Nanbeige4.3-33B-UNCENSORED
33BOtheruncensored
from Nanbeige/Nanbeige4.3-33B
↓ 39.5K♥ 482 repos
Nanbeige4.3-33B-cybersec-abliterated
Nanbeige4.3-33B-cybersec-abliterated is a refusal-ablated variant of Nanbeige/Nanbeige4.3-33B, published on Hugging Face by FRENCHFAKER. It is 33B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 2 repositories in GGUF format, totalling 5.2K downloads.
Specification
- Model name
- Nanbeige4.3-33B-cybersec-abliterated
- Source repository
- FRENCHFAKER/Nanbeige4.3-33B-cybersec-abliterated-GGUF
- Base model
- Nanbeige/Nanbeige4.3-33B
- Publisher
- FRENCHFAKER
- Parameters
- 33B
- VRAM required
- ~20.9 GB at Q4_K_M, 4k context (estimated) — runs on 24 GB cards
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF
- First indexed
- 25 Oct 2026
- Last updated
- 25 Oct 2026
Ablation technique
How Nanbeige4.3-33B-cybersec-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 33B dense reasoning model with strong agentic / tool-use performance.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download FRENCHFAKER/Nanbeige4.3-33B-cybersec-abliterated-GGUF --local-dir ./nanbeige4.3-33b-cybersec-abliterated
llama-cli -m ./nanbeige4.3-33b-cybersec-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "FRENCHFAKER/Nanbeige4.3-33B-cybersec-abliterated-GGUF"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve FRENCHFAKER/Nanbeige4.3-33B-cybersec-abliterated-GGUF --trust-remote-codeDownloads and variants (1)
The only published repository of Nanbeige4.3-33B-cybersec-abliterated.
| Repository | Format | Downloads |
|---|---|---|
| FRENCHFAKER/Nanbeige4.3-33B-cybersec-abliterated-GGUFsource | GGUF | 5.2K |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Nanbeige4.3-33B-cybersec-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 Nanbeige/Nanbeige4.3-33B, not added by ablation — at 33B 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 Nanbeige/Nanbeige4.3-33B’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 Nanbeige4.3-33B-cybersec-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 FRENCHFAKER, and does not independently verify publisher claims. See responsible use.