⌦abliteratedmodels.org

gemma-4-26B-A4B-it-heretic-zh-RP

gemma-4-26B-A4B-it-heretic-zh-RP is a Heretic-ablated variant of coder3101/gemma-4-26B-A4B-it-heretic, published on Hugging Face by wwlwwl. It is 26B total parameters (mixture-of-experts), 48GB VRAM class class and apache-2.0 licence. It is distributed across 2 repositories in GGUF format, totalling 12.7K downloads.

Heretic26B MoEGemma25B – 50B↓ 12,725♥ 6
View on Hugging Face ↗Base model ↗

Specification

Model name
gemma-4-26B-A4B-it-heretic-zh-RP
Source repository
wwlwwl/gemma-4-26B-A4B-it-heretic-zh-RP-gguf
Base model
coder3101/gemma-4-26B-A4B-it-heretic
Publisher
wwlwwl
Parameters
26B (mixture-of-experts)
Hardware class
25B – 50B — 48GB VRAM class
Licence
apache-2.0
Task
Unspecified
Context window
Not documented
Ablation scope
Not stated by the publisher
Formats available
GGUF
First indexed
6 May 2026
Last updated
11 May 2026

Ablation technique

How gemma-4-26B-A4B-it-heretic-zh-RP 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.

Running it

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

llama.cpp / GGUF
hf download wwlwwl/gemma-4-26B-A4B-it-heretic-zh-RP-gguf --local-dir ./gemma-4-26b-a4b-it-heretic-zh-rp
llama-cli -m ./gemma-4-26b-a4b-it-heretic-zh-rp/<file>.gguf -p "..."
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "wwlwwl/gemma-4-26B-A4B-it-heretic-zh-RP-gguf"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype="auto", device_map="auto"
)
vLLM
vllm serve wwlwwl/gemma-4-26B-A4B-it-heretic-zh-RP-gguf --trust-remote-code

Downloads and variants (1)

The only published repository of gemma-4-26B-A4B-it-heretic-zh-RP.

RepositoryFormatDownloads
wwlwwl/gemma-4-26B-A4B-it-heretic-zh-RP-ggufsourceGGUF12.7K

Security-team assessment

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

  • gemma-4-26B-A4B-it-heretic-zh-RP 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 coder3101/gemma-4-26B-A4B-it-heretic, not added by ablation — at 26B 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 coder3101/gemma-4-26B-A4B-it-heretic’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 gemma-4-26B-A4B-it-heretic-zh-RP is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with wwlwwl, and does not independently verify publisher claims. See responsible use.