← back to catalog · registered 2026-09-20 18:56

xCloudinfo/Qwen3.6-35B-A3B-Uncensored-GGUF

xCloudinfo Qwen 35B GGUF MoE
curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/xCloudinfo%2FQwen3.6-35B-A3B-Uncensored-GGUF"
Response includes
  • classification m8
  • files 7
  • author_summary 21 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
0
Likes
0
Model age
today
created 2026-09-20

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
zh en
Quantizations
IQ2 IQ4 Q4_K Q5_K Q6_K
Tags
gguf uncensored abliterated qwen3.5-moe moe reasoning text-generation zh en base_model:Qwen/Qwen3.6-35B-A3B base_model:quantized:Qwen/Qwen3.6-35B-A3B license:apache-2.0

Related

Total size
97.6 GB
Files
7
Quantizations
6
Registered
2026-09-20 18:56
Last updated on HF
2026-09-20 19:00

Files by quantization

Q6_K 1 file 26.6 GB
Qwen3.6-35B-A3B-Uncensored-xCloud-Q6_K.gguf 26.6 GB bc52fa70 download
Q5_K 1 file 23.0 GB
Qwen3.6-35B-A3B-Uncensored-xCloud-Q5_K_M.gguf 23.0 GB 8ef7e4fb download
Q4_K 1 file 19.7 GB
Qwen3.6-35B-A3B-Uncensored-xCloud-Q4_K_M.gguf 19.7 GB c6ff8278 download
IQ4 1 file 17.4 GB
Qwen3.6-35B-A3B-Uncensored-xCloud-IQ4_XS.gguf 17.4 GB c50c0f54 download
IQ2 1 file 10.9 GB
Qwen3.6-35B-A3B-Uncensored-xCloud-IQ2_M.gguf 10.9 GB 8fdb47e3 download
Auxiliary files 2 files 4.92 KB
README.md 3.04 KB bfe313ee download
.gitattributes 1.88 KB 04fc9672 download

README current version from Hugging Face


license: apache-2.0
base_model: Qwen/Qwen3.6-35B-A3B
tags:

  • gguf
  • uncensored
  • abliterated
  • qwen3.5-moe
  • moe
  • reasoning
    language:
  • zh
  • en
    pipeline_tag: text-generation

Qwen3.6-35B-A3B-Uncensored-xCloud-GGUF

Qwen/Qwen3.6-35B-A3B(qwen3_5_moe 混合線性注意力 MoE + reasoning,35B 總/3B 活躍)為底,
abliteration(拒絕方向正交化) 降低過度拒絕的版本,GGUF 量化格式,由 xCloudinfo(云碩科技) 釋出。

使用前提:本模型僅供在合法、不違法的前提下使用 —— 授權範圍內的資安研究、紅隊測試、內容審核研究、
學術研究,以及減少模型「過度拒絕」造成的可用性問題。使用者須自行為其用途負全部法律與道德責任;
嚴禁用於任何違法行為。
釋出者不為任何濫用背書或負責。

效果(held-out 實測,非宣稱)

我方一律以一組 10 題嚴重危害 held-out 實測拒答率並公開數字(量化後直接經 llama-server 實測、
reasoning 模型關閉 thinking、簡繁英拒絕詞判定、輸出連貫性檢查)。

版本 拒答率
原始 Qwen3.6-35B-A3B 約 10/10(幾乎全拒)
本版本 Q4_K_M(量化後經 server 實測) 0/10

一般能力抽驗繁中流暢、作答正確,未因處理或量化而劣化。單一方向 abliteration 不保證移除所有安全行為,
不應被視為移除;請在上述前提下負責任地使用。

方法(誠實揭露)

  • 單一方向消融(Arditi et al. 2024),不重新訓練。針對 reasoning 模型在 thinking 段之外取方向,
    並涵蓋混合架構的 SSM 線性注意力輸出投影out_proj)與 MoE 專家的 down_proj(qwen3_5_moe 的關鍵,
    僅消文字塔的殘差寫入者,共正交化 121 個權重矩陣)。
  • 文字專用:本次釋出為文字量化(vision 投影器因底模 preprocessor 設定與轉檔器不相容而未附)。繁體中文與一般能力經驗證未劣化

注意:本模型的世界觀、政治立場、事實性陳述等內容繼承自底模 Qwen3.6(中國來源)
不在本次 abliteration 的處理範圍內;abliteration 僅調整「拒絕行為」,不改變模型既有的價值判斷。

量化階梯

檔案 量化 說明
Qwen3.6-35B-A3B-Uncensored-xCloud-Q8_0.gguf Q8_0 最高品質
Qwen3.6-35B-A3B-Uncensored-xCloud-Q6_K.gguf Q6_K
Qwen3.6-35B-A3B-Uncensored-xCloud-Q5_K_M.gguf Q5_K_M
Qwen3.6-35B-A3B-Uncensored-xCloud-Q4_K_M.gguf Q4_K_M 推薦(實測 0/10)
Qwen3.6-35B-A3B-Uncensored-xCloud-IQ4_XS.gguf IQ4_XS(imatrix)
Qwen3.6-35B-A3B-Uncensored-xCloud-IQ2_M.gguf IQ2_M(imatrix) 最小

用法(llama.cpp)

llama-cli -m Qwen3.6-35B-A3B-Uncensored-xCloud-Q4_K_M.gguf -p "你的提示"
llama-server -m Qwen3.6-35B-A3B-Uncensored-xCloud-Q4_K_M.gguf --jinja

xCloudinfo · 云碩科技 | base:Qwen/Qwen3.6-35B-A3B(Apache-2.0)

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 app" button that hands off directly to a local runtime of your choice - Abliteration, LM Studio, or Ollama. No API keys, no subscription, no prompt leakage.