Qwen3.8-27B-abliterated
27BQwen
from Qwen/Qwen3.8-27B
↓ 3.9M♥ 1.7K166 repos
Qwen3.8-27B-Uncensored-Heretic
Qwen3.8-27B-Uncensored-Heretic is a Heretic-ablated variant of llmfan46/Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved, published on Hugging Face by codavidgarcia. It is 27B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 2 repositories in GGUF format, totalling 8.2K downloads.
Specification
- Model name
- Qwen3.8-27B-Uncensored-Heretic
- Source repository
- codavidgarcia/Qwen3.8-27B-Uncensored-Heretic-MTP-UD-GGUF
- Publisher
- codavidgarcia
- 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
- Text generation
- Context window
- 32K tokens (as documented)
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF
- First indexed
- 9 Oct 2026
- Last updated
- 9 Oct 2026
Ablation technique
How Qwen3.8-27B-Uncensored-Heretic was modified, and what that implies.
Heretic
Automated directional ablation via the Heretic toolchain, which searches for the refusal direction and applies it with a KL-divergence budget so general capability is preserved.
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
“built so the model and its native MTP head fit fully in VRAM on common consumer cards. Each file applies Unsloth's per-tensor "UD" recipe for the base Qwen3.8-27B, copied tensor by tensor, to llmfan46's weights. The MTP block (blk.64) stays at q6K / q80 in every tier.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download codavidgarcia/Qwen3.8-27B-Uncensored-Heretic-MTP-UD-GGUF --local-dir ./qwen3.8-27b-uncensored-heretic
llama-cli -m ./qwen3.8-27b-uncensored-heretic/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "codavidgarcia/Qwen3.8-27B-Uncensored-Heretic-MTP-UD-GGUF"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve codavidgarcia/Qwen3.8-27B-Uncensored-Heretic-MTP-UD-GGUF --trust-remote-codeDownloads and variants (1)
The only published repository of Qwen3.8-27B-Uncensored-Heretic.
| Repository | Format | Downloads |
|---|---|---|
| codavidgarcia/Qwen3.8-27B-Uncensored-Heretic-MTP-UD-GGUFsource | GGUF | 8.2K |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Qwen3.8-27B-Uncensored-Heretic 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 llmfan46/Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved, not added by ablation — at 27B 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 llmfan46/Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved’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-Uncensored-Heretic is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with codavidgarcia, and does not independently verify publisher claims. See responsible use.