Gemma4-12B-QAT-Uncensored-HauhauCS-Balanced
12BGemmauncensored
from google/gemma-4-12B-it
↓ 696.7K♥ 5178 repos
gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored
gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored is a Heretic-ablated variant of 0xSero/Gemma-4-19B, published on Hugging Face by DavidAU. It is 19B total parameters (mixture-of-experts), 24GB VRAM class class and apache-2.0 licence. It is distributed across 5 repositories in GGUF (imatrix), Transformers and GGUF formats, totalling 6.2K downloads.
Specification
- Model name
- gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored
- Source repository
- DavidAU/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored
- Base model
- 0xSero/Gemma-4-19B
- Publisher
- DavidAU
- Parameters
- 19B (mixture-of-experts)
- VRAM required
- ~12.4 GB at Q4_K_M, 4k context (estimated) — runs on 16 GB cards
- Hardware class
- 10B – 25B — 24GB VRAM class
- Licence
- apache-2.0
- Task
- Image text to text
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF (imatrix), Transformers, GGUF
- First indexed
- 12 Apr 2026
- Last updated
- 30 Oct 2026
Ablation technique
How gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored 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
“A strictly instruct trained (on heretic/uncensored base) version of Gemma's 26B-A4B , shrunk down to 19B-A4B [90 of 128 experts].”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored-i1-GGUF --local-dir ./gemma-4-19b-a4b-it-instruct-heretic-uncensored
llama-cli -m ./gemma-4-19b-a4b-it-instruct-heretic-uncensored/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "DavidAU/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve DavidAU/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored --trust-remote-codeDownloads and variants (4)
Every published repository of gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored-i1-GGUF | GGUF (imatrix) | 3.9K |
| mradermacher/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored-GGUF | GGUF | 1.8K |
| kainah/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored-Q4_K_M-GGUF | GGUF | 286 |
| DavidAU/gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensoredsource | Transformers | 195 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- gemma-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored 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 0xSero/Gemma-4-19B, not added by ablation — at 19B 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 0xSero/Gemma-4-19B’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-4-19B-A4B-it-INSTRUCT-Heretic-Uncensored is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with DavidAU, and does not independently verify publisher claims. See responsible use.