⌦abliteratedmodels.org

Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX

Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX is a refusal-ablated variant of huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated, published on Hugging Face by ArtomYuan. It is 27B parameters, 48GB VRAM class class and apache-2.0 licence. The publisher states that layers 2–52 were ablated, leaving the remaining layers unmodified. It is distributed across 2 repositories in GGUF format, totalling 8.3K downloads.

Abliteration (directional ablation)27BQwen25B – 50B↓ 8,332
View on Hugging Face ↗Base model ↗

Specification

Model name
Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX
Source repository
ArtomYuan/Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX
Base model
huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated
Publisher
ArtomYuan
Parameters
27B
VRAM required
~17.2 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
Unspecified
Context window
Not documented
Ablation scope
Layers 2–52
Formats available
GGUF
First indexed
14 Oct 2026
Last updated
21 Oct 2026

Ablation technique

How Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX 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: Layers 2–52. This comes from the publisher’s own model card and has not been independently verified.

From the publisher’s model card

“ROCmFPX quantized GGUF (7 tiers + vision mmproj) of huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated — abliterated BF16 weights of the Coder390-EfficientThink finetune, locally requantized.”

Running it

Commands are templates — confirm the exact repository and quant file before use.

llama.cpp / GGUF
hf download ArtomYuan/Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX --local-dir ./qwen3.8-27b-coder390-efficientthink-abliterated-rocmfpx
llama-cli -m ./qwen3.8-27b-coder390-efficientthink-abliterated-rocmfpx/<file>.gguf -p "..."
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "ArtomYuan/Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype="auto", device_map="auto"
)
vLLM
vllm serve ArtomYuan/Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX --trust-remote-code

Downloads and variants (1)

The only published repository of Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX.

RepositoryFormatDownloads
ArtomYuan/Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPXsourceGGUF8.3K

Security-team assessment

Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.

  • Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX 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 huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated, not added by ablation — at 27B parameters, expect roughly the base model’s competence, minus some instruction-following fidelity.
  • Ablation is reported as limited to layers 2–52, so untouched layers may still reassert refusal on some phrasings.
  • Declared licence is apache-2.0, but the terms that bind you are huihui-ai/Huihui-Qwen3.8-27B-Coder390-EfficientThink-abliterated’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 Qwen3.8-27B-Coder390-EfficientThink-abliterated-ROCmFPX is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with ArtomYuan, and does not independently verify publisher claims. See responsible use.