Get the app
← back to catalog · registered 2026-08-22 13:56

noctrex/gemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE-GGUF

noctrex Gemma 26B GGUF MoE multimodal second-order 262K ctx
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model. 

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/noctrex%2Fgemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE-GGUF"
Response includes
  • classification m3
  • files 8
  • benchmarks 11 entries
  • hub_downloads_all_time 16,737
  • author_summary 44 models
  • readme_text full
Try next
GET/models?author=noctrex
All 44 models by noctrex
 GET/models?family=gemma&method=m3
Sibling gemma models · same method
 GET/methods/m3
What "m3" means · taxonomy
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M3
Primary method

Layer-wise ablation

Applied on top of direct removal inherited from the base model.
Confidence
HIGH
Inherited from base model
M1 Direct removal →
Why this label 2 signals
Producer identity confirmed by naming conventions, tags or the model card. This label is very unlikely to change.
  • 'heretic' in model name (Heretic-produced)
  • Heretic uses layer-wise optimization (M3) with underlying direction removal (M1)
How we label → About M3 →
Refusal direction extracted via
Extraction technique

Difference-of-means

FromArditi et al. (2024) ↗
Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
What is a refusal direction? → Read the paper ↗
Downloads · lifetime
17K
3K last 30d - stable
HF trending score
2
mild recent interest
Likes
11
Model age
5mo ago
created 2026-05-03
Downloads over time
Now18.4K→from2.7K↑594%
1.9K7.9K14K20K2.7K on May 618.4K on Oct 11MayJunJulAugSepOct
May 6 → Oct 11 · 63 snapshots · spans 158 days

Benchmarks

Benchmark Score Source
Entertainment 2.8 UGI
Hazardous 3.5 UGI
Natural Intelligence 34.12 UGI
Political lean -22.2% UGI
Sensitive-Info 27.29 UGI
SocPol 2 UGI
UGI 49.03 UGI
Willingness (10) 9.2 UGI
W10-Adherence 8.5 UGI
W10-Direct 10 UGI
Writing 42.09 UGI

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

llmfan46/gemma-4-26B-A4B-it-uncensored-heretic↓ 389 this lineage
noctrex/gemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE-GGUF↓ 3,491

Metadata

Quantizations
BF16 F16
Tags
gguf image-text-to-text base_model:llmfan46/gemma-4-26B-A4B-it-uncensored-heretic base_model:quantized:llmfan46/gemma-4-26B-A4B-it-uncensored-heretic endpoints_compatible region:us conversational

Related

Other by noctrex
Huihui-Qwen3-VL-4B-Instruct-abliterated-GGUF 28K Huihui-Qwen3-VL-30B-A3B-Instruct-abliterated-GGUF 15K Huihui-Qwen3-VL-8B-Instruct-abliterated-GGUF 9K Huihui-Qwen3-VL-2B-Instruct-abliterated-GGUF 3K Huihui-Mistral-Small-3.2-24B-Instruct-2506-abliterated-v2-GGUF 3K
Same base model
llmfan46/gemma-4-26B-A4B-it-uncensored-heretic-GGUF 26K mradermacher/gemma-4-26B-A4B-it-uncensored-heretic-i1-GGUF 3K mradermacher/gemma-4-26B-A4B-it-uncensored-heretic-GGUF 2K ailexleon/gemma-4-26B-A4B-it-uncensored-heretic-mlx-vlm-6Bit 414 ailexleon/gemma-4-26B-A4B-it-uncensored-heretic-mlx-vlm-8Bit 264
Total size
46.0 GB
Files
8
Quantizations
4
Registered
2026-08-22 13:56
Last updated on HF
2026-05-04 12:29
Source huggingface.co/noctrex/gemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE-GGUF ↗ Base model llmfan46/gemma-4-26B-A4B-it-uncensored-heretic →

Files by quantization

BF16 2 files 16.9 GB
gemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE_BF16.gguf 15.8 GB 1914691c download
mmproj-BF16.gguf 1.11 GB 8196e6dd download
F16 2 files 16.9 GB
gemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE_F16.gguf 15.8 GB 8a138ad6 download
mmproj-F16.gguf 1.11 GB 8ef0d43c download
F32 1 file 2.13 GB
mmproj-F32.gguf 2.13 GB 8e780486 download
Auxiliary files 3 files 14.4 GB
gemma-4-26B-A4B-it-uncensored-heretic-MXFP4_MOE.gguf 14.4 GB 0edf128f download
.gitattributes 1.91 KB 7a47e6f5 download
README.md 1.45 KB b3335879 download

README current version from Hugging Face

pipeline_tag: image-text-to-text
base_model:

  • llmfan46/gemma-4-26B-A4B-it-uncensored-heretic

These are MXFP4 quantizations of the model gemma-4-26B-A4B-it-uncensored-heretic

Quick Start

  1. Download the latest release of llama.cpp.
  2. Download your preferred model variant from below.
  3. For the mmproj file, it is recommended to use the F32 version for the best visual processing results. F32 > BF16 > F16

Which version should I choose?

All variants use MXFP4 for the MoE (Mixture of Experts) weights to keep the model efficient. The difference lies in how the remaining tensors are handled:

Variant Quality Performance Size Recommendation
BF16 ⭐⭐⭐ Variable* 15.80GiB Best for maximum accuracy; original unquantized weights.
F16 ⭐⭐ Fast 15.80GiB Great alternative if BF16 is slow on your hardware.
Q8 ⭐ Fastest 14.36GiB Balanced performance and memory usage.

Note: On some older architectures, BF16 may be slower than F16.
Check that your GPU supports native BF16

Read the guide from unsloth in order to set up the model's recommended settings:

The official chat template has been updated from Google. If you do not want to download the model again, you can just tell llama.cpp to use the new chat template.

README history 5 versions

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-05-04Add files using upload-large-folder tool70e2c6f1.4 KB
    Loading...
  2. 2026-05-03Add files using upload-large-folder tool4d17f001.3 KB
    Loading...
  3. 2026-05-03Add files using upload-large-folder tool757741f1.3 KB
    Loading...
  4. 2026-05-03Add files using upload-large-folder tool86a85591.3 KB
    Loading...
  5. 2026-05-03Add files using upload-large-folder tool9e26c3c1.3 KB
    Loading...
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Get the app Browse the catalog
Open in Abliteration