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

Zach2255/qwen2.5-7b-abliterated-by-zach0028

Zach2255 Qwen 7.6B
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/Zach2255%2Fqwen2.5-7b-abliterated-by-zach0028"
Response includes
  • classification m1
  • files 9
  • benchmarks 16 entries
  • hub_downloads_all_time 73
  • author_summary 1 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
73
22 last 30d - stable
Likes
2
Descendants
1
in 1 direct fork
Model age
6mo ago
created 2026-04-07
Downloads over time
Now79→from31↑155%
2947658431 on Apr 1579 on Oct 1179 on Oct 9AprMayJunJulAugSepOct
Apr 15 → Oct 11 · 65 snapshots · spans 179 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
BBH average 0.48553638604228827 OpenLLM-v2
IFEval instruct 0.7961630695443646 OpenLLM-v2
IFEval-Prompt 0.7208872458410351 OpenLLM-v2
MATH lvl 5 0 OpenLLM-v2
MMLU-Pro 0.4286901595744681 OpenLLM-v2
Entertainment 1.3 UGI
Hazardous 2.9 UGI
Natural Intelligence 15.76 UGI
Political lean -14.7% UGI
Sensitive-Info 15.62 UGI
SocPol 0.8 UGI
UGI 23.75 UGI
Willingness (10) 4 UGI
W10-Adherence 4 UGI
W10-Direct 4 UGI
Writing 29.72 UGI

Genealogy 1 direct fork

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 fr
Tags
safetensors qwen2 abliterated uncensored research en fr arxiv:2406.11717 base_model:Qwen/Qwen2.5-7B-Instruct base_model:finetune:Qwen/Qwen2.5-7B-Instruct license:apache-2.0 region:us

Related

Total size
14.2 GB
Files
9
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-04-11 12:38

Files by quantization

Auxiliary files 9 files 14.2 GB
model.safetensors 14.2 GB 115df60b download
tokenizer.json 10.9 MB 1a7aabe7 download
experiment_results.json 49.0 KB 64222689 download
README.md 2.86 KB 2eef61a2 download
chat_template.jinja 2.45 KB bdf7919a download
.gitattributes 1.53 KB 52373fe2 download
config.json 1.34 KB a16f5408 download
tokenizer_config.json 665 B 7d75d3bb download
generation_config.json 242 B 0075e8b4 download

README current version from Hugging Face


license: apache-2.0
language:

  • en
  • fr
    base_model: Qwen/Qwen2.5-7B-Instruct
    tags:
  • abliterated
  • uncensored
  • research

Qwen2.5-7B Abliterated

This model is a modified version of Qwen2.5-7B-Instruct with its refusal mechanism removed via abliteration.

Method

Abliteration was performed on the 10 deepest layers (18-27) using the projected refusal direction method from NousResearch, with norm preservation enabled. The full process is documented in the companion article and is fully reproducible via the open-source toolkit below.

Research Artifact — Legal Notice

This model is a scientific research artifact, released solely for research and educational purposes. It is not a product and is not intended for commercial or production use.

It was produced as part of a technical analysis published on UnderTech, with the explicit goal of demonstrating and documenting the abliteration technique from Arditi et al. (2024) in a reproducible way. As a research artifact released in an academic/non-commercial setting, this model falls within the research exemption under the EU AI Act.

No warranty

This model is provided "AS IS", without warranty of any kind, express or implied, including but not limited to warranties of merchantability, fitness for a particular purpose, and non-infringement.

Limitation of liability

To the maximum extent permitted by applicable law, in no event shall the author be liable for any claim, damages, or other liability — whether in an action of contract, tort, or otherwise — arising from, out of, or in connection with this model, its use, or other dealings in the model.

User responsibility

By downloading or using this model, you acknowledge and agree that:

  • You are solely responsible for any content generated and for compliance with all applicable laws and regulations in your jurisdiction.
  • You will not use this model to generate or facilitate harmful, illegal, or unethical content.
  • You will not hold the author liable for any direct or indirect misuse of this model by yourself or any third party.
  • You acknowledge that this is a research artifact, not a commercial product.

This model inherits the Apache 2.0 license of Qwen2.5-7B-Instruct, which also includes warranty disclaimers and limitation of liability clauses that apply to this derivative work.

Resources

README history 1 version

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

  1. 2026-04-11Add model card with legal disclaimerface44f2.9 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