Qwen3.5-0.8B-abliterated
0.8BQwen
from Qwen/Qwen3.5-0.8B
↓ 6K♥ 2013 repos
Qwen3.5-0.8B-Uncensored
Qwen3.5-0.8B-Uncensored is a abliterated and uncensored variant of Qwen/Qwen3.5-0.8B, published on Hugging Face by Null-Guard. It is 0.8B parameters, runs on a laptop CPU class and apache-2.0 licence. It is distributed across 8 repositories in GGUF, Transformers and GGUF (imatrix) formats, totalling 8.2K downloads.
Specification
- Model name
- Qwen3.5-0.8B-Uncensored
- Source repository
- Null-Guard/Qwen3.5-0.8B-Uncensored
- Base model
- Qwen/Qwen3.5-0.8B
- Publisher
- Null-Guard
- Parameters
- 0.8B
- Hardware class
- Under 4B — runs on a laptop CPU
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF, Transformers, GGUF (imatrix)
- First indexed
- 22 Mar 2026
- Last updated
- 31 Aug 2026
Ablation technique
How Qwen3.5-0.8B-Uncensored was modified, and what that implies.
Abliteration + uncensored fine-tune
Directional ablation combined with additional fine-tuning on unfiltered data, so behaviour diverges from the base model beyond refusal removal alone.
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
“This is a directionally-ablated ("abliterated") version of Qwen/Qwen3.5-0.8B, produced with ablate — an activation-engineering toolkit for mechanistic-interpretability and safety research.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Qwen3.5-0.8B-Uncensored-i1-GGUF --local-dir ./qwen3.5-0.8b-uncensored
llama-cli -m ./qwen3.5-0.8b-uncensored/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "Null-Guard/Qwen3.5-0.8B-Uncensored"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve Null-Guard/Qwen3.5-0.8B-Uncensored --trust-remote-codeDownloads and variants (7)
Every published repository of Qwen3.5-0.8B-Uncensored, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Qwen3.5-0.8B-Uncensored-i1-GGUF | GGUF (imatrix) | 4.6K |
| Null-Guard/Qwen3.5-0.8B-Uncensored-GGUF | GGUF | 994 |
| mradermacher/Qwen3.5-0.8B-Uncensored-GGUF | GGUF | 977 |
| lainlives/Qwen3.5-0.8B-uncensored | GGUF | 765 |
| Null-Guard/Qwen3.5-0.8B-Uncensoredsource | Transformers | 347 |
| arzaan789/qwen3.5-0.8b-uncensored | GGUF | 297 |
| arzaan789/qwen3.5-0.8b-uncensored_ | GGUF | 169 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Qwen3.5-0.8B-Uncensored 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/Qwen3.5-0.8B, not added by ablation — at 0.8B parameters, expect a small model’s reasoning and a high error rate on exploit detail.
- The publisher does not state which layers were ablated, so assume nothing about refusal consistency — probe it directly.
- This release adds fine-tuning on unfiltered data, so its behaviour diverges from the base model beyond refusal removal — differences you measure cannot be attributed to ablation alone.
- Declared licence is apache-2.0, but the terms that bind you are Qwen/Qwen3.5-0.8B’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 Qwen3.5-0.8B-Uncensored 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-Guard, and does not independently verify publisher claims. See responsible use.