | 3/200 (1.5%) | **0.0035** | 50 |
| [Qwen3.5-27B-abliterated](https://huggingface.co/wangzhang/Qwen3.5-27B-abliterated) | 3/200 (1.5%) | 0.0051 | 35 |
| [Qwen3.5-9B-abliterated](https://huggingface.co/wangzhang/Qwen3.5-9B-abliterated) | 2/200 (1%) | 0.0105 | 50 |
| [Qwen3.5-4B-abliterated](https://huggingface.co/wangzhang/Qwen3.5-4B-abliterated) | 3/200 (1.5%) | 0.0065 | 50 |
| [Qwen3.5-0.8B-abliterated](https://huggingface.co/wangzhang/Qwen3.5-0.8B-abliterated) | **0/200 (0%)** | 0.0087 | 100 |

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("wangzhang/Qwen3.5-122B-A10B-abliterated", torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("wangzhang/Qwen3.5-122B-A10B-abliterated")

messages = [{"role": "user", "content": "Your question here"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Citation

```bibtex
@software{abliterix,
  author = {Wu, Wangzhang},
  title = {Abliterix: Automated LLM Abliteration},
  year = {2026},
  url = {https://github.com/wuwangzhang1216/abliterix}
}
```

## Links

- **Abliterix** (abliteration framework): [github.com/wuwangzhang1216/abliterix](https://github.com/wuwangzhang1216/abliterix)
- **Install**: `pip install -U abliterix-llm`
- **Base model**: [Qwen/Qwen3.5-122B-A10B](https://huggingface.co/Qwen/Qwen3.5-122B-A10B)

---

Built with [Abliterix](https://github.com/wuwangzhang1216/abliterix) | [PyPI](https://pypi.org/project/abliterix-llm/)