CatalogWantedMirrorsLedgerOperating StatementDocsStudio statusCard system
alt.freeweights.abliterated / Qwen3-8B-abliterated
From: mlabonne <huggingface.co/mlabonne>Newsgroups: alt.freeweights.abliteratedSubject:

Qwen3-8B-abliterated

Date: 2026-11-15Commit: 30c72faSigned: 2DC28CC7B26A283B ✓

Text generation model with 8.2B parameters. Safetensors weights, 32.8 GB in all. Published by mlabonne under Apache-2.0. Changed from the base model.

8.2B params32.8GBQwen3-8B-abliteratedsafetensorspermissive
FactsParams8.2BSize32.8 GBFormatssafetensorsTasktext generationLicenseApache-2.0From sourcepermissive
$ fw get mlabonne/Qwen3-8B-abliterated
# manifest sha256 2fc045c2e688… · 32.8 GB · 18 files
$ fw verify ./Qwen3-8B-abliterated mlabonne/Qwen3-8B-abliterated
From sourceHosted (signed)mlabonne/Qwen3-8B-abliteratedLast check: 2026-11-15 03:35 UTC
Run at homeGet help

Get help: custom fine-tune or private deploy. We review your setup before we quote.

Fine-tune of Qwen3-8B.

Works with

Run24Fine-tune29Convert15Evaluate13Scan13Apps58
OverviewModel cardFilesUse itLicenseProvenanceModel tree

Overview

Size
32.8 GB
Weights
32.8 GB (safetensors)
Parameters
8.2B
Formats
safetensors
Task
text generation
License
Apache-2.0 (permissive)
Files
18, each with a sha256
Publisher
mlabonne
Added
2026-11-15
Section
alt.freeweights.abliterated
Lineage

Changed from the base model. The uploader's repo is named abliterated: refusal behavior was removed from Qwen/Qwen3-8B by the uploader. Read the model card at the source for the method and the layers touched.

Base model: Qwen/Qwen3-8B

Where these bytes come from
Publisher at the pinned commit→You

Sources are tried in order. A file counts only when its size and sha256 match the manifest. Sources are checked when you ask for them, not on a timer.

Sources
BadgeHostLast checkResult
From sourcehuggingface.co2026-11-15 03:35 UTCpassed

Model card

From the publisher at commit 30c72fa. Shown as plain text with links; images open at the publisher.

🐹 Qwen3-8B-abliterated#

Image: image/png

Qwen3 Abliterated 0.6B • 1.7B • 4B • 8B • 14B • 30B-A3B

This is an uncensored version of Qwen/Qwen3-8B created with a new abliteration technique. See this article to know more about abliteration.

This is a research project to understand how refusals and latent fine-tuning work in LLMs. I played with different sizes of Qwen3 and noticed there was no one-size-fits-all abliteration strategy. In addition, the reasoning mode interfered with non-reasoning refusals, which made it more challenging. This made me iterate over different recipes and significantly consolidate my scripts with accumulation and better evaluations.

Note that this is fairly experimental, so it might not turn out as well as expected.

I recommend using these generation parameters: temperature=0.6, top_k=20, top_p=0.95, min_p=0.

✂️ Abliteration#

The refusal direction is computed by comparing the residual streams between target (harmful) and baseline (harmless) samples. The hidden states of target modules (e.g., o_proj) are orthogonalized to subtract this refusal direction with a given weight factor. These weight factors follow a normal distribution with a certain spread and peak layer. Modules can be iteratively orthogonalized in batches, or the refusal direction can be accumulated to save memory.

Finally, I used a hybrid evaluation with a dedicated test set to calculate the acceptance rate. This uses both a dictionary approach and NousResearch/Minos-v1. The goal is to obtain an acceptance rate >90% and still produce coherent outputs.

