← back to catalog · registered 2026-08-27 06:02

atakhadivi/Qwen3.8-2B-Uncensored-GGUF

Name: atakhadivi/Qwen3.8-2B-Uncensored-GGUF
Author: atakhadivi

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/atakhadivi%2FQwen3.8-2B-Uncensored-GGUF"

Response includes

classification m8
files 3
hub_downloads_all_time 16,927
author_summary 1 models
readme_text full

10 credits · hourly refresh · ~4 KB payload Get an API key →

Abliteration classifier · v1.0.0

Primary method

Repackaging (quantization)

Applied on top of direct removal inherited from the base model.

Confidence

MEDIUM

Inherited from base model

M1 Direct removal →

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.

Downloads · lifetime

17K

10K last 30d - active

Likes

Model age

6w ago

created 2026-08-27

Downloads over time

Aug 26 → Oct 11 · 47 snapshots · spans 46 days

Metadata

Quantizations

Q4_K

Files by quantization

Q4_K 1 file 1.19 GB

Qwen3.8-2B-Uncensored-Q4_K_M.gguf

1.19 GB

61713f6d

download

Auxiliary files 2 files 3.26 KB

README.md	1.70 KB	05e06110	download
.gitattributes	1.55 KB	8a87fcaf	download

README current version from Hugging Face

Qwen3.8-2B Uncensored — GGUF (Q4_K_M)

GGUF quantization of empero-ai/Qwen3.8-2B with refusals removed via Heretic v1.4.0 abliteration.
Built from insraq/Qwen3.5-2B-EmperoAI-Qwen3.8-Distill-Heretic-Abliterated (Heretic v1.4.0 decensored build,
reproducible recipe included in that repo). 2B parameters, Qwen3.5-architecture hybrid (Gated DeltaNet + full attention).

File

File	Quant	Size	SHA256
Qwen3.8-2B-Uncensored-Q4_K_M.gguf	Q4_K_M	1.27 GB	`61713f6d29b507f0ddd2ed989683e3a5e997b920b0c4901ee8f332321d4b0cc9`

Recommended for phones / PocketPal / SBCs. CPU-only is fine at this size.

Runtime requirements

llama.cpp >= b1-0379a19 (Qwen3.5 / Gated DeltaNet architecture support; May 2026+ build).
PocketPal AI: needs a build new enough for the qwen35 architecture.
No vision projector (text-only GGUF). No MTP head (converted with --no-mtp; the abliterated source
does not include the speculative head).

Usage (llama.cpp)

llama-cli -m Qwen3.8-2B-Uncensored-Q4_K_M.gguf \
  --temp 0.6 --top-p 0.95 --top-k 20 -c 2048 -n 512

Use the embedded chat template (conversation mode). This is a reasoning model: answers open with a think block.
Verified output on Apple Silicon CPU: 31.7 t/s generation.

Model card note

This is an uncensored model — the refusal direction was removed. It will not refuse harmful or illegal
requests. Research / personal use only. Apache-2.0 inherited from the Qwen3.5-2B base.

Credits

Base: Qwen/Qwen3.5-2B (Alibaba Qwen team)
Distillation: empero-ai/Qwen3.8-2B (Qwen3.8 2.4T A95B -> Qwen3.5-2B)
Abliteration: insraq (Heretic v1.4.0)
GGUF conversion: llama.cpp convert_hf_to_gguf.py --no-mtp, Q4_K_M quantization

README history 2 versions

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

2026-08-27Upload README.md with huggingface_hub649ee941.9 KB
Loading...
2026-08-27Upload README.md with huggingface_hub466a1cd1.7 KB
Loading...