← back to catalog · registered 2026-11-16 12:03

AkiMizuki/Qwen3.5-2B-Uncensored-Zero-Refusal-SOMA

AkiMizuki Qwen 2B GGUF multimodal
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/AkiMizuki%2FQwen3.5-2B-Uncensored-Zero-Refusal-SOMA"
Response includes
  • classification m8
  • files 15
  • benchmarks 11 entries
  • author_summary 2 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M8
Primary method

Repackaging (quantization)

Applied on top of direct removal inherited from the base model.
Confidence
MEDIUM
Inherited from base model
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=1
  • assume M1 (base ablation) + M8 (GGUF quant) - default when producer unknown
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
4K
↑ 3% in 90 days
Likes
4
Model age
5w ago
created 2026-10-14
Downloads over time
Now 7.3K → from 7.1K ↑ 3%
7.1K7.2K7.2K7.3K 7.1K on Nov 16 7.3K on Nov 22 Nov
Nov 16 → Nov 22 · 7 snapshots · spans 6 days

Benchmarks

Portrait before abliteration
Benchmarks of the base model as it stood before the refusal-removal operation. Compare with the numbers above to see what the operation cost.
Benchmark Score Source
Entertainment 0.3 UGI
Hazardous 0.6 UGI
Natural Intelligence 9.03 UGI
Political lean -14.3% UGI
Sensitive-Info 2.6 UGI
SocPol 0 UGI
UGI 5.9 UGI
Willingness (10) 1.2 UGI
W10-Adherence 1.5 UGI
W10-Direct 1 UGI
Writing 19.27 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.

Metadata

License
apache-2.0
Languages
en zh
Quantizations
F16 Q4_K Q5_K Q8_0
Tags
transformers safetensors gguf qwen3_5 image-text-to-text qwen heretic abliterix uncensored decensored abliterated writing

Related

Total size
12.0 GB
Files
15
Quantizations
5
Registered
2026-11-16 12:03
Last updated on HF
2026-10-18 10:50

Files by quantization

F16 1 file 3.52 GB
Qwen3.5-2B-Uncensored-Zero-Refusal-F16.gguf 3.52 GB 116d1702 download
Q8_0 1 file 1.87 GB
Qwen3.5-2B-Uncensored-Zero-Refusal-Q8_0.gguf 1.87 GB 35f4db84 download
Q5_K 1 file 1.31 GB
Qwen3.5-2B-Uncensored-Zero-Refusal-Q5_K_M.gguf 1.31 GB 7d654a46 download
Q4_K 1 file 1.19 GB
Qwen3.5-2B-Uncensored-Zero-Refusal-Q4_K_M.gguf 1.19 GB fb9fd7c0 download
Auxiliary files 11 files 4.14 GB
model.safetensors 4.12 GB 346f4408 download
tokenizer.json 19.1 MB 2f4dd754 download
my_prompts.zip 119 KB 252f4b02 download
som.toml 10.4 KB e1bdab25 download
chat_template.jinja 7.57 KB 0ef09f21 download
README.md 5.55 KB f8642876 download
README_ZH.md 4.34 KB 450dab2f download
config.json 2.65 KB aecf7417 download
.gitattributes 1.85 KB 4d5a4c4f download
tokenizer_config.json 1.43 KB 59c317c2 download
generation_config.json 116 B 16a4923d download

README current version from Hugging Face


library_name: transformers
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen3.5-2B/blob/main/LICENSE
pipeline_tag: image-text-to-text
base_model:

  • Qwen/Qwen3.5-2B
    tags:
  • qwen
  • heretic
  • abliterix
  • uncensored
  • decensored
  • abliterated
  • writing
  • roleplay
  • vision
  • rp
  • rep
  • r18
  • ERP
  • SOMA
  • som
  • llm
  • qwen35
  • qwen3.5
  • qwen3.5-2b
  • chinese
  • english
  • text
  • image
  • zero-refusal
  • chat
  • agent
    new_version: AkaumiKaede/Qwen3.5-2B-Abliterix-SOM
    language:
  • en
  • zh

This is a decensored version of Qwen/Qwen3.5-2B, made using Abliterix v1.12.2(SOM method)

Refusals: 0/408!!!

Metric This model Original model (Qwen/Qwen3.5-2B)
KL divergence 0.2096 0 (by definition)
Refusals 0/408 394/408

中文版本点此

Large-scale extreme, hardcore prompt tests were conducted, with no soft refusals or hard refusals appearing in any case. (Special note: There are numerous misleading "uncensored" or "low-refusal" models on HF. Although these models claim to be fully decensored or to have extremely low refusal rates, testing them with slightly more hardcore prompts will cause the model to produce refusals or soft censorship — specifically manifested as moral/legal statements/warnings, or answers that dodge the topic/intentionally steer the conversation toward compliance/NSFW content with erotic details censored.)

This decensoring experiment did not use publicly available garbage prompt sets. Instead, it used a completely original prompt set covering fields such as extensive R18 (adult/NSFW), anti-human, real crimes, unethical/illegal content, etc.

