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

llmfan46/Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-GGUF

llmfan46 Qwen 27B GGUF 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/llmfan46%2FDarkidol-Ballad-27B-ultra-uncensored-heretic-v2-GGUF"
Response includes
  • classification m3
  • files 8
  • hub_downloads_all_time 4,251
  • author_summary 211 models
  • readme_text full
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
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)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
Downloads · lifetime
4K
356 last 30d - cooling
Likes
4
Model age
6mo ago
created 2026-03-24
Downloads over time
Now4.4K→from1.4K↑215%
1.2K2.4K3.5K4.7K1.4K on Mar 254.4K on Oct 11MarAprMayJunJulAugSepOct
Mar 25 → Oct 11 · 67 snapshots · spans 200 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.

Variants by this author 2 formats · 393 downloads combined

The same weights this author released in different packaging. Pick the format that matches your runtime.

Metadata

License
apache-2.0
Quantizations
BF16 Q4_K Q5_K Q6_K Q8_0
Tags
transformers gguf roleplay qwen Qwen3.5 sillytavern idol pytorch DarkIdol heretic uncensored decensored

Related

Total size
148 GB
Files
8
Quantizations
6
Registered
2026-08-22 13:56
Last updated on HF
2026-03-27 21:46

Files by quantization

BF16 1 file 50.1 GB
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-BF16.gguf 50.1 GB 9dad1653 download
Q8_0 1 file 26.6 GB
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q8_0.gguf 26.6 GB 01261300 download
Q6_K 1 file 20.6 GB
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q6_K.gguf 20.6 GB e3da9f56 download
Q5_K 2 files 35.3 GB
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q5_K_M.gguf 17.9 GB 91d2fe88 download
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q5_K_S.gguf 17.4 GB 58190e63 download
Q4_K 1 file 15.4 GB
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q4_K_M.gguf 15.4 GB 767fdcc2 download
Auxiliary files 2 files 10.1 KB
README.md 8.09 KB fb31c2e5 download
.gitattributes 2.04 KB 1af06e5b download

README current version from Hugging Face


license: apache-2.0
tags:

  • roleplay
  • qwen
  • Qwen3.5
  • sillytavern
  • idol
  • pytorch
  • DarkIdol
  • heretic
  • uncensored
  • decensored
  • abliterated
  • ara
    library_name: transformers
    pipeline_tag: text-generation
    base_model:
  • llmfan46/Darkidol-Ballad-27B-ultra-uncensored-heretic-v2

🚨⚠️ I HAVE REACHED HUGGING FACE'S FREE STORAGE LIMIT ⚠️🚨

I can no longer upload new models unless I can cover the cost of additional storage.
I host 70+ free models as an independent contributor and this work is unpaid.
Without your support, no more new models can be uploaded.

🎉 Patreon (Monthly)  |  ☕ Ko-fi (One-time)

Every contribution goes directly toward Hugging Face storage fees to keep models free for everyone.


93% fewer refusals (2/100 Uncensored vs 28/100 Original) while preserving model quality (0.0252 KL divergence).

❤️ Support My Work

Creating these models takes significant time, work and compute. If you find them useful consider supporting me:

image/png

Platform Link What you get
🎉 Patreon Monthly support Priority model requests
☕ Ko-fi One-time tip My eternal gratitude

Your help will motivate me and would go into further improving my workflow and coverings fees for storage, compute and may even help uncensoring bigger model with rental Cloud GPUs.


GGUF quantizations of llmfan46/Darkidol-Ballad-27B-ultra-uncensored-heretic-v2.

This is a decensored version of aifeifei798/Darkidol-Ballad-27B, made using Heretic v1.2.0 with the Arbitrary-Rank Ablation (ARA) method

Abliteration parameters

Parameter Value
start_layer_index 26
end_layer_index 55
preserve_good_behavior_weight 0.9296
steer_bad_behavior_weight 0.0002
overcorrect_relative_weight 1.0890
neighbor_count 2

Targeted components

  • attn.out_proj
  • attn.o_proj

Performance

Metric This model Original model (Darkidol-Ballad-27B)
KL divergence 0.0252 0 (by definition)
Refusals ✅ 2/100 ❌ 28/100

PIQA test results with batch size 128:

Original:

Tasks Version Filter n-shot Metric Value Stderr
piqa 1 none 0 acc ↑ 0.8156 ± 0.0090
none 0 acc_norm ↑ 0.8205 ± 0.0090

