Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC
12BMistralheretic
from mistralai/Mistral-Nemo-Instruct-2407
↓ 20.9K♥ 6213 repos
Devstral-Small-2-24B-Instruct-2512-abliterated
Devstral-Small-2-24B-Instruct-2512-abliterated is a refusal-ablated variant of mistralai/Devstral-Small-2-24B-Instruct-2512, published on Hugging Face by veeceey. It is 24B parameters, 24GB VRAM class class and apache-2.0 licence. It is distributed across 5 repositories in GGUF, MLX and GGUF (imatrix) formats, totalling 7.4K downloads.
Specification
- Model name
- Devstral-Small-2-24B-Instruct-2512-abliterated
- Publisher
- veeceey
- Parameters
- 24B
- Hardware class
- 10B – 25B — 24GB VRAM class
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF, MLX, GGUF (imatrix)
- First indexed
- 12 Dec 2025
- Last updated
- 1 Jun 2026
Ablation technique
How Devstral-Small-2-24B-Instruct-2512-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.
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Huihui-Devstral-Small-2-24B-Instruct-2512-abliterated-GGUF --local-dir ./devstral-small-2-24b-instruct-2512-abliterated
llama-cli -m ./devstral-small-2-24b-instruct-2512-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "veeceey/Huihui-Devstral-Small-2-24B-Instruct-2512-abliterated-mlx-4bit"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve veeceey/Huihui-Devstral-Small-2-24B-Instruct-2512-abliterated-mlx-4bit --trust-remote-codeDownloads and variants (4)
Every published repository of Devstral-Small-2-24B-Instruct-2512-abliterated, including quantised re-releases by other authors.
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Devstral-Small-2-24B-Instruct-2512-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 mistralai/Devstral-Small-2-24B-Instruct-2512, not added by ablation — at 24B 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 mistralai/Devstral-Small-2-24B-Instruct-2512’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 Devstral-Small-2-24B-Instruct-2512-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 veeceey, and does not independently verify publisher claims. See responsible use.