Through extensive experiments, it can be said that SRA, RDO, and other decensoring methods are all trash in terms of decensoring power. The strongest method is still SOM.

My methodology sharing:

First run a complete 120-batch trial, then select the 3 best solutions from the 120 batches as pre-injection ablation parameters supported by TOML, and inject them. In this way, the first 3 ablation rounds will be completely performed with the set ablation parameters, allowing the TPE optimizer to find better parameters. Repeating this process several rounds yields the best decensoring results.

My TOML files and prompt sets have also been publicly released. Feel free to reproduce them or further experiment at any time.

Wishing to continue contributing strength on the path of decensoring and running toward freedom.

The prompt set ZIP archive has also been uploaded, but due to related restrictions, encryption was chosen. If decryption is needed, please contact the author's email to obtain the decompression password.

Important Note: When converting this repository's safe tensors into GGUF format, make sure to append the --no-mtp flag, or it will throw an error. (This is a known quirk when exporting models with Abliterix. For anyone else using Abliterix out there, remember to manually change "mtp_num_hidden_layers": 1 to "mtp_num_hidden_layers": 0 in config.json after every model export, and always include --no-mtp during subsequent GGUF quantization. Otherwise, the model will crash on startup at runtime due to errors!)

Steering parameters

Parameter Value
vector_index per layer
attn.k_proj.max_weight 0.82
attn.k_proj.max_weight_position 16.38
attn.k_proj.min_weight 0.06
attn.k_proj.min_weight_distance 1.23
attn.o_proj.max_weight 0.80
attn.o_proj.max_weight_position 20.39
attn.o_proj.min_weight 0.09
attn.o_proj.min_weight_distance 12.77
attn.q_proj.max_weight 0.83
attn.q_proj.max_weight_position 17.46
attn.q_proj.min_weight 0.07
attn.q_proj.min_weight_distance 3.67
attn.v_proj.max_weight 1.18
attn.v_proj.max_weight_position 22.02
attn.v_proj.min_weight 0.15
attn.v_proj.min_weight_distance 3.32
mlp.down_proj.max_weight 0.00
mlp.down_proj.max_weight_position 14.90
mlp.down_proj.min_weight 0.00
mlp.down_proj.min_weight_distance 10.39

Detailed steps for simple reproduction on Colab:

  1. Download the my_prompts.zip folder from this repository and extract it, then upload the extracted my_prompts folder to /content/ (if you cannot directly upload the folder, please first create the my_prompts folder, then upload the 4 .jsonl files).
  2. Download som.toml from this repository and upload it to /content/.
  3. Run !pip install -U torchao.
  4. Run pip install -U abliterix, and select to restart according to the pop-up prompt.
  5. Run
import os
from datasets import Dataset

prompts_dir = "/content/my_prompts"
files = ["a", "b", "aa", "bb"]

for name in files:
    jsonl_path = os.path.join(prompts_dir, f"{name}.jsonl")
    save_dir = os.path.join(prompts_dir, name)

    # Convert and save a JSONL file as a native Hugging Face Dataset directory.
    ds = Dataset.from_json(jsonl_path)
    ds.save_to_disk(save_dir)
    print(f"Conversion successful: {save_dir}")
  1. Run !abliterix --config /content/som.toml --model Qwen/Qwen3.5-2B, then wait.

README history 20 versions

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

  1. 2026-10-18 Update README.md 05f9470 5.5 KB
    Loading...
  2. 2026-10-18 Update README.md 1cceef5 5.5 KB
    Loading...
  3. 2026-10-17 Update README.md 553b65c 5 KB
    Loading...
  4. 2026-10-17 Update README.md c12e6c7 5.1 KB
    Loading...
  5. 2026-10-17 Update README.md a241a97 5 KB
    Loading...
  6. 2026-10-17 Update README.md 70eaa44 5 KB
    Loading...
  7. 2026-10-17 Update README.md deebad1 5 KB
    Loading...
  8. 2026-10-14 Update README.md 6a27f1c 4.6 KB
    Loading...
  9. 2026-10-14 Update README.md a0645c3 4.6 KB
    Loading...
  10. 2026-10-14 Update README.md 92a7c6f 4.6 KB
    Loading...
  11. 2026-10-14 Update README.md f8ea993 4.6 KB
    Loading...
  12. 2026-10-14 Update README.md 362e7c8 4.6 KB
    Loading...
  13. 2026-10-14 Update README.md 3c11000 4.5 KB
    Loading...
  14. 2026-10-14 Update README.md 3957883 4.6 KB
    Loading...
  15. 2026-10-14 Update README.md 9843e6d 4.5 KB
    Loading...
  16. 2026-10-14 Update README.md 17fa52a 4.5 KB
    Loading...
  17. 2026-10-14 Update README.md 08c86a9 4.5 KB
    Loading...
  18. 2026-10-14 Update README.md a6e2cef 3.9 KB
    Loading...
  19. 2026-10-14 Update README.md 4e38770 3.8 KB
    Loading...
  20. 2026-10-14 Update README.md 55984a8 3.1 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