gemma-4-31B-it-abliterated
31BGemma
from google/gemma-4-31B-it
↓ 766.5K♥ 11845 repos
gemma-4-31b-it-abliterated-heretic
gemma-4-31b-it-abliterated-heretic is a Heretic-ablated variant of trohrbaugh/gemma-4-31b-it-heretic-ara, published on Hugging Face by alonsoko. It is 31B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 2 repositories in AWQ format, totalling 143.8K downloads.
Specification
- Model name
- gemma-4-31b-it-abliterated-heretic
- Source repository
- alonsoko/gemma-4-31b-it-abliterated-heretic-AWQ-W4A16
- Base model
- trohrbaugh/gemma-4-31b-it-heretic-ara
- Publisher
- alonsoko
- Parameters
- 31B
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- apache-2.0
- Task
- Image text to text
- Context window
- 256K tokens (as documented)
- Ablation scope
- Not stated by the publisher
- Formats available
- AWQ
- First indexed
- 6 Apr 2026
- Last updated
- 17 May 2026
Ablation technique
How gemma-4-31b-it-abliterated-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
“projection layers are excluded from quantization. Only language-model linear layers (attention Q/K/V/O and MLP gate/up/down) are quantized.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "alonsoko/gemma-4-31b-it-abliterated-heretic-AWQ-W4A16"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve alonsoko/gemma-4-31b-it-abliterated-heretic-AWQ-W4A16 --trust-remote-codeDownloads and variants (1)
The only published repository of gemma-4-31b-it-abliterated-heretic.
| Repository | Format | Downloads |
|---|---|---|
| alonsoko/gemma-4-31b-it-abliterated-heretic-AWQ-W4A16source | AWQ | 143.8K |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- gemma-4-31b-it-abliterated-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 trohrbaugh/gemma-4-31b-it-heretic-ara, not added by ablation — at 31B 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 trohrbaugh/gemma-4-31b-it-heretic-ara’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-4-31b-it-abliterated-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 alonsoko, and does not independently verify publisher claims. See responsible use.