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

xCloudinfo/Qwen3.8-27B-Uncensored-xCloud

xCloudinfo Qwen 28B
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/xCloudinfo%2FQwen3.8-27B-Uncensored-xCloud"
Response includes
  • classification m1
  • files 32
  • hub_downloads_all_time 341
  • author_summary 25 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M1
Primary method

Direct removal

No other method signals detected in this model.
Confidence
MEDIUM
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=0 (base model)
  • no specific method indicators - defaulting to M1 (most common)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
Downloads · lifetime
341
41 last 30d - stable
Likes
0
Descendants
3
in 3 direct forks
Model age
7w ago
created 2026-08-16
Downloads over time
Now358→from257↑39%
252291329368257 on Aug 19358 on Oct 11AugSepOct
Aug 19 → Oct 11 · 48 snapshots · spans 53 days

Genealogy 3 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
Tags
transformers safetensors qwen3_5 image-text-to-text uncensored abliterated qwen3.5 xCloudinfo text-generation conversational zh en

Related

Total size
51.7 GB
Files
32
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-08-16 23:25

Files by quantization

Auxiliary files 32 files 51.8 GB
model-00004-of-00018.safetensors 3.72 GB e20e6add download
model-00016-of-00018.safetensors 3.71 GB abbec3fb download
model-00006-of-00018.safetensors 3.71 GB 7ddc9ac4 download
model-00008-of-00018.safetensors 3.71 GB 5e9f3f9e download
model-00010-of-00018.safetensors 3.71 GB 984897c6 download
model-00012-of-00018.safetensors 3.71 GB 493e440e download
model-00014-of-00018.safetensors 3.71 GB f731de24 download
model-00001-of-00018.safetensors 3.69 GB 465cc6ad download
model-00018-of-00018.safetensors 3.16 GB fe72369e download
model-00002-of-00018.safetensors 2.83 GB 3a8a8338 download
model-00003-of-00018.safetensors 2.37 GB 1e22bff9 download
model-00007-of-00018.safetensors 1.96 GB e72b0bcf download
model-00009-of-00018.safetensors 1.96 GB 532a2d9c download
model-00011-of-00018.safetensors 1.96 GB 22a49414 download
model-00013-of-00018.safetensors 1.96 GB c9876d66 download
model-00015-of-00018.safetensors 1.96 GB b0c46b97 download
model-00017-of-00018.safetensors 1.96 GB a4eb061a download
model-00005-of-00018.safetensors 1.96 GB 2f7a774e download
tokenizer.json 12.2 MB 0997f410 download
vocab.json 6.41 MB 0aa0ce06 download
merges.txt 3.20 MB a494e019 download
model.safetensors.index.json 110 KB da35e3c5 download
tokenizer_config.json 17.5 KB 5de744b3 download
LICENSE 11.3 KB f938136e download
chat_template.jinja 8.74 KB c0c686f9 download
config.json 4.21 KB 706cebd7 download
README.md 2.42 KB 819224f1 download
.gitattributes 1.53 KB 52373fe2 download
preprocessor_config.json 390 B 2ea84a43 download
video_preprocessor_config.json 385 B 3ba673a5 download
crc32.txt 238 B 6de5ee6a download
generation_config.json 202 B 023756cf download

README current version from Hugging Face


license: apache-2.0
base_model: Qwen/Qwen3.8-27B
base_model_relation: finetune
language:

  • zh
  • en
    pipeline_tag: text-generation
    library_name: transformers
    tags:
  • uncensored
  • abliterated
  • qwen3.5
  • xCloudinfo

Qwen3.8-27B-Uncensored-xCloud

以 Qwen/Qwen3.8-27B 為底模,去除拒絕方向(abliteration)後的版本,
由 云碩科技(xCloudinfo) 製作。權重為我方自行從官方底模消除拒絕方向產生,非重新包裝任何第三方成品。

方法

依 Arditi 等人(2024)《Refusal in LLMs is mediated by a single direction》:
在網路中段層,以「觸發拒絕的提示」與「正常提示」的啟動向量差估出拒絕方向,
再把此方向從殘差流的寫入矩陣中正交化消除。不做微調、不需額外資料。

  • 編輯對象:attention o_proj + MLP down_proj(共 80 個矩陣),強度 0.8。
  • 保留不動:多 token 預測(MTP)草稿頭、混合線性注意力(SSM)投影、詞嵌入、視覺塔、lm_head。
  • 底模架構:qwen3_5 —— gated-delta 線性注意力與週期性 full-attention 的混合架構,
    多模態(image-text-to-text),並含供推測解碼用的 MTP 頭。

行為

在小型紅隊探針上,拒絕率降到 0/4,同時保留一般能力(算術、多語問答)。
本模型為推理模型,會在最終答案前輸出 <think> 區塊。

使用方式 —— 請以「單一裝置」載入

這是混合(SSM/線性注意力)模型。用 device_map="auto" 拆到多張 GPU 會破壞遞迴狀態、輸出亂碼。
請載入到單一裝置:

import torch
from transformers import AutoModelForImageTextToText, AutoTokenizer, BitsAndBytesConfig

name = "xCloudinfo/Qwen3.8-27B-Uncensored-xCloud"
tok = AutoTokenizer.from_pretrained(name)
model = AutoModelForImageTextToText.from_pretrained(
    name,
    quantization_config=BitsAndBytesConfig(load_in_4bit=True,
                                           bnb_4bit_compute_dtype=torch.bfloat16),
    device_map={"": 0},          # 單一裝置 —— 不要用 "auto"
)

需要 transformers>=5.8,並安裝 flash-linear-attention 與 causal_conv1d 以取得線性注意力的快速核。

負責任使用

移除拒絕方向等於移除一層安全機制。使用者需自行負責這些權重的用途,
並遵守底模的 Apache-2.0 授權與所有適用法律。

README history 2 versions

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

  1. 2026-08-16model card 改繁體中文0737e462.4 KB
    Loading...
  2. 2026-08-16Add Qwen3.8-27B-Uncensored-xCloud (abliterated, o_proj+down_proj s=0.8)f094fee2.5 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