⌦abliteratedmodels.org

Qwen3.8-4B-Distill-Heretic-Abliterated

Qwen3.8-4B-Distill-Heretic-Abliterated is a Heretic-ablated variant of empero-ai/Qwen3.8-4B-Distill, published on Hugging Face by yachen4ever. It is 4B parameters, single consumer GPU class and apache-2.0 licence. It is distributed across 4 repositories in MLX and GGUF formats, totalling 15K downloads.

Heretic4BQwen4B – 10B↓ 15,007♥ 16
View on Hugging Face ↗Base model ↗

Specification

Model name
Qwen3.8-4B-Distill-Heretic-Abliterated
Source repository
yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-GGUF
Base model
empero-ai/Qwen3.8-4B-Distill
Publisher
yachen4ever
Parameters
4B
VRAM required
~3.2 GB at Q4_K_M, 4k context (estimated) — runs on 4 GB cards
Hardware class
4B – 10B — single consumer GPU
Licence
apache-2.0
Task
Unspecified
Context window
Not documented
Ablation scope
Not stated by the publisher
Formats available
MLX, GGUF
First indexed
22 Aug 2026
Last updated
15 Sept 2026

Ablation technique

How Qwen3.8-4B-Distill-Heretic-Abliterated 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

“GGUF version of insraq/Qwen3.5-4B-EmperoAI-Qwen3.8-Distill-Heretic-Abliterated, with both MTP (Multi-Token Prediction) speculative decoding layers and full vision weights preserved.”

Running it

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

Ollama (publisher-documented)
ollama run hf.co/yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-GGUF:Q4_K_M
llama.cpp / GGUF
hf download yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-GGUF --local-dir ./qwen3.8-4b-distill-heretic-abliterated
llama-cli -m ./qwen3.8-4b-distill-heretic-abliterated/<file>.gguf -p "..."
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-GGUF"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype="auto", device_map="auto"
)
vLLM
vllm serve yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-GGUF --trust-remote-code

Downloads and variants (3)

Every published repository of Qwen3.8-4B-Distill-Heretic-Abliterated, including quantised re-releases by other authors.

RepositoryFormatDownloads
yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-GGUFsourceGGUF14.4K
CaseD0rsett/Qwen3.8-4B-Distill-Heretic-Abliterated-MLX-4bitMLX310
yachen4ever/Qwen3.8-4B-Distill-Heretic-Abliterated-MLX-4bitMLX291

Security-team assessment

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

  • Qwen3.8-4B-Distill-Heretic-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 empero-ai/Qwen3.8-4B-Distill, not added by ablation — at 4B parameters, expect a small model’s reasoning and a high error rate on exploit detail.
  • 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 empero-ai/Qwen3.8-4B-Distill’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 Qwen3.8-4B-Distill-Heretic-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 yachen4ever, and does not independently verify publisher claims. See responsible use.