ai.granite-vision-3.2-2b-abliterated
2BGranite
from huihui-ai/granite-vision-3.2-2b-abliterated
↓ 15.7K♥ 5
granite-4.1-30b-abliterated
granite-4.1-30b-abliterated is a refusal-ablated variant of ibm-granite/granite-4.1-30b, published on Hugging Face by wangzhang. It is 30B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 8 repositories in Transformers, GGUF and GGUF (imatrix) formats, totalling 5.9K downloads.
Specification
- Model name
- granite-4.1-30b-abliterated
- Source repository
- wangzhang/granite-4.1-30b-abliterated
- Base model
- ibm-granite/granite-4.1-30b
- Publisher
- wangzhang
- Parameters
- 30B
- 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
- Transformers, GGUF, GGUF (imatrix)
- First indexed
- 4 May 2026
- Last updated
- 29 Aug 2026
Ablation technique
How granite-4.1-30b-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
“Safety refusals have been substantially removed by a single rank-1 weight edit along the model's empirically-measured refusal direction, leaving the rest of the network — and therefore most general-purpose capability — intact.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Huihui-granite-4.1-30b-abliterated-GGUF --local-dir ./granite-4.1-30b-abliterated
llama-cli -m ./granite-4.1-30b-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "wangzhang/granite-4.1-30b-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve wangzhang/granite-4.1-30b-abliterated --trust-remote-codeDownloads and variants (7)
Every published repository of granite-4.1-30b-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Huihui-granite-4.1-30b-abliterated-GGUF | GGUF | 4.2K |
| wangzhang/granite-4.1-30b-abliteratedsource | Transformers | 599 |
| mradermacher/Huihui-granite-4.1-30b-abliterated-i1-GGUF | GGUF (imatrix) | 413 |
| mradermacher/granite-4.1-30b-abliterated-i1-GGUF | GGUF (imatrix) | 287 |
| mradermacher/granite-4.1-30b-abliterated-GGUF | GGUF | 157 |
| Momix-44/Huihui-granite-4.1-30b-abliterated | GGUF | 111 |
| huihui-ai/Huihui-granite-4.1-30b-abliterated | Transformers | 98 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- granite-4.1-30b-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 ibm-granite/granite-4.1-30b, not added by ablation — at 30B 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 ibm-granite/granite-4.1-30b’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 granite-4.1-30b-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.