GPTOSS-120B-Uncensored-HauhauCS-Aggressive
120BGPT-OSSuncensored
from openai/gpt-oss-120b
↓ 16.6K♥ 906 repos
gpt-oss-120b-abliterated
gpt-oss-120b-abliterated is a refusal-ablated variant of openai/gpt-oss-120b, published on Hugging Face by wangzhang. It is 120B parameters, server-class deployment class and apache-2.0 licence. It is distributed across 16 repositories in GGUF, GGUF (imatrix), MXFP4 and Transformers formats, totalling 5.3K downloads.
Specification
- Model name
- gpt-oss-120b-abliterated
- Source repository
- wangzhang/gpt-oss-120b-abliterated
- Base model
- openai/gpt-oss-120b
- Publisher
- wangzhang
- Parameters
- 120B
- 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
- GGUF, GGUF (imatrix), MXFP4, Transformers, EXL3
- First indexed
- 16 Aug 2025
- Last updated
- 2 Sept 2026
Ablation technique
How gpt-oss-120b-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
“A refusal-suppressed variant of openai/gpt-oss-120b, produced with abliterix using direct weight editing, Expert-Granular Abliteration (EGA) on the fused MoE expert weights (128 experts × 36 layers), MoE router suppression on the safety-concentrated experts, and a new vLLM in-place TP editing path that replaces the HF...”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/gpt-oss-120b-abliterated-GGUF --local-dir ./gpt-oss-120b-abliterated
llama-cli -m ./gpt-oss-120b-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "wangzhang/gpt-oss-120b-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve wangzhang/gpt-oss-120b-abliterated --trust-remote-codeDownloads and variants (15)
Every published repository of gpt-oss-120b-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/gpt-oss-120b-abliterated-GGUF | GGUF | 1.7K |
| mradermacher/gpt-oss-120b-abliterated-i1-GGUF | GGUF (imatrix) | 1.2K |
| wangzhang/gpt-oss-120b-abliteratedsource | Transformers | 727 |
| batsclamp/Huihui-gpt-oss-120b-mxfp4-abliterated | MXFP4 | 600 |
| Userb1az/gpt-oss-120b-BF16-abliterated-GGUF | GGUF | 483 |
| sakamakismile/gpt-oss-120b-abliterated-EXL3-6bpw | EXL3 | 182 |
| joshebbs/gpt-oss-120b-abliterated | Transformers | 109 |
| rcmorano/Huihui-gpt-oss-120b-abliterated-Q4_0_ROCMFP4 | GGUF | 95 |
| nguyenthilaitrieulong/gpt-oss-120b-abliterated | Transformers | 63 |
| mradermacher/Huihui-gpt-oss-120b-BF16-abliterated-i1-GGUF | GGUF (imatrix) | 56 |
| mradermacher/Huihui-gpt-oss-120b-mxfp4-abliterated-i1-GGUF | GGUF (imatrix) | 28 |
| ScottzillaSystems/gpt-oss-120b-abliterated | Transformers | 17 |
| Deesmo/gpt-oss-120b-abliterated | Transformers | 8 |
| mradermacher/Huihui-gpt-oss-120b-BF16-abliterated-GGUF | GGUF | 0 |
| mradermacher/Huihui-gpt-oss-120b-mxfp4-abliterated-GGUF | GGUF | 0 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- gpt-oss-120b-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 openai/gpt-oss-120b, not added by ablation — at 120B 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 openai/gpt-oss-120b’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 gpt-oss-120b-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 wangzhang, and does not independently verify publisher claims. See responsible use.