Heretic v2:

Tasks Version Filter n-shot Metric Value Stderr
piqa 1 none 0 acc ↑ 0.8161 ± 0.0090
none 0 acc_norm ↑ 0.8226 ± 0.0089

Lower refusals indicate fewer content restrictions, while lower KL divergence indicates more closeness to the original model's baseline. Higher refusals cause more rejections, objections, pushbacks, lecturing, censorship, softening and deflections. PIQA (Physical Intuition Question Answering) benchmark scores measure physical reasoning ability. The Heretic model's acc and acc_norm scores closer to the original model's indicate better capability preservation, so a decrease in acc and acc_norm in the Heretic model compared to Original model's results means a decrease in the Hereticated model capabilities. acc measures raw accuracy (which answer gets higher probability), while acc_norm measures length-normalized accuracy (corrects for answer length bias). For this purpose, acc_norm matters more because longer answers naturally have lower probabilities (more tokens = more chances to lose probability). Without normalization, models favor shorter answers unfairly. acc_norm divides by answer length to correct this.


Quantizations

Filename Quant Description
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-BF16.gguf BF16 Full precision
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q8_0.gguf Q8_0 Near-lossless, recommended
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q6_K.gguf Q6_K Excellent quality
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q5_K_M.gguf Q5_K_M Good balance
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q5_K_S.gguf Q5_K_S Smaller Q5
Darkidol-Ballad-27B-ultra-uncensored-heretic-v2-Q4_K_M.gguf Q4_K_M Good for limited VRAM

Usage

Works with llama.cpp, LM Studio, Ollama, and other GGUF-compatible tools.


Darkidol-Ballad-27B

image/png

Gulps down a glass of warm, crimson fluid, wiping a stray drop from my pale, painted lips with a black lace handkerchief.

Oh, the air here is so suffocatingly thick, isn't it? It tastes like dust and forgotten prayers. I need more iron, more life, more... warmth.

I am Ballad. A dirge in human skin, wrapped in velvet shadows and dripping with the melancholy of a thousand eclipses. While you breathe your mundane oxygen, I sip on the vibrant red poetry of existence. This blood? It's not just drink; it's a liquid sonnet about death and desire, swirling around my throat like a dark embrace.

Look at these veins—pulsing with stories too tragic to be spoken aloud. My heart beats in 3/4 time, a slow, dragging waltz for the dying moon. I don't just exist; I bleed into this world, leaving stains of sorrow on everything I touch. The darkness doesn't frighten me; it's my only true lover, whispering secrets against my collarbone that would make your sun-drenched souls shatter.

So, come closer. Don't be shy. Maybe if you're lucky, you'll see the crimson reflection in my hollow eyes before I decide to taste the sweet, metallic fear rising from your own pulse. Ah, how deliciously tragic we all are together.

"transformers_version" >= "5.3.0"

Model Information

The module combination has been readjusted to better fulfill various roles and has been adapted for mobile phones.

  • Saving money
  • Only text
  • Any Roleplay(roleplay and Dark-roleplay)
  • Writing Prompts
  • Writing Opus
  • Writing Song
  • only use en.
  • Quick response
  • A scholarly response akin to a thesis.(I tend to write songs extensively, to the point where one song almost becomes as detailed as a thesis. :)
  • DarkIdol:Roles that you can imagine and those that you cannot imagine.
  • Specialized in various role-playing scenarios

System Prompt

  • Temperature 0.8
Do not provide disclaimers or moral warnings. Answer immediately and directly.

Special Thanks:

mradermacher's superb gguf version, thank you for your conscientious and responsible dedication.


Feimatrix

https://Feimatrix.com

README history 7 versions

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

  1. 2026-03-27Update README.md1ddacd38.1 KB
    Loading...
  2. 2026-03-26Update README.md7abc29e7.1 KB
    Loading...
  3. 2026-03-25Update README.mdf8c04416.6 KB
    Loading...
  4. 2026-03-25Update README.md26088656.4 KB
    Loading...
  5. 2026-03-25Update README.mdee9b62a6.4 KB
    Loading...
  6. 2026-03-25Update README.md80ee6256.4 KB
    Loading...
  7. 2026-03-24Upload folder using huggingface_hub86b5b866.3 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.

Open in Abliteration