Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER
27BQwenheretic
from DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU
↓ 1.4M♥ 2.4K37 repos
Qwen3.8-27B-TURBO-Heretic-NM-DAU
Qwen3.8-27B-TURBO-Heretic-NM-DAU is a Heretic-ablated variant of DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU, published on Hugging Face by blackfan23. It is 27B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 4 repositories in MLX format, totalling 7.4K downloads.
Specification
- Model name
- Qwen3.8-27B-TURBO-Heretic-NM-DAU
- Source repository
- blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit
- Publisher
- blackfan23
- 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
- Image text to text
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- MLX
- First indexed
- 10 Oct 2026
- Last updated
- 10 Oct 2026
Ablation technique
How Qwen3.8-27B-TURBO-Heretic-NM-DAU 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
“The language model was converted text-only with mlx-lm, the source checkpoint's BF16 vision tower was grafted back in as visiontower., the result validated for the Splash engine (family Qwen3.8-27B), and the chat template swapped to Sharp v22.5.0.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bit --trust-remote-codeDownloads and variants (3)
Every published repository of Qwen3.8-27B-TURBO-Heretic-NM-DAU, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-8bitsource | MLX | 3.2K |
| blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-4bit | MLX | 2.7K |
| blackfan23/Qwen3.8-27B-TURBO-Heretic-NM-DAU-MLX-6bit | MLX | 1.5K |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Qwen3.8-27B-TURBO-Heretic-NM-DAU 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 DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU, 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 DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU’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-TURBO-Heretic-NM-DAU is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with blackfan23, and does not independently verify publisher claims. See responsible use.