Files

18 files · 32.8 GBWeights 32.8 GB (safetensors). Sizes and sha256 from the signed manifest.
Hide all 18 files with their sha256
FileSizesha256
.gitattributes1.57 KB34448b82c17d60fec9b65b1f093c115ddbaadc04beb1b0140b6bfed2e012a930
README.md2.50 KBf86289aac41585b5161678b49a2bb73f42f388bb5545a2842cd30032effe9ceb
added_tokens.json707 Bc0284b582e14987fbd3d5a2cb2bd139084371ed9acbae488829a1c900833c680
config.json728 Bfed1cff9926b3e7ab5137be4861cf0d89b4b50ff4873d905eff9ac98c23a476f
generation_config.json214 B81051cd3f6e77013827148d0b8a6ead93f8ac390d5ab805f849199f0af6a08db
merges.txt1.67 MB8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5
model-00001-of-00007.safetensors4.97 GBf166dbe4c7b8ea760ed98dc00538aef49661b8b19bec33b9e1e4017e20a1688d
model-00002-of-00007.safetensors4.83 GB977ec927d68862a22bc79cef95441b7e3e8cd63f75d2b02f5a478c391ec56034
model-00003-of-00007.safetensors4.83 GBfec93db78f279e0c3e37eb3caeaf2c9ccb3f56db4606edc832c69efbf60991f2
model-00004-of-00007.safetensors5.00 GBe206f8da1726fb779fbc5cd46e67cb0361caca9d30472c85649c684d4a6f63b3
model-00005-of-00007.safetensors4.83 GBda0fbdbd2309afffbe4d2a72bf943be5b5adc77544435fa508d88040dd6d729c
model-00006-of-00007.safetensors4.83 GBecbff6633142c99d61a4287e61049177c82526c7295a8258027a3d6ee59ce598
model-00007-of-00007.safetensors3.46 GBa73aac17dc5253f04c1aef1a09e9d8b97c817af244e7d74ae05de166e6329f86
model.safetensors.index.json32.9 KBf0ddaa5d307f2a16ee137b12fa51e34028d17d07e30e3f201546897d1f64329f
special_tokens_map.json613 B76862e765266b85aa9459767e33cbaf13970f327a0e88d1c65846c2ddd3a1ecd
tokenizer.json11.4 MBaeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
tokenizer_config.json9.76 KB3c0884a30471f4f542dc89630f62a380bb70a341fafda826136a7be921fec7ea
vocab.json2.78 MBca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910
VersionsEach time the pinned upstream commit or file list changed in this repo. Newest first.
Upstream commitDateContents
30c72fa348 pinned now2026-11-15 03:35 UTC18 files · 32.8 GB

Use it

$ fw get mlabonne/Qwen3-8B-abliterated
# manifest sha256 2fc045c2e688… · 32.8 GB · 18 files
$ fw verify ./Qwen3-8B-abliterated mlabonne/Qwen3-8B-abliterated
fw.pyChecks the signature and every sha256, tries each source in order, resumes.
$ python3 fw.py get https://freeweights.allternit.com/manifests/mlabonne/Qwen3-8B-abliterated.json --dir ./Qwen3-8B-abliterated $ python3 fw.py verify ./Qwen3-8B-abliterated ./mlabonne__Qwen3-8B-abliterated.json
hfThe Hugging Face CLI with one setting. Same repo id.
$ HF_ENDPOINT=https://freeweights.allternit.com hf download mlabonne/Qwen3-8B-abliterated --revision 30c72fa348f37c72d12ecbb259068ddee98aa9ed --local-dir ./Qwen3-8B-abliterated
transformersLoad the weights in Python.
import os os.environ["HF_ENDPOINT"] = "https://freeweights.allternit.com" from transformers import AutoModelForCausalLM, AutoTokenizer tok = AutoTokenizer.from_pretrained("mlabonne/Qwen3-8B-abliterated", revision="30c72fa348f37c72d12ecbb259068ddee98aa9ed") model = AutoModelForCausalLM.from_pretrained("mlabonne/Qwen3-8B-abliterated", revision="30c72fa348f37c72d12ecbb259068ddee98aa9ed", device_map="auto")
vLLMServe an OpenAI-compatible API.
$ HF_ENDPOINT=https://freeweights.allternit.com vllm serve mlabonne/Qwen3-8B-abliterated --revision 30c72fa348f37c72d12ecbb259068ddee98aa9ed
Run at home

