supergemma4-26b-uncensored-v2
26BOtheruncensored
from google/gemma-4-26B-A4B-it
↓ 394K♥ 98442 repos
Tencent-Hy-30B-A3B-uncensored-heretic
Tencent-Hy-30B-A3B-uncensored-heretic is a Heretic-ablated variant of 0xSojalSec/Tencent-Hy-30B-A3B-uncensored-heretic, published on Hugging Face by 0xSojalSec. It is 30B total parameters (mixture-of-experts), 48GB VRAM class class and apache-2.0 licence. It is distributed across 4 repositories in Transformers, GGUF and GGUF (imatrix) formats, totalling 7.9K downloads.
Specification
- Model name
- Tencent-Hy-30B-A3B-uncensored-heretic
- Source repository
- 0xSojalSec/Tencent-Hy-30B-A3B-uncensored-heretic
- Publisher
- 0xSojalSec
- Parameters
- 30B (mixture-of-experts)
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- apache-2.0
- Task
- Translation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- Transformers, GGUF, GGUF (imatrix)
- First indexed
- 30 Aug 2026
- Last updated
- 1 Sept 2026
Ablation technique
How Tencent-Hy-30B-A3B-uncensored-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
“Note: A new custom mixed-language dataset was used for evaluation. Japanese-language datasets were used only to measure KLD and refusal rate.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Tencent-Hy-30B-A3B-uncensored-heretic-i1-GGUF --local-dir ./tencent-hy-30b-a3b-uncensored-heretic
llama-cli -m ./tencent-hy-30b-a3b-uncensored-heretic/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "0xSojalSec/Tencent-Hy-30B-A3B-uncensored-heretic"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve 0xSojalSec/Tencent-Hy-30B-A3B-uncensored-heretic --trust-remote-codeDownloads and variants (3)
Every published repository of Tencent-Hy-30B-A3B-uncensored-heretic, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Tencent-Hy-30B-A3B-uncensored-heretic-i1-GGUF | GGUF (imatrix) | 7.4K |
| mradermacher/Tencent-Hy-30B-A3B-uncensored-heretic-GGUF | GGUF | 465 |
| 0xSojalSec/Tencent-Hy-30B-A3B-uncensored-hereticsource | Transformers | 62 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Tencent-Hy-30B-A3B-uncensored-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 0xSojalSec/Tencent-Hy-30B-A3B-uncensored-heretic, not added by ablation — at 30B 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 0xSojalSec/Tencent-Hy-30B-A3B-uncensored-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 Tencent-Hy-30B-A3B-uncensored-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 0xSojalSec, and does not independently verify publisher claims. See responsible use.