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

nightmedia/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-qx86-hi-mlx

nightmedia Gemma 31B multimodal second-order
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/nightmedia%2Fgemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-qx86-hi-mlx"
Response includes
  • classification m3
  • files 14
  • hub_downloads_all_time 2,651
  • author_summary 51 models
  • readme_text full
Try next
GET/models?author=nightmedia
All 51 models by nightmedia
 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
3K
63 last 30d - cooling
Likes
0
Model age
6mo ago
created 2026-04-06
Downloads over time
Now2.7K→from1.8K↑52%
1.7K2.1K2.4K2.8K1.8K on Apr 152.7K on Oct 11AprMayJunJulAugSepOct
Apr 15 → Oct 11 · 65 snapshots · spans 179 days

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.

DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking↓ 271 this lineage
nightmedia/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-qx86-hi-mlx↓ 63

Metadata

License
apache-2.0
Tags
mlx safetensors gemma4 nightmedia gemma google Deckard(qx) image-text-to-text conversational base_model:DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking base_model:quantized:DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking license:apache-2.0

Related

Other by nightmedia
Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored 2K OpenAI-gpt-oss-20B-INSTRUCT-Heretic-Uncensored-MXFP4-q8-hi-mlx 2K Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC-qx86-hi-mlx 1K Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODER-qx4-mlx 714 Qwen3.5-9B-Claude-4.6-Opus-Deckard-V4.2-Uncensored-Heretic-Thinking-qx86-hi-mlx 583
Same base model
DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-Instruct-GGUF 3K mradermacher/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-GGUF 2K mradermacher/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-i1-GGUF 1K Jackxuanxuan/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-i1-GGUF 1K Jackxuanxuan/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-GGUF 468
Total size
28.1 GB
Files
14
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-04-13 21:09
Source huggingface.co/nightmedia/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-qx86-hi-mlx ↗ Base model DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking →

Files by quantization

Auxiliary files 14 files 28.1 GB
model-00002-of-00006.safetensors 5.00 GB 1e193d4a download
model-00001-of-00006.safetensors 5.00 GB 88432353 download
model-00005-of-00006.safetensors 4.98 GB 04d3cdbf download
model-00003-of-00006.safetensors 4.95 GB 73d17d72 download
model-00004-of-00006.safetensors 4.95 GB 80bfce72 download
model-00006-of-00006.safetensors 3.18 GB 76f23921 download
tokenizer.json 30.7 MB cc8d3a0c download
model.safetensors.index.json 201 KB caea1e9f download
config.json 104 KB aa2e47f0 download
chat_template.jinja 16.0 KB 3a5e1df6 download
tokenizer_config.json 2.63 KB 872a4ded download
processor_config.json 2.60 KB f08bdba3 download
.gitattributes 1.53 KB 52373fe2 download
README.md 871 B 8633fa73 download

README current version from Hugging Face

license: apache-2.0
license_link: https://ai.google.dev/gemma/docs/gemma_4_license
pipeline_tag: image-text-to-text
base_model: DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking
tags:

  • nightmedia
  • gemma4
  • gemma
  • google
  • Deckard(qx)
  • mlx

gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking-qx86-hi-mlx

This is a Deckard(qx) experimental quant.

Brainwaves

         arc   arc/e boolq hswag obkqa piqa  wino
qx86-hi  ...coming soon
Instruct
mxfp8    0.583,0.747,0.894,0.734,0.430,0.787,0.719

Base model

gemma-4-31B-it (Instruct)
qx86-hi  0.496,0.653,0.901,0.624,0.380,0.732,0.653

Previous model

gemma-3-27b-it-heretic
q8       0.557,0.711,0.868,0.533,0.452,0.706,0.695

As I don't have an easy way to test this until LMStudio supports it, please Like it only if you had a good experience.

Thank you,

-G

README history 7 versions

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

  1. 2026-04-07Update README.mdab9952b871 B
    Loading...
  2. 2026-04-06Update README.mdd5ab9eb851 B
    Loading...
  3. 2026-04-06Update README.md583b486781 B
    Loading...
  4. 2026-04-06Update README.mdd0799ed681 B
    Loading...
  5. 2026-04-06Update README.mde4d0057630 B
    Loading...
  6. 2026-04-06Update README.md3c71767840 B
    Loading...
  7. 2026-04-06Add files using upload-large-folder tool5568a9e581 B
    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