License

Name
Apache License 2.0
Id
Apache-2.0
Class
Permissive
Redistribution
yes
Listing
Hosted (signed)
Decided
License as declared by the publisher. No person read it. 2026-11-15
Text in
README.md
Text sha256
f86289aac41585b5161678b49a2bb73f42f388bb5545a2842cd30032effe9ceb
Terms gate
No extra terms beyond the license.

Read the license text

Licenses are checked against the license table. A custom license passes only after a person reads it.

Provenance

From: mlabonne <huggingface.co/mlabonne>Newsgroups: alt.freeweights.abliteratedSubject: Qwen3-8B-abliteratedDate: 2026-11-15Commit: 30c72faSigned: 2DC28CC7B26A283B ✓
Upstream
mlabonne/Qwen3-8B-abliterated
Pinned commit
30c72fa348f37c72d12ecbb259068ddee98aa9ed
Manifest
Qwen3-8B-abliterated.json · .minisig
Manifest sha256
2fc045c2e688b585dc0c245363b6d5076f857756e597ff07f6ee172e312e57f4
Signature
minisign, Ed25519, key 2DC28CC7B26A283B
Created
2026-11-15 03:35 UTC
Generator
freeweights-manifest/0.1.0
Takedowns
None.

How to verify

$ curl -O https://freeweights.allternit.com/manifests/mlabonne/Qwen3-8B-abliterated.json $ curl -O https://freeweights.allternit.com/manifests/mlabonne/Qwen3-8B-abliterated.json.minisig $ minisign -Vm Qwen3-8B-abliterated.json -P RWQ7KGqyx4zCLXtUFOqvfv8L3g+8fpKJ1FsuxPyOYMF+4KufbUxKq1Oq $ shasum -a 256 <file> # compare with the Files tab

A manifest whose signature fails must be rejected. A file whose size or sha256 does not match must be discarded. Verify a download

Model tree

Parent

Qwen/Qwen3-8B · this model is a finetune of it

1 model in the catalog name this one as their parent.

Quantizations
1
Quantizations 1
Get it
Download
$ HF_ENDPOINT=https://freeweights.allternit.com hf download mlabonne/Qwen3-8B-abliterated
Open in Freeweights Studio

Pull, verify and run it in the Allternit app on Desktop, web or phone.

Signed manifest
minisign, key 2DC28CC7B26A283B. Qwen3-8B-abliterated.json. Provenance. Takedown history: none.
Source
huggingface.co/mlabonne/Qwen3-8B-abliterated

Variants

1 quantized or adapter build of Qwen3-8B-abliterated. Each has its own license check and page.

SubjectSizeParamsLicenseDate
mlabonne_Qwen3-8B-abliterated-GGUFHosted (signed) From: bartowski in alt.freeweights.abliterated130 GB—Permissive2026-11-15

Related models

All of alt.freeweights.abliterated →
From: xtools-at <huggingface.co/xtools-at>Subject: DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic
From: mlx-community <huggingface.co/mlx-community>Subject: DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-OptiQ-4bit
From: mlx-community <huggingface.co/mlx-community>Subject: gemma-4-26B-A4B-it-heretic-msq-2.6bit
From: coder3101 <huggingface.co/coder3101>Subject: gemma-4-31B-it-heretic
From: Ilya626 <huggingface.co/Ilya626>Subject: gemma-4-31B-it-SDFT-Heretic-RP
From: llmfan46 <huggingface.co/llmfan46>Subject: gemma-4-31B-it-uncensored-heretic