gemma-4-31B-it-abliterated
31BGemma
from google/gemma-4-31B-it
↓ 766.5K♥ 11845 repos
gemma-4-31B-it-heretic-GGUF_ND
gemma-4-31B-it-heretic-GGUF_ND is a Heretic-ablated variant of coder3101/gemma-4-31B-it-heretic, published on Hugging Face by NeuralDreamer. It is 31B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 2 repositories in GGUF (imatrix) format, totalling 5.2K downloads.
Specification
- Model name
- gemma-4-31B-it-heretic-GGUF_ND
- Source repository
- NeuralDreamer/gemma-4-31B-it-heretic-i1-GGUF_ND
- Base model
- coder3101/gemma-4-31B-it-heretic
- Publisher
- NeuralDreamer
- Parameters
- 31B
- 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 (imatrix)
- First indexed
- 5 Aug 2026
- Last updated
- 5 Aug 2026
Ablation technique
How gemma-4-31B-it-heretic-GGUF_ND 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
“For a convenient overview and download list, visit our model page for this model.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download NeuralDreamer/gemma-4-31B-it-heretic-i1-GGUF_ND --local-dir ./gemma-4-31b-it-heretic-gguf-nd
llama-cli -m ./gemma-4-31b-it-heretic-gguf-nd/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "NeuralDreamer/gemma-4-31B-it-heretic-i1-GGUF_ND"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve NeuralDreamer/gemma-4-31B-it-heretic-i1-GGUF_ND --trust-remote-codeDownloads and variants (1)
The only published repository of gemma-4-31B-it-heretic-GGUF_ND.
| Repository | Format | Downloads |
|---|---|---|
| NeuralDreamer/gemma-4-31B-it-heretic-i1-GGUF_NDsource | GGUF (imatrix) | 5.2K |
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-heretic-GGUF_ND 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-4-31B-it-heretic, 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 coder3101/gemma-4-31B-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-4-31B-it-heretic-GGUF_ND is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with NeuralDreamer, and does not independently verify publisher claims. See responsible use.