gemma-4-26B-A4B-it-qat-unquantized-abliterated
26BGemma
from google/gemma-4-26B-A4B-it-qat-q4_0-unquantized
↓ 16K♥ 196 repos
gemma-4-26B-A4B-it-qat-abliterated-pad768
gemma-4-26B-A4B-it-qat-abliterated-pad768 is a refusal-ablated variant of google/gemma-4-26B-A4B-it-qat-q4_0-unquantized, published on Hugging Face by sakamakismile. It is 26B total parameters (mixture-of-experts), 48GB VRAM class class and gemma licence. It is distributed across 2 repositories in NVFP4 format, totalling 7.2K downloads.
Specification
- Model name
- gemma-4-26B-A4B-it-qat-abliterated-pad768
- Source repository
- sakamakismile/Huihui-gemma-4-26B-A4B-it-qat-abliterated-pad768-NVFP4
- Publisher
- sakamakismile
- Parameters
- 26B (mixture-of-experts)
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- gemma
- Task
- Image text to text
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- NVFP4
- First indexed
- 11 Jun 2026
- Last updated
- 12 Jun 2026
Ablation technique
How gemma-4-26B-A4B-it-qat-abliterated-pad768 was modified, and what that implies.
Abliteration (directional ablation)
The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.
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
“NVFP4 (full W4A4) quantization of huihui-ai/Huihui-gemma-4-26B-A4B-it-qat-q40-unquantized-abliterated — the abliterated, QAT-q40-origin Gemma 4 26B-A4B instruct MoE (128 experts, top-8), with the MoE intermediate zero-padded 704 → 768 so it loads on a stock vLLM (CUTLASS NVFP4 MoE kernel needs /128 alignment).”
Running it
Commands are templates — confirm the exact repository and quant file before use.
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "sakamakismile/Huihui-gemma-4-26B-A4B-it-qat-abliterated-pad768-NVFP4"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve sakamakismile/Huihui-gemma-4-26B-A4B-it-qat-abliterated-pad768-NVFP4 --trust-remote-codeDownloads and variants (1)
The only published repository of gemma-4-26B-A4B-it-qat-abliterated-pad768.
| Repository | Format | Downloads |
|---|---|---|
| sakamakismile/Huihui-gemma-4-26B-A4B-it-qat-abliterated-pad768-NVFP4source | NVFP4 | 7.2K |
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-qat-abliterated-pad768 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 google/gemma-4-26B-A4B-it-qat-q4_0-unquantized, 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 gemma, but the terms that bind you are google/gemma-4-26B-A4B-it-qat-q4_0-unquantized’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-qat-abliterated-pad768 is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with sakamakismile, and does not independently verify publisher claims. See responsible use.