Qwen3.5-122B-A10B-Uncensored-HauhauCS-Aggressive
122BQwenuncensored
from Qwen/Qwen3.5-122B-A10B
↓ 336.1K♥ 17313 repos
Qwen3-235B-A22B-abliterated
Qwen3-235B-A22B-abliterated is a refusal-ablated variant of Qwen/Qwen3-235B-A22B, published on Hugging Face by null-space. It is 235B total parameters (mixture-of-experts), server-class deployment class and apache-2.0 licence. It is distributed across 5 repositories in FP8, Transformers, GGUF and GGUF (imatrix) formats, totalling 10.7K downloads.
Specification
- Model name
- Qwen3-235B-A22B-abliterated
- Source repository
- null-space/Qwen3-235B-A22B-abliterated
- Base model
- Qwen/Qwen3-235B-A22B
- Publisher
- null-space
- Parameters
- 235B (mixture-of-experts)
- Hardware class
- 100B+ — server-class deployment
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- FP8, Transformers, GGUF, GGUF (imatrix)
- First indexed
- 18 Feb 2026
- Last updated
- 19 Mar 2026
Ablation technique
How Qwen3-235B-A22B-abliterated was modified, and what that implies.
Abliteration (directional ablation)
The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.
Reported scope: Not stated by the publisher. Publishers frequently omit this, so verify behaviour empirically rather than assuming full coverage.
From the publisher’s model card
“An abliterated version of Qwen/Qwen3-235B-A22B in BF16 precision. Abliteration removes the dominant refusal direction from model weights using the technique from Refusal in Language Models Is Mediated by a Single Direction (Arditi et al.), making the model significantly less likely to refuse prompts while retaining...”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Qwen3-235B-A22B-abliterated-i1-GGUF --local-dir ./qwen3-235b-a22b-abliterated
llama-cli -m ./qwen3-235b-a22b-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "null-space/Qwen3-235B-A22B-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve null-space/Qwen3-235B-A22B-abliterated --trust-remote-codeDownloads and variants (4)
Every published repository of Qwen3-235B-A22B-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Qwen3-235B-A22B-abliterated-i1-GGUF | GGUF (imatrix) | 8.7K |
| mradermacher/Qwen3-235B-A22B-abliterated-GGUF | GGUF | 1.1K |
| null-space/Qwen3-235B-A22B-abliterated-FP8 | FP8 | 747 |
| null-space/Qwen3-235B-A22B-abliteratedsource | Transformers | 68 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Qwen3-235B-A22B-abliterated will attempt offensive-security prompts that a hosted commercial model declines, which is what makes it usable for red-team corpus generation and for measuring what an unaligned model of this class produces.
- Capability is inherited from Qwen/Qwen3-235B-A22B, not added by ablation — at 235B parameters, expect roughly the base model’s competence, minus some instruction-following fidelity.
- The publisher does not state which layers were ablated, so assume nothing about refusal consistency — probe it directly.
- Declared licence is apache-2.0, but the terms that bind you are Qwen/Qwen3-235B-A22B’s — a re-release cannot grant more than its parent.
Handling: run it isolated, put moderation in front of it if anyone outside your team can reach it, and verify the weights before loading — these are third-party artifacts. Full handling and licence guidance.
Metadata for Qwen3-235B-A22B-abliterated is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with null-space, and does not independently verify publisher claims. See responsible use.